text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: |
The nature and origin of the cold interstellar medium (ISM) in early type galaxies are still a matter of debate, and understanding the role of this component in galaxy evolution and in fueling the central supermassive black holes requires more observational constraints. Here, we present a multi-wavelength study of the ISM in eight nearby, X-ray and optically bright, giant elliptical galaxies, all central dominant members of relatively low mass groups. Using far-infrared spectral imaging with the [*Herschel*]{} Photodetector Array Camera & Spectrometer (PACS), we map the emission of cold gas in the cooling lines of \[[C]{}\]$\lambda157\mu$m, \[[O]{}\]$\lambda$63$\mu$m, and \[[O]{}b\]$\lambda$145$\mu$m. Additionally, we present H$\alpha$+\[[N]{}\] imaging of warm ionized gas with the Southern Astrophysical Research (SOAR) telescope, and a study of the thermodynamic structure of the hot X-ray emitting plasma with [*Chandra*]{}. All systems with extended H$\alpha$ emission in our sample (6/8 galaxies) display significant \[[C]{}\] line emission indicating the presence of reservoirs of cold gas. This emission is co-spatial with the optical H$\alpha$+\[[N]{}\] emitting nebulae and the lowest entropy soft X-ray emitting plasma. The entropy profiles of the hot galactic atmospheres show a clear dichotomy, with the systems displaying extended emission line nebulae having lower entropies beyond $r
\gtrsim1$ kpc than the cold-gas-poor systems. We show that while the hot atmospheres of the cold-gas-poor galaxies are thermally stable outside of their innermost cores, the atmospheres of the cold-gas-rich systems are prone to cooling instabilities. This provides considerable weight to the argument that cold gas in giant ellipticals is produced chiefly by cooling from the hot phase. We show that cooling instabilities may develop more easily in rotating systems and discuss an alternative condition for thermal instability for this case. The hot atmospheres of cold-gas-rich galaxies display disturbed morphologies indicating that the accretion of clumpy multiphase gas in these systems may result in variable power output of the AGN jets, potentially triggering sporadic, larger outbursts. In the two cold-gas-poor, X-ray morphologically relaxed galaxies of our sample, NGC 1399 and NGC 4472, powerful AGN outbursts may have destroyed or removed most of the cold gas from the cores, allowing the jets to propagate and deposit most of their energy further out, increasing the entropy of the hot galactic atmospheres and leaving their cores relatively undisturbed.
author:
- |
N. Werner$^{1,2}$, J. B. R. Oonk$^{3}$, M. Sun$^{4}$, P. E. J. Nulsen$^{5}$, S. W. Allen$^{1,2,6}$, R. E. A. Canning$^{1,2}$,A. Simionescu$^{7}$, A. Hoffer$^{8}$, T. Connor$^{8}$, M. Donahue$^{8}$, A. C. Edge$^{9}$, A. C. Fabian$^{10}$, A. von der Linden$^{1,2,11}$, C. S. Reynolds$^{12}$, M. Ruszkowski$^{13,14}$\
$^1$Kavli Institute for Particle Astrophysics and Cosmology, Stanford University, 452 Lomita Mall, Stanford, CA 94305-4085, USA\
$^2$Department of Physics, Stanford University, 382 Via Pueblo Mall, Stanford, CA 94305-4060, USA\
$^3$ASTRON, Netherlands Institute for Radio Astronomy, P.O. Box 2, 7990 AA Dwingeloo, The Netherlands\
$^4$Department of Physics, University of Alabama in Huntsville, Huntsville, AL 35899, USA\
$^{5}$Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138, USA\
$^6$SLAC National Accelerator Laboratory, 2575 Sand Hill Road, Menlo Park, CA 94025, USA\
$^7$Institute of Space and Astronautical Science (ISAS), JAXA, 3-1-1 Yoshinodai, Chuo-ku, Sagamihara, Kanagawa, 252-5210 Japan\
$^8$Physics & Astronomy Department, Michigan State University, East Lansing, MI 48824-2320, USA\
$^9$Institute for Computational Cosmology, Department of Physics, Durham University, Durham, DH1 3LE, UK\
$^{10}$Institute of Astronomy, Madingley Road, Cambridge CB3 0HA, UK\
$^{11}$Dark Cosmology Centre, Niels Bohr Institute, University of Copenhagen, Juliane Maries Vej 30, DK-2100 Copenhagen, Denmark\
$^{12}$Department of Astronomy and the Maryland Astronomy Center for Theory and Computation, University of Maryland, College Park, MD 20742, USA\
$^{13}$Department of Astronomy, University of Michigan, 500 Church Street, Ann Arbor, MI 48109, USA\
$^{14}$Michigan Center for Theoretical Physics, 3444 Randall Lab, 450 Church St, Ann Arbor, MI 48109, USA\
bibliography:
- 'clusters.bib'
title: |
The origin of cold gas in giant elliptical galaxies\
and its role in fueling radio-mode AGN feedback
---
galaxies: active – galaxies: elliptical and lenticular, cD – galaxies: ISM – accretion – infrared: galaxies – X-rays: galaxies
Introduction
============
Recent observations and simulations strongly suggest that the growth and evolution of giant early type galaxies is closely tied to that of their central supermassive black holes through a well regulated feedback cycle [e.g. @silk1998; @magorrian1998; @croton2006; @sijacki2007]. Many fundamental aspects of this feedback process, such as the nature of the material feeding the black holes in active galactic nuclei (AGN), jet formation, and the heating, cooling and detailed physics of the interstellar medium (ISM) are not well understood. Among the best laboratories to test theories for the formation and growth of massive galaxies are nearby giant ellipticals, groups and clusters of galaxies: systems where hot X-ray emitting plasma may be cooling and accreting onto the galaxies and interacting with their AGN.
-------------- ------------------ --------------------- ---------- -------------------------------- ------------------------------------- ------------------------------ -------------------------------- -------------------------------- ---------------------
Galaxy $d$ Scale $z$ $L_{\rm K}$ $L_{\rm{H}\alpha+[{N\,{\sc{II}}}]}$ $L_{\rm radio}$ $L_{\rm X}$ $P_{\rm jet}$ SFR
(Mpc) (arcsec kpc$^{-1}$) ($\times10^{40}$ erg s$^{-1}$) ($\times10^{39}$ erg s$^{-1}$) ($ ($\times10^{41}$ erg s$^{-1}$) ($\times10^{41}$ erg s$^{-1}$) $M_\odot$ yr$^{-1}$
\times10^{38}$ erg s$^{-1}$)
NGC 1399 20.9$^\ddagger $ 9.9 0.004753 2.69 9.3$^\star $ 1.52 5.68 $21.9\pm7$ 0.171
NGC 4472 16.7$^\ddagger $ 12.4 0.003326 3.97 5.8 1.20 2.96 $80.7\pm23.5$ 0.099
NGC 4636 14.7$^\dagger$ 14.0 0.003129 1.20 5.5$^\star $ 0.28 3.32 $3.0\pm0.8$ 0.027
NGC 5044 31.2$^\dagger$ 6.6 0.009280 1.65 45.6$^\star $ 0.59 5.87 $6.0$ 0.073
NGC 5813 32.2$^\dagger$ 6.4 0.006578 2.30 14.0$^\star $ 0.26 10.63 $1.2$ 0.063
NGC 5846 27.1$^\dagger$ 7.6 0.005717 2.11 24.6 0.26 6.25 $7.4\pm2.6$ 0.074
NGC 6868 26.8$^\dagger$ 5.6 0.009520 1.74 23.8 1.00 1.01 – 0.078
NGC 7049 29.9$^\dagger$ 7.1 0.007622 2.32 31.8 0.84 1.23 – 0.117
\[galaxies\]
-------------- ------------------ --------------------- ---------- -------------------------------- ------------------------------------- ------------------------------ -------------------------------- -------------------------------- ---------------------
In massive galaxy clusters, with central cooling times shorter than the Hubble time, the central brightest cluster galaxies (BCG) are often surrounded by spectacular, filamentary optical H$\alpha$+\[[N]{}\] emission-line nebulae extending up to 70 kpc from the core of the BCG . These nebulae of ionized gas seem to be co-spatial with warm (1000-2000 K) molecular hydrogen seen in the near-infrared [e.g. @jaffe1997; @falcke1998; @donahue2000; @edge2002; @hatch2005; @jaffe2005; @johnstone2007; @oonk2010; @lim2012] and with large quantities ($10^{8}-10^{11.5}$ M$_{\odot}$) of cold ($<50$ K) molecular gas traced by CO [e.g. @edge2001; @edge2003; @salome2003; @mcdonald2012]. Recently, observations with the [*Herschel Space Observatory*]{} [@pilbratt2010] also revealed the presence of far-infrared (FIR) cooling lines of \[[C]{}\], \[[O]{}\], and \[[N]{}\] in the X-ray bright cores of Abell 1068, Abell 2597 [@edge2010], and the Centaurus, Perseus, and Virgo clusters [@mittal2011; @mittal2012; @werner2013]. In nearby systems, such as the Centaurus, Perseus, and Virgo clusters, where the spatial distribution of the \[[C]{}\] line could be mapped, it was found to be extended and co-spatial with optical H$\alpha$+\[[N]{}\], far-ultraviolet [C]{} [in M 87, @sparks2012], and soft X-ray emission. The filamentary nebulae in the centers of massive cool-core galaxy clusters thus contain multi-phase material spanning a temperature range of over 5 orders of magnitude, from $<50$ K to $\sim10^{7}$ K. This gas often appears to be interacting with the jets and the buoyant relativistic plasma from the central AGN [@fabian2003b; @hatch2006; @canning2013]. Despite the large quantities of cold gas, many extended nebulae show no evidence for recent star-formation.
The majority of AGN in these systems are in a quiescent, so called ‘radio’-mode, where they are accreting at a modest rate. The accretion flows in this mode, though optically faint, often drive powerful relativistic jets, extending to large distances, which can have a profound impact on their surroundings. Using [*Chandra*]{} X-ray observations of nine nearby, X-ray luminous elliptical galaxies, @allen2006 found a tight correlation between the Bondi accretion rates [see @bondi1952] from the hot X-ray emitting ISM and the power in the relativistic jets [though see @russell2013]. The jet powers were determined from the work required to inflate bubbles of relativistic plasma associated with the most recent cavities in the hot X-ray emitting atmospheres of the galaxies [@churazov2002]. The relationship between jet power and accretion rate of the hot ISM, however, raises the question: what is the role of the cold gas in the AGN feedback cycle?
--------- ---------------- ---------------- --------------- ---------- --------------- --------------- --------------------------------------------- -----------------------------
Galaxy [*Herschel*]{} [*Herschel*]{} [*Chandra*]{} Detector [*Chandra*]{} SOAR/SOI SOAR/SOI SOAR/SOI
Obs. ID Exp. (s) Obs. ID Exp. (ks) Obs. date filter Exp (s)
NGC1399 1342239492 8161 319 ACIS-S 49.9 2012 Oct. 9 CTIO 656375-4/6916-78 $3\times600$ / $3\times600$
4172 ACIS-I 36.7 2013 Aug. 4 Goodman 1.68$''$ long slit, KOSI600 grating $3\times480$
9530 ACIS-S 59.3 2013 Sep. 8 Goodman 3.0$''$ long slit, KOSI600 grating $4\times480$
NGC4472 1342234992 8161 321 ACIS-S 19.1
NGC4636 1342236884 5442 323 ACIS-S 35.8 2009 June 1 CTIO 6600/75 $3\times780$
CTIO 6120/140 $3\times500$
NGC5044 1342238376 24723 9399 ACIS-S 53.4 2010 April 10 CTIO 6649/76 $3\times1200$
3926 ACIS-I 54.5 CTIO 6520/76 $3\times720$
4415 ACIS-I 73.3
NGC5813 1342238158 10880 5907 ACIS-S 47.4 2008 July 6 CTIO 6600/75 $3\times900$
9517 ACIS-S 98.8 CTIO 6120/140 $3\times720$
NGC5846 1342238157 5442 788 ACIS-S 17.3 2009 June 1 CTIO 6600/75 $3\times900$
7923 ACIS-I 75.9 CTIO 6120/140 $3\times540$
NGC6868 1342215929 5442 3191 ACIS-I 18.6 2012 Oct. 9 CTIO 6600/75 $2\times900$, $1\times600$
11753 ACIS-I 56.8 2012 Oct. 9 CTIO 6563/75 $2\times900$, $1\times600$
NGC7049 1342216658 5442 5895 ACIS-I 2.2 2012 Oct. 9 CTIO 6600/75 $2\times900$, $1\times600$
2012 Oct. 9 CTIO 6563/75 $1\times900$, $2\times600$
\[obs\]
--------- ---------------- ---------------- --------------- ---------- --------------- --------------- --------------------------------------------- -----------------------------
\
Here we present a multi-wavelength study of 8 nearby, X-ray bright, giant elliptical galaxies, all central dominant members of relatively low mass groups. We use FIR data obtained with the [*Herschel*]{} Photodetector Array Camera & Spectrometer (PACS) in the lines of \[[C]{}\]$\lambda157$ $\mu$m, \[[O]{}\]$\lambda$63$\mu$m, and \[[O]{}b\]$\lambda$145$\mu$m that are excellent probes of cold $\sim100$ K gas; optical data from the Southern Astrophysical Research (SOAR) telescope in the lines of H$\alpha$+\[[N]{}\] probing warm, ionized $\sim$10,000 K gas; and X-ray data from the [*Chandra X-ray Observatory*]{}, tracing the hot 5-20 million K plasma permeating these systems.
The galaxy sample {#sample}
-----------------
Our target list is drawn from the parent sample of @dunn2010, who identified the optically and X-ray brightest giant elliptical/S0 galaxies within a distance $d\leq100$ Mpc and with declination $\rm{Dec.} \geq -45$. Motivated by the goal of studying the properties of cold gas in these types of systems in detail, we selected only those galaxies with distance $d<35$ Mpc and with relatively bright H$\alpha$+\[[N]{}\] emission as reported by @macchetto1996. All of the systems in our sample are the central dominant galaxies of their respective groups. The properties of the galaxies are summarized in Table \[galaxies\]. The table lists the distances of the galaxies [@tonry2001; @blakeslee2009], their K-band near-infrared luminosities determined using the magnitudes measured by the 2MASS survey [@jarrett2003], H$\alpha$+ \[[N]{}\] luminosities [@macchetto1996], radio luminosities at 1.4 GHz [@condon1998; @condon2002], bolometric X-ray luminosities [@osullivan2001], jet-powers [@allen2006; @david2009; @shurkin2008; @randall2011], and star-formation rates (SFR) estimated using a combination of far-ultraviolet (FUV) data from [*Galex*]{} and mid- infrared (MIR) data from [*Wise*]{}, employing the technique developed by @hao2011.
The near-infrared K-band luminosities, and by implication the stellar masses, span a range of approximately a factor of three, with NGC 4636 being the least and NGC 4472 the most luminous. The H$\alpha$+\[[N]{}\] luminosities span a range of a factor of eight, with NGC 5044 being the most luminous, and with NGC 4636 and NGC 4472 on the low luminosity end. NGC 5044 boasts a particularly spectacular network of radial filaments, extending out to a radius of at least $r\sim$10 kpc [@gastaldello2009; @david2011]. The SFRs are small, typically of the order of 0.1 $M_\odot$ yr$^{-1}$, and they do not correlate with the H$\alpha$ luminosities. All eight galaxies have central, active radio jets, also spanning a range of radio luminosities and jet-powers. The X-ray luminosities of the galaxies span a range of an order of magnitude.
Observations and data analysis {#analysis}
==============================
Far-infrared spectroscopy with [*Herschel*]{} PACS
--------------------------------------------------
We observed the FIR cooling lines of \[[C]{}\]$\lambda157\mu$m, \[[O]{}\]$\lambda63\mu$m, and \[[O]{}b\]$\lambda145\mu$m in our sample of 8 galaxies with the PACS integral-field spectrometer [@poglitsch2010] on the [*Herschel Space Observatory*]{}. Table \[obs\] gives a summary of the observations. The observations were taken in line spectroscopy mode with chopping-nodding to remove the telescope background, sky background and dark current. A chopper throw of 6 arcmin was used. For 7 systems, with a relatively compact H$\alpha$ emission region, the observations were taken in pointed mode targeting the centers of the galaxies. For NGC 5044, we used raster mapping in $3\times3$ steps of 23.5 arcsec to match the extent of the H$\alpha$ nebula. The observations were reduced using the [HIPE]{} software version 8.2.0, using the PACS [ChopNodLineScan]{} pipeline script. This script processes the data from level 0 (raw channel data) to level 2 (flux calibrated spectral cubes).
During the final stage of the reduction the data were spectrally and spatially rebinned into $5\times5\times\lambda$ cubes. In the following we will refer to these cubes as the rebinned cubes. Each spatial pixel, termed spaxel, in these cubes has a size of $9.4\times9.4$ arcsec$^{2}$. The cubes thus provide us with a field of view (FoV) of $47\times47$ arcsec$^{2}$. For the wavelength regridding the parameters [oversample]{} and [upsample]{} were set to 2 and 1, respectively. This means that one spectral bin corresponds to the native resolution of the PACS instrument.
The integrated \[[C]{}\] line fluxes were in all cases, except for NGC 5044, obtained by spatially integrating the $5\times5$ spaxels from the rebinned cubes. For NGC 5044 the integrated line flux is obtained after projecting the individual rebinned cubes onto the sky. No point-spread function (PSF) correction is applied as the \[[C]{}\] emission, in all cases where it is detected, is found to be extended.
To visualize the extent of the \[[C]{}\] emission in our galaxies, we have created sky maps of the \[[C]{}\] emission by using the [specProject]{} task in [HIPE]{} and the [hrebin]{} task in [IDL]{}. In the following we will refer to these data cubes as the projected cubes. A pixel size of 6 arcsec was chosen in order to Nyquist sample the beam, the full-width-at-half-maximum (FWHM) of which is 12 arcsec at the observed wavelength of the \[[C]{}\] line. We only consider spatial bins where the signal-to-noise ratio of the integrated line flux is greater than 2. We have compared the integrated line fluxes obtained from the level 2 rebinned cubes with those from the projected cubes and find that they are consistent. Velocity and velocity width maps were constructed by fitting a single Gaussian to the projected data.
For the \[[O]{}\] and \[[O]{}b\] lines, we report line fluxes obtained from either the central spaxel or the central $3\times3$ spaxels. The PSF corrected fluxes should be viewed as lower limits to the true \[[O]{}\] and \[[O]{}b\] line emission in these galaxies. For NGC 5044 the integrated \[[O]{}\] and \[[O]{}b\] fluxes are obtained after projecting the individual rebinned cubes onto the sky.
-------------- --------------------------- ---------------------- ------------------------------------- --------------- --------------- -------------------------------------
galaxy Line Observation duration Line Flux in central spaxel Observed FWHM Line shift Integrated Line Flux
(s) ($10^{-14}$ erg s$^{-1}$ cm$^{-2}$) (km s$^{-1}$) (km s$^{-1}$) ($10^{-14}$ erg s$^{-1}$ cm$^{-2}$)
NGC 1399 [C]{}$\lambda157.7\mu$m 2250 $<0.2$ 300$^{\rm f}$ - -
[O]{}$\lambda63.2\mu$m 2484 $<2.8$ 300$^{\rm f}$ - -
[O]{}b$\lambda$145.5 3360 $<0.2$ 300$^{\rm f}$ - -
NGC 4472 [C]{}$\lambda157.7\mu$m 2250 $0.2\pm0.04$ $255\pm36$ $99\pm15$ $1.04\pm0.16$
[O]{}$\lambda63.2\mu$m 2484 $<2.5$ 255$^{\rm f}$ - -
[O]{}b$\lambda145.5\mu$m 3360 $<0.2$ 255$^{\rm f}$ - -
NGC 4636 [C]{}$\lambda157.7\mu$m 1500 $2.6\pm0.06$ $361\pm6$ $22\pm3$ $10.52\pm0.37$
[O]{}$\lambda63.2\mu$m 1656 $1.3\pm0.2$ (1.9) $233\pm29$ $75\pm12$ -
[O]{}b $\lambda145.5\mu$m 2240 $<0.5$ 233$^{\rm f}$ - -
NGC 5044 [C]{}$\lambda157.7\mu$m 6750 - - - $31.73\pm4.76$
[O]{}$\lambda63.2\mu$m 7452 - - - $7.1\pm1.8$
[O]{}b$\lambda145.5\mu$m 10080 - - - $1.0\pm0.5$
NGC 5813 [C]{}$\lambda157.7\mu$m 3000 $1.4\pm0.04$ $419\pm10$ $96\pm4$ $8.55\pm0.26$
[O]{}$\lambda63.2\mu$m 3312 $1.0\pm0.2$ (1.5) $273\pm36$ $30\pm15$ -
[O]{}b$\lambda145.5\mu$m 4480 $0.2\pm0.04$ (0.4) $608\pm88$ $100\pm37$ $1.4\pm0.2$
NGC 5846 [C]{}$\lambda157.7\mu$m 1500 $2.2\pm0.06$ $477\pm10$ $-25\pm4$ $13.61\pm0.36$
[O]{}$\lambda63.2\mu$m 1656 $<2.5$ 477$^{\rm f}$ - -
[O]{}b$\lambda145.5\mu$m 2240 $<0.3$ 477$^{\rm f}$ - -
NGC 6868 [C]{}$\lambda157.7\mu$ 1500 $5.7\pm0.08$ $510\pm6$ $125\pm3$ $21.61\pm0.44$
[O]{}$\lambda63.2\mu$m 1656 $3.0\pm0.3$ $506\pm31$ $137\pm13$ $5.9\pm0.8$
[O]{}b$\lambda145.5\mu$m 2240 $0.5\pm0.06$ $579\pm50$ $125\pm21$ $2.0\pm0.3$
NGC 7049 [C]{}$\lambda157.7\mu$m 1500 $2.5\pm0.06$ $395\pm7$ $78\pm3$ $22.49\pm0.45$
[O]{}$\lambda63.2\mu$m 1656 $<3.3$ 395$^{\rm f}$ - -
[O]{}b$\lambda145.5\mu$m 2240 $<0.2$ 395$^{\rm f}$ - -
\[FIRlines\]
-------------- --------------------------- ---------------------- ------------------------------------- --------------- --------------- -------------------------------------
Notes: In brackets, we indicate the PSF corrected fluxes determined assuming the emission comes from a point-source, correcting the measured spaxel flux upward to account for the beam size. The PSF corrected fluxes can therefore be considered a lower limit to the flux integrated over the whole PACS area.
---------------- -------------------------------------
Galaxy $f_{\rm{H}\alpha+\rm{[NII]}}$
($10^{-13}$ erg s$^{-1}$ cm$^{-2}$)
NGC 1399 $<0.34$
NGC 4636 $2.7\pm0.4$
NGC 5044 $7.6\pm0.9$
NGC 5813 $2.2\pm0.3$
\[SOARfluxes\]
---------------- -------------------------------------
: H$\alpha$+\[[N]{}\] fluxes from the SOAR telescope.
----------------- ----------------- ----------------- ----------------- --------------- ----------------
Galaxy $n_{\rm e}$ $kT$ $P_{\rm e}$ $K$ $t_{\rm cool}$
(cm$^{-3}$) (keV) (keV cm$^{-3}$) (keV cm$^2$) ($10^7$ yr)
NGC 1399 $0.167\pm0.005$ $0.950\pm0.009$ $0.158\pm0.005$ $3.14\pm0.07$ 4.6
NGC 4472 $0.154\pm0.008$ $0.860\pm0.012$ $0.132\pm0.007$ $2.99\pm0.12$ 4.3
NGC 4636 $0.072\pm0.001$ $0.535\pm0.020$ $0.039\pm0.001$ $3.11\pm0.07$ 5.2
NGC 5044 $0.068\pm0.005$ $0.582\pm0.037$ $0.039\pm0.004$ $3.51\pm0.27$ 5.9
NGC 5813 $0.074\pm0.003$ $0.616\pm0.041$ $0.046\pm0.005$ $3.50\pm0.31$ 5.7
NGC 5846 $0.076\pm0.006$ $0.677\pm0.033$ $0.052\pm0.005$ $3.77\pm0.26$ 6.2
NGC 6868 $0.072\pm0.002$ $0.768\pm0.036$ $0.055\pm0.003$ $4.45\pm0.22$ 7.8
\[deprojected\]
----------------- ----------------- ----------------- ----------------- --------------- ----------------
H$\alpha$+\[[N]{}\] imaging and spectroscopy {#SOARanalysis}
--------------------------------------------
We performed narrow-band imaging observations for all systems in our sample, except for NGC 4472, using the SOAR Optical Imager (SOI) on the 4.1 m SOAR telescope. See Table \[obs\] for details on the observations. For each galaxy, we obtained two narrow-band images, one centered on the H$\alpha$ emission and the other in an adjacent emission-line free band. We reduced the images using standard procedures in the [IRAF MSCRED]{} package. The pixels were binned by a factor of two, for a scale of 0.154 arcsec per pixel. The typical seeing was $\sim$1$''$. Spectrophotometric standard stars were observed for each exposure. More detail on the SOI data reduction can be found in @sun2007. For continuum subtraction, we follow the isophote fitting method described in @goudfrooij1994. We assumed line ratios of \[[N]{}\]$
\lambda6583$/H$\alpha = 1.5$ and \[[N]{}\]$\lambda6548$/\[[N]{}\]$\lambda6583=1/3$, which are consistent with the typical values in @goudfrooij1994.
On August 4 and September 8 2013, we also took spectra of NGC 1399 using the Goodman spectrograph on the SOAR telescope. We used a 600 lines per mm grating for the wavelength range of 4350–6950 Å. The first run was taken with a 1.68$''$ slit, along the NS direction. The second run was taken with a 3.0$''$ slit, at a position angle of 302 degrees (almost along the major axis of the galaxy). We took three 480 second exposures on the first night and four 480 second exposures on the second night. Before and after each exposure we took a quartz flat and a comparison spectrum of an Fe lamp. Bias correction, flat-fielding, and wavelength calibration were all performed using standard [IRAF]{} procedures. For flux calibration, we took spectra of the spectrophotometric standard stars LTT7379 and LTT1020.
[*Chandra*]{} X-ray data
------------------------
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
The archival [*Chandra*]{} data were reprocessed using the CIAO (version 4.3) software package. We cleaned the data to remove periods of anomalously high background. Table \[obs\] summarizes the net exposure times after cleaning. Our data reduction procedures are described in detail in @million2010a [@million2010b].
To produce 2D maps of projected thermodynamic quantities, we identified spatial regions using the contour binning algorithm [@sanders2006b], which groups neighbouring pixels of similar surface brightness until a desired signal-to-noise ratio threshold is met. We adopted a signal-to-noise ratio of 18–25 ($\sim$320–630 counts per region) in the 0.6–2.0 keV band, which gives us small enough regions to resolve substructure, yet provides enough counts to achieve better than 5 per cent precision on the temperature. Point sources were identified using the CIAO task WAVDETECT and excluded from all regions used for spectral analysis. Separate photon-weighted response matrices and effective area files were constructed for each spectral region.
We modelled the spectra extracted from each spatial region with the SPEX package [@kaastra1996]. The spectrum for each region was fitted with a model consisting of an absorbed single-phase plasma in collisional ionization equilibrium, with the temperature and spectral normalization (emission measure) as free parameters. The line-of-sight absorption column densities, $N_{\rm H}$, were fixed to the values determined by the Leiden/Argentine/Bonn radio survey of [H]{} [@kalberla2005].
From the best fit emission measures, $Y=\int n_{\rm H} n_{\rm e} {\rm d}V$, where $n_{\rm H} = n_{\rm e} /1.2$ is the hydrogen number density and $V$ is the emitting plasma volume, we determined the projected electron densities, $n_{\rm e}$, for each region assuming a fixed column depth of $l=20$ kpc. Using these electron densities and the best fit plasma temperatures ($kT$), we determined the pressures ($P_{\rm e}=n_{\rm e}kT$) and entropies ($K=kT/n_{\rm e}^{2/3}$). We note that the choice of the assumed column depth for these projected maps is arbitrary and it does not affect the magnitude of the observed azimuthal variations in the derived thermodynamic properties.
Because the statistical quality of the data for the X-ray faintest galaxies in our sample, NGC 6868 and NGC 7049, does not allow us to produce 2D maps of thermodynamic properties, we only present X-ray images for these systems. Background-subtracted images were created in six narrow energy bands, spanning 0.5–2.0 keV. These were flat fielded with respect to the median energy for each image and then co-added to create the broad-band X-ray images.
To determine the deprojected thermodynamic properties of the hot gas, we extracted, for each galaxy (except NGC 7049), a set of azimuthally averaged spectra from concentric annuli. We modeled these spectra simultaneously in the 0.6–2.0 keV band, with the [XSPEC]{} [version 12.5 @arnaud1996] spectral fitting package, using the [projct]{} model [see @werner2012 for details]. We have determined the azimuthally-averaged deprojected radial profiles of electron density and temperature, from which we derived the deprojected radial profiles for entropy, electron pressure, and cooling time. We define the cooling time as the gas enthalpy divided by the energy radiated per unit volume of the plasma: $$t_{\mathrm{cool}}=\frac{\frac{5}{2}(n_{\mathrm{e}}+n_{\mathrm{i}})kT}{n_{\mathrm{e}}n_{\mathrm{i}}\Lambda(T)},$$ where the ion number density $n_{\mathrm{i}}=0.92n_{\mathrm{e}}$, and $\Lambda(T)$ is the cooling function for Solar metallicity tabulated by @schure2009. We caution that, because the deprojected thermodynamic properties are determined assuming spherical symmetry, for the 5/7 analyzed galaxies with disturbed morphologies, the derived values have significant systematic uncertainties.
Results
=======
Using [*Herschel*]{} PACS, we detect FIR \[[C]{}\]$\lambda157\mu$m line emission in 7/8 systems (see Fig. \[fig:C2sample\] and Table \[FIRlines\]). The detections of \[[O]{}\]$\lambda63.2\mu$m and \[[O]{}b\]$\lambda145.5\mu$m lines are relatively weak, and are only seen in 4/8 and 3/8 systems, respectively (see Fig. \[fig:O1sample\], \[fig:O1bsample\] and Table \[FIRlines\]). No cold neutral gas is detected in NGC 1399 and the \[[C]{}\] detection in NGC 4472 is relatively weak. Table \[FIRlines\] lists the observed lines for all the target galaxies, their rest frame wavelengths and observation durations. For the central spaxel, we present the observed line fluxes and the $2\sigma$ upper limits where appropriate, the observed FWHM, and the observed line shifts with respect to the systemic velocity determined based on optical data. In the brackets, we give the PSF corrected fluxes determined assuming that the emission comes from a point-source, correcting the measured spaxel flux upward to account for the beam size. The PSF corrected fluxes can therefore be considered a lower limit to the flux integrated over the whole PACS area. For sources that allow us to confirm that the emission is extended we report non-PSF corrected fluxes. The last column gives the line fluxes spatially integrated over the $5\times5$ spaxel ($47''\times47''$) [*Herschel*]{} PACS field of view.
The \[[O]{}\]/\[[C]{}\] line ratios determined from the central spaxel in systems with significant \[[O]{}\] detections are in the range of 0.4–0.8. The $2\sigma$ upper limits on the fluxes of the \[[O]{}\]$\lambda63\mu$m and \[[O]{}b\]$\lambda145\mu$m lines given in Table \[FIRlines\] were determined assuming that their velocity widths are equal to the FWHM of the \[[C]{}\] line.
![The H$\alpha$+\[[N]{}\] luminosity (bottom panel) from SOAR for NGC 1399, NGC 5044, NGC 5813, and NGC 4636 (see Table \[SOARfluxes\]) and from @macchetto1996 for NGC 4472, NGC 5846, NGC 6868, and NGC 7049 (see Table \[galaxies\]), 70/24 $\mu$m infrared continuum ratio , and K-band luminosity [top panel, based on the 2MASS survey, @jarrett2003] plotted against the spatially integrated \[[C]{}\] luminosity measured by [ *Herschel*]{} PACS. The grey doted lines on the bottom panel indicate constant \[[C]{}\] over H$\alpha$+\[[N]{}\] luminosity ratios of 0.4 and 0.8. These ratios are remarkably similar for 6/8 galaxies in our sample. The outliers, NGC 4472 and NGC 1399, have little or no cold gas. They are indicated in red. []{data-label="Ha_CII"}](Ha_Ir_mK.pdf){width="1\columnwidth"}
![Jet powers [@allen2006; @shurkin2008; @david2009; @randall2011] plotted against the spatially integrated \[[C]{}\] luminosity measured by [*Herschel*]{} PACS. Galaxies with little or no cold gas are indicated in red. For the jet-powers in NGC 5044 and NGC 5813 no errorbars were published [@randall2011; @david2009].[]{data-label="Pj_CII"}](Pj.pdf){width="1\columnwidth"}
As indicated in Section \[sample\] and in Table \[galaxies\], according to @macchetto1996 all our targets are H$\alpha$ luminous. Fig. \[fig:Ha\] shows narrow band H$\alpha$+\[[N]{}\] images obtained with the SOAR telescope of our six targets with extended line-emitting nebulae. The nebulae in NGC 6868 and NGC 7049 show regular extended disk-like morphologies reaching radii $r\sim2-3$ kpc [see also @macchetto1996]. On the other hand, NGC 5044, NGC 5813, NGC 5846, and NGC 4636 show filamentary emission. NGC 5044 has the largest optical emission line luminosity, with a rich, dense network of filaments extending from the core out to a radius of $\sim10$ kpc [see also @david2011; @gastaldello2009]. The nebulae in NGC 5813 show a radial bipolar distribution in the direction of buoyant bubbles rising in the hot X-ray emitting atmosphere of the galaxy [see also @randall2011]. The bright H$\alpha$+\[[N]{}\] emission in NGC 4636 and NGC 5846 appears relatively compact, concentrated in the innermost $r\lesssim1.5$ kpc of the galaxies, but the morphology of the nebulae is clearly filamentary exhibiting signs of interaction with the central AGN.
Our new SOAR data allow robust constraints on the H$\alpha$+\[[N]{}\] flux in four galaxies, NGC 5044, NGC 4636, NGC 5813, and NGC 1399 (see Table \[SOARfluxes\]). The NGC 5846 data were not taken in photometric conditions, and for NGC 6868 and NGC 7049, the filters available at the time of the observations made the continuum subtraction less reliable than what is achieved for other galaxies. Our H$\alpha$+\[[N]{}\] flux for NGC 4636 is consistent with the values published by @macchetto1996 and @buson1993. For NGC 5813, our result is consistent with @goudfrooij1994 but is almost a factor of two higher than the flux measured by @macchetto1996. For NGC 5044, our flux is significantly higher than the values measured by @goudfrooij1994 and @macchetto1996, but is consistent with the result of @rickes2004. The large spatial extent of the nebula surrounding NGC 5044 combined with the small field-of-view of the instrument (1.4$'\times2.2'$) may have prevented @macchetto1996 from a robust determination of the local background.
{width="95.00000%"}
{width="95.00000%"}
{width="95.00000%"}
{width="105.00000%"}
{width="105.00000%"}
{width="105.00000%"}
{width="105.00000%"}
{width="105.00000%"}
{width="105.00000%"}
{width="102.00000%"}
{width="102.00000%"}
{width="102.00000%"}
{width="99.00000%"}
{width="99.00000%"}
{width="99.00000%"}
{width="102.00000%"}
{width="102.00000%"}
{width="102.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
{width="100.00000%"}
Contrary to the results of @macchetto1996, we detected no H$\alpha$+\[[N]{}\] emission in NGC 1399. @goudfrooij1994 derived an H$\alpha$+\[[N]{}\] flux of $4.2\pm0.2\times10^{-14}$ erg s$^{-1}$ cm$^{-2}$, much smaller than the value measured by @macchetto1996. Our SOI H$\alpha$+\[[N]{}\] imaging analysis of NGC 1399, with either adjacent narrow-band or R band image for the continuum subtraction, results in negative net flux around the center of the galaxy, indicative of H$\alpha
$ absorption. Therefore, we took additional long-slit spectra with the Goodman spectrograph on SOAR. Our Goodman spectra give a 5 $\sigma$ limit on the EW of the emission line of 0.1 $\AA$. Assuming that all the H$\alpha$+\[[N]{}\] emission is from the central 12$''$ radius, as shown by the H$\alpha$+\[[N]{}\] image from @macchetto1996, this EW upper limit translates to an H$\alpha$+\[[N]{}\] flux limit of $3.4\times10^{-14}$ erg s$^{-1}$ cm$^{-2}$. The lack of a spectroscopic detection of optical emission lines in NGC 1399 is inconsistent with previous reports of detections through narrow-band filters. We suspect that the previously reported detections of morphologically smooth line emission, where the emitting region has the same morphology as the old stellar populations, may be due to color gradients in the stellar population of the galaxy. Currently, we do not have SOAR data for NGC 4472.
In Fig. \[Ha\_CII\], we plot the H$\alpha$+\[[N]{}\] luminosities (bottom panel), 70/24 $\mu$m infrared continuum ratios [middle panel, @temi2007], and the K-band luminosity [top panel, based on the 2MASS survey, @jarrett2003] against the spatially integrated \[[C]{}\] luminosity measured by [*Herschel*]{} PACS. The ratio of the \[[C]{}\] over H$\alpha$+\[[N]{}\] luminosity is remarkably similar $\sim0.4$–0.8 for 6/8 galaxies in our sample. For the same 6/8 systems, the \[[C]{}\]$\lambda157\mu
$m line emission is spatially extended and appears to follow the distribution of the H$\alpha$+\[[N]{}\] filaments (see the left panels of Fig. \[fig:C2maps1\] and \[fig:C2maps2\]). In the extended network of filaments of NGC 5044 at $r\gtrsim1$ kpc, the \[[C]{}\]/(H${\alpha}$+\[[N]{}\]) ratios also remain similar as a function of position. In the two FIR faint systems (shown by the red data points) the ionized H$\alpha$ emission is either concentrated in the nucleus of the galaxy or is not present (NGC 1399).
The \[[C]{}\] luminosities, which are likely excellent tracers of the cold molecular gas mass [@crawford1985], do not correlate with the K-band luminosities (see Table \[galaxies\]), which trace the stellar masses of the galaxies. There is also no correlation between the \[[C]{}\] luminosities of systems with extended emission line nebulae and the 70/24 $\mu$m infrared continuum ratios. In the FIR faint systems indicated by the red data points these ratios are smaller (see the central panel of Fig. \[Ha\_CII\]). The 70/24 $\mu$m infrared continuum ratio is a good tracer of interstellar dust [@temi2007].
Fig. \[Pj\_CII\] shows that the \[[C]{}\] luminosities in systems with spatially extended emission do not correlate with the jet powers determined from the work required to inflate bubbles of relativistic plasma associated with the cavities in the hot X-ray emitting atmospheres of the galaxies listed in Table \[galaxies\]. In our relatively small sample, the galaxies with little or no cold gas (indicated in red) have higher jet powers than the systems with significant \[[C]{}\] detections.
The central panels of Fig. \[fig:C2maps1\] and \[fig:C2maps2\] show the velocity distributions of the \[[C]{}\] line emitting gas with respect to the velocities of the host galaxies. In NGC 6868 and NGC 7049 the velocity distribution of the \[[C]{}\] emitting gas indicates that the cold gas forms a rotating disk. This is consistent with the overall morphology of the optical line emission nebulae. The velocity distribution of the \[[C]{}\] emitting gas in the other four systems with extended emission does not show any coherent structure.
The projected velocity dispersions measured from the \[[C]{}\] lines along our line of sight are generally in the range between 100 km s$^{-1}$ and 200 km s$^{-1}$, with the highest measured values reaching $\sim280$ km s$^{-1}$.
Fig. \[fig:Chandrasample\] shows maps of the pressure and entropy in the hot X-ray emitting intra-group/interstellar medium for the galaxies where the cold gas has a filamentary morphology. In all of these systems, the filaments are co-spatial with the lowest entropy hot X-ray emitting gas. They also appear to anti-correlate and interact with the AGN inflated bubbles of relativistic plasma, which provide significant non-thermal pressure in the X-ray atmospheres of the galaxies and therefore appear as regions of low projected X-ray pressure on our maps. This is most clearly seen for the southwestern filament in NGC 5813, which appears to be pushed aside by the large, older, outer bubble. The inner portion of the filament also appears on the side and partly on the top of the small, younger, inner southwestern bubble that is currently being inflated in the core [for detailed discussion on the bubbles see @randall2011]. Similar interaction, with the filaments being pushed around by the bubbles, is seen in NGC 5846 and NGC 4636, both of which show ‘cavities’ in the H$\alpha$+\[[N]{}\] images that coincide with pressure decrements in their X-ray spectral maps. NGC 5044 shows a strongly disturbed morphology due to both AGN activity and the intra-group medium sloshing in the gravitational potential of the system [@gastaldello2009; @david2009; @david2011; @osullivan2013].
Fig. \[fig:Chandrasample2\] shows the pressure and entropy maps for NGC 4472 and NGC 1399 - systems with little or no cold and warm gas. Compared to the systems with filamentary H$\alpha$ nebulae, their thermodynamic maps have relatively regular morphologies. These systems are among the nearby giant ellipticals with the most relaxed X-ray morphologies at small radii [@werner2012].
NGC 6868 and NGC 7049 do not have X-ray data with the quality that would allow us to extract detailed maps of thermodynamic properties and in Fig. \[xrayimages\] we only show their X-ray images. The images show extended X-ray haloes, co-spatial with the H$\alpha$+\[[N]{}\] emission, surrounding the galaxies. The X-ray image of NGC 6868 shows a relatively complex disturbed morphology [see also @machacek2010]. The X-ray data of NGC 7049 are very shallow and only allow us to detect the extended X-ray emission. Four galaxies - NGC 5044, NGC 6868, NGC 7049, and NGC 5813 - have an X-ray point source associated with the central AGN.
The deprojected central densities and pressures measured at $r=0.5$ kpc in the morphologically relaxed NGC 1399 and NGC 4472 appear significantly higher than in the other, more disturbed, \[[C]{}\] bright systems (see Table \[deprojected\]). All \[[C]{}\] bright galaxies appear to have similar central densities and pressures. The entropies at $r=0.5$ kpc span a small range of 3.0–4.5 keV cm$^{2}$ and the central cooling times are 4–$8\times10^7$ yr. At radii $r\gtrsim1$ kpc the entropies of the galaxies containing cold gas (except NGC 6868) are systematically lower than those of NGC 1399 and NGC 4472 (see Fig. \[entropies\]). The measured relatively low central densities and pressures of the morphologically disturbed systems may partly be due to the additional non-thermal pressure support in the central regions of the these systems. However, we also suspect a bias due to departures from spherical symmetry. Due to the non-uniform surface brightness within the circular annuli used in the deprojection analysis, densities in the outer shells may be somewhat overestimated, leading to underestimated central densities. For the morphologically disturbed systems, we repeated the deprojection analysis using wedges and found, that despite the systematic uncertainties, at radii $r\gtrsim1$ kpc the difference in the shapes of the profiles shown in Fig. \[entropies\] remains robust.
![0.5–2 keV [*Chandra*]{} images of the X-ray faintest galaxies in our sample, NGC 6868 and NGC 7049, with the H$\alpha$+\[[N]{}\] contours overlaid. The statistical quality of the X-ray data does not allow us to produce 2D maps of thermodynamic properties for these two systems.[]{data-label="xrayimages"}](NGC6868im.pdf "fig:"){width="22.00000%"} ![0.5–2 keV [*Chandra*]{} images of the X-ray faintest galaxies in our sample, NGC 6868 and NGC 7049, with the H$\alpha$+\[[N]{}\] contours overlaid. The statistical quality of the X-ray data does not allow us to produce 2D maps of thermodynamic properties for these two systems.[]{data-label="xrayimages"}](NGC7049im.pdf "fig:"){width="22.00000%"}
![Entropy profiles for a sample of morphologically relaxed giant ellipticals [NGC 4472, NGC 1399, NGC 4649, NGC 1407, NGC 4261; @werner2012], with no extended optical emission line nebulae (in black), and for five galaxies from our sample with \[[C]{}\] emission indicating the presence of reservoirs of cold gas (NGC 5044, NGC 5813, NGC 5846, NGC 4636, NGC 6868 - in red). The entropy in galaxies containing cold gas is systematically lower at $r\gtrsim1$ kpc. The only exception is NGC 6868 the entropy profile of which follows that of the cold-gas-poor galaxies. []{data-label="entropies"}](entropies_s.pdf){width="1\columnwidth"}
Discussion
==========
Properties of the cold ISM
--------------------------
The observed high \[[C]{}\] luminosities indicate the presence of large reservoirs of cold gas. Cold gas is evidently prevalent in massive giant elliptical galaxies with spatially extended H$\alpha$+\[[N]{}\] nebulae, even though their stellar populations are old and appear red and dead [@annibali2007]. NGC 5044 has previously also been detected in the CO(2–1) line with the 30 m IRAM telescope with a total flux of $6.6\times10^{-17}$ erg s$^{-1}$ cm$^{-2}$ (Jeremy Lim & Francoise Combes, private communication). Assuming a CO(2–1) to CO(1–0) ratio of 0.8, we obtain a \[[C]{}\]$\lambda158$$\mu$m over CO(1–0) flux ratio of 3817, which is within the range of the ratios (1500–6300) observed in normal and star forming galaxies and in Galactic molecular clouds [@crawford1985; @stacey1991].
Assuming the cold $\sim$100 K \[[C]{}\] emitting gas is in thermal pressure equilibrium with the surrounding hot X-ray emitting plasma, it will presumably form filaments with densities $\sim10^4$ cm$^{-3}$ and small volume filling fractions. As discussed in @fabian2003b [@fabian2008; @fabian2011] and @werner2013, the emission line filaments are likely to consist of many strands that have small volume filling fractions but large area covering factors. The soft X-ray emission that is always associated with filaments in cool core clusters [@sanders2007; @sanders2008; @sanders2009; @sanders2009b; @deplaa2010; @werner2011; @werner2010; @werner2013] indicates the presence of cooling hot plasma distributed between the threads. The hot plasma may be cooling both radiatively and by mixing with the cold gas in the filaments. Part of the soft X-ray emission from this cooling plasma gets absorbed by the cold gas in the strands [@werner2013].
The velocities inferred from the \[[C]{}\] line emission are consistent with those measured from the H$\alpha$+\[[N]{}\] lines by @caon2000, indicating that the different observed gas phases form multiphase filaments where all the phases move together. The velocity structure measured in NGC 6868 and NGC 7049 indicates that the cold gas in these systems is distributed in large extended rotating disks. Given the expected densities of the \[[C]{}\] emitting gas, its volume filling fraction must be low and the disks are also likely to be formed from many small clouds, sheets, and filaments.
The observed velocity dispersions of the \[[C]{}\] emitting gas are similar to the range of values measured from CO line widths and using optical and near-infrared integral field spectroscopy of 2000–10,000 K gas in the cooling cores of clusters [e.g. @edge2001; @ogrean2010; @oonk2010; @farage2012; @canning2013]. Observations indicate that, due to their small volume filling fraction and large area covering factors, filaments may be blown about by the ambient hot gas, potentially serving as good tracers of its motions [@fabian2003b]. The measured velocity dispersions of the \[[C]{}\] emitting gas, which reflect the range of velocities of the many small gas clouds/filaments moving through the galaxy, may therefore be indicative of the motions in the hot ISM. They are in general consistent with the limits and measurements obtained through X-ray line broadening and resonant line scattering observations with the Reflection Grating Spectrometers on [*XMM-Newton*]{} [@werner2009; @deplaa2012; @sanders2013].
The mid-infrared [*Spitzer*]{} spectra of all of our \[[C]{}\] bright galaxies overlapping with the sample of @panuzzo2011 - NGC 4636, NGC 5044, NGC 5813, NGC 5846, NGC 6868 - show the presence of dust, warm H$_2$ molecular gas, and for NGC 5044, NGC 6868, and NGC 4636 also PAH emission. @temi2007 show that, for NGC 4472 and NGC 1399, galaxies with little or no \[[C]{}\] detected, the infrared spectral energy distributions (SEDs) can be fully explained by circumstellar dust in the context of a steady state model of dust production in normal stellar mass loss followed by sputtering by the hot X-ray emitting gas. However, for all of the galaxies with extended \[[C]{}\] emission, the infrared SEDs observed with [*IRAS*]{} and [*Spitzer*]{} indicate the presence of true interstellar dust, well in excess of the predictions of the steady state model [@knapp1989; @temi2007; @temi2007b]. Although there are early reports of detections of [H]{} emission in NGC 5846 and NGC 4636 [@bottinelli1977; @bottinelli1979; @knapp1978], they were not confirmed by later re-observations [@krishna1983; @lake1984; @knapp1985].
@werner2013 showed that the \[[S]{}\]$\lambda6716/6731$ line ratios of the filaments in M 87 indicate very low densities in the H$\alpha$+\[[N]{}\] emitting 10,000 K phase. They concluded that, assuming subsonic turbulence in the filaments, the presence of significant magnetic pressure is required to keep this warm ionized gas in pressure equilibrium with the surrounding intra-cluster medium, indicating that the filaments are supported by magnetic fields of $B = 28-73\mu$G. The assumption of subsonic turbulence is motivated by the lack of \[[O]{}\] line emission, which would be produced by strong shocks. For 5/8 galaxies overlapping with the sample of @annibali2010, the \[[S]{}\]$\lambda6717/6731$ line ratios point to very low densities, $n_{\rm e}<26$ cm$^{-3}$, in the ionized gas, indicating that it is supported by magnetic fields. Such low densities appear to be common in the ionized, extended emission line nebulae in the cores of galaxy clusters. Significant magnetic fields threading the emission line nebulae were also inferred using arguments based on the integrity of the filaments in the Perseus Cluster [@fabian2008] and based on radio observations of the Faraday rotation measure in cooling core clusters [@taylor2001; @taylor2007; @allen2001; @feretti1999]. This magnetic support may be slowing or preventing the gravitational collapse of any molecular gas clouds traced by the \[[C]{}\] line emission that exceed the Jeans mass, preventing them from forming stars [e.g. see the discussion for NGC 1275 in @ho2009].
Heating and ionization of the cold ISM
--------------------------------------
In the six systems with significant cold gas mass reservoirs, the \[[C]{}\] line emission appears to be co-spatial with the H$\alpha$+\[[N]{}\] emission, and with the lowest entropy X-ray emitting plasma. In the same galaxies, the ratios of \[[C]{}\]/(H$\alpha$+\[[N]{}\]) emission appear to be similar (0.4–0.8), indicating that in these systems the \[[C]{}\] and H$\alpha$+\[[N]{}\] emission are powered by the same energy source.
Photoionization by young hot stars or by the central AGN is negligible as a source of excitation for the nebulae in these galaxies. @ferland2009 showed that the broad band emission-line spectra of the filamentary nebulae around central galaxies of cooling core clusters most likely originate in gas exposed to ionizing particles, either relativistic cosmic rays or hot X-ray emitting plasma penetrating into the cold gas. If the magnetized, ionized H$\alpha$ emitting phase forms a thin skin on the underlaying cold neutral gas, then the hot plasma particles must somehow overcome the obstacle presented by the magnetic fields. Based on the observations of the emission line filaments in M 87, @werner2010 and @werner2013 propose that shocks propagating within the hot X-ray emitting plasma as well as the movement of the filaments through this ambient hot gas (both as they are being uplifted and as they fall back) induce shearing around the filaments, thereby promoting mixing of the cold gas with the ambient hot medium via instabilities [e.g. @friedman2012]. @fabian2011 propose that the ionizing hot plasma in the core of the Perseus Cluster penetrates the cold filaments through magnetic reconnection diffusion [@lazarian2010; @lazarian2011], which may be induced by shearing instabilities and turbulence. The penetration of the cold gas by the hot plasma particles implies that the X-ray emitting gas cools through mixing and thus the filaments of cold gas grow continuously in mass. If the mixing of the cold and hot gas phases turns more violent and the energy input rate due to penetrating hot plasma becomes larger than the cooling rate, this process may also lead to the destruction of the filaments.
On the other hand, according to the scenario proposed by @churazov2013, the filaments may be powered by the reconnection of the stretched magnetic fields in the wakes of AGN inflated buoyantly rising bubbles, and do not necessarily grow in mass. The fact that the ratios of the \[[C]{}\]/(H$\alpha$+\[[N]{}\]) emission appear similar in both filamentary nebulae and rotating disks indicates that, despite the different morphologies, the gas is heated and ionized by the same energy source in both types of systems. But because the cold gas in the disky nebulae is not distributed in the wakes of AGN inflated bubbles, the scenario proposed by @churazov2013 could not explain their energy source. However, in the reconnection model, the powering of the cold gas could also be driven by the orbiting motion of the magnetized blobs/filaments through the hot plasma. This may lead to the sliding of the cold gas along the opposite-polarity magnetic fields and to the subsequent reconnection.
Despite its relatively low X-ray luminosity, the central pressure of the hot X-ray emitting plasma in the disky NGC 6868 is similar to the central pressures in the filamentary systems (see Table \[deprojected\]) indicating that in the absence of magnetic fields the heat flux from the hot into the cold phase would also be similar in both types of systems. It therefore appears that if the cold and hot phase come into direct contact - either due to turbulence, shearing motions, or the magnetic fields tied to the ionized phase becoming unstable - hot gas particles penetrating into the cold gas provide a viable mechanism for powering the filaments in all of the systems in our sample. The penetrating particles will heat and increase the degree of ionization of the molecular gas clouds present within the filaments, increasing their Jeans mass and slowing down the collapse of the gas clouds in the presence of magnetic fields.
This model, calculated by @ferland2009 considering emission from an optically thin cell of gas, however, predicts \[[O]{}\]/\[[C]{}\]$\sim21$, significantly higher than our observed range of 0.5–0.7. The ratios previously observed in the Perseus and Centaurus clusters, and in Abell 1068 and Abell 2597 are also low, in the range of 0.3–1 [@edge2010; @mittal2011; @mittal2012]. @mittal2012 conclude that these line ratios suggest that the lines are optically thick, implying a large reservoir of cold gas, which was not accounted for in previous inventories of the filament masses. Our measured ratios are consistent with the values determined previously in normal and starburst galaxies [@malhotra2001].
Origin of the cold ISM
----------------------
![The Field criterion, ${\Pi_{\rm F}}= {\kappa T/ ({n_{\rm e}}{n_{\rm H}}\Lambda(T) r^2)}$, as a function of radius for the galaxies that display extended emission line nebulae and \[[C]{}\] line emission (shown in red) and for a sample of arguably cold-gas-poor systems (shown in black). The clear dichotomy, with the cold gas-rich systems being thermally unstable out to relatively large radii, indicates that the cold gas originates from the cooling hot phase. The galaxy NGC 6868, which shows evidence for a rotating disk of gas, is shown with a dashed line. []{data-label="stability"}](field.pdf){width="1\columnwidth"}
A hotly debated question in the literature is whether the cold gas seen in the central giant elliptical galaxies in clusters and groups, and in early type galaxies with X-ray haloes, originated from the radiative cooling of the hot plasma, from stellar mass loss, or whether it was accreted in mergers with gas rich galaxies.
The difference between the entropy profiles of the systems with cool gas and those without it in Fig. \[entropies\] supports the argument that the cold gas originates by cooling from the hot phase. In more massive clusters, H$\alpha$ filaments are always co-spatial with soft X-ray emitting $\sim$0.5 keV gas [@sanders2007; @sanders2008; @sanders2009; @sanders2009b; @deplaa2010; @werner2011; @werner2010; @werner2013], indicating that the cold gas must be related to the hot phase. Furthermore, significant star formation and line emitting gas around the central galaxy are only found in systems where the central cooling time is short, or, equivalently, the central entropy is low [@rmn08; @cdv08].
The parameter $$\label{eqn:field_crit}
{\Pi_{\rm F}}= {\kappa T \over {n_{\rm e}}{n_{\rm H}}\Lambda(T) r^2},$$ where $\kappa$ is the thermal conductivity, $T$ is the temperature, $\Lambda$ is the cooling function, and ${n_{\rm e}}$ and ${n_{\rm H}}$ are the electron and hydrogen number densities, respectively, is a measure of the ratio of the conductive heating rate to the radiative cooling rate on scales comparable to $r$, i.e. the “Field criterion” for thermal instability [@field1965]. Fig. \[stability\] shows the Field criterion, ${\Pi_{\rm F}}$, as a function of the radius, where the systems with substantial quantities of cold gas are plotted in red. There is a clear dichotomy with the cold-gas-rich systems remaining unstable out to relatively large radii. This result strongly indicates that the cold gas is produced chiefly by thermally unstable cooling from the hot phase. A similar result was found for a sample of central galaxies in more massive clusters by @vcd08, who found minimum values of ${\Pi_{\rm F}}\lesssim 5$ in all systems with star formation. @vcd08 argued that the effective conductivity is suppressed by a factor of about 5 by magnetic fields, in which case all of their systems with young stars are thermally unstable by the Field criterion.
The Field criterion alone is insufficient to ensure that the gas is thermally unstable. In a spherical atmosphere, the gas is supported by buoyancy, so that an overdense blob tends to fall inwards on the free-fall timescale towards its equilibrium position where the surrounding gas has the same specific entropy and requires the same specific heating rate. This suppresses the growth of thermal instability unless the cooling time is comparable to the free-fall time or shorter [@cfn80; @bs89]. Using numerical simulations, @smq12 found that thermal instability is only significant when the cooling time is less than $\sim 10$ free fall times [see also @mccourt2012; @kunz2012; @gaspari2012; @gaspari2013]. However, this condition can be circumvented if an overdense gas blob is prevented from falling to its equilibrium position by the magnetic fields which pin the cloud to the ambient ICM [@nulsen1986], or by rotational support in addition to buoyancy.
If a cooling cloud has sufficient angular momentum that is retained while it cools, it can cool unstably onto a non-radial orbit. This provides an alternative condition for thermal instability, that the viscous diffusion length in a cooling time, $\sqrt{\nu {t_{\rm c}}}$, where $\nu$ is the kinematic viscosity and ${t_{\rm c}}$ is the cooling time, must be smaller than the size of a cloud. Analogous to the Field criterion, we define the stability parameter $$\label{eqn:visc_crit}
{\Pi_{\rm v}}= {\nu {t_{\rm c}}\over r^2},$$ which needs to be smaller than about unity for thermal instability to develop on scales comparable to $r$. Viscous stresses in a plasma [@b65] are much less sensitive to the structure of the magnetic field than conduction [e.g., @nm13] and, although the stresses have a different tensor character in magnetized and unmagnetized plasmas, the viscosity itself is the same. Because the underlying process is diffusion controlled by Coulomb collisions in both cases (electrons for ${\Pi_{\rm F}}$ and ions for ${\Pi_{\rm v}}$), the ratio of these two parameters is almost constant at ${\Pi_{\rm v}}/ {\Pi_{\rm F}}\simeq
0.0253$. Numerical simulations are needed to determine a more accurate stability threshold, but we expect that the viscous diffusion length needs to be somewhat smaller than $r$ for instability by this mechanism. The condition ${\Pi_{\rm F}}< 5$ translates to ${\Pi_{\rm v}}\lesssim 0.13$, or the diffusion length in one cooling time being $\lesssim 0.36 r$. Although it is unclear which stability threshold takes precedence, we conclude that it is sufficient to test only one of these parameters, which we take to be ${\Pi_{\rm F}}$, for stability.
Our [*Herschel*]{} data reveal that in two galaxies, NGC 6868 and NGC 7049, the cold gas forms rotating disks with radii $r\sim3$ kpc. The disks of cold gas in these systems rotate with orbital velocities of up to $250$ km s$^{-1}$. If this gas cooled from the hot phase, its rotation implies that the hot atmospheres of these systems have significant net angular momentum and the hot gas cools onto non- radial orbits. In the case of NGC 6868, this angular momentum may have been provided by gas sloshing indicated by the [*Chandra*]{} data [@machacek2010]. Because the atmosphere is aspherical, heating originating from the center of the galaxy cannot balance the cooling rate locally throughout the atmosphere and so should not be expected to stop hot gas from cooling onto an extended disk. Because of this, and because rotational support prevents infall, instabilities may develop more easily and the gas may become thermally unstable at higher ${\Pi_{\rm F}}$ than in non-rotating systems. This is consistent with our results presented in Fig. \[stability\], where ${\Pi_{\rm F}}$ for NGC 6868, shown with a dashed line, is higher than ${\Pi_{\rm F}}$ for the other systems containing non-rotating cold gas. The Field criterion still needs to be met, but the behavior of the stability parameter in NGC 6868 also indicates that the viscous stability criterion takes precedence.
However, in the context of the radiative cooling model, the presence of dust and PAHs within the cold gas clouds [@panuzzo2011] is intriguing. It suggests that at least some of the cold material originated from stellar mass loss [see also @voitdonahue2011; @donahue2011 for galaxy clusters]. Our understanding of dust formation is, however, limited and it cannot be ruled out that dust may somehow form within the cold gas clouds [@fabian1994b]. Recent [*Herschel*]{} observations suggest that in the most massive galaxies, the dust mass is unconnected to the stellar populations, leading to the suggestion that the dust and the cold gas in these systems have been acquired externally [@smith2012]. @rawle2012 show that the dust-to-stellar mass ratios in BCGs without line emitting nebulae are much smaller than in systems with cold gas. In the absence of cold gas clouds, the dust produced by stellar mass loss will get destroyed by sputtering and the tenuous gas, lost by stars, may get more easily mixed with and assimilated into the hot ISM. Therefore, if a galaxy loses all of its cold gas content (e.g. due to AGN activity), then stellar mass loss will likely become inefficient in rebuilding its cold gas reservoir. A galaxy that loses its cold gas content will therefore remain free of gas and dust until it again accretes cold material from the cooling hot phase or in a wet merger. This may have happened to NGC 1399 and NGC 4472. On the other hand, if a galaxy already contains clouds of cold gas (it either never lost its cold ISM or it accreted some cold material externally) then these clouds and filaments will help preserving the products of stellar mass loss (both the gas and the dust) by assimilating them as they plow through the galaxy.
In the systems with filamentary cold gas, the AGN jets and the buoyant bubbles of relativistic plasma seem to interact with the filaments, dragging them out from the center of the galaxy. As these filaments move through the hot ISM they may get penetrated by hot particles, which deposit energy into the cold gas heating and ionizing it. This interaction can both destroy the cold gas and contribute to the growth of its mass, depending on the energy input rate from the hot into the cold phase. In the context of this model, the total gas/dust mass will not correlate with the stellar mass of the galaxy, but it will depend on the balance between heating and cooling and on the interaction of the cold nebulae with the AGN [see also @mathews2013].
Fueling the AGN
---------------
The jet-powers, determined from the work required to inflate bubbles of relativistic plasma associated with the cavities in the hot X-ray emitting atmospheres, do not increase with the amount of cold gas in these galaxies. On the contrary, the two galaxies that lack any significant reservoirs of cold gas have the largest jet-powers in our sample. These galaxies, NGC 1399 and NGC 4472, also have the largest hot ISM densities in their cores. Accretion of hot gas in galaxies with high core densities can naturally result in higher jet powers, as has been demonstrated by the correlation between the Bondi accretion rates of hot gas and the observed jet powers in such ellipticals [@allen2006]. The pressure and entropy distributions of the hot ISM in the cores of these two galaxies look remarkably regular, with the X-ray cavities/radio-lobes appearing at larger radii of $\sim$4 kpc and $\sim$8 kpc for NGC 4472 and NGC 1399, respectively.
Other nearby giant elliptical galaxies that appear morphologically relaxed in their cores, such as NGC 4649, NGC 1407, NGC 4261, or NGC 1404, also do not display extended optical emission line nebulae [@macchetto1996; @baldi2009; @tremblay2009; @moustakas2010], indicating that they too are relatively depleted of cold gas. Interestingly, some of these morphologically relaxed cold-gas-poor systems, such as NGC 4261, have very large jet powers, with jets puncturing through the X-ray emitting galactic atmosphere to form giant lobes in the surrounding intragroup medium [@osullivan2011]. For others, such as NGC 1404, NGC 4649, and NGC 1407, however, the current jet powers are more modest [@shurkin2008; @giacintucci2012].
Galaxies containing large reservoirs of cold gas, on the other hand, in general display strongly disturbed X-ray morphologies, possibly relating to earlier AGN outbursts impacting the cores of these systems. In principle, the presence of cold gas may affect both accretion and jet heating in these systems. Accretion of clumpy multiphase gas (hot plasma mixed with cooler, filamentary gas) may result in variable accretion rates and power output of the AGN jets, potentially triggering sporadic, larger outbursts than would be typical from the steady accretion of hot gas alone. Because of the relatively small volume filling fraction of this cooler gas, these outbursts may be relatively short [arguments for sporadic cold/multi-phase accretion in radio mode AGN in galaxy clusters have also been presented by @rafferty2006; @russell2013].
As demonstrated by @morganti2013, radio-mechanical feedback does not only operate on the hot tenuous atmospheres of galaxies but also on the dense cold gas, which can be accelerated by jets to high speeds. Recent observations using ALMA also show that radio mode AGN activity can drive massive outflows of molecular gas from BCGs [@russell2013b; @mcnamara2013]. In systems with large cold gas reservoirs, jets may be more easily slowed by coupling to this material, and may therefore deposit most of their energy at smaller radii. This will result in more disturbed central thermodynamic distributions and, in particular may decrease the central hot ISM density, resulting in lower hot gas accretion rates between the relatively brief periods of multiphase accretion. Any reduction in the accretion rates from the hot gas would result in smaller jet powers, consistent with the observations presented here.
Numerical simulations [@wagner2011; @wagner2012] have shown that radio jets can be efficient in clearing the central regions of galaxies of their cold gas, provided that the cold gas clouds are porous, clumpy, and have small volume filling fractions, with individual clouds being relatively small, which again are characteristics consistent with the observations discussed here. Powerful jets may thus eventually destroy or remove most of the cold gas in the cores of the galaxies, allowing the jets to propagate further out and deposit their energy at larger radii, increasing the entropy of the hot galactic atmospheres. This in turn will allow the density of the hot gas in the core of the galaxy to increase, thus increasing the jet power. The central regions of galaxies cleared of cold gas may acquire relaxed morphologies within the central $r\sim5$ kpc over $
\sim5\times10^7$ yr or a few sound crossing times. The similarity of the thermodynamic profiles of elliptical galaxies with highly relaxed X-ray morphologies [see black data points in Fig. \[entropies\] and @werner2012] indicates that, after elliptical galaxies are cleared of cold gas, jet heating and cooling of their hot atmospheres comes to an equilibrium in an approximately universal manner. The galaxies may maintain these equilibria until they are disturbed by a merger event.
Conclusions
===========
Using FIR, optical, and X-ray data, we study the ISM in eight nearby, X-ray and optically bright, giant elliptical galaxies, all central dominant members of relatively low mass groups. We find that:
- All systems with extended H$\alpha$ emission in our sample (6/8 galaxies) display significant \[[C]{}\] line emission that traces $\sim$100 K gas.
- This emission is co-spatial with the optical H$\alpha$+\[[N]{}\] emitting nebulae and the lowest entropy soft X-ray emitting plasma.
- These systems have similar \[[C]{}\]/(H$\alpha$+\[[N]{}\]) ratios of 0.4–0.8, indicating that the \[[C]{}\] and H$\alpha$+\[[N]{}\] emission are powered by the same energy source. The likely dominant source of energy is the hot X-ray emitting plasma penetrating into the cold gas.\
- The entropy profiles of the hot galactic atmospheres show a clear dichotomy, with the systems displaying extended emission line nebulae having lower entropies beyond $r\gtrsim1$ kpc than the cold-gas-poor systems. We show that while the hot atmospheres of the cold-gas-poor galaxies are thermally stable outside of their innermost cores, the atmospheres of the cold-gas-rich systems are prone to cooling instabilities. This provides considerable weight to the argument that cold gas in giant ellipticals is produced chiefly by cooling from the hot phase. We show that cooling instabilities may develop more easily in rotating systems and discuss an alternative condition for thermal instability for this case.
- The hot atmospheres of cold-gas-rich galaxies display disturbed morphologies indicating that the accretion of clumpy multiphase gas in these systems may result in variable power output of the AGN jets, potentially triggering sporadic, larger outbursts. The jets may be more easily slowed by coupling to the dense cold material and may therefore deposit most of their energy at smaller radii, resulting in the observed disturbed central thermodynamic distributions.
- In the two cold-gas-poor, X-ray morphologically relaxed galaxies of our sample, NGC 1399 and NGC 4472, powerful AGN outbursts may have destroyed or removed most of the cold gas from the cores, allowing the jets to propagate and deposit most of their energy further out, increasing the entropy of the hot galactic atmospheres and leaving their cores relatively undisturbed. These observations indicate that radio-mechanical AGN feedback is likely to play a crucial role in clearing giant elliptical galaxies of their cold gas, keeping them ‘red and dead’.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work is based in part on observations made with Herschel, a European Space Agency Cornerstone Mission with significant participation by NASA. Support for this work was provided by NASA through award number 1428053 issued by JPL/Caltech. This work is based in part on observations obtained at the Southern Astrophysical Research (SOAR) telescope, which is a joint project of the Ministério da Ciência, Tecnologia, e Inovação (MCTI) da República Federativa do Brasil, the U.S. National Optical Astronomy Observatory (NOAO), the University of North Carolina at Chapel Hill (UNC), and Michigan State University (MSU). SWA acknowledges support from the U.S. Department of Energy under contract number DE-AC02-76SF00515. MR acknowledges the NSF grant AST 1008454 and NASA ATP grant (12-ATP12-0017).
Plots of the FIR line spectra\
==============================
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
[ngc5044\_red\_1\_pC\_cii\_obsv\_sub\_wav\_mapp\_3p\_cen21\_hwd13.pdf]{} (50,68)[\[CII\]]{} (65,68)[NGC 5044]{}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We explore a particular approach to the analysis of dynamical and geometrical properties of autonomous, Pfaffian non-holonomic systems in classical mechanics. The method is based on the construction of a certain auxiliary constrained Hamiltonian system, which comprises the non-holonomic mechanical system as a dynamical subsystem on an invariant manifold. The embedding system possesses a completely natural structure in the context of symplectic geometry, and using it in order to understand properties of the subsystem has compelling advantages. We discuss generic geometric and topological properties of the critical sets of both embedding and physical system, using Conley-Zehnder theory and by relating the Morse-Witten complexes of the ’free’ and constrained system to one another. Furthermore, we give a qualitative discussion of the stability of motion in the vicinity of the critical set. We point out key relations to sub-Riemannian geometry, and a potential computational application.'
address: ' Courant Institute of Mathematical Sciences, New York University, 251 Mercer Street, New York, NY 10012-1185. [email protected] '
author:
- Thomas Chen
title: ' CRITICAL MANIFOLDS AND STABILITY IN HAMILTONIAN SYSTEMS WITH NON-HOLONOMIC CONSTRAINTS '
---
Å[A\^]{}
[ H]{} §[[S]{}]{}
\[section\] \[section\] \[section\] \[section\] \[section\] \[section\] \[section\]
Introduction {#introduction .unnumbered}
============
We introduce and explore a particular approach to the analysis of autonomous, Pfaffian non-holonomic systems in classical mechanics, which renders them naturally accessible to the methods of symplectic and sub-Riemannian geometry. We note that typical examples of systems encountered in sub-Riemannian geometry emerge from optimal control, or ’vakonomic’ problems, which are derived from a different variational principle (minimization of the Carnot-Caratheodory distance) than the Euler-Lagrange equations of classical mechanical systems with non-holonomic constraints (the Hölder variational principle, cf. [@Ar1], and section III in this paper). The strategy is based on the introduction of an artificial Hamiltonian system with constraints that are compatible with the symplectic structure, constructed in a manner that it comprises the non-holonomic mechanical system as a dynamical subsystem on an invariant manifold. The main focus of the discussion in this paper aims at the geometrical and topological properties of the critical sets of both embedding and mechanical system, on the stability of equilibria, and an application of the given analysis to a computational problem.
There exists a multitude of different approaches to the description and analysis of non-holonomic systems in classical mechanics, stemming from various subareas of application. The geometrical approach given here has been strongly influenced by [@We] and [@Bra; @Bra2; @SoBr]. A construction for the Lagrangian case, which is closely related to what will be presented in section III, has been given in [@CaFa]. A different approach in the Hamiltonian picture is dealt with in [@vdSMa]. A geometrical theory of non-holonomic systems with a strong influence of network theory has been developed in [@YoKa]. The geometrical structure of non-holonomic systems with symmetries and the associated reduction theory, as well as aspects of their stability theory has been at the focus in the important works [@BlKrMaMu; @KoMa; @KoMa1; @ZeBlMa], and other papers by the same authors.
This paper is structured as follows. In section I, we introduce a class of Hamiltonian systems with non-integrable constraints. Given a symplectic manifold $(M,\omega)$ and a non-integrable, symplectic distribution $V$, we focus on the flow $\tPhi_t$ generated by the component $X_H^V$ of the Hamiltonian vector field $X_H$ in $V$. In section II, we study the geometry and topology of the critical set $\crit$ of the constrained Hamiltonian system. The main technical tool used for this purpose is a gradient-like flow $\phi_t$, whose critical set $\crit$ is identical to that of $\tPhi_t$. Assuming that the Hamiltonian $H:M\rightarrow\R$ is a Morse function, it is proved that generically, $\crit$ is a normal hyperbolic submanifold of $M$. Using Conley-Zehnder theory, we prove a topological formula for closed, compact $\crit$, that is closely related to the Morse-Bott inequalities. A second, alternative proof is given, based on the use of the Morse-Witten complex, to elucidate relations between the ’free’ and the constrained system. In section III, we give a qualitative, partly non-rigorous discussion of the stability of the constrained Hamiltonian system, and conjecture a stability criterion for the critically stable case. A proof of the asserted criterion, which would involve methods of KAM and Nekhoroshev theory, is beyond the scope of the present work. We derive an expression for orbits in the vicinity of a critically stable equilibrium that is adapted to the flag of $V$, and point out relations to sub-Riemannian geometry.
In section IV, we consider Hamiltonian mechanical systems with Pfaffian constraints. We show that for any such system, there exists an auxiliary constrained Hamiltonian system of the type introduced in section I. We study the global topology of the critical manifold of the constrained mechanical system, and again discuss the stability of equilibria. Finally, we propose a computational application, a method to numerically determine the generic connectivity components of the critical manifold.
A NON-INTEGRABLE GENERALIZATION OF DIRAC CONSTRAINTS {#gendir}
====================================================
Let $(M,\omega,H)$ be a Hamiltonian system, where $M$ is a smooth, symplectic $2n$-manifold with $C^\infty$ symplectic structure $\omega\in \Lambda^2(M)$, and where $H\in C^\infty(M)$ is the Hamiltonian function. For $p=1,\dots,2n$, let $\Lambda^p(M)$ denote the $C^\infty(M)$-module of $p$-forms on $M$. The Hamiltonian vector field $X_H\in \Gamma(TM)$ is determined by $$i_{X_H}\omega=-dH\;,$$ where $i$ stands for interior multiplication. Given a smooth distribution $W\subset TM$, $\Gamma(W)$ will denote the $C^\infty(M)$-module of smooth sections of $W$. The Hamiltonian flow is the 1-parameter group $\Phi_t\in {\rm Diff}(M)$ generated by $X_{H}$, with $t\in \R$, and $\Phi_0={\rm id}$. Its orbits are solutions of \[hameqmo\]\_t\_t(x)=X\_H(\_t(x))for all $x\in M$, and $t\in\R$.
Let us first recall some standard facts about Dirac constraints that will be useful in the subsequent construction. Let, for $f,g\in C^\infty$, \[poisson\]f,g=(X\_f,X\_g) denote the smooth, non-degenerate Poisson structure on $M$ induced by $\omega$. It is a derivative in both of its arguments, and satisfies the Jacobi identity $\lbrace f,\lbrace g,h\rbrace\rbrace+($cyclic$) = 0$, thus $(C^\infty(M),\lbrace\cdot,\cdot\rbrace)$ is a Lie algebra. Then, ( \[hameqmo\]) translates into \[Poisseqsmo\]\_t f(\_t(x))= H, f(\_t(x)) , for all $f\in C^\infty(M)$, and all $x\in M$, $t\in\R$.
Let $j:M'\hookrightarrow M$ be an embedded, smooth, $2k$-dimensional symplectic submanifold of $M$, endowed with the pullback symplectic structure $j^*\omega$. The Dirac bracket corresponds to the induced Poisson bracket on $M'$, $$\lbrace f,g\rbrace_D\;=\;
(j^*\omega)(X_{\tilde{f}},X_{\tilde{g}}) \; ,$$ defined for any pair of extensions $\tilde{f},\tilde{g}\in C^\infty(M)$ of $f,g\in C^\infty(M')$. If $M'$ is locally characterized as the locus of common zeros of some family of functions $G_i\in C^\infty(M)$, $i=1,\dots,2(n-k)$, the following explicit construction of the Dirac bracket can be given, [@MaRa]. Since $M'\subset M$ is symplectic, the $(n-k)^2$ functions locally given by $D_{ij}:=\lbrace G_i,G_j\rbrace$ can be patched together to define a matrix-valued $C^\infty$ function that is invertible everywhere on $M'$. The explicit formula for the Dirac bracket is locally given by \[dirbra\]f,g\_D= ,- ,G\_i D\^[ij]{}G\_j,,where $D^{ij}$ denotes the components of the inverse of $[D_{ij}]$.
This construction can be put into the more general context of a symplectic distribution.
A distribution $V$ over the base manifold $M$ is symplectic if $V_x$ is a symplectic subspace of $T_x M$ with respect to $\omega_x$, for all $x\in M$. Its symplectic complement $V^\perp$ is the distribution which is fibrewise $\omega$-skew orthogonal to $V$. Furthermore, an embedding $I\subset\R\hookrightarrow M$ that is tangent to $V$ is called $V$-horizontal.
Clearly, $V^\perp$ is by itself symplectic, and smoothness of $V$ and $\omega$ implies smoothness of $V^\perp$. Furthermore, the Whitney sum bundle $V\oplus V^\perp$ is $TM$. Thus, let $V$ denote an integrable, smooth, symplectic rank $2k$-distribution $V$ over $M$. Clearly, any section $X\in\Gamma(TM)$ has a decomposition $$X=X^V + X^{V^\perp}\;,$$ where $X^{V^{(\perp)}}\in\Gamma(V^{(\perp)})$, so that $\omega(X^V , X^{V^\perp})=0$. Furthermore, there exists an $\omega$-skew orthogonal tensor $\PV:TM\longrightarrow TM$ with $${\rm Ker}(\PV)=V^\perp\;\;\;\;,
\;\;\;\;\PV(X)=X\;\;\;\;\forall X\in\Gamma(V)\;,$$ which satisfies ((X),Y)=(X,(Y))\[omskewsymmeq\]for all $X,Y\in\Gamma(TM)$. It will be referred to as the $\omega$-skew orthogonal projection tensor associated to $V$. Let $Y_1,\dots,Y_{2k}$ denote a local spanning family of vector fields for $V$. Then, $V$ being symplectic is equivalent to the matrix $[C_{ij}]:=[\omega(Y_i,Y_j)]$ being invertible.
\[Pconstr\] Let $[C^{kl}]$ denote the inverse of $[C_{ij}]$, and let $\theta_j :=i_{Y_j}\omega \in\Lambda^1(TM)$. Then, locally, $\PV=C^{ij}Y_i\otimes\theta_j$.
The fact that $C^{ij}Y_i\otimes\theta_j$ is a projector, and that ( \[omskewsymmeq\]) holds, follows from $\theta_i(Y_j)=C_{ij}$, and $C_{ij}C^{jk}=\delta_i^l$. Its rank is clearly $2k$, and it is straightforward to see that its kernel is given by $\Gamma(V^\perp)$.
**Non-Integrable Constraints**
------------------------------
The quadruple $(M,\omega,H,V)$ naturally defines a dynamical system whose orbits are all $V$-horizontal. Its flow is simply the 1-parameter group of diffeomorphisms $\tPhi_t$ generated by $X_H^V:=\PV( X_H)\in\Gamma(V)$, with \_t \_t(x) = X\_H\^V (\_t(x) ) \[eqsofmo\] for every $x\in M$. In a local Darboux chart, where $\omega$ is represented by \[sympJ\]= (
[cc]{}0&\_n\
-\_n&0
),and where $x(t)$ stands for the vector of coordinate components of $\tPhi_t(\xa)$, ( \[eqsofmo\]) is given by \_t x(t) = (P\_x\_x H)(x(t)) = ( P\_x\^\_x H)(x(t)). \[eqsmocoord\]$P$ denotes the matrix of $\PV$, and $P^\dagger$ is its transpose.
If the condition of integrability imposed on $V$ is dropped, this dynamical system will allow for the description of non-holonomic mechanics. If $V$ is integrable, $M$ is foliated into $2k$-dimensional symplectic integral manifolds of $V$. On every leaf $j:M'\rightarrow M$, the induced dynamical system corresponds to the pullback Hamiltonian system $(M,j^*\omega,H\circ j)$. In this sense, ( \[eqsofmo\]) generalizes the Dirac constraints.
A new class of dynamical systems is obtained by discarding the requirement of integrability on $V$. Let $[\cdot,\cdot]$ denote the Lie bracket. We recall that the distribution $V$ is non-integrable if there exists a filtration \[filt\]V\_0V\_1V\_2V\_r, inductively defined by $V_0=V$, and $V_i=V_{i-1}+[V_0,V_{i-1}]$, where $V_1\neq V_0$. The sequence $\lbrace V_i\rbrace_1^r$ is called the [*flag*]{} of $V$. If the fibre ranks of all $V_i$ are base point independent, $V$ is called equiregular. The smallest number $\rV $ at which the flag of $V$ stabilizes, that is, for which $V_s=V_{\rV }$, $\forall s\geq \rV $, is called the degree of non-holonomy of $V$. If $V_{\rV }=TM$, one says that $V$ satisfies Chow’s condition, or that it is totally non-holonomic.
\[Frob\](Frobenius condition) $V$ is integrable if and only if locally, \^k\_[ij]{} := ()\^r\_i ()\^s\_j (\_r()\^k\_s- \_s()\^k\_r) = 0 everywhere on $M$.
$V$ is integrable if and only if $\PVc\left([\PV (X),\PV (Y)]\right)=0$ for all sections $X,Y$ of $TM$, which is equivalent to $V_1=V$. The asserted formula is the local coordinate representation of this condition.
An Auxiliary Almost Kähler Structure
-------------------------------------
For the analysis in subsequent sections, it will be useful to introduce an almost Kähler structure on $M$ that is adapted to $V$. To this end, let us briefly recall some basic definitions. Let $g$ denote a Riemannian metric on $M$. An almost complex structure $J$ is a smooth bundle isomorphism $J:TM\rightarrow TM$ with $J^2=-\1$. Together with $g$, it defines a two form satisfying \[sksadness\]\_[g,J]{}(X,Y)=g(JX,Y) for all sections $X$, $Y\in\Gamma(TM)$. $g$ is hermitean if $g(JX,JY)=g(X,Y)$, and Kähler if $\omega_{g,J}$ is closed. The triple $(g,J,\omega_{g,J})$ is called compatible. Every symplectic manifold admits an almost complex structure $J$, and a Kähler metric $g$, such that $(g,J,\omega)$ is compatible.
\[comptripprp\] For any symplectic manifold $(M,\omega)$, together with a symplectic distribution $V$, there exists a compatible triple $(g,J,\omega)$, such that $\PV$ is symmetric with respect to $g$, and $\PV JX=J\PV X$ for all $X\in\Gamma(TM)$.
We pick a smooth Riemannian metric $\tilde{g}$ on $M$, relative to which $\PV$ is symmetric, for instance by choosing an arbitrary Riemannian metric $g'$ on $M$, and defining $\tilde{g}(X,Y):= g'(\PV X,\PV Y)+g'(\bar{\pi}_V X,\bar{\pi}_V Y)$, where $\bar{\pi}_V=\1-\PV$. We consider the non-degenerate, smooth bundle map $K$ defined by $\omega(X,Y)=\tilde{g}(K X,Y)$, which is skew symmetric with respect to $\tilde{g}$, that is, its $\tilde{g}$-adjoint is $K^*=-K$. $K^* K=-K^2$ is smooth, positive definite, non-degenerate and $\tilde g$-symmetric, hence there is a unique smooth, positive definite, $\tilde g$-symmetric bundle map $A$ defined by $A^2=-K^2$, which commutes with $K$. Consequently, the bundle map $J=KA^{-1}$ satisfies $J^2=-\1$, and defines an almost complex structure. Since $A$ is positive definite and $\tilde g$-symmetric, $g(X,Y):=\tilde{g}(AX,Y)$ is a Riemannian metric with $\omega(X,Y)=g(JX,Y)$. Moreover, this metric is hermitean, since $g(JX,JY) = \tilde{g}(KX, A^{-1}K Y)
= -\tilde{g}(X,K^2 A^{-1}Y)
= \tilde{g}(X,AY)
= g(X,Y)$. In fact, since $\omega$ is closed, $g$ is Kähler.
To show that $\PV$ is $g$-symmetric, we note that $\PV$ commutes with $K$, since $\tilde{g}(K\PV X,Y)=\omega(\PV X,Y)=\omega(X,\PV Y)=\tilde{g}(KX,\PV Y)=
\tilde{g}(\PV KX,Y)$ for all $X,Y\in\Gamma(TM)$, using that $\PV$ is symmetric with respect to $\tilde{g}$. Hence, $\PV$ commutes with $A^2=-K^2$, and it straightforwardly follows from the $\tilde g$-orthogonality of $A$, $\PV$, and the positivity of $A$ that $\PV$ and $A$ commute. Hence, $\PV$ is $g$-symmetric, and it is also clear that $\PV$ commutes with $J=KA^{-1}$. Thus, $J$ in particular restricts to a bundle map $J:V\rightarrow V$.
Further Properties
------------------
Some key properties of Hamiltonian systems concerning symmetries, Poisson brackets, energy conservation, and, to some degree, symplecticness, can be generalized to the constrained system.
### Symmetries
Let us assume that the Hamiltonian system $(M,\omega,H)$ admits a symplectic $G$-action ($G$ some Lie group) $\Psi:G\rightarrow {\rm Diff}(M)$, such that $\Psi_h^*\omega=\omega$ and $H\circ\Psi_h=H$ for all $h\in G$. Then, we will say that the constrained system $(M,\omega,H,V)$ admits a $G$-symmetry if $\Psi_{h*} V=V$ holds for all $h\in G$.
### Generalized Dirac bracket
The smooth, $\R$-bilinear, antisymmetric pairing on $C^\infty(M)$ associated to $(M,\omega,V)$ given by \[Vbra\]f,g\_V:=((X\_f),(X\_g))is a straightforward generalization of the Dirac and Poisson brackets. Along orbits of $\tPhi_t$, one has $$\partial_t f(\tPhi_t(x))\;=\;\lbrace H,f\rbrace_V(\tPhi_t(x))\;$$ for all $x\in M$, in analogy to ( \[Poisseqsmo\]). However, the bracket ( \[Vbra\]) does not satisfy the Jacobi identity if $V$ is non-integrable, but it satisfies a Jacobi identity on every (symplectic) integral manifold if $V$ is integrable.
### Energy conservation
This key conservation law also holds for the constrained system.
The energy $H$ is an integral of motion of the dynamical system ( \[eqsofmo\]).
This follows from the antisymmetry of the generalized Dirac bracket, which implies that $\partial_t H=\lbrace H,H\rbrace_V=0$.
### Symplecticness
The flow $\tPhi_t$ is not symplectic, but the following holds. Let us consider \_t\_t\^[\*]{}&=&\_t\^\* \_[X\_H\^V]{} = \_t\^\*(d i\_[X\_H\^V]{}+ i\_[X\_H\^V]{}d)\
&= & -\_t\^\*d(()\^i\_k \_i H dx\^k) = - \_t\^\*(\_l()\^i\_k\_i H dx\^ldx\^k)\
&=& - \_t\^\* ( ( \_l()\^i\_k - \_k ()\^i\_l ) \_i H dx\^l dx\^k ).Hence, the restriction of $\partial_t\tPhi_t^*\omega$ to $X,Y\in\Gamma(V)$ is given by \_t\_t\^\*(X,Y)= - \_t\^\* ( \^i\_[rs]{} \_i HX\^r Y\^s ) ,where $\cF^i_{rs}$ is defined in lemma [ \[Frob\]]{}. Thus, the right hand side vanishes identically if and only if $V$ is integrable. In the latter case, the restriction of $\tPhi_t^*\omega$ to $V\times V$ equals its value for $t=0$, given by $\omega(\PV(\cdot),\PV(\cdot))$. On every integral manifold $j:M'\rightarrow M$ of $V$, $\tPhi_t$ is symplectic with respect to the pullback symplectic structure $j^*\omega$.
THE GEOMETRY AND TOPOLOGY OF THE CRITICAL MANIFOLD {#sectionII}
==================================================
In this main section, we address geometrical and topological properties of the critical set $\crit$ of the constrained Hamiltonian system $(M,\omega,H,V)$. The main result of the subsequent analysis is, for $M$ compact and without boundary, the topological formula ( \[Conley-Zehnder0\]) that interrelates the Poincaré polynomials of $M$ and $\cg$ in a manner closely akin to the Morse-Bott inequalities. This result implies that the topology of $M$ necessitates the existence of certain connectivity components of $\cg$ of a prescribed index. The analysis is structured as follows.
In section [ \[genpropsubsec\]]{}, we prove that $\crit$ is, in the sense of Sard’s theorem, generically a smooth $2(n-k)$-dimensional submanifold $\cg\subset M$. For the special case in which $V$ is integrable, it is shown that the intersection of any integral manifold of $V$ with $\cg$ is a discrete set, in agreement with the usual understanding that critical points in Hamiltonian systems - on every leaf of the foliation in the integrable case - are typically isolated.
In section [ \[topcritmfsubsec\]]{}, we introduce the main tool for the analysis of $\crit$, an auxiliary gradient-like flow $\phi_t\in{\rm Diff}(M)$ generated by the vector field $\PV\grad H$, where $g$ is the Kähler metric of the compatible quadruple introduced after proposition [ \[comptripprp\]]{}. From this point on, we assume that $H:M\rightarrow\R$ is a Morse function. Let $j:\cg\hookrightarrow M$ denote the embedding. We show that $\cg$ is normal hyperbolic with respect to $\phi_t$, and that the critical points of $j^*H$ on $\cg$ are precisely those of $H$ on $M$. The latter is quintessential for our discussion of the topology of $\cg$ via comparison of the Morse-Witten complexes of $(\cg,j^*H)$ and of $(M,H)$ in section [ \[mowiproofsubsec\]]{}.
In section [ \[cozeproofsubsec\]]{}, we prove ( \[Conley-Zehnder0\]) by an application of Conley-Zehnder theory, [@CoZe], to the auxiliary gradient-like system. The assumptions on $\crit$ are slightly less strict than genericity. In particular, assuming that $\crit\setminus \cg$ is a disjoint union of $C^1$ manifolds, we show that every connectivity component of $\crit\setminus \cg$ is contained in a $H$-level surface, and that $\crit\setminus \cg$ can be deformed away by an infinitesimal perturbation of the vector field.
In section [ \[mowiproofsubsec\]]{}, we assume that $\crit=\cg$, and give a second proof based on the comparison of the Morse-Witten complexes of $(M,H)$ and $(\cg,H|_{\cg})$. Our construction only uses the theory for non-degenerate Morse functions, not for Morse-Bott functions. The interest in this discussion is to elucidate the relationship between critical points of the ’free’ system $(M,H)$, and of the critical manifold $\cg$ of the constrained system $(M,H,V)$. The special case of mechanical systems (where $M$ is noncompact) will be analyzed in a later section.
Generic Properties of the Critical Set {#genpropsubsec}
--------------------------------------
Let us to begin with recall some basic definitions. Critical points of $H$ are given by zeros of $dH$, and a corresponding value of $H$ is called a critical value. A critical level surface $\Sigma_{E}$ corresponds to a critical value $E$ of $H$, whereas a regular level surface $\Sigma_E$ contains no critical points of $H$ (the corresponding value of $E$ is then called regular). The critical set of the constrained Hamiltonian system $(M,\omega,H,V)$ is given by $$\crit\;=\;\left\lbrace x\in M\;|\;X_H^V(x)\;=\;0\right\rbrace\;
\;\subset\;\;M\;.$$ The following theorem holds independently of the fact whether $V$ is integrable or not.
\[Sardthm\] In the generic case, the critical set is a piecewise smooth, $2(n-k)$-dimensional submanifold of $M$.
Let $\lbrace Y_i\rbrace_{i=1}^{2k}$ denote a smooth, local family of spanning vector fields for $V$ over an open neighborhood $U\subset M$. Since $V$ is symplectic, the fact that $X_H^V$ is a section of $V$ implies that $\omega(Y_i,X_H^V)$ cannot be identically zero for all $i$ and everywhere in $U$. Due to the $\omega$-skew orthogonality of $\PV$, and $\PV Y_i=Y_i$, $$\omega(Y_i,X_H^V)\;=\;\omega(\PV (Y_i),X_H)\;=\;\omega(Y_i,X_H)
\;=\;Y_i(H)\;.$$ Thus, with $\underline{F}:=(Y_1(H),\dots,Y_{2k}(H))\in
C^\infty(U, \R^{2k})$, it is clear that $\crit\cap U=\underline{F}^{-1}(\underline{0})$. Since $\underline{F}$ is smooth, Sard’s theorem implies that regular values, having smooth, $2(n-k)$-dimensional submanifolds of $U$ as preimages, are dense in $\underline{F}(U)$ [@Mi2].
For future technical convenience, we pick a local spanning family $\left\lbrace Y_i\in\Gamma(V)\right\rbrace_{i=1}^{2k}$ for $V$ that satisfies $$\omega(Y_i,Y_j)\;=\;\tJ_{ij}\;,$$ with $\tJ:=\left(\begin{array}{cc}0&\1_k\\-\1_k&0\end{array}\right)$. This choice is always possible.
Furthermore, introducing the associated family of 1-forms $\lbrace\theta_i\rbrace$ by $\theta_i(\cdot):=\omega(Y_i,\cdot)$, theorem [ \[Pconstr\]]{} implies that $$\PV\;=\;\tJ^{ij}Y_i\otimes \theta_j\;,$$ where $\tJ^{ij}$ are the components of $\tJ^{-1}=-\tJ$. Expanding $X_H^V$ with respect to the basis $\lbrace Y_i\rbrace$ gives \[XHVexp\] X\_H\^V=(X\_H\^V)=-Y\_i(H)\^[ij]{} Y\_j,where one uses the relationship $\theta_j(X_H^V)=Y_j(H)$ obtained in the proof of theorem [ \[Sardthm\]]{}. Then, the following proposition holds, which is in the subsequent discussion interpreted as the property of normal hyperbolicity of $\crit$ with respect to a certain gradient-like flow if the genericity assumption is satisfied.
\[invert\] Under the genericity assumption of theorem [ \[Sardthm\]]{}, the $2k\times 2k$-matrix given by $[Y_k(Y_i(H))(\xa)]$ is invertible for all $\xa\in\crit$, and every local spanning family $\lbrace Y_i\in\Gamma(V)\rbrace$ of $V$.
Let us pick a local basis $\lbrace Y_i\rbrace_1^{2k}$ for $V$, and $\lbrace Z_j\rbrace_1^{2(n-k)}$ for $V^\perp$, which together span $TM$. Let $\xa\in\cg$, and assume the generic situation of theorem [ \[Sardthm\]]{}. Because $\cg$ is defined as the set of zeros of the vector field ( \[XHVexp\]), the kernel of the linear map $$\left. dF_i(\cdot)\tJ^{ik}Y_k\right|_{\xa}\;:\;
T_{\xa}M\;\rightarrow\;V_{\xa}\;,$$ where $F_i:= Y_i(H)$, is precisely $T_{\xa}\cg$, and has a dimension $2(n-k)$.
In the basis $\lbrace Y_1|_{\xa},\dots,Y_{2k}|_{\xa},Z_1|_{\xa},\dots,
Z_{2(n-k)}|_{\xa}\rbrace$, its matrix is given by $$A\;=\;\left[ A_V\;A_{V^\perp}\right]\;,$$ where $A_V:=[Y_i(F_j)\tJ^{jk}Y_k|_{\xa}]$, and $A_{V^\perp}:=[Z_i(F_j)\tJ^{jk}Y_k|_{\xa}]$. Bringing $A$ into upper triangular form, $A_V$ is likewise transformed into upper triangular form. Because the rank of $A$ is $2k$, and $A_V$ is a $2k\times 2k$-matrix, its upper triangular form has $2k$ nonzero diagonal elements. Consequently, $A_V$ is invertible, and due to the invertibility of $\tJ$, one arrives at the assertion.
\[SpecQaminPalm\] Let $\{Y_i\}_{i=1}^{2k}$ denote a local spanning family for $V$, and let $\{X_i\}_{i=1}^{2k}$ be a local spanning family (of $C^\infty$ sections) of $N\cg$, interpreted as a vector bundle over $\cg$ that is embedded in $\cup_{x\in\cg}T_xM$. Then, the matrix $[g(Y_i,X_j)(x)]_{i,j=1}^{2k}$ is invertible for every $x\in\cg$.
\[intcasecor\] Let $\cg$ satisfy the genericity assumption of theorem [ \[Sardthm\]]{}. If $V$ is integrable, the intersection of any integral manifold of $V$ with $\cg$ is a discrete set.
The previous proposition implies that generically, integral manifolds of $V$ intersect $\cg$ transversely. Their dimensions are mutually complementary, hence the intersection set is zero-dimensional.
Normal Hyperbolicity and an Auxiliary Gradient-Like System {#topcritmfsubsec}
----------------------------------------------------------
In this section, we introduce the main tool necessary for the analysis of the topology of $\crit$, given by an auxiliary gradient-like flow on $M$ whose same critical set is also $\crit$. Furthermore, we show that generically, $\crit$ is normal hyperbolic with respect to the latter.
### A Generalized Hessian
To begin with, we define a generalized Hessian for $\crit$. The usual coordinate-free definition of the Hessian of $H$ is $\nabla dH$, evaluated at the critical points of $H$, where $\nabla$ is the Levi-Civita connection of the Kähler metric $g$. Let $\PV^\dagger$ denote the dual projection tensor associated to $\PV$, which acts on sections of the cotangent bundle $T^*M$, so that for any 1-form $\theta$, $\langle \PV^\dagger\theta, X\rangle =\langle\theta,\PV X\rangle$. The generalization of the Hessian in our context is the tensor $\nabla (\PV^\dagger dH)$, which acts as a bilinear form on $\Gamma(TM)\times\Gamma(TM)$ by way of (\^dH)(X,Y)&:=&\_X (\^dH), Y\
&=&((()\^j\_r H\_[,j]{})\_[,s]{} -\^s\_[ri]{}()\_s\^j H\_[,j]{} ) X\^r Y\^s ,where $\Gamma^s_{ri}$ are the Christoffel symbols. Evaluating this quantity on $\crit$, the second term in the bracket on the lower line is zero. The non-vanishing term is determined by the matrix \[hessgen\]\[K\_[rs]{}\] := \[(()\^j\_r H\_[,j]{})\_[,s]{}\] .One straightforwardly verifies that $(\PV)^j_i K_{jk} = K_{ik}$ is satisfied everywhere on $\crit$, hence rank$\{K\}\leq$ rank$\{\PV\}=2k=$rank$\{V\}$. Clearly, the corank of $K|_a$ equals the dimension of the connectivity component of $\crit$ containing $a$.
### Definition of the Gradient-Like System {#auxgradlkflowsssec}
A flow is gradient-like if there exists a function $f:M\rightarrow\R$ that decreases strictly along all of its non-constant orbits. The flow $\tPhi_t$ of the constrained Hamiltonian system is not gradient-like, and hence turns out to be of limited use for the study of the global topology of $\crit$, because invariant sets of $\tPhi_t$ do in general not only contain fixed points, but also periodic orbits.
Instead, we introduce the auxiliary dynamical system \[gradlikeODE\]\_t(t) =-(\_g H )((t)), where $\gamma:I\subset\R\rightarrow M$, which turns out to be an extremely powerful tool for our purpose. Let us denote its flow by $\phi_t\in{\rm Diff}(M)$. The orbits of ( \[gradlikeODE\]) are clearly $V$-horizontal, and both $\Phi_t^c$ and $\phi_t$ exhibit the same critical set $\crit$.
The flow $\phi_t$ is gradient-like.
Since \_t H((t))&=&dH((t)) , \_t (t) = - g(H , H)((t))\
&=&- g(H , H)((t)) 0 ,it follows that $H$ decreases strictly along the non-constant orbits of $\phi_t$. We have here used the fact that $(g,J,\omega,\PV)$ is a compatible quadruple.
$(g,J,\omega,\PV)$ has been constructed for this precise reason. It is immediately clear that $\phi_t$ generates no periodic trajectories, hence $\crit$ comprises all invariant sets of $\phi_t$.
### Morse Functions and Non-Degenerate Critical Manifolds
Let us next recall some standard definitions from Morse- and Morse-Bott theory that will be needed in the subsequent discussion. The dimensions of the zero and negative eigenspaces of the Hessian of $f$ at a critical point $a$ are called the nullity and the index of the critical point $a$. If all critical points of $f:M\rightarrow\R$ have a zero nullity, $f$ is called a Morse function, and the index is then called the Morse index of $a$. If the critical points of $f$ are not isolated, but elements of critical manifolds that are non-degenerate in the sense of Bott, $H$ is called a Morse-Bott function [@Bo]. Throughout this section, we will assume that $H$ is a Morse function.
Furthermore, we recall some standard definitions related to normal hyperbolicity, applied to the case of $\crit$. A connectivity component $\crit_i$ is locally normal hyperbolic at the point $a\in\crit$ with respect to $\phi_t$ if it is a manifold at $a$, and if the restriction of $\K$ to the normal space $N_a\crit$ is non-degenerate. A connectivity component $\crit_i$ is called non-degenerate if it is a manifold that is everywhere normal hyperbolic with respect to $\phi_t$. The index of a non-degenerate connectivity component $\crit_i$ is the number of eigenvalues of the constrained Hessian $\K$ on $\crit_i$ that are contained in the negative half plane.
\[normhypgenprop\] If $\crit$ is generic in the sense of Sard’s theorem, it is normal hyperbolic with respect to the gradient-like flow $\phi_t$.
This follows straightforwardly from proposition [ \[invert\]]{}.
Let $\crit_i$, $i=1,..,l$ denote the connectivity components of $\crit=\cup \crit_i$, and let $j_i:\crit_i
\hookrightarrow M$ denote the embedding of the $i$-th components.
\[critMorseH1\] Assume that $\crit$ satisfies the genericity assumption in the sense of Sard’s theorem, and that $H:M\rightarrow\R$ is a Morse function. Then, $H_i:=H\circ j_i:\crit_i\rightarrow\R$ is a Morse function, and $x\in \crit_i$ is a critical point of $H_i$ if and only if it is a critical point of $H$.
It is trivially clear that every critical point of $H$ is a critical point of $H_i$. To prove the opposite direction, suppose that $a$ is an extremum of $H|{\crit_i}$. Then, $\grad H|_a\in N_a\crit$, but also, by definition of $\crit$, $P_a\grad H|_a = 0$. By corollary [ \[SpecQaminPalm\]]{}, this can only be true if $\grad H_a=0$. The Hessian of the restriction of $H$ at any critical point of $H_i$ is nondegenerate, thus $H_i$ is a Morse function on $\crit_i$.
\[critMorseH2\] The critical points of $H|_{\cg}:\cg\rightarrow\R$ are precisely the critical points of $H:M\rightarrow\R$. If $\crit_i$ is a non-generic connectivity component that is a normal hyperbolic submanifold of $M$, $\crit_i\subset\Sigma_{H(\crit_i)}$.
The first assertion follows trivially from the previous proposition. Assuming that $\crit_i$ is a non-generic connectivity component of $\crit$ that is a manifold and normal hyperbolic, the previous proposition implies that there are no extrema of $H|_{\crit_i}$. Thus, $\crit_i$ is a submanifold of the level surface $\Sigma_{H(\crit_i)}$.
Approach via Conley-Zehnder Theory {#cozeproofsubsec}
----------------------------------
The goal in this and the next section is to derive the relationship ( \[Conley-Zehnder0\]) between the Poincaré polynomials of $\crit$, and $M$. We first approach this problem by use of Conley-Zehnder theory, [@CoZe], and under slightly less restrictive assumptions than genericity.
Let us for convenience first recall some of the key elements from Conley-Zehnder theory, [@CoZe]. Let $\crit_i$ be any compact component of $\crit$. An index pair associated to $\crit_i$ is a pair of compact sets $(N_i,\tilde{N}_i)$ that possesses the following properties. The interior of $N_i$ contains $\crit_i$, and moreover, $\crit_i$ is the maximal invariant set under $\phi_t$ in the interior of $N_i$. $\tilde{N}_i$ is a compact subset of $N_i$ that has empty intersection with $\crit_i$, and the trajectories of all points in $N_i$ that leave $N_i$ at some time under the gradient-like flow $\phi_t$ intersect $\tilde{N}_i$. $\tilde{N}_i$ is called the exit set of $N_i$.
The homotopy type of the pointed space $N_i/ \tilde{N}_i$ only depends on $\crit_i$, by a result proven in [@CoZe], so that the relative cohomology $H^*(N_i,\tilde{N}_i)$ (with coefficients appropriately chosen) is independent of the particular choice of index pairs (the space $N_i/ \tilde{N}_i$ is obtained from collapsing the subspace $\tilde{N}_i$ of $N_i$ to a point). The equivalence class $[N_i/ \tilde{N}_i]$ of pointed topological spaces under homotopy is called the [*Conley index*]{} of $\crit_i$.
Let $I$ denote a compact invariant invariant set under $\phi_t$. A Morse decomposition of $I$ is a finite, disjoint family of compact, invariant subsets $\lbrace M_1,\dots,M_n\rbrace$ that satisfies the following requirement on the ordering. For every $x\in I\setminus \cup_i
M_i$, there exists a pair of indices $i<j$, such that $\lim_{t\rightarrow-\infty}\phi_t(x)\subset M_i$, and $\lim_{t\rightarrow \infty}\phi_t(x)\subset M_j$. Such an ordering, if it exists, is called admissible, and the $M_i$ are called Morse sets of $I$.
For every compact invariant set $I$ admitting an admissibly ordered Morse decomposition, there exists an increasing sequence of compact sets $N_i$ with $N_0\subset
N_1\subset\dots\subset N_m$, such that $(N_i,N_{i-1})$ is an index pair for $M_i$, and $(N_m,N_0)$ is an index pair for $I$, [@CoZe].
Consider compact manifolds $A\supset B \supset C$. It is a standard fact that the exact sequence of relative cohomologies $$\dots\stackrel{\delta^{k-1}}{\rightarrow} H^k(A,B)\rightarrow
H^k(A,C)\rightarrow H^k(B,C) \stackrel{\delta^k}{\rightarrow}
H^{k+1}(A,B)\rightarrow\dots$$ implies that, with $r_{i,p}$ denoting the rank of $H^p(N_i,N_{i-1})$, $$\sum_{i,p}\lambda^p r_{i,p}=\sum_p b_p \lambda^p + (1+\lambda)\Q(\lambda)$$ for the indicated Poincaré polynomials (cf. for instance [@Jo]). $b_j$ is the $j$-th Betti number of the index pair $(N_m,N_0)$ of $I$, and $\Q(\lambda)$ is a polynomial in $\lambda$ with non-negative integer coefficients. Due to the positivity of the coefficients of $\Q(\lambda)$, it is clear that $\sum_i r_{i,p}\geq b_p$.
If $M$ is compact and closed, and if $\crit$ is non-degenerate, the following holds. The invariant set $I$ can be chosen to be equal to $M$. We let $N_m=M$ and $N_0=\emptyset$ denote the top and bottom elements of the sequence defined above, and order the connected elements of $\crit$ according to the descending values of the maximum of $H$ attained on each $\crit_i$. Then, $\crit$ furnishes a Morse decomposition for $M$. The homology groups of $M$ are isomorphic to the relative homology groups of the index pair $(N_m,N_0)$. Hence the numbers $b_p$ are the Betti numbers of $M$.
Let $\crit_i\subset\cg$ be a generic connectivity component, compact and without boundary, and let $(N_i,\tilde{N}_i)$ denote any associated index pair. Then, H\^[q+\_i]{}(N\_i,\_i) H\^q(\_i),where $\mu_i$ is the index of $\crit_i$, and $q=0,\dots,dim(\crit_i)$.
We consider, for $\epsilon_0>0$ small, a compact tubular $\epsilon_0$-neighborhood $U$ of $\crit_i$ (of dimension $2n$), and let $$W_U^{cu}(\crit_i)\;:=\;(W^-(\crit_i)\cup \crit_i)\cap U$$ denote the intersection of the center unstable manifold of $\crit_i$ with $U$. $W^-(\crit_i)$ denotes the unstable manifold of $\crit_i$. Pick some small, positive $\epsilon <\epsilon_0$, and let $U_\epsilon$ be the compact tubular $\epsilon$-neighborhood of $W_U^{cu}(\crit_i)$ in $U$.
Letting $\epsilon$ continuously go to zero, we obtain a homotopy equivalence of tubular neighborhoods, for which $W_U^{cu}(\crit_i)$ is a deformation retract. Let $$U_\epsilon^{out}\;:=\;\partial U_\epsilon \cap \phi_\R(U_\epsilon)$$ denote the intersection of $\partial U_\epsilon$ with all orbits of the gradient-like flow that contain points in $U_\epsilon$. Then, $(U_\epsilon,U_\epsilon^{out})$ is an index pair for $\crit_i$, and by letting $\epsilon$ continuously go to zero, $U_\epsilon^{out}$ is homotopically retracted to $\partial W_U^{cu}(\crit_i)$.
Thus, by homotopy invariance, H\^\*(U\_,U\^[out]{}\_) H\^\*(W\_U\^[cu]{}(\_i),W\_U\^[cu]{}(\_i)).Since $\crit_i$ is normal hyperbolic with respect to the gradient-like flow, $W_U^{cu}(\crit_i)$ has a constant dimension $n_i+\mu(\crit_i)$ everywhere, where $n_i={\rm dim}\crit_i$. Therefore, by Lefschetz duality, [@DuFoNo], $$H^{n_i+\mu_i-p}(W_U^{cu}(\crit_i),\partial W_U^{cu}(\crit_i))\cong
H_p(W_U^{cu}(\crit_i)\setminus\partial W_U^{cu}(\crit_i)) ,$$ where $\mu_i=\mu(\crit_i)$, the index of $\crit_i$. Since $\crit_i$ is a deformation retract of the interior of $W_U^{cu}(\crit_i)$, the respective cohomology groups are isomorphic.
Due to $dim(\crit_i)=n_i$, we have by Poincaré duality $$H_p(W_U^{cu}(\crit_i)\setminus\partial W_U^{cu}(\crit_i))
\cong H_p(\crit_i)\cong H^{n_i-p}(\crit_i),$$ so that with $q:=n_i-p$, \[holy\]H\^[q+\_i]{}(U\_,U\^[out]{}\_) H\^q(\_i) ,which proves the claim.
From ( \[holy\]), we deduce that $r_{i,p} = {\rm dim} H^{i,p-\mu_i}(\crit_i) $ (recalling that $\mu_i$ is the index of $\crit_i$), hence $r_{i,p}=b_{p-\mu_i}(\crit_i)$. Assuming that the number of connected components of $\crit$ is finite, one thus obtains \[Conley-Zehnder0\]\_[i,p]{} \^[p+\_i]{}b\_[p]{}(\_i)= \_p \^p b\_p(M) + (1+) () ,which in particular implies $\sum_{i} b_{p-\mu_i}(\crit_i) \geq b_{p}(M)$. Setting $\lambda=-1$, $$\sum_{p} (-1)^{p+\mu_i} b_{p}(\crit_i) = \sum_{i} (-1)^{\mu_i} \chi(\crit_i)
= \chi(M) ,$$ where $\chi$ denotes the Euler characteristic.\
[**Remark.**]{} In the case of mechanical systems, the phase space of the relevant constrained Hamiltonian system is non-compact, and the critical manifold is generally unbounded. Therefore, the arguments used here do not apply. However, since in that case, $M$ and $\crit$ are vector bundles, we are nevertheless able to prove results that are fully analogous to ( \[Conley-Zehnder0\]).\
We can prove a slightly more general result by relaxing the assumption of genericity.
\[Conley-Zehnderineqprp1nongeneric\] Assume that $\crit\setminus\cg$ is a disjoint union of $C^1$-manifolds. Then, \[Conley-Zehnder1\]\_ b\_[i,p]{}\^[p+\_i]{}=\_p b\_p \^p + (1+)() . $b_{i,p}$ are the $p$-th Betti numbers of the connectivity components $\crit_i$ of $\cg$, $b_p$ are the Betti numbers of $M$, and $\tilde{\Q}$ is a polynomial with non-negative integer coefficients.
We show that $X_H^V$ can be infinitesimally perturbed such that $\crit\setminus\cg$ is removed. Consider, for $\epsilon>0$, the compact tubular neighborhoods \[neighb\]U\_(\_i)=xM dist\_g(x,\_i)of connectivity components $\crit_i\subset \crit\setminus\cg$, where $dist_g$ denotes the Riemannian distance function induced by $g$. We introduce a vector field $X_\epsilon$, given by $\PV\grad H$ in $M\setminus U_\epsilon(\crit_i)$, and in the interior of every $U_\epsilon(\crit_i)$ with $\crit_i\subset\crit\setminus\cg$, by \[kiwi\]X\_|\_x=H|\_x +h(x) H|\_x.Here, $h\in C^1(U_\epsilon(\crit_i),[0,1])$, obeying $h|_{\crit_i}=1$ and $h|_{\partial U_\epsilon(\crit_i)}=0$ is strictly monotonic along the the flow lines generated by $\PV\grad H$. $h$ exists because $\crit\setminus\cg$ is a disjoint union of $C^1$-manifolds.
For all $\crit_i\subset\crit\setminus\cg$, $\grad H$ is strictly non-zero in $U_\epsilon(\crit_i)$, as shown above. We have $$g(X_\epsilon,\grad H)=(\|\PV\grad H\|_g^2)(x)+\epsilon h(x) (\|\grad
H\|_g^2)(x) ,$$ where we have used the $g$-symmetry of $\PV$, and $\|X\|_g^2\equiv g(X,X)$. The first term on the right hand side is non-zero on the boundary of $U_\epsilon(\crit_i)$, while the second term vanishes. Moreover, the second term is non-zero everywhere in the interior of $U_\epsilon(\crit_i)$. Therefore, $X_\epsilon$ vanishes nowhere in $U_\epsilon(\crit_i)$. Hence, $X_\epsilon$ is a deformation of $\PV\grad H$, with critical set $\cg$. Notably, $\cg$ cannot be removed in this manner, since it contains critical points of $H$.
Since $g(X_\epsilon,\grad H)$ is strictly positive in every $U_\epsilon(\crit_i)$, $X_\epsilon$ generates a gradient-like flow. Since $\|X_\epsilon-\PV\grad H\|_g\leq O(\epsilon)$ everywhere on $M$, one can pick $X_\epsilon$ arbitrarily close to $\PV\grad H$ in the $L^\infty$-norm on $\Gamma(TM)$ induced by $\|\cdot\|_g$. Carrying out the Conley-Zehnder construction with respect to the flow generated by $X_\epsilon$ yields ( \[Conley-Zehnder1\]). This result does not require the assumption of normal hyperbolicity on $\crit$.
Approach via the Morse-Witten Complex {#mowiproofsubsec}
-------------------------------------
We will next provide a different derivation of ( \[Conley-Zehnder1\]), based on the construction of the Morse-Witten differential complex. The motivation is to clarify the orbit structure of the gradient-like system, and to devise an explicit construction that relates the Morse-Witten complexes of the free and constrained system to one another. This in particular only involves the corresponding theory for [*non*]{}-degenerate Morse functions.
Let us to begin with briefly recall the basic framework of this construction. Let $M$ be a compact, closed, orientable and smooth $n$-manifold, and let $f:M\rightarrow\R$ be a Morse function. Let $\Cb^p$ denote the free $\Z$-module generated by the critical points of $f$ with a Morse index $p$. The set $\Cb=\oplus_p \Cb^p$ is the free $\Z$-module generated by the critical points of $f$, and graded by their Morse indices. There exists a natural coboundary operator $\delta:\Cb^p\rightarrow\Cb^{p+1}$, with $\delta\circ\delta=0$, whose construction we recall next, cf. [@AuBr; @Bo1; @Fl; @Wi].
Introducing an auxiliary Riemannian structure on $M$, we let $W_a^-$ and $W_a^+$ denote the unstable and stable manifold of the critical point $a$ of $f$ under the gradient flow, respectively, and assign an arbitrary orientation to every $W_a^-$. The orientation of $M$, together with the orientation of $W_a^-$ at every critical point $a$ induces an orientation of $W_a^+$. Morse functions, for which all $W_a^-$ and $W_{a'}^+$ intersect transversely, are dense in $C^\infty(M)$. The dimension of $W_a^-$ equals the Morse index $\mu(a)$ of $a$, and the dimension of the intersection $M(a,a'):= W_a^-\cap W_{a'}^+$ is given by ${\rm max}\lbrace\mu(a)-\mu(a'),0\rbrace$. For pairs of critical points $a$ and $a'$ with relative Morse index $1$, say $\mu(a)=\mu(a')+1$, $M(a,a')$ is a finite collection of gradient lines that connect $a$ with $a'$.
The intersection of $M(a,a')$ with any regular level surface $\Sigma_c$ of $f$ with $f(a)<f(\Sigma_c)=c<f(a')$ is transverse, and consists of a finite collection of isolated points. Then, one picks the orientation of $\Sigma_c$, which, combined with the section $\grad f$ of its normal bundle, shall agree with the orientation of $M$. The submanifolds $W_{a,c}^-:= W_a^-\cap\Sigma_c$ and $W_{a',c}^+:= W_{a'}^+\cap\Sigma_c$ of $\Sigma_c$ are smooth, compact and closed, with complementary dimensions in $\Sigma_c$, and orientations picked above. Hence, their intersection number, which is often in this context written as $\langle a,\delta a'\rangle := \sharp(W_{a,c}^-,W_{a',c}^+)$, is well-defined, [@Hi]. The coboundary operator of the Morse-Witten complex is defined as the $\Z$-linear map $\delta:\Cb^p\rightarrow\Cb^{p+1}$, defined by $$\delta a' = \sum_{\mu(a)=p+1} \langle a,\delta a'\rangle a .$$
\[CohWitdRhthm\] The cohomology of the differential complex $(\Cb,\delta)$ is isomorphic to the de Rham cohomology of $M$, ${\rm ker}\delta / {\rm im}\delta \cong H^*(M,\Z)$.
The proof can for instance be found in [@AuBr; @Fl; @Sw; @Wi]. If $\langle a,\delta a'\rangle\neq0$ for a pair $a$ and $a'$ of critical points with a relative Morse index $1$, we will say that they are effectively connected (by gradient lines).
As is well-known, the existence of the Morse-Witten complex implies the strong Morse inequalities in the following manner. Let $\Zb^p:={\rm ker}\delta\cap\Cb^p$ denote the $p$-th cocyle group, $\B^p\subset\Cb^p$ the $p$-th coboundary group, and $\H^p:=\Zb^p\setminus\B^p$ the $p$-th cohomology group under $\delta$. Thus, ${\rm dim}\H^{p}=b_p(M)$ by theorem [ \[CohWitdRhthm\]]{}. From $${\rm dim}\Cb^p= b_p(M)+ {\rm dim}\B^p + {\rm dim}\B^{p+1} ,$$ where ${\rm dim}\Cb^p=N_p$ (the number of critical points of $f$ with Morse index $p$), follows \[oregon\] \_[p=0]{}\^n \^p N\_p =\_[p=0]{}\^n \^p b\_p(M)+ (1+)\_[p=1]{}\^n \^[p-1]{}[dim]{}\^p (both $\B^0$ and $\B^{2n+1}$ are empty). The coefficients of the polynomial ${\mathcal Q}(\lambda)=\sum \lambda^{p-1}{\rm dim}\B^p$ are evidently non-negative and integer. Clearly, ${\rm dim}\B^p$ is the number of critical points of Morse index $p$ that are effectively connected to critical points of Morse index $p-1$ via gradient lines of $f$.
### **Comparing the Complexes for the Free and Constrained System**
The goal of our discussion here is to devise an explicit construction that relates the Morse-Witten complex of the free system $(M,H)$ to the one on the critical manifold $(\cg,H|_{\cg})$, by a deformation of the gradient-like flow $\phi_t$. This will yield ( \[Conley-Zehnder1\]).
Let $\crit_i$ denote the $i$-th connectivity component of $\cg$, and $\A_i:=\lbrace a_{i,1},\dots,a_{i,m}\rbrace$ the critical points of $H$ contained in $\crit_i$. Furthermore, let $\mu(a_{i,r})$ be the associated Morse indices of $H:M\rightarrow\R$, and $H_i:= H|_{\crit_i}$ denote the restriction of $H$ to $\crit_i$. By proposition [ \[critMorseH2\]]{} and corollary [ \[critMorseH2\]]{}, $H_i:\crit_i\rightarrow \R$ is a Morse function, whose critical points are precisely the elements of $\A_i$. The index $\mu(\crit_i)$ of $\crit_i$ equals the number of negative eigenvalues of the Hessian of $H$ at any $a_{i,r}\in\A_i$ whose eigenspace is normal to $\crit_i$.
The Morse index of $a_{i,r}$ with respect to $H_i$ is thus $\mu(a_{i,r})-\mu(\crit_i)$. To define the Morse-Witten complex associated to $\crit_i$, we introduce the free $\Z$-module generated by the elements of $\A_i$, graded by the Morse indices $p$ of the critical points of $H_i$, $$\Cb_i=\oplus_p \Cb^p_i.$$ To construct the coboundary operator $\delta_i:\Cb^p_i\rightarrow\Cb_i^{p+1}$, one uses the gradient flow on $\crit_i$ generated by $H_i$, thus obtaining \_i / [im]{}\_i H\^\*(\_i,) .Application of ( \[oregon\]) shows that for every $\crit_i\in\cg$, \[ontario\]\_[p]{} \^p N\_[i,p]{} &=& \_[p]{} \^p b\_p(\_i)+ (1+)\_[p]{} \^[p-1]{}[dim]{}\^p\_i ,where $\B^p_i$ is the $p$-th coboundary group of the Morse-Witten complex of $\crit_i$, and $N_{i,p}$ is the number of critical points of $H_i$ on $\crit_i$ of Morse index $p$.
Since every critical point of $H$ lies on precisely one generic component $\crit_i$, the number $N_q$ of critical points of $H$ with a Morse index $q$ is given by $$N_p=\sum_{i}N_{i;p-\mu(\crit_i)} .$$ Thus, combining ( \[ontario\]) with ( \[oregon\]), one obtains \_[i,p;\_i]{}\^q b\_[q-(\_i)]{}(\_i)=\_q \^q b\_q(M)\
+(1+)\_q \^[q-1]{} ([dim]{}\^q-\_[\_i]{}[ dim]{}\_i\^[q-(\_i)]{}).Hence, ( \[Conley-Zehnder1\]) is equivalent to the statement that the polynomial on the last line, which is multiplied by $(1+\lambda)$, has non-negative integer coefficients.
By a homotopy argument, we will now prove that for all $q$, \[wdder\][dim]{}\^q \_[\_i]{}[ dim]{}\_i\^[q-(\_i)]{}holds, thus obtaining an alternative proof of ( \[Conley-Zehnder1\]). The main motivation here is to give an explicit construction that geometrically elucidates this relation, noting that the left hand side is defined by the flow of the ’free’ gradient-like system corresponding to $(M,\omega,H)$, while the right hand side is defined by the constrained gradient-like system corresponding to $(M,\omega,H,V)$. We note that ${\rm dim}\B_i^{q-\mu(\crit_i)}$ denotes the number of critical points of $H$ with a Morse index $q$ in $\crit_i$, which are effectively connected to critical points of Morse index $p+1$ in $\crit_i$ via gradient lines of the Morse function $H_i$ on $\crit_i$. Therefore, the sum on the right hand side of ( \[wdder\]) equals the number of those critical points of $H$ with a Morse index $q$, which are effectively connected to critical points of Morse index $q+1$ via gradient lines of the functions $H\circ j_i$ on all generic $\crit_i$. Here, $j_i:\crit_i\rightarrow M$ denotes the corresponding inclusion maps.
### Proof of ( \[wdder\])
Our strategy consists of constructing a homotopy of vector fields $v_\sig $, with $\sig \in[0,1]$, whose zeros are hyperbolic and independent of $\sig $, which generate gradient-like flows. They interpolate between $v_1:=\grad H$, and $v_0$, which is a vector field that is tangent to $\cg$. For every $\sig \in[0,1]$, we construct a coboundary operator via the one-dimensional integral curves of $v_\sig $ that connect its zeros. These coboundary operators are independent of $\sig $, and act on the free $\Z$-module $\Cb$ of the Morse-Witten complex associated to $(M,H)$. ( \[wdder\]) then follows from a simple dimensional argument.
There exists $v_0\in\Gamma(TM)$, which is gradient-like, and tangent to $\cg$. Furthermore, the zeros of $v_0$ are hyperbolic, and identical to the critical points of $H$. The dimension of any unstable manifold of the flow generated by $-v_0$ equals the Morse index of the critical point of $H$ from which it emanates.
We recall the vector field $X_\epsilon$ constructed in the proof of proposition [ \[Conley-Zehnderineqprp1nongeneric\]]{}, and consider the compact tubular $\epsilon$-neighborhoods $U_\epsilon(\cg)$, defined similarly as in ( \[neighb\]). Furthermore, let $\bar{Q}=\bar{Q}^2$ (resp. $Q=\1-\bar Q$) be $g$-orthogonal, smooth tensors on $TU_\epsilon(\cg)$ of fixed rank $2(n-k)$ (resp. $2k$), with ${\rm Ker}\{\bar Q(a)\}=N_a\cg$ (resp. ${\rm Ker}\{ Q(a)\}=T_a\cg$) for every $a\in\cg$.
We define $v_0$ as follows. In $M\setminus U_\epsilon(\cg)$, it shall equal $X_\epsilon$, and that for $x$ in $U_\epsilon(\cg)$, it shall be given by $$v_0(x):= (\PV\grad H)(x)+h(x)(\bar{Q}\grad H)(x) ,$$ where $h:U_\epsilon(\cg)\rightarrow[0,1]$ is a smooth function obeying $h|_{\cg}=1$ and $h|_{\partial U_\epsilon(\cg)}=0$. In particular, $h$ is assumed to be strictly monotonic along all non-constant trajectories of the flow generated by $\PV\grad H$, and $dh$ shall vanish on $\cg$.
It can be easily verified that $v_0$ possesses all of the desired properties. It generates a gradient-like flow, since outside of $U_\epsilon(\cg)$, $g(\grad H,v_0)=g(\grad H,X_\epsilon)>0$, as has been shown in the proof of proposition [ \[Conley-Zehnderineqprp1nongeneric\]]{}. In the interior of $U_\epsilon(\cg)$, one finds $g(\grad H,v_0)=\|\PV\grad H\|_g^2 +
h\|\bar{Q}\grad H\|_g^2$, due to the $g$-orthogonality both of $\PV$ and $\bar{Q}$. The first term on the right hand side vanishes everywhere on $\cg$, but at no other point in $U_\epsilon(\cg)$. The second term equals $\|\bar{Q}\grad H\|_g^2$ on $\cg$. Since evidently, $\bar{Q}\grad H|_{\cg}$ is the gradient field of the Morse function $H|_{\cg}:\cg\rightarrow\R$ relative to the Riemannian metric on $T\cg$ induced by $g$, its zeros are precisely the critical points of $H$ on $\cg$, and it possesses no other zeros. Because $g(v_0,\grad H)>0$ except at the critical points of $H$, it is clear that $-v_0$ generates a gradient-like flow $\psi_{0,t}$, so that $H$ is strictly decreasing along all non-constant orbits. Furthermore, it is clear from the given construction that $v_0$ is tangent to $\cg$.
To prove the remaining statements of the lemma, we note that the Jacobian matrix of $v_0$ at $a$ in a local chart is given by Dv\_0(a)= (D\^2\_a H)\^+(P\_a- Q\_a)(D\^2\_a H)\^. \[armgdd\]There is no dependence on $h$ because $dh|_{\cg}=0$. Furthermore, $(D^2_a H)^\sharp$ is defined as the matrix $[g^{ij}H_{,jk}|_a]$ in the given chart, and $P_a$ denotes the matrix of $\PV(a)$. Normal hyperbolicity follows from the invertibility of $Dv_0(a)$, which is verified in the proof of lemma [ \[SpecQaminPalm\]]{} below.
Let $v_\sig := \sig \grad H + (1-\sig )v_0$ with $\sig \in[0,1]$. Then, the flow $\psi_{\sig ,t}$ generated by $-v_\sig $ is gradient-like for any $\sig \in[0,1]$. The zeros of $v_\sig $ are hyperbolic fixed points of $\psi_{\sig ,t}$, and independent of $\sig $. Thus, the dimensions of the corresponding unstable manifolds equal the Morse indices of the critical points of $H$ from which they emanate, for all $\sig $.
We consider $g(\grad H,v_\sig )=\sig \|\grad H\|_g^2 +(1-\sig )g(\grad H,v_0)$. The first term on the right hand side is obviously everywhere positive except at the critical points of $H$, and the same has been proved previously for the second term. Thus, $H$ decreases strictly along all non-constant orbits of $\psi_{\sig ,t}$, hence the latter is gradient-like. The Jacobian of $v_\sig $ at a critical point of $H$ is given by Dv\_(a )= ([**1**]{}\_[2n]{}+(1-)(P\_a - Q\_a))(D\^2\_a H)\^. $Dv_\sig(a )$ is invertible for all $\sig \in [0,1]$, since $(D^2_a H)^\sharp$ is invertible, and ${\rm spec}\{\IPa P_a - \IQa Q_a\}\subset(-1,1)$. To prove the latter, we first observe that ${\rm spec}\{\IPa P_a - \IQa Q_a\}\subset[-1,1]$ is trivial, because $P_a$ and $Q_a$ both have a spectrum $\{0,1\}$. $\{-1,1\}$ is not included, because otherwise, $\bar P_a Q_a$, respectively $P_a \bar Q_a$, would not have a full rank, in contradiction to corollary [ \[SpecQaminPalm\]]{}.
By smoothness of $v_\sig $, it follows that $\psi_{\sig ,t}$ is $C^\infty$ in $\sig $. Thus, $\sig$ smoothly parametrizes a homotopy of stable and unstable manifolds of $\psi_{\sig ,t}$ emanating from the critical points of $H$. Since the fixed points of $\psi_{\sig ,t}$ are independent of $\sig $, and the dimensions of the corresponding unstable manifolds are equal to the Morse indices of the critical points of $H$, we consider, for every value of $\sig\in[0,1]$, the free $\Z$-module $\Cb=\oplus_p \Cb^p$ that is generated by the critical points of $H$, and graded by their Morse indices. For every $\sig$, we define a coboundary operator on $\Cb$, using $\psi_{\sig ,t}$ as follows. Picking a pair of critical points of $H$ with a relative Morse index $1$, we consider the unstable manifold $W^-_{\sig ,a}$ of $a$, and the stable manifold $W^+_{\sig ,a'}$ of $a'$ associated to $\psi_{\sig ,t}$. Since $\sig $ parametrizes a homotopy of such manifolds, they naturally inherit an orientation from the one picked for $\sig =1$ in the construction of the coboundary operator of the Morse-Witten complex for $(M,H)$.
Let $\Sigma_E$ denote a regular energy surface for $H(a)<E<H(a')$. $W_{\sig ,a}^\pm$ intersects $\Sigma_E$ transversely, because $H$ is strictly decreasing along all non-constant orbits generated by $-v_\sig $. $W^-_{\sig ,a}\cap\Sigma_E$ and $W^+_{\sig ,a'}\cap\Sigma_E$ define two homotopies of oriented submanifolds of $\Sigma_E$. By homotopy invariance of their intersection number, the coboundary operators are independent of $\sig $, and thus identical to the $\delta$-operator of the Morse-Witten complex given for $\sig =1$.
The stable and unstable manifolds of $\psi_{0,t}$ are either confined to some $\crit_i$, or connect critical points lying on different $\crit_i$’s. Let us consider pairs of critical points of $H$ with a relative Morse index $1$ that lie on the same component $\crit_i\in\cg$, and the corresponding stable and unstable manifolds of $\psi_{0,t}$ which are contained in $\crit_i$. Since $v_0|_{\crit_i}$ is the projection of $\grad H|_{\crit_i}$ to $T\crit_i$, these stable and unstable manifolds are the same as those which were used to define the Morse-Witten complex on $(\crit_i,H_i)$. Using only stable and unstable manifolds of $\psi_{0,t}$ contained in $\cg$, we construct an operator $\tilde{\delta}$ acting on $\Cb$ in the same manner in which the coboundary operator was defined, thus obtaining $\tilde{\delta}\equiv \oplus_i \delta_i $, where $\delta_i$ is the coboundary operator of the Morse-Witten complex associated to the pair $(\crit_i,H_i)$. Let $P_i:\Cb\rightarrow\Cb_i$ stand for the projection of the free $\Z$-module $\Cb$ generated by all critical points of $H$ to the one generated by the critical points contained in $\crit_i$. Eliminating all integral lines of $-v_0$ that connect critical points on different connectivity components of $\cg$ in the above construction, one sees that $\delta_i=P_i\delta P_i$, thus $\tilde \delta=P_i\delta P_i$. Hence, clearly, $${\rm dim} ({\rm im}\delta|_{\Cb^p})\geq{\rm dim}
({\rm im}\tilde{\delta}|_{\Cb^p})\;,$$ which precisely corresponds to ( \[wdder\]). This completes the proof.
QUALITATIVE ASPECTS RELATED TO CRITICAL STABILITY {#sectionIII}
=================================================
So far, we have established that in the generic case, the connectivity components of $\crit=\cg$ are embedded submanifolds of dimension $2(n-k)$ equal to the corank of $V$. Furthermore, we have seen that the topology of the symplectic manifold $M$ enforces the existence of connectivity components of $\cg$ of certain prescribed indices with respect to the auxiliary gradient-like flow $\phi_t$.
In this section, we focus on the physical dynamics, characterized by the flow $\tPhi_t$ generated by $X_H^V$, of the constrained Hamiltonian system $(M,\omega,H,V)$ in a tubular $\epsilon$-neighborhood of $\cg$, and particularly on the issue of stability. Let $g$ again denote the auxiliary Kähler metric introduced in section [ \[sectionII\]]{}, with the induced Riemannian distance function given by $dist_R$ (in contrast to the Carnot-Caratheodory distance function $d_{C-C}$ induced by $g$, which will also be considered). We recall that a point $\xa\in\cg$ is stable if there exists $\delta(\epsilon)>0$ for every $\epsilon>0$, so that for all $t$, $dist_R(\tPhi_t(x),\xa)< \epsilon$ holds for all $x$ with $dist_R(x,\xa)<\delta(\epsilon)$.
To elucidate the key differences between the local dynamics in the vicinity of $\cg$ for the cases of integrable and non-integrable $V$, let us first describe the situation where $V$ is integrable. As proved in corollary [ \[intcasecor\]]{}, $M$ is foliated into $2k$-dimensional symplectic submanifolds which intersect $\cg$ transversely. Thus, on every leaf ${\mathcal N}$, the equilibrium solutions are generically isolated points. Let the linear operator $\Omega_a$ correspond to the linearization of $X_H^V$ on $T_a M$ for some $a\in {\mathcal N}\cap\cg$, and restricted to the fibre $V_a=T_a {\mathcal N}\subset T_a M$. Its spectrum, if it is not purely imaginary, conclusively characterizes the stability of $a$; we refer to this as the asymptotically (un)stable case. If the spectrum of $\Omega_a$ is purely imaginary, which we refer to as the critically stable case, is is well-known that if there exists a local Lyapunov function $L_{{\mathcal N}}: U(a)\cap{\mathcal N}\rightarrow\R$ for $a$, then $a$ is stable.
If $V$ is non-holonomic, the situation is similar in the asymptotically (un)stable case, but drastically different in the critically stable situation. In the critically stable case, the presence of a local degenerate Lyapunov function for a single equilibrium $a\in\cg$ is of limited use, since there is a whole submanifold $\cg\cap\Sigma_E$ (the $H$-level set for the energy $E$ of the initial condition) of valid equilibria for a given energy $E$. One may relax this condition to the existence of the following class of functions.
\[LDLFdef\] Let $\nabla_g^\perp$ denote the component of the gradient $\nabla_g$ normal to $\cg$ with respect to $g$ at $\cg$. Let $U(a)$ be a $dist_R$-small open neighborhood of $a\in\cg$. A local, degenerate, almost Lyapunov function for $a$ is a class $C^1$ function $L:U(a)\rightarrow\R$, which satisfies $(\nabla_g^\perp L)(a')=0$ for all $a'\in\cg\cap U(a)$, and $\|\nabla_g L\|_g>0$ for all $x\in U(a)\setminus \cg$. Furthermore, $(\nabla_g dL)|_{a'}$ is positive definite quadratic form on $N_{a'}\cg$ for all $a'\in U(a)\cap\cg$, and $L(\tPhi_t(x_0))\leq L(x_0)$ for all $x_0\in U(x_0)$, and all $t$ such that $\tPhi_t(x_0)\in U(a)$.
Notably, $L$ defined here is not a local degenerate Lyapunov function, because $\crit_i\cap U(a)$ is not a critical level set (we remark that this would be equivalent to $L$ being a Morse-Bott function in $U(a)$), on which $L$ is extremal.
While the existence of $L$ guarantees that the orbit $\tPhi_t(x_0)$ remains within a tubular $\epsilon$-neighborhood of $\cg\cap U(a)$ for all $t$ such that $\tPhi_t(x_0)\in U(a)$, it does not imply stability of $a\in\cg\cap U$. There is an additional, necessary condition on the rational independence of the frequencies of the oscillatory linear problem that must be imposed. Otherwise, an inner resonance, connected to the appearance of small divisors, occurs, and $\tPhi_t(x_0)\in U(a)$ may evolve away from $a$, in a diffusive motion along the higher flag elements of $V$ that are approximately tangent to $\cg$, while along $V$, which is transverse to $\cg$, the motion is bounded and oscillatory.
From the analysis in section [ \[sectionII\]]{}, it is clear that for every connectivity component $\crit_i\subset\cg$ of index $\mu(\crit_i)=0$ (with respect to $\phi_t$), the Hamiltonian $H$ is a local degenerate, almost Lyapunov function for all of its points. The minimum $a^*$ of $H|_{\crit_i}$ on $\crit_i$ is a local minimum of $H$, and hence stable (since $H$ serves as a Lyapunov function for $a^*$). Hence, in particular, if $V$ is integrable, all points on $\crit_i$ are stable if $\mu(\crit_i)=0$. We also note that on the connectivity components $\crit_j$ with index $\mu(\crit_j)>0$, $H$ is never a local degenerate, almost Lyapunov function.
The main focus in this section will be to discuss issues of this type. However, an essential part of sections [ \[avtheorsubsub\]]{} and [ \[Loptsubsubsect\]]{} will be in mathematically non-rigorous terms, since a rigorous treatment of the matters addressed there would fall into the domain of KAM and Nekhoroshev theory, and is beyond the scope of the present work.
A concrete aim in this discussion is to arrive at stability criteria for equilibria of the constrained Hamiltonian system $(M,\omega,H,V)$. From an instructive, despite elementary, application of averaging theory, we conjecture a condition for the critically stable case that involves an incommensurability condition imposed on the frequencies of the linearized problem, as remarked above. In order to elucidate its geometric content, we study the dynamics in the vicinity of a critically stable equilibrium in a geometrically invariant form that is adapted to the flag of $V$. Invoking a perturbation expansion based on this description, we argue that the incommensurability condition, which might merely correspond to an artefact of the averaging method, cannot be omitted. A rigorous proof of the conjectured stability criterion is left for future work.
Stability Criteria
------------------
Let $\xa\in\cg$, and pick some small neighborhood $U(a)\subset M$ together with an associated Darboux chart, with its origin at $\xa$. The equations of motion are given by $\partial_t x_t = P(x_t)\mcJ H_{,x}(x_t) = X_H^V(x_t)$, where the coordinates are given by $x=(x^1,\dots,x^n,x_{n+1},\dots,x_{2n})$, and $\mcJ $ is the symplectic standard matrix. Furthermore, $H_{,x}$ abbreviates $\partial_{x}H$, and $P$ is the $2n\times2n$-matrix representing the tensor $\PV$. $\omega$-skew orthogonality of $\PV$ translates into $P(x)\mcJ X(x)=\mcJ P^\dagger(x)X(x)$ for all vector fields $X$.
There exists a chart in which the equations of motion have the form \_t (y\_t,z\_t)=( y\_t + Y(z\_t,y\_t) , Z(z\_t,y\_t) ) \^[2k]{}\^[2(n-k)]{} .\[parttzZeq\]In particular, $\Al$ corresponds to the restriction of $DX_H^V(0) \IP$ to $V_0$, and $|Y(y,z)|$, $|Z(y,z)|=O(|y|\,|z|)+O(|y|^2)$.
In a sufficiently small vicinity $U\subset\R^{2n}$ of the origin (corresponding to $a$), one infers from corollary [ \[SpecQaminPalm\]]{} that $T_a\cg\oplus V_0= \R^{2n}$, for $a\in U\cap \cg$. Accordingly, we choose local coordinates $z\in U'(0)\subset\R^{2(n-k)}$ on $\cg$, and $\tilde y\in V_{0}$, noting that the decomposition $x=a(z)+\tilde y$ for any $x\in U\subset\R^{2n}$ is unique, where $a:\R^{2{n-k}}\hookrightarrow U$ is the (smooth) embedding. Let $y$ denote the coordinates of $\tilde y$ with respect to some family of basis vectors for $V_0$. Then, ( \[parttzZeq\]) evidently follows from Taylor expansion.
### Asymptotic (In)stability
If spec$\{\Al\}\cup i\R=\emptyset$, there exists, by the center manifold theorem, a coordinate transformation $(y,z)\rightarrow(\bar{y},\bar{z})$, such that ( \[parttzZeq\]) becomes \_t(|[y]{}\_t , z\_t)=(|[y]{}\_t + |[Y]{}(|[y]{}\_t,|[z]{}\_t) , 0 ) [@ZeBlMa], where $\bar{Y}(0,\bar{z})=0$ for all $\bar{z}$. Thus, $\xa\in\cg$ is asymptotically unstable if there are eigenvalues with a positive real part, and asymptotically stable if all eigenvalues have a negative real part. If $V$ is integrable, asymptotic stability is impossible, because the dynamics is Hamiltonian on every integral manifold. However, if $V$ is non-integrable, there is, to the author’s knowledge, no obstruction to the existence of asymptotically stable equilibria, since the flow map is not symplectic.
### An Elementary Application of Averaging Theory {#avtheorsubsub}
In the case of critical stability, one has spec$\{\Al\}=\lbrace i\omega_1,\dots,i\omega_{2k}
\rbrace$, with $\omega_i\in \R\setminus\{0\}$ for $i=1,\dots,2k$. Let us for the context of an averaging analysis assume that the vector fields on the r.h.s. of ( \[parttzZeq\]) are real analytic with respect to $(y,z)$. We apply a complex linear coordinate transformation that diagonalizes $\Al$, and denote the complexified, new coordinates and vector fields again by $(y,z)$, and $Y(y,z)$, $Z(y,z)$, respectively, by which we find \[beginning\] \_t ( y\_t,z\_t) = ([diag]{}(i) y\_t + Y(y\_t,z\_t) ,Z(y\_t,z\_t)) \^[2k]{}\^[2(n-k)]{}, where $\omega:=(\omega_1,\dots,\omega_{2k})$. Complexifying ( \[parttzZeq\]), the continuation of $\cg$ into $\C^{2n}$ is defined as the common zeros of $Y(0,z)$ and $Z(0,z)$ for $z\in\C^{2(n-k)}$.
We next introduce polar coordinates $(I,\phi)\in\R^{2k}\times[0,2\pi)^{2k}$ and $(J,\theta)\in\R^{2(n-k)}\times[0,2\pi)^{2(n-k)}$ in terms of $y^r =: e^{i\phi_r} I^r$ and $z^s=: e^{i\theta_s} J^s$, with $r=1,\dots,2k$ and $s=1,\dots,2n-2k$. In particular, $I\in\R^{2k}$, $J\in\R^{2n-2k}$, $\phi\in[0,2\pi]^{2k}=\TT^{2k}$ (the $2k$-dimensional torus), and $\theta\in[0,2\pi]^{2n-2k}=\TT^{2n-2k}$. For brevity, let $e^{i\phi} v:=(e^{i\phi_1}v^1,\dots,e^{i\phi_{2k}}v^{2k})$ and $e^{i\theta}w :=(e^{i\theta_{1}} w^{1},\dots,e^{i\theta_{2(n-k)}} w^{2(n-k)})$, for $v\in\C^{2k}$ and $w\in\C^{2n-2k}$. ( \[beginning\]) is then easily seen to be equivalent to (the dot abbreviates $\partial_t$) = [Re]{}e\^[-i]{} Y(e\^[i]{}I,e\^[i]{}J)&, & = + [Im]{}e\^[-i]{} [diag]{}(\_I) Y(e\^[i]{}I,e\^[i]{}J). \[castor\]\
= [Re]{}e\^[-i]{} Z(e\^[i]{}I,e\^[i]{}J)& , & = [Im]{}e\^[-i]{} [diag]{}(\_J) Z(e\^[i]{}I,e\^[i]{}J). \[pollux\] Let us assume that $\epsilon:=|I(0)|\ll 1$, and $|J(0)|\leq
O(\epsilon^2)$. We then introduce rescaled variables $I\rightarrow\epsilon I$ and $J\rightarrow \epsilon^2 J$.
Analyticity of $Y(y,z)$ and $Z(y,z)$ with respect to $(y,z)$ implies that the power series expansion with respect to $e^{i\phi}I$ and $e^{i\theta}J$ converges for $\epsilon$ sufficiently small. Accordingly, ( \[castor\]) and ( \[pollux\]) yield \[bog1\] \^r&=&\_[|m|+|p|2]{}\^[|m|+2|p|-1]{}F\^r\_[mp]{}(I,J) e\^[i(m,-\_r)]{} e\^[ip,]{}\
\[bog2\] \^s&=&\_[|m|+|p|2]{}\^[|m|+2|p|-2]{}G\^s\_[mp]{}(I,J) e\^[im,]{}e\^[ip,]{}\
\[bog3\] \_r&=&\_r + \_[|m|+|p|2]{}\^[|m|+2|p|-1]{}\_[r;mp]{}(I,J) e\^[i(m,-\_r)]{} e\^[ip,]{}\
\[bog4\] \_s&=& \_[|m|+|p|2]{}\^[|m|+2|p|-2]{}\_[s;mp]{}(I,J) e\^[im,]{}e\^[ip,]{} ,where we introduced the multiindices $m\in\Z^{2k}$ and $p\in\Z^{2n-2k}$, with $|m|:=\sum|m_r|$ and $|p|:=\sum|p_s|$. In this Fourier expansion with respect to the $2\pi$-periodic angular variables $\phi$ and $\theta$, every Fourier coefficient labeled by a pair of indices $(m,p)$, is a homogenous polynomial of degree $|m|$ in $I$, and of degree $|p|$ in $J$.
If the components of $\omega$ are all mutually rationally independent, one may consider the averaged quantities $f_t(\phi)\rightarrow\bar{f}_t:=
(2\pi)^{-n}\int_{\TT^{n}}d\phi f_t(\phi)$. From ( \[beginning\]), $Y(y,z)$ and $Z(y,z)$ are $O(|y|)$, thus their power series involve terms $e^{i\langle m,\phi\rangle}$ with $|m|\geq 1$, but none with $|m|=0$. Averaging ( \[bog1\]) $\sim$ ( \[bog4\]) with respect to $\phi$ thus gives =\^2 (|[I]{},|[J]{},|) , =0 , =0 for some function $\tilde{F}$, where the bars account for averaged variables. Thus, if we in addition assume that there exists a local degenerate almost Lyapunov function with respect to $\cg\cap U$, it follows for the averaged equations of motion that $|\bar I|$ is bounded for all $t$. In particular, if the incommensurability condition holds, $|\bar J|$ is then also bounded for all $t$, and $a$ (respectively $0$) is, for the averaged system, stable. Based on these insights, and on intuition stemming from KAM and Nekhoroshev theory, it is thus natural to conjecture the following stability criterion.
\[stabcriteriaconj\] Let $\crit_i\subset\cg$ be a connectivity component of the critical manifold, and let $\xa\in\crit_i$, with [spec]{}$\{DX_H^V(a)\}\setminus\{0\}=\lbrace i\omega_1,\dots,
i\omega_{2k}\rbrace$, and $\omega_i\in\R\setminus\lbrace0\rbrace$ for $i=1,\dots,2k$. Assume that (1) the frequencies $\omega_r$ are rationally independent, and (2) that there exists a local degenerate, almost Lyapunov function with respect to $\crit_i\cap U(a)$, in the sense of definition [ \[LDLFdef\]]{}. Then, $\xa$ is stable in the sense of Nekhoroshev. In particular, condition (2) is always satisfied (by the Hamiltonian $H$) if the index of $\crit_i$ is $\mu(\crit_i)=0$.
The Relationship to Sub-Riemannan Geometry
------------------------------------------
To elucidate the geometric nature of the requirement of rationally independent frequencies, we will now approach the discussion of critical stability from a different point of view. This discussion involves issues that are central to sub-Riemannian geometry, [@BeRi; @Ge; @Gr2; @Str].
We study the time evolution map in a tubular $\epsilon$-vicinity of $U(a)\cap\cg$ by invoking a geometrically invariant Lie series that is adapted to the elements of the flag of $V$. By an asymptotic analysis, we explain the mechanism by which an instability can arise. The reason is that if the eigenfrequencies of the linear problem are not incommensurable, the problem of small divisors appears. This picture seems to be familiar from the perturbation theory of integrable Hamiltonian systems, but we note once more that the lack of integrability here originates from the non-holonomy of the constraints. A rigorous treatment of this last part of the analysis is beyond our current scope, and left for future work.
### Dynamics Along the Flag of V
Let $U$ denote a small open neighborhood $U$ of $\xa\in\crit$, and assume that $\cg:=\crit\cap U$ satisfies the genericity condition of theorem [ \[Sardthm\]]{}.
\[tubneigh\] Let $\cg=\crit\cap U$ have the genericity property formulated in theorem [ \[Sardthm\]]{}. Then, there exists $\epsilon>0$ such that every point $x\in U$ with $d_R(x,\cg)<\epsilon$ is given by $$x=\exp_s Y (\xa)\;\;\;\;,\;\;\;\;|s| < \epsilon$$ for some $Y\in\Gamma(V)$ with $\|Y\|_{g_M}\leq 1$, $\xa\in\cg$ ($\exp_s Y$ denotes the 1-parameter group of diffeomorphisms generated by $Y$, with $\exp_0 Y=$id).
We choose a spanning family $\lbrace Y_i\in\Gamma(V)\rbrace_1^{2k}$ of $V$, with $\|Y_i\|_{g_M}= 1$. If for all $\xa\in\cg$, $T_{\xa}\cg$ contains no subspace of $V_{\xa}$, then $$\exp_1(t_1Y_1+\dots+t_{2k}Y_{2k})(\cg)\;\;\cap\;\;U$$ is an open tubular neighborhood of $\cg$ in $U$, for $t_i\in (-\epsilon,\epsilon)$. Because the normal space $N_{\xa}\cg$ is dual to the span of the 1-forms $dF_i$ at $\xa$, this condition is satisfied if and only if the matrix $[dF_j(Y_i)]=[Y_i(Y_j(H))]$ is invertible everywhere on $\cg$. According to proposition [ \[invert\]]{}, this condition is indeed fulfilled.
Hence, there is an element $Y\in\Gamma(V)$ with $\|Y\|_{g_M}\leq 1$, so that $x=\Psi_\epsilon(\xa)$ for some $0<\epsilon\ll 1$. Since $\xa\in\cg$, it is clear that under the flow generated by $X_H^V$, $\tPhi_{\pm t}(\xa)=\xa$, thus the solution of ( \[eqsofmo\]) with initial condition $x$ is given by \_\^t():= \_t\_()=(\_t \_ \_[-t]{})().$\Psi_\epsilon^t$ is, in particular, the 1-parameter group of diffeomorphisms with respect to the variable $\epsilon$ that is generated by the pushforward vector field \[pushf\]Y\_t(x):=\_[t]{}Y(x)= d\_[t]{}Y(\_[-t]{}(x)),where $d\tPhi_t$ denotes the tangent map associated to $\tPhi_t$. From the group property $Y_{s+t}=\tPhi_{s\,*} Y_t$ follows that \[Ytder\]\_t Y\_t=.\_s|\_[s=0]{} \_[s]{} Y\_t =,everywhere in $U$.
Next, we pick a local spanning family $\left\lbrace Y_i\in\Gamma(V)\right\rbrace_{i=1}^{2k}$ for $V$ that satisfies $\omega(Y_i,Y_j)=\tJ_{ij}$, with $\tJ:=\left(\begin{array}{cc}0&\1_k\\-\1_k&0\end{array}\right)$. Furthermore, defining $\theta_i(\cdot):=\omega(Y_i,\cdot)$, $\PV=\tJ^{ij}Y_i\otimes \theta_j$, where $\tJ^{ij}$ are the components of $\tJ^{-1}=-\tJ$. In particular, X\_H\^V=(X\_H\^V)=-Y\_i(H)\^[ij]{} Y\_jin the basis $\left\lbrace Y_i\right\rbrace_{i=1}^{2k}$.
The following proposition characterizes the orbit emanating from $x$ in terms of nested commutators with respect to $Y_t$.
Let $f,F_i\in C^\infty(U)$, where $F_i := Y_i(H)$, $i=1,\dots,2k$, and assume that $F_i(\Psi_\epsilon^t(\xa))$, $f(\Psi_\epsilon^t(\xa))$ are real analytic in $\epsilon$. For $X,Y\in \Gamma(TM)$, let $$\cL_Y^r X \;=\;[Y,\dots,[Y,X]]$$ denote the $r$-fold iterated Lie derivative. Then, for sufficiently small $\epsilon$, \[dertflag\]\_t f(\_\^t())=- F\_i(\_\^t())\^[ik]{}\_[r0]{} (\_[Y\_t]{}\^[r]{} Y\_k)(f\_\^t)().
Clearly, \_t f(\_\^t())&=&X\_H\^V(f) (\_\^t())\
&=&- F\_i(\_\^t())\^[ik]{} Y\_k(f)(\_\^t())\
&=&- F\_i(\_\^t())\^[ik]{} (\_\^t Y\_k ) (f\_\^t)().Using the Lie series $\Psi_{\epsilon\;*}^t Y_k = \sum_r \frac{\epsilon^r}{r!}
\cL_{Y_t}^{\;r} Y_k$, we arrive at the assertion.
\[commflag\] Assume that $Y_{t=0}\in\Gamma(V)$, and let $\lbrace Y_j\rbrace_1^{2k}$ be the given local spanning family of $V$. Then, $\cL_{Y_t}^{\;i}Y_j\in\Gamma(V_i)$, where $V_i$ is the $i$-th flag element of $V$.
Since $\tPhi_{t\;*}:\Gamma(V)\rightarrow\Gamma(V)$, $Y_t$ is a section of $V$ for all $t$ if it is for $t=0$. The claim immediately follows from the definition of the flag of $V$.
Proposition [ \[commflag\]]{} implies that there are functions $a^i(t,\cdot)\in C^\infty(U)$, $i=1,\dots,2k$, so that $Y_t(x) = a^i(t,x)Y_i$. The next proposition determines their time evolution.
Let $Y_{t=0}=a_0^i Y_i$ define the initial condition, and introduce the matrix $$\Omega_x:=[Y_l(F_i)(x)\tJ^{ij}]\;.$$ Then, pointwise in $x$, \[abODE\] a\^m(t,x)&=&((-t\_x))\^m\_j a\_0\^j +F\_j(x)R\^[jm]{}\_i(t,x)a\_0\^i ,where $$R^{jm}_i(t,x):= \tJ^{jl}\tJ^{nk} \int_0^t ds
(\exp(-(t-s)\Omega_x))^m_k
\omega\left([Y_l,\tPhi_{s\,*}Y_i], Y_n\right)\;.$$
The initial condition at $t=0$ is given by $Y_0=a^i_0 Y_i$, that is, by $a^i(0,x)=a^i_0$. Thus, by the definition of $Y_t$ in ( \[pushf\]), one has $Y_t=a^i_0\, \tPhi_{t\;*}Y_i$, so that $a^i(t,x)Y_i=a^i_0\tPhi_{t\;*}Y_i$. From $\omega(Y_i,Y_j)=\tJ_{ij}$, $\tJ_{ik}=-\tJ_{ki}$ and $\tJ_{im}\tJ^{ml}=-\delta^l_i$, $$a^l(t,x)=-a^i_0\omega\left(\tPhi_{t\;*}Y_i\,,\,Y_{j}\right)
\tJ^{jl}\;.$$ Now, taking the $t$-derivative on both sides of the equality sign, one finds \_t a\^m(t,x)&=&- a\^i\_0 (\[X\_H\^V,\_[t]{}Y\_i\],Y\_[k]{})\^[km]{}\
&=&- a\^i(t,x) Y\_i(F\_j)(x) \^[jm]{}\
&& - a\^i\_0 F\_j(x) \^[jl]{} \^[km]{} (\[Y\_l,\_[t]{}Y\_i\],Y\_[k]{}).Using the variation of constants formula pointwise in $x$, one arrives at the assertion.
### Non-holonomy and small divisors {#Loptsubsubsect}
Using the description of the dynamics in the vicinity of $a$ derived above, we will here use the small parameter $\epsilon$ for an asymptotic expansion. The intention of this part of the discussion, which is not rigorous, is to explain the geometric origin of the incommensurability condition on frequencies in conjecture [ \[stabcriteriaconj\]]{}.
We consider the following simplified situation:
[()]{}
$\Omega_x=\Omega$, constant for all $x$ in $U$.
spec$\{\Omega\}=\lbrace i\omega_1,\dots,i\omega_{2k}\rbrace$, with $\omega_r\in \R$.
$\|\Omega\|:=\sup_r|\omega_r|\ll\frac{1}{\epsilon}$.
Let us briefly comment on the generic properties of $\lbrace\omega_r\rbrace$. Writing $\Omega=\tJ A$, we decompose the matrix $A=[Y_i(Y_j(H))(\xa)]$ into its symmetric and antisymmetric parts $A_+$ and $A_-$, respectively. $A_-=[[Y_j,Y_i](H)(\xa)]/2$ vanishes if $V$ is integrable, which one deduces from $X_H|_{\xa}\in V_{\xa}^\perp$ for all $\xa\in \cg$, and the Frobenius condition. The linear system $\underline{\dot{a}}=\tJ A_+ \underline{a}$ is Hamiltonian, hence the spectrum of $\tJ A_+$ consists of complex conjugate pairs of eigenvalues in $i\R$ if it is purely imaginary (here, $\underline{a}:=(a^1,\dots,a^{2k})$). Considering $\tJ A_-$ as a perturbation of $\tJ A_+$, we may generically assume that all frequencies $\omega_r$ are distinct from one another, and that there are both negative and positive frequencies.
Under the simplifying assumptions at hand, let us compute ( \[abODE\]) to leading order in $\epsilon$. From ( \[abODE\]), one infers $$Y_t=a^j_0(\exp(-t\Omega))^i_j Y_i + \sum_i O(|x|) Y_i\;,$$ since $|F_j(x)|=O(|x|)=O(\epsilon)$, which follows from $F_j(\xa)=0$. Thus, $$[Y_t,X]= a^j_0 \exp(-t \Omega)_j^i [Y_i,X] +
\sum_i O(\epsilon) [Y_i,X]
+\sum_i O(1) Y_i$$ for all $X\in\Gamma(TM)$, and $x\in U_\epsilon(\xa)$. Assuming that all objects in question are $C^\infty$, iterating the Lie bracket $\cL_{Y_t}$ $r$ times produces $$\Big(\prod_{m=1}^r
a^{j_m}_0 \big(\exp(-t \Omega)\big)^{i_m}_{j_m} + O(\epsilon)
\Big)
[Y_{i_1},[Y_{i_2},\dots,[Y_{i_r},Y_l]\cdots]]\;,$$ plus a series of terms with less than $r$ nested Lie commutators that contribute to higher order corrections.
Let us, for the discussion of the leading order terms along each flag element of $V$, omit the relative errors of order $O(\epsilon)$. By the assumption of smoothness, our considerations are valid for $t\leq O(\epsilon^{-1})$. Let us consider the term \[Vrcontr\] F\_i(\_\^t()) \^[ik]{} (\^[r]{}\_[Y\_t]{}Y\_k) (f\_\^t)(), for fixed $r$. It is easy to see that \[Fiser\] F\_i(\_\^t()) = Y\_t(F\_i)() + O(\^2),due to $F_i(\xa)=0$. Therefore, F\_i(\_\^t())\^[ik]{}= (-t)\^m\_j a\_0\^j\^k\_m+ O(\^2),from a straightforward calculation.
Hence, the terms with $r$ nested commutators in ( \[Vrcontr\]) are a\^i\_0 (-t)\^j\_i \_j\^l(\_[m=1]{}\^r a\^[j\_m]{}\_0 ((-t))\^[i\_m]{}\_[j\_m]{} ) \[Y\_[i\_1]{},\[ …,\[Y\_[i\_r]{},Y\_l\]\]\](f)()\
+ O(\^[r+2]{}),as long as $dist_R(\Psi_\epsilon^t(\xa),\xa)\leq O(\epsilon)$. This implies that for $f\in C^\infty(U)$, f(\_\^t()) &\~& f() + \_[r0]{} \_0\^t ds a\^i\_0 (-s )\^j\_i \_j\^l\
&& (\_[m=1]{}\^r a\^[j\_m]{}\_0 ((-s))\^[i\_m]{}\_[j\_m]{} ) \[Y\_[i\_1]{},…,\[Y\_[i\_r]{},Y\_l\]\](f)(), \[fflagexp\]up to relative errors of higher order in $\epsilon$ for every fixed $r$.
If $f$ is chosen as the $i$-th coordinate function $x^i$, so that $f(\Psi_\epsilon^t(\xa))=x^i_t$, the quantity $[Y_{i_1},\dots,[Y_{i_r},Y_l]\cdots](f)(\xa)$ is the $i$-th coordinate of the vector field defined by the brackets at $\xa$. Consequently, ( \[fflagexp\]) is the component decomposition of $x^i_t$ relative to the flag of $V$ at $\xa$, to leading order in $\epsilon$.
By the given simplifying assumptions, spec$\{\Omega\}\subset i\R\setminus\{0\}$, and the norm of $\exp(-s\Omega)$ is 1, independently of $s$. Consequently, the integrand of ( \[fflagexp\]) is bounded for all $s$. It follows that if the $r$-th integral in the sum diverges, it will become apparent only for $t\geq O\big(\frac{1}{\epsilon^{r}}\big)$. This would correspond to an instability along the direction of the flag element $V_r$. While the leading term with $r=0$ is bounded for all $t$, terms with $r>0$ can diverge.
We next write (s)=(-s ) \_0 = \_[=1]{}\^[2k]{} A\_\_(-i\_s) ,where $\lbrace \underline{e}_\alpha\rbrace$ is an orthonormal eigenbasis of $\Omega$ with respect to the standard scalar product in $\C^{2k}$, and spec$\{\Omega\}=\lbrace i\omega_\alpha\rbrace$. The amplitudes $A_\alpha\in \C$ are determined by the initial condition $a^i(t=0)=a_0^i$, which we assume to be nonzero. By linear recombination of the vector fields $Y_i$, one can set $e^i_\alpha=\delta_{i,\alpha}$. Then, ( \[fflagexp\]) can be written as \[intnestcomm\]\_[r0]{} \_[l;i\_1,…,i\_r]{} I\_[l;i\_1,…,i\_r]{}(t) \[Y\_[i\_1]{},…,\[Y\_[i\_r]{},Y\_l\]\](f)(),where \[Iindt\] I\_[l;i\_1,…,i\_r]{}(t)&:=& \_0\^t ds \_l A\_l (\_[m=1]{}\^r A\_[j\_m]{} ) e\^[-is (\_l + \_[m=1]{}\^r \_[j\_m]{} )]{}\
&=& (\_[m=1]{}\^r A\_[j\_m]{} ) ( e\^[ -it (\_l + \_[m=1]{}\^r \_[j\_m]{} )]{} - 1 ). \[Ili1dotsirdef\]We note that the sum of frequencies (of generically indefinite signs) in the denominator on the last line raises the problem of small divisors. We also remark that evidently, the nested commutators vanish if all indices $i_1,\dots,i_r,l$ have equal values, as it should be (otherwise, the solutions would always diverge).
### Rational frequency dependence and blow-up of solutions
Let us next discuss the situation in which the small divisors approach zero. To this end, we introduce the set \^[(r)]{}(t):=I\_[l;i\_1,…,i\_r]{}(t) \_[l,i\_j=1]{}\^[2k]{} I\_[l;l,…,l]{}(t) \_[l=1]{}\^[2k]{},which we endow with the norm $\|\I^{(r)}(t)\|:=\sup_{I(t)\in\I^{(r)}(t)} |I(t)|$, and let $\|A\|:=\sup_{i=1,\dots,2k}\lbrace|A_i|\rbrace$, where $A_i$ are $\C$-valued amplitudes.
Furthermore, let $\Aa:=\lbrace\omega_1,\dots,\omega_{2k}\rbrace$, and let \_r := \_[r]{} ,denote its $r$-fold sumset, which is the set containing all sums of $r$ elements of $\Aa$.
For two sets of real numbers $\Aa$ and ${\mathfrak B}$, we define d(,[B]{}) := \_[i,j]{}|a\_i-b\_j || a\_i ,b\_j . Then, it follows from ( \[Ili1dotsirdef\]) that if $d(\Aa_r,-\Aa)>0$, \^[(r)]{}(t) d(\_r,-)\^[-1]{} A\^r(the sum over frequencies $\sum_{m=1}^r
\omega_{j_m}$ in ( \[Ili1dotsirdef\]) is an element of $\Aa_r$, and can only equal $-\omega_l$ if $d(\Aa_r,-\Aa)=0$). However, if $d(\Aa_r,-\Aa)=0$, there is a tuple of indices $\lbrace l;i_1,\dots,i_r\rbrace$ such that I\_[l;i\_1,…,i\_r]{}(t) = - t \_l A\_l \_[m=1]{}\^r A\_[j\_m]{} ,in case of which $\|\I^{(r)}(t)\| \sim t$, that is, a divergence linear in $t$ for large $t$ (recalling that the present asymptotic considerations require $t\leq\epsilon^{-1}$). Only if there are simultaneously positive and negative frequencies, $d(\Aa_r,-\Aa)=0$ is possible, but due to the remark at the beginning of subsection [ \[Loptsubsubsect\]]{}, this situation must generically assumed to be given.
As an illustration, the following picture holds for $r\leq2$. The fact that for $r=0$, $\|\I^{(0)}(t)\|$ is bounded for all $t$ is clear. For $r=1$, the first flag element $V_1=[V,V]$ is in question. The condition for the emergence of a divergence is that $d(\Aa,-\Aa)=0$. This is precisely given if there is a pair of frequencies $\pm\omega_i$ of equal modulus, but opposite sign. For $r=2$, assuming that $d(\Aa,-\Aa)>0$, the condition $d(\Aa_2,-\Aa)=0$ implies that there is a triple of frequencies such that $\omega_{i_1}+\omega_{i_2}=
-\omega_{i_3}$, $i_j\in\lbrace1,\dots,2k\rbrace$. If this occurs, the solution will diverge in the direction of the second flag element, $V_2=[V,[V,V]]$. The discussion for $r>2$ continues in the same manner.
Hence, our conclusion from this asymptotic analysis is that if $d(\Aa_r,-\Aa)=0$ for some $r$, then $\|\I^{(r)}(t)\|= O( t)$ for $t\rightarrow\infty$.
The physical insight gained from the above discussion can be summarized as follows. If the frequencies of the linearized problem fail to satisfy the incommensurability condition $d(\Aa_r,-\Aa)>0$ for all $r$, the equilibrium $\xa$ is unstable. However, the time required for an orbit to exit from a Riemannian $\epsilon$-neighborhood $U_\epsilon(\xa)$ is very large. In fact, assuming that $d(\Aa_r,-\Aa)=0$ for some $r\leq \rV $ (the degree of non-holonomy of $V$), a time $t\sim O(\frac{1}{\epsilon^{r}})$ is necessary to exit from $U_\epsilon(\xa)$ in the direction of the flag element $V_r$ (due to the factor $\frac{\epsilon^r}{r!}$ in ( \[fflagexp\])). We note that the orbit does not drift out from $U_\epsilon(\xa)\cap\cg$ in the direction of $V_{\xa}$ owing to the existence of a local degenerate Lyapunov function required in the conjectured stability criterion. Therefore, this discussion suggests that the incommensurability condition imposed on the frequencies of the linearized system can indeed not be omitted.
### Instabilities in the Context of Carnot-Caratheodory Geometry
The constrained Hamiltonian system $(M,\omega,H,V)$ shares many characteristics with systems typically encountered in sub-Riemannian geometry [@BeRi; @Ge; @Gr2; @Str]. The natural metric structure in this context is given by the Carnot-Caratheodory distance function $dist_{C-C}$ induced by the Riemannian metric $g$. It assigns to a pair of points $x,y\in M$ the length of the shortest $V$-horizontal $g$-geodesic.
If $V$ satisfies the Chow condition, $dist_{C-C}(x,y)$ is finite for all $x,y\in M$, by the Rashevsky-Chow theorem [@BeRi; @Gr2]. In this case, the Carnot-Caratheodory $\epsilon$-ball $$B^{C-C}_\epsilon(\xa) := \Big\lbrace x\in M \Big|
dist_{C-C}(x,\xa) < \epsilon \Big\rbrace$$ is open in $M$.
If $V$ fails to satisfy the Chow condition, pairs of points that cannot be joined by $V$-horizontal $g_M$-geodesics are assigned a Carnot-Caratheodory distance $\infty$. Then, $M$ is locally foliated into submanifolds $N_\lambda$ of dimension $(2n-{\rm rank}V_{\rV })$ (we recall that $\rV $ denotes the degree of non-holonomy of $V$), with $\lambda$ in some index set, which are integral manifolds of the (necessarily integrable) final element $V_{\rV }$ of the flag of $V$. On every $N_\lambda$, the distribution $V_\lambda:=j_\lambda^* V$ satisfies the Chow condition, where $j_\lambda:N_\lambda\rightarrow M$ is the inclusion. Therefore, all points $x,y\in N_\lambda$ have a finite distance with respect to the Carnot-Caratheodory metric induced by the Riemannian metric $j_\lambda^*g_M$. Every leaf $N_\lambda$ is an invariant manifold of the flow $\tPhi_t$.
Let $\lbrace Y_{i_r}\rbrace_{r=1}^{\rV }$ denote a local spanning family of $TM$ such that $\lbrace Y_{i_r}\rbrace$ spans the flag element $V_r$. Let the $g$-length of all $Y_{i_r}$’s be 1. Then, we define the ’quenched’ box $${\rm Box}_\epsilon(x) := \Big\lbrace
\exp_1\Big(\sum_{r=1}^{\rV }
\epsilon^r\sum_{i_r=1}^{{\rm dim}V_r}t_{i_r}\;Y_{i_r}\Big)(x)\;
\Big|\;t_{i_r} \in (-1,1)\Big\rbrace\;$$ in $N_\lambda$, where $\lambda$ is suitably picked so that $x\in N_\lambda$. Evidently, if $V$ satisfies Chow’s condition, $N_\lambda=M$. According to the ball-box theorem [@BeRi; @Gr2], there are constants $C>c>0$, such that $${\rm Box}_{c\epsilon}(x)\;\subset\;B^{C-C}_\epsilon(x)\;
\subset\;{\rm Box}_{C\epsilon}(x)\;.$$ Carnot-Caratheodory $\epsilon$-balls can be approximated by quenched boxes in Riemannian geometry.
The above perturbative results imply that if there is some $r<\rV $, for which $d(\Aa_r,-\Aa)=0$, the flow $\tPhi_t$ blows up the quenched boxes, and thus the Carnot-Caratheodory $\epsilon$-ball around $\xa\in\cg$, linearly in $t$, and along the direction of $V_r$. In fact, $B^{C-C}_\epsilon(\xa)$ is widened along $V_r$ at a rate linear in $t$. For $t=O(\frac{1}{\epsilon})$, $\tPhi_t$ maps the Carnot-Caratheodory $\epsilon$-ball containing the initial condition to a Carnot-Caratheodory ball of radius $O(1)$. Thus, in the context of Carnot-Caratheodory geometry, these instabilities, which have no counterpart in systems with integrable constraints, are far more significant than in the Riemannian picture.
AUTONOMOUS NON-HOLONOMIC SYSTEMS IN CLASSICAL MECHANICS {#sectionIV}
=======================================================
In this main section, we focus on the analysis of non-holonomic mechanical systems, and their relationship to the constrained Hamiltonian systems considered previously. The discussion is restricted to linear non-holonomic, [*Pfaffian*]{} constraints.
Let $(Q,g,U)$ be a Hamiltonian mechanical system, where $Q$ is a smooth Riemannian $n$-manifold with a $C^\infty$ metric tensor $g$, and where $U\in C^\infty(Q)$ denotes the potential energy. No gyroscopic forces are taken into consideration. Let $g^*$ denote the induced Riemannian metric on the cotangent bundle $T^*Q$. For $X\in\Gamma(TM)$, let $\theta_X$ be the 1-form defined by $\theta_X(Y)=g(X,Y)$ for all $Y\in\Gamma(TQ)$. Clearly, $g(X,Y)=g^*(\theta_X,\theta_Y)$ for all $X,Y\in\Gamma(TQ)$.
The Kähler metric of the previous discussion, also denoted by $g$, will not appear in this section. From here on, $g$ will denote the Riemannian metric on $Q$, which should not give rise to any confusion.
In a local trivialization of $T^*Q$, a point $x\in T^*Q$ is represented by a tuple $(q^i,p_j)$, where $q^i$ are coordinates on $Q$, and $p_k$ are fibre coordinates in $T^*_q Q$, with $i,j=1,\dots,n$. The natural symplectic 2-form associated to $T^*Q$, can be written in coordinates as $$\omega_0 = \sum_i dq^i \wedge dp_i\;=\;-d\theta_0\;.$$ $\theta_0=p_i dq^i$ is referred to as the symplectic 1-form.
We will only consider Hamiltonians of the form \[hamham\]H(q,p)=g\_q\^\*(p,p)+U(q) . In local bundle coordinates, the corresponding Hamiltonian vector field $X_H$ is given by $$X_H\;=\; \sum_i \left( (\partial_{p_i}H)\partial_{q^i} -
(\partial_{q^i}H)\partial_{p_i}\right)\; .$$ The orbits of the associated Hamiltonian flow $\Phi_t$ satisfy \[herodot\]\^i = \_[p\_i]{} H(q,p), \_j = -\_[q\^j]{}H(q,p) .The superscript dot abbreviates $\partial_t$, and will be used throughout the discussion.
Let ${\mathcal A}_I$ denote the space of smooth curves $\gamma:I\subset\R\rightarrow T^*Q$, with $I$ compact and connected, and let $t$ denote a coordinate on $\R$. The basis one form $dt$ defines a measure on $\R$. The action functional is defined by ${\mathcal I}:{\mathcal A}_I\rightarrow \R$, \[sagittarius\][I]{}\[\]&=&\_Idt (\^\*\_0 - H)\
&=&\_I dt( p\_i(t)\^i(t)-H(q(t),p(t))),with $\dot{\gamma}=\sum(\dot{q}^i\partial_{q^i}+\dot{p}_i\partial_{p_i})$. Denoting the base point projection by $\pi:T^*Q\longrightarrow Q$, let $c:=(\pi\circ\gamma):I\longrightarrow Q$ denote the projection of $\gamma$ to $Q$. We assume that $\|c(I)\|$ is sufficiently small so that solutions of ( \[herodot\]) exist, which connect the end points $c(\partial I)$. Among all curves $\gamma:I\rightarrow T^*Q$ with fixed projected endpoints $c(\partial I)$, the ones that extremize ${\mathcal I}$ are physical orbits of the system.
Linear Non-Holonomic Constraints
---------------------------------
Let us next impose linear, ’Pfaffian’ constraints on the Hamiltonian mechanical system $(Q,g,U)$, by adding a rank $k$ distribution $W$ over $Q$ to the existing data, and invoke the Hölder variational principle, [@Ar1], that generates the correct physical flow on $T^*Q$. The orbits of the resulting constrained dynamical system possess $W$-horizontal projections to $Q$.
We introduce the $g$-symmetric projection tensor associated to $W$ given by $\alph=\alph^2:TQ\rightarrow TQ$, with $${\rm Ker}(\alph)\;=\;W^\perp\;\;\;\;,\;\;\;\;
\alph(X)\;=\;X\;\;\;\;\forall\;X\;\in\;\Gamma(TQ)\;,$$ and its orthogonal complement $\bbeta=\1-\alph$. We note that in local coordinates, $\alph$ is represented by a $n\times n$ matrix of rank $k$. The dual of $W$, denoted by $W^*$, is defined as the image of $W$ under the isomorphism $g:TQ\rightarrow T^*Q$, and likewise for $(W^*)^\perp := g\circ W^*$. The corresponding $g^*$-orthogonal projection tensors on $T^*Q$ are denoted by $\alph^\dagger$ and $\bbeta^\dagger$, respectively. Our inspiration to introduce $\alph$ and $\bbeta$ for this analysis stems from [@Bra].
### Dynamics of the Constrained Mechanical System
Next, we derive the equations of motion of the constrained mechanical system from the Hölder variational principle. For a closely related approach to the Lagrangian theory of constrained mechanical systems, cf. [@CaFa].
A projective $W$-horizontal curve in $T^*Q$ is an embedding $\gamma:I\subset\R\hookrightarrow T^*Q$ whose image $c=\pi\circ\gamma$ under base point projection $\pi:T^*Q\rightarrow Q$ is tangent to $W$.
Let $\gamma_s:I\rightarrow T^*Q$, with $s\in[0,1]$, be a smooth one parameter family of curves for which the end points $c_s(\partial I)$ are independent of $s$ (where $c_s:=\pi\circ\gamma_s$).
A $W$-horizontal variation of a projective $W$-horizontal curve $\gamma$ is a smooth one parameter family $\gamma_s:\R\rightarrow T^*Q$, with $s\in[0,1]$, for which $\frac{\partial}{\partial s}(\pi\circ\gamma_s)$ is tangent to $W$, and $\gamma_0=\gamma$.
Let $$\delq^i(t):=\left.\partial_s\right|_{s=0}q^i(s,t)\;\;\;\;,\;\;\;\;
\delp_k(t):=\left.\partial_s\right|_{s=0}p_k(s,t)\;.$$ To any $W$-horizontal variation $\gamma_s$ of a $W$-horizontal curve $\gamma_0$ with fixed projections of the boundaries \[varcond\](\_s)(I)= (\_0)(I),so that $\delq^i|_{\partial I}=0$, we associate the action functional $${\mathcal I}[\gamma_s]=\int_I \left(\sum
p_i(s,t)\dot{q}^i(s,t)-H(q(s,t),p(s,t))\right)dt\;.$$
(Hölder principle) A physical orbit of the constrained mechanical system $(Q,g,U,W)$ is a projective $W$-horizontal curve $\gamma_0:I\rightarrow T^*Q$ that extremizes ${\mathcal I}[\gamma_s]$ among all $W$-horizontal variations $\gamma_s$ which satisfy ( \[varcond\]).
Hence, if \[varact\] \[\_s\]=p\_i \^i|\_[I]{} +\_I ((\_i-\_[q\^i]{}H)\^i - (\^i+ \_[p\_i]{}H)\_i)=0for all $W$-horizontal variations of $\gamma_0$ that satisfy $\delq^i|_{\partial I}=0$, then $\gamma_0$ is a physical orbit.
\[dae\] In the given local bundle chart, the Euler-Lagrange equations of the Hölder variational principle are the differential-algebraic relations \[qdotcon\] &=& (q)\_p H(q,p)\
\[pdotcon\] \^(q) &=& -\^(q) \_q H(q,p)\
\[physleaf\] (q)\_p H(q,p)&=&0.
The boundary term vanishes due to $\delq^i|_{\partial I}=0$.
For any fixed value of $t$, one can write $\delq(t)$ as $$\delq(t) \;=\; \sum_{\alpha=1}^k \;f_\alpha(q(t))\; Y_\alpha(q(t))\; ,$$ where $Y_\alpha$ is a $g$-orthonormal family of vector fields over $c(I)$ that spans $W_{c(I)}$. Furthermore, $f_\alpha\in C^\infty(c(I))$ are test functions obeying the boundary condition $f_\alpha(c(\partial I))=0$.
Since $f_\alpha$ and $\delp$ are arbitrary, the terms in ( \[varact\]) that are contracted with $\delq$, and the ones that are contracted with $\delp$ vanish independently. In case of $\delq$, one finds $$\int_I\; dt\;f_\alpha\;\;
(\dot{p}+\partial_q H)_i\;Y_\alpha^i\;=0$$ for all test functions $f_\alpha$. Thus, $(\dot{p}+\partial_q H)_i\,Y_\alpha^i=0$ for all $\alpha=1,\dots,k$, or equivalently, $\alph^\dagger(\dot{p}+\partial_q H)=0$, which proves ( \[pdotcon\]).
Since $\gamma_0$ is $W$-horizontal, $\bbeta(q)\dot{q}=0$, so the $\delp$-dependent term in $\delta{\mathcal I}[\gamma_s]$ gives \_I dt (-\_p H)\^i(\^)\_i +\_I dt (\_p H)\^i(\^)\_i =0.The components of $\delp$ in the images of $\alph^\dagger(q)$ and $\bbeta^\dagger(q)$ can be varied independently. Thus, both terms on the second line must vanish separately, as a consequence of which one obtains ( \[qdotcon\]) and ( \[physleaf\]).
The smooth submanifold $$\phys:=\Big\{(q,p)
\Big|\bbeta(q)\partial_p H(q,p) =0\Big\}
\subset T^*Q$$ locally characterized by ( \[physleaf\]) is called the physical leaf.
$\phys$ contains all physical orbits of the system, that is, all smooth paths $\gamma:\R\rightarrow\phys\subset T^*Q$ that satisfy the differential-algebraic relations of theorem [ \[dae\]]{}.
Let $H$ be of the form ( \[hamham\]). Then, there exists a unique physical orbit $\gamma:\R^+\rightarrow\phys$ with $\gamma(0)=x$ for every $x\in \phys$.
We cover $\phys$ with local bundle charts of $T^*Q$ with coordinates $(q,p)$. For the Hamiltonian ( \[hamham\]), ( \[physleaf\]) reduces to $$\bbeta(q)\;g^{-1}(q)\;p\;=\;g^{-1}(q)\;\bbeta^\dagger(q)\;p\;=\;0\;,$$ where one uses the $g$-orthogonality of $\bbeta$. Hence, ( \[physleaf\]) is equivalent to $\bbeta^\dagger(q) p=0$. Since $\phys$ is the common zero level set of the $n$ component functions $(\bbeta^\dagger(q) p)_i$, every section $$X\;=\;v^r(q,p)\;\partial_{q^r}\;+\;w_s(q,p)\;\partial_{p_s}$$ of $T\phys$ is annihilated by the 1-forms $$d(\bbeta^\dagger p)_i\;=\;\partial_{q^r}(\bbeta^\dagger p)_i\;dq^r\;+\;
\partial_{p_s}(\bbeta^\dagger p)_i \;dp_s$$ for $i=1,\dots,n$ (of which only $n-k$ are linearly independent), on $\phys$.
This is expressed by 0&=& (v\^r\_[q\^r]{}) \^p + (w\_s\_[p\_s]{})\^p\
&=& (v\^r\_[q\^r]{}) \^p+\^w, which shows that the components $v$ of $X$ determine the projection $\bbeta^\dagger w$. Hence, the components $v$ and $\alph^\dagger w$ suffice to uniquely reconstruct $X$. Consequently, the right hand sides of ( \[qdotcon\]) and ( \[pdotcon\]) determine a unique section $X$ of $T\phys$, so that every curve $\gamma:\R^+\rightarrow\phys$, with arbitrary $\gamma(0)\in\phys$, that satisfies $\partial_t\gamma(t)=X(\gamma(t))$ automatically fulfills ( \[qdotcon\]) $\sim$ ( \[physleaf\]). This proves the assertion.
### Equilibria
The constrained Hamiltonian mechanical system $(Q,g,U,W)$ possesses \[pegasus\]\_Q :=qQ|\^(q) \_q U(q)=0as its critical set. An application of Sard’s theorem fully analogous to the proof of theorem [ \[Sardthm\]]{} shows that generically, this is a piecewise smooth, $n-k$-dimensional submanifold of $Q$, (recall that the rank of $\alph(q)$ is $k$).
### Symmetries
Let $G$ be a Lie group, and let $\psi:G \rightarrow {\rm Diff}(Q)$, $h \mapsto \psi_h$ with $\Psi_e = {\rm id}$, denote a group action. The constrained Hamiltonian mechanical system $(Q,g,U,W)$ is said to exhibit a $G$-symmetry if the following hold. (1) Invariance of the Riemannian metrics: $g\circ\psi_{h}=g$ and $g^*\circ\psi_{h}=g^*$ for all $h\in G$. (2) Invariance of the potential energy: $U\circ\psi_{h}=U$ for all $h\in G$. (3) Invariance of the distributions: $\psi_{h\,*}W=W$ and $\psi_h^* W^*=W^*$ for all $h\in G$.
Construction of the Auxiliary Extension
---------------------------------------
We are now prepared to embed the non-holonomic mechanical system into a constrained Hamiltonian system of the type considered in the previous sections.
To this end, we will introduce a set of generalized Dirac constraints over the symplectic manifold $(T^*Q,\omega_0)$ in the way presented in section [ \[gendir\]]{}. They define a symplectic distribution $V$, in a manner that the constrained Hamiltonian system $(T^*Q,\omega_0,H,V)$, with $H$ given by ( \[hamham\]), contains the constrained mechanical system as a dynamical subsystem. Thus, the auxiliary constrained Hamiltonian system $(T^*Q,\omega_0,H,V)$ extends the mechanical system in the sense announced in the introduction. An early inspiration for this construction stems from [@SoBr]. We require the following properties to be satisfied by $(T^*Q,\omega_0,H,V)$.
[.]{}
$\phys$ is an invariant manifold under the flow $\tPhi_t$ generated by ( \[eqsofmo\]).
All orbits $\tPhi(x)$ with initial conditions $x\in\phys$ satisfy the Euler-Lagrange equations of the Hölder principle.
$\phys$ is marginally stable under $\tPhi_t$.
The critical set $\crit$ of $\tPhi_t$ is a vector bundle over $\crit_Q$, hence equilibria of the constrained mechanical system are obtained from equilibria of the extension by base point projection.
Symmetries of the constrained mechanical system extend to those of $\tPhi_t$.
Let us briefly comment on (iii) $\sim$ (v). (iii) is of importance for numerical simulations of the mechanical system. (iv) makes it easy to extract information about the behaviour of the mechanical system from solutions of the auxiliary system. Condition (v) allows to apply reduction theory to the auxiliary system, in order to reduce the constrained mechanical system by a group action, if present. The choice for $V$ is by no means unique, and depending on the specific problem at hand, other conditions than (iii) $\sim$ (v) might be more useful.
### Construction of $V$
Guided by the above requirements, we shall now construct $V$.
To this end, we pick a smooth, $g^*$-orthonormal family of 1-forms $\lbrace \zeta_I\rbrace_{I=1}^{n-k}$ with $$\zeta_I\;=\;\zeta_{Ik}(q)\;dq^k\;,$$ so that locally, $$\langle\lbrace\zeta_1,\dots,\zeta_{n-k}\rbrace\rangle =
\left(W^*\right)^\perp\;.$$ The defining relationship $\bbeta^\dagger(q)p=0$ for $\phys$ is equivalent to the condition \[holcon\]f\_I(q,p):= g\_q\^\*(p,\_I(q))=0 I=1,…,n-k. It is clear that $f_I\in C^\infty(T^*Q)$.
1. To satisfy conditions (i) and (iii), we require that the level surfaces \[invman\] \_:=(q,p)|f\_I(q,p)=\_I; I=1,…,n-k , with $\underline{\mu}:=(\mu_1,\dots,\mu_{n-k})$, are integral manifolds of $V_{\rV }$. Here, $\rV $ denotes the degree of non-holonomy of $V$, and evidently, $\Mm_{\underline{0}}=\phys$.
Condition (iii) is satisfied because $$L(q,p)\;:=\;\sum_I\;\left|f_I(q,p)\right|^2$$ is an integral of motion for orbits of $\tPhi_t$. Since $L$ grows monotonically with increasing $|\underline{\mu}|$, and attains its (degenerate) minimum of value zero on $\phys$, it is a Lyapunov function for $\phys$. Anything better than marginal stability is prohibited by energy conservation.
2. To satisfy condition (ii), we demand that $\bbeta(q)\dot{q}=0$, or equivalently, that \[nonhcon\]\_I()=0, I=1,…,n-k, shall be satisfied along all orbits $(q(t),p(t))$ of ( \[eqsofmo\]), owing to ( \[qdotcon\]).
3. If the constrained mechanical system exhibits a $G$-symmetry, characterized by a group action $\psi:G\rightarrow {\rm Diff}(Q)$ so that $\psi_{h*}W=W\;\forall h\in G$, the local family of 1-forms $\lbrace\zeta_I\rbrace$ can be picked in a manner that $\psi_{h}^*\zeta_I=\zeta_I$ is satisfied for all $h\in G$ in a vicinity of the unit element $e$. Consequently, the functions $f_I(q,p)=h^*_q(\zeta_I,p)$ and their level sets $\Mm_{\underline{\mu}}$ are invariant under the group action.
The condition that ( \[invman\]) are integral manifolds of $V_{\rV }\supset V$ implies that all sections of $V$ are annihilated by the 1-forms $df_I$, for $I=1,\dots,n-k$. Furthermore, the condition ( \[nonhcon\]) requires $V$ to be annihilated by the 1-forms \[defxi\]\_I:=\_[Ir]{}(q)dq\^r+\_s0dp\_sthat are obtained from lifting $\zeta_I$ to $T^*(T^*Q)$, with $I=1,\dots,n-k$.
\[defV\] The distribution $$V\;:=\;\Big(\bigcap_I{\rm ker}\;df_I\Big)\;\;\;\bigcap\;\;\;
\Big(\bigcap_I{\rm ker}\;\xi_I\Big)\;\;\subset\;T(T^*Q)$$ is symplectic.
$V$ is symplectic iff its symplectic complement $V^\perp$ is. With the given data, the latter condition is more convenient to check. $V^\perp$ is locally spanned by the vector fields $(Y_1,\dots,Y_{2k})$ obtained from \[granite\]\_0(Y\_I,)=\_I(), (Y\_[I+k]{},)=d f\_I() ,where $I=1,\dots, k$, and $\omega_0=-dp_i\wedge dq^i$.
$V^\perp$ is symplectic if and only if $D:=\;[\omega(Y_I,Y_J)]$ has values in $GL_\R(2(n-k))$.
We remark that in the present notation, capital indices range from $1$ to $k$ if they label 1-forms, and from $1$ to $2k$ if they label vector fields.
In local bundle coordinates, $$\label{kabana}df_I \;=\; (\partial_{q^i}f_I)(q,p)\; dq^i \;\;+\;\;
\zeta_{Ii}(q)\;g^{ij}(q)\;dp_j \;,$$ where $g_{ij}$ are the components of the metric tensor $g$ on $Q$, as before. Let us introduce the functions $E(q):=[\zeta_{Ji}(q)]$ and $F(q,p):=[\partial_{q^j}f_K(q,p)]$, both with values in ${\rm Mat}_\R(n\times (n-k))$, which we use to assemble K:=(
[cc]{} E\^&0\
F\^&E\^g\^[-1]{}
): T\^\*Q\_(2(n-k)2n).Any component vector $v:T^*Q\rightarrow \R^{2n}$ that locally represents an element of $\Gamma(V)$ satisfies $K v=0$. The symplectic structure $\omega_0$ is locally represented by $J$, defined in ( \[sympJ\]). One can easily verify that the $I$-th row vector of the matrix $K \Jj^{-1}$ is the component vector of $Y_I$. In conclusion, introducing the matrices G(q)&:=&E\^(q) g\^[-1]{}(q)E(q)\
S(q,p)&:=& F\^(q,p) g\^[-1]{}(q)E(q)- E\^(q) g\^[-1]{}(q)F(q,p) ,one immediately arrives at D& =&KK\^= (
[cc]{} 0& G\
-G&S \[Dmat\]
).Since $\zeta_I$ has been picked a $g^*$-orthonormal family of 1-forms on $Q$, it is clear that $G(q)=\1_{n-k}$. Thus, $D$ is invertible. This proves that $V^\perp$ is symplectic.
### Construction of the projection tensors
Next, we determine the matrix of the $\omega_0$-orthogonal projection tensor $\PV$, which is associated to $V$, in the present bundle chart. Again, it is more convenient to carry out the construction for its complement first.
The matrix of the $\omega_0$-orthogonal projection tensor $\PVc$ associated to $V^\perp$ (considered as a tensor field that maps $\Gamma(T(T^*Q))$ to itself, with kernel $V$) is given by =(
[cc]{}&0\
T&\^
)in the local bundle chart $(q,p)$. The matrix $T=T(q,p)$ is defined in ( \[eqncs\]).
The proof of lemma [ \[Pconstr\]]{} can be used for this proof. The inverse of ( \[Dmat\]) is D\^[-1]{}=(
[cc]{} S &-\_[n-k]{}\
\_[n-k]{}&0
),where we recall that $G(q)=\1_{n-k}$. The $I$-th column vector of the matrix $K \Jj^{-1}$ is the component vector of $Y_I$ (we have required that $\lbrace Y_1,\dots Y_{2(n-k)}\rbrace$ spans $V^\perp$). This implies that $\PVc=\Jj K^\dagger D^{-1} K$.
The matrix of $\bbeta$ in the given chart is given by \[betaeq\](q)=g\^[-1]{}(q)E(q)E\^(q).
The construction presently carried out for $\PVc$ can also be applied to $\bbeta$. One simply replaces $V^\perp$ by $W^\perp$, and $\omega_0$ by the Riemannian metric $g$ on $Q$. An easy calculation immediately produces the asserted formula. The matrix of $\alph$ is subsequently obtained from $\alph+\bbeta=\1$. For more details, cf. [@Bra].
Introducing \[eqncs\] T(q,p):=E(q)F\^(q,p)(q)- \^(q)F(q,p) E\^(q) ,a straightforward calculation produces the asserted formula for $\PVc$.
In the given bundle coordinates, the matrix of $\PV$ is =(
[cc]{}&0\
-T&\^
), where $T=T(q,p)$ is defined in ( \[eqncs\]).
This is obtained from $\PV+\PVc =\1_{2n}$.
In this chart, $\PV(x)\Jj=\Jj\PV^\dagger(x)$, by $\omega_0$-skew orthogonality of $\PV$.
Let $H$ be as in ( \[hamham\]). Then, the dynamical system locally represented by \[salvation\](
[c]{}\
) =(
[cc]{}0&\
-\^&-T
) (
[c]{}\_q H\
\_p H
) , corresponding to the contrained Hamiltonian system $(T^*Q,\omega_0,H,V)$, is an extension of the constrained mechanical system $(Q,g,U,W)$.
By construction, $\phys$ is an invariant manifold of the associated flow $\tPhi_t$, hence ( \[physleaf\]) is fulfilled for all orbits of ( \[salvation\]) with initial conditions in $\phys$.
The equation $\dot{q}=\alph \partial_p H$ in ( \[salvation\]) obviously is ( \[qdotcon\]).
Next, using the notation $\underline{f}:=\left(f_1,\dots,f_{n-k}\right)^\dagger$, =E\^ g\^[-1]{}p,and substituting ( \[eqncs\]) for $T(q,p)$, the equation for $\dot{p}$ in ( \[salvation\]) becomes =-\^ \_q H-E F\^ +\^F .
Since $M_{\underline{\mu}}$ are invariant manifolds of the flow $\tPhi_t$ generated by ( \[salvation\]), $\partial_t f_I(q(t),p(t))$ vanishes along all orbits of ( \[salvation\]), so that $F^\dagger\dot{q}+E^\dagger g^{-1}\dot{p}=0$. This implies that =-\^\_q H+ E E\^ g\^[-1]{} +\^\_q (\^ ).\[dotpeq1\]Recalling that $\bbeta=g^{-1}E E^\dagger$ from ( \[betaeq\]), and using the fact that $\underline{f}=
\underline{0}$ on $\phys$, one arrives at ( \[pdotcon\]) by multiplication with $\alph^\dagger$ from the left.
### Equilibria of the extension
The critical set of the extension constructed above is characterized by the following theorem.
\[critsetextthm\] The critical set of ( \[salvation\]) is given by the vector bundle $$\crit\;=\;\bigcup_{q\in \crit_Q}\{q\}\times (W_q^*)^\perp
\; \;\;$$ with base space $\crit_Q$, cf. ( \[pegasus\]).
Let us first consider ( \[dotpeq1\]). As has been stated above, the second term on its right hand equals $\bbeta^\dagger(q)\dot{p}$, and moreover, from ( \[holcon\]), one concludes that $$\underline{f}^\dagger
\underline{f}\;=\; \|\bbeta^\dagger p \|_{g^*}^2\;.$$ The Hamiltonian ( \[hamham\]) can be decomposed into $$H(q,p)\;=\;H(q,\alph^\dagger p)\;+\;
\frac{1}{2}\|\bbeta^\dagger p\|_{g^*}^2\;,$$ due to the $g^*$-orthogonality of $\alph^\dagger$ and $\bbeta^\dagger$, so that ( \[dotpeq1\]) can be written as $$\dot{p}\;=\;-\;\alph^\dagger \partial_q H(q,\alph^\dagger p)\;+\;\bbeta^\dagger\dot{p}\;.$$
The equilibria of ( \[salvation\]) are therefore determined by the conditions $$\alph^\dagger(q)p\;=\;0\;\;\;,\;\;\;\alph^\dagger(q)\;
\partial_q H(q,\alph^\dagger p)\;=\;0\; .$$ Because $H$ depends quadratically on $\alph^\dagger p$, the second condition can be reduced to $$\alph^\dagger(q)\;\partial_q U(q)\;=\;0$$ using the first condition. Comparing this with ( \[pegasus\]), the assertion follows.
In particular, this fact implies that every equilibrium $(q_0,p_0)$ of the extension defines a unique equilibrium $q_0$ on $\crit_Q$ by base point projection.
To analyze the stability of a given equilibrium solution $q_0\in \crit_Q$, it is necessary to determine the spectrum of the linearization of $X_H^V$ at $\xa=(q_0,0)$.
A straightforward calculation along the lines of the previous discussion shows that in the present bundle chart, \[DXHVlinexpl\]DX\_H\^V()=(
[cc]{}0& g\^[-1]{}\^\
-\^D\^2\_[q\_0]{} U-R&0
)() , where \[corrosion\]:= \_(nn) .Furthermore, $D^2_{q_0} U$ is the matrix of second derivatives of $U$. The stability discussion in the previous section can now straightforwardly be applied to $DX_H^V(\xa)$.
### Extension of symmetries
Let us assume that the constrained mechanical system $(Q,g,U,W)$ exhibits a $G$-symmetry $\psi:G\rightarrow {\rm Diff}(Q)$. Then, we claim that it is extended by $(T^*Q,\omega_0,H,V)$. To this end, we recall that the 1-forms $\zeta_I$ satisfy $\psi^*_h \zeta_I$ for all $h\in G$ close to the unit.
Via its pullback, $\psi$ induces the group action :=\^\*:GT\^\*Q&& T\^\*Qon $T^*Q$. This group action is symplectic, that is, $\Psi_h^*\omega_0=\omega_0$ for all $h\in G$. For a proof, consider for instance [@AbMa].
The 1-forms $\xi_I$, defined in ( \[defxi\]), satisfy $\Psi_h^*\xi_I=\xi_I$, and likewise, $f_I\circ\Psi_h=f_I$ is satisfied for all $h\in G$ close to the unit. The definition of $V$ in proposition [ \[defV\]]{} thus implies that $$\Psi_{h\,*}V\;=\;V$$ is satisfied for all $h\in G$. Due to the fact that $\omega$ and $V$ are both $G$-invariant, $\PV$ and $\PVc$ are also invariant under the $G$-action $\Psi$.
The Hamiltonian $H$ in ( \[hamham\]) is $G$-invariant under $\Psi$, by assumption on the constrained Hamiltonian mechanical system. Thus, $X_H$ fulfills $\Psi_{h*}X_H=X_H$ for all $h\in G$, which implies that $X_H^V=\PV (X_H)$ is $G$-invariant.
The Topology of the Critical Manifold
-------------------------------------
Since $\crit$ is not a compact submanifold of $T^*Q$, our previous results cannot be applied directly. However, owing to the vector bundle structure of $\crit$ and $T^*Q$, the result \[eqnhd\] \_[i,p]{} \^[p+\_i]{} [dim]{}H\^p\_c(\_i) = \_[p]{} \^p [dim]{}H\^p\_c(T\^\*Q) +(1+) () still holds, where $H^*_c$ denotes the de Rham cohomology based on differential forms with compact supports. The polynomial $\Q(t)$ has non-negative integer coefficients.
In a first step, the arguments of section [ \[sectionII\]]{} can be straightforwardly applied to $\crit_Q$. $\crit_Q$ is normal hyperbolic with respect to the gradient-like flow $\psi_t$ generated by $$\partial_t q(t)=-\alph(q(t))\nabla_g U(q(t)) ,$$ it contains all critical points of the Morse function $U$, but no other conditional extrema of $U|_{\crit_Q}$ apart from those (it is gradient-like because along all of its non-constant orbits, $\frac{d}{dt}U(t)=-g(\alph\nabla_g
U,\alph\nabla_g U)|_{q(t)}<0$ holds, since $\alph$ is an orthogonal projection tensor with respect to the Riemannian metric $g$ on $Q$). This can be proved by substituting $M\rightarrow Q$, $H\rightarrow U$, $\PV\rightarrow\alph$, $g_{({\rm Kahler})}\rightarrow g$, and $\crit\rightarrow\crit_Q$ in section [ \[sectionII\]]{}, and by applying the arguments used there. Hence, letting $\mu_i$ denote the index of the connectivity component $\crit_{Qi}$ of $\crit_Q$, ( \[Conley-Zehnder1\]) implies that for compact, closed $Q$, \[eqnhdbum\] \_[i,p]{} \^[p+\_i]{} [dim]{}H\^p(\_[Qi]{}) = \_[p]{} \^p [dim]{}H\^p(Q) +(1+) () , where $\Q(t)$ is a polynomial with non-negative integer coefficients.
$\crit_Q$, being the zero section of $\crit$, is a deformation retract of $\crit$, and likewise, $Q$ is a deformation retract of $T^*Q$. Thus, ( \[eqnhd\]) follows trivially from the invariance of the de Rham cohomology groups under retraction, $H^p_c(\crit_i)\cong H^p(\crit_{Qi})$, $H^p_c(T^*Q)\cong H^p(Q)$. Hence, ( \[eqnhd\]) is equivalent to \[MorseBottCZQcase\] \_[i,p]{} \^[p+\_i]{} b\_p(\_[Qi]{}) = \_[p]{} \^p b\_p(Q) +(1+) (),where $b_p$ is the $p$-th Betti number.
Consequently, one finds $\sum_{i} b_{p-\mu_i}(\crit_{Qi}) \geq b_{p}$, and in particular, for $\lambda=-1$, one obtains $$\sum_{i,p} (-1)^{p+\mu_i} b_p(\crit_{Qi}) =
\sum_{i} (-1)^{\mu_i} \chi(\crit_{Qi}) =\chi(Q) ,$$ where $\chi$ denotes the Euler characteristic.
Applications, Illustrations and Examples
========================================
Let us conclude our analysis with the discussion of some simple applications and examples.
A Computational Application
---------------------------
Let us first formulate an application of our analysis for the computational problem of finding the equilibria in a large constrained multibody system. It is in this context also desirable to determine whether a given set of parameters and constraints implies the existence of non-generic critical points. This is due to the circumstance that in practice, manufacturing imprecisions can have a significant effect on the latter.
For large multibody systems, equilibria can realistically only be determined by numerical routines. The strategy presented in chapters [ \[sectionII\]]{} and [ \[sectionIV\]]{} suggests the following method.
If $U$ is a Morse function whose critical points are known, and if $Q$ is compact and closed, it is possible to numerically construct all generic connectivity components of $\crit_Q$. This is because generic components of $\crit_Q$ are smooth, $n-k$-dimensional submanifolds of $Q$ containing all critical points of $U$, and no other critical points of $U|_{\crit_Q}$. This information can be exploited to find sufficiently many points on $\crit_Q$, so that a suitable interpolation routine enables the approximate reconstruction of an entire connectivity component. To this end, one chooses a vicinity of a critical point $a$ of $U$, and uses a fixed point solver to determine neighboring zeros of $|\alph(q)\nabla_g U(q)|^2$, which are elements of $\crit_Q$ close to $a$. Iterating this procedure with the critical points found in this manner, pieces of $\crit_Q$ of arbitrary size can be determined.
If all critical points of $U$ are a priori known, one can proceed in this manner to construct all connectivity components of $\crit_Q$ that contain critical points of $U$. Then, one is guaranteed to have found all of the generic components of $\crit$ if the numerically determined connectivity components are closed, compact, and contain all critical points of $U$.
We remark that determining the critical points of a Morse function $U:Q\rightarrow\R$ is a difficult numerical task by itself. Attempting to find critical points by simulating the gradient flow generated by $-\nabla_g U$ is time costly, because the critical points define a thin set in $M$. Their existence, however, is of course ensured by the topology of $Q$.
Another remark is that all critical points $a$ at which $D(\alph \nabla_g U)(a)$ has a reduced rank, are elements of the non-generic part of $\crit_Q$. Thus, the latter condition is an indicator for non-genericity. If there are such exceptional critical points in a technically relevant region of $Q$, they can be removed by a small local modification of the system parameters or constraints.
A disc in a periodic potential, sliding on the plane
----------------------------------------------------
Let us consider a mechanical example, consisting of a thin disk of radius $r$ and mass $m$ on the plane $\R^2$, which is attached to a massless skate. The connecting line between the center of the disc and the contact point at the center of the skate with the plane is normal to the plane, precisely if the disc is horizontal. We assume that the disk remains horizontal during its motion, and that the translational motion of the disc is only possible in the direction of the skate.
Let $(x_1,x_2)$ denote the position of the center of mass of the disc with respect to some Euclidean coordinate system on $\R^2$, and let $\phi$ denote the angle enclosed by the skate and the $x_1$-axis.
The kinetic energy of this system is given by $$T=\frac{m }{2}(\dot{x}_1^2+\dot{x}_2^2)
+\frac{1}{2}\frac{mr^2}{2}
\dot{\phi}^2 \; ,$$ which defines a Riemannian metric on $TQ$ with metric tensor $$[g_{ij}(\phi,\theta,\psi)]=
\left(\begin{array}{ccc}m &0&0\\
0&m &0
\\0&0&\frac{mr^2}{2}
\end{array}\right) .$$ Furthermore, we assume that it moves against the background of a $(2\pi\Z)^3$-periodic potential energy $$U(x_1,x_2,\phi)=\sum_{i=1,2} c_i (1-\cos x_i)+ c_\phi (1-\cos\phi) \; ,$$ where $c_1$, $c_2$, and $c_\phi$ are coupling constants.
Dividing out the translational symmetry with respect to $(2\pi\Z)^3$, the configuration manifold of this mechanical system is given by $Q=[0,2\pi]^3\cong T^3$ (periodic boundary conditions). Clearly, $U:T^3\rightarrow \R$ is a real analytic Morse function, with 8 critical points in the corners of $[0,\pi]^3$, while each of the remaining critical points in $[0,2\pi]^3$ is identified with one of the former by periodicity. Correspondingly, we will from here on consider $(x_1,x_2)$ as coordinates on $T^2$, that is, mod $2\pi$.
The requirement that the disk shall slide in the direction of the skate is expressed by the non-holonomic constraint $$\dot{x}_1\sin\phi- \dot{x}_2\cos\phi=0 .$$ The matrix $E^\dagger(x_1,x_2,\phi)$, introduced in the proof of theorem [ \[defV\]]{}, thus corresponds to $$E^\dagger(x_1,x_2,\phi)=(\sin\phi,-\cos\phi,0)\;,$$ so that $E^\dagger g^{-1}E=\frac{1}{m}$.
The orthoprojectors $\bbeta$ and $\alph$ are thus straightforwardly obtained as (x\_1,x\_2,)&=&(
[ccc]{} \^2&-&0\
-&\^2&0\
0&0&0
)\
(x\_1,x\_2,)&=&(
[ccc]{} \^2&&0\
&\^2&0\
0&0&1
) .The critical set is given by $$\crit_Q=\Big\{(x_1,x_2,\phi)\Big|\big(\alph^\dagger \nabla U\big)(x_1,x_2,\phi)
=0\Big\}$$ (where $\nabla:=(\partial_{x_1},\partial_{x_2},\partial_\phi)$). Let $$\crit_{a,b}:=\Big\{(x_1,x_2,\phi)\Big|x_1=a \, , \,
x_2\in[0,2\pi]\, , \, \phi=b\Big\}\;.$$ Then, $$\crit_Q=\bigcup_{a,b\in\{0,\pi\}}\crit_{a,b} \;.$$ It is trivially clear that $\crit_Q$ contains all critical points of $U$. Let $q_c\in\crit_{a,b}$, where $a,b\in\{0,\pi\}$. Noting that $\alph={\rm diag}(1,0,1)$ on $\crit_Q$, we have \[icestorm\] ((\^U))(q\_c) &=&(\^( U)) (q\_c)+(q\_c)\
&=& (
[ccc]{} c\_1a&0&c\_2x\_2\
0&0&0\
0&0&c\_b
) . Clearly, $${\rm spec}\Big(\Big(\nabla\otimes(\alph^\dagger \nabla U)\Big)(q_c)\Big)
=\Big\{0,c_1\cos a,c_\phi \cos b\Big\} \;,$$ which is, for each fixed $a,b$, independent of $x_2$. Thus, the indices of the connectivity components $\crit_{a,b}$ with respect to the gradient-like flow generated by $-\alph^\dagger \nabla U$ are given by (\_[0,0]{})=2 , (\_[0,]{})=(\_[,0]{})=1 , (\_[,]{})=0 , and clearly, $\crit_{a,b}\cong S^1$ for all $a,b\in\{0,\pi\}$. Since the Betti numbers of $T^3$ are given by $b_0=b_3=1$, $b_1=b_2=3$, and those of $\crit_{a,b}$ by $b_0(\crit_{a,b})=b_1(\crit_{a,b})=1$, $b_2(\crit_{a,b})=b_3(\crit_{a,b})=0$, one finds that $$\sum_{a,b} b_{p-\mu(\crit_{a,b})}(\crit_{a,b})=b_p(Q)$$ for $p=0,\dots,3$, or explicitly, b\_[3-2]{}(\_[0,0]{})&=&1=b\_3(T\^3)\
b\_[2-2]{}(\_[0,0]{})+b\_[2-1]{}(\_[0,]{})+b\_[2-1]{}(\_[,0]{})&=&3 =b\_2(T\^3)\
b\_[1-0]{}(\_[,]{})+b\_[1-1]{}(\_[0,]{})+b\_[1-1]{}(\_[,0]{}) &=&3=b\_1(T\^3)\
b\_[0-0]{}(\_[,]{})&=&1=b\_0(T\^3), in agreement with ( \[MorseBottCZQcase\]).
Next, we determine the spectrum of the linearization of $X_H^V$ at $(q_c,0)\in T^*Q$, cf. ( \[DXHVlinexpl\]). To this end, $$\Big(\alphc g^{-1}\alphc^\dagger\Big)(q_c)
=\left(\begin{array}{ccc}
\frac1m&0&0\\0&0&0
\\0&0&\frac{2}{mr^2}
\end{array}\right) ,$$ and multiplying this matrix from the right with ( \[icestorm\]) yields $$\Omega(q_c,\theta):=\left(\begin{array}{ccc}
\frac{c_1\cos a}{m}&0&
\frac{c_2 \sin x_2}{m}
\\0&0&0\\0&0& \frac{2 c_\phi \cos b}{mr^2}
\end{array}\right) .$$ Clearly, $${\rm spec}\Big(\Omega(q_c,\theta)\Big) =
\Big\{0,\frac{c_1\cos a}{m},\frac{2 c_\phi \cos b}{mr^2}\Big\} \;.$$ From ( \[DXHVlinexpl\]), it is easy to see that $${\rm spec}\Big(DX_H^V(q_c,0)\Big)=
\Big\{0,\pm\sqrt{\frac{c_1\cos a}{m}},
\pm\sqrt{\frac{2 c_\phi \cos b}{mr^2}}\Big\}\;,$$ hence critical stability occurs for the case $a=b=\pi$, while in all other cases, there is an asymptotically unstable direction.
We conclude that all components $\crit_{a,b}$, where $a+b\leq \pi$, are unstable. In the critically stable case $a=b=\pi$, the linear problem is oscillatory, and the eigenfrequencies are given by $\sqrt{\frac{c_1}{m}}$ and $\sqrt{\frac{2 c_\phi}{mr^2}}$, independently of $x_2$. Since $\mu(\crit_{\pi,\pi})=0$, our discussion in section [ \[sectionIII\]]{} suggests that the connectivity component $\crit_{\pi,\pi}$ of $\crit_Q$ is stable in the sense of Nekhoroshev if the ratio $\sqrt{\frac{c_1 mr^2}{2c_\phi}}$ is irrational.
**Acknowledgements** {#acknowledgements .unnumbered}
--------------------
This work is based on the thesis [@Ch], which was carried out at the center of mechanics (IMES), ETH Zürich. I warmly thank Prof. H. Brauchli for suggesting this area of problems, for his insights, and for the possibility to carry out this work. I am profoundly grateful to Prof. E. Zehnder for his generosity, and discussions that were most enlightening and helpful. It is a pleasure to thank M. von Wattenwyl, M. Sofer, H. Yoshimura, O. O’Reilly, and especially M. Clerici, for highly interesting discussions. I also thank the referee for his helpful suggestions. The author is supported by a Courant Instructorship.
[99]{}
R. Abraham, J. E. Marsden, ’Foundations of mechanics’, Benjamin/Cummings, (1978).
V. I. Arnol’d, ’Mathematical methods of classical mechanics’, Second Edition, Graduate Texts in Mathematics [**60**]{}, Springer Verlag (1989).
V. I. Arnol’d, ’Dynamical systems III’, Encyclopedia of Mathematics [**3**]{}, Springer Verlag (1988).
D. M. Austin, P. J. Braam, ’Morse-Bott theory and equivariant cohomology’, in ’The Floer memorial volume’, eds. H. Hofer, C. H. Taubes, A. Weinstein, E. Zehnder, Birkhäuser Verlag (1995).
A. Bellaiche, J.-J. Risler (eds.), ’Sub-Riemannian geometry’, Birkhäuser Verlag (1996).
J. - M. Bismut, ’The Witten complex and the degenerate Morse inequalities’, J. Diff. Geom., [**23**]{}, 207 - 240 (1986).
A.M. Bloch, P.S. Krisnaprasad, J.E. Marsden, R.M. Murray, ’Non-holonomic mechanical systems with symmetry’, Arch. Rational Mech. Anal., [**136**]{}, 21-99, (1996).
R. Bott, ’Nondegenerate critical manifolds’, Ann. Math. [**60**]{}, No. 2, 248 - 261, (1954).
R. Bott, ’Morse theory indomitable’, Publications mathématiques, [**68**]{}, 99 - 114 (1989).
R. Bott, ’Lectures on characteristic classes and foliatons’, in R. Bott, S. Gitler, I.M. James, ’Lectures on algebraic topology’, Lecture notes in mathematics, [**279**]{}, Springer Verlag (1972).
H. Brauchli, ’Mass-orthogonal formulation of equations of motion for multibody systems’, J. Appl. Math. Phys. (ZAMP), [**42**]{} , 169 - 182 (1991).
H. Brauchli, ’Efficient description and geometrical interpretation of the dynamics of constrained systems’, Computational Methods in Mechanical Systems ’97, J. Angeles, E. Zakhariev (eds.), Springer Verlag Berlin Heidelberg (1998).
F. Cardin, M. Favretti, ’On non-holonomic and vakonomic dynamics of mechanical systems with nonintegrable constraints’, J. Geom. Phys., [**18**]{}, 295 - 325 (1996).
T. Chen, ’Non-holonomy, critical manifolds and stability in constrained Hamiltonian systems’, ETH-Dissertation 13017 (1999).
C. Conley, E. Zehnder, ’Morse type index theory for flows and periodic solutions of Hamiltonian equations’, Comm. Pure Appl. Math., [**37**]{}, 207 - 253 (1984).
B. A. Dubrovin, A. T. Fomenko, S. P. Novikov, ’Modern geometry - methods and applications’, Vol. III, Springer Verlag, (1985).
A. Floer, ’Witten’s complex and infinite dimensional Morse theory’, J. Diff. Geom., [**30**]{}, 207 - 221 (1989).
Z. Ge, ’Betti numbers, characteristic classes and sub-Riemannian geometry’, Illinois J. Math., [**36**]{}, No. 3, 372 - 403 (1992).
M. Gromov, ’Carnot - Caratheodory spaces seen from within’, in ’Sub-Riemannian geometry’, eds. A. Bellaiche, J.-J. Risler, Birkhäuser Verlag, (1996).
M. W. Hirsch, ’Differential topology’, Springer Verlag New York, (1976).
H. Hofer, E. Zehnder, ’Symplectic invariants and Hamiltonian dynamics’, Birkhäuser Verlag, (1994).
J. Jost, ’ Riemannian geometry and geometric analysis’, Springer Verlag Berlin Heidelberg, (1995).
W. S. Koon, J. E. Marsden, ’The Hamiltonian and Lagrangian approaches to the dynamics of non-holonomic systems’, Rep. Math. Phys., [**40**]{}, 21-62 (1997).
W. S. Koon, J. E. Marsden, ’The Poisson reduction of nonholonomic mechanical systems’, Reports on Math. Phys., [**42**]{}, 101-134 (1998).
J. E. Marsden, T. Ratiu, ’Introduction to mechanics and symmetry’, Springer Verlag New York (1994).
D. McDuff, D. Salamon, ’Introduction to symplectic topology’, Clarendon Press (1995).
J. Milnor, ’Morse theory’, Princeton Univ. Press, Princeton, N. J. (1963).
J. Milnor, ’Topology from the differentiable viewpoint’, Princeton Landmarks in Mathematics, Princeton University Press (1997).
M. Schwarz, ’Morse Homology’, Birkhäuser Verlag (1993).
S. Smale, ’Morse inequalities for a dynamical system’, Bull. Amer. Math. Soc., [**66**]{}, 43 - 49 (1960).
S. Smale, ’On gradient dynamical systems’, Ann. Math., [**74**]{}, No. 1, 199 - 206 (1961).
M. Sofer, O. Melliger, H. Brauchli, ’Numerical behaviour of different formulations for multibody dynamics’, Numerical Methods in Engineering ’92, Ch. Hirsch et al (eds.), Elsevier Science Publishers, Amsterdam (1992).
E. Spanier, ’Algebraic Topology’, Springer Verlag New York (1966).
R. Strichartz, ’Sub-Riemannian geometry’, J. Diff. Geom., [**24**]{}, 221 - 261 (1986).
J. Van der Schaft, B.M. Maschke, ’On the Hamiltonian formulation of non-holonomic mechanical systems’, Rep. Math. Phys., [**34**]{}, 225-233 (1994).
E. Witten, ’Supersymmetry and Morse theory’, J. Diff. Geom. [**17**]{}, 661 - 692, (1982).
R. W. Weber, ’Hamiltonian systems with constraints and their meaning in mechanics’, Arch. Rat. Mech. Anal. [**91**]{}, 309 - 335 (1986).
H. Yoshimura, T. Kawase, ’A duality principle in non-holonomic mechanical systems’, Nonconvex Optim. Appl., [**50**]{}, 447–471 (2001).
E. Zehnder, ’The Arnold conjecture for fixed points of symplectic mappings and periodic solutions of Hamiltonian systems’, Proceedings of the International Congress of Mathematicians, Berkeley, California, USA (1986).
D. V. Zenkov, A. M. Bloch, J. E. Marsden, ’The energy-momentum method for the stability of non-holonomic systems’, Dyn. Stab. of Systems, [**13**]{}, 123-166 (1998).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Erez Berg
- Eduardo Fradkin
- 'Steven A. Kivelson'
title: Charge $4e$ superconductivity from pair density wave order in certain high temperature superconductors
---
**A number of spectacular experimental anomalies[@li-2007; @fujita-2005] have recently been discovered in certain cuprates, notably [La$_{2-x}$Ba$_x$CuO$_4$]{} and [La$_{1.6-x}$Nd$%
_{0.4}$Sr$_x$CuO$_{4}$]{}, which exhibit unidirectional spin and charge order (known as “stripe order”). We have recently proposed to interpret these observations as evidence for a novel “striped superconducting” state, in which the superconducting order parameter is modulated in space, such that its average is precisely zero. Here, we show that thermal melting of the striped superconducting state can lead to a number of unusual phases, of which the most novel is a charge $4e$ superconducting state, with a corresponding fractional flux quantum $hc/4e$. These are never-before observed states of matter, and ones, moreover, that cannot arise from the conventional Bardeen-Cooper-Schrieffer (BCS) mechanism. Thus, direct confirmation of their existence, even in a small subset of the cuprates, could have much broader implications for our understanding of high temperature superconductivity. We propose experiments to observe fractional flux quantization, which thereby could confirm the existence of these states.**
It is widely accepted that the mechanism of superconductivity in the cuprate high temperature superconductors is different than in conventional superconductors, and that the “normal state” above $T_c$ is anything but normal. In contrast, the low energy low temperature properties of the superconducting state appears to be largely accounted for by the conventional BCS theory, albeit with a d-wave order parameter and a much suppressed superfluid density. In short, the perception is that the superconducting (SC) phase is simple while the “normal” phase is not. However, recent experiments [@li-2007; @fujita-2005] in the “stripe ordered” materials, in which a novel form of highly two-dimensional partial SC order was found, suggest that the SC phases may have unconventional aspects. We have interpreted[@berg-2007; @berg-2008a; @berg-2009] these anomalies as evidence of a new type of order, “striped superconductivity” or unidirectional pair density wave (PDW) order. Subsequent experiments[tranquada08,basov08]{} have provided additional indirect evidence in favor of this interpretation.
In the present paper, we study the thermal melting of a striped superconducting groundstate by the proliferation of topological defects leads to a complex phase diagram, shown in Fig.\[fig:phase-diagram\], with several interesting phases, of which the most novel is a charge $4e$ superconducting state. The existence of such a state can be directly established experimentally by observing magnetic flux quantization with period $hc/4e$, half that of the usual superconducting flux quantum, in a variety of experimental geometries, as shown in Fig.[fig:squid]{}. Moreover, it is likely that the charge 4e SC, resulting from partial melting the striped SC, has a finite density of states for gapless quasiparticles, reflecting the fact that it does not arise from the Bose condensation of literal four electron bound states.
In related developments, Agterberg and Tsunetsugu[@agterberg08; @agterberg-2009] and Radzihovsky and Vishwanath, [@Radzihovsky-2008] have discussed related phenomena in connection with Larkin-Ovchinnikov (LO) phases in, respectively, the heavy fermion material CeCoIn$_5$ at high fields and in partially spin-polarized ultra-cold atomic gases. The states we discuss here differ from the LO states in that there is no explicit time-reversal symmetry breaking. Also the structure of the striped SC reflects the discrete point group symmetry of the underlying (cuprate) lattice.
A striped superconductor is a unidirectional PDW state in which a spin singlet superconducting order parameter, $\Delta(\mathbf{r})$ oscillates in space with a wave vector $\mathbf{Q}$,[@commensuration] (shown schematically in Fig. \[fig:phase-diagram\]): $$\Delta(\mathbf{r})=\Delta_{\mathbf{Q}}(\mathbf{r}) \, e^{i \mathbf{Q} \cdot
\mathbf{r}}+ \Delta_{-\mathbf{Q}}(\mathbf{r}) \, e^{-i \mathbf{Q} \cdot
\mathbf{r}}. \label{eq:op-pdw}$$ Of course, there will always be higher harmonics of the order parameter at wave vectors $n\vec Q$, but in the ordered state, these are slaved to the fundamentals and hence are not independent dynamical degrees of freedom. Generally, therefore, we will not treat them explicitly. However, two subsidiary order parameters play a special role in the thermal melting: Charge density wave (CDW) order, $\rho(\mathbf{r})=\rho_0+\rho_{\mathbf{K}}(%
\mathbf{r}) \, e^{i \mathbf{K} \cdot \mathbf{r}}+ \rho^\star_{\mathbf{K}}(%
\mathbf{r}) \, e^{-i \mathbf{K} \cdot \mathbf{r}}$, appears as a second harmonic of the fundamental ordering, with wave vector $\mathbf{K}=2\mathbf{Q%
}$, where $\rho_{\mathbf{K}}(\mathbf{r}) \propto \Delta^\star_{-\mathbf{Q}}(%
\mathbf{r}) \Delta_{\mathbf{Q}}(\mathbf{r})$. Similarly, charge $4e$ SC order, represented by the complex scalar field $\Delta_{4e}(\mathbf{r})
\propto \Delta_{-\mathbf{Q}}(\mathbf{r}) \Delta_{\mathbf{Q}}(\mathbf{r})$, occurs parasitically along with the fundamental striped SC order.
Although the stripe ordered superconductor need not be associated with a non-zero magnetization, it is in many ways similar to a LO state.[Larkin-1964]{} As such it has a $U(1) \times U(1)$ symmetry corresponding to the independent uniform shifts of the phases of the complex components of the order parameter: $$\begin{aligned}
\Delta_{\pm \mathbf{Q}}(\mathbf{r}) && \to e^{i\theta_{\pm \mathbf{Q}}}
\Delta_{\pm \mathbf{Q}}(\mathbf{r}) =e^{i(\theta\pm \phi)} \Delta_{\pm
\mathbf{Q}}(\mathbf{r}), \\
\rho_{\mathbf{K}}(\mathbf{r}) && \to e^{i2\phi} \rho_{\mathbf{K}}(\mathbf{r}%
)\ \ \mathrm{with } \ \ \phi\equiv (\theta_{\mathbf{Q}}-\theta_{-\mathbf{Q}%
})/2 \notag \\
\Delta_{4e}(\mathbf{r}) &&\to e^{i 2\theta}\Delta_{4e}(\mathbf{r}) \ \
\mathrm{with } \ \ \theta\equiv (\theta_{\mathbf{Q}}+\theta_{-\mathbf{Q}})/ 2
\notag \label{eq:symmetries}\end{aligned}$$ Much as other stripe electronic liquid crystal states, the PDW also typically breaks the point group symmetry; to be concrete, we will consider the case of a tetragonal crystal, in which the choice of direction for the unidirectional order breaks the $C_{4v}$ point group symmetry down to $C_2$. In analogy with classical liquid crystals, we refer to this as nematic ordering, although since the symmetry breaking is in fact, discrete, it is an “Ising nematic.” This is the minimal set of broken symmetries associated with the striped superconductor. For simplicity, we will consider only the interplay of charge stripe and SC stripe orders, and ignore the striped spin order, even though all three are intertwined where striped superconductivity is conjectured to occur in the cuprates.[@berg-2007; @berg-2008a; @berg-2009]
We consider a system which has a striped superconducting groundstate, and consider the manner in which thermal fluctuations gradually restore all the broken symmetries as the temperature, $T$, increases. Deep in the PDW phase, other than in a vortex core, we can ignore fluctuations in the magnitude of the order parameter, and write $\Delta_{\pm \mathbf{Q}}(\mathbf{r}%
)=\Delta_{SC}\, \exp\big\{ i\left(\theta(\mathbf{r})\pm \phi (\mathbf{r}%
)\right)\big\}$, $\rho_{\mathbf{K}}(\mathbf{r})=\rho_{K} \exp\left[i2\phi(%
\mathbf{r})\right]$, and $\Delta_{4e}(\mathbf{r})=\Delta_{4e}\exp\left[%
i2\theta(\mathbf{r})\right]$. Because the phase of the CDW order is subject to pinning by quenched disorder, the PDW is considerably more fragile than a uniform SC state, and quenched randomness easily leads to an XY superconducting glass phase. (Unless stated otherwise, below we ignore the effects of disorder.) Moreover, the striped superconductor state admits novel topological excitations, which play a central role in the thermal melting of the groundstate.
Deep in the striped SC phase, treating the system as 2D, and taking $\vec Q$ in the $x$ direction, the effective Hamiltonian for the low energy thermal fluctuations is $$\mathcal{H}[\theta,\varphi]= \frac{\rho_s}{2} \left|\boldsymbol{D_s}%
\theta\right|^2+ \frac{\kappa}{2} \left(\boldsymbol{D_c} \varphi\right)^2,
\label{eq:london}$$ $\boldsymbol{D_s} = \alpha_s[-i\hbar \partial_x -(2e/c) A_x]\hat x
+ \alpha_s^{-1}[-i\hbar \partial_y -(2e/c) A_y]\hat y$, $\boldsymbol{D_c} = \alpha_c[-i\hbar \partial_x ]\hat x +
\alpha_c^{-1}[-i\hbar \partial_y]\hat y $, $\rho_s$ and $\kappa$ are, respectively, the superfluid stiffness and the CDW elastic constant, and $\alpha_s$ and $\alpha_c$ are the corresponding (finite) anisotropies.[@continuum] We have shown, explicitly, the coupling to an external gauge field, but henceforth, unless otherwise specified, we will take $\mathbf{A}=\mathbf{0}$ [@comment-A]. The correlation functions of the PDW, CDW and charge $4e$ SC order parameters in this phase are $$\begin{aligned}
&& \langle \Delta_{\pm \boldsymbol{Q}}(\mathbf{r}) \Delta_{\pm \boldsymbol{Q}%
}(\mathbf{r}^\prime)^*\rangle \propto K_s(\mathbf{r} -\mathbf{r}^\prime) K_c(%
\mathbf{r} -\mathbf{r}^\prime) \notag \\
&& \langle \rho_{ 2\boldsymbol{Q}}(\mathbf{r}) \rho_{2 \boldsymbol{Q}}(%
\mathbf{r}^\prime)^*\rangle \propto [K_c(\mathbf{r} -\mathbf{r}^\prime)]^4
\notag \\
&& \langle \Delta_{4e}(\mathbf{r}) \Delta_{4e}(\mathbf{r}^\prime)^*\rangle
\propto [K_s(\mathbf{r} -\mathbf{r}^\prime) ]^4 \label{eq:qlro}\end{aligned}$$ where $$\begin{aligned}
&& K_s(\mathbf{r}) \sim\big[(x\alpha_s)^2+(\alpha_s^{-1} y)^2\big]^{-\frac{%
\eta_s}{2}} \notag \\
&& K_c(\mathbf{r}) \sim \big[(x\alpha_c)^2+(\alpha_c^{-1} y)^2\big]^{-\frac{%
\eta_c}{2}} \notag \\
&& \eta_s= 2\pi T/\rho_s \ \ \ \eta_c=2\pi T/\kappa.\end{aligned}$$
Since the order parameters $\Delta_{\pm \mathbf{Q}}(\mathbf{r})$ must be separately single valued, their phase fields must be invariant under the transformations $\theta_{\pm \mathbf{Q}}(\mathbf{r}) \to \theta_{\pm \mathbf{%
Q}}(\mathbf{r}) +2\pi m_{\pm \mathbf{Q}}$, where $m_{\pm \mathbf{Q}}$ are integers. Correspondingly, the fields $\theta(\mathbf{r})$ and $\varphi(%
\mathbf{r})$ must obey the conditions $\theta(\mathbf{r}) \to \theta(\mathbf{%
r})+\pi (m_{ \mathbf{Q}}+ m_{- \mathbf{Q}})$, $\varphi(\mathbf{r}) \to
\varphi(\mathbf{r}) + \pi (m_{ \mathbf{Q}}-m_{-\mathbf{Q}})$. The integers $%
m_{\pm \mathbf{Q}}$ then classify the topological excitations supported by the PDW state: vortices with topological charge $q_s=(m_{ \mathbf{Q}}+ m_{-
\mathbf{Q}})/2$ and dislocations with topological charge $q_c=(m_{ \mathbf{Q}%
}- m_{- \mathbf{Q}})/2$. We have three types of topological excitations $%
(q_s,q_c)$:
1. Full vortices with $q_s =\pm 1$ and $q_c=0$.
2. Double dislocations, with $q_c=\pm 1$ and $q_s=0$.
3. Half-vortices, $q_s=\pm 1/2$, bound to single dislocations, $q_c =\pm
1/2$.
Much as in the case of the well understood theory of the Kosterlitz-Thouless (KT) phase transition,[@kosterlitz73; @nelson-1979; @young-1979] the thermodynamic behavior of this system is also represented by a generalized (neutral) Coulomb gas of vortices $\{q_s\}$ and dislocations $\{ q_c\}$, both with logarithmic interactions, where the strength of the interactions between vortices (and anti-vortices), both fractional and integral, is controlled by the superfluid density $\rho_s$ and the interaction between dislocations (and anti-dislocations) is controlled by the CDW stiffness $%
\kappa$. Let us denote by $g_{({q_s,q_c})}$ the fugacity of a topological excitation with topological charges $(q_s,q_c)$. In the dilute limit, in which the fugacities are small, the partition function of the generalized vector Coulomb gas[@young-1979] can be represented by a sine-Gordon type effective field theory (for a review see Refs.[@Nienhuis1987]), that in this case requires two fields $\tilde{\theta}$ and $\tilde{\varphi}$, the *dual* of the superconducting phase $\theta$ and the CDW phase $\varphi$. The effective Hamiltonian density of the 2D field theory, dual to the degrees of freedom of Eq., is $$\begin{aligned}
&&\!\!\!\!\!\!\!\!\mathcal{H}_{\mathrm{dual}}[\tilde{\theta},\tilde{\varphi}%
]= \frac{T}{2\rho_s} \left(\boldsymbol{D}_s \tilde{\theta}\right)^2+ \frac{T%
}{2\kappa} \left(\boldsymbol{D}_c \tilde{\varphi}\right)^2 \notag \\
&& \!\!\!\!\!\!\!\! -g_{({1,0})} \cos(2\pi \tilde{\theta}) - g_{({0,1})}
\cos(2\pi \tilde{\varphi}) \notag \\
&& \!\!\!\!\!\!\!\!- 2g_{(\frac{1}{2},\frac{1}{2})} \cos (\pi \tilde{\theta}%
)\cos(\pi \tilde{\varphi}) \label{eq:SG1}\end{aligned}$$
![Schematic phase diagram: KT transitions between the PDW, CDW (stripe), nematic, and charge $4e$ SC, and the (Ising) nematic-isotropic transition. $M_1$ and $M_2$ are multi-critical points (see text). PDW, CDW, and charge $4e$ SC orders have QLRO in their respective phases, and are short ranged in the nematic phase.[]{data-label="fig:phase-diagram"}](phase_diagram_1.eps){width="48.00000%"}
There are three distinct pathways for the thermal melting of the PDW state[@spin] by condensation of its topological excitations, as shown in the schematic phase diagram in Fig.\[fig:phase-diagram\]. In constructing this figure, the location of the phase boundaries were estimated by computing the scaling dimensions of the cosine operators at the non-interacting fixed point, $\Delta_{q_s,q_c}=\frac{\pi}{T} (q_s^2
\rho_s+q_c^2 \kappa)$, and then identifying the lines, $\Delta_{q_s,q_c} = 2$, at which each of the cosine operators in Eq. first becomes relevant. Alternatively, a mean-field phase diagram can be obtained by treating Eq. in a self-consistent phonon approximation. More sophisticated RG treatments of a Coulomb gas system with precisely the same *formal* structure as the above have been discussed previously for a model of the thermal melting of a charge-spin stripe state,[@kruger-2002] and the context of spinor $S=1$ Bose condensates.[@podolsky-2007] All approaches yield the same topology of the phase diagram as the one shown in Fig.\[fig:phase-diagram\], although there are some unresolved differences concerning the shape of the various phase boundaries and the nature of the multicritical points, $M_1$ and $M_2$. Moreover, since the specific problems being addressed in [@podolsky-2007] and [@kruger-2002] were different than those being considered here, the interpretation and the physics of the phases was different.
#### PDW Phase:
At low temperatures, all three cosine operators are irrelevant and the topological excitations are uncondensed. The system is in a PDW state with quasi-long range order (QLRO), as given in Eq. .
#### Charge $4e$ SC Phase:
This phase is accessed when the double dislocations proliferate, *i.e.* when the operator $%
\cos(2\pi \tilde{\varphi})$ becomes relevant, Hence, the PDW-Charge $4e$SC phase boundary is the line $\Delta_{0,1}=4\pi \kappa/T=2$, or $\frac{\kappa}{%
\rho_s}=2 \pi \frac{T}{\rho_s}$ (see Fig.\[fig:phase-diagram\].) In the charge $4e$ SC phase, the field $\tilde{\varphi}$ is pinned at integer values $n_{\varphi} \in \mathbb{Z}$, and has massive (gapped) fluctuations. In this phase, the dislocations of the CDW are screened, and the dual the CDW phase field $\varphi$ has wild fluctuations, leading to exponentially decaying correlations of both the PDW and CDW order parameters. In contrast, the field $\tilde{\theta}$ remains gapless, and the correlation function of the charge $4e$ SC have power-law correlations, with the same exponent $%
\eta_{4e}$. The phase transition from the charge $4e$ SC to the disordered non-superconducting nematic higher temperature phase proceeds through the subsequent unbinding and proliferation of vortices with fractional (half) topological charge.
#### Stripe (CDW) Phase:
This phase is accessed through the proliferation of SC vortices with integer topological charge when the operator $\cos(2\pi \tilde{\theta}_+)$ becomes marginal, $\Delta_{1,0}=\pi \rho_s/T=2$. There is CDW QLRO, with correlation functions that decay with a power law with an exponent $\eta_{CDW}$, and no SC order of any type. In this phase the field $\tilde{\theta}$ is pinned at the integer values $n_\theta \in \mathbb{Z}$, and its fluctuations are massive. The phase transition from this phase to the normal (nematic) phase occurs by the unbinding of *single dislocations* whose fractional vortex charge is screened in this phase.
#### Direct PDW-nematic normal state phase transition:
This transition proceeds through the liberation of half-vortex-integer dislocation composite excitations, *i.e.* when the topological excitations $(\pm \frac{1}{2},\pm \frac{1}{2})$ condense. This phase boundary is the $M_1 M_2$ line shown in Fig. \[fig:phase-diagram\], determined by the condition $\Delta_{\pm \frac{1}{2},\pm \frac{1}{2}}=2$.
#### Multicritical Points:
The schematic phase diagram of Fig.\[fig:phase-diagram\] has two multi-critical points, $M_1$ and $M_2$, whose existence follows from the topology of the phase diagram. From the structure of Eq. \[eq:SG1\], it also follows that there exist two special critical points with an emergent higher symmetry where two operators become marginal simultaneously: $(0,1)$ and $(\frac{1}{2},\pm \frac{1}{2})$ at $M_1^\prime$, and $(1,0)$ and $(\frac{%
1}{2},\pm \frac{1}{2})$ at $M_2^\prime$. At these points, the general $%
U(1)\times U(1)$ symmetry of the model is enlarged to an $SU(3)_1$ symmetry,[@kondev-henley96] and in their vicinity the correlation length exhibits KT behavior with a modified exponent.[@young-1979; @kruger-2002] The simple scaling arguments used to construct Fig.\[fig:phase-diagram\] and the RG analysis in [@kruger-2002] suggest that the high symmetry and multicritical points are one and the same, $M_j=M_j^\prime$, while the self-consistent phonon calculation and the RG treatment of [podolsky-2007]{} suggest that they are distinct, with $M_1^\prime$ and $%
M_2^\prime$ lying part way along the segment of the phase boundary between $%
M_1$ and $M_2$.
While the present discussion of thermal melting was confined to 2D, it can be readily extended to the case of a 3D layered material with sufficiently weak inter-layer coupling. Some aspects of this extension depend on details of the interlayer geometry[@berg-2009]. For instance, depending on whether or not the inter-layer coupling frustrates the stripe ordering, the low temperature state may or may not have a gentle, time-reversal symmetry breaking superconducting spiral superimposed on the basic single-plane stripe order. But, in general, all of the phases indicated in Fig.[fig:phase-diagram]{} persist in 3D, with the QLRO replaced by true long-range order, and a crossover, very near the phase boundaries, to 3D criticality.
At $T=0$ the power-law correlations of the PDW phase are replaced by true long range order and gapless collective (Goldstone) modes. A simple mean-field treatment[@dror; @berg-2009] of the spectrum of Bogoliubov quasiparticles in this phase reveals that, in common with a CDW, it generically has an only partially gapped Fermi surface, and hence a finite density of zero energy states. In principle, it should also be possible for the PDW phase to undergo *quantum melting*, possibly also by a quantum analog of the vortex unbinding mechanism described above.[@quantum-melting] This is a more complex problem than the thermal melting we have discussed here. In particular, the *quantum* phase transitions are affected by the existence of gapless quasiparticles. Hence, the structure and quantum dynamics of the vortices (both integer and fractional) is expected to be generally damped and anisotropic, which affects the physics of quantum melting.
There are many unusual physical consequences of the nature of the PDW phase and of its thermally melted daughter phases. We have previously discussed[@berg-2007; @berg-2008a; @berg-2009] some remarkable bulk features of a fully ordered PDW phase, including the possibility of dynamical layer decoupling (in a 3D layered material), an anomalous sensitivity to quenched disorder, the existence (in principle testable by STM and scanning SQUID) of pinned half-vortices associated with each disorder induced dislocation, and a tendency to formation of a superconducting glass phase which likely spontaneously breaks time reversal symmetry. Here we focus on “phase sensitive” measurements, which have not yet been explored, and offer the possibility of directly establishing the existence of either the fully ordered phase or its uniform, translationally invariant charge 4e SC descendant.
Clearly, a superconducting SQUID loop, made of a charge 4e SC, will exhibit all the familiar features of a SQUID, but with a distinct half-flux quantum, $hc/4e$ replacing the usual supeconducting flux quantum, $hc/2e$. From a practical point of view, however, it is undoubtedly easier to fabricate a SQUID loop in which only a single link consists of a charge 4e SC, as shown schematically in Fig.\[fig:squid\]a. Such a SQUID will exhibit the same half-flux quantum as if it were an entire loop of charge 4e SC. In addition to the practical advantage, such a geometry will detect PDW related phases under a wider range of circumstances. For instance, in the disordered phase near criticality, so long as the width of the putative charge 4e link is not large compared to the SC coherence length, it will simply act as a Josephson weak link in a charge 4e SC SQUID. Moreover, under many circumstances, even if the link consists of a PDW state, the SQUID will still exhibit $hc/4e$ periodicity.
To see this, consider a Josephson junction between a striped SC and an ordinary SC, as shown schematically in Fig.\[fig:squid\]b. The dependence of the free energy on the phase difference, $\Delta \theta $, between the two SC can be expanded as $$F(\Delta \theta )=S\sum_{n=1}^{\infty }J_{n}\cos [n\Delta \theta ]\text{,}$$where $S$ is the junction area, and $J_{n}$ involves tunneling processes of $%
2n$ electrons, so that generally $J_{n}$ decays exponentially with increasing $n$. For the geometry shown in Fig. 2b, which is representative of the putative striped SC phase in [La$_{2-x}$Ba$_{x}$CuO$_{4}$]{} near $x=1/8$, $J_{1}$ (and indeed, all $J_{2n+1}$) vanish identically, since a $\pi $ phase change of the striped SC is equivalent to a translation by half a period. As a result, the (resistively shunted) Josephson coupling between the two SC is dominated by the intrinsically smaller higher order coupling, $J_{2}$.[@half-flux] Not only does that mean that a striped SC in the place of the charge 4e SC link in Fig.\[fig:squid\]a, will act in the same way, it also means that all the standard characteristics of a single Josephson junction, including Josephson oscillations and Shapiro steps, will occur with twice the usual frequencies.
$J_{2}$ is proportional to $J^{2}$, where $J$ is the bare (microscopic) Josephson coupling. A quantitative estimate of $J_{2}$ is difficult, since it depends exponentially on microscopic parameters. However, assuming that the coherence length $\xi $ is smaller than the period of the striped superconductor, an estimate based on an effective $x-y$ model (similar to that of Ref. [@berg-2009], but adapted for a c-axis junction between a striped and a uniform superconductor) gives that $J_{2}\sim J^{2}/\left\vert J^{\prime
}\right\vert $, where $J^{\prime }$ is the inter-stripe Josephson coupling in the striped superconductor. Thus the energy denominator is the smallest energy scale of the single-plane problem. Moreover, since the total coupling energy is proportional to the area of the junction, it should be possible to use a large enough junction such that $SJ_{2}$ is substantial (i.e., much larger than $k_{B}T$).
It should be possible to perform the proposed experiments using small crystals of [La$_{2-x}$Ba$_x$CuO$_4$]{} with $x=1/8$, where there is already strong circumstantial evidence of the existence of a striped SC phase. The observation either of $hc/4e$ flux quantization in a SQUID of the sort shown in Fig. 2a, or a charge $4e$ in the Josephson relation in a Josephson junction of the sort shown in Fig. 2b, would constitute dramatic and direct evidence of the existence of these exotic superconducting phases.
E. Berg, [email protected], is the corresponding author.
We thank Dimitri Basov, Tony Leggett, Daniel Podolsky, Leo Radzihovsky, Doug Scalapino, John Tranquada, and Dale Van Harlingen for discussions. This work was supported in part by the National Science Foundation, under grants DMR 0758462 (E.F.) and DMR 0531196 (S.A.K.), and by the Office of Science, U.S. Department of Energy under Contracts DE-FG02-91ER45439 through the Frederick Seitz Materials Research Laboratory at the University of Illinois (E.F.), DE-FG02-06ER46287 through the Geballe Laboratory of Advanced Materials at Stanford University (S.A.K. and E.B.).
All three authors contributed equally to all parts of this work. The authors declare that they have no competing financial interests.
[10]{} url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, , , & . ** ****, ().
, , , & . ** ****, ().
*et al.* . ** ****, ().
, & . ** ****, ().
, , & (). , .
*et al.* . ** ****, ().
. (). .
& . ** ****, ().
, & (). , .
& (). , .
.
& . ** ****, (). .
.
& . ** ****, ().
& . ** ****, ().
. ** ****, ().
. In & (eds.) **, vol. , (, , ).
& . ** ****, ().
, & (). , .
& . ** ****, ().
& . ** ****, ().
& . ** ****, ().
*et al.* . ** ****, ().
| {
"pile_set_name": "ArXiv"
} |
[OG 3.4.4\
]{} [GAMMA-RAY ABSORPTION AT HIGH REDSHIFTS AND THE GAMMA-RAY BACKGROUND\
]{}
F. W. Stecker$^{1}$ and M. H. Salamon$^2$\
[*$^1$LHEA, NASA/Goddard Space Flight Center, Greenbelt, MD 20770, USA\
$^2$Physics Dept., University of Utah, Salt Lake City, UT 84112, USA\
*]{}
[ABSTRACT\
]{} We present results of a calculation of absorption of 10-500 GeV $\gamma$-rays at high redshifts (Salamon and Stecker, 1997). This calculation requires the determination of the high redshift evolution of the full spectral energy distribution of the intergalactic photon field. For this, we have primarily followed the recent analysis of Fall, Charlot & Pei. We give our results for the $\gamma$-ray opacity as a function of redshift out to a redshift of 3. We then give predicted $\gamma$-ray spectra for selected blazars and also extend our results on the background from unresolved blazars to an energy of 500 GeV. Absorption effects are predicted to significantly steepen the background spectrum above 20 GeV. Our absorption calculations can be used to place limits on the redshifts of $\gamma$-ray bursts. Our background calculations can be used to determine the observability of multi-GeV lines from dark matter (neutralino) particles.
INTRODUCTION
============
Absorption of $\gamma$-rays from blazars and extragalactic $\gamma$-ray bursts is strongly dependent on the redshift of the source (Stecker, De Jager & Salamon 1992). Stecker & De Jager (1997) have calculated the absorption of $\gamma$-rays at above 0.3 TeV at redshifts up to 0.54. The study of extragalactic absorption below 0.3 TeV at higher redshifts is a more complex and physically interesting subject. In order to calculate such absorption properly, one must determine the spectral evolution of galaxy starlight photons from the IR through the UV range out to high redshifts. Pei & Fall (1995) have devised a clever method for calculating stellar emissivity as a function of redshift, one which is consistent with all recent data. We adopt this method and extend it by calculating the additional effect of metallicity evolution on stellar emissivity. We then calculate the $\gamma$-ray opacity of the universe to stellar photons at various redshifts and apply our results to selected blazar spectra and the blazar background.
CALCULATION OF STELLAR EMISSIVITY
=================================
The basic idea of the Pei & Fall (1995) approach, which we follow, is to relate the star formation rate to the evolution of neutral gas density in damped Ly$\alpha$ systems and then to use the population synthesis models (Bruzal & Charlot 1993) to calculate the mean volume emissivity of the universe from stars as a function of redshift and frequency. Damped Ly$\alpha$ systems are believed to be either the precursors to galaxies or young galaxies themselves. It is in these systems that initial star formation probably took place, so there is a relationship between the mass content of stars and gas in these clouds. The results obtained by Fall, 1996 show excellent agreement with observational data obtained by the Canada-France redshift survey group for redshifts out to 1 (Lilly, 1996) and are consistent with lower limits obtained on the emissivity at higher redshifts (Madau 1996). The stellar emissivity is found to peak between a redshift of 1 and 2 which is consistent with the results of ongoing observations from both the Hubble and Keck telescopes. We have made one significant modification to the calculations of Fall, (1996). We have attempted to account for the significantly lower metallicity of early generation stars at higher redshifts which results in increased emission at shorter wavelengths and lowered emission at longer wavelengths. In order to estimate this effect, we have used the results of Worthey (1994) and moderately extrapolated them to both lower and higher wavelengths (Salamon & Stecker 1997). We have also considered the effect of dust opacity and have assumed a reasonable escape factor to account for the fact that a small fraction of Lyman continuum photons escape from galaxies unattenuated by stars and dust. The effect of this escape factor on our subsequent opacity calculations is negligible, since there are not enough ionizing photons in intergalactic space to provide a significant opacity to multi-GeV . Because the metallicity corrections are less certain for the more massive stars ($M > 2M_{\odot}$), our metallicity-corrected UV radiation density should be viewed as an upper limit.
OPACITY OF THE UNIVERSE TO GAMMA-RAYS
=====================================
Once the spectral energy density distribution of stellar photons in intergalactic space as a function of redshift is determined, the opacity of the universe to as a function of energy and redshift can be calculated (Stecker, 1992). The basic processes which causes the attenuation of is the interaction of a with a starlight photon which results in the production of an electron-positron pair. Our results indicate that above an energy of $\sim$15 GeV will be attenuated if they are emitted at redshifts greater than or equal to $\sim$3. The burst observed by EGRET on 17 Feb 1994 contained a photon of energy $\sim$18 GeV. Figure 1 shows the calculated opacity as a function of energy for various source redshifts, with and without the metallicity correction included; the true opacities likely lie between the values shown in the left and right halves of Fig. 1.
1.0truecm
0.0truecm
-0.0truecm [*Figure 1: High energy opacities calculated with and without metallicity correction factor included.* ]{}
Because the stellar emissivity peaks between a redshift of 1 and 2, there is little increase in the opacity when one goes to redshifts greater than 2. This weak dependence indicates that the opacity is not determined by the initial epoch of galaxy formation (which may be at $z \ge 5$), contrary to the speculation of MacMinn & Primack (1996).
EFFECT OF ABSORPTION ON BLAZARS AND THE BACKGROUND
==================================================
Figure 2 shows the attenuation of spectra resulting from the opacities given in Figure 1 for the blazar sources 1633+382 ($z=1.81$), 3C279 ($z=0.54$), 3C273 ($z=0.15$), and Mrk421 ($z=0.031$). The solid (dashed) lines result from the opacities shown in left (right) half of Figure 1. The redshift dependence of the break energies is evident, as is the absence of any breaks below 10 GeV. Future measurements of blazar spectral break energies will discriminate between models of extragalactic extinction (such as this one) and those involving cutoffs [*intrinsic*]{} to the source.
1.0truecm
0.0truecm
-0.0truecm [*Figure 2: High energy spectra of selected blazars attenuated according to the opacities of Figure 1. The solid (dashed) lines correspond to the opacities calculated with (without) the metallicity correction.* ]{}
Figure 3 shows the effect of absorption on the extragalactic background computed using the unresolved blazar model of Stecker & Salamon (1996). The solid (dashed) lines correspond to the metallicity correction being included (neglected) in the opacity calculation. The two families of curves correspond to point source sensitivities of EGRET (top curves) and GLAST (bottom curves). (A better point source sensitivity results in the reduction in the number of [*unresolved*]{} sources which contribute to the background.) Also shown are the preliminary EGRET data on the extragalactic background spectrum (Fichtel 1996). The cutoff observed beyond $\sim$20 GeV reduces the effect of the extragalactic background on searches for lines from neutralino-neutralino annihilation in the galactic halo, should the extragalactic background be due to unresolved flat-spectrum radio quasars.
1.0truecm
0.0truecm
-0.0truecm [*Figure 3: The extragalactic background, calculated with the model of Stecker and Salamon (1996), and attenuated with the opacities of Fig. 1. (See text.)* ]{}
ACKNOWLEDGEMENTS
================
We thank M. Fall, M. Malkan, Y. Pei, and G. Worthey for helpful discussions and comments.
REFERENCES
==========
[ -5mm]{}
Bruzal, A. G. & Charlot, S. 1993, [$\;\,$**405, 538.**]{}
Fall, S. M., Charlot, S. & Pei, Y. C. 1996, [$\;\,$**402, 479.**]{}
Fichtel, C. E. 1996, [$\;\,$**120, 23.**]{}
Lilly, 1996, [$\;\,$**460, L1.**]{}
MacMinn, D, & Primack, J. 1966, [$\;\,$**75, 413.**]{}
Madau, P. 1996 [$\;\,$**283, 138.**]{}
Pei, Y. C. & Fall, S. M., 1995, [$\;\,$**454, 69.**]{}
Salamon, M. H. & Stecker, F. W. 1997, submitted to
Stecker, F. W. & De Jager, O. C. 1997, [$\;\,$**476, 712.**]{}
Stecker, F. W., De Jager, O. C. & Salamon, M. H. 1992, [$\;\,$**390, L49.**]{}
Stecker, F.W., and Salamon, M.H. 1996, [$\;\,$**464, 600.**]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The opacity of typical objects in the world results in occlusion — an important property of natural scenes that makes inference of the full 3-dimensional structure of the world challenging. The relationship between occlusion and low-level image statistics has been hotly debated in the literature, and extensive simulations have been used to determine whether occlusion is responsible for the ubiquitously observed power-law power spectra of natural images. To deepen our understanding of this problem, we have analytically computed the 2- and 4-point functions of a generalized “dead leaves" model of natural images with parameterized object transparency. Surprisingly, transparency alters these functions only by a multiplicative constant, so long as object diameters follow a power law distribution. For other object size distributions, transparency more substantially affects the low-level image statistics. We propose that the universality of power law power spectra for both natural scenes and radiological medical images – formed by the transmission of x-rays through partially transparent tissue – stems from power law object size distributions, independent of object opacity.'
author:
- Joel Zylberberg
- David Pfau
- Michael Robert DeWeese
title: 'Dead leaves and the dirty ground: low-level image statistics in transmissive and occlusive imaging environments'
---
Introduction
============
Natural images are surprisingly statistically uniform. The autocorrelation function, a measure of how similar nearby pixels tend to be, is virtually universal for natural images [@stephens; @ruderman; @dong_atickB; @field; @olshausen_simoncelli; @torralba; @vandersschaff] (Fig. 1). This is typically quantified by measuring image power spectra (Fourier transform of the autocorrelation function), which are well-described by scale-invariant power law functions with power $\mathcal{P}$ and spatial frequency $k$ related by $\mathcal{P}(k)\propto k^{-\alpha}$, with exponents $\alpha \approx 2$. The exponents $\alpha$ vary slightly from image-to-image, and there are small differences in average exponent $\alpha$ between terrestrial [@field; @ruderman; @vandersschaff] and aquatic [@balboa] environments, and between natural and man-made ones [@torralba].
Intriguingly, even radiological images like mammograms have power law power spectra [@heine_velthuizen; @li], typically with larger $\alpha$ values, despite the fact that the physics of image formation are very different for radiological and natural images. In natural images, formed by reflection of light off of surfaces, objects tend to be opaque, and thus they occlude one another, whereas in mammograms, formed by the transmission of x-rays through breast tissue, objects are more transmissive and do not completely occlude one another. The statistics of radiological images have received less attention and are less well understood. Interestingly, however, the powers $\alpha$ typically vary between mammogram images of patients with low vs. high risk of developing breast cancer [@li], and vary as a function of the density of the breast tissue [@metheany], highlighting the potential clinical importance of these image statistics.
The statistical regularity of natural scenes implies that engineers can design, and evolution might have selected for, coding schemes that exploit this structure [@barlow61; @olshausen_simoncelli]. Indeed, the peripheral mammalian visual system appears to exploit this homogeneity by using simple filters to decorrelate the incoming signal [@dong_atickA; @atick_redlich; @dan] and more complex feature dictionaries to efficiently encode the decorrelated signal [@jz_plos; @olshausen_simoncelli; @rehn_sommer].
{width="3.6in"}
Using the intuition that the environment is composed of distinct objects, Ruderman studied a “dead leaves" model [@matherton_1968; @bordenave] for natural scenes, in which images are created by sequentially placing opaque, potentially overlapping circles of random brightness in random locations on a 2-dimensional image plane [@ruderman_scaling] (Fig. 2A). Ruderman modeled correlations between pixels by assuming a different correlation function for points falling within a visible circle than for points falling in different visible circles. Using analytical calculations he demonstrated that, so long as the diameters $s$ of the circles follow a power law distribution with probabilities $p(s) \propto s^{-(3 + \eta)}$, the images exhibit power law correlation functions, $C(q) \propto q^{-\eta}$, where $q$ is the separation between pixels, and power law power spectra, $\mathcal{P}(k) \propto k^{-(2-\eta)}$. If the circle sizes are drawn from other distributions, Ruderman’s analytical calculations suggest that the power spectra that could be made to differ from a power law, contrary to the old notion [@carlson] that the $1/k^2$ power spectra result from the mere presence of edges, each of which has a $1/k^2$ 1-dimensional power spectrum (*cf.* Balboa et al. [@balboa_2001]). More recently, Balboa et al. [@balboa_2001] simulated the analytical examples presented by Ruderman [@ruderman_scaling], including images with the exponential distribution of object sizes that was claimed [@ruderman_scaling] to yield non-power-law power spectra. They found that these images had nearly power law power spectra, and subsequently reiterated the previous claim that occlusion, and not object size distributions, are the cause of power law power spectra in natural images.
This “edges vs. size distributions" debate was subsequently resolved when Hsiao and Milane demonstrated, via numerical simulations, that dead leaf models with partially transparent objects (and thus only partial occlusion) whose sizes follow a power law distribution yield power law power spectra, and that dead leaf models with opaque objects from other size distributions can have non power-law power spectra [@hsiao]. In other words, occlusion is neither necessary, nor sufficient, to yield power law image power spectra. In the same paper, Hsiao and Milane computed the power spectrum of a simplified ensemble of images formed by summing the intensities of different randomly placed disks. This model was simpler than the images with partially occluding leaves that they simulated. The linearity of this model makes it relatively straightforward to compute the Fourier transform of the model images, and thus to estimate the power spectra.
Thus, to date, the 2-point statistics of dead leaf image models have been analytically calculated for both fully opaque leaves [@ruderman_scaling], and for fully transmissive leaves [@hsiao]. What remains is to solve for the 2-point function of images with partial occlusion, which will deepen our understanding of how opacity and image statistics inter-relate along this continuum of object properties. Thusly motivated, we studied a generalized dead leaves model, in which the leaves have variable transparency. While general feature probabilities have been solved exactly for the fully opaque dead leaves model [@pitkow], our transparent generalization requires other methods and has not previously been systematically explored. We show herein that, so long as leaf sizes follow a power-law distribution, transparency results in an overall multiplicative factor in the 2- and 4-point functions but does not change their functional (power-law) form. For other size distributions, transparency does change the form of the autocorrelation function, suggesting that power-law size distributions, unify the observed power spectra of natural and radiological images.
Analytical calculation of the 2-point function in the transmissive dead leaves model
====================================================================================
We begin by analytically computing the 2-point functions of images in our “transmissive dead leaves" environment. For image pixels values $I(\vec{x})$, the 2-point function is given by $C(\vec{x},\vec{x}') = \left< I(\vec{x}) I(\vec{x}') \right> = C(|\vec{x}-\vec{x}'|)$, where the angle brackets denote averaging over images drawn from this ensemble and the second step stems from the fact that, since our model world is invariant under both translations and rotations, the 2-point function depends only on the distance $|\vec{x}-\vec{x}'| = q$ between sample points.
The image is formed by randomly placing a circle whose diameter $s$ is drawn from some distribution, with brightness value $b$, and transparency $a$, on a surface of diameter $L$. The brightnesses $b$ will be drawn from a zero-mean distribution, and the transparencies $a \in [0,1]$ can also be random. A value $a=1$ specifies a fully transparent (invisible) circle, while a value of $a=0$ specifies a fully opaque circle, as in Ruderman’s model [@ruderman_scaling]. When a new circle is added, the pixel value $I(\vec{x})$ at a point $\vec{x}$ that falls within the circle undergoes the transformation $$I(\vec{x}) \to (1-a)b + aI(\vec{x}).$$ Pixels not lying under the circle are unaffected by its addition. This process is continued ad infinitum to create model images (Fig. 2).
![ (Color online) **For power law object size distributions, the 2-point statistics of opaque and transmissive dead leaves images differ by a multiplicative constant.** (**A**) A representative image from the opaque ($a=0$) dead leaves model with circle diameters drawn from the distribution $p(s) \propto s^{-3.2}$ for $s>s_0 = 1$ pixel and circle brightnesses drawn uniformly within $b \in [-1,1]$. (**B**) When the circles are partially transparent ($a=0.25$ for all circles), but all other parameters are the same, previously occluded circles are partially visible. (**C**) A higher level of transparency ($a = 0.75$) results in an image that begins to approximate Gaussian pink noise, as expected from the central limit theorem [@pitkow]. (**D**) Autocorrelation functions of dead leaves image ensembles of different opacity levels differ only by a multiplicative constant for power-law object size distributions. The 2-point functions are power law functions of distance, with power $\sim-0.2$, in good agreement with our analytical calculation. (**E**) Similarly, the power spectra of these image ensembles are roughly power-law functions and are all the same up to a multiplicative constant. The ratio of the opaque and most transparent power spectra is nearly flat. At relatively high spatial frequencies (above $\sim 20$ cycles/image), corresponding to small length scales, the $q \gg s_0$ approximation in our analytical calculation fails, and slight deviations from power-law power spectra can be observed, as can deviations from constancy in the ratio.](fig2){width="3.6in"}
We will compute $\left< I(\vec{x})^2 \right>$ and $C(q)$ recursively by noting that adding another leaf to an image creates a new image from the same transmissive dead leaves ensemble and thus the (average) statistical properties must remain unchanged by this transformation [@ruderman_scaling].
Using Eq. (1), we can compute the pixel variance $$\begin{aligned}
\left< I^2(\vec{x}) \right> &=& \left( 1-P_{in} \right) \left< I^2(\vec{x}) \right> \\
&+& P_{in} \left< \left( aI(\vec{x}) + (1-a)b \right)^2 \right> \nonumber \\ \nonumber
\Rightarrow \left< I^2(\vec{x}) \right> &=& \frac{ \left< b^2 \right> \left< (1-a)^2\right>} {1- \left< a^2 \right> } \nonumber,\end{aligned}$$ where $P_{in}$ is the probability that the point in question falls within the newly added circle. The quantity $P_{in}$, and thus the distribution of circle sizes, does not affect the pixel variance. It will however, affect the spatial properties of the image, including $C(q)$.
To compute $C(q)$, consider how the pixel values of a pair of points with separation $q$ are affected by the addition of a new leaf. After adding the leaf, either one, both, or neither of the sample points lie under the leaf, resulting in three different possible modifications to the pixel values (Eq. 1). These outcomes occur with probabilities $P_1(q)$, $P_2(q)$, or $P_0(q)$, respectively, which we will later compute. Equating the 2-point functions before and after the addition of a new leaf, we obtain $$\begin{aligned}
&C(q)& = P_0(q) C(q) + P_1(q) \left< \left[ a I(\vec{x}) + (1-a)b \right] I(\vec{x}') \right> \nonumber \\
&+& P_2 (q) \left< \left[ a I(\vec{x}) + (1-a)b \right] \left[ a I(\vec{x}') + (1-a)b \right] \right>. \end{aligned}$$ Recalling the definition of the autocorrelation function and the normalization $P_0 (q)+ P_1(q) + P_2(q) = 1$, we find $$C(q) = \frac{ \left< b^2 \right> \left< (1-a)^2 \right> P_2(q)}{P_1(q)\left< 1-a \right> + P_2(q) \left< 1-a^2 \right> }.$$ The quantities $\left< b^2 \right>$, $\left< a^2 \right>$, and $\left< a \right>$ depend on the distributions of circle brightnesses and opacities.
To calculate $P_1(q)$, we first define $P^{\star} = \left< s^2 \right> / L^2$, which is the probability that any given point in the image falls within a newly-deposited leaf. Here $L$ is the diameter of the circular image area, $s$ is the diameter of the newly added circle, and we assume $\left<s^2 \right> \ll L^2$. The probability $P_1(q)$ that either point, but not both, falls within the circle is then $P_1(q) = 2 \left( P^{\star} - P_2(q) \right)$, where the factor of 2 comes in because there are two such points to consider.
To determine the probability $P_2(q)$, note that, for a circle of diameter $s$, given that one particular point $\vec{x}$ is within the circle (which occurs with probability $s^2/L^2$), the probability that another point, a distance $q$ away, is also within the circle, is given by [@ruderman_scaling] $g(q/s \in [0,1]) = \frac{2}{\pi} \left[ \cos^{-1}(q/s) - (q/s) \sqrt{1-(q/s)^2} \right]$, and thus $$P_2(q) = \int_{0}^{\infty} \frac{s^2}{L^2} g(q/s)p(s) ds.$$ For a power law size distribution $p(s) = (A/s_0) (s/s_0)^{-\alpha}$, where $\alpha > 3$, $A$ is a unitless normalization constant, and $s_0$ is the small-size cutoff, the change of variables $u = s/q$ in the above integral yields $$P_2(q) = A \left( \frac{s_0}{L} \right)^2 \left( \frac{q}{s_0} \right)^{-(\alpha -3)} \int_{1}^{\infty} g(1/u) u^{2-\alpha} du.$$ Define the integral to be $B(\alpha)$. For pixel separations much larger than the small-size cutoff of our leaf diameter distribution, $q \gg s_0$ (in which case $P^{\star} = \frac{A}{\alpha-3} \left( \frac{s_0}{L} \right)^2 \gg P_2(q)$), Eq. (4) becomes $$C(q) = \frac{ B(\alpha) \left(\alpha-3 \right) \left< b^2 \right> \left< (1-a)^2 \right>}{2\left< 1-a \right>} \left( \frac{q}{s_0} \right)^{-(\alpha - 3)},$$ yielding an image power spectrum [@ruderman_scaling] $\mathcal{P}(k) \propto \frac{ \left< b^2 \right> \left< (1-a)^2 \right>}{\left< 1-a \right>} k^{-(5-\alpha)}$ in which the opacity affects the power spectrum only as a multiplicative prefactor. When $a=0$ for all circles (opaque limit), our result is equal to that of Ruderman [@ruderman_scaling], as it must be. Also note that, as one might expect, the 2-point function does not depend on the size $L$ of the image surface.
To demonstrate that leaf opacity can affect the functional form of the 2-point function, we repeat the above calculations, but now have all leaves be the same size $s^{\star}$. The size distribution is thus $p(s) = \delta(s-s^{\star})$, in which case the correlation function is $$C_{\delta} (q) = \frac{ \left<b^2 \right> \left<\left( 1-a \right)^2 \right> g(q/s^{\star})}{2 \left<1-a \right> - \left< \left(1+a\right)^2 \right> g(q/s^{\star})},$$ which depends non-trivially on $a$: for $q>s^{\star}, g(q/s^{\star})=0$ and the correlation function vanishes, so the large-$q$ limit in which Eq. (7) was derived is irrelevant for delta-function size distributions. Furthermore, even for fully opaque leaves, it is clear that this correlation function, which is identically zero for $q>s^{\star}$, is not described by a power-law function of distance.
A comparison of Eqs. (2) and (7) shows that the pixel variance, and the image autocorrelation function, are multiplied by different opacity dependent pre-factors. For $q=0$, the variance and the 2-point function are equal, so the fact that for $q \gg s_0$, they scale differently with changing opacity highlights that there is a qualitative change in the 2-point function near the $q \sim s_0$ boundary. For natural images, the minimum object size is much smaller than our cameras can resolve, so this boundary is never encountered in practice. Furthermore, this comparison demonstrates that not all image statistics vary in the same way with changing leaf opacity.
Analytical calculation of the 4-point function for collinear points in the transmissive dead leaves model
=========================================================================================================
As we have seen, the form of the 2-point function is independent of leaf opacity for power-law object size distributions. At the same time, the images generated with different leaf opacities (Fig. 2) are visibly different, so there must be some difference in the image statistics (aside from the overall pixel variance) from ensembles with different object opacities. To understand this difference, we consider higher-order statistics beyond the 2-point function. If the leaf brigthnesses $\left<b \right>$ are symmetrically distributed about zero, then the 3-point function will vanish, and so the next possible candidate beyond the 2-point function is the 4-point function.
In this section, we will compute the 4-point function $C^{coll}_4(\vec{x}, \vec{x}',\vec{x}'',\vec{x}''') = \left< I(\vec{x}) I(\vec{x}') I(\vec{x}'') I(\vec{x}''') \right>$ for equidistant collinear points; $|\vec{x}-\vec{x}'| = |\vec{x}'-\vec{x}''| = |\vec{x}''-\vec{x}'''| = q$ and $|\vec{x}-\vec{x}''| = |\vec{x}'-\vec{x}'''| = 2q$, for the dead leaves model with power-law leaf size distribution. We chose this arrangement of points because it considerably simplifies the analysis of the 4-point function, for reasons that will become apparent during the calculation. Nevertheless, the calculation itself is still somewhat tedious, so some readers may wish to skip to the result at the end of this section. As in the case of the 2-point function described above, since our image ensemble is invariant under translations and rotations, the result depends only on the pixel spacing $q$: $C^{coll}_4(\vec{x}, \vec{x}',\vec{x}'',\vec{x}''') = C^{coll}_4(q)$. We apply the same recursive logic that we used for computing the 2-point function in order to infer the 4-point function, and start by enumerating all of the possible modifications to the 4-point function upon the addition of a new circle. We will number the points from left to right. Thus,
$$\begin{aligned}
C^{coll}_4(q) &=& P^{coll}_{\o}(q) C^{coll}_4(q) \\
&+& P^{coll}_{1,\o} (q) \left< \left[ a I(\vec{x}) + (1-a)b \right] I(\vec{x}') I(\vec{x}'') I(\vec{x}''') \right> \nonumber \\
&+& P^{coll}_{2,\o}(q) \left< I(\vec{x}) \left[a I(\vec{x}') + (1-a)b \right] I(\vec{x}'') I(\vec{x}''') \right> \nonumber \\
&+& P^{coll}_{3,\o}(q) \left< I(\vec{x}) I(\vec{x}') \left[a I(\vec{x}'') + (1-a)b \right] I(\vec{x}''') \right> \nonumber \\
&+& P^{coll}_{4,\o}(q) \left< I(\vec{x}) I(\vec{x}') I(\vec{x}'') \left[a I(\vec{x}''') + (1-a)b \right] \right> \nonumber \\
&+& P^{coll}_{1,2}(q) \left< \left[a I(\vec{x}) + (1-a)b \right]\left[a I(\vec{x}') + (1-a)b \right] I(\vec{x}'') I(\vec{x}''') \right> \nonumber \\
&+& P^{coll}_{2,3}(q) \left< I(\vec{x}) \left[a I(\vec{x'}) + (1-a)b \right]\left[a I(\vec{x}'') + (1-a)b \right] I(\vec{x}''') \right> \nonumber \\
&+& P^{coll}_{3,4}(q) \left< I(\vec{x}) I(\vec{x}') \left[a I(\vec{x''}) + (1-a)b \right]\left[a I(\vec{x}''') + (1-a)b \right] \right> \nonumber \\
&+& P^{coll}_{1,2,3}(q) \left< \left[a I(\vec{x}) + (1-a)b \right] \left[a I(\vec{x'}) + (1-a)b \right] \left[a I(\vec{x''}) + (1-a)b \right] I(\vec{x}''' )\right> \nonumber \\
&+& P^{coll}_{2,3,4}(q) \left< I(\vec{x}) \left[a I(\vec{x'}) + (1-a)b \right] \left[a I(\vec{x''}) + (1-a)b \right] \left[a I(\vec{x'''}) + (1-a)b \right] \right> \nonumber \\
&+& P^{coll}_{1,2,3,4}(q) \left< \left[a I(\vec{x}) + (1-a)b \right] \left[a I(\vec{x'}) + (1-a)b \right] \left[a I(\vec{x''}) + (1-a)b \right] \left[a I(\vec{x'''}) + (1-a)b \right] \right> \nonumber, \end{aligned}$$
where $P^{coll}_{\o}$ is the probability that none of the four collinear points fall under the newly-deposited circle, $P^{coll}_{i,\o}$ is the probability that only the $i^{th}$ point falls under the newly-deposited circle, $P^{coll}_{i,j}$ is that probability that only the $i^{th}$ and $j^{th}$ collinear points fall under the newly-deposited circle, and so on. Because the points are collinear, it is impossible for non-neighboring pixels to fall under a given circle unless all of the pixels in between them also fall under that circle. Hence, there are no terms like $P^{coll}_{1,3}$ or $P^{coll}_{1,2,4}$ in the above equation, since they would require there to be “gaps" between neighboring pixels. Alternatively, one can include those terms but note that the probabilities associated with them are zero.
To simplify Eq.(9) to the point that we can easily solve for $C^{coll}_4(q)$, we will first expand and simplify all of the average products $\left< \cdot \right>$, then compute all of the probabilities $P^{coll}_{\{ \cdot \}}$, and finally assemble all of these pieces.
Expanding and simplifying the average pixel-value-products
----------------------------------------------------------
Since the circle brightnesses $b$ are zero-mean and independently drawn, each of the terms in which a single pixel is modified (the second through fifth terms in Eq. (9)) reduces to $\left<a\right> P^{coll}_{i,\o} C^{coll}_4(q)$. Similarly, expanding the terms in which 2 points fall under the circle (the sixth through eighth terms in Eq. (9)), recalling that $\left<b \right> = 0$, and performing a bit of algebra, each of those terms can be simplified to
$$P^{coll}_{i,j}(q) \left[ \left< a^2 \right> C^{coll}_4(q) + \left< (1-a)^2 \right> \left< b^2 \right> C_2(|k-m|q) \right],$$
where $k \ne m$, $k,m \in \{1,2,3,4\} \backslash \{i,j\} $, $C_2(.)$ is the 2-point function that we calculated in the previous section (Eqs. (4) and (7) for power-law object size distributions), and we now denote it with a subscript 2 to avoid confusion with the 4-point function.
Assuming that the circle brightnesses are symmetrically distributed about zero (and thus $\left<b^3 \right> = 0$), the $P^{coll}_{i,j,k}$ terms in which 3 points fall under the circle reduce to
$$P^{coll}_{i,j,k}(q) \left[ \left< a^3\right> C^{coll}_4(q) + \left< a(1-a)^2 \right> \left< b^2 \right> \left( C_2(q) + C_2(2q) + C_2(3q) \right) \right].$$
Finally, the last term in Eq. (9), in which all 4 points fall under the new circle, simplifies to
$$P^{coll}_{1,2,3,4}(q) \left[ \left< a^4\right> C^{coll}_4(q) + \left<a^2 (1-a)^2 \right> \left<b^2 \right> \left( 3C_2(q) + 2C_2(2q) + C_2(3q) \right) + \left< (1-a)^4 \right> \left<b^4 \right> \right].$$
Computing the probabilities $P^{coll}_{\{ \cdot \}}$
----------------------------------------------------
We now require the probabilities $P^{coll}_{\o}, P^{coll}_{1,\o}, P^{coll}_{2,\o}, P^{coll}_{1,2}, P^{coll}_{2,3}, P^{coll}_{1,2,3}$, and $P^{coll}_{1,2,3,4}$. The remaining probabilities in Eq. (9) are equivalent to these because of the symmetry of the arrangement of points (and of the image ensemble).
Because all intervening pixels must lie under the circle if the bounding ones do, $P^{coll}_{1,2,3,4}(q) = P_2(3q)$, where $P_2(.)$ is the probability that 2 pixels of a given separation lie under the same circle, and is calculated in the previous section (Eq. (5) for power-law distributions of circle sizes). We will use similar arguments to obtain the other 6 probability functions that we require.
The “triplet" probability $P^{coll}_{1,2,3}(q)$ is thus given by the probability that 3 of the (adjoining) pixels fall under the circle, minus the probability that all four pixels fall under it: $P^{coll}_{1,2,3} = P_2(2q) - P_2(3q)$. And by the same logic, $P^{coll}_{1,2} = P_2(q) - P_2(2q)$.
For the “inner" pairs, we compute the probability of the 2 “inner" points falling under the circle minus the probability that those two points *and any adjoining ones* all fall under the circle. Thus,
$$\begin{aligned}
P^{coll}_{2,3}(q) &=& P_2(q) - P^{coll}_{1,2,3} - P^{coll}_{2,3,4} - P^{coll}_{1,2,3,4} \\
\Rightarrow P^{coll}_{2,3}(q) &=& P_2(q) - 2P_2(2q) + P_2(3q). \nonumber\end{aligned}$$
Similarly, $P^{coll}_{1,\o}(q) = P^{\star} - P_2(q)$, where $P^{\star} = \left< s^2 \right> / L^2$ is the probability of any given point falling under the newly-deposited circle, and
$$\begin{aligned}
P^{coll}_{2,\o}(q) &=& P^{\star} - P^{coll}_{1,2}(q) - P^{coll}_{2,3}(q) - P^{coll}_{1,2,3}(q) - P^{coll}_{2,3,4}(q) - P^{coll}_{1,2,3,4}(q) \\
\Rightarrow P^{coll}_{2,\o}(q) &=& P^{\star}- 2P_2(q) + P_2(2q) \nonumber.\end{aligned}$$
Finally,
$$\begin{aligned}
P^{coll}_{\o}(q) &=& 1 - \sum_{i} P^{coll}_{i,\o}(q) - \sum_{i,j\ne i} P^{coll}_{i,j}(q) - \sum_{i,j\ne i,k\ne i,j}P^{coll}_{i,j,k}(q) - P^{coll}_{1,2,3,4}(q) \\
\Rightarrow P^{coll}_{\o}(q) &=& 1 - 4P^{\star} + 3P_2(q) \nonumber.\end{aligned}$$
Assembling the pieces to find $C^{coll}_4(q)$
---------------------------------------------
Before substituting all of our results into Eq. (9) and solving for $C^{coll}_4(q)$, it will be useful to first consider the $q \gg s_0$ limit, in which we derived the 2-point function. In that limit (Eq. (6)),
$$P_2(q) = A B(\alpha) \left( \frac{s_0}{L} \right)^2 \left( \frac{q}{s_0} \right)^{-(\alpha-3)} \ll 1$$
and
$$C_2(q) = \frac{ B(\alpha) \left(\alpha-3 \right) \left< b^2 \right> \left< (1-a)^2 \right>}{2\left< 1-a \right>} \left( \frac{q}{s_0} \right)^{-(\alpha - 3)} \ll 1,$$
so only the lowest-order terms in these quantities need to be considered. Because of the power-law nature of these functions, $C_2(2q)$ and $P_2(2q)$ have the same dependence on distance $q$ as do the $C_2(q)$ and $P_2(q)$ terms, but are smaller by a factor of $2^{-(\alpha-3)} $, and similarly for the $f(3q)$ type terms.
Substituting all of the products and probabilities derived in the preceding subsections into Eq. (9), keeping only the lowest-order terms in $(q/s_0)^{-(\alpha-3)}$, which dominate for $q \gg s_0$, and solving for $C^{coll}_4(q)$, we find that
$$\begin{aligned}
C^{coll}_4(q) &\approx& \frac{ B(\alpha) \left(\alpha-3 \right) \left< b^4 \right> \left< (1-a)^4 \right>}{4\left< 1-a \right>} \left( \frac{3q}{s_0} \right)^{-(\alpha - 3)}.\end{aligned}$$
Thus, the 4-point function for this arrangement of points (in the $q \gg s_0$ limit) has the same power-law form as does the 2-point function (Eq. 7), and it also only depends on opacity by a multiplicative pre-factor. Given that this (collinear) arrangement of points is so similar to the arrangement of points in the 2-point function (two points will always be collinear), this result is perhaps unsurprising. To test the generality of this result, we will compute the 4-point function for a square arrangement of points in the next section.
Analytical calculation of the 4-point function for a square arrangement of points in the transmissive dead leaves model
=======================================================================================================================
In this section, we calculate the 4-point function for our transmissive dead leaves ensemble, for the case in which the 4 points lie on the vertices of a square with edge length $q$. Similar to the collinear arrangement of points, the symmetry in this arrangement will greatly simplify our calculations and, since it has non-trivial geometry when compared to the collinear arrangement, there is a possibility for interesting features to arise in this 4-point function that are not apparent in either the 2-point function, or the 4-point function for collinear points.
We will label these points ${1,2,3,4}$, going clockwise, and beginning in the upper left-hand corner. Similar to the calculation for the collinear case, we first list all of the possible modifications to the 4-point function, and the probabilities with which they occur. We will then simplify this expression, calculate the relevant probabilities, and use recursion to solve for the 4-point function. Similar to the previous calculations, the translation and rotation invariance of our image ensemble means that this 4-point function will depend only on the edge length of the square: $C^{square}_4(\vec{x}_1,\vec{x}_2,\vec{x}_3,\vec{x}_4) = C^{square}_4(q)$.
Enumerating all possible modifications caused by the addition of a new circle, we find that
$$\begin{aligned}
C^{square}_4(q) &=& P^{square}_{\o}(q) C^{square}_4(q) \\
&+& 4 P^{square}_{1,\o} (q) \left< \left[ a I(\vec{x}_1) + (1-a)b \right] I(\vec{x}_2) I(\vec{x}_3) I(\vec{x}_4) \right> \nonumber \\
%&+& P_{2,\o}(q) \left< I(\vec{x}_1) \left[a I(\vec{x}_2) + (1-a)b \right] I(\vec{x}_3) I(\vec{x}_4) \right> \nonumber \\
%&+& P_{3,\o}(q) \left< I(\vec{x}_1) I(\vec{x}_2) \left[a I(\vec{x}_3) + (1-a)b \right] I(\vec{x}_4) \right> \nonumber \\
%&+& P_{4,\o}(q) \left< I(\vec{x}_1) I(\vec{x}_2) I(\vec{x}_3) \left[a I(\vec{x}_4) + (1-a)b \right] \right> \nonumber \\
&+& 4 P^{square}_{1,2}(q) \left< \left[a I(\vec{x}_1) + (1-a)b \right]\left[a I(\vec{x}_2) + (1-a)b \right] I(\vec{x}_3) I(\vec{x}_4) \right> \nonumber \\
%&+& P_{2,3}(q) \left< I(\vec{x}_1) \left[a I(\vec{x}_2) + (1-a)b \right]\left[a I(\vec{x}_3) + (1-a)b \right] I(\vec{x}_4) \right> \nonumber \\
%&+& P_{3,4}(q) \left< I(\vec{x}_1) I(\vec{x}_2) \left[a I(\vec{x}_3) + (1-a)b \right]\left[a I(\vec{x}_4) + (1-a)b \right] \right> \nonumber \\
%&+& P_{1,4}(q) \left< \left[a I(\vec{x}_1) + (1-a)b \right] I(\vec{x}_2) I(\vec{x}_3) \left[a I(\vec{x}_4) + (1-a)b \right] \right> \nonumber \\
&+& 4 P^{square}_{1,2,3}(q) \left< \left[a I(\vec{x}_1) + (1-a)b \right] \left[a I(\vec{x}_2) + (1-a)b \right] \left[a I(\vec{x}_3) + (1-a)b \right] I(\vec{x}_4)\right> \nonumber \\
%&+& P_{2,3,4}(q) \left< I(\vec{x}_1) \left[a I(\vec{x}_2) + (1-a)b \right] \left[a I(\vec{x}_3) + (1-a)b \right] \left[a I(\vec{x}_4) + (1-a)b \right] \right> \nonumber \\
&+& P^{square}_{1,2,3,4}(q) \left< \left[a I(\vec{x}_1) + (1-a)b \right] \left[a I(\vec{x}_2) + (1-a)b \right] \left[a I(\vec{x}_3) + (1-a)b \right] \left[a I(\vec{x}_4) + (1-a)b \right] \right> \nonumber, \end{aligned}$$
where $P^{square}_{\o}(q)$ is the probability that none of the four corners of the square fall under the newly-deposited circle, $P^{square}_{i,\o}$ is the probability that only the $i^{th}$ corner falls under the newly-deposited circle, $P^{square}_{i,j}$ is that probability that only the $i^{th}$ and $j^{th}$ corners fall under the newly-deposited circle, and so on. The symmetries in the square configuration (all edges are equivalent, and all corners are equivalent) allow us to collapse the (equivalent) $P^{square}_{i,\o}$ terms, and similarly for the $P^{square}_{i,j}$ terms and the $P^{square}_{i,j,k}$ terms. We further note that terms like $P^{square}_{1,3}$ and $P^{square}_{2,4}$, which contain opposite corners of the circle, are omitted because it is impossible for a circle to cover diagonally opposite corners of the square without covering at least one other corner. The factors of $4$ in the above equation come in because there are 4 corners to a square, and 4 edges to a square, and ${ 4 \choose 3} = 4$ different ways to choose groupings of three of the four corners.
We can expand and simplify the averages of the products of the pixel values, as in the previous section, to find
$$\begin{aligned}
C^{square}_4(q) &=& P^{square}_{\o}(q) C^{square}_4(q) \\
&+& 4 \left< a \right > P^{square}_{1,\o} (q) C^{square}_4(q) \nonumber \\
&+& 4 P^{square}_{1,2}(q) \left[ \left< a^2 \right> C^{square}_4(q) + \left< (1-a)^2 \right> \left< b^2 \right> C_2(q) \right] \nonumber \\
&+& 4 P^{square}_{1,2,3}(q) \left[ \left< a^3 \right> C^{square}_4(q) + 2 \left< a(1-a)^2 \right> \left< b^2 \right> C_2(q) + \left< a(1-a)^2 \right> \left< b^2 \right> C_2(\sqrt{2}q) \right] \nonumber \\
&+& P^{square}_{1,2,3,4}(q) \left[ \left< a^4\right> C^{square}_4(q) + \left<a^2 (1-a)^2 \right> \left<b^2 \right> \left( 4 C_2(q) + 2C_2(\sqrt{2} q) \right) + \left< (1-a)^4 \right> \left<b^4 \right> \right] \nonumber, \end{aligned}$$
where the function $C_2(\cdot)$ is the 2-point function we discuss in Eq. 7.
Computing the probabilities $P^{square}_{\{ \cdot \}}$
------------------------------------------------------
To finish our calculation of the 4-point function for square geometries, we require the probabilities $P^{square}_{\o}(q)$, $P^{square}_{1,\o}(q)$, $P^{square}_{1,2}(q)$, $P^{square}_{1,2,3}(q)$, and $P^{square}_{1,2,3,4}(q)$.
For the calculation of $P^{square}_{1,2,3,4}(q)$, we first note that, given that one of the corners of the square falls under a newly-deposited circle (with diameter $s$), the probability that all 4 points fall under it is $g_4(q/s \in [0,1/\sqrt{2}]) = \frac{4}{\pi} \left[ \cos^{-1}(q/s) - (\pi/4) + (q/s)^2 - (q/s) \sqrt{1 - (q/s)^2} \right]$.
Using the same logic (and variable substitution) as in Eq. 6, we find that
$$\begin{aligned}
P^{square}_{1,2,3,4}(q) &=& \int_{0}^{\infty} \frac{s^2}{L^2} g_4(q/s)p(s) ds \\
&=& A \left(\frac{s_0}{L} \right)^2 \left( \frac{q}{s_0} \right)^{-(\alpha - 3)} B_4(\alpha), \nonumber \\\end{aligned}$$
where $B_4(\alpha) = \int_{\sqrt{2}}^{\infty} g_4(1/u) u^{2-\alpha} du$.
To derive $P^{square}_{1,2,3}(q)$, we seek the probability that 3 of the points, but not all 4, lie under the newly-deposited circle. If the two diagonal points are under the circle, so will at least one of the corners, and thus $P^{square}_{1,2,3}(q) = (P_2(\sqrt{2} q) - P^{square}_{1,2,3,4}(q))/2$, where $P_2(x)$ is the probability that two points a distance $x$ apart lie under a newly-deposited circle, and is calculated in Eqs. 5 and 6 (above).
The “doublet" probability $P^{square}_{1,2}(q)$ is the probability that 2, but not 3 or 4 of the points fall under the circle, and thus is given by $P^{square}_{1,2}(q) = P_2(q) - P^{square}_{1,2,3}(q) - P^{square}_{1,2,4}(q) - P^{square}_{1,2,3,4}(q) = P_2(q) - P_2(\sqrt{2}q)$. The “singlet" probability $P^{square}_{1,\o}(q)$ is the probability that 1, but not more, of the points fall under the circle, and is thus given by $P^{square}_{1,\o} (q)= P^{\star} - P^{square}_{1,2}(q) - P^{square}_{1,4}(q) - P^{square}_{1,2,3} (q) - P^{square}_{1,3,4}(q) - P^{square}_{1,2,4}(q) - P^{square}_{1,2,3,4}(q)$, where $P^{\star} = \left< s^2 \right> / L^2$ is the probability that a newly-deposited circle covers any given point, and is calculated in the previous sections. Simplifying this expression using our previously-derived results, we find that $P^{square}_{1,\o} (q)= P^{\star} - 2P_2(q) + \frac{1}{2} P_2(\sqrt{2} q) + \frac{1}{2} P^{square}_{1,2,3,4} (q)$.
Finally, the probability that none of the points falls under a newly-deposited circle is given by $P^{square}_{\o}(q) = 1 - \sum_i P^{square}_{i,\o}(q) - \sum_{i,j\ne i} P^{square}_{i,j} - \sum_{i,j\ne i,k \ne i,j} P^{square}_{i,j,k} - P^{square}_{1,2,3,4} = 1 - 4 P^{\star} + 4 P_2(q) - P^{square}_{1,2,3,4}(q)$.
Combining the pieces to find $C^{square}_4(q)$
----------------------------------------------
As in our calculation of the 4-point function for collinear points, we again consider the $q/s_0 \gg 1$ limit, in which we need only consider the lowest-order terms in $(q/s_0)^{-(\alpha-3)}$. In that limit, we find that
$$\begin{aligned}
C^{square}_4(q) &\approx& \frac{ B_4(\alpha) \left(\alpha-3 \right) \left< b^4 \right> \left< (1-a)^4 \right>}{4\left< 1-a \right>} \left( \frac{q}{s_0} \right)^{-(\alpha - 3)}.\end{aligned}$$
Like the other n-point functions computed thus far, the 4-point function for square geometries is a power law with power $-(\alpha-3)$, and it depends on opacity only as a multiplicative pre-factor. We note that, for $\alpha = 3.2$, $B(\alpha) \approx 4.014$, while $B_4(\alpha) \approx 3.581$, where these values come from numerical integration using Simpson’s method. These values are similar in magnitude, and thus the 4-point function is not inherently much smaller than the 2-point function.
Finally, we note that the 2- and 4-point functions depend differently on object opacity, and thus the visible difference in the different image ensembles likely arises from the relative amplitudes of these (power-law) functions, and not any difference in their functional forms.
Numerical analysis of the transmissive fallen-leaf images
=========================================================
To confirm our analytical calculations of the 2-point functions, we simulated 500-frame ensembles of $256 \times 256$ pixel images, using the procedure described in Eq. 1: circles of random size (following a power law distribution $p(s) \propto s^{-3.2}$ above the cutoff of $s_0 = 1$ pixel), brightness, and position were iteratively placed on the image frame to build up the images. For each frame, $10^6$ circles were deposited, which is the number required to cover the image surface $\sim 100$ times. To avoid edge effects, circle centers were allowed to fall up to $256 + s/2$ pixels away from the center of the image frame, where $s$ is the circle diameter in pixels. We used a large maximum circle size, $s_{max} = 10^8$ pixels, because prior work [@huang] on dead leaves models found that the functional form of the measured autocorrelation function approaches the analytically calculated curve only in the $ s_{\max} \to \infty $ limit. The heavy tail of the power-law distribution contains a non-negligible number of very large leaves, which contribute to the long-range correlations in the images.
{width="3.6in"}
We then measured the difference functions $D(q) = \left<| I(\vec{x}) - I(\vec{x'})|^2 \right> = 2\left<I(\vec{x})^2\right> -2C(q)$ for the image ensembles. $D(q)$ is clearly related to the autocorrelation function $C(q)$, but is easier to measure [@ruderman_scaling] as it is unaffected by the mean values of the individual images. We fit the measured difference functions to power law functions of the form $D(q) = \eta \times q^{\mu} + \nu$, as is suggested by our analytical calculations (Eq. 7). The best-fit parameters $(\eta,\mu,\nu)$ for the image ensembles with $a= \{0,0.25,0.5,0.75 \}$ were $(-0.48 \pm 0.01, -0.24 \pm 0.04, 0.69 \pm 0.03)$, $(-0.32 \pm 0.01, -0.23 \pm 0.03, 0.41 \pm 0.02)$, $(-0.191 \pm 0.004, -0.22 \pm 0.03, 0.23 \pm 0.01)$, $(-0.086 \pm 0.002, -0.21 \pm 0.02, 0.098 \pm 0.005)$, respectively, where the uncertainties represent $95\%$ confidence intervals. These values are in good agreement with the analytical calculations that predict $\mu = -0.2$ for all ensembles, and $\nu = \{ 0.66,0.396,0.22,0.094 \} $ for the ensembles with $a= \{0,0.25,0.5,0.75 \}$, respectively. The correlation functions shown (Fig. 2D) are the measured difference functions subtracted from the constants $\nu$ measured in the fit: $C(q) = [\nu - D(q)]/2$. These correlation functions are power-law functions of distance (linear on the log-log plot), and differ by a multiplicative constant. Similarly, the power spectra of the image ensembles (Fig. 2E), differ only by a multiplicative constant for low spatial frequencies, where the $q \gg s_0$ approximation holds.
Fig. 3 demonstrates that the 2-point function is affected substantially by leaf opacity for delta-function size distributions. In particular, the modulation depth of the “ripples" in the power spectra depend on the leaf opacity, and thus the opacity does not modify the power spectra simply by a multiplicative factor. The procedures used to generate the data shown in Fig. 3 were the same as for the power-law object size distribution, except for the different distribution of object sizes.
A more realistic model of radiological images
=============================================
![ **A shadowing dead leaves model with finite optical depth also exhibits scale invariant 2-point statistics for power law object size distributions.** (**A**) An example image from a dead leaves model with the same power law distributed leaf sizes as in Fig. 2, in which each leaf leaves a shadow by multiplying the brightness of the pixels it subtends by a factor no greater than one, drawn uniformly within $[0.5,1]$. Unlike the previous models, each pixel starts out at full brightness, and only a finite number of circles is added to generate the image. The autocorrelation function (**B**) and power spectrum (**C**) of this ensemble show scale invariance (for relatively low frequencies, which corresponds to $q \gg s_0$), just like the previous models.](fig4){width="3.6in"}
Our transmissive dead leaves model is not a perfect model for radiological images. Image formation in mammograms and other projectional radiographs results from the partial blockage of a roughly uniform illumination of x-rays due to local regions of dense tissue, unlike our dead leaves model. Moreover, imaged tissue is typically much thinner than the path length required to fully block the x-rays throughout the image, unlike the effectively infinite optical depth of our “additive" transparent dead leaves model (Eq. (1)). It is thus natural to ask whether our conclusions about variable object opacity generalize to these types of images. Analytically computing the 2-point statistics for this radiographic model is more involved than for the infinite depth models, since recursion is more complex in this case. For this reason, we chose to verify via simulation that the qualitative results from our analytical calculations hold for these types of images.
Fig. 4A shows a typical image from a shadowing dead leaves model with finite optical depth and the same power law leaf size distribution as in the previous models. To generate these model images, a uniform background illumination (of 1) was imposed across the whole image. Randomly sized and located circles were then deposited onto the image plane, with each leaf multiplying the brightness of the pixels it subtends by a factor drawn uniformly within $[0.5,1]$. The circle sizes were drawn from the same power law distribution as in the previous simulations, and the simulation code was thus very similar.
For an ensemble of these “radiographic" images, the empirically measured 2-point function (Fig. 4B) and power spectrum (Fig. 4C) exhibit the same power laws as we found for our previous models (Figs. 2,3), suggesting that our calculation holds more generally than for the specific model for which we performed the analytical calculations.
Intuitively, one might expect the same scale-invariant 2-point function for this model as for the previous one since no new length scale has been introduced.
Conclusions
===========
For the special case of power-law object size distributions, object opacity does not affect the form of either the 2- or 4-point functions, or the power spectrum of images: it is manifest only by a multiplicative constant in these power-law functions. Ours is the first analytic calculation that demonstrates these facts, and thus deepens our understanding of image statistics.
For object size distributions other than power-law, object opacity can (potentially dramatically) alter the low-level image statistics. Occlusion is important for natural image formation, but we find that it does not change the form of the power spectrum. Since images formed by opaque leaves that are all the same size have oscillatory, non-power-law, power spectra (Fig. 3), and transmissive leaves can yield power law power spectra (Figs. 2 and 4), occlusion is likely not responsible for scale invariance of images. We propose that the universality of power law power spectra in both occlusive imaging environments, such as natural photographic images, and transmissive ones, such as mammography, is likely due to power-law object size distributions in both settings.
JZ’s contribution to this work was supported by an international student research fellowship from the Howard Hughes Medical Institute (HHMI). This material is based upon work supported by a National Science Foundation Graduate Research Fellowship to DP under Grant No. DGE Ð 11-44155. MRD thanks the Hellman Family Foundation, the James S. McDonnell Foundation, and the McKnight Foundation for support.
[99]{}
Stephens, G.J., Mora, T., Tkačik, G., and Bialek, W. (2008). arXiv:0806.2694.
Ruderman, D.L. and Bialek, W. (1994). Phys. Rev. Lett. 73, 814-817.
Dong, D.W. and Atick, J.J. (1995). Network: Comput. Neural Syst. 6, 345-358.
Field, D.J. (1987). J. Opt. Soc. Am. A. 4, 2379-2394.
Simoncelli, E.P. and Olshausen, B.A. (2001). Annu. Rev. Neurosci. 24, 1193-1216.
Torralba, A. and Oliva, A. (2003). Network: Comput. Neural Syst. 14, 391-412.
van der Schaff, A. and van Hateren, J. (1996). Vis. Res. 36, 2759-2770.
Balboa, R.M. and Grzywacz, N.M. (2003). Vis. Res. 43, 2527-2537.
Heine, J.J. and Velthuizen, R.P. (2002). Med. Phys. 29, 647-661.
Li, H., M.L. Giger, O.I. Olopade, and M.R. Chinander (2008). J. Digit. Imaging 21, 145-152.
K.G. Metheany, C.K. Abbey, N. Packard, and J.M. Boone (2008). Med. Phys. 35, 4685-4694.
Barlow, H.B. (1961). In Sensory Communication, W.A. Rosenblith, ed. (Cambridge, MA: MIT Press), pp. 217-234.
Dong, D.W. and Atick, J.J. (1995). Network: Comput. Neural Syst. 6, 159-178.
Atick, J.J. and Redlich, A.N. (1992). Neural Comput. 4, 196-210.
Dan, Y., Atick, J.J. and Reid, R.C. (1996). J. Neurosci. 16, 3351-3362.
Zylberberg, J., Murphy, J.T. and DeWeese, M.R. (2011). PLoS Comput. Biol. 7, e1002250.
Rehn, M. and Sommer, F.T. (2007). J Comput. Neurosci. 22, 135-146.
Matheron, G. (1968). Modèle séquentiel de partition aléatoire. Tech. Rep., Centre de Morphologie Mathématique, Fontainebleau.
Bordenave, C., Gousseau, Y., and Roueff, F. (2006). Adv. in Appl. Probab. 38, 31-46. Ruderman, D.L. (1997). Vis. Res. 37, 3385-3398.
Carlson, C. R. (1978). Photographic science and engineering. 22, 69-71.
Balboa, R.M., Tyler, CW., and Grzywacz, N.M. (2001). Vis. Res. 41, 955-964.
Hsiao, W.H. and Milane, R.P. (2005). J. Opt. Soc. Am. A 22, 1789-1797.
Pitkow, X. (2010). J. Vis. 10, 42. Lee, A.B., Mumford, D. and Huang, J. (2001). Int. J. Comp. Vis. 41, 35-59.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We numerically investigate the dynamic heterogeneity and its length scale found in the coarse-grained ionic liquid model systems. In our ionic liquid model systems, cations are modeled as dimers with positive charge, while anions are modeled as monomers with negative charge, respectively. To study the effect of the charge distributions on the cations, two ionic liquid models with different charge distributions are used and the model with neutral charge is also considered as a counterpart. To reveal the heterogeneous dynamics in the model systems, we examine spatial distributions of displacement and time distributions of exchange and persistence times. All the models show significant increase of the dynamic heterogeneity as the temperature is lowered. The dynamic heterogeneity is quantified via the well-known four-point susceptibility, $\chi_4(t)$, which measures the fluctuation of a time correlation function. The dynamic correlation length is calculated by fitting the dynamic structure factor, $S_4(k,t)$, with Ornstein-Zernike form at the time scale at which the dynamic heterogeneity reaches the maximum value. Obtained time and length scales exhibit a power law relation at the low temperatures, similar to various supercooled liquid models. Especially, the charged model systems show unusual crossover behaviors which are not observed in the uncharged model system. We ascribe the crossover behavior to the enhanced cage effect caused by charges on the particles.'
author:
- Soree Kim
- 'Sang-Won Park'
- YounJoon Jung
bibliography:
- 'main\_v2.bib'
title: 'Heterogeneous dynamics and its length scale in simple ionic liquid models: A computational study'
---
Introduction
============
Room-temperature ionic liquids (RTILs) have attracted great attention because of their uncommon physical properties and various applications including non-toxic solvents, electrolytes, and supercapacitors.[@holbrey1999ionic; @wasserscheid2000ionic; @weingartner2008understanding; @deyoung2014graphene; @shim2012graphene] Widely known features of RTILs are their thermal stability, high polarity, high viscosity, very low vapor pressure, and low combustibility. [@weingartner2008understanding] Usually, RTILs are composed of bulky, asymmetric cations and small, symmetric anions. Due to their considerable size difference, RTILs exist in a liquid phase near the room temperature in spite of the presence of strong Coulomb interaction. One of the intriguing features of RTILs is their heterogeneous dynamics. As reported by theoretical[@park2015lifetime; @jeong2010fragility; @del2004structure; @wang2007understanding; @kim2014dynamic] and experimental studies,[@russina2012mesoscopic; @guo2011fluorescence; @hideaki2005physical] the evidences of the heterogeneous dynamics such as a non-exponential decay of correlation functions have been found. Computer simulation studies also have found the glassy dynamics of RTILs which is characterized by the breakdown of the Stokes-Einstein relation and decoupling of exchange and persistence events of defined excitations.[@park2015lifetime; @jeong2010fragility]
There have been previous computational studies on the description of the heterogeneous dynamics in the ionic liquid systems. [@urahata2004structure; @wang2005unique; @habasaki2008heterogeneous; @hu2006heterogeneity; @bhargava2007nanoscale] However, the length scale of the dynamic heterogeneity has not been investigated thoroughly[@pal2014slow] because of the difficulties on performing massive simulation with complex structures and long range interactions. To overcome this difficulty and to enhance computational efficiency, various levels of the coarse-grained models have been proposed.[@malvaldi2008molten; @roy2010dynamics; @del2004structure; @wang2007understanding; @jeong2010fragility] Among these models, we use simple models of RTILs which are introduced by a previous study[@malvaldi2008molten] and investigated intensively by Park et al..[@park2015lifetime] In the previous study, Ref , we examined the structural and dynamic properties of RTILs thoroughly. Especially, the relation between the time scale of the heterogeneous dynamics and the length scale of the structural relaxation were studied. According to the simulation results, calculated lifetime of the heterogeneous dynamics, ${\tau}_{\text{dh}}$, could be regarded as a distinctive time scale from the relaxation time. While the lifetime of the heterogeneous dynamics is calculated using the three-time correlation functions, the length scale was not examined. In the present study, we further investigate the heterogeneous dynamics of RTILs initiated in Ref . We show the simulation results that support the existence of the heterogeneous dynamics in our model systems. Furthermore, we present the simulation results on the length scale of the heterogeneous dynamics and the scaling relation between the relaxation time to show the distinctive nature of RTILs against the model without Coulomb interaction. The heterogeneous dynamics found in RTILs have similar aspects found in the supercooled liquids. When the liquids are cooled down rapidly, they exist in supercooled liquids rather than forming a crystal structure. The viscosity and the structural relaxation time grow dramatically as the temperature of the system is lowered. The complete understanding of these physical phenomena and their theoretical explanations are still lacking. Among the distinctive behaviors of the supercooled liquids, the correlations between the time-dependent local density fluctuations are found to play an important role in the slowing down of system. This phenomenon, typically called dynamic heterogeneity, has been investigated through diverse theoretical [@park2015lifetime; @choi2015dynamic; @kim2014dynamic; @jung2005dynamical; @jung2004excitation; @pan2005heterogeneity; @charbonneau2013decorrelation; @karmakar2014growing; @berthier2011theoretical; @berthier2005length; @berthier2004time; @kim2013dynamic; @lacevic2003spatially; @lavcevic2002growing; @glotzer2000time; @toninelli2005dynamical; @stein2008scaling; @hedges2007decoupling] and experimental studies.[@berthier2005direct; @wang1999long; @cicerone1995relaxation; @schmidt1991nature; @heuer1995rate; @ediger2000spatially]
In previous theoretical and computational studies, the time and length scales of the dynamic heterogeneity have been obtained using the four-point density correlation functions, [@dasgupta1991there; @chandler2006lengthscale; @flenner2014universal; @flenner2013dynamic; @flenner2011analysis; @flenner2010dynamic; @flenner2009anisotropic; @kim2013dynamic; @lacevic2003spatially; @lavcevic2002growing; @glotzer2000time; @toninelli2005dynamical] which has its origin in the study of spin glasses.[@young1997spin] A four-point correlation function is defined as $$\label{four-point}
\begin{split}
g_4(\mathbf{r},t)=\langle \delta\rho(0,0)\delta\rho(0,t)\delta\rho(\mathbf{r},0)\delta\rho(\mathbf{r},t) \rangle \\
-\langle \delta\rho(0,0)\delta\rho(0,t)\rangle\langle\delta\rho(\mathbf{r},0)\delta\rho(\mathbf{r},t) \rangle,
\end{split}$$ where $\delta\rho(\mathbf{r},t)$ is the deviation of the local density at the position $\mathbf{r}$ and at time $t$. $g_4(\mathbf{r},t)$ measures the correlation of relaxation of the density fluctuation between the two points separated by $\mathbf{r}$. The dynamic susceptibility and the dynamic structure factor can be derived from this function, by integrating and by performing Fourier transform, respectively. We use the dynamic susceptibility as an index of the quantification of the dynamic heterogeneity and define the time value that makes the dynamic susceptibility maximum as a characteristic time scale of the dynamic heterogeneity. The dynamic structure factor is also calculated to extract the dynamic length scale, $\xi_4(t)$, which could be interpreted as a length scale of dynamically correlated regions. With these schemes, we find the characteristic time scale and the length scale of the dynamic heterogeneity in the ionic liquid model systems. The contents of this article are organized as follows: In Section 2, we introduce ionic liquid models and describe shortly the simulation methods. In Section 3, the evidences of heterogeneous dynamics in the ionic liquid models are shown using displacement distributions and the decoupling of the mean exchange time and the mean persistence time. Furthermore, the dynamic length scale obtained by calculating the four-point correlation functions and related scaling behavior will be illustrated. Finally, the conclusions on this work are shown in the Section 4.
Models
======
We use simple coarse-grained models to investigate the heterogeneous dynamics of the room-temperature ionic liquid systems. In order to study the effect of charge distribution on the cation, the symmetrically charged model (SCM) and the asymmetrically charged model (ACM) are used. In addition, the uncharged model (UCM) is also used as a comparison group. Three model systems have the cation composed of two particles and the anion of single particle (Fig.\[model\_new\]). All the physical parameters are the same for those models except the charge distribution. SCM has $+0.5 e$ (where $e$ is the elementary charge) on each particle in the cation and $-1.0 e$ on the anion, while ACM has $+1.0 e$ on C1 particle, zero charge on C2 particle, and also $-1.0 e$ for the anion. UCM has zero charge for all particles. We also use the term “cation” and “anion” for the UCM, for convenience, even if the model does not have charges on the particles.
\[displacement\_dist\]
The total potential energy is given by the sum of the pairwise interactions, $$U_{\text{total}} =\sum_{i,j}\{ U_{\text{LJ}}(r_{ij})+U_{\text{Coulomb}}(r_{ij}) \}$$ where, $$U_{\text{LJ}}(r_{ij}) =4\epsilon_{ij}\left\lbrack \left(\frac{\sigma_{ij}}{r_{ij}}\right)^{12}-\left(\frac{\sigma_{ij}}{r_{ij}}\right)^{6}+\frac{1}{4}\right\rbrack H(r_{\text{cut}}-r_{ij}),$$ and $$U_{\text{Coulomb}}(r_{ij}) = \frac{1}{4\pi\epsilon_0}\frac{q_i q_j e^2}{r_{ij}}.$$ $H(r_{\text{cut}}-r_{ij})$ is the Heaviside step function, where the cutoff distance is set to be $r_{\text{cut}}=2^{1/6}\sigma_{ij}$. Note that $U_{\text{LJ}}(r_{ij})$ is purely repulsive and it is called the Weeks-Chandler-Andersen (WCA) potential.[@weeks1971role] In all of the three models, $\epsilon_{ij}=\epsilon=2$ kJ/mol and $\sigma_{ij}= \sigma=0.5$ nm for all $i,j$ pairs. The length of rigid bond between two particles (C1 and C2) in the cation is set to be $0.8\sigma$. The mass of the particles in the cation is $m=100$ amu and the mass of the anion is 200 amu, so the total mass of the cation and that of the anion are the same. We use the length, energy, and mass scaled by the units of $\sigma$, $\epsilon$, and $m$. The other units are converted by the following relations: unit time, $t_0=(m\sigma^2/\epsilon)^{1/2}=5$ ps, unit temperature, $T_0=\epsilon/k_{\text{B}}=240.5$ K, unit charge, $q_0=(4\pi \epsilon_0 \sigma \epsilon)^{1/2}=0.08484$ $e$, and unit pressure, $P_0 =262.2$ atm. We use 2048 pairs of RTIL molecules in a cubic simulation box of $L=17.88$, where $L$ is the length of each side. All the system have the reduced number density $\rho^{*}=\rho \sigma^3 =0.716$.
We perform molecular dynamics (MD) simulation using GROMACS 4.5 MD package program[@pronk2013gromacs] under $NVT$ ensemble condition with Nos$\acute{\text{e}}$-Hoover thermostat. Periodic boundary condition is applied to each direction and the finite size effect is carefully checked by comparing physical quantities calculated from systems with different size of 512, 1024, 2048 and 4096 RTIL pairs. For all the systems, ten independent trajectories are used and the length of production run is about 40 times of the $\alpha$-relaxation time of each system. The details of the molecular dynamics simulation conditions are given in Ref..
Results and discussion
======================
Heterogeneous dynamics
----------------------
Our ionic liquid model systems are expected to have the heterogeneous dynamics because of the size difference between the cation and the anion. To investigate the heterogeneous dynamics in detail, we first calculate the displacement of each particle. Fig.\[dis\_tscan\_scm\] shows the probability of the logarithm of displacements, $P[\text{log}_{10}(\delta r);t]$, of the cation in SCM. The time $t$ is set to be the $\alpha$-relaxation time, $t=\tau_{\alpha}$, at each temperature, where the $\alpha$-relaxation time is defined by the time at which the normalized overlap function, $Q(t)/N$, falls into $1/e$ (See Fig.\[q\_t\]). The definition of the overlap function will be introduced in Section 3.2. $P[\text{log}_{10}(\delta r);t]$ is related to the self-van Hove function, $G_s(\delta r;t)$, through the equation, $P[\text{log}_{10}(\delta r);t]=4\pi \text{log}(10){\delta r}^3 G_s(\delta r;t)$. Since $G_s(\delta r;t)$ follows a Gaussian function when the particle experienced the Fickian diffusion, $P[\text{log}_{10}(\delta r);t]$ would show a single peak. Therefore, the broadening or split of the distribution is a clear evidence for non-Fickian motion and heterogeneous dynamics.
In Fig.\[dis\_tscan\_scm\], the distribution of the cation in SCM is getting broader as the temperature is lowered. Compared to the anion case, the cation clearly shows more heterogeneous dynamics at lower termperatures. (See the supporting information for the data of the ACM, UCM cases.) The results is consistent with the previous study that demonstrates the cation moves faster than the anion.[@park2015lifetime] For the ACM case, the distribution of the anion is more heterogeneous than the cation. This opposite result comes from the different charge distribution of the cation. Compared to the SCM, the cations in the ACM make irregular structures around the anions because of their asymmetric charge distribution. As a result, it is expected that relatively small anion could have fast movement. In the UCM system, alternating structure of the cations and the anions are not observed, since there is no charge on the cations.[@park2015lifetime] The cage effect would be suppressed and the distributions of two particles show the similar results. The difference of two distributions are not profound but the anions have higher ratio of fast particles because of lower steric hindrance.
Fig.\[dis\_scm\] and Fig.\[g\_s\_cat\] show the time dependence of the probability distributions and the corresponding self-van Hove functions of the cations in SCM. A single peak at short time evolves into a broad distribution at $t\sim1\tau_{\alpha}$ and becomes a single peak again in the long time limit. From this results, we can infer that the dynamics are most heterogeneous at time around from $t\sim1\tau_{\alpha}$ to $t\sim5\tau_{\alpha}$. In the ACM and UCM, similar tendency is found for the calculated distributions. At long time limit, the self-van Hove function is approaching the Gaussian distribution shown as a dashed line in Fig.\[g\_s\_cat\], $G_s(\delta r;t)=(4\pi Dt)^{-3/2}\text{exp}(-\delta r^2/(4Dt))$, where $D$ is the diffusion coefficient. However, even at $t=50\tau_{\alpha}$, there exists a mismatch at small and large $r$, which means that the dynamic heterogeneity still remains at this long time case.
To characterize the dynamic heterogeneity in a different way, we calculate the excitation of each particle.[@jeong2010fragility; @hedges2007decoupling] The excitation is defined as an event that single particle $i$ moves more than distance $d$. For example, when particle $i$ moved more than $d$ at time $t_1$, $\lvert\mathbf{r}_i(t_1)-\mathbf{r}_i(0)\rvert>d$, than the first excitation takes place at $t=t_1$. Further, when particle moved more than $d$ from the $\mathbf{r}_i(t_1)$ after $t_2$, $\lvert\mathbf{r}_i(t_1+t_2)-\mathbf{r}_i(t_1)\rvert>d$, than the second excitation is at $t=t_1+t_2$. For the third excitation at $t=t_1+t_2+t_3$, same rule is applied, $\lvert\mathbf{r}_i(t_1+t_2+t_3)-\mathbf{r}_i(t_1+t_2)\rvert>d$, and so on. Using the series of the excitations, we can define two time scales, the persistence time and the exchange time. The persistence time is the time value that the first excitation occurs, so it is the set of all $t_1$ for every particle and trajectories. Another time scale, the exchange time, is defined by the waiting time between two excitations. It is the set of all $t_2, t_3, ...$. It has been known that the decoupling of the two time scales occurs when the system is dynamically heterogeneous.[@jeong2010fragility; @hedges2007decoupling; @jung2005dynamical] Jung et al. first calculated the relationship of the two time scales in the kinetically constrained model (KCM) and showed that the dynamic heterogeneity provokes the decoupling of the two time scales.[@jung2005dynamical] Although the definition of the excitation is different from the case of the KCM, the physical meanings are similar in the ionic liquids systems and it can be applied for studying the heterogeneous dynamics in those systems. This analysis scheme has been applied to study the heterogeneous nature of the supercooled liquids, the ionic liquids, and the ring polymer melts.[@jung2005dynamical; @hedges2007decoupling; @jeong2010fragility; @lee2015slowing]
Before analyzing the decoupling of the two time scales, we visualize the heterogeneous dynamics using the persistence time of each particle. Fig.\[het\_ca\_scm\] shows the spatial distributions of the persistence times. The radius of each particle represents the size of the persistence time in logarithm scale. For the cufoff distance $d$, we used $d=1$ which is comparable to the size of the particles. At the high temperature, Fig.\[het\_ca\_scm\_high\], the persistence time distribution is relatively homogeneous than that at the low temperature, Fig.\[het\_ca\_scm\_low\]. Note that not only the size of sphere is heterogeneous at the low temperature but the spatial distribution also shows the heterogeneity. This means that there is a correlation between the slow particles and this could be the evidence of growing dynamic length scale.
Now, we investigate the decoupling of the two time scales and relate this phenomenon with the breakdown of the Stokes-Einstein relation. When the probability distribution of the exchange time is exponential, the distribution of the two time scales is identical because the persistence time is related with the exchange time through the integral.[@jung2005dynamical] At high temperatures, there is a weak correlation between the excitation events, which indicates that excitation events follow the Possion process. At low temperature, however, the correlations between the excitation is pronounced and the excitation events would experience non-Possion process that results in decoupling of the persistence time and the exchange time distributions, Fig.\[per\_exc\_scm\] and Fig.\[per\_exc\_scm\_ani\]. We can interpret this phenomenon that the correlation between excitations increases because of the clustering of the slow particles. The mean values of the two time scales also show the decoupling. The mean persistence time and the mean exchange time, which are defined by the ensemble average of the persistence times, $\tau_{\text{p}}=\langle t_{\text{p}}\rangle$, and the exchange times, $\tau_{\text{e}}=\langle t_{\text{e}}\rangle$, are related to the transport coefficients, the relaxation time and the diffusion coefficient, respectively.[@jeong2010fragility] It has been known that the mean persistence time would be proportional to the relaxation time, $\tau$, when $d$ is comparable with $2\pi/q$ where $q$ is the first peak position of the structure factor. Furthermore, the exchange event is governed by the diffusion of particle so that $1/\tau_{\text{e}}$ would have relation between diffusion constant, $D$, through power law relation. Since the ionic liquid system is a kind of fragile liquids,[@park2015lifetime; @jeong2010fragility] it has been shown that there is a sublinear relation between $1/\tau_{\text{e}}$ and $D$, irrespective of $d$.[@jeong2010fragility] This is due to the correlation between the excitation events.[@jung2005dynamical] The power law relations between $\tau_{\text{p}}$ and $\tau_{\alpha}$, and the relations between $\tau_{\text{e}}$ and $1/D$ are shown in supporting information. Length scale $d$ dependence is also investigated in FIG.S7. Fig.\[violation\] shows the ratio of $\tau_{\text{p}}$ and $\tau_{\text{e}}$, which shows similar divergent behavior of $D\tau$ at low temperatures.[@park2015lifetime] In addition, there are power law relations, $\tau_{\text{e}}\sim\tau_{\text{p}}^{\nu}$, between the two physical time scales as it can be seen in Fig.\[per\_exc\_power\]. The value of the power law exponents are 0.81 (SCM-cation), 0.85 (SCM-anion), 0.78 (ACM-cation), 0.77 (ACM-anion), 0.89 (UCM-cation) and 0.83 (UCM-anion). The exponent of the SCM cation is analogous to that of the coarse-grained ionic liquids system which is 0.80 for the cations.[@jeong2010fragility] Comparing the values of the exponents, we can infer that the fragility of the system increases in the order of UCM, SCM and ACM. More detailed informations can be found in the supporting information.
In this section, we confirmed that the heterogeneous dynamics found in the supercooled liquids system are also found in our ionic liquids model systems. From the displacement distribution, we find the clue that the mobility of the particles are heterogeneously distributed. However, the correlations between slow or fast particles can not be obtained from this analysis. In order to observe this correlated behavior, a four-point correlation function is introduced to calculate the dynamic length scale in the next section. It is noteworthy that the distribution of the charge on the cations not only changes the local structure but also affects the fragility of the whole system. From the scaling analysis, we find that the difference between the cation and the anion is smaller compared to the differences between the model systems. Thus, we concentrate on the differences between the models rather than the type of ions.
The dynamic susceptibility and the dynamic structure factor
-----------------------------------------------------------
In the previous studies of glassy dynamics, there have been several different schemes to define the length scale of the dynamic heterogeneity. [@donati1999spatial; @donati2002theory; @lavcevic2002growing; @lacevic2003spatially; @chandler2006lengthscale; @flenner2011analysis; @kim2013multiple] Among them, the dynamic length scale defined from the four-point correlation function has been widely adopted for many systems. The four-point correlation function given in Eq.\[four-point\] measures the correlation of the relaxation of the density fluctuation. Since the dynamic susceptibility is obtained by integrating $g_4(\mathbf{r},t)$ over the space, the dynamic susceptibility can be regarded as a volume of the correlated motion. Furthermore, the dynamic susceptibility can be written in the form of the fluctuation of the dynamic quantity as will be shown below. We employ this framework that has been previously established and applied to analyze the supercooled liquids systems.[@lacevic2003spatially; @lavcevic2002growing; @glotzer2000time]
The dynamic susceptibility, $\chi_4(t)$, is defined as, $$\label{eq-chi4}
\chi_4(t)=\frac{1}{N}\lbrack \langle Q(t)^2 \rangle- \langle Q(t) \rangle^2 \rbrack,$$
$$%Q(t)=\sum_{i=1}^{N}\sum_{j=1}^{N}w(\vert \mathbf{r}_i(0)-\mathbf{r}_j(t) \vert),
Q(t)=\sum_{i=1}^{N}w(\vert \mathbf{r}_i(0)-\mathbf{r}_i(t) \vert),
\label{eq-q_t}$$
where $w(\vert \mathbf{r}_i(0)-\mathbf{r}_i(t) \vert)$ is a overlap function which is 0 when $\vert \mathbf{r}_i(0)-\mathbf{r}_i(t) \vert > a$ and 1 when $\vert \mathbf{r}_i(0)-\mathbf{r}_i(t) \vert \le a$. $Q(t)$ counts the number of self overlapping particles using the configurations separated by a time interval $t$. Thus, $Q(t)/N$ could be regarded as an index how much the system has been relaxed. Fig.\[q\_t\] shows that $Q(t)/N$ decays from $1$ to $0$ as the time is passed, showing similar functional behavior with the self-intermediate scattering function. In Fig.\[q\_t\], we find that $Q(t)/N$ in charged systems shows more stretched form compared to that in the UCM. The difference between the model systems comes from the different local environment that the particles experience. Detailed discussion will be given in Section 3.3. Eq.\[eq-chi4\] tells that $\chi_4(t)$ can be interpreted as a quantification of a fluctuation of $Q(t)$. In this study, we choose overlap cutoff $a=0.3$ as used in other studies.[@glotzer2000time; @lavcevic2002growing] The $\alpha$-relaxation time, $\tau_{\alpha}$, can be defined as the time at which $Q(\tau_{\alpha})/N=1/e$. This definition gives analogous value of $\tau_{\alpha}$ with the result of the conventional use of self-intermediate scattering function. Additionally, for the cations, $\chi_4(t)$ calculated using the coordinates of the center of mass and the particle itself did not show distinctive difference for the scaling law. We will use the simulation data obtained by calculations with center of mass for each cation.
Fig.\[dyn\_sus\] shows the dynamic susceptibility calculated using the Eq.\[eq-chi4\] at various temperatures. At a fixed temperature, $\chi_4(t)$ increases as the time passes. $\chi_4(t)$ has a maximum peak at a certain time scale, namely $t_{4}^{*}(T)$, which is comparable to the relaxation time, $\tau_{\alpha}(T)$. We find that $t_4^*$ is proportional and almost equal to $\tau_{\alpha}$ for all the three model systems (See the supporting information). This time scale, $t_4^*$, which shows the maximum value of the dynamic susceptibility, $\chi_4^*=\chi_4(t_4^*)$, is defined as a characteristic time scale of the dynamic heterogeneity. As the time increases further, $\chi_4(t)$ decreases to zero. The functional form of $\chi_4(t)$ confirms that the dynamic heterogeneity is transient in time. As the y-axis in the inset of Fig.\[dyn\_sus\] is on log scale, $\chi_4^*$ would show power law relation with $t_4^*$. Note that the maximum values, $\chi_4^*$, are marked with black dots. We find the power law relations between $t_4^*$ and $\chi_4^*$ for all systems. Moreover, the crossover behaviors at short time regime are found for SCM and ACM. The existence of crossover behavior is a unique phenomenon in our model compared to the models of supercooled liquids. This phenomenon is based on the enhanced separation of sub-diffusive regime and diffusive regime for charged model. The details about the crossover behavior will be discussed in the next section.
An alternative definition of the dynamic susceptibility can be used,[@chandler2006lengthscale] $$\tilde{\chi}_4(k,t)=\frac{1}{N}\lbrack \langle \tilde{Q}(\mathbf{k},t)^2 \rangle- \langle \tilde{Q}(\mathbf{k},t) \rangle^2 \rbrack,
\label{eq-chi4-isf}$$
$$\tilde{Q}(\mathbf{k},t)=\sum_{l=1}^{N}e^{i{\mathbf{k}}\cdot(\mathbf{r}_l(t)-\mathbf{r}_l(0))},
\label{eq-q-isf}$$
where $k=\vert\mathbf{k}\vert$. Here, $k$ is a wavevector that regulates the length scale of local area and has the similar role of $a$ in the overlap function. Using the Eq.\[eq-chi4-isf\] and Eq.\[eq-q-isf\], $\tilde{\chi}_4(k,t)$ can be expressed in the form, $(1/N) \sum_{j=1}^{N} \sum_{l=1}^{N} \langle {\delta}\tilde{Q}_{j}(\mathbf{k},t) {\delta}\tilde{Q}_{l}(-\mathbf{k},t) \rangle$, where ${\delta}\tilde{Q}_{j}(\mathbf{k},t)=e^{i{\mathbf{k}}\cdot(\mathbf{r}_j(t)-\mathbf{r}_j(0))}-\langle e^{i{\mathbf{k}}\cdot(\mathbf{r}_j(t)-\mathbf{r}_j(0))}\rangle$. In this form, the self part of $\tilde{\chi}_4(k,t)$ can be easily obtained applying $j=l$ condition, $\tilde{\chi}_4^{\text{self}}(k,t)=1-F_s(k,t)^2$, where $F_s(k,t)=\langle e^{i{\mathbf{k}}\cdot(\mathbf{r}_j(t)-\mathbf{r}_j(0))}\rangle$ is the self-intermediate scattering function. Fig.\[dyn\_sus\_isfs\] shows the $\tilde{\chi}_4(k,t)-\tilde{\chi}_4^{\text{self}}(k,t)$ for the cation and the anion in SCM at various temperatures. The magnitude of the wavevector is set by $k=2{\pi}/{\lambda}_{\text{max}}$, where ${\lambda}_{\text{max}}$ is the shortest peak position of radial distribution function between the cation and the anion. The overall functional form of $\tilde{\chi}_4(k,t)-\tilde{\chi}_4^{\text{self}}(k,t)$ is similar with $\chi_4(t)$, while the peak at short time scale is more pronounced. The crossover behavior for the maximum point of $\tilde{\chi}_4(k,t)$ also can be found in the inset of Fig.\[dyn\_sus\_isfs\].
In order to obtain the length scale of the dynamic heterogeneity, we calculate the dynamic structure factor with the follwing equations, $$S_4(q,t)=\frac{1}{N}\lbrack \rho(\mathbf{q},t)\rho(\mathbf{-q},t)\rbrack,
\label{eq-str4}$$ $$\rho(\mathbf{q},t)=\sum_{i=1}^{N}\text{exp}[i\mathbf{q}\cdot\mathbf{r}_{i}(0)]w(\vert \mathbf{r}_i(0)-\mathbf{r}_i(t) \vert),
%\rho(\mathbf{q},t)=\sum_{i=1}^{N}\sum_{j=1}^{N}\text{exp}[i\mathbf{q}\cdot\mathbf{r}_{i}(0)]w(\vert \mathbf{r}_i(0)-\mathbf{r}_j(t) \vert),$$ where $q=\vert \mathbf{q} \vert$. The dynamic correlation length, $\xi_4(t)$, is obtained by fitting the small wavevector regime into the Ornstein-Zernike (OZ) equation, $$S_4(q,t)=\frac{S_4(q=0,t)}{1+(q\xi_4(t))^2},$$ where, $S_4(0,t)$ and $\xi_4(t)$ are fitting parameters. $S_4(q,t)$ is fitted in the regime of $q\le1.5$ which is corresponding to the condition, $4.19\le r\le17.88$. Obtained correlation length is shown in Fig.\[dyn\_length\] at various temperature and time. Similar to the dynamic heterogeneity, the correlation length is also transient in time. At first, the correlation length is growing until it reaches the maximum, and then it decreases. The functional form of the correlation length resembles that of $\chi_4(t)$, however, the time values that the peaks occur are not identical. Analogous to the previous studies on the supercooled liquids, the time for which $\xi_4(t)$ is maximum is larger than $t_4^*$.[@toninelli2005dynamical; @flenner2011analysis] From this result, we find that the characteristic time scale of the correlation length is longer than the time scale of the dynamic heterogeneity.
In order to determine the dynamic correlation length of the system at a fixed temperature, we use $\xi_4^*=\xi_4(t=t_4^*)$. $\xi_4^*$ represents the dynamic correlation length when the dynamic heterogeneity is maximum. All the dynamic structure factors collapse into single functional form of $f(x)=1/(1+x^2)$, when the x-axis is $q\xi_4^*$ and y-axis is $S_4(q,t_4^*)/S_4(0,t_4^*)$. (See FIG.S11 in the supporting information) The definition of the dynamic correlation length can be varied using different empirical functions. We compare various correlation length obtained by fitting $S_4(q,t_4)$ into functions: (1)$S_4(q,t_4^*)=S_4(0,t_4^*)/(1+(q\xi_4^{(1)})^2)$; (2)$S_4(q,t_4^*)=(S_4(0,t_4^*)-C)/(1+(q\xi_4^{(2)})^2)+C$; (3)$S_4(q,t_4^*)=(S_4(0,t_4^*)-C)/(1+(q\xi_4^{(3)})^{\zeta})+C$; (4)$S_4(q,t_4^*)=(S_4(0,t_4^*)-C)/(1+(q\xi_4^{(4)})^2+(q\xi_4^{(4)})^4)+C$. Here, the $C$ and $\zeta$ are fitting parameters and $C$ is included to improve the fitting against the baseline problem.[@toninelli2005dynamical] The fitted function to the data is shown in Fig.\[str4\_fit\_0.56\] for the cation in SCM at T=1.16.
Fig.\[dyn\_length\_compare\_Temp\] shows the temperature dependence of these length scales for the cation in SCM. Among these correlation lengths, $\xi_4^{(1)}$ grows faster than the other lengths and shows clear crossover behavior. Moreover, the scaling behavior of $\xi_4^{(1)}$ and $t_4^*$ reveals most reasonable power law exponent. In this sense, we use $\xi_4^{(1)}$ as a dynamic correlation length ,$\xi_4^*$ , in the rest part of this article.
[c|M[2.2cm]{}|M[2cm]{}|M[2.4cm]{}|M[2.6cm]{}]{} & $\nu$ & $\nu'$ & $\beta$ & $\eta$\
$d=1$ & 1.063 $\pm$ 0.007 & 1.06 $\pm$ 0.02 & 1.43 $\pm$ 0.02 &[-0.09 $\pm$ 0.04]{}\
$d=2$ & 0.50 $\pm$ 0.03 & 0.57 $\pm$ 0.02 & 1.36 $\pm$ 0.02 &[-0.51 $\pm$ 0.02]{}\
$d=3$ & 0.30 $\pm$ 0.01 & 0.40 $\pm$ 0.02 & 1.31 $\pm$ 0.02 &[-0.58 $\pm$ 0.06]{}\
$d=4$ & 0.24 $\pm$ 0.02 & 0.34 $\pm$ 0.02 & 1.27 $\pm$ 0.04 &[-0.64 $\pm$ 0.09]{}\
& $\xi \propto c^{-\nu}$ & $\xi' \propto c^{-\nu'}$ & $\mathbf{C}(\frac{r}{\xi},\tau) \propto$ $ exp(-(r/\xi))^{\beta}$ &$\mathbf{S}(q\xi') \propto$ $1/(1+(q\xi')^{2-\eta})$\
-- -------- ------- -------- ------- ------ ------ --
high T low T high T low T
cation 0.51 0.24 0.61 0.33 0.48
anion 0.50 0.25 0.71 0.33 0.48
cation 2.1 4.8 2.0 4.1 2.5
anion 1.9 4.5 1.6 3.8 2.7
-- -------- ------- -------- ------- ------ ------ --
Scaling laws
------------
We now investigate the power law relations between the dynamic physical quantities we calculated. In various model systems of the supercooled liquids, the scaling law has been found for the dynamic length and time scales.[@stein2008scaling; @whitelam2005renormalization; @berthier2005numerical; @whitelam2004dynamic; @biroli2006inhomogeneous] It is noted that these kinds of relations are originally found in the critical behavior of phase transitions. In our ionic liquid model system, a similar power law relation is discovered. First, we show the relation between $t_4^*$ and $\chi_4^*$ in Fig.\[pow\_t4\_chi4\]. For all three systems, the power law relation, $\chi_4^* \sim {t_4^*}^{\theta}$ is found. Another power relation for $t_4^*$ and $\xi_4^*$ is also found, $t_4^* \sim {\xi_4^*}^{\gamma}$, as shown in Fig.\[pow\_t4\_xi4\]. The power law exponents are listed in Table \[table\_exponent\]. As previously observed, there is clear crossover behavior. Interestingly, this crossover behavior is prominent in SCM and ACM which include the charge on the particles. Kim et al. also found the crossover behavior for a glass-forming binary soft-sphere mixture, and addressed that this is due to different physical behaviors at different time scales, which are $\beta$-relaxation time, $\tau_{\beta}$, and $\alpha$-relaxation time, $\tau_{\alpha}$.[@kim2013dynamic] In the article, $\tau_{\beta}$ is defined by the minimum value of $\text{d}(\text{ln}\langle \delta \mathbf{r}^2(t) \rangle)/\text{d}(\text{ln} t)$, where $\delta \mathbf{r} = \mathbf{r}(t)-\mathbf{r}(0)$. This means that $\tau_{\beta}$ is the time value of the plateau of mean-squared displacement at each temperature. The mean-squared displacement and its derivative of the cation in SCM is shown in Fig.\[cal\_tau\_beta\]. As the temperature is lowered, $\tau_{\beta}$ increases until it reaches maximum value of $t=2$. Note that our definition of $\tau_{\alpha}$ and $\tau_{\beta}$ allows $\tau_{\beta}$ to be longer than $\tau_{\alpha}$ at the high temperatures.$\tau_{\alpha}$ and $\tau_{\beta}$ are also marked in $Q(t)/N$, Fig.\[q\_t\]. Note that $\tau_{\alpha}$ is a time scale of the structural relaxation, while $\tau_{\beta}$ is a time scale of plateau of $Q(t)/N$. In this sense, $\tau_{\beta}$ can be interpreted as a characteristic time scale that particles stay in the cage. When the temperature is high, ${\beta}$-relaxation regime is not clearly observed in the time correlation function $Q(t)/N$. However, at the low temperature, slowing down of local dynamics due to the cage effect makes ${\beta}$-relaxation regime distinctive. The onset temperature of this phenomenon is related to the onset of the crossover behavior. When $\xi_4(t)$ is calculated at $t={\tau}_{\beta}$, the cations and the anions of SCM and ACM have the power law exponent ${\gamma} \sim 2$ (Fig.\[pow\_t4\_xi4\]). However, when $t=t_4^*\sim\tau_{\alpha}$, the exponent is much larger at low temperatures. Such phenomenon is not found in the UCM system since the cage effect of uncharged system is weak compared to the charged systems. The evidence of enhanced cage effect in SCM and ACM can be found in the behavior of $Q(t)/N$, Fig.\[q\_t\]. For SCM and ACM, $Q(t)/N$ show highly stretched form and the plateau is clearly observed at short time. However, the plateau in $Q(t)/N$ of UCM is not profound compared to SCM and ACM cases. From this observation, we argue that the existence of the charges on the particles strengthens the cage effect around certain particle. Furthermore, this enhanced cage effect causes the crossover behavior in the power law relation. This crossover behavior observed more profoundly in SCM and ACM can be thought as a distinguishing property of ionic liquids in our model systems. When we compare the values of the exponent ${\gamma}$, we find ${\gamma}_{SCM}^{C} \sim 4.8$, ${\gamma}_{ACM}^{C} \sim 4.1$, and ${\gamma}_{UCM}^{C} \sim 2.5$. The exponents of the anions are similar to those of the cations. The exponent value of UCM is similar to the Lennard-Johns mixture studied by La$\check{\text{c}}$evi$\acute{\text{c}}$ et al.[@lacevic2003spatially] It is notable that the lifetime of the dynamic heterogeneity calculated from Ref has the same exponent, $\zeta_{\text{dh}}$, in all the models in terms of $\tau$, $\tau_{\text{dh}} \sim \tau^{\zeta_{\text{dh}}}$.[@park2015lifetime] Fig.\[xi\_t4\_together\] clearly shows that the $t_4^*$ of SCM and ACM increases much faster than the $t_4^*$ of UCM when $\xi_4^*$ is increased. As we can interpret $\xi_4^*$ as a size of the dynamic cluster, the same size of the dynamic cluster is preserved longer in time for charged systems. This is because the local structure of charged system is highly ordered compared to the structure of uncharged system. Meanwhile, the difference between two charged system, SCM and ACM, is not profound. In spite of the different charge distributions of the cation, the crossover behavior is very similar for two systems and the exponents of the power law between the correlation length and the characteristic time scale of the dynamic heterogeneity have similar values.
Finally, we demonstrate the three fitting schemes to find the relation between $t_4^*$ and $\xi_4^*$. These fitting schemes have different theoretical bases. First, mode-coupling theory predicts there is a power law relation between two quantities as we already observed, [@biroli2004diverging; @biroli2006inhomogeneous; @szamel2008divergent] $t_4^* \sim {{\xi}_4^*}^{\gamma}$. Second, the Random-First-Order Transition (RFOT) theory suggests the exponential relation,[@kirkpatrick1989scaling; @lubchenko2007theory] $t_4^* \sim \text{exp}({\xi_4^*}^z)$. Lastly, the view of the facilitation picture suggests the following relation,[@keys2011excitations] $t_4^* \sim \text{exp}(A(\text{log}({\xi_4^*}/B))^2)$. Fig.\[xi\_t4\_together\] shows the functions fitted on the data of the cation of SCM using different schemes. The lower fitting range is set to be $0.6$ for all functional forms. It seems that the power law relation is most appropriate to describe the data at long length scale. The other two functions show similar exponential behavior. At short length scales, it seems that the exponential function well matches to the data. However, the data at short length scales are governed by a different physical environment and it may be a coincidence that the functions agree with the data. Note that different behavior of short length regime is due to the cage effect of ionic liquid model. It is notable that Flenner et al. reported that there is a universal behavior of supercooled liquids which is an exponential relation between $t_4^*$ and $\xi_4^*$.[@flenner2014universal]
Conclusions
===========
In the previous works, heterogeneous dynamics has been found in RTILs using the theoretical schemes which are applied to supercooled liquids. In this study, we find the evidences of heterogeneous dynamics in the simple ionic liquids models. As the temperature decreases, displacement distribution of the cation and the anion is getting broaden as shown in Fig.\[dis\_tscan\_scm\] and Fig.\[dis\_tscan\_scm\_ani\]. It seems that the broadening of the distribution reaches its maximum when the time is comparable to the relaxation time of each model system. Furthermore, the decoupling of the mean exchange time and the mean persistence time is observed. At low temperatures, the mean persistence time is growing much faster than the mean exchange time as expected from the previous studies.[@jung2005dynamical; @jeong2010fragility] From the result, we can infer that the defined excitation events are correlated each other at sufficiently low temperatures, which is caused by the correlated local motion of the particles.
We adopt the four-point correlation function analysis to study how the local densities are correlated in our ionic liquids model systems. To quantify the heterogeneous dynamics, the dynamic susceptibility is calculated based on two different time-correlation functions which are $Q(t)/N$ and $F_s(k,t)$. The time dependence and the temperature dependence of calculated dynamic susceptibility, $\chi_4(t)$ and $\tilde{\chi}_4(t)$, are investigated. Our results illustrate that the dynamic heterogeneity found in RTILs is transient in time analogous to the situation in the supercooled liquids.
![ Comparison of relations between $t_4^*$ and $\xi_4^*$ for SCM, ACM and UCM. The power law exponent, $\gamma$, of the cation and the anion in SCM and ACM is much larger than the exponent of cation and anion in UCM. Three different functional form is used to fit the data of the cation in SCM: (1) $t_4^* \sim {{\xi}_4^*}^{\gamma}$, (2) $t_4^* \sim \text{exp}({\xi_4^*}^z)$, (3) $t_4^* \sim \text{exp}(A(\text{log}({\xi_4^*}/B))^2)$. []{data-label="xi_t4_together"}](scale_xi_t4_fit.eps){width="50.00000%"}
We also successfully extract the dynamic correlation length, $\xi_4^*$, by fitting the dynamic structure factor, $S_4(t)$, into the Ornstein-Zernike equation. Calculated quantities such as the characteristic time scale of the dynamic heterogeneity, $t_4^*$, the maximum value of the dynamic susceptibility, $\chi_4^*$, and the dynamic correlation length, $\xi_4^*$, are connected via the power law relations at low temperatures. Interestingly, the crossover behavior around $t_4^* \sim 1$ and $\xi_4^* \sim 1$ is prominent in the charged model, SCM and ACM. We count this phenomenon on the enhanced cage effect due to the existence of charge. The crossover behavior and the peak of the dynamic susceptibility in the short time region $t \sim \tau_{\beta}$ are not noticeable in the UCM and in the previous studies on the glassy systems. Note that, in the Ref , the power law relations between the lifetime of the dynamic heterogeneity, $\tau_{\text{dh}}$, and $\tau$ was investigated. The power law exponents, $\zeta_{\text{dh}}$, were found to be the same in all the models studied in this work. In this study, on the other hand, the power law exponents related with the length scale of the dynamic heterogeneity, $\theta$ and $\gamma$, show different values depending on the models.
As we vary the charge distributions on the cation particle, the effect of different charge distributions on the glassy dynamics is observed. When the result of UCM is compared with the charged model systems, all three models show the heterogeneous dynamics at sufficiently low temperatures. Even if the particles do not have the charge, the mixture of different shape of particle shows the glassy behavior like other supercooled liquids models. The existence of charge on the particles mainly affects two aspects of the system. First, the onset temperature of heterogeneous dynamics increases. Because the structures of the charged systems are more stable than the structure of UCM at the same temperature, the dynamics is much slower and the temperature which shows heterogeneous dynamics is much higher. Second, the cage effect is enhanced. From the simulation results of dynamic susceptibility and the power law analysis, we confirm that the alternating local structure of the cations and the anions results strong cage effect. As a result, we find crossover behavior for the power law relation between the time scale and the length scale of the dynamic heterogeneity. Furthermore, the comparison between SCM and ACM reveals that the asymmetric charge distribution makes the system more fragile. However, two models basically show similar behaviors except the onset temperature of the dynamic heterogeneity.
The simple models we used are designed to reveal heterogeneous dynamics which can be observed in RTILs. Because of their simplicity, a clear comparison of different models is possible and relatively long and large simulation is available compared to the all-atom models of RTILs. While the models can provide the insight on the role of the charge distributions on the cations, they are highly coarse-grained models so that the effect of the molecular details are ignored. In order to extend the models to more realistic systems, detailed molecular structure can be considered. In the future study, less coarse-grained model such as 4-atom cation model[@jeong2010fragility] may be used for investigation of the dynamic heterogeneity in RTILs.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'From a generalization to $\Z^n$ of the concept of congruence we define a family of regular digraphs or graphs called multidimensional circulants, which turn out to be Cayley (di)graphs of Abelian groups. This paper is mainly devoted to show the relationship between the Smith normal form for integral matrices and the dimensions of such (di)graphs, that is the minimum ranks of the groups they can arise from. In particular, those $2$-step multidimensional circulants which are circulants, that is Cayley (di)graphs of cyclic groups, are fully characterized. In addition, a reasoning due to Lawrence is used to prove that the cartesian product of $n$ circulants with equal number of vertices $p>2$, $p$ a prime, has dimension $n$.'
author:
- |
M.A. Fiol\
\
[Universitat Politècnica de Catalunya, BarcelonaTech]{}\
[Departament de Matemàtica Aplicada IV]{}\
[Barcelona, Catalonia]{}\
[(e-mail: [[email protected]]{})]{}
title: On Congruence in $Z^n$ and the Dimension of a Multidimensional Circulant
---
Introduction
============
Throughout this paper we make use of standard concepts and terminology of graph theory and group theory, see for instance [@r6] and [@r15] respectively. We will recall here the most relevant definitions. Let $\Gamma$ be a group with identity element $e$, and let $A\subseteq \Gamma\setminus \{ e\}$ such that $A^{-1} = A$. The Cayley graph of $\Gamma$ with respect to $A$, denoted by $G(\Gamma ;A)$, is the graph whose vertices are labelled with the elements of $\Gamma$, and an edge $(u,v)$ if and only if $u^{-1}v \in
A$. The Cayley digraph $G(\Gamma ;A)$ is defined similarly, but now we do not require $A^{-1} = A$. Since left translations in $\Gamma$ are automorphisms of $G(\Gamma ;A)$, a Cayley graph is always vertex-transitive. Moreover, the group of such automorphisms, $\Aut G(\Gamma ;A)$, contains a regular subgroup (that is a transitive subgroup whose order coincides with the order of the graph) isomorphic to $\Gamma$. In fact, Sabidussi [@r16] first showed that this is also a sufficient condition for a graph to be a Cayley graph (of the group $\Gamma$). Clearly, the same statements also hold for Cayley digraphs. In particular, this result or its digraph analog will be simply referred to as [*Sabidussi’s result*]{}.
Because of both theoretical and practical reasons, the class of Cayley (di)graphs obtained from Abelian groups have deserved special attention in the literature. This is the case, for instance, for circulant (di)graphs the definition of which follows. Let $m$ be a positive integer, and $A =\{ a_{1} ,a_{2} , \ldots ,a_{d} \}
\subseteq \Z/m\Z\setminus \{ 0\}$. The [*$($d-step$)$ circulant digraph*]{} $G(m;A)$ has as set of vertices the integers modulo $m$, and vertex $u$ is adjacent to the vertices $u+A = \{ u+a_i\pmod{m} : a_i \in A \}$. The names [*multiple loop digraph*]{} and [*multiple fixed step digraph*]{} [@r3; @r12] are also used. The [*$($d-step$)$ circulant graph*]{} $G(m;A)$ is defined analogously with $A = -A = \{ \pm a_{1}, \pm a_{2}, \ldots ,
\pm a_{d} \}$. These graphs have also received other names such as [*starred polygons*]{} [@r17] and [*multiple loop graphs*]{} [@r4]. In both the directed and undirected case, the elements of $A$ are called [*jumps*]{} or [*steps*]{}. Henceforth, we will use the word [*circulant*]{} to denote either a circulant digraph or a circulant graph. As stated above, notice that the circulant $G(m;A)$ is just the Cayley (di)graph of the cyclic group $Z/mZ$ with respect to $A$. Therefore, from Sabidussi’s result, [*a $($di$)$graph is a circulant iff its automorphism group contains a regular cyclic subgroup*]{}. As Leighton showed in [@r13], this characterization can be used to easily prove Turner’s result [@r17], which states that every transitive graph on a prime number of vertices is a circulant. (It suffices to use Cauchy’s group theorem: ‘If a prime $p$ divides the order of a finite group then it contains an element of order $p$.’) In fact the same reasoning shows that Turner’s result also holds for digraphs.
Recall that, given two graphs, $G_{1} = (V_{1},E_{1})$ and $G_{2} =
(V_{2},E_{2})$, their [*cartesian product*]{} $G_{1} \times G_{2}$ is the graph with set of vertices $V_{1} \times V_{2}$ and an edge between $(u_{1},u_{2})$ and $(v_{1},v_{2})$ iff either $(u_{1},v_{1})
\in E_{1}$ and $u_{2} =v_{2}$ or $u_{1} = v_{1}$ and $(u_{2},v_{2})
\in E_{2}$. The cartesian product of two digraphs is defined analogously. Another well known family of Cayley graphs of Abelian groups are the (Boolean) $n$-cubes, which are sometimes defined as the cartesian product of $n$ copies of the complete graph $K_{2}$. From our point of view, the [*n-cube*]{} (or [*binary n-dimensional hypercube*]{}) is the Cayley graph $G(\Gamma ;A)$, where $\Gamma = \Z/2\Z \times \cdots \times \Z/2\Z$ ($n$ factors) and $A$ is the set of unitary vectors $\e_{i}$, $1\leq i \leq n$.
This paper studies the structure of the so-called ($d$-step) multidimensional circulants (that is Cayley graphs or digraphs of Abelian groups), which are defined from integral matrices in Section 3. More precisely, given such a (di)graph, we are interested in finding its ‘dimension’, that is the minimum rank of the group(s) it can arise from. (The [*rank*]{} of a finitely generated Abelian group is the minimum number of elements which generate it.) For instance, from its definition it is clear that the $n$-cube has dimension not greater than $n$. As far as we know, this study was initialized by Leighton [@r13], where it was shown that the dimension of the $n$-cube is in fact $\lfloor (n+1)/2 \rfloor$. Other results, concerning $2$-step circulant digraphs (dimension $1$) can be found in [@eaf; @r12]. In this paper we continue this study by deriving some new results, which are contained in Section 3. For instance, using some facts about integral matrices and a theorem in [@r7], we give a full characterization of those $2$-step multidimensional circulants which are $1$-dimensional. Moreover, it is proved that if $G_{1}, G_{2}, \ldots ,G_{n}$ are circulant (di)graphs with $p>2$ vertices, $p$ a prime, then the cartesian product $G_{1}\times G_{2}\times \cdots \times G_{n}$ has dimension $n$.
As stated above, our approach uses some results of integral matrix theory which are summarized in the next section. In particular, we deal with the concept of congruence in $\Z^{n}$, also discussed there. The reason is that, in the same way that congruence in $\Z$ (periodicity in one dimension) leads to the consideration of cyclic groups, congruence in $\Z^{n}$ (related to periodicity in $n$ dimensions) induces quotient structures which are Abelian groups.
Congruences in $Z^n$ and the Induced Abelian Groups
===================================================
This section mainly deals with the concept of congruence in $\Z^{n}$ and its consequences to our study. In this context, the theory of integral matrices (that is matrices whose entries are integers) proves to be very useful and, hence, we begin by recalling some of its main results. The reader interested in the proofs is referred to [@r14].
Let $\Z^{n\ast n}$ be the ring of $n \times n$ matrices over $\Z$. Given $\A,\B\in \Z^{n\ast n}$, $\A$ is said to be [*right equivalent*]{} to $\B$ if there exists a unimodular (with determinant $\pm 1$) matrix $\V\in
\Z^{n\ast n}$ such that $\A=\B\V$; and $\A$ is [*equivalent*]{} to $\B$ if $\A=\U\B\V$ for some unimodular matrices $\U,\V\in \Z^{n\ast n}$. Clearly, both of them are equivalence relations.
Henceforth, $\M=(m_{ij})$ will denote a nonsingular matrix of $\Z^{n\ast n}$ with columns $\m_{j}=(m_{1j},m_{2j},\ldots
,m_{nj})^\top$, $j=1,2, \ldots ,n$, and $m=|\det \M|$. By the Hermite normal form theorem, $\M$ is right equivalent to an upper triangular matrix $\H(\M)=\H=(h_{ij})$ with positive diagonal elements $h_{ii}$ and with each element above the main diagonal $h_{ij}$, $j>i$, $i=1,2,\ldots
,n-1$, lying in a given complete set of residues modulo $h_{ii}$ (for instance, $0\leq h_{ij} \leq h_{ii}-1)$. This normal form is unique.
Let $k\in \Z$, $1\leq k\leq n$. The $k$th [*determinantal divisor*]{} of $\M$, denoted by $d_{k}(\M)=d_{k}$, is defined as the greatest common divisor of the $(^{n}_{k})^{2}$ $k\times k$ determinantal minors of $\M$. Since $\M$ is nonsingular, not all of them are zero. Notice that $d_{k}|
d_{k+1}$ for all $k=1,2,\ldots ,n-1$ and $d_{n}=m$. For convenience, put $d_{0}=1$. The [*invariant factors*]{} of $\M$ are the quantities $$s_{k}(\M) = s_{k} = \frac{d_{k}}{d_{k-1}},\quad k=1,2,\ldots ,n.$$ It can be shown that $s_{i}|s_{i+1}$, $i=1,2,\ldots ,n-1$.
By the Smith normal form theorem, $\M$ is equivalent to the diagonal matrix $\S(\M)=\S=\;$diag$(s_{1},s_{2},\ldots ,s_{n})$. This canonical form is unique.
For instance, the matrix $\M=\;$diag$(2,2,3)$, with determinantal divisors $d_{1}=1$, $d_{2} =2$, $d_{3}=12$, and invariant factors $s_{1} =1$, $s_{2}=2$, $s_{3}=6$, is equivalent to $\S=\;$diag$(1,2,6)$ since there exist the unimodular matrices $$\U=\left( \begin{array}{rrr}
-1 & 0 & 1 \\ 0 & 1 & 0 \\ -3 & 0 & 2
\end{array} \right),\quad
\V=\left( \begin{array}{rrr}
1 & 0 & 3 \\ 0 & 1 & 0 \\ 1 & 0 & 2
\end{array} \right) \label{eq1}$$ such that $\S=\U\M\V$.
As usual, the greatest common divisor of the integers $a_{1},a_{2},
\ldots ,a_{n}$ will be denoted by $\gcd(a_{1},a_{2}, \ldots ,a_{n} )$. When they are the coordinates of a vector $\a$, we will simply write $\gcd(\a)$.
Most of the remaining material in this section is drawn from [@r9] where additional details can be found.
Let $\Z^{n}$ denote the additive group of column $n$-vectors with integral coordinates. The set $\M\Z^{n}$ , whose elements are linear combinations (with integral coefficients) of the (column) vectors $\m_{j}$, is said to be the lattice generated by $\M$. Clearly, $\M\Z^{n}$ with the usual vector addition is a normal subgroup of $\Z^{n}$.
The concept of congruence in $\Z$ has the following natural generalization to $\Z^{n}$. Let $\a,\b\in \Z^{n}$ .We say that [*$\a$ is congruent with $\b$ modulo $\M$*]{}, and write $\a \equiv \b\pmod{\M}$, if $$\label{eq2}
\a-\b \in \M\Z^{n}.$$ The quotient group $\Z^{n}/\M\Z^{n}$ can intuitively be called the [*group of integral vectors modulo* ]{}. Henceforth, we follow the usual convention of identifying each equivalence class by any of its representatives.
Note that whenever $\M=\;$diag$(m_{1},m_{2}, \ldots ,m_{n})$ the vectors =$(a_{1},a_{2}, \ldots ,a_{n})^\top$ and =$(b_{1},b_{2}, \ldots ,b_{n})^\top$ are congruent modulo $\M$ iff the system of congruences in $\Z$ $$a_{i} \equiv b_{i}\pmod{m_{i}},\quad i=1,2, \ldots ,n$$ holds. In this case $\Z^{n}/\M\Z^{n}$ is the direct product of the cyclic groups $\Z/m_{i}\Z$, $i=1,2, \ldots ,n$.
If $\A$ and $\B$ are $n\times r$ matrices over $\Z$ with columns $\a_{j}$ and $\b_{j}$, $j=1,2,\ldots ,r$, respectively, we will write $\A \equiv \B\pmod{\M}$ to denote that $\a_{j} \equiv \b_{j}\pmod{\M}$ for all $j=1,2,\ldots ,r$.
Let $\H=\M\V$ be the Hermite normal form of $\M$. Then (\[eq2\]) holds iff $\a-\b \in \H\V^{-1}\Z^{n} = \H\Z^{n}$ since $\V$, and hence $\V^{-1}$, are unimodular. Therefore we conclude that $$\a \equiv \b\pmod{\M}\quad \Leftrightarrow \quad
\a \equiv \b\pmod{\H} \label{eq3}$$ or, what is the same, $$\Z^{n}/\M\Z^{n} \cong \Z^{n}/\H\Z^{n}. \label{eq4}$$
Let us now consider the Smith normal form of $\M$, $\S=\;$diag$(s_{1},s_{2},
\ldots ,s_{n})=\U\M\V$. Then (\[eq2\]) holds iff $\U\a \equiv
\U\b\pmod{\S}$ or, equivalently, $$\u_{i}\a \equiv \u_{i}\b\pmod{s_i},\quad
i=1,2, \ldots ,n \label{eq5}$$ where $\u_{i}$ stands for the $i$th row of $\U$. Moreover, if $r$ is the smallest integer such that $s_{n-r}=1$ (thus, $s_{1}=s_{2}=
\cdots =s_{n-r-1} = 1$), (if there is no such a $r$, let $r=n$), the first $n-r$ equations in (\[eq5\]) are irrelevant, and we only need to consider the other ones. This allows us to write $$\a \equiv \b\pmod{\M} \quad \Leftrightarrow \quad
\U'\a \equiv \U'\b\pmod{\S'} \label{eq6}$$ where $\U'$ stands for the $r\times n$ matrix obtained from $\U$ by leaving out the first $n-r$ rows, and $\S'=\;$diag$(s_{n-r+1},s_{n-r+2},
\ldots ,s_{n})$. So, the (linear) mapping $\phi$ from the vectors modulo $\M$ to the vectors modulo $\S'$ given by $\phi (\a)=\U'\a$ is a group isomorphism, and we can write $$\Z^{n}/\M\Z^{n} \cong \Z^{r}/\S'\Z^{r} = \Z/s_{n-r+1}\Z \times \cdots \times \Z/s_{n}\Z.
\label{eq7}$$ Analogously, it may be shown that the $n \times r$ matrix of the inverse mapping $\phi^{-1}$ is obtained from $\U^{-1}$ by leaving out its first $n-r$ columns.
The next proposition contains some easy consequences of the above results. For instance, $(b)$ follows from the fact that $s_{1}s_{2} \cdots s_{n} = d_{n} = m$ and $s_{i}|s_{i+1}$, $i=1,2, \ldots ,n-1$.
\[pro2.1\]
- The number of equivalence classes modulo $\M$ is $ |\Z^{n}/\M\Z^{n}| = m = |\det \M| $.
- If $p_{1}^{r_{1}}p_{2}^{r_{2}} \cdots p_{t}^{r_{t}}$ is the prime factorization of $m$, then $\Z^{n}/\M\Z^{n} \cong \Z^{r}/\S'\Z^{r}$ for some $r\times r$ matrix $\S'$ with $r \leq \max \{r_{i} :1 \leq i \leq t \}$.
- The $($Abelian$)$ group of integral vectors modulo $\M$ is cyclic iff $d_{n-1}=$1.
- Let $r$ be the smallest integer such that $s_{n-r}=1$. Then $r$ is the rank of $\Z^{n}/\M\Z^{n}$ and the last $r$ columns of $\U^{-1}$ form a basis of $\Z^{n}/\M\Z^{n}$. $\Box$
Given any element of $\Z^{n}/\M\Z^{n}$, simple reasoning shows that its order is given by the formula $$\label{eq8}
{\rm o}(\a) = \frac{m}{\gcd(m,\gcd(m\M^{-1}\a))},$$ (see [@r9]). For instance, if $\M=(m_{ij})$ is a $2\times 2$ matrix and $\a=(a_{1},a_{2})^\top$ we have $${\rm o}(\a) = \frac{m}{\gcd(m,a_{1}m_{22}-a_{2}m_{12},
a_{2}m_{11}-a_{1}m_{21})}. \label{eq9}$$ According to (\[eq7\]), for any given $\M\in \Z^{n \ast n}$ there exists an Abelian group $\Gamma$ such that $\Gamma=\Z^{n}/\M\Z^{n}$. Conversely, let $\Gamma$ be a finite Abelian group generated by the elements $g_{1},g_{2}, \ldots g_{n}$. Then $\Gamma$ is isomorphic to $\Z^{n}/K$, where $K$ is the kernel of the surjective homomorphism $\Psi : \Z^{n} \longrightarrow \Gamma$ defined by $\Psi(\x)=
\g\x$, where $\g$ denotes the row vector $(g_{1},g_{2}, \ldots ,g_{n})$ and $\x\in \Z^{n}$. (Note that $\Psi(\e_{i})=g_{i}$, $1\leq i\leq n$, where $\e_{i}$ stands for the $i$th unitary vector.) More precisely, $K$ is the lattice of $\Z^{n}$ generated by the upper triangular $n\times n$ matrix $\H=(m_{ij})$ defined as follows:
- $m_{11}={\rm o}(g_{1})=|\langle g_{1}\rangle|$;
- $m_{jj}= \min\{ \mu \in \Z^{+}:\mu g_{j} \in \langle g_{1},g_{2}, \ldots
,g_{j-1}\rangle \}$, $j=2,3, \ldots ,n $; and
- $m_{ij}$, $j=2,3, \ldots ,n$, $i<j$, are any integers such that $m_{1j}g_{1}+m_{2j}g_{2}+ \cdots +m_{jj}g_{j} = 0$, (they can be chosen in a given complete set of residues modulo $h_{ii}$, e.g., $0 \leq m_{ij} \leq h_{ii} -1$),
where, as usual, $\langle g_{1},g_{2},
\ldots ,g_{j}\rangle$ denotes the group generated by $g_{1},g_{2}, \ldots
,g_{j}$, and $0$ is the identity element of $\Gamma$. Clearly, $\H$ is the Hermite normal form of any matrix $\M$ which generates the lattice $K$.
Multidimensional circulants and their dimension
===============================================
Congruence in $\Z^{n}$ leads to the following generalization of circulants. Let $\M$ be an $n \times n$ integral matrix as in Section 2. Let $A=\{ \a_{j}=(a_{1j},a_{2j}, \ldots ,a_{nj})^\top :
1 \leq j \leq d \} \subseteq \Z^{n}/\M\Z^{n}$. The [*multidimensional $($d-step$)$ circulant digraph G(;A)*]{} has as vertex-set the integral vectors modulo $\M$, and every vertex $\u$ is adjacent to the vertices $\u+A\pmod{\M}$. As in the case of circulants, the [*multidimensional $($d-step$)$ circulant graph $G(\M;A)$*]{} is defined similarly just requiring $A=-A$.
In [@r13], Leighton considered multidimensional circulant graphs with diagonal matrix $\M$, and characterized them by showing that [*the automorphism group of these graphs must contain a regular Abelian subgroup*]{}. Clearly, a multidimensional circulant (digraph or graph) is a Cayley (di)graph of the Abelian group $\Z^{n}/\M\Z^{n}$. As a consequence, Sabidussi’s result implies that Leighton’s statement holds in fact for a multidimensional circulant obtained from any matrix $\M$.
As another consequence of the above, if $\alpha$ is the index of the subgroup $\Gamma=\langle \a_{1},\a_{2}, \ldots ,\a_{d}\rangle$ in $\Z^{n}/\M\Z^{n}$ , the multidimensional circulant $G(\M;A)$ consists of $\alpha$ copies of the Cayley (di)graph of $\Gamma$ generated by $A$. Besides, from the comments in the last paragraph of Section 2, $\Gamma \cong \Z^{d}/\H\Z^{d}$, where $\H$ is an upper triangular $d \times d$ matrix, and each such copy is isomorphic to $G(\H;\e_{1},\e_{2}, \ldots \e_{d})$.
In particular, $G(\M;\a_{1},\a_{2}, \ldots \a_{d})$ (respectively $G(\M;\pm\a_{1},\pm\a_{2}, \ldots \pm\a_{d})$) is strongly connected (respectively connected), that is $\alpha=1$, iff $\{ \a_{1},\a_{2}, \ldots \a_{d} \}$ generates $\Z^{n}/\M\Z^{n}$, that is, there exist $n$ integral $d$-vectors $\x^{j} = (x_{1j},x_{2j}, \ldots ,x_{dj})^\top$, $j=1,2, \ldots ,n$, such that $$x_{1j}\a_{1}+x_{2j}\a_{2}+ \cdots +x_{dj}\a_{d} \equiv
\e_{j}\pmod{\M},\quad j=1,2, \ldots ,n$$ or, in matrix form, $$\A\X \equiv \I\pmod{\M}$$ where $\A$ now denotes the $n\times d$ matrix $(a_{ij})$, $\X$ is the $d\times n$ matrix $(x_{ij})$, and $\I$ stands for the identity matrix.
A certain (di)graph may be a multidimensional circulant for several different values of $n$. For instance, the digraph $G(\M;A)$ with $\M=\mbox{ diag}(2,2,3)$ and $A=\{ (1,0,0)^\top ,(0,1,0)^\top,$ $
(0,0,2)^\top \}$ is isomorphic to the digraph $G(\M';A')$ with $\M'=\mbox{ diag}(2,6)$ and $A'=\{ (0,3)^\top ,(1,0)^\top ,
(0,4)^\top \}$ since, if $\U'$ is the $2\times 3$ matrix obtained by taking the two last rows of the matrix $\U$ in (\[eq1\]), we have $\U'A\equiv A'\pmod{\M'}$. Following Leighton’s terminology [@r13], if $k$ is the smallest value of such $n$ we will say that the multidimensional circulant has [*dimension*]{} $k$ or that it is [*k-dimensional*]{}. Notice that this parameter is in fact the minimum rank of the groups such a (di)graph can arise from. Then the class of circulants is precisely the class of $1$-dimensional circulants.
In studying the dimension of a given multidimensional circulant $G(\M;A)$ we only need to consider the connected case. Indeed suppose that the $\alpha$ disjoint components of $G(\M;A)$ are, say, $k$-dimensional and isomorphic to $G(\M';A'), \M' \in \Z^{k \ast k}$. Then $G(\M;A)$ is isomorphic to $G(\alpha \M';\alpha A')$ where $\alpha \M'$ and $\alpha A'$ denote the matrix and set obtained from $\M'$ and $A'$ by simply multiplying by ${\alpha }$ any, say the first, component of the corresponding (column) vectors. As a corollary, the dimension of $G(\M;A)$ cannot be greater than the cardinality of the minimum subset of $A$ that generates $\Gamma=\langle \a_{1},\a_{2}, \ldots
,\a_{d}\rangle$.
To obtain other results about the dimension of multidimensional circulants it is useful to introduce the concept of Ádám isomorphism. Let $\M\in \Z^{n \ast n}$ and $\M'\in \Z^{n' \ast n'}$. Then the multidimensional circulants $G(\M;A)$ and $G(\M';A')$ are said to be [*Ádám isomorphic*]{} if there exists an isomorphism $\phi$ between the groups $\Z^{n}/\M\Z^{n}$ and $\Z^{n'}/\M'\Z^{n'}$ such that $\phi (A)=A'$. For instance, if $u$ is a unit of $\Z/m\Z$, that is $\gcd (u,m)=1$, the circulants $G(m;A)$ and $G(m;uA)$ are Ádám isomorphic. In [@r1] it was first conjectured that any two isomorphic circulant digraphs are Ádám isomorphic, but in subsequent papers more attention was paid to the corresponding statement for circulant graphs. For instance, Djokovic [@r8] and Turner [@r17] independently proved that Ádám’s conjecture is true for circulant graphs with prime order, and this is also the case for circulant digraphs [@r11]. The first counter-examples to this conjecture, both for graphs and digraphs were given by Elpas and Turner in [@r11]. In [@r2], Alspach and Parsons characterized, in terms of a condition on automorphism groups, the validity of Ádám’s conjecture for a given order $m$. In particular, the authors used this characterization to show that it holds for $m=p_{1}p_{2}$ where $p_{1}$ and $p_{2}$ are different primes. In [@r5], Boesch and Tindell conjectured that all isomorphic $2$-step circulant graphs are Ádám isomorphic. This was independently proved in [@r7] and [@r19]. The same result for $2$-step circulant digraphs was given in [@r10]. In fact, Delorme, Favaron and Mahéo [@r7] proved some more general results concerning Cayley (di)graphs of Abelian groups. Using our terminology, they are stated in the following theorem.
\[the3.1\] Let $\M\in \Z^{n \ast n}$ and $\M'\in \Z^{n' \ast n'}$ and suppose that $A=\{ \a_{1},\a_{2} \}$ and $A'=\{ \b_{1},\b_{2} \}$ are generating sets for $\Z^{n}/\M\Z^{n}$ and $\Z^{n'}/\M'\Z^{n'}$, respectively. Then the two $($connected$)$ multidimensional circulant digraphs $G(\M;A)$ and $G(\M';A')$ are isomorphic iff they are Ádám isomorphic, except in the case when there exist two group isomorphisms $$\phi : \Z^{n}/\M\Z^{n} \longrightarrow \Z/2\eta \Z\times \Z/2\Z,\quad
\eta \in \Z^{+}$$ and $$\phi ' : \Z^{n'}/\M'\Z^{n'} \longrightarrow \Z/4\eta \Z$$ such that $\phi(A)=\{ (1,0)^\top ,(1,1)^\top \}$ and $\phi'(A')=\{1,2\eta +1\}$. $($In this case the two digraphs are isomorphic but, clearly, they are not Ádám isomorphic.$)$ Moreover, this result is also true for connected multidimensional circulant graphs if we change $A$ and $A'$ by $\pm A$ and $\pm A'$, respectively. $\Box$
Note that the exceptional case $\{ \Z/2\eta \Z\times \Z/2\Z$, $\a_{1}
=(1,0)^\top , \a_{2}=(1,1)^\top \}$ could also be characterized by the defining relations $\a_{1}+\a_{2}=\a_{2}+\a_{1}$ (Abelian group), $2\eta\a_{1}=\vec0$ and $2\a_{1}=
2\a_{2}$. Moreover, this last relation is equivalent to writing ${\rm o}(\a_{1}-\a_{2})=2$ (which holds indeed if we substitute the above values in (\[eq9\])).
Let $\H=\M\V$ be the Hermite normal form of the matrix $\M\in \Z^{n \ast n}$. Let $\S=\mbox{ diag}(s_{1},s_{2}, \ldots ,s_{n})=\U\M\V$ be its Smith normal form with $s_{1}=s_{2}= \cdots =s_{n-r}=1$ and consider the $r \times r$ and $r \times n$ matrices $\S'$ and $\U'$ defined as in Section 2. From the results (\[eq3\]), (\[eq4\]), (\[eq6\]) and (\[eq7\]) given there we have the following theorem.
\[the3.2\] The multidimensional circulants $G(\M;A)$, $G(\H;A)$ and $G(\S';\phi(A))$, where $\phi(A)=\{ \U'\a:\a\in A \}$, are Ádám isomorphic. $\Box$
As an example of application of this theorem, we can again consider the two isomorphic multidimensional circulants with matrices $\M=\mbox{ diag}
(2,2,3)$ and $\M'=\mbox{ diag}(2,6)$ mentioned before.
\[cor3.3\] Let $G(\M;A)$ be a $k$-dimensional circulant. Then $k \leq r$. In particular, if $r=1$ $(d_{n-1}=s_{n-1}=1)$ such a $($di$)$graph is a circulant. $\Box$
From the above corollary and Proposition \[pro2.1\](b) we get
\[cor3.4\] Let $G(\M;A)$ be a $k$-dimensional circulant with $m=p_{1}^{r_{1}}p_{2}^{r_{2}} \cdots p_{t}^{r_{t}}$ vertices. Then $k\leq \max\{r_{i}:1\leq i\leq t \}$. In particular, if $m$ is square free (that is, $m$ is not divisible by the square of a prime) $G(\M;A)$ is a circulant. $\Box$
This coincides with the result obtained by Leighton in [@r13] for a multidimensional circulant graph $G(\M;A)$ with $\M$ a diagonal matrix.
In the case of multidimensional $2$-step circulants we can give a complete characterization of circulants and, hence, of their dimension.
\[the3.5\] Let $\M$ be an $n\times n$ matrix with $(n-1)$th determinantal divisor $d_{n-1}$. Let $A=\{ \a_{1},\a_{2} \}$ be a generating set of $\Z^{n}/\M\Z^{n}$. Then the $($connected$)$ multidimensional $2$-step circulant digraph $G(\M;A)$ $($respectively, graph $G(\M;\pm A)$$)$, on $m=|\det \M|$ vertices, is a circulant iff one of the following conditions holds:
- $d_{n-1} = 1$; or
- $d_{n-1} = 2$ and $m=2\gcd(m,\gcd(m\M^{-1}(\a_{1}-\a_{2})))$; $($respectively, or
- $d_{n-1} = 2$ and $m=2\gcd(m,\gcd(m\M^{-1}(\a_{1}+\a_{2})))$$)$.
[**Proof.**]{} From Theorem \[the3.1\] and Corollary \[cor3.3\] it is clear that $(a)$ is a necessary and sufficient condition for $G(\M;A)$ or $G(\M;\pm A)$ to be a circulant except in the case $\Z^{n}/\M\Z^{n} \cong
\Z/2\eta \Z \times \Z/2\Z$ and o$(\a_{1}-\a_{2})=2$ (or possibly, in the case of graphs, o$(\a_{1}+\a_{2})=2$.) (According to this theorem, in this case we also have a circulant.) But then from the results of Section 2, and in particular (\[eq8\]), condition $(b)$ (or condition $(c)$, in the case of graphs) must hold. $\Box$
In [@r12; @eaf] it was shown that the study of some distance-related parameters, such as the diameter, of $2$-step circulant digraphs is best accomplished by considering them as particular instances of multidimensional circulants digraphs $G(\M;\e_{1},\e_{2})$ with $\M$ a $2\times 2$ matrix. Such (strongly connected) digraphs have been called [*commutative 2-step digraphs*]{} [@r10]. The reason is that the matrix $\M=(m_{ij})$ can always be chosen so that the studied parameter is easily related to its entries $m_{ij}$. (In some cases the same fact is true for $2$-step circulant graphs, see [@r4] and [@r18].) Hence, it is of some interest to characterize those commutative $2$-step (di)graphs which are circulants. As a particular case of Theorem 3.5, the next corollary gives a complete answer to this question.
\[cor3.6\] Let $\M=(m_{ij})$ be a $2\times 2$ integer matrix with $|\det \M|=m$. Then the commutative $2$-step digraph $G(\M;\e_{1} ,\e_{2})$ is a circulant digraph iff either $$d_{1}=\gcd (m_{11},m_{12},m_{21},m_{22})=1$$ or $$d_{1}=2 \mbox{\ and\ } m=2\gcd (m,m_{22}+m_{12},m_{11}+m_{21}). \ \Box$$
Theorem \[the3.5\] illustrates the fact that, although the knowledge of the structure of $\Z^{n}/\M\Z^{n}$ (Proposition \[pro2.1\]) gives an upper bound for the dimension of a multidimensional circulant (Corollary \[cor3.3\]), the computation of its exact value may require more sophisticated and particular techniques. This is also made apparent for the next result, which gives the dimension of the direct product of $n$ circulants, all of them with equal prime number of vertices. The proof is similar to that of Lemma 1 in [@r13], which was suggested by Lawrence (personal communication to Leighton.)
\[the3.7\] Let $G_{1}=G(p;A_{1}), G_{2}=G(p;A_{2} ),\ldots ,G_{n}=G(p;A_{n})$ be $($connected$)$ circulants with $p>2$ vertices, $p$ a prime. Let $\M\in \Z^{n \ast n}$ be the diagonal matrix [diag]{}$(p,p,\ldots p)$ and $A=\{ a\e_{i}:a\in A_{i} ,1\leq i\leq n \}$. Then the multidimensional circulant $G(\M;A)$ has dimension $n$.
[**Proof.**]{} First note that the (di)graph $G(\M;A)$, with vertex-set $V=\Z/p\Z \times \cdots \times \Z/p\Z$ ($n$ factors), is nothing more than the cartesian product $G_{1}\times \cdots \times G_{n}$. Let $\Omega$ be any regular Abelian subgroup of ${\sl Aut}G(\M;A)$, $|\Omega |=|V|=p^{n}$. By Sabidussi’s result it suffices to show that $\Omega \cong \Z^{n}/\M\Z^{n} \cong
\Z/p\Z \times \cdots \times \Z/p\Z$. For some fixed $1\leq i\leq n$ and $j\in \Z/p\Z$, let $G_{ij}$ denote the sub(di)graph of $G(\M;A)$ spanned by the vertices whose labels have their $i$th component equal to $j$. Now, let us consider the group $\Omega$ as acting on the set ${\cal G}=\{ G_{ij}:1\leq i\leq n, 0\leq j< p \}$. To show that any automorphism $\omega \in \Omega$ preserves the set ${\cal G}$ (that is, either $\omega(G_{ij}) \bigcap G_{kl} = \emptyset$ or $\omega(G_{ij}) = G_{kl}$), it suffices to prove that $\omega$ preserves the set of $n$ directions. (As expected, a [*direction*]{} is defined as the set of edges whose endvertices only differ in a given coordinate.) To know whether different edges belong to the same direction we can apply the following algorithm:
Choose a vertex $\u \in V$ and consider any shortest odd cycle containing it. Then, all the edges of this cycle clearly belong to one direction, say $i$. If the cycle has length $p$, then the sub(di)graph spanned by its $p$ vertices, $G_i(\u)$, is the copy of $G_{i}=G(p;A_{i})$ that contains vertex $\u$ and whose edges belong to direction $i$. Otherwise, we consider an edge of the cycle and look for a different shortest odd cycle (of the same length as before) containing it. In this way we successively find the edges (and vertices) of $G_i(\u)$. If, in some step, there is no such a shortest cycle we consider another of the (already found) edges of $G_i(\u)$. Because of the nature of $G_{i}$, it is not difficult to realize that we eventually find the searched $p$ vertices spanning $G_i(\u)$. To find the other sub(di)graphs $G_j(\u)$, $j \neq i$, we start again from vertex $\u$ and look, in the same way as before, for shortest odd cycles not containing edges in the already found directions. This is done until no edge incident to $\u$ is left out of discovered directions. To identify the directions of the edges incident to other vertices, different from $\u$, we can apply the following procedure: Consider two adjacent edges with endvertex $\u$ and different directions, say $(\z,\u) \in G_j(\u)$ and $(\u,\v) \in G_i(\u)$. Look for a shortest cycle containing them, $\u,\v,\ldots,\z,\u$, (note that its length must be at least 4). Then, the first edge (of the cycle) not in $G_i(\u)$ has direction $j$ and, similarly, the last edge not in $G_j(\u)$ has direction $i$. Finally, once the directions of a sufficient number of edges incident to a vertex, say $\w$, have been determined, we can search again for appropriates odd cycles going through it, in order to locate the (di)graphs $G_i(\w)$, $i=1,2,\ldots ,n$.
From the above, the action of an automorphism $\omega \in
\Omega$ on $V$ completely determines its action on ${\cal G}$. Conversely, let $\u=(u_{1},u_{2}, \ldots ,u_{n}) \in V$. Then $\u$ is the only vertex the sub(di)graphs $G_{iu_{i}}, 1\leq i\leq n$, have in common. Hence, the action of $\omega$ on ${\cal G}$ also determines its action on $V$.
Let ${\cal G}_{1}, \ldots ,{\cal G}_{k}$ be the orbits of ${\cal G}$ under the action of $\Omega$. Let $\Omega_h$, $1\leq h\leq k$, be the restriction of $\Omega$ to ${\cal G}_{h}$ with duplicates eliminated. Then, $\Omega \subseteq \Omega_1 \times
\cdots \times \Omega_k$ and hence $$\prod_{h=1}^k |\Omega_{h}| \geq |\Omega| = p^{n}. \label{eq10}$$ Moreover, since $\Omega_h$ is Abelian and transitive on ${\cal G}_h$, it is also regular. Therefore $|\Omega_h| = |{\cal G}_h|$, $1\leq h\leq k$, and then $$\sum_{h=1}^k |\Omega_{h}| = \sum_{h=1}^k |{\cal G}_{h}| = |{\cal G}| = np. \label{eq11}$$ In addition, the order of an orbit, $|\Omega_h|= |{\cal G}_h|$, divides the order of the permutation group $|\Omega |=p^n$, see [@r15]. Thus there exist integers $r_h \geq 0$, $1\leq h\leq k$, such that $|\Omega_h|=p^{r_{h}}$ , and formulas (\[eq10\]), (\[eq11\]) yield $$\sum_{h=1}^k r_h \geq n,\quad \sum_{h=1}^k p^{r_{h}} = np,$$ respectively. Hence, we must have $\sum_{h=1}^k pr_{h} \geq pn = \sum_{h=1}^k p^{r_{h}}$. But, for $p>2$, $pr_{h} < p^{r_{h}}$ if $r_{h} \neq 1$ and $pr_{h}=p^{r_{h}}$ otherwise. Thus $|\Omega_h|=p$ for any $1\leq h\leq k$, so that $\Omega_h$ is isomorphic to the cyclic group $\Z/p\Z$ and, from (\[eq10\]), $\Omega \cong \Z/p\Z \times \cdots
\times \Z/p\Z$ ($n$ factors) as claimed. $\Box$
As an example of application of the above theorem, we can state the following special case:
\[cor3.8\] The cartesian product $K_p \times \cdots \times K_p$ $($n factors$)$, and the cartesian product of $n$ $p$-cycles $($directed or not$)$ both have dimension $n$. $\Box$
In the above examples there is an easy way to know whether edges incident to a given vertex $\u$ belong to the same direction (or to locate the (di)graphs $G_i(\u)$), owing to the connected components of the neighbourhood of $\u$ in the case of complete graphs $K_p$, and to the (shortest) $p$-cycles in the case of cycles. Another example comes when each set $A_i$ has the property that if $x, y \in A_i$, $x \neq y$, then one at least of the elements $x-y$, $x+y$, $-x+y$, $-x-y$ belongs to $A_i$ (for example, if $A_i$ is stable under multiplication by 2): the directions are then given by the connected components of the neigbourhood of $\u$.
This work has been supported in part by the Spanish Research Council (Comisi' on Interministerial de Ciencia y Tecnología, CICYT) under Projects TIC90-0712 and TIC-92-1228-E. The author thanks Charles Delorme for his valuable comments on the proof of Theorem \[the3.7\] and Corollary \[cor3.8\], and one of the referees for many helpful suggestions.
[99]{}
A. Ádám, Research problem 2-10, [*J. Combin. Theory*]{} [**2**]{} (1967) 393.
B. Alspach and T.D. Parsons, Isomorphism of circulant graphs and digraphs, [*Discrete Math.*]{} [**25**]{} (1979) 97–108.
J.-C. Bermond, F. Comellas and D.F. Hsu, Distributed loop computer networks: a survey, to appear in [*J. Parallel and Distributed Computing*]{} [**24**]{} (1995) 2–10.
J.-C. Bermond, G. Illiades and C. Peyrat, An optimization problem in distributed loop computer networks, Proc. 3rd Int. Conf. Combin. Math., New York, 1985. [*Ann. of the New York Acad. Sci.*]{} [**555**]{} (1989) 45–55.
F. Boesch and R. Tindell, Circulants and their connectivities, [*J. Graph Theory*]{} [**8**]{} (1984) 487–499.
G. Chartrand and L. Lesniak, [*Graphs and Digraphs*]{}, Wadsworth, Monterrey, 1986.
C. Delorme, O. Favaron and M. Mahéo, Isomorphisms of Cayley multigraphs of degree 4 on finite Abelian groups, [*European J. Combin.*]{} [**13**]{} (1992), [*no.*]{} 1, 59–61.
D.Z. Djokovic, Isomorphism problem for a special class of graphs, [*Acta Math. Acad. Sci. Hung.*]{} [**21**]{} (1970) 267–270.
B. Elpas and J. Turner, Graphs with circulant adjacency matrices, [*J. Combin. Theory*]{} [**9**]{} (1970) 297–307.
P. Esqué, F. Aguiló and M.A. Fiol, Double commutative-step digraphs with minimum diameters, [*Discrete Math.*]{} [**114**]{} (1993) 147–157.
M.A. Fiol, Congruences in $Z^n$, finite Abelian groups and the Chinese remainder theorem, [*Discrete Math.*]{} [**67**]{} (1987) 101–105.
M.A. Fiol and P. Morillo, Congruences in $Z^2$ and commutative two step digraphs, [*3ème Coll. Int. Theorie des Graphes et Combinatoire*]{}, Marseille, France, June 1986.
M.A. Fiol, J.L.A. Yebra, I. Alegre and M. Valero, A discrete optimization problem in local networks and data alignment, [*IEEE Trans. Comput.*]{} [**C-36**]{} (1987) 702–713.
F.T. Leighton, Circulants and the characterization of vertex-transitive graphs, [*J. Res. Natl. Bur. Standards*]{} [**88**]{} (1983), no. 6, 395–402.
M. Newman, [*Integral Matrices*]{}, Pure and Appl. Math. Series Vol. [**45**]{}, Academic Press, New York, 1972.
D.J.S. Robinson, [*A Course in the Theory of Groups*]{}, Springer, New York, 1982.
G. Sabidussi, On a class of fixed-point-free graphs, [*Proc. Amer. Math. Soc.*]{} [**9**]{} (1958) 800–804.
J. Turner, Point-symmetric graphs with a prime number of points, [*J. Combin. Theory*]{} [**3**]{} (1967) 136–145.
J.L.A. Yebra, M.A. Fiol, P. Morillo and I. Alegre, The diameter of undirected graphs associated to plane tessellations, [*Ars Combin.*]{} [**20B**]{} (1985) 159–171.
S.C. Zhou, On Ádám isomorphism of circulants, [*J. Changsha Railway Inst.*]{} [**5**]{} (1987), no. 2, 11–18.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'P. P. Marvol$^1$'
title: A Tip of the TOE
---
physics/0309055
Institute for Applied Quantum Acoustics,\
Pyramid College Cacania, Hatetepe Boulevard 12,\
CC-12000 Cacania City, CACANIA
[*Abstract.*]{} Using standard methods from string theory, this paper presents a comprehensive survey about the most important aspects of the theory of sand. Special interest is put on the examination of the sand-wind duality and the interaction of ordinary (non-supersymmetric) sand with the heat field of the earth, as solution of the field inequalities of a stone. This will lead us, in a natural way, to completely new insights into the theory of sandstorms.
Keywords: [*many particle systems; string theory; field theory; quantum chaos; GRT; SUSY; thermal QFT; GRB; quantum cosmology; noncommutative QT;*]{}
Introduction
============
The matter of sand has ever since fascinated mankind. Already twenty thousand years ago attempts to classify the grainy zoo have been made, [@abc]. In the Cretacious Age many advances with respect to sand have been undertaken, especially it has been proved by [@unspellable] that chalk and sand are—at a fundamental level—two manifestations of the same thing. Unfortunately, since the end of the Cretatious Age, sand research activities suffered from a considerable lack of financial resources, a time period which is called the Big Sand Crisis (BSC). This led to the the well-known revolutionary movement “Every Scientist Needs a Camel”, [@marx]. After all, at least a few new concepts have been invented, among them the so-called [*wheel*]{}, which then, however, has been discarded due to its impracticability, [@tuareg]. (However, there are some unteachable fanatics who still try to demonstrate the usefulness of wheels on sand in the annual Paris-Dakar competition.)
Some centuries after the BSC a completely new impulse came from ancient Lei Li Ga, [@lei], who was the first to put special interest in the dynamics of sand.
The so far highly disordered efforts in sand have been fibre bundled by the famous philosoph Goe-The, who formulated the central question of today’s complex theory of sand, [@fist]:
[*Herauszufinden, was den Sand im Innersten zusammenband.$^{2}$* ]{}\
Starting from this point, many interesting theories about sand dynamics have been developed. One of the most ambitious (and ambiguous) models was the well-known sandstring theory developed by the old Egyptian priest Edua ’Wit X [@cheops]. With the postulate of the non-existence of infinitely small grains of sand, several severe problems of ordinary sand-theory (e.g. incurable sand-divergencies, causality problems etc.) seemed to be repaired. However, over the millennia, this model did not bear any fruits in sand. (After all, there have been some interesting side results for weavers, in particular new techniques for knotting.) This led to a complete restart of research in the field of sand. Fortunately, at least some interesting questions have been solved independently of sandstring theory, [@cameroon], [@unknown], [@acta].
Modern, post-string sand theories have also their origin in Egypt. In particular, a comprehensive quantum theory of the dynamics of sand dunes has been developed, [@annals], [@quant]. Piles of stones (so called “pyramides”) have been constructed for a macroscopic test of the dynamic theory, especially for the prominent dune-tunnel effect. However, the pyramid-models emerged as too roughly textured, so there could not be proven anything with these (although they are still very impressive). Indeed, experimental clues for the correctness of the Egyptian Theory of Quantum Sand Dynamics (ETQSD) has been found only recently by Swedish Inger Zeil, [@zei]. Of course, to people living in North-Africa, his results have always been obvious. However, as they aroused some interest outside the Big Desert, he eventually will receive the prize without any bells.
Presently, research in sand dynamics came to an interlocutory end with a stone’s$^3$ field inequalities, [@stone], which form the basis of our approach.
The paper is organized as follows. In section two, the main aspects of the standard sand-wind duality are recapitulated. In section three a stone’s field inequalities (specialized for the heat kernel of the earth) are presented. From these fundamental considerations, section four will lead us in a natural way to a completely new and attractive theory of sandstorms. In section five follows a brief conclusion and outlook with respect to some recent aspects of sand theory will be given.
Sand-Wind Duality
=================
This section is devoted to a brief recapitulation of the sand-wind duality of grain dynamics. Since sand is quantized (’grained’), one has to use the well-known cat-equation, which reads (in natural units) $$(W + V_H)|\mathrm{sand}\rangle = -i\frac\partial{\partial t}|\mathrm{sand}
\rangle.$$ $W$ is the wind operator, $V_H$ stands for the heat field operator. We use the idempotency of sand dunes (two sand dunes thrown upon each other give again a sand dune, since the superfluous grains merely drain in the sea of sand, as introduced by the great wizard Pamdirac) and $WV_H=V_HW=0$ (heat and wind never occur at the same time). Now, multiplying this equation by its complex conjugate and dressing the cat with its bra leads immediately to $$\langle \mathrm{sand}|W|\mathrm{sand}\rangle =
\langle \mathrm{sand}|(\frac\partial{\partial t})^2\mathrm{|sand}\rangle.$$ Due to the quadratic time derivative on the right hand side of this equation, by interpreting the operators as states and the states as operators we can immediately read off the duality between sand and wind. Most interesting consequences of this phenomenon follow from the fact that a (sand theoretical) distinction between the sand-field and the wind-field is impossible. This is the basis of our modern understanding of sandstorms.
In classical sand theory, it has been believed that the genesis of sandstorms lies merely in the wind field, interacting with the grains of sand, pushing them up and around. Modern understanding of sandstorms using the sand-wind duality of grain dynamics, however, has taught that the wind is a mere manifestation of the sand field dynamics. Thus, the usual theory of sandstorms is simply that the interaction of the sand-wind-field with itself is sufficient to create a sandstorm. Indeed, it has been shown by [@private] that the movement of the grains of sand functions as source term for the wind field, not the other way as suggested by classical theory: It’s the grains generating the wind.
So far, this is somwhat awkward but, after all, well understood. Unfortunately, combinig the principles of grain dynamics with the Relative Generality Theory of a stone, we will find that this simple, non-general model is not sufficient to a full understanding of sandstorms.
The Heatfield
=============
In the preceding section we have deliberately assumed that the reader is familiar with some of the basic notions describing the coupling of wind to sand. However, in order to fully exploit the origin of the breakdown of sand-wind-duality in sandstorms, we shall need a more thorough understanding of the generalistic physics of sand with respect to the heat field, i.e. of a stone’s theory and its grainization. An excellent recent introduction can be found in [@fein] – in chapter 42, of course.
Let’s begin with a brief historical survey. As it is well known, the experiments of Michael’s son (an ethiopian sprinter) and a cheetah [@mic] have given a first evidence that nothing moves faster than sand and that all grains (in a sandstorm) have the same speed, at least within the experimental accuracy of that time. This led to a relatively special view of matters. However, it still remained unclear how such a theory could be combined with the interaction of wind, sand and heat in the spirit of the old ton theory [@cern]. Only a few millennia later, this problem has been solved in an ingeniously simple manner by a stone.
Obviously, the sand in the desert is rather hot, in fact even hotter than the air surrounding it. Up to a stone’s era many scientists desperately tried to understand this phenomenon. A stone turned it into a new paradigm instead. In brief, his idea can be stated as follows: The heat moves the grains, but the grains are the source for the heatfield at the same time. (If we could switch off the dynamics and wait long enough then the air would be as hot as the sand.) More quantitatively, a stone’s inequalities read $$\begin{aligned}
T_g^{\mu\nu} &\geq& T_w^{\mu\nu} \quad , \\
G^{\mu\nu} &\approx& T^{\mu\nu}_{\mbox{?}} \end{aligned}$$ from which one immediately derives his famous saying “E=m ceh ceh”. (The ancient letter $=$ has no analogue nowadays.) Note that the sand is still treated as a classical (in the sense of continuous) field (of characteristic zero) here. The coupling of the spin-2-heatfield to wind, which is spin-1, i.e. a vector-field, is described by the force $ F^\mu = h^{\mu\nu} W_{\nu} $ acting on the wind quanta.
Due to its many nontrivial predictions, the success of a stone’s theory was overwhelming. Among these predictions were the delay of the time shown by a sandglass in an external heatfield and, most striking, the deflection of light in heatfields, often referred to as “Fata Morgana”. (By the way, this effect is also the origin of the common misbelief that dromedars have two hunches.)
Even more so, it turned out that the grainization of this theory was no great deal $^4$. A typical effect that can only be understood in the framework of the full theory of grain dynamics, as it is a typical grain effect of the sand-wind-interaction, is the movement of sandpiles [@quant]. We have already given a detailed account of this application in the preceding section. We have also shown therein that the resulting sand-wind duality leads to our present understanding of the genesis of sandstorms.
But we also mentioned, that this model has some drawbacks to which we shall turn our attention now.
First of all and honestly speaking, the above described mechanism for the genesis of sandstorms is only of limited theoretical value. In fact, no one has ever been able to prove that such extreme states of the sand-wind-field like sandstorms really do exist. (Actually, it is not even clear, whether this theory describes anything realistic at all.) A modern researcher can hardly take a “problem”of this kind serious, of course, but unfortunately it is not the end of the story:
Evidently, there is only a finite amount of energy in the heatfield of the earth. It came as a big surprise, when it has been discovered that certain kinds of sandstorm, so called grain ray bursts (GRB) had energies seemingly exceeding this upper bound, if a stone’s (grainized) theory was valid in this regime.
For a few hundred years this discovery aroused a lot of confusion and many scientists even started to doubt the spherical shape of the earth, until Re-Ez [@reez] remembered the forgotten singularity theorems of and $\;$ , [@Penhawk], for the heatfield . The former, , has also shown that such singularities will “radiate” large amounts of sand.
Assuming that the GRBs stem from the creation of sand due to the presence of a singularity of the heatfield he overcame all problems and eventually caused the (long overlooked) second revolution in the field of sand.
In fact, it soon became clear that there do exist many visible singularities of the heatfield. For instance, the sun is nothing but such a singularity, arising periodically and then moving at the sky. Due to the emittance of sand$^5$ , it disappears again after a certain time, depending on the state of the heatfield, i.e. the temperature. (In summer it will last longer than in winter.)
Quite recently, even the existence of a very massive heat-kernel in the center of earth has been established.
Most importantly and even more surprisingly, however, it turned out that many ordinary (non-GBR) sandstorms can be traced back to such singularities. This motivates our conjecture that [*all*]{} sandstorms are caused by such singularities.
Finally, the interpretation of the genesis of sandstorms as caused by singularities also resolved another puzzle:
Since sand is fermionic (it obviously respects the exclusion principle and, of course, the grains have spin $\frac{1}{2}$) and interacts only via heat-interactions, the system should be stable, seemingly in contradiction to the existence of sandstorms. However, sandstorms are only a local instability in the superfield, just like oases [@acta]. (For this reason, the energy stored in a sandstorm does not exceed the maximal available energy.) Globally, the system is in equilibrium! This fact might be of some importance.
Sandstorms Revisited
====================
So let’s assume that in the center of each sandstorm a singularity of the heatfield is sitting.$^6$ As it is well known, in the eye of a (sand)storm the wind vanishes, and thus it is commonly believed that $WV_H=V_HW=0$ holds even there. But that is wrong, since $V_H$ is infinite in the center of a sandstorm. Accordingly, the sand-wind-duality breaks down in this region, and “that is the poodle’s core ”[@fist].
Before we can explore this fact and its consequences in detail we should describe the mechanism of the emergence of these singularities, however. This mechanism is quite similar to the mechanism discovered by Msw the Elder [@msw], when he tried to understand the so-called solar-new-tiro puzzle. (The problem why the sun emits so few sand, which is solved by a resonance mechanism. At the same time the Msw-resonance explains why the sun emerges only periodically and why it is red around the times of its appearance, respectively disappearance.)
Thus, we immediately infer that small fluctuations in the heatfield will eventually lead, by a similar resonance, to the emergence of a singularity if we apply the following well-established result [@myself]:
[**Theorem 1.**]{} Under the above assumptions, there exists a point $x\in \mathbf{R}^3 $ in which the degree of singularity of the heatfield-distribution $\langle
\mathrm{sand} |h^{\mu\nu}
|\mathrm{sand} \rangle $ is strictly larger than $C_{storm}$.
Thus, the genesis of a sandstorm will be inevitable. Moreover, it is now evident, that the standard sand-wind duality is broken, viz. $$\lim_{r\to x }\langle \mathrm{sand} \label{result}
|V_H W |\mathrm{sand} \rangle (r) \neq 0.$$ Hence, [**the grains in a sandstorm are driven by the wind!**]{} This is the most nontrivial and surprising result of this paper. It can and should be verified empirically.
Conclusion and Outlook
======================
In this paper we presented in a highly suggestive and self-explaining way a new idea about the emergence of sandstorms. We have shown that the movement of the grainized sand is caused by the wind field. This result seems to be in consistency with the predictions of the original (but wrong) classical sand-theory. Note, however, that this “accordance” is not even qualitative: There are no similarities of the classical theory with our completely new and astonishing result (\[result\]).
Of course, this paper has only sketched some basic features of the new sand theory. A complete review is in preparation. There furthergoing questions about noncommutative sand, the super-connection between quicksand and worme holes and some quite interesting (although avantgardistic) new features in sand theory will be presented [@greek]. Moreover, we will overcome some of the flaws this paper is suffering from, e.g. a renormalization of the infantilities due to the self-citation [@myself] will be done. This works the following way: Assuming that the problem is already removed at the first order. Then it follows from an inductive proof that no infantilities will arise at any higher order due to recursive citations. Now, curing the problem at the first loop level is rather simple. This completes the proof of renormalizability of this work.
We should also admit that this work suffers from a considerable lack of computer assistance. Unfortunately, the current generation of the abacus (used here at the Pyramid College) seems unable to perform simulations of such complexity. Not to speak of resolving a heart-tree in the fog.
[**Acknowledgements**]{}\
I would like to thank my colleagues Mario Paschke and Volkmar Putz who made me aware of the reference [@myself]. Without their support and friendship I would not exist. Furthermore I have to thank my namesake P.P. Longstockings, from whom much inspiration has been drawn in making this paper: [*“No one really knows...”*]{}
[99]{}
see fig.1, facsimile taken from [**Sand in our times 15, 199985 .**]{}
$\ \ $
.
K. Murx, [*“Proletariat and Sand,”*]{} work in regress.
Abou Tsa Id, [*“The Wheel and its Possible Applications,”* ]{} [**Tuareg Journal of Modern Physics A 8000 ,**]{} stone tables 12-12.5.
Lei Li Ga, [*“But still it moves,”*]{} [**Crackpot Rev. Let. of Common Sand Vol. MCMDXXXIII.**]{}
Goe-The, [*“Fist I,”*]{} [**Reclam excavations,**]{} edited by the police department of Leipzig.
Edua ’Wit X, [*“M-, F-, Sand-Strings and all that*]{}”, [**Pyramide of Cheops, corridor 6B (periodically under water),**]{} table 1267.
Shrinkhead No. 7, [*“Some least interesting questions of sand”*]{}, [**Cameroon Jungle Drums 6000 .**]{}
reference unknown, but widely distributed.
He, who drank too much of the muddy water (if the hyroglyphes are translated correctly), [*“Do oases really exist? IV : The unreasonable effectiveness of washing clothes with sand.”*]{}, [**Acta Mesopotamica Vol. 4.8 %.**]{}
Pamdirac, [*“Spells on Grain mechanics”*]{}, [**Beduin Acad. Press.**]{}
A most recent review – though written in a somewhat old-fashioned language – can be found under:
[**www.marxist.com/science/dialecticalmaterialism.html**]{}
I. Zeil \[[*innumerable references*]{}\], [**Comm. in Priv. Mat. Vol. 12 - 400.**]{}
\[Author unknown\], [*“The Relative Generality Theory of the Heat Field,”*]{} [**found on a large stone near the Sahara.**]{}
Probably going back to Apu, who had told this to H. Simpson, who had told it to his wife, who had told it to her hair-cutter...
1237 anonymous authors, [*“On Temperature and Evaporation of apples stored in new and old tons,”*]{} [**Clear and Evident Research of Nubia,**]{} reprint of sheepskins 11-27.
Re-Ez, [*“I don’t get the problem ”*]{} contribution to the [**2. Masai-meeting of heat and wind**]{} 2000 .
& $\;$, [**Severe Lectures in Sand,**]{} Knight, 12 .
Msw the Elder [*“What the great god Fe y Nman told me,”*]{} [**Annals of Nil-Pollution, Thebia-Series,**]{} pp. 387-402.
P.P.Marvol [*“A Tip of the TOE,”*]{} physics/0309055.
X. Inachos & Y. Theseus, “[*A Brief Introduction to Brain Theory,*]{}” [**Artemis-Temple-Library of Athens**]{}, destroyed by the Spanish Inquisition.
Figure 1 (The so-called $\Omega^{- -}$-facsimile).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is proved that the space of differential forms with weak exterior- and co-derivative, is compactly embedded into the space of square integrable differential forms. Mixed boundary conditions on weak Lipschitz domains are considered. Furthermore, canonical applications such as Maxwell estimates, Helmholtz decompositions and a static solution theory are proved. As a side product and crucial tool for our proofs we show the existence of regular potentials and regular decompositions as well.'
address: 'Fakultät für Mathematik, Universität Duisburg-Essen, Campus Essen, Germany'
author:
- Sebastian Bauer
- Dirk Pauly
- Michael Schomburg
bibliography:
- 'paule-michael.bib'
date: '; [*Corresponding Author*]{}: Dirk Pauly'
title: 'Weck’s Selection Theorem: The Maxwell Compactness Property for Bounded Weak Lipschitz Domains with Mixed Boundary Conditions in Arbitrary Dimensions'
---
In Memoriam of our Dear Friend and Mentor Karl-Josef (Charlie) Witsch (1948-2017)
Introduction
============
The aim of this contribution is to prove a compact embedding, so called “Weck’s selection theorem” or (generalized) Maxwell compactness property [@weckmaxcomp; @weckmax; @picardcomimb], of differential $q$-forms with weak exterior and co-derivative into the space of square integrable $q$-forms subject to mixed boundary conditions on bounded weak Lipschitz domains $\om\subset\rN$, i.e., $$\begin{aligned}
\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)\hookrightarrow\Ltqom\end{aligned}$$ is compact. The main result is given by Theorem \[satzMKE\]. Here $N\geq2$ and $0\leq q\leq N$ are natural numbers, the dimension of the domain $\om$ and the rank of the differential forms, respectively. This generalises the results from [@bauerpaulyschomburgmaxcompweaklip], where bounded weak Lipschitz domains in the classical setting of $\rz^3$ were considered. In fact, the results from [@bauerpaulyschomburgmaxcompweaklip] can be recovered by setting $N=3$ and $q=1$ or $q=2$.
Similar results for strong Lipschitz domains in three dimensions can be found in [@jochmanncompembmaxmixbc; @FeGi97]. For a historical overview of the mathematical treatment of Weck’s selection theorem (Maxwell compactness property) see [@bauerpaulyschomburgmaxcompweaklip; @leisbook; @picardweckwitschxmas] and the literature cited therein. In particular, let us mention the important papers [@weckmaxcomp; @webercompmax; @picardcomimb; @costabelremmaxlip; @witschremmax; @jochmanncompembmaxmixbc; @picardweckwitschxmas]. We emphasise that in [@witschremmax] Witsch was able to go even beyond Lipschitz regularity ($p$-cusps). In [@weckelacomp] Weck applied Witsch’s ideas to the theory of elasticity.
The central role of compact embeddings of this type can for example be seen in connection with Hilbert space complexes, where the compact embeddings immediately provide closed ranges, solution theories by continuous inverses, Friedrichs/Poincaré-type estimates, and access to Hodge-Helmholtz-type decompositions, Fredholm theory, div-curl-type lemmas, and a-posteriori error estimation, see [@paulyapostfirstordergen; @paulydivcurl; @paulyzulehnerbiharmonic]. In exterior domains, where local versions of the compact embeddings hold, one obtains radiation solutions (scattering theory) with the help of Eidus’ limiting absorption principle [@eiduslabp; @eiduslamp; @eiduslamptwo], see [@paulydiss; @paulytimeharm; @paulystatic; @paulydeco; @paulyasym; @paulypoly]. We elaborate on some of these applications in our Section \[sectApplications\].
Finally we note that by the same arguments as in [@picardcomimb] our results extend to Riemannian manifolds.
Notations, Preliminaries and Outline of the Proof
=================================================
Let $ \om\subset\rN $ be a bounded weak Lipschitz domain. For a precise definition of weak Lipschitz domains, see Definitions \[defilipmani\] and \[defilipsubmani\]. In short, $ \om$ is an $N$-dimensional $\Czo$-submanifold of $\rN$ with boundary, i.e., a manifold with Lipschitz atlas. Let $ \Gamma := \partial \om $, which is itself an $(N-1)$-dimensional Lipschitz-manifold without boundary, consist of two relatively open subsets $\Gamma_\tau$ and $\Gamma_\nu$ such that $\overline\Gamma_\tau \cup \overline\Gamma_\nu = \Gamma$ and $\Gamma_{\tau}\cap\Gamma_{\nu}=\emptyset$. The separating set $ \overline\Gamma_\tau \cap \overline\Gamma_\nu $ (interface) will be assumed to be a, not necessarily connected, $(N-2)$-dimensional Lipschitz-submanifold of $\Gamma$. We shall call $(\om, \Gamma_{\tau})$ a weak Lipschitz pair.
We will be working in the framework of alternating differential forms, see for example [@Janich2001VA]. The vector space $\Ciqc(\om)$ is defined as the subset of $\Ciq(\om)$, the set of smooth alternating differential forms of rank $q$, having compact support in $\om$. Together with the inner product $$\scpLtqom{E}{H} := \int_\om E \wedge \star H$$ it is an inner product space[^1]. We may then define $\Ltqom$ as the completion of $\Ciqc(\om)$ with respect to the corresponding norm. $\Ltqom$ can be identified with those q-forms having $\Lt$-coefficients with respect to any coordinate system. Using the weak version of Stokes’ theorem $$\begin{aligned}
\label{defweakderi}
\scpLtqpoom{\ed E}{H} = -\scpLtqom{E}{\cd H}, \qquad E\in\Ciqc(\om),\,H\in\Ciqpoc(\om),\end{aligned}$$ weak versions of the exterior derivative and co-derivative can be defined. Here $\ed$ is the exterior derivative, $\cd=(-1)^{N(q-1)}\star\ed\star$ the co-derivative and $\star$ the Hodge-star-operator on $\om$. We thus introduce the Sobolev (Hilbert) spaces (equipped with their natural graph norms) $$\begin{aligned}
\Dq(\om):=\left\{E\in\Ltqom: \ed E \in\Ltqpoom\right\},\quad \Deq(\om):=\left\{E\in\Ltqom: \cd E \in\Ltqmo(\om)\right\}\end{aligned}$$ in the distributional sense. It holds $$\star\Dqom=\Degen{N-q}{}{}(\om),\qquad \star\Deqom=\Dgen{N-q}{}{}(\om).$$
We further define the test forms $$\begin{aligned}
\Ciqct(\om) :=\bset{\varphi|_{\om}}{\varphi\in\Ciqc(\rN),~ \dist(\supp \varphi, \Gamma_\tau) > 0}\end{aligned}$$ and note that $ \Ciqc_\emptyset(\om)= \Ciq(\ol\om) $. We now take care of boundary conditions. First we introduce strong boundary conditions as closures of test forms by $$\begin{aligned}
\label{defstark}
\Dqct(\om) :=\overline{\Ciqct(\om)}^{\Dq(\om)}, \quad
\Deqcn(\om) :=\overline{\Ciqcn(\om)}^{\Deq(\om)}.\end{aligned}$$ For the full boundary case $ \Gamma_\tau = \Gamma $ (resp. $ \Gamma_\nu = \Gamma $) we set $$\Dqc(\om) := \Dqct(\om),\quad \Deqc(\om) := \Deqcn(\om).$$ Furthermore, we define weak boundary conditions in the spaces $$\begin{aligned}
\label{defschwach}
\begin{split}
\cDqct(\om) &:=\bset{E\in\Dq(\om)}{\scpLtqom{E}{\cd \varphi} = -\scpLtqpoom{\ed E}{\varphi}~\text{for all}~ \varphi\in\Ciqpocn(\om)},\\
\cDeqcn(\om) &:=\bset{H\in\Deq(\om)}{\scpLtqom{H}{\ed\varphi} = -\scpLtqmoom{\cd H}{\varphi}~\text{for all}~ \varphi\in\Ciqmoct(\om)},
\end{split}\end{aligned}$$ and again for $ \Gamma_\tau = \Gamma $ (resp. $ \Gamma_\nu = \Gamma $) we set $$\cDqc(\om) := \cDqct(\om),\quad \cDeqc(\om) := \cDeqcn(\om).$$ We note that in definitions and the smooth test forms can by mollification be replaced by their respective Lipschitz continuous counterparts, e.g. $\Ciqct(\om)$ can be replaced by $\Czoqct(\om)$. Similarly, in definition the smooth test forms can by completion be replaced by their respective closures, i.e., $\Ciqpocn(\om)$ and $\Ciqmoct(\om)$ can be replaced by $\Deqpocn(\om)$ and $\Dqmoct(\om)$, respectively. In and homogeneous tangential and normal traces on $ \Gamma_\tau $, respectively $ \Gamma_\nu $, are generalised. Clearly $$\Dqct(\om)\subset\cDqct(\om),\quad\Deqcn(\om)\subset\cDeqcn(\om)$$ and it will later be shown that in fact equality holds under our regularity assumptions on the boundary. In case of full boundary conditions the equality even holds without any assumptions on the regularity of the boundary, as can be seen by a short functional analytic argument, see [@bauerpaulyschomburgmaxcompweaklip], but which is unavailable for the mixed boundary case.
We define the closed subspaces $$\begin{aligned}
\Dqz(\om):=\bset{E\in\Dq(\om)}{\ed E = 0},\quad \Deqz(\om):=\bset{E\in\Deq(\om)}{\cd E = 0}\end{aligned}$$ as well as $ \Dqczt(\om) := \Dqct(\om) \cap \Dqz(\om) $ and $\Deqczn(\om) := \Deqcn(\om) \cap \Deqz(\om) $. Analogously for the weak spaces $$\begin{aligned}
\cDqczt(\om) := \cDqct(\om) \cap \Dqz(\om) ,\quad \cDeqczn(\om) := \cDeqcn(\om) \cap \Deqz(\om).\end{aligned}$$
In addition to the latter canonical Sobolev spaces we will also need the classical Sobolev spaces for the Euclidean components of $q$-forms. Note that $\om$, together with the global identity chart, is an $N$-dimensional Riemannian manifold. In particular, $q$-forms $E\in\Ltqom$ can be represented globally in Cartesian coordinates by their components $E_I$, i.e., $E = \sum_{I} E_I \mathrm{d} x^I$. Here we use the ordered multi index notation $\mathrm{d}x^I = \mathrm{d}x^{i_1}\wedge
\cdot\cdot\cdot\wedge\mathrm{d}x^{i_q}$ for $I=(i_1,...,i_q)\in\{1,...,N\}^q$. The inner product for $E,H\in\Ltqom$ is given by $$\begin{aligned}
\scpLtqom{E}{H}
&= \int_{\om} E\wedge\star H
= \sum_I \int_{\om} E_I H_I
= \sum_I \scpLtom{E_I}{H_I}
= \scpLtom{\vec E}{\vec H},\end{aligned}$$ where we introduce the vector proxy notation $$\vec E =[E_I]_I \in \Lt(\om;\rz^{N_q}),\quad N_q := \binom{N}{q}.$$ For $k\in\nz$ we can now define the Sobolev space $\Hkqom$ as the subset of $\Ltqom$ having each component $E_I$ in $\Hkom$. In these cases, we have for $\norm{\alpha} \leq k$ $$\partial^\alpha E
= \sum_I \partial^\alpha E_I \mathrm{d}x^I \quad\text{and}\quad \scp{E}{H}_{\Hkqom}
:= \sum_{0\leq\norm{\alpha}\leq k}\scpLtqom{\partial^\alpha E}{\partial^\alpha H}$$ and we use the vector proxy notation also for the gradient, i.e., $$\na\vec E = [\partial_n E_I]_{n,I} = [...\na E_I...]_I \in \Lt(\om;\rz^{N\times N_q}).$$ In particular, for $E,H\in\Hoqom$ $$\begin{aligned}
\scp{E}{H}_{\Hoqom}
&=\scpLtqom{E}{H}+\sum_{n=1}^N \scpLtqom{\partial_n E}{\partial_n H}
=\sum_I\big(\int_\om E_I H_I + \sum_n \int_\om \partial_n E_I \partial_n H_I\big)\\
&=\sum_I(\scpLtom{E_I}{H_I} + \scpLtom{\na E_I}{\na H_I})
= \scpLtom{\vec E}{\vec H} + \scpLtom{\na\vec E}{\na\vec H}
= \scp{\vec E}{\vec H}_{\Hoom}.\end{aligned}$$ Boundary conditions for $\Hoqom$-forms can again be defined strongly and weakly, i.e., by closure $$\begin{aligned}
\Hoqct(\om) := \overline{\Ciqc_{\Gamma_{\tau}}(\om)}^{\Hoqom}\end{aligned}$$ and by integration by parts $$\cHoqct(\om) :=\bset{E\in\Hoqom}{\scpLtom{E_{I}}{\p_{n}\phi} = -\scpLtom{\p_{n}E_{I}}{\phi}
~\text{for all}~ n,I~\text{and all}~\phi\in\Cicn(\om)},$$ respectively. Le us also introduce the following Sobolev type spaces $$\begin{aligned}
\Dqk(\om):&=\bset{E\in\Hkq(\om)}{ \ed E \in\Hkqpo(\om)}, \\
\Deqk(\om):&=\bset{E\in\Hkq(\om)}{ \cd E \in\Hkqmo(\om)}.\end{aligned}$$
We emphasise that by switching $ \Gamma_\tau $ and $ \Gamma_\nu $ we can define the respective boundary conditions on the other part of the boundary as well. Moreover, all definitions of our spaces extend literally to any open subset $\om\subset\rN$ and any relatively open complementary boundary pairs $\Gamma_\tau$ and $\Gamma_\nu$.
Finally we introduce our transformations $\eps$.
\[defitrafo\] A transformation $\eps:\Ltqom\to\Ltqom$ will be called admissible, if $\eps$ is bounded, symmetric, and uniformly positive definite. More precisely, $\eps$ is a self-adjoint operator on $\Ltqom$ and there exists $\ul{\eps},\ol{\eps}>0$ such that for all $E\in\Ltqom$ $$\ul{\eps}\normLtqom{\eps E}\leq\normLtqom{E}\leq\ol{\eps}\sqrt{\scpLtqom{\eps E}{E}}.$$
Lipschitz Domains
-----------------
Let $\om\subset\rN$ be a bounded domain with boundary $\Gamma:=\p\!\om$. We introduce the setting we will be working in. Define (cf. Figure \[fig:cube\]) $$\begin{aligned}
I&:=(-1,1),&
B&:=I^N\subset\rN, &B_{\pm}&:=\set{x\in B}{\pm x_N>0}, &
B_0&:=\set{x\in B}{x_N = 0}, \\
&&&&B_{0,\pm}&:=\set{x\in B_0}{\pm x_1 > 0 },&
B_{0,0}&:=\set{ x\in B_0 }{ x_1 = 0 }.\end{aligned}$$
\[defilipmani\] $ \om $ is called weak Lipschitz, if the boundary $ \Gamma $ is a Lipschitz submanifold of the manifold $\ol{\om}$, i.e., there exist a finite open covering $ U_1,\dots, U_K\subset \rN $ of $ \Gamma $ and vector fields $ \phi_k : U_k \rightarrow B$, such that for $ k = 1,\dots, K $
- $ \phi_k \in \Czo(U_k,B) $ is bijective and $ \psi_k := \phi_k^{-1} \in \Czo(B,U_k) $,
- $ \phi_k(U_k \cap \om) = B_{-}$
hold.
For $ k = 1,\dots, K $ we have $\phi_k (U_k \setminus \ol{\om}) = B_+ $ and $\phi_k (U_k \cap \Gamma) = B_0$.
\[defilipsubmani\] Let $ \om $ be weak Lipschitz. A relatively open subset $ \Gamma_\tau $ of $ \Gamma $ is called weak Lipschitz, if $ \Gamma_\tau $ is a Lipschitz submanifold of $ \Gamma $, i.e., there are an open covering $ U_1,\dots, U_K \subset \rN $ of $ \Gamma $ and vector fields $ \phi_k := U_k \rightarrow B $, such that for $ k = 1,\dots, K $ and in addition to (i), (ii) in Definition \[defilipmani\] one of
- $ U_k \cap \Gamma_\tau = \emptyset$,
- $ U_k \cap \Gamma_\tau = U_k \cap \Gamma \qimpl \phi_k(U_k \cap \Gamma_\tau) = B_{0} $,
- $ \emptyset \neq U_k \cap \Gamma_\tau \neq U_k \cap \Gamma \qimpl \phi_k(U_k \cap \Gamma_\tau) = B_{0,-} $
holds. We define $ \Gamma_\nu := \Gamma \setminus \ol\Gamma_\tau $ to be the relatively open complement of $\Gamma_\tau$.
A pair $ (\om,\Gamma_\tau) $ conforming to Definitions \[defilipmani\] and \[defilipsubmani\] will be called weak Lipschitz.
If $(\om,\Gamma_{\tau})$ is weak Lipschitz, so is $(\om,\Gamma_{\nu})$. Moreover, for the cases (iii), (iii$'$) and (iii$''$)in Definition \[defilipsubmani\] we further have
- $ U_k \cap \Gamma_\tau = \emptyset \;\impl\; U_k \cap \Gamma_\nu = U_k \cap \Gamma \;\impl\;
\phi_k(U_k \cap \Gamma_\nu) = B_{0}$,
- $ U_k \cap \Gamma_\tau = U_k \cap \Gamma \;\impl\; U_k \cap \Gamma_\nu = \emptyset $,
- $ \emptyset \neq U_k \cap \Gamma_\tau \neq U_k \cap \Gamma \;\impl \;
\emptyset \neq U_k \cap \Gamma_\nu \neq U_k \cap \Gamma \;\impl\; \phi_k(U_k \cap \Gamma_\nu) = B_{0,+} $ and $ \phi_k(U_k \cap \ol{\Gamma}_\tau \cap \ol{\Gamma}_\nu) = B_{0,0} $.
In the literature the notion of a Lipschitz domain $\om\subset\rN$ is often used for a strong Lipschitz domain. For this let us define for $x\in\rN$ $$x':=(x_1,x_2,\dots,x_{N-1}),\quad
x'':=(x_2,\dots,x_{N-1}).$$
\[defistrlip\] $ \om $ is called strong Lipschitz, if there are an open covering $ U_1,\dots, U_K \subset \rN $ of $\Gamma$, rigid body motions $ R_k = A_k + a_k $, $ A_k\in\rNtN $ orthogonal, $ a_k \in \rN $, and $ \xi_k \in \Czo(I^{N-1},I)$, such that for $ k = 1,\dots, K $
- $R_k(U_k \cap \om) = \bset{ x \in B }{ x_N < \xi_k(x') }$.
For $ k = 1,\dots, K $ we have $$R_k(U_k \setminus \ol{\om}) = \bset{ x \in B}{x_N > \xi_k (x')},\quad
R_k(U_k \cap \Gamma) = \bset{ x \in B}{x_N = \xi_k (x')}.$$
\[defistrlipinterface\] Let $ \om $ be strong Lipschitz. A relatively open subset $ \Gamma_\tau $ of $ \Gamma $ is called strong Lipschitz, if there exist an open covering $ U_1,\dots, U_K \subset \rN $ of $ \Gamma $, rigid body motions $R_k$, and $ \xi_k \in \Czo(I^{N-1},I)$, $ \zeta_k \in \Czo(I^{N-2},I) $, such that for $ k = 1,\dots, K $ and in addition to (i) in Definition \[defistrlip\] one of
- $ U_k \cap \Gamma_\tau = \emptyset$,
- $ U_k \cap \Gamma_\tau = U_k \cap \Gamma
\qimpl
R_k(U_k \cap \Gamma_{\tau}) = \bset{ x \in B}{x_N = \xi_k (x')}$,
- $ \emptyset \neq U_k \cap \Gamma_\tau \neq U_k \cap \Gamma
\qimpl
R_k(U_k \cap \Gamma_\tau)
= \bset{ x\in B }{ x_N = \xi_k(x'),~ x_1 < \zeta_k (x'') }$
holds. We define $ \Gamma_\nu := \Gamma \setminus \ol\Gamma_\tau $ to be the relatively open complement of $\Gamma_\tau$.
A pair $ (\om,\Gamma_\tau) $ conforming to Definitions \[defistrlip\] and \[defistrlipinterface\] will be called strong Lipschitz.
If $(\om,\Gamma_{\tau})$ is strong Lipschitz, so is $(\om,\Gamma_{\nu})$. Moreover, for the cases (ii), (ii$'$) and (ii$''$) in Definition \[defistrlipinterface\] we further have
- $ U_k \cap \Gamma_\tau = \emptyset
\;\impl\;
U_k \cap \Gamma_\nu = U_k \cap \Gamma
\;\impl\;
R_k(U_k \cap \Gamma_{\nu}) = \bset{ x \in B}{x_N = \xi_k (x')}$,
- $ U_k \cap \Gamma_\tau = U_k \cap \Gamma \;\impl\; U_k \cap \Gamma_\nu = \emptyset $,
- $ \emptyset \neq U_k \cap \Gamma_\tau \neq U_k \cap \Gamma
\;\impl \;
\emptyset \neq U_k \cap \Gamma_\nu \neq U_k \cap \Gamma
\;\impl\; $ $$\begin{aligned}
R_k(U_k \cap \Gamma_\nu) &= \bset{ x \in B }{ x_N = \xi_k (x'),~ x_1 > \zeta_k (x'') }, \\
R_k(U_k \cap \ol{\Gamma}_\tau \cap \ol{\Gamma}_\nu) &= \bset{ x \in B }{ x_N = \xi_k (x'),~ x_1 = \zeta_k (x'') }.\end{aligned}$$
The following holds:
- $ \om $ strong Lipschitz $ \qimpl $ $ \om $ weak Lipschitz
- $ (\om,\Gamma_\tau) $ strong Lipschitz pair $ \qimpl $ $ (\om,\Gamma_\tau) $ weak Lipschitz pair
For a proof just define $\phi_{k}:=\varphi_{k}\circ R_{k}$ with $\varphi_{k}:U_{k}\to B$ given by $$\varphi_{k}(x):=\dvec{x_{1}-\zeta_{k}(x'')}{x''}{x_{N}-\xi_{k}(x')}.$$ Note that the contrary does not hold as the implicit function theorem is not available for Lipschitz maps.
For later purposes we introduce special notations for the half-cube domain $$\begin{aligned}
\label{halfcube}
\Xi:=B_{-},\quad \gamma:=\p\Xi\end{aligned}$$ and its relatively open boundary parts $\gamma_{\tau}$ and $\gamma_{\nu}:=\gamma\setminus\ol{\gamma_{\tau}}$. We will only consider the cases $$\begin{aligned}
\label{halfcubegn}
\gamma_{\nu}=\emptyset,\quad\gamma_{\nu}=B_{0},\quad\gamma_{\nu}=B_{0,+}\end{aligned}$$ and we note that $\Xi$ and $\gamma$, $\gamma_\tau$, $\gamma_\nu$ are strong Lipschitz, see Figure \[figureone\].
Outline of the Proof
--------------------
Let $(\om,\Gamma_{\tau})$ be a weak Lipschitz pair for a bounded domain $\om\subset\rN$.
- As a first step, we observe $ \Hoqct(\om) = \cHoqct(\om) $, i.e., for the $ \Hoqom$-spaces the strong and weak definitions of the boundary conditions coincide, see Lemma \[Hoct\].
- In the second and essential step, we construct various regular $ \Hoq $-potentials on simple domains, mainly for the half-cube $ \Xi$ from with the special boundary constellations , i.e., $$\cDqczn(\Xi) = \Dqczgn(\Xi) = \ed\Hoqmocgn(\Xi), \quad
\cDeqczn(\Xi) = \Deqczn(\Xi) = \cd\Hoqpocgn(\Xi),$$ see Section \[secregpot\]. Potentials of this type are called regular potentials.
- In the third step, Section \[secstrweakXi\], it is shown that the strong and weak definitions of the boundary conditions coincide on the half-cube $\Xi$ from with the special boundary constellation , i.e., $$\begin{aligned}
\label{pillemann}
\cDqcn(\Xi) = \Dqcn(\Xi), \quad\cDeqcn(\Xi) = \Deqcn(\Xi). \end{aligned}$$
- The fourth step proves the compact embedding on the half-cube $\Xi$ from with the special boundary constellations , i.e., $$\begin{aligned}
\label{MCPXi}
\Dqct(\Xi)\cap\eps^{-1}\Deqcn(\Xi)\hookrightarrow\Ltq(\Xi)\end{aligned}$$ is compact, see Section \[seccompembhalfcube\].
- In the fifth step, Theorem \[theoweakeqstrong\], is established for weak Lipschitz domains, i.e., $$\cDqct(\om) = \Dqct(\om),\quad\cDeqcn(\om) = \Deqcn(\om).$$
- In the last step, we finally prove the compact embedding for weak Lipschitz pairs, i.e., $$\begin{aligned}
\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)\hookrightarrow\Ltqom\end{aligned}$$ is compact, see our main result Theorem \[satzMKE\].
Some Important Results
----------------------
Within our proofs we need a few important technical lemmas. First, the strong and weak definitions of the boundary conditions coincide for $\Hoqom$-forms, which is a density result for $\Hoqom$-forms. This is an immediate consequence of the corresponding scalar result, whose proof can be found in [@jochmanncompembmaxmixbc Lemma 2, Lemma 3] and with a simplified proof in [@bauerpaulyschomburgmaxcompweaklip Lemma 3.1].
\[Hoct\] Let $\om\subset\rN$ be a bounded domain and let $(\om, \Gamma_{\tau})$ be a weak Lipschitz pair as well as $$\Hoqcttr(\om):=\bset{u\in\Hoqom}{u|_{\Gamma_{\tau}}=0}$$ in the sense of traces. Then $\cHoqct(\om)=\Hoqcttr(\om)=\Hoqct(\om)$.
Another crucial tool in our arguments is a universal extension operator for the Sobolev spaces $\Dqk(\om)$ and $\Deqk(\om)$ given in [@hiptmairlizouunivextdiffforms], which is based on the universal extension operator for standard Sobolev spaces $\Hk(\om)$ introduced by Stein in [@steinsingintbook]. “Universality” in this context means that the operator, which is given by a single formula, is able to extend all orders of Sobolev spaces simultaneously. More precisely, the following theorem, which is taken from [@hiptmairlizouunivextdiffforms Theorem 3.6], holds:
\[steinexop\] Let $\om\subset\rN$ be a bounded strong Lipschitz domain. Then for $k\in\nz_0$ and $0\leq q\leq N$ there exists a (universal) linear and continuous extension operator $$\E\,:\,\Dqk(\om)\rightarrow \Dqk(\rN).$$ More precisely, $\E$ satisfies $\E E = E$ a.e. in $\om$ and there exists $c>0$ such that for all $E\in\Dqk(\om)$ $$\normDqkrn{\E E} \leq c\normDqkom{E}.$$ Furthermore, $\E$ can be chosen such that $\E E$ has a fixed compact support in $\rN$ for all $E\in\Dqk(\om)$.
Our third lemma summarises well known and fundamental results for the theory of Maxwell’s equations from [@picardpotential; @picardcomimb]. For this, we denote orthogonality and the orthogonal sum in $\Ltqom$ by $\bot$ and $\oplus$, respectively, and introduce the harmonic Dirichlet and Neumann forms $$\mathcal{H}^{q}_{D}(\om):=\Dqcz(\om)\cap\Deqz(\om),\qquad
\mathcal{H}^{q}_{N}(\om):=\Dqz(\om)\cap\Deqcz(\om),$$ respectively.
\[picardlem\] Let $\om\subset\rN$ be a bounded weak Lipschitz domain. Then the embeddings $$\begin{aligned}
\Dqcom\cap\Deqom\hookrightarrow\Ltqom,\qquad
\Dqom\cap\Deqcom\hookrightarrow\Ltqom\end{aligned}$$ are compact and $\mathcal{H}^{q}_{D}(\om)$, $\mathcal{H}^{q}_{N}(\om)$ are finite-dimensional. Moreover, the Helmholtz decompositions $$\begin{aligned}
\Ltqom&=\ed\Dqmocom\oplus\Deqzom
&
\Ltqom&=\ed\Dqmoom\oplus\Deqczom\\
&=\Dqczom\oplus\cd\Deqpoom
&
&=\Dqzom\oplus\cd\Deqpocom\\
&=\ed\Dqmocom\oplus\mathcal{H}^q_{D}(\om)\oplus\cd\Deqpo(\om),
&
&=\ed\Dqmoom\oplus\mathcal{H}^q_{N}(\om)\oplus\cd\Deqpoc(\om)\end{aligned}$$ are valid. In particular, all ranges are closed subspaces of $\Ltqom$ and $$\begin{aligned}
\ed\Dqmocom
&=\Dqczom\cap\mathcal{H}^q_{D}(\om)^{\bot},
&
\ed\Dqmoom
&=\Dqzom\cap\mathcal{H}^q_{N}(\om)^{\bot},\\
\cd\Deqpoom
&=\Deqzom\cap\mathcal{H}^q_{D}(\om)^{\bot},
&
\cd\Deqpocom
&=\Deqczom\cap\mathcal{H}^q_{N}(\om)^{\bot}.\end{aligned}$$ Furthermore, there exists $c>0$ such that $$c\normLtqom{E}\leq\normLtqpoom{\ed E}+\normLtqmoom{\cd E}$$ holds for all $E\in\Dqcom\cap\Deqom\cap\mathcal{H}^q_{D}(\om)^{\bot}$ and all $E\in\Dqom\cap\Deqcom\cap\mathcal{H}^q_{N}(\om)^{\bot}$, i.e., the Maxwell (or Friedrichs-Poincaré type) estimates are valid.
\[picardcorone\] Let $\om\subset\rN$ be a bounded weak Lipschitz domain. Then $$\begin{aligned}
\Dqcom&=\Dqczom\oplus\big(\Dqcom\cap\cd\Deqpoom\big),
&
\ed\Dqcom&=\ed\big(\Dqcom\cap\cd\Deqpoom\big),\\
\Dqom&=\Dqzom\oplus\big(\Dqom\cap\cd\Deqpocom\big),
&
\ed\Dqom&=\ed\big(\Dqom\cap\cd\Deqpocom\big),\\
\Deqom&=\big(\ed\Dqmocom\cap\Deqom\big)\oplus\Deqzom,
&
\cd\Deqom&=\cd\big(\ed\Dqmocom\cap\Deqom\big),\\
\Deqcom&=\big(\ed\Dqmoom\cap\Deqcom\big)\oplus\Deqczom,
&
\cd\Deqcom&=\cd\big(\ed\Dqmoom\cap\Deqcom\big).\end{aligned}$$
Let $\pi_{q,\om}:\Ltqom\to\cd\Deqpocom$ be the orthonormal Helmholtz projector onto $\cd\Deqpocom$. By the latter corollary $\pi_{q,\om}$ maps $\Dqom$ to $$\Dqom\cap\cd\Deqpocom=\Dqom\cap\Deqczom\cap\mathcal{H}^q_{N}(\om)^{\bot}.$$
\[picardcortwo\] Let $\om\subset\rN$ be a bounded weak Lipschitz domain. Then for all $E\in\Dqom$ it holds $\pi_{q,\om}E\in\Dqom\cap\cd\Deqpocom$ and $\ed\pi_{q,\om}E=\ed E$ as well as $$c\normLtqom{\pi_{q,\om}E}\leq\normLtqpoom{\ed E},$$ with $c$ from Lemma \[picardlem\].
If $\om=\rN$ a similar theory holds true utilising polynomially weighted Sobolev spaces, see [@picardpotential] for details. Let $\pi_{q,\rN}:\Ltq(\rN)\to\Deqz(\rN)$ be the orthonormal Helmholtz projector onto $\Deqz(\rN)$.
\[picardlemrN\] It holds $\mathcal{H}^q_{N}(\rN)=\mathcal{H}^q_{D}(\rN)=\{0\}$ and $$\Ltq(\rN)=\Dqz(\rN)\oplus\Deqz(\rN),\quad
\Dq(\rN)=\Dqz(\rN)\oplus\big(\Dq(\rN)\cap\Deqz(\rN)\big).$$ Moreover, for all $E\in\Dq(\rN)$ it holds $\pi_{q,\rN}E\in\Dq(\rN)\cap\Deqz(\rN)$ and $\ed\pi_{q,\rN}E=\ed E$ as well as $$\norm{\pi_{q,\rN}E}_{\Dq(\rN)}\leq\norm{E}_{\Dq(\rN)}.$$
Regularity in the whole space, see e.g. [@kuhnpaulyregmax (4.7) or Lemma 4.2 (i)], shows the following result.
\[kuhnpaulylemrN\] $\Dq(\rN)\cap\Deq(\rN)=\Hoq(\rN)$ with equal norms. More precisely, $E\in\Dq(\rN)\cap\Deq(\rN)$ if and only if $E\in\Hoq(\rN)$ and $$\norm{E}_{\Hoq(\rN)}^2
=\norm{E}_{\Ltq(\rN)}^2+\norm{\ed E}_{\Ltqpo(\rN)}^2+\norm{\cd E}_{\Ltqmo(\rN)}^2.$$
Regular Potentials {#secregpot}
==================
As one of our main steps (step 4), in Section \[seccompembhalfcube\] the compact embedding is proved on the half-cube $\Xi\subset\rN$. This will be achieved (in step 2) by constructing regular $\Ho(\Xi)$-potentials for $ \ed $-free and $ \cd $-free $\Ltq(\Xi)$-forms, which will then enable us to use Rellich’s selection theorem. This section is devoted to the construction and existence of these regular potentials, i.e., to step 2.
Regular Potentials Without Boundary Conditions
----------------------------------------------
Let us recall $$\ed\Dqmoom=\Dqzom\cap\mathcal{H}^q_{N}(\om)^{\bot},\quad
\cd\Deqpoom=\Deqzom\cap\mathcal{H}^q_{D}(\om)^{\bot}$$ from Lemma \[picardlem\]. The next two lemmas ensure the existence of $ \Hoqom $-potentials without boundary conditions for strong Lipschitz domains.
\[Dqzpot\] Let $\om\subset\rN$ be a bounded strong Lipschitz domain. Then there exists a continuous linear operator $$\T_{\ed}:\Dqzom\cap\mathcal{H}^{q}_{N}(\om)^{\perp}\rightarrow\Hoqmo(\rN)\cap\Deqmoz(\rN)$$ such that for all $E\in\Dqzom\cap\mathcal{H}^{q}_{N}(\om)^{\perp}$ $$\ed\T_{\ed} E=E\quad\text{in }\om.$$ Especially $$\Dqzom\cap\mathcal{H}^{q}_{N}(\om)^{\perp}
=\ed\Hoqmoom=\ed\big(\Hoqmoom\cap\Deqmozom\big)$$ and the regular potential depends continuously on the data. Particularly, these are closed subspaces of $\Ltqom$ and $\T_{\ed}$ is a right inverse to $\ed$. By a simple cut-off technique $\T_{\ed}$ may be modified to $$\T_{\ed}:\Dqzom\cap\mathcal{H}^{q}_{N}(\om)^{\perp}\rightarrow\Hoqmo(\rN)$$ such that $\T_{\ed}E$ has a fixed compact support in $\rN$ for all $E\in\Dqzom\cap\mathcal{H}^{q}_{N}(\om)^{\perp}$.
Suppose $ E\in\Dqzom\cap\mathcal{H}^{q}_{N}(\om)^{\perp} $. By Lemma \[picardlem\] there exists $ H\in\Dqmoom $ with $ \ed H = E $ in $ \om $. Applying Corollary \[picardcortwo\] we get $\pi_{q-1,\om}H\in\Dqmoom\cap\cd\Deqcom $ with $\ed\pi_{q-1,\om}H=\ed H=E$ and $$\norm{\pi_{q-1,\om} H}_{\Dqmoom}\leq c\normLtqom{E}.$$ Note that $\pi_{q-1,\om} H$ is uniquely determined. By the Stein extension operator $\E:\Dgen{0,q-1}{}{}(\om)\rightarrow \Dgen{0,q-1}{}{}(\rN)$ from Lemma \[steinexop\] we have $\E\pi_{q-1,\om}H\in\Dgen{0,q-1}{}{}(\rN)$ with compact support. Projecting again, now with Lemma \[picardlemrN\] onto $ \Deqmoz(\rN) $, we obtain $\pi_{q-1,\rN}\E\pi_{q-1,\om}H\in\Dqmo(\rN)\cap\Deqmoz(\rN)$ (again uniquely determined) with $\ed\pi_{q-1,\rN}\E\pi_{q-1,\om}H=\ed\E\pi_{q-1,\om}H$ and $$\norm{\pi_{q-1,\rN}\E\pi_{q-1,\om}H}_{\Dqmo(\rN)}
\leq\norm{\E\pi_{q-1,\om}H}_{\Dqmo(\rN)}
\leq c\norm{\pi_{q-1,\om}H}_{\Dqmo(\om)}.$$ Lemma \[kuhnpaulylemrN\] shows $\pi_{q-1,\rN}\E\pi_{q-1,\om}H\in\Hoqmo(\rN)\cap\Deqmoz(\rN)$ with $$\norm{\pi_{q-1,\rN}\E\pi_{q-1,\om}H}_{\Hoqmo(\rN)}
=\norm{\pi_{q-1,\rN}\E\pi_{q-1,\om}H}_{\Dqmo(\rN)}.$$ Finally, $\T_{\ed}E:=\pi_{q-1,\rN}\E\pi_{q-1,\om}H\in\Hoqmo(\rN)\cap\Deqmoz(\rN)$ meets our needs as $$\norm{\T_{\ed}E}_{\Hoqmo(\rN)}
\leq c\normLtqom{E}$$ and $\ed\T_{\ed}E=\ed\pi_{q-1,\rN}\E\pi_{q-1,\om}H=\ed\E\pi_{q-1,\om}H=\ed\pi_{q-1,\om}H=\ed H=E$ in $\om$.
By Hodge-$\star$-duality we get a corresponding result for the $ \cd $-operator.
\[Deqzpot\] Let $\om\subset\rN$ be a bounded strong Lipschitz domain. Then there exists a continuous linear operator $$\T_{\cd}:\Deqzom\cap\mathcal{H}_{D}^{q}(\om)^\perp\rightarrow\Hoqpo(\rN)\cap\Dqpoz(\rN),$$ such that for all $E\in\Deqzom\cap\mathcal{H}_{D}^{q}(\om)^\perp$ $$\cd\T_{\cd}E=E\quad\text{in }\om.$$ Especially $$\Deqzom\cap\mathcal{H}_{D}^{q}(\om)^\perp
=\cd\Hoqpoom=\cd\big(\Hoqpoom\cap\Dqpozom\big)$$ and the regular potential depends continuously on the data. In particular, these are closed subspaces of $\Ltqom$ and $\T_{\cd}$ is a right inverse to $\cd$. By a simple cut-off technique $\T_{\cd}$ may be modified to $$\T_{\cd}:\Deqzom\cap\mathcal{H}^{q}_{D}(\om)^{\perp}\rightarrow\Hoqpo(\rN)$$ such that $\T_{\cd}E$ has a fixed compact support in $\rN$ for all $E\in\Deqzom\cap\mathcal{H}^{q}_{D}(\om)^{\perp}$.
Regular Potentials With Boundary Conditions for the Half-Cube
-------------------------------------------------------------
Now we start constructing $\Hoq(\Xi)$-potentials on $\Xi$ with boundary conditions. Let us recall our special setting on the half-cube $$\Xi=B_{-}\quad\text{and}\quad
\gamma_{\nu}=\emptyset,\quad\gamma_{\nu}=B_{0}\quad\text{or}\quad\gamma_{\nu}=B_{0,+}.$$ Furthermore, cf. Figure \[fig:cube\], we extend $ \Xi $ over $ \gamma_\nu $ by $$\begin{aligned}
\widetilde \Xi&\,=
\text{\rm int}(\ol\Xi\cup\ol{\widehat \Xi}),&
\widehat \Xi&:=
\begin{cases}
\set{x\in B}{x_N>0}=B_{+}
&\text{, if }\gamma_{\nu}=B_{0},\\
\set{x\in B}{x_N,x_1>0}
=\set{x\in B_{+}}{x_1>0}=:B_{+,+}
&\text{, if }\gamma_{\nu}=B_{0,+}.
\end{cases}\end{aligned}$$
\[satzD0L6\] There exists a continuous linear operator $$\S_{\ed}:\cDqczgn(\Xi)\rightarrow\Hoqmo(\rN)\cap\Hoqmocgn(\Xi),$$ such that for all $H\in\cDqczgn(\Xi)$ $$\ed\S_{\ed} H=H\quad\text{in }\Xi.$$ Especially $$\cDqczgn(\Xi)=\Dqczgn (\Xi)=\ed\Hoqmocgn(\Xi)=\ed\Dqmocgn(\Xi)=\ed\cDqmocgn(\Xi)$$ and the regular $\Hoqmocgn(\Xi)$-potential depends continuously on the data. In particular, these spaces are closed subspaces of $\Ltq(\Xi)$ and $\S_{\ed}$ is a right inverse to $\ed$. Without loss of generality, $\S_{\ed}$ maps to forms with a fixed compact support in $\rN$.
The case $\gamma_{\nu}=\emptyset$ is done in Lemma \[Dqzpot\]. Hence let $\gamma_{\nu}=B_{0}$ or $\gamma_{\nu}=B_{0,+}$. Suppose $H\in\cDqczgn(\Xi)$ and define $\widetilde H\in\Ltq(\widetilde \Xi)$ as extension of $H$ by zero to $\widehat{\Xi}$ by $$\begin{aligned}
\label{defHschlange}
\widetilde H:=
\begin{cases}
H&\text{in }\Xi,\\
0&\text{in }\widehat \Xi.
\end{cases}\end{aligned}$$ By definition of $\cDqczgn(\Xi)$ (weak boundary condition) it follows $\ed \widetilde H=0$ in $\widetilde \Xi$, i.e., $\widetilde H \in\Dqz(\widetilde \Xi)$. Because $\widetilde \Xi$ is strong Lipschitz and topologically trivial, especially $\mathcal{H}^{q}_{N}(\widetilde{\Xi})=\{0\}$, Lemma \[Dqzpot\] yields a regular potential $ E = \T_{\ed}\widetilde H\in\Hoqmo(\rN)\cap\Dqmoz(\rN)$ with $\ed E=\widetilde H$ in $\widetilde \Xi$ and $$\norm{E}_{\Hoqmo(\rN)}
\leq c\norm{\widetilde{H}}_{\Ltq(\widetilde{\Xi})}
\leq c\norm{H}_{\Ltq(\Xi)}.$$ In particular, $E\in\Hoqmo(\widehat \Xi)$ and $\ed E = 0$ in $\widehat \Xi$, i.e., $E\in\Hoqmo(\widehat\Xi)\cap\Dqmoz(\widehat\Xi)$. Using Lemma \[Dqzpot\] again, this time in $\widehat{\Xi}$, we obtain $F=\T_{\ed}E\in\Hoqmt(\rN)\subset\Hoqmt(\widehat \Xi)$ with $\ed F = E \text{ in }\widehat \Xi$ and $$\norm{F}_{\Hoqmt(\rN)}
\leq c\norm{E}_{\Ltq(\widehat{\Xi})}.$$ Since $ E \in \Hoqmo(\widehat \Xi) $ we have $F\in \Dqomt(\widehat \Xi)$. Let $\E:\Dqomt(\widehat \Xi)\rightarrow\Dqomt(\rN)$ be the Stein extension operator from Lemma \[steinexop\]. Then $$\begin{aligned}
\Abb{\S_{\ed}}{\cDqczgn(\Xi)}{\Hoqmo(\rN)}{H}{E -\ed(\E F)}\end{aligned}$$ is linear and continuous as $$\begin{aligned}
\norm{\S_{\ed}H}_{\Hoqmo(\rN)}
&\leq\norm{E}_{\Hoqmo(\rN)}
+\norm{\E F}_{\Dqomt(\rN)}\\
&\leq\norm{E}_{\Hoqmo(\rN)}
+\norm{F}_{\Dqomt(\widehat\Xi)}
\leq\norm{E}_{\Hoqmo(\rN)}
\leq c\norm{H}_{\Ltq(\Xi)}.\end{aligned}$$ Since $\S_{\ed} H = 0$ in $\widehat \Xi$, we have $\S_{\ed} H|_{\gamma_{\nu}} = 0$, which means $\S_{\ed} H\in\Hoqmocgntr(\Xi)$. Therefore, by Lemma \[Hoct\] we see $\S_{\ed} H \in\Hoqmocgn(\Xi)\subset\Dqmocgn(\Xi)\subset\cDqmocgn(\Xi)$. Moreover, $\ed(\S_{\ed} H)=\ed E=\widetilde{H}$ in $\widetilde\Xi$, especially $\ed(\S_{\ed} H)=H$ in $\Xi$. Finally we note $$\ed\Hoqmocgn(\Xi)
\subset\ed\Dqmocgn(\Xi)
\subset\Dqczgn(\Xi),\,
\ed\cDqmocgn(\Xi)
\subset\cDqczgn(\Xi)
\subset\ed\Hoqmocgn(\Xi),$$ completing the proof.
Again by Hodge-$ \star $-duality we obtain the following.
\[satzLtL6\] There exists a continuous linear operator $$\S_{\cd}:\cDeqczgn(\Xi)\rightarrow\Hoqpo(\rN)\cap\Hoqpocgn(\Xi),$$ such that for all $H\in\cDeqczgn(\Xi)$ $$\cd\S_{\cd} H=H\quad\text{in }\Xi.$$ Especially $$\cDeqczgn(\Xi)=\Deqczgn (\Xi)=\cd\Hoqpocgn(\Xi)=\cd\Deqpocgn(\Xi)=\cd\cDeqpocgn(\Xi)$$ and the regular $\Hoqpocgn(\Xi)$-potential depends continuously on the data. In particular, these spaces are closed subspaces of $\Ltq(\Xi)$ and $\S_{\cd}$ is a right inverse to $\cd$. Without loss of generality, $\S_{\cd}$ maps to forms with a fixed compact support in $\rN$.
Weak and Strong Boundary Conditions Coincide for the Half-Cube {#secstrweakXi}
--------------------------------------------------------------
Now the two main density results immediately follow. We note that this has already been proved for the $\Hoqom$-spaces in Lemma \[Hoct\], i.e., $\cHoqct(\om)=\Hoqct(\om)$.
\[rss\] $\cDqcgn(\Xi)=\Dqcgn(\Xi)$ and $\cDeqcgn(\Xi) =\Deqcgn(\Xi)$.
Suppose $E\in\cDqcgn(\Xi)$ and thus $\ed E\in\cDqpoczgn(\Xi)$. By Lemma \[satzD0L6\] there exists $H = \S_{\ed} \ed E\in\Hoqcgn(\Xi)$ with $\ed H= \ed E$. By Lemma \[satzD0L6\] we get $E-H\in\cDqczgn(\Xi)=\Dqczgn(\Xi)$ and hence $E\in\Dqcgn(\Xi)$.
Weck’s Selection Theorem {#seccompemb}
========================
The Compact Embedding for the Half-Cube {#seccompembhalfcube}
---------------------------------------
First we show the main result on the half-cube $\Xi=B_{-}$ with the special boundary patches $$\gamma_{\nu}=\emptyset,\quad\gamma_{\nu}=B_{0}\quad\text{or}\quad\gamma_{\nu}=B_{0,+}$$ from the latter section. To this end let $\eps$ be an admissible transformation on $\Ltq(\Xi)$ and let us consider the densely defined and closed (unbounded) linear operator $$\ed_{\tau}^{q-1}:\Dqmocgt(\Xi)\subset\Ltqmo(\Xi)\rightarrow\Ltqeps(\Xi)\,;\quad
E\mapsto\ed E$$ together with its (Hilbert space) adjoint $$-\cd^q_{\nu}\eps:=(\ed^{q-1}_{\tau})^*:\eps^{-1}\cDeqcgn(\Xi)\subset\Ltqeps(\Xi)\rightarrow\Ltqmo(\Xi)\,;\quad
H\mapsto-\cd\eps H.$$ Note that by Lemma \[rss\] we have $\cDeqcgn(\Xi)=\Deqcgn(\Xi)$. Here, $\Ltqeps(\Xi)$ denotes $\Ltq(\Xi)$ equipped with the inner product $\scp{\,\cdot\,}{\,\cdot\,}_{\LtqepsXi} := \scpLtqXi{\eps\,\cdot\,}{\,\cdot\,}$. Let $\oplus_{\eps}$ denote the orthogonal sum with respect to the $\Ltqeps$-scalar product. The projection theorem yields immediately:
\[lemHDXi\] The Helmholtz decompositions $$\begin{aligned}
\Ltqeps(\Xi)
&= \Dqczgt (\Xi)\oplus_{\eps}\eps^{-1}\Deqczgn(\Xi),
&
\Dqczgt (\Xi)&=\ed\Hoqmocgt(\Xi),
&
\Deqczgn(\Xi)&=\cd\Hoqpocgn(\Xi)\end{aligned}$$ hold. Moreover, the refined Helmholtz decompositions $$\begin{aligned}
\Dqcgt(\Xi)
&=\ed\Hoqmocgt(\Xi)\oplus_{\eps}\big(\Dqcgt(\Xi)\cap\eps^{-1}\cd\Hoqpocgn(\Xi)\big),\\
\eps^{-1}\Deqcgn(\Xi)
&=\big(\ed\Hoqmocgt(\Xi)\cap\eps^{-1}\Deqcgn(\Xi)\big)\oplus_{\eps}\eps^{-1}\cd\Hoqpocgn(\Xi),\\
\Dqcgt(\Xi)\cap\eps^{-1}\Deqcgn(\Xi)
&=\big(\ed\Hoqmocgt(\Xi)\cap\eps^{-1}\Deqcgn(\Xi)\big)\oplus_{\eps}\big(\Dqcgt(\Xi)\cap\eps^{-1}\cd\Hoqpocgn(\Xi)\big)\end{aligned}$$ are valid, and the respective regular potentials, given by the operators $\S_{\ed}$ and $\S_{\cd}$ from Lemma \[satzD0L6\] and Lemma \[satzLtL6\], respectively, depend continuously on the data.
The projection theorem yields $\Ltqeps(\Xi)=\overline{\ed\Dqmocgt(\Xi)}\oplus_{\eps}\eps^{-1}\cDeqczgn(\Xi)$. Furthermore, $$\overline{\ed\Dqmocgt(\Xi)} = \ed\Dqmocgt(\Xi) = \ed\Hoqmot(\Xi)=\Dqczgt (\Xi)$$ by Lemma \[satzD0L6\] and $$\cDeqczgn(\Xi) =\Deqczgn(\Xi)=\cd\Hoqpon(\Xi)$$ by Lemma \[satzLtL6\]. The other assertions follow immediately.
\[HSG\] The embedding $\Dqcgt(\Xi)\cap\eps^{-1}\Deqcgn(\Xi) \hookrightarrow \Ltqeps(\Xi)$ is compact.
Let $(H_n)_{n\in\nz}$ be a bounded sequence in $\Dqcgt(\Xi)\cap\eps^{-1}\Deqcgn(\Xi)$. By Lemma \[lemHDXi\] we can decompose $$\begin{aligned}
H_n = H_n^{\ed} + H_n^{\cd} = \ed E_n^{\ed} + \eps^{-1}\cd E_n^{\cd}
\in \big(\ed\Hoqmocgt(\Xi) \cap \eps^{-1}\Deqcgn(\Xi)\big)
\oplus_{\eps}\big( \Dqcgt(\Xi)\cap\eps^{-1}\cd\Hoqpocgn(\Xi)\big),\end{aligned}$$ with $E_n^{\ed}=\S_{\ed}H_n^{\ed}$ and $E_n^{\cd}=\S_{\cd}H_n^{\cd}$. Then $\ed H^{\cd}_{n} = \ed H_{n}$ and $\cd\eps H^{\ed}_{n} = \cd\eps H_{n}$ as well as $$\begin{aligned}
\normHoqmoxi{E_n^{\ed}}\leq c\,\normLtqXi{H_n^{\ed}}\leq c\,\normLtqepsXi{H_n},\\
\normHoqpoxi{E_n^{\cd}}\leq c\,\normLtqXi{H_n^{\cd}}\leq c\,\normLtqepsXi{H_n}.\end{aligned}$$ By Rellich’s selection theorem and without loss of generality $ (E_n^{\ed}) $ and $ (E_n^{\cd}) $ converge in $ \Ltqmo(\Xi) $ and $ \Ltqpo(\Xi) $, respectively. Moreover, $$\begin{aligned}
\normLtqepsXi{H_n^{\ed} - H_m^{\ed}}^2
&=\scpLtqepsXi{H_n^{\ed} - H_m^{\ed}}{\ed(E_n^{\ed}-E_m^{\ed})}\\
&=-\scpLtqmoXi{\cd\eps(H_n^{\ed} - H_m^{\ed})}{E_n^{\ed}-E_m^{\ed}}
\leq c\,\normLtqmoXi{E_n^{\ed}-E_m^{\ed}},\\
\normLtqepsXi{H_n^{\cd} - H_m^{\cd}}^2
&=\scpLtqepsXi{H_n^{\cd} - H_m^{\cd}}{\eps^{-1}\cd(E_n^{\cd}-E_m^{\cd})}\\
&=-\scp{\ed(H_n^{\cd} - H_m^{\cd})}{E_n^{\cd}-E_m^{\cd}}_{\Ltqpo(\Xi)}
\leq c\,\norm{E_n^{\cd}-E_m^{\cd}}_{\Ltqpo(\Xi)}.\end{aligned}$$ Thus $(H_n^{\ed})$ and $(H_n^{\cd})$ converge in $\Ltqeps(\Xi)$ and altogether $(H_n)$ converges in $\Ltqeps(\Xi)$ as well.
\[HSGrem\] The use of Helmholtz decompositions and regular potentials in the proof of Lemma \[HSG\] demonstrates the main idea behind an elegant proof of a compact embedding. This general idea carries over to proofs of compact embeddings related to other kinds of Hilbert complexes as well, arising, e.g., in elasticity, general relativity, or biharmonic problems, see for example [@paulyzulehnerbiharmonic].
The Compact Embedding for Weak Lipschitz Domains {#mcpweaklip}
------------------------------------------------
The aim of this section is to transfer Lemma \[HSG\] to arbitrary weak Lipschitz pairs $(\om,\Gamma_{\tau})$. To this end we will employ a technical lemma, whose proof is sketched in [@picardcomimb Section 3] and [@wecktrace Remark 2]. We give a detailed proof in the appendix. Let us consider the following situation: Let $\Theta$, $\widetilde{\Theta}$ be two bounded domains in $\rN$ with boundaries $\Upsilon:=\p\Theta$, $\widetilde\Upsilon:=\p\widetilde\Theta$ and let $\Upsilon_0\subset\Upsilon$ be relatively open. Moreover, let $$\phi:\Theta\to\widetilde\Theta,\qquad
\psi:=\phi^{-1}:\widetilde\Theta\to\Theta$$ be Lipschitz diffeomorphisms, this is, $\phi\in\Czo(\Theta,\widetilde\Theta)$ and $\psi=\phi^{-1}\in\Czo(\widetilde\Theta,\Theta)$. Then $\widetilde\Theta=\phi(\Theta)$, $\widetilde\Upsilon=\phi(\Upsilon)$ and we define $\widetilde\Upsilon_0:=\phi(\Upsilon_0)$.
\[lemtrafo\] Let $E\in\cDgenc{q}{\Upsilon_0}(\Theta)$ resp. $\Dgenc{q}{\Upsilon_0}(\Theta)$ and $H\in\eps^{-1}\cDegenc{q}{\Upsilon_0}(\Theta)$ resp. $\eps^{-1}\Degenc{q}{\Upsilon_0}(\Theta)$ for an admissible transformation $\eps$ on $\Ltq(\Theta)$. Then $$\begin{aligned}
\psi^* E&\in\cDgenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)
\text{ resp. }\Dgenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)&
&\text{and}&
\ed\psi^* E&=\psi^*\ed E,\\
\psi^*H&\in\mu^{-1}\cDegenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)
\text{ resp. }\mu^{-1}\Degenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)&
&\text{and}&
\cd\mu\psi^* H&=\pm\star\ed\psi^*\star\eps H = \pm\star\psi^*\star\cd\eps H,\end{aligned}$$ where $\mu:=(-1)^{qN-1}\star\psi^*\star\eps\phi^*$ is an admissible transformation. Moreover, there exists $c>0$, independent of $E$ and $H$, such that $$\norm{\psi^*E}_{\Dq(\widetilde\Theta)}\leq c\norm{E}_{\Dq(\Theta)}, \qquad
\norm{\psi^*H}_{\mu^{-1}\Deq(\widetilde\Theta)}\leq c\norm{H}_{\eps^{-1}\Deq(\Theta)}.$$
Let $(\om,\Gamma_{\tau})$ be a bounded weak Lipschitz pair as introduced in Definitions \[defilipmani\] and \[defilipsubmani\]. We adjust Lemma \[lemtrafo\] to our situation: Let $U_1,\dots,U_K$ be an open covering of $\Gamma$ according to Definitions \[defilipmani\] and \[defilipsubmani\] and set $U_{0}:=\om$. Therefore $U_0,\dots,U_K$ is an open covering of $\ol\om$. Moreover let $\chi_k\in\Cic(U_k)$, $k\in\{0,\dots,K\}$, be a partition of unity subordinate to the open covering $U_0,\dots,U_K$. Now suppose $k\in\{1,\dots,K\}$. We define $$\begin{aligned}
\om_k&:=U_k\cap\om,&
\Gamma_k&:=U_k\cap\Gamma,&
\Gamma_{\tau,k}&:=U_k\cap\Gamma_{\tau},&
\Gamma_{\nu,k}&:=U_k\cap\Gamma_{\nu},\\
\widehat\Gamma_{k}&:=\p\om_{k},&
\Sigma_k&:=\widehat\Gamma_{k}\setminus\Gamma,&
\widehat\Gamma_{\tau,k}&:=\textrm{int}(\Gamma_{\tau,k}\cup\ol\Sigma_k),&
\widehat\Gamma_{\nu,k}&:=\textrm{int}(\Gamma_{\nu,k}\cup\ol\Sigma_k),\\
&&
\sigma&:=\gamma\setminus\ol B_{0},&
\widehat\gamma_{\tau}&:=\textrm{int}(\gamma_{\tau}\cup\ol\sigma),&
\widehat\gamma_{\nu}&:=\textrm{int}(\gamma_{\nu}\cup\ol\sigma).\end{aligned}$$ Lemma \[lemtrafo\] will from now on be used with $$\Theta:=\om_k,\quad
\widetilde\Theta:=\Xi,\qquad
\phi:=\phi_k:\om_{k}\to\Xi,\quad
\psi:=\psi_k:\Xi\to\om_{k}$$ and with one of the following cases $$\Upsilon_0:=\Gamma_{\tau,k},\quad
\Upsilon_0:=\widehat\Gamma_{\tau,k},\quad
\Upsilon_0:=\Gamma_{\nu,k},\quad
\Upsilon_0:=\widehat\Gamma_{\nu,k}.$$ Then $\Upsilon=\widehat\Gamma_{k}$ and $\widetilde\Upsilon=\phi_{k}(\widehat\Gamma_{k})=\gamma$ as well as (depending on the respective case) $$\begin{aligned}
\widetilde\Upsilon_0
&=\phi_{k}(\Gamma_{\tau,k})
=\gamma_{\tau},&
\widetilde\Upsilon_0
&=\phi_{k}(\widehat\Gamma_{\tau,k})
=\widehat\gamma_{\tau},&
\gamma_{\tau}
&\in\{\emptyset,B_{0},B_{0,-}\},&
\gamma_{\nu}
&=\gamma\setminus\ol\gamma_{\tau},\\
\widetilde\Upsilon_0
&=\phi_{k}(\Gamma_{\nu,k})
=\gamma_{\nu},&
\widetilde\Upsilon_0
&=\phi_{k}(\widehat\Gamma_{\nu,k})
=\widehat\gamma_{\nu},&
\gamma_{\nu}
&\in\{\emptyset,B_{0},B_{0,+}\},&
\gamma_{\tau}
&=\gamma\setminus\ol\gamma_{\nu}.\end{aligned}$$
\[Bminus\] Lemmas \[satzD0L6\], \[satzLtL6\], \[rss\], \[lemHDXi\], \[HSG\] hold for $\gamma_{\nu}=B_{0,-}$ without any (substantial) modification as well.
It is straightforward to show the following:
\[kor320\] Let $(\om,\Gamma_{\tau})$ be a bounded weak Lipschitz pair. Then for $E\in\cDqct(\om)$, respectively, $E\in\Dqct(\om)$ and $H\in\cDeqcn(\om)$, respectively, $H\in\Deqcn(\om)$ we have for $k\in\{1,\dots,K\}$ $$\begin{aligned}
E&\in\cDgenc{q}{\Gamma_{\tau,k}}(\om_k),
&
\chi_k E&\in\cDgenc{q}{\widehat\Gamma_{\tau,k}}(\om_k),
&
H&\in\cDegenc{q}{\Gamma_{\nu,k}}(\om_k),
&
\chi_k H&\in\cDegenc{q}{\widehat\Gamma_{\nu.k}}(\om_k),\\
E&\in\Dgenc{q}{\Gamma_{\tau,k}}(\om_k),
&
\chi_k E&\in\Dgenc{q}{\widehat\Gamma_{\tau,k}}(\om_k),
&
H&\in\Degenc{q}{\Gamma_{\nu,k}}(\om_k),
&
\chi_k H&\in\Degenc{q}{\widehat\Gamma_{\nu.k}}(\om_k).\end{aligned}$$
\[theoweakeqstrong\] Let $(\om,\Gamma_{\tau})$ be a bounded weak Lipschitz pair. Then $\cDqct(\om)=\Dqct(\om)$ and $\cDeqcn(\om)=\Deqcn(\om)$.
Suppose $E\in\cDqct(\om)$. Then $\chi_{0}E\in\Dqc(\om)\subset\Dqct(\om)$ by mollification. Let $k\in\{1,\dots,K\}$. Then $\chi_{k}E\in\cDgenc{q}{\widehat\Gamma_{\tau,k}}(\om_k)$ by Lemma \[kor320\]. Lemma \[lemtrafo\], Lemma \[rss\] (with $\gamma_{\nu}:=\gamma_{\tau}$) and Remark \[Bminus\] yield $$\psi_{k}^*(\chi_{k}E)\in\cDgenc{q}{\widehat\gamma_{\tau}}(\Xi)=\Dgenc{q}{\widehat\gamma_{\tau}}(\Xi),\qquad
\widehat\gamma_{\tau}=\phi_{k}(\widehat\Gamma_{\tau,k}),\qquad
\gamma_{\tau}\in\{\emptyset,B_{0},B_{0,-}\}.$$ Then $\chi_{k}E=\chi_{k}\phi_{k}^*\psi_{k}^*E\in\Dgenc{q}{\widehat\Gamma_{\tau,k}}(\om_k)\subset\Dqct(\om)$ by Lemma \[lemtrafo\]. Hence we see $E=\sum_{k}\chi_k E\in\Dqct(\om)$. $\cDeqcn(\om)=\Deqcn(\om)$ follows analogously or by Hodge-$\star$-duality.
Now the compact embedding for bounded weak Lipschitz pairs $(\om,\Gamma_{\tau})$ can be proved.
\[satzMKE\] Let $(\om,\Gamma_{\tau})$ be a bounded weak Lipschitz pair and let $\eps$ be an admissible transformation on $\Ltqom$. Then the embedding $$\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)\hookrightarrow\Ltqepsom$$ is compact.
Suppose $(E_n)$ is a bounded sequence in $\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)$. Then by mollification $$E_{0,n}:=\chi_0E_n\in\Dqcom\cap\eps^{-1}\Deqcom$$ and $E_{0,n}$ even has compact support in $\om$. By classical results, see [@weckmaxcomp; @weckmax; @picardcomimb], $(E_{0,n})$ contains a subsequence, again denoted by $(E_{0,n})$, converging in $\Ltqepsom$. Let $k\in\{1,\dots,K\}$. By Lemma \[kor320\] $$E_{k,n}:=\chi_kE_n\in\Dgenc{q}{\widehat\Gamma_{\tau,k}}(\om_k),\qquad
\eps E_{k,n}\in\Degenc{q}{\widehat\Gamma_{\nu,k}}(\om_k),$$ and the sequence $(E_{k,n})$ is bounded in $\Dgenc{q}{\widehat\Gamma_{\tau,k}}(\om_k)\cap\eps^{-1}\Degenc{q}{\widehat\Gamma_{\nu,k}}(\om_k)$ by the product rule. By Lemma \[lemtrafo\] we have $\psi_{k}^* E_{k,n}\in\Dgenc{q}{\widehat\gamma_{\tau}}(\Xi)$ and $$\begin{aligned}
\norm{\psi_{k}^* E_{k,n}}_{\Dq(\Xi)}
\leq c\norm{E_{k,n}}_{\Dq(\om_{k})},\end{aligned}$$ showing that $(\psi_{k}^* E_{k,n})$ is bounded in $\Dgenc{q}{\widehat\gamma_{\tau}}(\Xi)$. Analogously, $(\psi_{k}^*E_{k,n})\subset\mu_{k}^{-1}\Degenc{q}{\widehat\gamma_{\nu}}(\Xi)$ is bounded in $\mu_{k}^{-1}\Degenc{q}{\widehat\gamma_{\nu}}(\Xi)$ with the admissible transformation $\mu_{k}:=(-1)^{qN-1}\star\psi_{k}^*\star\eps\phi_{k}^*$. Thus $(\psi_{k}^* E_{k,n})$ is bounded in $$\Dgenc{q}{\widehat\gamma_{\tau}}(\Xi)\cap\mu^{-1}_k\Degenc{q}{\widehat\gamma_{\nu}}(\Xi)
\subset\Dgenc{q}{\widehat\gamma_{\tau}}(\Xi)\cap\mu^{-1}_k\Degenc{q}{\gamma_{\nu}}(\Xi),\quad
\gamma_{\nu}\in\{\emptyset,B_{0},B_{0,+}\},\quad
\widehat\gamma_{\tau}=\gamma\setminus\ol\gamma_{\nu}.$$ Thus, by Lemma \[HSG\] and without loss of generality, $(\psi_{k}^*E_{k,n})$ is a Cauchy sequence in $\Ltq(\Xi)$. Now $$E_{k,n}=\phi_{k}^*\psi_{k}^*E_{k,n}\in\Ltq(\om_{k})$$ and Lemma \[lemtrafo\] yields $$\begin{aligned}
\norm{E_{k,n}-E_{k,m}}_{\Ltq(\om_k)}
\leq c\norm{\psi_{k}^*E_{k,n}-\psi_{k}^*E_{k,m}}_{\Ltq(\Xi)}.\end{aligned}$$ Hence $(E_{k,n})$ is a Cauchy sequence in $\Ltq(\om_k)$ and so in $\Ltqepsom$ for their extensions by zero to $\om$. Finally, extracting convergent subsequences for $k=1,\dots,K$, we see that $$(E_n)=\big(\sum_{k=0}^{K}\chi_kE_n\big)=\big(\sum_{k=0}^{K}E_{k,n}\big)$$ is a Cauchy sequence in $\Ltqepsom$.
Applications {#sectApplications}
============
From now on, let $\om\subset\rN$ be a bounded domain and let $(\om, \Gamma_{\tau})$ be a weak Lipschitz pair as well as $\eps:\Ltqom\rightarrow\Ltqom$ be admissible. Then by Theorem \[satzMKE\] the embedding $$\begin{aligned}
\label{compembappsec}
\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)\hookrightarrow\Ltqom\end{aligned}$$ is compact. The results of this section immediately follow in the framework of a general functional analytic toolbox, see [@paulyapostfirstordergen; @paulydivcurl; @paulyzulehnerbiharmonic]. For details, see also the proofs in [@bauerpaulyschomburgmaxcompweaklip] for the classical case of vector analysis.
The Maxwell Estimate {#secappmaxest}
--------------------
A first consequence of is that the space of so-called “harmonic” Dirichlet-Neumann forms $$\harmdiqeps:=\Dqczt(\om)\cap\eps^{-1}\Deqczn(\om)$$ is finite-dimensional, as the unit ball in $\harmdiqeps$ is compact by . By a standard indirect argument, immediately implies the so-called Maxwell estimate:
\[MA\] There is $c_{\mathsf{m}}>0$, such that for all $E\in\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)\cap\harmdiqeps^{\perp_\eps}$ $$\begin{aligned}
\norm{E}_{\Ltqepsom}
\leq c_{\mathsf{m}}\,\big(\normLtqpoom{\ed E}^2
+\normLtqmoom{\cd\eps E}^2\big)^{1/2}.\end{aligned}$$
Here we denote by $\perp_{\eps}$ orthogonality with respect to the $\Ltqeps(\om)$-inner product.
Helmholtz Decompositions {#sechelmdeco}
------------------------
Applying the projection theorem to the densely defined and closed (unbounded) linear operators $$\ed_{\tau}^{q-1}:\Dqmoct(\om)\subset\Ltqmo(\om)\rightarrow\Ltqepsom\,;\quad
E\mapsto\ed E$$ with (Hilbert space) adjoint (see Theorem \[theoweakeqstrong\]) $$-\cd^q_{\nu}\eps:=(\ed^{q-1}_{\tau})^*:\eps^{-1}\Deqcn(\om)\subset\Ltqepsom\rightarrow\Ltqmoom\,;\quad
H\mapsto-\cd\eps H$$ and $$-\eps^{-1}\cd^{q+1}_{\nu}:\eps^{-1}\Deqpocn(\om)\subset\Ltqpoom\rightarrow\Ltqepsom\,;\quad
H\mapsto-\eps^{-1}\cd H$$ with adjoint (see Theorem \[theoweakeqstrong\]) $$\ed^{q}_{\tau}:=(-\eps^{-1}\cd^{q+1}_{\nu})^*:\Dqct(\om)\subset\Ltqepsom\rightarrow\Ltqpoom\,;\quad
E\mapsto\ed E$$ we obtain the Helmholtz decompositions $$\begin{aligned}
\label{HDom1}
\Ltqeps(\om)
&=\overline{\ed\Dqmoct(\om)}\oplus_{\eps}\eps^{-1}\Deqczn(\om),\\
\label{HDom2}
\Ltqeps(\om)
&=\Dqczt(\om)\oplus_\eps\overline{\eps^{-1}\cd\Deqpocn(\om)}.\end{aligned}$$ Therefore, $\Dqczt(\om)=\overline{\ed\Dqmoct(\om)} \oplus_{\eps}\harmdiqeps$ and, altogether, we get the refined Helmholtz decomposition $$\begin{aligned}
\label{HDom3}
\Ltqeps(\om)=\overline{\ed\Dqmoct(\om)}\oplus_\eps\harmdiqeps\oplus_\eps\overline{\eps^{-1}\cd\Deqpocn(\om)}.\end{aligned}$$
\[HZ\] The orthonormal decompositions $$\begin{aligned}
\Ltqepsom
&=\ed\Dqmoct(\om)\oplus_\eps \eps^{-1}\Deqczn(\om)\\
&=\Dqczt(\om)\oplus_\eps\eps^{-1}\cd\Deqpocn(\om)\\
&=\ed\Dqmoct(\om)\oplus_\eps\harmdiqeps\oplus_\eps\eps^{-1}\cd\Deqpocn(\om)\end{aligned}$$ hold. Furthermore $$\begin{aligned}
\ed\Dqct(\om)
&=\ed\big(\Dqct(\om)\cap\eps^{-1}\cd\Deqpocn(\om)\big)
=\ed\big(\Dqct(\om)\cap\eps^{-1}\Deqczn(\om)\cap\harmdiqeps^{\perp_\eps}\big),\\
\cd\Deqcn(\om)
&=\cd\big(\Deqcn(\om)\cap\eps\ed\Dqmoct(\om)\big)
=\cd\Big(\Deqcn(\om)\cap\eps\big(\Dqczt(\om)\cap\harmdiqeps^{\perp_\eps}\big)\Big)\end{aligned}$$ and $$\begin{aligned}
\ed\Dqmoct(\om)
&=\Dqczt(\om)\cap\harmdiqeps^{\perp_\eps},&
\cd\Deqpocn(\om)
&=\Deqczn(\om)\cap\harmdiqeps^{\perp},\\
\Dqczt(\om)
&=\ed\Dqmoct(\om)\oplus_{\eps}\harmdiqeps,&
\Deqczn(\om)
&=\cd\Deqpocn(\om)\oplus_{\eps^{-1}}\eps\harmdiqeps.\end{aligned}$$ The ranges $\ed\Dqmoct(\om)$ and $\cd\Deqpocn(\om)$ are closed subspaces of $\Ltqepsom$. Moreover, the $\ed$- resp. $\cd$-potentials are uniquely determined in $\Dqct(\om)\cap\eps^{-1}\Deqczn(\om)\cap\harmdiqeps^{\perp_\eps}$ and $\Deqcn(\om)\cap\eps\big(\Dqczt(\om)\cap\harmdiqeps^{\perp_\eps}\big)$, respectively, and depend continuously on their respective images.
For $\eps=\id$ and yield $$\begin{aligned}
\Deqcn(\om)
&=\big(\overline{\ed\Dqmoct(\om)}\cap\Deqcn(\om)\big)
\oplus\Deqczn(\om),\\
\Dqct(\om)
&=\Dqczt(\om)
\oplus\big(\Dqct(\om)\cap\overline{\cd\Deqpocn(\om)}\big)\end{aligned}$$ and thus with , , and $$\begin{aligned}
\cd\Deqcn(\om)
&=\cd\big(\Deqcn(\om)\cap\ol{\ed\Dqmoct(\om)}\big)
=\cd\big(\Dqczt(\om)\cap\Deqcn(\om)\cap\harmdiq^{\perp}\big),\\
\ed\Dqct(\om)
&=\ed\big(\Dqct(\om)\cap\ol{\cd\Deqpocn(\om)}\big)
=\ed\big(\Dqct(\om)\cap\Deqczn(\om)\cap\harmdiq^{\perp}\big).\end{aligned}$$ Now Theorem \[MA\] implies the closedness of the ranges and the continuity of the potentials. The other assertions follow immediately.
\[HZcor\] It holds $$\begin{aligned}
\Dqct(\om)
&=\ed\Dqmoct(\om)\oplus_\eps\big(\Dqct(\om)\cap\eps^{-1}\Deqczn(\om)\big)\\
&=\Dqczt(\om)\oplus_\eps\big(\Dqct(\om)\cap\eps^{-1}\cd\Deqpocn(\om)\big)\\
&=\ed\Dqmoct(\om)\oplus_\eps\harmdiqeps\oplus_\eps\big(\Dqct(\om)\cap\eps^{-1}\cd\Deqpocn(\om)\big),\\
\eps^{-1}\Deqcn(\om)
&=\big(\ed\Dqmoct(\om)\cap\eps^{-1}\Deqcn(\om)\big)\oplus_\eps\eps^{-1}\Deqczn(\om)\\
&=\big(\Dqczt(\om)\cap\eps^{-1}\Deqcn(\om)\big)\oplus_\eps\eps^{-1}\cd\Deqpocn(\om)\\
&=\big(\ed\Dqmoct(\om)\cap\eps^{-1}\Deqcn(\om)\big)\oplus_\eps\harmdiqeps\oplus_\eps\eps^{-1}\cd\Deqpocn(\om).\end{aligned}$$
Static Solution Theory
----------------------
As a further application we turn to the boundary value problem of generalized electro- and magnetostatics with mixed boundary values: Let $F\in\Ltqpoom$, $G\in\Ltqmoom$, $E_\tau,\,E_\nu\in\Ltqepsom$, and let $\eps$ be admissible. The problem is to find $E\in\Dq(\om)\cap\eps^{-1}\Deq(\om)$ such that $$\begin{aligned}
\begin{split}
\label{EMS}
\ed E &= F,\\
\cd \eps E &= G,\\
E - E_{\tau} &\in\Dqct(\om),\\
\eps(E - E_{\nu}) &\in\Deqcn(\om).
\end{split}\end{aligned}$$ For uniqueness, we require the additional conditions $$\begin{aligned}
\label{ONBproj}
\scpLtqepsom{\eps E}{D_\ell} = \alpha_\ell \in\rz,\quad \ell=1,\dots,d,\end{aligned}$$ where $d$ is the dimension and $\{D_\ell\}$ an $\eps$-orthonormal basis of $\harmdiqeps$. The boundary values on $\Gamma_{\tau}$ and $\Gamma_{\nu},$ respectively, are realised by the given volume forms $E_{\tau}$ and $E_{\nu}$, respectively.
\[satzloesung\] admits a solution, if and only if $$E_{\tau}\in\Dq(\om),\quad
E_{\nu}\in\eps^{-1}\Deq(\om),$$ and $$\begin{aligned}
\label{perploes}
F-\ed E_{\tau}\perp\Deqpoczn(\om),
\quad G-\cd \eps E_{\nu}\perp\Dqmoczt(\om).\end{aligned}$$ The solution $E\in\Dq(\om)\cap\eps^{-1}\Deq(\om)$ can be chosen in a way such that condition with $\alpha\in\rz^d$ is fulfilled, which then uniquely determines the solution. Furthermore, the solution depends linearly and continuously on the data.
Note that is equivalent to $$F-\ed E_{\tau}\in\ed\Dqct(\om),\qquad
G-\cd\eps E_{\nu}\in\cd\Deqcn(\om).$$ For homogeneous boundary data, i.e., $E_{\tau}=E_{\nu}=0$, the latter theorem immediately follows from a functional analytic toolbox, see [@paulyapostfirstordergen; @paulydivcurl; @paulyzulehnerbiharmonic], which even states a sharper result: The linear static Maxwell-operator $$\Abb{M}{\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)}{\ed\Dqct(\om)\times\cd\Deqcn(\om)\times\rz^d}
{E}{\big(\ed E,\cd\eps E,(\scpLtqepsom{\eps E}{D_\ell})_{\ell=1}^{d}\big)}$$ is a topological isomorphism. Its inverse $M^{-1}$ maps not only continuously onto $\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)$, but also compactly into $\Ltqepsom$ by . For homogeneous kernel data, i.e., for $$\Abb{M_{0}}{\Dqct(\om)\cap\eps^{-1}\Deqcn(\om)\cap\harmdiqeps^{\perp_\eps}}{\ed\Dqct(\om)\times\cd\Deqcn(\om)}
{E}{(\ed E,\cd\eps E)},$$ we have $\norms{M_{0}^{-1}}\leq(c_{\mathsf{m}}^2+1)^{1/2}$. For details and a proof of Theorem \[satzloesung\] in the classical setting of vector analysis see [@bauerpaulyschomburgmaxcompweaklip].
Remarks on the Transformations
==============================
Let us mention some observations on the transformations, in particular that some results are independent of the admissible transformation $\eps$. For this, let $\om\subset\rN$ be an open set, let $\eps$ and $\mu$ be admissible transformations on $\Ltqom$, and let us recall the arguments leading to in Section \[sechelmdeco\] for the densely defined and closed (unbounded) linear operator $$\ed_{\tau}^{q-1}:\Dqmoct(\om)\subset\Ltqmo(\om)\rightarrow\Lgen{2,q}{\mu}(\om)\,;\quad
E\mapsto\ed E.$$ If Theorem \[theoweakeqstrong\] (weak and strong boundary conditions coincide) is not available, its adjoint is given by $$-\cd^q_{\nu}\mu:=(\ed^{q-1}_{\tau})^*:\mu^{-1}\cDeqcn(\om)\subset\Lgen{2,q}{\mu}(\om)\rightarrow\Ltqmoom\,;\quad
H\mapsto-\cd\mu H$$ yielding instead of the Helmholtz (and refined Helmholtz) decompositions $$\begin{aligned}
\label{HDom1weakL}
\Lgen{2,q}{\mu}(\om)
&=\overline{\ed\Dqmoct(\om)}\oplus_{\mu}\mu^{-1}\cDeqczn(\om),\\
\label{HDom1weakR}
\Dqct(\om)
&=\ol{\ed\Dqmoct(\om)}\oplus_{\mu}\big(\Dqct(\om)\cap\mu^{-1}\cDeqczn(\om)\big),\\
\label{HDom1weakD}
\mu^{-1}\cDeqcn(\om)
&=\big(\ol{\ed\Dqmoct(\om)}\cap\mu^{-1}\cDeqcn(\om)\big)\oplus_{\mu}\mu^{-1}\cDeqczn(\om).\end{aligned}$$
\[epstheo\] Let $\om\subset\rN$ be an open set and let $\eps$ be an admissible transformation on $\Ltqom$.
- Weck’s selection theorem is independent of the transformation $\eps$, i.e., the compactness of the embedding in Theorem \[satzMKE\] does not depend on $\eps$.
- The dimension of $\harmdiqeps$ does not depend on $\eps$, in particular $\dim\harmdiqeps=\harmdiq$.
- If Weck’s selection theorem (Theorem \[satzMKE\]) holds, then the dimension of $\harmdiqeps$ is finite.
\(iii) has already been shown in the beginning of Section \[secappmaxest\].
To show (i), let us assume that the embedding $$\begin{aligned}
\label{comembproofepsind}
\Dqct(\om)\cap\mu^{-1}\cDeqcn(\om)\hookrightarrow\Lgen{2,q}{\mu}(\om)\end{aligned}$$ is compact. Moreover, let $(E_{n})$ be a bounded sequence in $\Dqct(\om)\cap\eps^{-1}\cDeqcn(\om)$. By we have $$\begin{aligned}
\label{helmdecoproofepsmu}
\Dqct(\om)\ni E_{n}=E_{\ed,n}+E_{0,n}
\in\ol{\ed\Dqmoct(\om)}\oplus_{\mu}\big(\Dqct(\om)\cap\mu^{-1}\cDeqczn(\om)\big)\end{aligned}$$ with $\ed E_{n}=\ed E_{0,n}$ and $\norm{E_{\ed,n}}_{\Lgen{2,q}{\mu}(\om)}\,,\,\norm{E_{0,n}}_{\Lgen{2,q}{\mu}(\om)}
\leq\norm{E_{n}}_{\Lgen{2,q}{\mu}(\om)}$. Hence $(E_{0,n})$ is a bounded sequence in $\Dqct(\om)\cap\mu^{-1}\cDeqczn(\om)$ and therefore contains by a $\Lgen{2,q}{\mu}(\om)$-converging subsequence, again denoted by $(E_{0,n})$. By we get $$\mu^{-1}\cDeqcn(\om)\ni\mu^{-1}\eps E_{n}=H_{\ed,n}+H_{0,n}
\in\big(\ol{\ed\Dqmoct(\om)}\cap\mu^{-1}\cDeqcn(\om)\big)\oplus_{\mu}\mu^{-1}\cDeqczn(\om)$$ with $\cd\eps E_{n}=\cd\mu H_{\ed,n}$ and $\norm{H_{\ed,n}}_{\Lgen{2,q}{\mu}(\om)}\,,\,\norm{H_{0,n}}_{\Lgen{2,q}{\mu}(\om)}
\leq\norm{\mu^{-1}\eps E_{n}}_{\Lgen{2,q}{\mu}(\om)}$. Therefore $(H_{\ed,n})$ is a bounded sequence in $\Dqczt(\om)\cap\mu^{-1}\cDeqcn(\om)$ and hence contains by a $\Lgen{2,q}{\mu}(\om)$-converging subsequence, again denoted by $(H_{\ed,n})$. Then by orthogonality, i.e., $\ol{\ed\Dqmoct(\om)}\,\bot\,\cDeqczn(\om)$, $$\begin{aligned}
&\qquad\scpLtqepsom{E_{n}-E_{m}}{E_{n}-E_{m}}\\
&=\scpLtqom{\eps(E_{n}-E_{m})}{E_{\ed,n}-E_{\ed,m}}
+\scpLtqom{\eps(E_{n}-E_{m})}{E_{0,n}-E_{0,m}}\\
&=\scpLtqom{\mu(H_{\ed,n}-H_{\ed,m})}{E_{\ed,n}-E_{\ed,m}}
+\scpLtqom{\mu\mu^{-1}\eps(E_{n}-E_{m})}{E_{0,n}-E_{0,m}}\\
&\leq c\big(\norm{H_{\ed,n}-H_{\ed,m}}_{\Lgen{2,q}{\mu}(\om)}
+\norm{E_{0,n}-E_{0,m}}_{\Lgen{2,q}{\mu}(\om)}\big),\end{aligned}$$ which shows that $(E_{n})$ is a Cauchy sequence in $\Ltqepsom$.
To show (ii), we obtain by $$\begin{aligned}
\label{helmdecoproofepsmuedzero}
\Dqczt(\om)
=\ol{\ed\Dqmoct(\om)}\oplus_{\mu}\qharmdi{q}{\mu},\qquad
\qharmdi{q}{\mu}
=\Dqczt(\om)\cap\mu^{-1}\cDeqczn(\om),\end{aligned}$$ and denote the orthonormal projector on the second component by $\pi$. Then $$\Abb{\hat{\pi}}{\harmdiqeps}{\qharmdi{q}{\mu}}{H}{\pi H}$$ is injective, as $\hat{\pi}E=0$ implies $E\in\ol{\ed\Dqmoct(\om)}\cap\harmdiqeps=\{0\}$, and hence $\dim\harmdiqeps\leq\dim\qharmdi{q}{\mu}$. By symmetry we obtain $\dim\harmdiqeps=\dim\qharmdi{q}{\mu}$, completing the proof.
General Regular Potentials and Decompositions
=============================================
A closer inspection of the proof of Lemma \[satzD0L6\] shows that Lemma \[satzD0L6\] and Lemma \[satzLtL6\] hold for more general situations.
\[defirotpot\] Let $\om\subset\rN$ and let $(\om,\Gamma_{\nu})$ be a bounded strong Lipschitz pair. Moreover, let $\om$ and $\Gamma_{\nu}$ be topologically trivial (then so is $\Gamma_{\tau}$). The pair $(\om,\Gamma_{\nu})$ is called “extendable”, if $\om$ can be extended through $\Gamma_{\nu}$ by zero to $\widehat{\om}$, resulting in a topologically trivial strong Lipschitz domain $\widetilde{\om}=\text{\rm int}(\ol\om\cup\ol{\widehat\om})$.
\[regpotexted\] Let $\om\subset\rN$ and let $(\om,\Gamma_{\nu})$ be a bounded, topologically trivial, and extendable strong Lipschitz pair.
- There exists a continuous linear operator $$\S_{\ed}^{q}:\cDqczn(\om)\rightarrow\Hoqmo(\rN)\cap\Hoqmocn(\om),$$ such that $\ed\S_{\ed}^{q}=\id|_{\cDqczn(\om)}$, i.e., for all $H\in\cDqczn(\om)$ $$\ed\S_{\ed}^{q}H=H\quad\text{in }\om.$$ Especially $$\Dgenc{q}{\Gamma_{\nu},0}(\om)
=\cDqczn(\om)
=\ed\S_{\ed}^{q}\cDqmoczn(\om)
=\ed\Hoqmocn(\om)
=\ed\Dqmocn(\om)
=\ed\cDqmocn(\om)$$ and the regular $\Hoqmocn(\om)$-potential depends continuously on the data. In particular, these spaces are closed subspaces of $\Ltq(\om)$ and $\S_{\ed}^{q}$ is a right inverse to $\ed$. Without loss of generality, $\S_{\ed}$ maps to forms with a fixed compact support in $\rN$.
- The regular decompositions $$\begin{aligned}
\cDgenc{q}{\Gamma_{\nu}}(\om)
=\Dqcn(\om)
&=\Hoqcn(\om)+\ed\Hoqmocn(\om)\\
&=\S_{\ed}^{q+1}\ed\Dqcn(\om)
\dotplus\ed\S_{\ed}^{q}(1-\S_{\ed}^{q+1}\ed)\Dqcn(\om)\\
&=\S_{\ed}^{q+1}\ed\Dqcn(\om)
\dotplus\Dgenc{q}{\Gamma_{\nu},0}(\om)\end{aligned}$$ hold with linear and continuous regular decomposition resp. potential operators, where $\dotplus$ denotes the direct sum. More precisely, $\S_{\ed}^{q+1}\ed+\ed\S_{\ed}^{q}(1-\S_{\ed}^{q+1}\ed)
=\id|_{\Dqcn(\om)}$, i.e., for all $E\in\Dqcn(\om)$ $$E=\S_{\ed}^{q+1}\ed E+\ed\S_{\ed}^{q}(1-\S_{\ed}^{q+1}\ed)E
\in\Hoqcn(\om)+\ed\Hoqmocn(\om)$$ with the linear and continuous regular potential operators $$\begin{aligned}
\S_{\ed}^{q+1}\ed:\Dqcn(\om)
&\to\Hoqcn(\om),\\
\S_{\ed}^{q}(1-\S_{\ed}^{q+1}\ed):\Dqcn(\om)
&\to\Hoqmocn(\om).\end{aligned}$$
- Hodge-$\star$-duality yields the corresponding results for the co-derivative. In particular, there exists a continuous linear $\cd$-right inverse operator $$\S_{\cd}^{q}:\cDeqczn(\om)\rightarrow\Hoqpo(\rN)\cap\Hgenc{1,q+1}{\Gamma_{\nu}}(\om),$$ i.e., $\cd\S_{\cd}^{q}=id|_{\cDeqczn(\om)}$. Moreover, $\Degenc{q}{\Gamma_{\nu},0}(\om)
=\cDeqczn(\om)
=\cd\Hgenc{1,q+1}{\Gamma_{\nu}}(\om)$ and the regular $\Hgenc{1,q+1}{\Gamma_{\nu}}(\om)$-potential depends continuously on the data. Furthermore, the regular decompositions $$\begin{aligned}
\cDegenc{q}{\Gamma_{\nu}}(\om)
=\Degenc{q}{\Gamma_{\nu}}(\om)
&=\Hoqcn(\om)+\cd\Hgenc{1,q+1}{\Gamma_{\nu}}(\om)\\
&=\S_{\cd}^{q-1}\cd\Degenc{q}{\Gamma_{\nu}}(\om)
\dotplus\cd\S_{\cd}^{q}(1-\S_{\cd}^{q-1}\cd)\Degenc{q}{\Gamma_{\nu}}(\om)\end{aligned}$$ hold with the linear and continuous regular potential operators $$\begin{aligned}
\S_{\cd}^{q-1}\cd:\Degenc{q}{\Gamma_{\nu}}(\om)
&\to\Hoqcn(\om),\\
\S_{\cd}^{q}(1-\S_{\cd}^{q-1}\cd):\Degenc{q}{\Gamma_{\nu}}(\om)
&\to\Hgenc{1,q+1}{\Gamma_{\nu}}(\om),\end{aligned}$$ and $\S_{\cd}^{q-1}\cd+\cd\S_{\cd}^{q}(1-\S_{\cd}^{q-1}\cd)
=\id|_{\Degenc{q}{\Gamma_{\nu}}(\om)}$.
For (i) we follow the proof of Lemma \[satzD0L6\]. To show (ii), we first note $\cDgenc{q}{\Gamma_{\nu}}(\om)=\Dqcn(\om)$ by Theorem \[theoweakeqstrong\]. Let $E\in\Dqcn(\om)$. Then $\ed E\in\Dgenc{q+1}{\Gamma_{\nu},0}(\om)$ and by (i) we see $\S_{\ed}^{q+1}\ed E\in\Hoqcn(\om)$ with $\ed(\S_{\ed}^{q+1}\ed E)=\ed E$. Thus $E-\S_{\ed}^{q+1}\ed E\in\Dgenc{q}{\Gamma_{\nu},0}(\om)=\ed\Hoqmocn(\om)$ and $\S_{\ed}^{q}(E-\S_{\ed}^{q+1}\ed E)\in\Hoqmocn(\om)$ with $\ed\S_{\ed}^{q}(E-\S_{\ed}^{q+1}\ed E)=E-\S_{\ed}^{q+1}\ed E$ by (i), yielding $$E=\S_{\ed}^{q+1}\ed E+\ed\S_{\ed}^{q}(1-\S_{\ed}^{q+1}\ed)E
\in\Hoqcn(\om)+\ed\Hoqmocn(\om),$$ which proves the regular decompositions and also the assertions about the regular potential operators. To show the directness of the sums, let $H=\S_{\ed}^{q+1}\ed E\in\Dgenc{q}{\Gamma_{\nu},0}(\om)$ with some $E\in\Dqcn(\om)$. Then $0=\ed H=\ed E$ as $\ed E\in\Dgenc{q+1}{\Gamma_{\nu},0}(\om)$ and thus $H=0$.
\[regpotextedrem\] Let $\om\subset\rN$ and let $(\om,\Gamma_{\tau})$ be a bounded, topologically trivial, and extendable strong Lipschitz pair. Then the Dirichlet-Neumann forms are trivial, i.e., $\harmdiqeps=\{0\}$, which follows immediately by Theorem \[HZ\] and, interchanging $\Gamma_{\tau}$ and $\Gamma_{\nu}$, Lemma \[regpotexted\] (i) as $\ed\Dqmoct(\om)
=\Dqczt(\om)
=\ed\Dqmoct(\om)\oplus_\eps\harmdiqeps$.
Now, assume $(\om,\Gamma_{\tau})$ to be a bounded strong Lipschitz pair and let us recall the partition of unity from Section \[mcpweaklip\]. After some possible adjustments, $U_{k}$ and $\chi_{k}$ can be chosen such that $(\om_{k},\widehat\Gamma_{\nu,k})$ is a bounded, topologically trivial, and extendable strong Lipschitz pair for all $k=0,\dots,K$. Maybe $U_{0}$ has to be replaced by more neighbourhoods $U_{-L},\dots,U_{0}$ to ensure that all pairs $(\om_{k},\widehat\Gamma_{\nu,k})$, $k=-L,\dots,K$, are topologically trivial. Note that for all “inner” indices $k=-L,\dots,0$ we have $\om_{k}=U_{k}$ as well as $\widehat\Gamma_{\nu,k}=\widehat\Gamma_{k}=\p\om_{k}=\p U_{k}$. Then for $E\in\Dqcn(\om)$ we have $\chi_{k}E\in\Dgenc{q}{\widehat\Gamma_{\nu,k}}(\om_{k})$ for all $k$ by Lemma \[kor320\]. Lemma \[regpotexted\] (ii) shows the decomposition $$\begin{aligned}
\chi_{k}E=E_{k}+\ed H_{k}
\in\Hgenc{1,q}{\widehat\Gamma_{\nu,k}}(\om_{k})
+\ed\Hgenc{1,q-1}{\widehat\Gamma_{\nu,k}}(\om_{k})\end{aligned}$$ with potentials depending continuously on $\chi_{k}E$. Extending $E_{k}$ and $H_{k}$ by zero to $\om$ yields $\widetilde{E}_{k}\in\Hoqcn(\om)$ and $\widetilde{H}_{k}\in\Hoqmocn(\om)$ and $$E=\sum_{k}\chi_{k}E=\sum_{k}\widetilde{E}_{k}+\ed\sum_{k}\widetilde{H}_{k}
\in\Hoqcn(\om)+\ed\Hoqmocn(\om).$$ As all operations have been linear and continuous we obtain the regular decomposition and potential representation $$\begin{aligned}
\label{regdecogenLip}
\Dqcn(\om)
=\Hoqcn(\om)+\ed\Hoqmocn(\om),\qquad
\ed\Dqcn(\om)
=\ed\Hoqcn(\om)\end{aligned}$$ with linear and continuous potential operators $$\begin{aligned}
\P_{\ed}^{q}:\Dqcn(\om)&\to\Hoqcn(\om),
&
\S_{\ed}^{q+1}:\ed\Dqcn(\om)&\to\Hoqcn(\om),\\
\Q_{\ed}^{q}:\Dqcn(\om)&\to\Hoqmocn(\om).\end{aligned}$$ Note that by Theorem \[HZ\] $$\begin{aligned}
\label{hemldecospdeco}
\ed\Dqcn(\om)
=\Dgenc{q+1}{\Gamma_{\nu},0}{}(\om)\cap\qharmdi{q+1}{\eps}^{\perp_{\eps}},\qquad
\Dgenc{q+1}{\Gamma_{\nu},0}{}(\om)
=\ed\Dqcn(\om)\oplus_{\eps}\qharmdi{q+1}{\eps},\end{aligned}$$ where here $\Gamma_{\tau}$ and $\Gamma_{\nu}$ are interchanged in the definition of $$\qharmdi{q+1}{\eps}
:=\Dgenc{q+1}{\Gamma_{\nu},0}{}(\om)\cap\eps^{-1}\Degenc{q+1}{\Gamma_{\tau},0}{}(\om).$$
Let us summarise the results related to .
\[regpotextedtheo\] Let $\om\subset\rN$ and let $(\om,\Gamma_{\nu})$ be a bounded strong Lipschitz pair.
- There exists a continuous linear operator $$\S_{\ed}^{q}:\ed\Dgenc{q-1}{\Gamma_{\nu}}(\om)\rightarrow\Hoqmocn(\om),$$ such that $\ed\S_{\ed}^{q}=\id|_{\ed\Dgenc{q-1}{\Gamma_{\nu}}(\om)}$. Especially $$\ed\Dqmocn(\om)
=\ed\Hoqmocn(\om)$$ and the regular $\Hoqmocn(\om)$-potential depends continuously on the data. In particular, these spaces are closed subspaces of $\Ltq(\om)$ and $\S_{\ed}^{q}$ is a right inverse to $\ed$.
- The regular decompositions $$\begin{aligned}
\Dqcn(\om)
&=\Hoqcn(\om)
+\ed\Hoqmocn(\om)
&
\Dgenc{q}{\Gamma_{\nu},0}(\om)
&=\ed\Hoqmocn(\om)
+\big(\Hoqcn(\om)\cap\Dgenc{q}{\Gamma_{\nu},0}(\om)\big)\\
&=\S_{\ed}^{q+1}\ed\Dqcn(\om)
\dotplus\Dgenc{q}{\Gamma_{\nu},0}(\om),
&
&=\ed\Hoqmocn(\om)
\oplus\harmdiq\\
&&
&=\ed\Hoqmocn(\om)
\oplus_{\eps}\harmdiqeps\end{aligned}$$ hold with linear and continuous regular decomposition resp. potential operators, which can be defined explicitly by the orthonormal Helmholtz projectors and the operators $\S_{\ed}^{q}$.
\(i) and the first regular decomposition of (ii) together with the existence of the regular potential operators are clear from the considerations leading to . Let $E\in\Dqcn(\om)$. As $\ed\S_{\ed}^{q+1}\ed E=\ed E$ by (i), we have $E-\S_{\ed}^{q+1}\ed E\in\Dgenc{q}{\Gamma_{\nu},0}(\om)$, showing the second regular decomposition of (ii). As in the proof of Lemma \[regpotexted\] the sum is direct. Finally, (i) and complete the proof.
Note that $\harmdiq$ is a subspace of smooth forms, i.e., $$\harmdiq
=\Dgenc{q}{\Gamma_{\nu},0}{}(\om)
\cap\Degenc{q}{\Gamma_{\tau},0}{}(\om)
\cap\Ciqom.$$
Hodge-$\star$-duality yields the corresponding results for the co-derivative.
\[regpotextcdtheo\] Let $\om\subset\rN$ and let $(\om,\Gamma_{\nu})$ be a bounded strong Lipschitz pair.
- There exists a continuous linear operator $$\S_{\cd}^{q}:\cd\Degenc{q+1}{\Gamma_{\nu}}(\om)\rightarrow\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om),$$ such that $\cd\S_{\cd}^{q}=\id|_{\cd\Degenc{q+1}{\Gamma_{\nu}}(\om)}$. Especially $$\cd\Degenc{q+1}{\Gamma_{\nu}}(\om)
=\cd\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om)$$ and the regular $\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om)$-potential depends continuously on the data. In particular, these spaces are closed subspaces of $\Ltq(\om)$ and $\S_{\cd}^{q}$ is a right inverse to $\cd$.
- The regular decompositions $$\begin{aligned}
\Deqcn(\om)
&=\Hoqcn(\om)
+\cd\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om)
&
\Degenc{q}{\Gamma_{\nu},0}(\om)
&=\cd\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om)
+\big(\Hoqcn(\om)\cap\Degenc{q}{\Gamma_{\nu},0}(\om)\big)\\
&=\S_{\cd}^{q-1}\cd\Deqcn(\om)
\dotplus\Degenc{q}{\Gamma_{\nu},0}(\om),
&
&=\cd\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om)
\oplus\harmdiq\\
&&
&=\cd\Hgenc{1,q+1}{\Gamma_{\nu}}{}(\om)
\oplus\harmdiqeps\end{aligned}$$ hold with linear and continuous regular decomposition resp. potential operators, which can be defined explicitly by the orthonormal Helmholtz projectors and the operators $\S_{\cd}^{q}$.
In the latter theorem for $\cd$ the Dirichlet-Neumann forms have again the usual boundary conditions $$\harmdiqeps=\Dqczt(\om)\cap\eps^{-1}\Deqczn(\om).$$
For the case of no or full boundary conditions, related results on regular potentials and regular decompositions are presented in [@costabelmcintoshbogopoinlipop].
Proof of Lemma \[lemtrafo\] (Pull-Back Lemma for Lipschitz Transformations)
===========================================================================
We start out by proving the assertions for the exterior derivative.
Without Boundary Conditions {#appendixnoBC}
---------------------------
Let $E=\sum_{I}E_{I}\ed x^I \in\Dq(\Theta)$. We have to show $\psi^* E \in \Dq(\widetilde\Theta)$ with $\ed \psi^* E= \psi^* \ed E$.
- Let us first consider $\Phi=\sum_{I}\Phi_{I}\ed x^I\in\Czoq(\Theta)$, i.e., $\Phi_{I}\in\Czo(\Theta)$ for all $I$. In the following we denote by $\widetilde\cdot$ the composition with $\psi$. We have $$\begin{aligned}
\ed\psi_{j}
&= \sum_{i}\partial_{i}\psi_{j}\ed x^i,
&
\psi^* \Phi
&= \sum_{I}\widetilde \Phi_{I}\psi^*\ed x^I
= \sum_{I}\widetilde \Phi_{I}(\ed\psi_{i_{1}})\wedge\cdot\cdot\cdot\wedge(\ed\psi_{i_{q}}),\\
&&
\ed \Phi &=\sum_{I,j}\partial_{j}\Phi_{I}(\ed x_{j})\wedge(\ed x^I).\end{aligned}$$ By Rademacher’s theorem $\widetilde \Phi_{I} = \Phi_{I}\circ\psi$ and $\psi_{j}$ belong to $\Czo(\widetilde\Theta)\subset\Ho(\widetilde\Theta)$ and the chain rule holds, i.e., $\partial_{i}\widetilde \Phi_{I}=\sum_{j}\widetilde{\partial_{j}\Phi_{I}}\partial_{i}\psi_{j}$. As $\psi_{j}\in\Ho(\widetilde\Theta)$ we get $\ed\psi_{j}\in\DSobolev^1_{0}(\widetilde\Theta)$ by $$\scp{\ed\psi_{j}}{\cd\varphi}_{\Lebesgue^{2,1}(\widetilde\Theta)}
=-\scp{\psi_{j}}{\cd\cd\varphi}_{\Lebesgue^{2,0}(\widetilde\Theta)}=0$$ for all $\varphi\in\Citc(\widetilde\Theta) $. Thus by definition we see $$\begin{aligned}
\ed\psi^* \Phi &=\sum_{I}(\ed \widetilde \Phi_{I})\wedge(\ed \psi_{i_{1}})\wedge\cdot\cdot\cdot\wedge (\ed\psi_{i_{q}})
= \sum_{I,i}\partial_{i}\widetilde \Phi_{I}(\ed x^i)\wedge(\ed\psi_{i_{1}})\wedge\cdot\cdot\cdot\wedge (\ed\psi_{i_{q}})\\
&=\sum_{I,i,j}\widetilde{\partial_{j} \Phi_{I}}\partial_{i}\psi_{j}(\ed x^i)\wedge(\ed\psi_{i_{1}})\wedge\cdot\cdot\cdot\wedge (\ed\psi_{i_{q}})
=\sum_{I,j}\widetilde{\partial_{j} \Phi_{I}}(\ed\psi_{j})\wedge(\ed\psi_{i_{1}})\wedge\cdot\cdot\cdot\wedge (\ed\psi_{i_{q}}).\end{aligned}$$ On the other hand it holds $$\psi^*\ed \Phi = \sum_{I,j}\widetilde{\partial_{j} \Phi_{I}}(\psi^*\ed x_{j})\wedge(\psi^*\ed x^I)=\sum_{I,j}\widetilde{\partial_{j} \Phi_{I}}
(\ed\psi_{j})\wedge(\ed\psi_{i_{1}})\wedge\cdot\cdot\cdot\wedge (\ed\psi_{i_{q}}).$$ Therefore, $\psi^*\Phi\in\Dq(\widetilde\Theta)$ and $\ed\psi^* \Phi = \psi^*\ed \Phi$.
- For general $E\in\Dq(\Theta)$ we pick $\Phi\in\Ciqpoc(\widetilde\Theta)$. Note $\supp\Phi\subset\subset\widetilde\Theta=\phi(\Theta)$. Replacing $\psi$ by $\phi$ in (i) we have $\phi^*\star\Phi\in\DNmqmo(\Theta)$ with $\ed\phi^*\star\Phi = \phi^*\ed\star\Phi$ and, since $\phi^*\star\Phi = \sum_{I}\widetilde{(\star\Phi)_{I}}\phi^*\ed x^I$ holds, $\supp\phi^*\star\Phi\subset\subset\Theta$. By standard mollification we obtain a sequence $(\Psi_{n})\subset\CiNmqmoc(\Theta)$ with $\Psi_{n}\rightarrow\phi^*\star\Phi$ in $\DNmqmo(\Theta)$. Furthermore $\star\Psi_{n}\in\Ciqpoc(\Theta)$. Then $$\begin{aligned}
\scp{\psi^*E}{\cd\Phi}_{\Ltq(\widetilde\Theta)}
&=\int_{\widetilde\Theta} \psi^*E\wedge\star\cd\Phi
= \pm\int_{\widetilde\Theta}\psi^*E\wedge\psi^*\phi^*\ed\star\Phi
= \pm \int_{\widetilde\Theta}\psi^*(E\wedge\phi^*\ed\star\Phi)\\
&= \pm\int_{\Theta} E\wedge\phi^* \ed\star\Phi
=\pm\int_{\Theta} E\wedge\ed\phi^*\star\Phi
\leftarrow \pm\int_{\Theta} E\wedge\ed\Psi_{n}\\
&=\pm\int_{\Theta} E\wedge\star\star\ed\star\star\Psi_{n}
=\pm\scp{E}{\cd\star\Psi_{n}}_{\Ltq(\Theta)}\\
&=\pm\scp{\ed E}{\star\Psi_{n}}_{\Ltqpo(\Theta)}
\rightarrow\pm\scp{\ed E}{\star\phi^*\star\Phi}_{\Ltqpo(\Theta)}
=\pm\int_{\Theta}\ed E\wedge\phi^*\star\Phi\\
&=\pm\int_{\widetilde\Theta}\psi^*(\ed E\wedge\phi^*\star\Phi)
=\pm\int_{\widetilde\Theta}(\psi^*\ed E)\wedge\star\Phi
=-\scp{\psi^*\ed E}{\Phi}_{\Ltqpo(\widetilde\Theta)}\end{aligned}$$ and hence $\psi^* E\in\Dq(\widetilde\Theta)$ with $\ed\psi^*E = \psi^*\ed E$.
- Let $E\in\Dq(\Theta)$. By (ii) we know $\psi^* E\in\Dq(\widetilde\Theta)$ with $\ed\psi^*E = \psi^*\ed E$. Hence $$\begin{aligned}
\norm{\psi^* E}^2_{\Ltq(\widetilde\Theta)}
&=\int_{\widetilde\Theta}\psi^* E\wedge \star \psi^* E
=\int_{\Theta}\phi^*\psi^* E\wedge \phi^*\star \psi^* E\\
&=\pm\int_{\Theta} E\wedge \star(\star\phi^*\star\psi^*) E
\leq c\norm{E}^2_{\Ltq(\Theta)}\end{aligned}$$ and $$\begin{aligned}
\norm{\ed\psi^* E}_{\Ltqpo(\widetilde\Theta)} = \norm{\psi^*\ed E}_{\Ltqpo(\widetilde\Theta)}\leq c\norm{\ed E}_{\Ltqpo(\Theta)}.\end{aligned}$$
With Strong Boundary Condition. {#appendixsBC}
-------------------------------
Let $E\in\Dgenc{q}{\Upsilon_0}(\Theta)$ and $(E_{n})\subset\Cgenc{\infty,q}_{\Upsilon_0}(\Theta)$ with $E_{n}\rightarrow E$ in $\Dq(\Theta)$. By Appendix \[appendixnoBC\] (ii) we know $\psi^*E_{n},\psi^*E\in\Dq(\widetilde\Theta)$ with $\ed\psi^*E_{n}=\psi^*\ed E_{n}$ as well as $\ed\psi^*E = \psi^*\ed E$. Furthermore, $\psi^* E_{n}$ has compact support away from $\widetilde\Upsilon_{0}$. By standard mollification we see $\psi^*E_{n}\in\Dgenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)$. Moreover, by \[appendixnoBC\] (iii) $\psi^*E_{n} \rightarrow\psi^*E$ in $\Dq(\widetilde\Theta)$. Therefore $\psi^* E\in\Dgenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)$ with $\ed\psi^* E = \psi^*\ed E$.
With Weak Boundary Condition {#appendixwBC}
----------------------------
Let $E\in\cDgenc{q}{\Upsilon_0}(\Theta)$ and $\Phi\in\Cgenc{\infty,q+1}_{\widetilde\Upsilon_1}(\widetilde\Theta)$, where $\Upsilon_{1} = \Upsilon\setminus\ol\Upsilon_{0}$. By Appendix \[appendixnoBC\] (ii) we again know $\psi^*E\in\Dq(\widetilde\Theta)$ with $\ed\psi^*E=\psi^*\ed E$. Moreover by Appendix \[appendixsBC\] $\phi^*\star\Phi\in\Dgenc{N-q-1}{\Upsilon_{1}}(\Theta)$ and hence $\star\phi^*\star\Phi\in\Degenc{q+1}{\Upsilon_{1}}(\Theta)$. We repeat the calculation from Appendix \[appendixnoBC\] (ii) to arrive at $$\begin{aligned}
\scp{\psi^*E}{\cd\Phi}_{\Ltq(\widetilde\Theta)} &=\int_{\widetilde\Theta} \psi^*E\wedge\star\cd\Phi = \pm\scp{E}{\star\phi^*\ed\star\Phi}_{\Ltq(\Theta)}\\
&=\pm\scp{E}{\star\ed\phi^*\star\Phi}_{\Ltq(\Theta)}=\pm\scp{E}{\cd\star\phi^*\star\Phi}_{\Ltq(\Theta)}\\
&=\pm\scp{\ed E}{\star\phi^*\star\Phi}_{\Ltqpo(\Theta)} = -\scp{\psi^*\ed E}{\Phi}_{\Ltqpo(\widetilde\Theta)} =-\scp{\ed\psi^*E}{\Phi}_{\Ltqpo(\widetilde\Theta)}\end{aligned}$$ and therefore $\psi^* E\in\cDgenc{q}{\widetilde\Upsilon_0}(\widetilde\Theta)$.
Assertions for the Co-Derivative {#appendixcoderi}
--------------------------------
It holds by Appendix \[appendixnoBC\] (ii) $$\begin{aligned}
\eps H\in\Deq(\Theta) \qequi \star\eps H\in\Dgen{N-q}{}{}(\Theta) \qequi \psi^*\star\eps \phi^*\psi^* H\in\Dgen{N-q}{}{}(\widetilde\Theta) \qequi \mu\psi^* H\in\Deq(\widetilde\Theta). \end{aligned}$$ Moreover, using Appendix \[appendixnoBC\] (iii) $\mu$ is admissible since for all $H\in\Ltq(\widetilde\Theta)$ $$\begin{aligned}
\scp{\mu H}{H}_{\Ltq(\widetilde\Theta)} &= \pm\scp{\star\psi^*\star\eps\phi^* H}{H}_{\Ltq(\widetilde\Theta)}
=\pm\scp{\psi^*\star\eps\phi^* H}{\star H}_{\LtNmq(\widetilde\Theta)}\\
&=\pm\int_{\widetilde\Theta}\psi^*\star\eps\phi^* H \wedge H
=\pm\int_{\Theta} \star\eps\phi^* H \wedge \star\star\phi^* H\\
&=\pm\scp{\eps\phi^* H}{\phi^* H}_{\Ltq(\Theta)}
\geq c\norm{\phi^* H}^2_{\Ltq(\Theta)} \geq c\norm{H}^2_{\Ltq(\widetilde\Theta)}.\end{aligned}$$ Furthermore $$\begin{aligned}
\cd\mu\psi^* H&=\pm\star\ed\psi^*\star\eps H = \pm\star\psi^*\star\cd\eps H.\end{aligned}$$ The remaining assertions now follow by Appendix \[appendixnoBC\]-\[appendixwBC\] and Hodge-$\star$-duality.
[^1]: For simplicity we work in a real Hilbert space setting.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We implement an adaptation of the COLA approach, a hybrid scheme that combines Lagrangian perturbation theory with an N-body approach, to model non-linear collapse in chameleon and symmetron modified gravity models. Gravitational screening is modeled effectively through the attachment of a suppression factor to the linearized Klein-Gordon equations.
The adapted COLA approach is benchmarked, with respect to an N-body code both for the $\Lambda$CDM scenario and for the modified gravity theories. It is found to perform well in the estimation of the dark matter power spectra, with consistency of 1% to $k\sim2.5$ h/Mpc. Redshift space distortions are shown to be effectively modeled through a Lorentzian parameterization with a velocity dispersion fit to the data. We find that COLA performs less well in predicting the halo mass functions, but has consistency, within $1\sigma$ uncertainties of our simulations, in the relative changes to the mass function induced by the modified gravity models relative to $\Lambda$CDM.
The results demonstrate that COLA, proposed to enable accurate and efficient, non-linear predictions for $\Lambda$CDM, can be effectively applied to a wider set of cosmological scenarios, with intriguing properties, for which clustering behavior needs to be understood for upcoming surveys such as LSST, DESI, Euclid and WFIRST.
author:
- 'Georgios Valogiannis and Rachel Bean.'
nocite: '[@*]'
title: Efficient simulations of large scale structure in modified gravity cosmologies with comoving Lagrangian acceleration
---
\[firstpage\]
Introduction {#sec:intro}
============
The nature of the unknown mechanism responsible for the accelerated expansion of the universe, as measured by Type 1a supernovae [@Perlmutter:1998np; @Riess:2004nr], baryon acoustic oscillations (BAO) in galaxy clustering [@Eisenstein:2005su; @Percival:2007yw; @Percival:2009xn; @Kazin:2014qga], and the Cosmic Microwave Background (CMB) [@Spergel:2003cb; @Ade:2013zuv; @Ade:2015xua], commonly labeled as “Dark Energy", is one of the most challenging, open questions in modern cosmology. Assuming that Einstein’s General Relativity (GR) is the correct framework to describe gravity at large scales, the recent accelerative phase can be driven either by a cosmological constant term $\Lambda$ with negative pressure, or by introducing a scalar field called quintessence [@wetterich1988cosmology; @PhysRevD.37.3406; @Copeland:2006wr]. The necessary value of $\Lambda$ to account for the observed acceleration rate is extremely small, however, when compared to the predictions from high energy physics, and as a result it has to be fine-tuned [@Weinberg:1988cp]. Such an unattractive feature, together with the need to explore all other alternatives, has motivated the development of theories in which GR breaks down at large scales [@Carroll:2003wy; @Capozziello:2003tk; @Nojiri:2006ri], the so called modified gravity (MG) theories. The existence of MG theories with massive scalar fields can reproduce the recent accelerative phase, however, the fifth forces that arise as a result of their coupling to matter would, in principle, cause large deviations from the tight experimental constraints of GR in the solar system [@Will:2005va].
As a consequence, MG models can only be viable if they reduce to the successful GR phenomenology in the local dense environments (eg. Earth, Solar System) through a restoring screening mechanism [@Khoury:2010xi; @Khoury:2013tda]. Based on the qualitative features of the screening mechanism they exhibit, such schemes are commonly classified in various broad classes: the “chameleons” [@PhysRevLett.93.171104; @PhysRevD.69.044026], where the scalar fields become massive and decouple in regions of high Newtonian potential, the kinetic/“$k$-Mouflage" models [@Babichev:2009ee; @Dvali:2010jz], in which the deviations are screened when fifth forces exceed some critical model dependent value and the Vainshtein mechanism [@VAINSHTEIN1972393], that reproduces GR when large derivatives of the fifth forces are experienced. Similar, in terms of phenomenology, with the chameleons are the symmetrons [@PhysRevLett.104.231301; @Olive:2007aj], which exhibit the additional property of a vanishing coupling in dense regions through symmetry restoration.
This rich spectrum of MG models are theoretically viable and offer observational consequences that are potentially distinguishable from $\Lambda$CDM through a variety of astrophysical characteristics. A number of spectroscopic and photometric Large Scale Structure (LSS) surveys both currently underway, e.g. Dark Energy Survey (DES) [@Abbott:2005b], HyperSuprimeCam (HSC) [^1], and eBOSS [@Comparat:2012hz], and coming online in the coming decade, e.g. DESI [@Levi:2013gra], PSF, LSST [@Abell:2009aa], Euclid [@Laureijs:2011gra] and WFIRST [@Spergel:2013tha], will probe the properties of gravity with remarkable precision in both the linear and non-linear regimes, using galaxy clustering, cluster counts, gravitational lensing and peculiar velocities. They offer an unprecedented opportunity to test the landscape of modified gravity theories observationally with respect to the simplest, $\Lambda$CDM scenario. As a result, simulating the structure formation in the the linear, mildly non-linear and non-linear regimes is necessary for both $\Lambda$CDM and all the alternatives.
A variety of analytical, semi-analytical and numerical approaches have been used to study $\Lambda$CDM and dark energy scenarios in the non-linear regime. Lagrangian perturbative techniques up to first [@Zeldovich:1969sb; @1980lssu.book.....P] or second order [@Bouchet:1994xp], have been shown to produce accurate results for $\Lambda$CDM in the linear and mildly non-linear scales without having to perform a complete numerical treatment of structure formation. They fail to achieve the desired accuracy, however, at smaller, non-linear scales for which a full N-body simulation is required. In light of the computational resources necessary for N-body simulations, and given the successes of Lagrangian Perturbation Theory (LPT), hybrid schemes have been proposed, with the aim of combining the strengths of both approaches. In this paper we focus on the Comoving Lagrangian Acceleration (COLA) hybridization scheme [@Tassev:2013pn]. By evolving the large scales analytically using LPT and the small scales exactly with a full N-body treatment, the COLA method manages to produce accurate results deep in the mildly non-linear regime with only a few number of time steps, making it possible to produce fast results in exchange for some accuracy.
In modified gravity simulations, the need to accurately capture the effects of the fifth forces and the screening mechanism adds a new layer of complexity. For an exact description, one needs to solve the full Klein-Gordon equation, whose non-linearities render the procedure both challenging and computationally expensive. It is natural consequently to investigate whether an inexpensive, approximate scheme can be used instead. A linear treatment of the perturbation equations, together with the linearized Klein-Gordon equation it produces may seem efficient at first, but a more careful examination shows [@2012JCAP...10..002B; @Brax:2013mua] that it fails to incorporate the non-linear screening effects and gives poor results. Effective approaches [@Winther:2014cia] have managed to implement screening successfully, however, following a phenomenological path. An ineffective but computationally fast linearized scheme, can be combined with the attachment of a screening factor for a spherically symmetric configuration, to speed up MG simulations without the sacrifice of much accuracy.
Given the success of Lagrangian approaches in $\Lambda$CDM simulations and the need to develop effecient, but representative, realizations of the LSS in different cosmological scenarios, it is natural to see alternative routes in MG models. The benefits of LPT have already been discussed in the context of generating initial conditions, appropriate for coupled scalar field cosmologies [@Li:2010re] or MG models [@Valkenburg:2015dsa] . In this paper, we study the effectiveness of the COLA hybrid scheme, in which the linear scales are evolved exactly using LPT and the non-linear ones using N-body simulations, for MG scenarios. As far as the N-body component is concerned, the fifth force calculation lies in the solution of the linearized KG equation and an approximate screening implementation through the thin shell factor for a dense sphere, similar to [@Winther:2014cia]. In chameleon-type (and symmetron) models, a scalar field acquires a very large mass within a massive object and consequently decouples due to the Yukawa suppression, so essentially only a fraction of the total mass (thin shell) contributes to the fifth force.
The layout of the paper is as follows: in Sec. \[sec:Formalism\] we first review the MG models studied and the non-linear approaches used in the analysis. In Sec. \[sec:Analysis/Results\] we present our results, assessing the performance for the scheme to predict a number of LSS observables, including the matter power spectrum, the redshift space distortions, and halo mass function, before summarizing the findings and discussing implications for future work in Sec. \[sec:conclusions\].
Formalism {#sec:Formalism}
=========
Modified gravity and screening models {#sec:Modified gravity and screening models}
-------------------------------------
A wide class of viable scalar-tensor theories have been shown to be described by a Horndeski Lagrangian [@Horndeski1974; @PhysRevD.84.064039]. Using a general single scalar field Lagrangian, in the Einstein frame, written in terms of a scalar field $\phi$ and its derivatives, $$\mathcal{L}=\frac{M_{Pl}^2}{2}R+\mathcal{L}(\phi,\partial_{\mu} \phi,\partial_{\mu}\partial^{\mu}\phi)+\mathcal{L}_m(e^{2\beta(\phi)\phi/M_{pl}}g_{\mu\nu},\psi_m),$$ where R is the Ricci scalar, $\phi$ the scalar field, $M_{Pl}$ the reduced planck mass $M_{Pl}=\frac{m_{Pl}}{\sqrt{8\pi G}}$ and ${\mathcal L}_m$ is the Lagrangian for the matter sector, in which the matter fields $\psi_m$ are non-minimally coupled to the scalar field with a dimensionless coupling constant $\beta(\phi)$. In the chameleon and symmetron models, the properties of the single scalar field can be described by a simple, scalar field Lagrangian $$\mathcal{L} = -\frac{1}{2} \left(\nabla\phi \right)^2-V(\phi)$$ where $V(\phi)$ is the self-interacting potential. Varying the action gives us the equations of motion for the scalar field, the Klein-Gordon equation $$\label{klein}
\Box\phi = V_{eff,\phi}$$ where the effective potential combining the self-interaction potential and coupling term is given by $$V_{eff} = V(\phi) + \frac{e^{\beta\phi/M_{pl}}\rho_m}{M_{Pl}}$$ The chameleon screening mechanism lies in the fact that the effective mass of the scalar field calculated at the minimum, $m$, which is given by $$m^2 = \frac{d^2 V_{eff}}{d\phi^2},$$ has to be positive. For the chameleon theories, this requirement is guaranteed through the interplay between a monotonically decreasing potential $V(\phi)$ and an increasing coupling. In the symmetron model, on the other hand, the viability is restored using a “Mexican hat" symmetry breaking potential [@PhysRevLett.104.231301], the behavior of which still gives rise to a positive density-dependent mass.
The observational consequences of such models can be demonstrated by extracting the scalar field profile, $\phi(r)$, produced by the density profile (r) = {
[@c@l@]{} \_c &\
\_ &\
. where $r$ is the radial distance from the center of a compact spherically symmetric configuration of density $\rho_c$ and radius $\RR_c$ (not to be confused with the Ricci scalar R), that is isolated on a uniform density background $\rho_{\infty}$. Under spherical symmetry, (\[klein\]) becomes $$\label{kleinsphere}
\frac{1}{r^2}\frac{d}{dr}\left(r^2\frac{d\phi}{dr}\right) = \left(\frac{\partial V}{\partial \phi} + \frac{\beta(\phi) \rho_c (r)}{M_{Pl}}\right).$$ Even though (\[kleinsphere\]) does not have, in principle, an analytical solution, accurate approximations can be performed for two different configurations, that correspond to opposite regimes with respect to screening [@PhysRevLett.93.171104; @PhysRevD.69.044026]. The first case is that of a large, strongly perturbing object of very large density $\rho_c$, for which the interior field is forced to acquire the value that corresponds to the minimum of the effective potential, $\phi_c$ and the scalar field profile outside the object is given by $$\phi(r) = \phi_{\infty} + \frac{\left(\phi_c-\phi_{\infty}\right) \RR_c}{r}e^{-m_{\infty}r}, r > \RR_c.$$ The corresponding fifth-force experienced by a unit mass particle outside the object is $$\begin{aligned}
\label{force}
F_{\phi}(r) & =& 2\beta_{\infty}^2\left(\frac{\Delta \RR_c}{\RR_c}\right)\frac{GM}{r^2}\left(1+m_{\infty}r\right)e^{-m_{\infty}r}, \end{aligned}$$ where $m_{\infty},\beta_{\infty}$ are respectively the background values of the mass and coupling and M the mass of the object. Given the characteristic large values of the Compton wavelength $\lambda_c \equiv m_{\infty}^{-1}$, the scalar field is essentially free within our scales of interest and the Yukawa suppression can be neglected in (\[force\]), $$\begin{aligned}
\label{screenapprox}
F_{\phi}(r) & \approx & 2\beta_{\infty}^2\left(\frac{\Delta \RR_c}{\RR_c}\right)\frac{GM}{r^2}, \hspace{1cm}m_{\infty}r \ll 1.\end{aligned}$$ The above approximation is valid when the “screening factor" is $$\label{fac}
\frac{\Delta \RR_c}{\RR_c} = \frac{|\phi_{\infty}-\phi_{c}|}{2\beta_{\infty}M_{Pl} \Phi_N} \ll 1,$$ which also defines the criterion for the existence of a thin shell [@PhysRevLett.104.231301; @Khoury:2013tda], whose mass is the fraction of the total that actually contributes to the fifth force, due to the strong Yukawa suppression deep inside dense objects. The Newtonian gravitational potential is denoted by $\Phi_N$ in (\[fac\]). On the other hand, when linear perturbation theory is valid, which is the case when $\frac{\Delta \RR_c}{\RR_c} > 1$, the linearized form of (\[kleinsphere\]) gives $$\begin{aligned}
\label{forcelin}
F_{\phi}(r) & \approx & 2\beta_{\infty}^2\frac{GM}{r^2}, \hspace{1cm}m_{\infty}r \ll 1 \end{aligned}$$ for the fifth force. Based on (\[screenapprox\])-(\[forcelin\]), we see that in the linear regime the fifth force is the same as the Newtonian force with a coupling $2 \beta_{\infty}^2$ and deep in the non-linear (screened) regime, it is suppressed by the thin shell factor (\[fac\]). Furthermore, it should be also noted that, as shown in [@Brax:2012gr], one can derive a pair of functions $\beta(a),m(a)$, for the characterization of a model within the above framework. Unlike models with constant couplings, symmetrons exhibit an additional form of screening [@PhysRevLett.104.231301; @Olive:2007aj] due to the fact that in dense environments symmetry is restored and the coupling $\beta(\phi)$ vanishes.
Adopting this formulation, linear perturbation theory gives [@PhysRevD.80.044027; @Brax:2012gr] for the growth of CDM density perturbations in the quasi-static limit and for sub-horizon scales $$\label{growth}
\ddot{\delta}_m + 2H\dot{\delta}_m = \frac{3}{2} \Omega_m(a)H^2 \delta_m\frac{G_{eff}(k,a)}{G}$$ with $$\label{effective}
\frac{G_{eff}(k,a)}{G} = 1+ \frac{2\beta^2(a)k^2}{k^2 + a^2 m^2(a)}$$ where $a$ is scale factor, with $a=1$ today, and $k$ is the comoving wavenumber.
The effects of gravity modifications at the linear approximation are incorporated in the second term. For very large scales and/or early times (GR regime), $am(a)/k \gg1$ and (\[growth\]) reduces to the standard GR expression in the weak gravity regime, where the Newtonian gravitational potential is given by the Poisson equation, $$\label{poisson}
\nabla^2\Phi_N = \frac{3}{2} \Omega_{m0}\frac{H_0^2}{a}\delta_m.$$ When $am(a)/k \le1$ however (scalar-tensor regime), the second term becomes significant and gives the linearized Klein-Gordon equation for the fifth potential $\phi$ $$\label{fourklein}
\phi(k,a)=-\frac{\beta(a)}{k^2+a^2m^2(a)}\frac{\bar{\rho}_m a^2}{M_{pl}} \delta_m.$$ with the real space expression being $$\label{realklein}
\nabla^2\phi = a^2m^2(a)\phi + \frac{\beta(a) a^2 \bar{\rho}_m}{M_{pl}}\delta_m.$$
### The $f(R)$ model
$f(R)$ theories [@Carroll:2003wy] are widely-studied modified gravity scenarios, that give rise to acceleration on cosmic scales and can be incorporated [@Brax:2008hh] into the chameleon formalism with a constant coupling $\beta=1/\sqrt{6}$ . The first model we tested thus, was the Hu-Sawicky $f(R)$ model [@Hu:2007nk] with a scalar field mass $$m(a) = \left(\frac{1}{3(n+1)}\frac{\bar{R}}{|\bar{f}_{R_0}|}\left(\frac{\bar{R}}{\bar{R}_0}\right)^{n+1}\right)^{\frac{1}{2}}$$ where $$\bar{R}= -3(H_0^2\Omega_{m0})^2\left(a^{-3}+4\frac{\Omega_{\Lambda0}}{\Omega_{m0}}\right)$$ where $H_0$ is the Hubble Constant and $\Omega_{\Lambda0}$ and $\Omega_{m0}$ are, respectively, the dark energy and dark matter fractional energy densities today. The mass takes the form $$m(a)=\frac{1}{2997}\left(\frac{1}{2|\bar{f}_{R_0}|}\right)^{\frac{1}{2}}\frac{\left(\Omega_{m0} a^{-3}+4\Omega_{\Lambda0}\right)^{1+\frac{n}{2}}}{\left(\Omega_{m0}+4\Omega_{\Lambda0}\right)^{\frac{n+1}{2}}} [Mpc/h].$$ Furthermore, the screening factor is given by $$\frac{\Delta \RR_c}{\RR_c} = \frac{3}{2} \abs{\frac{\bar{f}_{R_0}}{\Phi_N}}\left(\frac{\Omega_{m0}+4\Omega_{\Lambda0}}{\Omega_{m0} a^{-3}+4\Omega_{\Lambda0}}\right)^{n+1}.$$ $\bar{f}_{R_0}=\frac{df(R)}{dR}\big|_{z=0}$ and $n$ are the model’s free parameters. In this paper, we consider the model for $n=1$ and $\abs{\bar{f}_{R_0}}=\{10^{-4},10^{-5},10^{-6}\}$. These describe cosmologically viable scenarios whose non-linear properties have been simulated using the full Klein-Gordon equation [@Zhao:2010qy; @Winther:2014cia] with which our results can be compared.
### The symmetron model {#symsec}
The general framework laid previously, can also incorporate the symmetron model, with a “Mexican hat" symmetry breaking potential [@PhysRevLett.104.231301], for which scalar fields couple to matter after $a>a_{ssb}$, with $$\label{symscreen}
\begin{split}
m(a) & = \frac{1}{\lambda_{\phi0}}\sqrt{1-\left(\frac{a_{ssb}}{a}\right)^3} \\
\beta(a) & = \beta_0\sqrt{1-\left(\frac{a_{ssb}}{a}\right)^3}
\end{split}$$ and the coupling vanishes for $a<a_{ssb}$, when symmetry is restored. The screening factor for this model becomes [@Davis:2011pj; @Winther:2014cia] $$\frac{\Delta \RR_c}{\RR_c} = \frac{\Omega_{m0}}{3.0 a_{ssb}^3}\left(\frac{\lambda_{\phi 0}}{Mpc/h}\right)^2\abs{\frac{10^{-6}}{\Phi_N}}$$ We consider this model with values $a_{ssb}=0.5,\beta_0=1$ and $\lambda_{\phi 0}=1Mpc/h$ which again have been shown [@Davis:2011pj] to predict deviations consistent with experimental constraints. It should be also pointed out that, as explained previously, models of this type exhibit field dependent couplings which cause additional screening due to the coupling suppression in dense environments, where symmetry is again restored. This effect is not taken into account in our approximate scheme.
Simulating non-linear clustering {#sec:Non-linear approaches}
--------------------------------
### The N-body method {#sec:Nbody}
The COLA code has been loosely based on A. Klypin’s PM code [@Klypin:1997sk], and this motivates the latter’s use as a comparison for our approximate scheme’s effectiveness. It is also a simple and representative implementation of a Particle-Mesh (PM) N-body code. N-body simulations for MG using the PM code have been performed previously [@Stabenau:2006td; @Laszlo:2007td; @Khoury:2009tk]. For each scenario, we consider 10 simulated realizations, initialized at an initial redshift $z_i=49$, at which density perturbations on the scales we study are linear. After providing a linear power spectrum from the cosmological code CAMB [@Lewis:1999bs] for the desired $\Lambda$CDM cosmology at the time $z_i$, $N_p=256^3$ particles are placed in our simulation box with side L=200 Mpc/h, in a mesh of $512^3$, using 1st order Lagrangian Perturbation Theory (Zel’dovich approximation) [@Zeldovich:1969sb]. The parameters that define our background $\Lambda$CDM cosmology are $\Omega_{m0}=0.25$, $\Omega_{\Lambda0}=0.75$, $h=0.7$, $n_s=1.0$ and $\sigma_8=0.8$. The particle positions are updated, using 500 time steps, through the displacement equation: $$\label{dispL}
\ddot{\mathbf{x}} + 2H \dot{\mathbf{x}} = -\frac{1}{a^2}\nabla_{\mathbf{x}} \Phi_N.$$ In Fig. \[figp\], it is shown that the choice of 500 iterations, which corresponds to steps of $\Delta a=0.00196$ in the scale factor, guarantees convergence at the 0.08% level.
In MG cosmologies, the modified geodesic equation gives, in the weak gravity regime, the modified version of (\[dispL\]), $$\label{dispMG}
\ddot{\mathbf{x}} + \left(2H+\frac{\beta}{M_{Pl}}\dot{\phi}\right) \dot{\mathbf{x}} = -\frac{1}{a^2}\left(\nabla_{\mathbf{x}} \Phi_N + \frac{\beta}{M_{Pl}}\nabla_{\mathbf{x}} \phi \right),$$ where the term $\abs{\frac{\beta}{M_{Pl}}\dot{\phi}}$ is negligible given observational constraints from variations of constants [@Winther:2014cia]. Equation (\[dispMG\]), which also holds for the full non-linear KG description, forms a closed system of equations with (\[poisson\]) and (\[realklein\]) that are solved in the Fourier space for the potentials $\Phi_N$ and $\phi$.
The linearized form of KG equation, (\[realklein\]), does not incorporate the screening effects. To account for the screening effect, we adopt an effective parameterization similar to the one proposed in [@Winther:2014cia]. In section Sec. \[sec:Modified gravity and screening models\], we showed that the linear solution for the fifth force, (\[screenapprox\]), is suppressed by the screening factor deep in the non-linear regime. As a result, we incorporate the screening effects by explicitly attaching the screening factor to the fifth force in accordance with (\[screenapprox\])-(\[forcelin\]) and (\[dispMG\]), $$\label{dispscreen}
\ddot{\mathbf{x}} + 2H \dot{\mathbf{x}}=-\frac{1}{a^2}\left(\nabla_{\mathbf{x}} \Phi_N+ \frac{\Delta \RR_c}{\RR_c} \frac{\beta}{M_{Pl}}\nabla_{\mathbf{x}} \phi \right).$$ To interpolate properly between the screened and the unscreened regime we set $$\label{sfac}
% \[
\frac{\Delta \RR_c}{\RR_c} = \left\{\def\arraystretch{1.2}
\begin{array}{@{}c@{\quad}l@{}}
\frac{\phi(a)}{2\beta(a) M_{Pl} \abs{\Phi_N}} & \text{if $ \frac{\phi(a)}{2\beta(a) M_{Pl} \abs{\Phi_N}} <1$ }\\
1 & \text{if $\frac{\phi(a)}{2\beta(a) M_{Pl} \abs{\Phi_N}}>1.$}\\
\end{array}\right.
%\]$$ Within our approximate scheme, the functions $|\phi_{\infty}-\phi_{c}|$ and $\beta_{\infty}$ have been set equal to the background ones $|\phi(a)|$ and $\beta(a)$ correspondingly, which has been shown to be a good approximation in [@Winther:2014cia].
### The COLA method {#sec:COLA}
The fact that N-body codes manage to simulate the Large Scale Structure accurately but at a significant computational cost, has motivated the development of several analytical perturbative techniques to avoid a full blown N-body simulation. Lagrangian Perturbation Theory (LPT) [@Zeldovich:1969sb; @Bouchet:1994xp] works perturbatively in a Lagrangian displacement field and manages to give accurate results in the Linear and the Mildly Non-Linear regime. However, it quickly fails to capture the non-linearities associated with the smaller scales and consequently it underestimates significantly the power at large $k$. Given that we have to choose between accurate, but expensive N-body simulations and fast but approximate perturbative techniques, it is reasonable to ask whether one can efficiently combine the benefits of both approaches. Such a hybrid method, named COmoving Lagrangian Acceleration (COLA) was proposed in [@Tassev:2013pn]. Here we outline the basic framework and its modifications for MG, while details can be found at [@Tassev:2013pn]. The particle comoving positions are decomposed as a sum of two pieces, in the “manifestly" exact form $$\label{basx}
\mathbf{x} = \mathbf{x}_{res} + \mathbf{x}_{LPT}$$ By defining a new time variable $d\theta \equiv H_0\frac{dt}{a^2}=\frac{H_0}{a}d\eta$, where $\eta$ is conformal time, (\[dispL\]) can be cast in the simpler form $$T^2(\mathbf{x})=-\frac{a^2}{H_0^2}\nabla_{\mathbf{x}} \Phi_N,$$ with $T\equiv \frac{d}{d\theta}=\frac{a}{H_0}\partial_{\eta}=Q(a)\partial_a$, and $Q(a)=a^3\frac{H(a)}{H_0}$. In the Lagrangian description $\mathbf{x}=\mathbf{q}+\mathbf{s}(\mathbf{q},a)$, with $\mathbf{q}$ the initial Eulerian position and $\mathbf{s}$ the Lagrangian displacement and $$T^2(\mathbf{s})=-\frac{a^2}{H_0^2}\nabla_{\mathbf{x}} \Phi_N.$$ One can now solve for the residual displacement in $\Lambda$CDM $$T^2(\mathbf{s}_{res})=-\frac{a^2}{H_0^2}\nabla_{\mathbf{x}} \Phi_N - T^2[D_1(a)] \mathbf{s}_1 - T^2[D_2(a)] \mathbf{s}_2,$$ where $D_1(a)$ and $D_2(a)$ are the first and second order growth factors, respectively, and $\mathbf{s}_1$, $\mathbf{s}_2$ are the Zel’dovich and second order LPT displacements. The fact that the LPT piece is evolved analytically and we only solve numerically for $\mathbf{s}_{res}$, can be interpreted as working on a frame that is co-moving with observers that follow LPT trajectories.
$T$ can be discretized using a Leapfrog scheme [@Quinn:1997iy] to get the core COLA equations for each particle’s position and velocity change between the times $a_i,a_f$ $$\label{fullMG}
\begin{split}
\mathbf{x}(a_f) & = \mathbf{x}(a_i) + \mathbf{\upsilon}(a_c)\int_{a_i}^{a_f}\frac{da}{Q(a)} + \\
& + \left(\mathbf{s}_1(\mathbf{q},a_f)-\mathbf{s}_1(\mathbf{q},a_i)\right)+\\
& + \left(\mathbf{s}_2(\mathbf{q},a_f)-\mathbf{s}_2(\mathbf{q},a_i)\right) \\
\mathbf{\upsilon}(a_f) & = \mathbf{\upsilon}(a_i) - \left(\int_{a_i}^{a_f}\frac{a}{a_cQ(a)}da\right) \times \\
& \left[ - 1.5\Omega_{m0} a_c\left(\nabla_{\mathbf{x}} \tilde{\Phi}_{N}(\mathbf{x})+\frac{\Delta \RR_c}{\RR_c}\frac{\beta}{M_{Pl}}\nabla_{\mathbf{x}} \tilde{\phi}(\mathbf{x})\right) - \right. \\
& \left. - T^2[\mathbf{s}_1](a_c)-T^2[\mathbf{s}_2](a_c)\right],
\end{split}$$ where a tilde denotes a quantity in units of $1.5 \Omega_{m0}H_0^2/a$.
Initial conditions are produced using the 2LPT initial conditions code (2LPTic) [@Scoccimarro:1997gr] which does so by performing LPT up to second order. In a $\Lambda$CDM cosmology, growth functions $D_1(a)$ and $D_2(a)$ are scale independent [@Zeldovich:1969sb; @Bouchet:1994xp] and one only needs to produce an LPT snapshot for $z=0$ for both generating initial conditions and obtaining the LPT terms at the different timesteps. In such a case, the LPT displacements are given by $\mathbf{s}_{1}(\mathbf{q},a)=D_{1}(a)\mathbf{s}_{1}(\mathbf{q},a_0)$, $\mathbf{s}_{2}(\mathbf{q},a)=D_{2}(a)\mathbf{s}_{2}(\mathbf{q},a_0)$ and (\[fullMG\]) reduces to the standard COLA $\Lambda$CDM scheme (with the fifth force term omitted). Initial conditions and background cosmology are produced, for 10 realizations, for the same cosmological parameters as used in the PM code, at the initial redshift z=9.0 which has been shown [@Tassev:2013pn] to work well for COLA in $\Lambda$CDM. The simulation box size, number of particles and mesh size are the same as used in the PM code. It should be noted though that we don’t perform a comparison of the codes by initiating both with identically seeded initial conditions, but instead, we compare the statistical consistency of the means of the 10 runs for each of the two techniques with the respective sets each using different random generated seeds. In its initial formulation, COLA was used with 10 time steps, which enables accurate predictions down to $k\sim 0.5 $ h/Mpc, which can be also seen in Fig. \[figp\], where the $\Lambda$CDM power spectrum by COLA is presented for various choices of time steps. By increasing the number of steps to 50, still significantly fewer than the typical number of iterations performed in a standard N-body code, we can provide accuracy down to smaller scales, $k \sim2 $ h/Mpc. The $\Lambda$CDM COLA run-time in this set up is $\sim$10 times shorter than that of the PM code. COLA’s accuracy as a function of the number of time steps used, is further discussed in section \[sec:Modified gravity results\].
When gravity is modified, the core equations need to be changed appropriately to account for the additional fifth forces and the screening effects. As in the N-body code, one can use an approximate framework to model the modified gravity effects both on the growth rate and screening: solving the linearized KG equation (\[fourklein\]) and attaching the screening factor (\[sfac\]) to fifth force term in (\[fullMG\]). For the LPT component of COLA, one must consider that, in MG theories, the growth factors $D_1$ and $D_2$ become scale dependent.
In Figure \[fig1\], we summarize the first and second order growth factors for a chameleon and symmetron scenario. In each case as we approach late times, $z<2$ for the chameleon model, and $z<0.5$ for the symmetron, we find significant scale dependent deviations from $\Lambda$CDM at the level of 10% for $D_1$ and 25% for $D_2$ at k=0.1 h/Mpc today, which means that not all Fourier modes evolve the same way with time [@Valkenburg:2015dsa]. This causes the LPT trajectories of a given particle to bend, in principle. As a consequence one has to be very cautious about how to obtain the LPT terms at the different times. We briefly outline the application of LPT to scale dependent growth functions in MG in appendix \[App:AppendixA\].
Unlike the $\Lambda$CDM case, here the growth factors’ scale dependent nature does not allow one to evolve the Zel’dovich and $2^{nd}$ order displacements with a single scale independent function for all scales. To account for that, we have considered two alternative modifications to COLA. In the first approach, we create an MG version of COLA that calculates the LPT displacements numerically at each time step using an MG version of 2LPTic. The relevant LPT terms in (\[fullMG\]) are calculated after Fourier transforming (\[fourdisps\]) and (\[fouraccels\]). Besides the modified N-body component, the fact that we have to solve numerically for the Lagrangian terms at every discrete time step increases the computational cost significantly. In a second approach, we utilize the fact that the LPT part of the scheme serves to evolve the linear scales, for which the MG deviations with respect to $\Lambda$CDM are known to be small for most times and adopt an approximate scheme in which only the N-body part is modified and the $\Lambda$CDM solutions are used for the Lagrangian displacements. The resulting scheme has the same N-body component as in (\[fullMG\]) and the known $\Lambda$CDM LPT terms, in which the Lagrangian displacements are evolved with $D_{1,\Lambda}(a)$ and $D_{2,\Lambda}(a)$. $$\label{approxMG}
\begin{split}
\mathbf{x}(a_f) & = \mathbf{x}(a_i) + \mathbf{\upsilon}(a_c)\int_{a_i}^{a_f}\frac{da}{Q(a)} + \\
& + \left(D_{1,\Lambda}(a_f)-D_{1,\Lambda}(a_i)\right)\mathbf{s}_{1}(\mathbf{q},a_0)+ \\
& + \left(D_{2,\Lambda}(a_f)-D_{2,\Lambda}(a_i)\right)\mathbf{s}_{2}(\mathbf{q},a_0) \\
\mathbf{\upsilon}(a_f) & = \mathbf{\upsilon}(a_i) - \left(\int_{a_i}^{a_f}\frac{a}{a_cQ(a)}da\right) \times \\
& \left[ - 1.5\Omega_{m0} a_c\left(\nabla_{\mathbf{x}} \tilde{\Phi}_{N}(\mathbf{x})+\frac{\Delta \RR_c}{\RR_c}\frac{\beta}{M_{Pl}}\nabla_{\mathbf{x}} \tilde{\phi}(\mathbf{x})\right) - \right. \\
& \left. - T^2[D_{1,\Lambda}](a_c)\mathbf{s}_{1}(\mathbf{q},a_0)-T^2[D_{2,\Lambda}](a_c)\mathbf{s}_{2}(\mathbf{q},a_0)\right].
\end{split}$$ A comparison of the two approaches for the $f(R)$ model with $\abs{f_{R_0}}=10^{-4}$, for which we expect the largest modifications, is shown in Fig. \[fig2\]. One comparison tracks a given particle inside our volume during the simulation and we also compare the resulting power spectra using both schemes. We find excellent agreement between the approximate and fully modified schemes, in both the linear and mildly non-linear regimes. We also find very small differences for the position and displacement vectors (magnitude & direction) with differences in angular orientation of at most $11$ arcseconds, and differences in the magnitude of steps less than $2,5\%$, which result in power spectra that have a fractional difference no larger than 0.3% today. The approximate scheme takes under half the run time of the full implementation. In light of these results, we adopt the approximate scheme in the COLA simulations used in this analysis. This has the great advantage of not having to solve numerically for the LPT displacements at every time step, without sacrificing much accuracy.
Analysis/Results {#sec:Analysis/Results}
================
Modified gravity results {#sec:Modified gravity results}
------------------------
In this section we present the results of the assessment of COLA’s performance with respect to the predicted power spectra, redshift space distortions (RSD) and dark matter halos for the modified gravity scenarios and $\Lambda$CDM. For every given model and choice of parameters, simulations have been performed using both COLA and the PM code.
### Power spectra {#powfr}
To appropriately benchmark the COLA performance for modified gravity, we first compare the performance of COLA for $\Lambda$CDM. In Fig. \[fig3\], we show the $\Lambda$CDM power spectra as obtained by both codes, together in comparison with the fit by [@Smith:2002dz]. The two results agree well within a standard deviation of each other for all scales, but start to, underestimate power, consistently with one another, by $k\sim2$ h/Mpc, relative to higher resolution simulations. For that reason, we choose to compare performance down to a scales with $k=2.5$ h/Mpc, while the Nyquist wavenumber, for our simulation, is $k\sim4$ h/Mpc.
In Fig. \[fig4\], the fractional difference in the power spectra is plotted for all our models and both codes are found to agree with each other well within one standard deviation, with the differences being smaller than 1%. Our results demonstrate the consistency between COLA and the N-body approach using the approximate scheme. In turn this connects with previous work that has shown, in general, the good degree of consistency of this approximate scheme with N-body simulations using the full Klein Gordon for the same models in the literature [@Zhao:2010qy; @Davis:2011pj; @2012JCAP...10..002B; @Winther:2014cia]. In particular the results for the $\abs{f_{R_0}}=10^{-5}$ & $\abs{f_{R_0}}=10^{-6}$ models are in excellent agreement with the literature for all scales. Our results confirm findings in [@Winther:2014cia], in studying the effectiveness of the linearized screening schema: for the lowest screening $f(R)$ model, with $\abs{f_{R_0}}=10^{-4}$, and the symmetron model the effective screening parameterization, respectively, under and overestimates the power, relative to the full KG simulation, at the non-linear scales.
Fig. \[figlp\] shows our COLA scheme’s accuracy in predicting the fractional difference in the power spectra for the highest deviation model, $\abs{f_{R_0}}=10^{-4}$, as a function of the number of time steps used, for one realization. We find that using 50 time steps provides excellent convergence, at the level of 0.9%, to the scales we want to consider, $k\sim 2$ Mpc/h. Using 30 time steps provides convergence at the level of 8% at $k\sim 2$ Mpc/h.
### Redshift space distortions
A great amount of observational effort is being invested in studying the three-dimensional Large Scale Structure (LSS) through spectroscopic galaxy surveys that measure precise redshifts. Among various challenges faced by such measurements, the observed clustering structures appear distorted in redshift space.
Density perturbations give rise to peculiar velocities with respect to the Hubble flow, which result in the redshift space position $\mathbf{r}_s$, being different than the real space position $\mathbf{r}_r$, with the relationship between them taking the form $$\label{map}
\mathbf{r}_s = \mathbf{r}_r + \frac{\mathbf{v}\cdot\hat{\mathbf{n}}}{H_0}\hat{\mathbf{n}}.$$
By $\mathbf{v}$ we denote the peculiar velocity and by $\hat{\mathbf{n}}$ the unit vector along the line of sight. At linear scales, coherent motions of galaxies that tend to collapse within an overdense region, cause it to appear squashed in redshift space. As shown by [@Kaiser01071987], in the distant observer approximation, such an overdensity will be distorted in the redshift space: $$\label{reddel}
\delta_s (\mathbf{k},a) = \left(1 + f \mu^2 \right)\delta_r(\mathbf{k},a),$$ where $\mu$ is the angle between the peculiar velocity and the line of sight in $k$ space, $\hat{\mathbf{k}}$, and $f$ the linear growth rate, $$f_g(a) = \frac{d\ln D_{1}(a)}{d\ln a},$$ with subscript ‘$g$’ to differentiate it from the $f(R)$ function. Such an effect gives rise to, based on (\[reddel\]), an overestimation of the power spectrum measured in the redshift space: $$\label{redpow}
P_s (k, \mu, a) = \left(1 + \beta_g \mu^2 \right)^2P_r(k,a),$$ where we introduced the factor $\beta_g=f_g/b$ (not to be confused with the coupling $\beta$) to account for the galaxy bias $b$, with $b=1$ for cold dark matter. Averaging (\[redpow\]) over all directions, gives the $0^{th}$ order piece $$P_s (k,a) = \left(1 + \frac{2}{3}\beta_g + \frac{1}{5}\beta_g^2 \right)P_r(k,a).$$ At smaller, non-linear, scales the random incoherent velocities of galaxies within virialized structures cause overdense regions to appear elongated along the line of sight (“Fingers of God"), causing suppression of power. An exact quantitative treatment of the phenomenon is hard, due to the complicated nature of the small-scale velocity correlations and as a result phenomenological approaches have been proposed. Such models [@1992MNRAS.259..494P] treat the line-of-sight distortion as a radial convolution of the correlation function $\xi_{r}$ (including the Kaiser boost) with an incoherent velocity distribution $f(v)$ \_s(r\_,r\_)=\_[-]{}\^\_[r]{}(r\_,r)f(r\_-r)dr, where $r_{\perp}$ and $r_{\parallel}$ are the perpendicular and parallel components. Assuming a Gaussian velocity distribution [@1992MNRAS.259..494P], the Fourier space expression would then be \[gaussf\] P\_[s]{}(k, , a)=P\_r(k,a)(1 + \_g \^2 )\^2(-k\^2\^2\_[com]{}\^2) with $\sigma_{com}$ being the comoving distance dispersion that is related [@2013MNRAS.430.2446W] to the velocity dispersion $\sigma_{p}$ through \_[p]{}=aH(a)\_[com]{} Even though the exponential term in (\[gaussf\]) is reasonable as a damping term for capturing the non-linear power suppressions, it has been noted [@1983ApJ...267..465D] that an exponential pairwise velocity distribution f(v)=(-/\_p) is a better fit. This gives rise to the dispersion model [@Peacock:1993xg] $$\label{rsdpow}
P_s (k, \mu, a) = \left(1 + \beta_g \mu^2 \right)^2P_r(k,a)\left(\frac{1}{1 + \frac{1}{2} k^2\mu^2\sigma_{com}^2}\right),$$ in which the damping effects are incorporated through a Lorentzian term and $\sigma_{com}$ (or $\sigma_{p}$) is considered a free parameter to be fitted to the data. It should be noted that $\sigma_{p}$ is actually scale and bias dependent, which is one of the limitations the dispersion model faces [@Scoccimarro:2004tg]. The above description can still prove to be a very useful tool for obtaining an effective non-linear velocity dispersion parameter and thus quantifying the non-linear FoG effect. Integrating (\[rsdpow\]) over all directions gives the monopole piece, which can be fitted over the results to obtain $\sigma_{p}$. This is slightly different than other approaches: [@Angulo:2007fw] proposed attaching a simple factor $\frac{1}{1+k^2\sigma^2}$ to the Kaiser boost with $\sigma$ being a free parameter, loosely related to $\sigma_{p}$, while [@2012ApJ...748...78K] suggested attaching a free function $F(k,\mu)=\frac{A}{1+B k^2\mu^2}+Ck^2\mu^2$ and marginalized over the parameters $A$, $B$ and $C$ for to account for the uncertainties in constraining the effects of modified gravity on the RSD power spectrum.
[ | m[8em]{}| C[4em]{} | C[4em]{} | C[4em]{} | C[3.99em]{} C[0.01em]{} |]{} & & &\
Scenario & ${P_{rsd}/P_{real}}$ & $f_g$ & $\gamma_{eff}$ & $ \sigma_p (km/s)$ &\
\[1ex\] $\Lambda$CDM & 1.35 & 0.462 & 0.556 & 567 &\
\[1ex\] $f(R)$, $\abs{f_{R_0}}=10^{-6}$ &1.35 & 0.463 & 0.555 & 605&\
\[1ex\] $f(R)$, $\abs{f_{R_0}}=10^{-5}$ & 1.38 & 0.491& 0.512& 714&\
\[1ex\] $f(R)$, $\abs{f_{R_0}}=10^{-4}$ &1.42 & 0.541 & 0.443 & 834&\
\[1ex\] Symmetron &1.35 & 0.464 & 0.554 & 611&\
\[1ex\]
Through the mapping (\[map\]), we obtained redshift space power spectra for all of the simulated models, with the results presented in Fig.\[fig5\]. We compared the large scale results to analytic predictions arising from the linear growth rate, and also used the monopole model in (\[rsdpow\]) to obtain an effective velocity dispersion damping factor for the FoG effect. The results are summarized in Fig. \[fig5\] and Table \[tab1\].
We first benchmarked COLA’s performance for $\Lambda$CDM. We see that the PM and COLA codes’ RSD predictions for $\Lambda$CDM do not differ by more than 0.5% at all the scales of interest and agree remarkably well with the analytical prediction, with expected values of $f_g(a=1)$=0.467 and $\frac{P_s(k)}{P_r(k)}$=1.354, assuming $f_g=\Omega_m(a)^{\gamma}$, with $\gamma=0.55$. At smaller scales , the “Fingers of God” effect quickly dominates, and causes power suppression and find this suppression is well modeled by (\[rsdpow\]) with $\sigma_{p}=567\ km/sec$.
For the MG models, the additional fifth forces cause the redshift space power spectra to have, in principle, different shapes. In large scales, the enhanced clustering results in higher coherent velocities of collapse into overdense regions which translates to a higher boost in the RSD power spectra with respect to GR, translating into higher values for the growth rate and a lower $\gamma$. For lower magnitude modifications, the suppression of the fifth forces gives results that tend to the $\Lambda$CDM prediction. At smaller scales, the fifth forces cause higher random velocity dispersions inside virialized structures, making the damping effects stronger in MG.
These combined effects cause the redshift space distortions to be more pronounced in MG compared to GR. This can be clearly seen in the upper left panel in Fig.(\[fig5\]). As expected, the redshift space distortions vary from the most pronounced, in the lowest screening model, to very small deviations from GR in the strong screening regime. For the same models, the redshift space power spectra from the PM code agree with COLA well within a standard deviation. The results using the approximate schema are in good agreement with full non-linear MG N-body simulations for redshift space distortions in $f(R)$ gravity performed by [@2012MNRAS.425.2128J]. For all the models, COLA predicts deviations that are 0.5% more pronounced (higher in large scales, smaller in small scales) than the PM code.
### Halo Mass Function
To determine the halo mass function we identify halos in the simulations using the Rockstar halo finder [@2013ApJ...762..109B] for all models. In Fig. \[fig6\] we show the comparison of the halo mass function predicted by COLA and the PM code, together with a high accuracy result by [@Murray:2013qza]. COLA and PM are found to be in a better than 2.5% agreement in the lower and intermediate mass range, while in the highest mass bins there is a maximum difference of 10%.
In Fig. \[fig7\], we plot the fractional difference in the halo mass function with respect to $\Lambda$CDM, for all of our models. The COLA and PM code results agree in general, well within the standard deviation from the averaged suite of simulations. In particular, In the $\abs{f_{R_0}}=10^{-4}$ and $\abs{f_{R_0}}=10^{-5}$ models, the PM code predicts a fractional boost in the halo mass function that is higher than COLA’s by $<$ 2% and 2.5%, for the lower and intermediate bins, while in the highest bin COLA gives a boost larger by 5% and 3% correspondingly. For the $\abs{f_{R_0}}=10^{-6}$ and symmetron models, the differences are 1% and smaller, with the PM code giving greater number counts for the two mass bins below $10^{14}M_{\odot}/h$ and COLA being higher for the bin over $10^{14}M_{\odot}/h$. The differences between the predictions in each case and especially in the high mass bin, are within, and likely largely resulting from, the differences observed in the $\Lambda$CDM benchmarking of the mass functions.
While we do not perform a simulation with the full non-linear Klein-Gordon equation, we note that compared to other full KG treatments in the literature [@Winther:2014cia; @Zhao:2010qy], our method performs well and only slightly underestimates the mass function for the $\abs{f_{R_0}}=10^{-4}$ & $\abs{f_{R_0}}=10^{-5}$ models, in accordance with the general features noticed in the power spectra discussion. In agreement with [@Winther:2014cia], we observe an underestimation of halos in the lower end of our mass range (around M $\sim10^{13}{M_{\odot}/h}$) for the $\abs{f_{R_0}}=10^{-6}$ model, indicating too much screening, and an overestimation of the mass function for the symmetron model.
Conclusions {#sec:conclusions}
============
In this paper, we have implemented a hybrid scheme, that combines Lagrangian Perturbation Theory and N-body approaches, to numerically characterize the evolution of large scale structure in chameleon and symmetron, modified gravity, theories which exhibit gravitational screening in the non-linear regime. LPT is used to evolve linear scales analytically in combination with a full N-body approach that is used for the non-linear scales to reduce computational costs. An effective screening scheme is implemented in place of a solution to the full Klein-Gordon equation for the fifth potential, in which an effective suppression factor is attached to the real-space linearized perturbations.
We demonstrate that while in MG spatial modes evolve differently in LPT (and can have deviations from the nominal GR geodesic paths), the scheme can be further simplified, for the models we studied, by using a displacement coordinate system based on scale-independent $\Lambda$CDM growing modes combined with a modified, screened Poisson equation. We note that, while this approximate scheme works well for the chameleon and symmetron models we consider, it should always be tested against the exact LPT solution for a new modified gravity model.
Our method was applied on the $f(R)$ and symmetron models and it was tested against power spectra, redshift space distortions and dark matter halo mass functions, using a fiducial number of 50 time steps. At the same time, we assessed our hybrid’s performance against simulations from a pure N-body code with the same screening implementation for the same models, using 500 iterations.
With regards to power spectra, we found COLA to be in better than 1% agreement with the N-body code at all scales for all the models studied. Note that the effective screening scheme we use has previously been shown to be in good agreement with results using the full non-linear Klein Gordon in an N-body implementation [@Winther:2014cia]. We find, as was discussed in [@Winther:2014cia], that the effective screening approach does underestimate power, relative to that found in solving the full Klein-Gordon [@Zhao:2010qy], as one moves into the fully non-linear regime ($k>\sim 2Mpc/h$), however this is also beyond the regime of applicability of COLA’s scheme.
COLA and the N-body code are in better than 0.5% agreement with respect to redshift space distortions for all the scales and models of interest. The distortions were modeled by attaching the linear Kaiser factor for the enhancement at large scales and a Lorentzian dispersion factor for the small scale suppression due to incoherent motions within virialized structures. We find that the monopole is a well fit using an effective pairwise velocity dispersion as a fitting parameter to quantify the suppressions at non-linear scales. The additional fifth forces present in the chameleon and symmetron models, cause the redshift space distortions to be more pronounced with respect to $\Lambda$CDM. This can be seen by the larger boosts in linear scales due to the higher coherent velocities, and by the stronger suppressions in the non-linear scales because of the higher values of the velocity dispersion. The adapted COLA scheme gives reasonable results for the predicted fractional boost in the halo mass function relative to $\Lambda$CDM, with the differences between the N-body and COLA results in the halo mass function estimation most likely being due to the difference between the two codes in $\Lambda$CDM.
In this paper, we have focused on chameleon and symmetron-type scalar-tensor theories, but it would be very interesting to see how well this scheme performs for the simulation of other screening mechanisms as well such as the Vainshtein mechanism [@VAINSHTEIN1972393], as well as other dark energy models, such as those with non-minimal couplings between dark matter and a quintessence scalar field [@Li:2010re]. Given the level of consistency between COLA and the N-body predictions for the monopole of the redshift power spectrum, it would also be interesting to investigate the COLA scheme’s ability to capture higher order moments of the angular power spectrum to, for example, calculate the ratio of the quadrupole to monopole moments to estimate $\beta_g$ in a way that is robust to systematic effects from incomplete modeling of the nonlinear distortions [@Landy:2002eq; @2012MNRAS.425.2128J].
Many theories being considered as explanations for cosmic acceleration have tantalizing predictions in the non-linear regime but also present computational challenges in modeling them. With a suite of next-generation large scale structure surveys, including LSST, DESI, Euclid and WFIRST, starting in next few years, there is an unprecedented opportunity to measure the properties of large scale structure clustering as it transitions from linear to mildly and then strongly non-linear scales, and using multiple tracers. The results presented here demonstrate that COLA, proposed to enable accurate and efficient, non-linear predictions for $\Lambda$CDM, is a viable approach to study non-linear collapse for a broader portfolio of cosmological scenarios. For example, in work that has followed our paper in [@Winther:2017jof], the effectiveness of the COLA approach has also been studied in the $f(R)$ and nDGP models, and was shown to perform very well in predicting the fractional deviations with respect to the $\Lambda$CDM power spectra and halo mass functions, using a small number of time steps.
Acknowledgments {#acknowledgments .unnumbered}
===============
We want to thank Hans Winther and Pedro Ferreira for useful comments on our paper and Risa Wechsler for helpful discussions on halo mass function performance with the COLA algorithm. We would also like to thank an anonymous referee, whose valuable comments helped improve and clarify this manuscript. The work of GV and RB is supported by NASA ATP grant NNX14AH53G, NASA ROSES grant 12-EUCLID12- 0004 and DoE grant DE-DE-SC0011838.
\
Lagrangian perturbation theory in Modified Gravity {#App:AppendixA}
==================================================
LPT [@Bouchet:1994xp] works perturbatively in a displacement field $\mathbf{s}(\mathbf{q},a)$ $$\mathbf{x}=\mathbf{q}+\mathbf{s}_{1}(\mathbf{q},a)+\mathbf{s}_{2}(\mathbf{q},a)+...,$$ where $\mathbf{q}$ and $\mathbf{x}$ are the initial and final comoving Eulerian particle positions. In this formulation, all the information is reflected in the mapping through the displacement field. Working up to first order gives the so-called Zel’dovich approximation in $\Lambda$CDM, for which the $\mathbf{s}(\mathbf{q},a)$ can be decomposed into a product of temporal and spatial factors $$\nabla_{\mathbf{q}}\mathbf{s}_{1}(\mathbf{q},a) = D_{1}(a)\nabla_{\mathbf{q}}\mathbf{s}_{1}(\mathbf{q},a_0)$$ and $$\label{gauss}
\nabla_{\mathbf{q}}\mathbf{s}(\mathbf{q},a_0) = -\delta(\mathbf{q},a_0)$$ with $\delta(\mathbf{q},a_0)$ being the Gaussian density field generated by an initial linear power spectrum and $D_{1}(a)$ the scale independent first order growth factor, given by $$\ddot{D}_{1} + 2H\dot{D}_{1} = \frac{3}{2} \Omega_m(a)H^2 D_{1}$$ In an MG scenario, the growth factor is not scale independent any more and $$\label{firstdisp}
\nabla_{\mathbf{q}}\mathbf{s}_{1} = D_{1}(\mathbf{q},a)\nabla_{\mathbf{q}}\mathbf{s}_1(\mathbf{q},a_0)$$ where $$\ddot{D}_{1}(\mathbf{k},a) + 2H\dot{D}_{1}(\mathbf{k},a) = \frac{3}{2} \Omega_m(a)H^2 D_{1}(\mathbf{k},a)\frac{G_{eff}}{G}$$ in Fourier space. This implies that particle trajectories, unlike in $\Lambda$CDM, are not straight lines [@Valkenburg:2015dsa]. (\[gauss\]) and (\[firstdisp\]) indeed give $$\label{fullZel}
\mathbf{s}_{1} = -D_{1}(\mathbf{q},a)\frac{\nabla_{\mathbf{q}}}{\nabla^2_{\mathbf{q}}}\delta(\mathbf{q},a_0)-\delta(\mathbf{q},a_0)\frac{\nabla_{\mathbf{q}}}{\nabla^2_{\mathbf{q}}}D_{1}(\mathbf{q},a)$$ The second term, responsible for the trajectory bending, vanishes when the growing mode is scale independent, in which case one recovers the standard $\Lambda$CDM Zel’dovich approximation. When working up to second order (2LPT), we have in a similar fashion $$\nabla_{\mathbf{q}}\mathbf{s}_{2}(\mathbf{q},a) = D_{2}(a)\nabla_{\mathbf{q}}\mathbf{s}_{2}(\mathbf{q},a_0),$$ where the second order growth factor is given by $$\ddot{D}_{2}(a) + 2H\dot{D}_{2}(a) = \frac{3}{2} \Omega_m(a)H^2 D_{2}(a)\left(1 - D_1^2(a)\right)$$ For the early times, the spatial part is given by $$\label{spatial}
\nabla_{\mathbf{q}}\mathbf{s}_{2}(\mathbf{q},a_0) = \frac{1}{2}\sum_{i\ne j}\left(\mathbf{s}_{1i,i}\mathbf{s}_{1j,j}-\mathbf{s}_{1i,j}\mathbf{s}_{1j,i}\right).$$ In the MG case, we will have again $$\label{seconddisp}
\nabla_{\mathbf{q}}\mathbf{s}_{2}(\mathbf{q},a) = D_{2}(\mathbf{q},a)\nabla_{\mathbf{q}}\mathbf{s}_{2}(\mathbf{q},a_0),$$ with the scale dependent second order growth factor that obeys $$\begin{split}
\ddot{D}_{2}(\mathbf{k},a) + 2H\dot{D}_{2}(\mathbf{k},a) & = \frac{3}{2} \Omega_m(a)H^2 D_{2}(\mathbf{k},a)\times \\
& (1 - D_1^2(\mathbf{k},a))\frac{G_{eff}}{G},
\end{split}$$ in the Fourier space. The fact that all of our models recover GR at early times, guarantees that the early time spatial part is still given by (\[spatial\]). In our implementation of the full MG COLA scheme, a suitably modified version of 2LPTic produces the LPT terms at every time step, through the Fourier space versions of (\[firstdisp\]) and (\[seconddisp\]) $$\label{fourdisps}
\begin{split}
\mathbf{s}_{1}(\mathbf{k},a) & = \frac{i \mathbf{k}}{k^2}D_{1}(\mathbf{k},a)\delta(\mathbf{k},a_0) \\
\mathbf{s}_{2}(\mathbf{k},a) & = -\frac{i \mathbf{k}}{k^2}D_{2}(\mathbf{k},a) \frac{1}{2}\sum_{i\ne j}\left(\mathbf{s}_{1i,i}\mathbf{s}_{1j,j}-\mathbf{s}_{1i,j}\mathbf{s}_{1j,i}\right),
\end{split}$$ and also the same for the accelerations $$\label{fouraccels}
\begin{split}
T^2[\mathbf{s}_{1}(\mathbf{k},a)] & = \frac{i \mathbf{k}}{k^2}T^2[D_{1}(\mathbf{k},a)]\delta(\mathbf{k},a_0) \\
T^2[\mathbf{s}_{2}(\mathbf{k},a)] & = -\frac{i \mathbf{k}}{k^2}T^2[D_{2}(\mathbf{k},a)] \frac{1}{2}\sum_{i\ne j}\left(\mathbf{s}_{1i,i}\mathbf{s}_{1j,j}-\mathbf{s}_{1i,j}\mathbf{s}_{1j,i}\right).
\end{split}$$
[^1]: http://sumire.ipmu.jp/en/
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper describes the systems submitted by the department of electronic engineering, institute of microelectronics of Tsinghua university and TsingMicro Co. Ltd. (THUEE) to the NIST 2019 speaker recognition evaluation CTS challenge. Six subsystems, including etdnn/ams, ftdnn/as, eftdnn/ams, resnet, multitask and c-vector are developed in this evaluation.'
bibliography:
- 'reference\_for\_sre19.bib'
title: THUEE system description for NIST 2019 SRE CTS Challenge
---
NIST 2019 SRE CTS challenge, eftdnn, multitask, c-vector, additive margin
Introduction {#sec:intro}
============
This paper describes the systems developed by the department of electronic engineering, institute of microelectronics of Tsinghua university and TsingMicro Co. Ltd. (THUEE) for the NIST 2019 speaker recognition evaluation (SRE) CTS challenge [@sre19]. Six subsystems, including etdnn/ams, ftdnn/as, eftdnn/ams, resnet, multitask and c-vector are developed in this evaluation. All the subsystems consists of a deep neural network followed by dimension deduction, score normalization and calibration. For each system, we begin with a summary of the data usage, followed by a description of the system setup along with their hyperparameters. Finally, we report experimental results obtained by each subsystem and fusion system on the SRE18 development and SRE18 evaluation datasets.
Data Usage
==========
For the sake of clarity, the datasets notations are defined as in table 1 and the training data for the six subsystems are list in table 2, 3, and 4.
notation datasets
-------------- -------------------------------
SRE SRE04/05/06/08/10/MIXER6
LDC98S75/LDC99S79/LDC2002S06/
LDC2001S13/LDC2004S07
Voxceleb Voxceleb 1/2
Fisher+SWB I Fisher + Switchboard I
CH+CF Callhome+Callfriend
: Datasets Notations
Components Data usage
---------------- ------------------
Neural Network SRE+SWB+Voxceleb
LDA/PLDA SRE+SRE16+SRE18
PLDA-adapt SRE+SRE16+SRE18
asnorm SRE18 unlabel
: Data usage for etdnn/ams, ftdnn/as, and resnet subsystems
Components Data usage
---------------- -------------------------------
GMM-HMM Fisher+SWB I
Neural Network SRE+SWB+Voxceleb+Fisher+SWB I
LDA/PLDA SRE+SRE16+SRE18
PLDA-adapt SRE16+SRE18+MIXER6+CH+CF
asnorm SRE18 unlabel
: Data usage for multitask and c-vector subsystems
Components Data usage
---------------- ------------------------
Neural Network SRE+SWB+Voxceleb+CH+CF
LDA/PLDA SRE+SRE16+SRE18 eval
PLDA-adapt SRE+SRE16+SRE18 eval
asnorm SRE18 unlabel
: Data usage for eftdnn subsystem
Systems
=======
Etdnn/ams
---------
Etdnn/ams system is an extended version of tdnn with the additive margin softmax loss [@AMsoftmax19]. Etdnn is used in speaker verification in [@JHUMITsre18]. Compared with the traditional tdnn in [@Snyder2017Deep], it has wider context and interleaving dense layers between each two tdnn layers. The architecture of our etdnn network is shown in table \[tab:etdnn\]. It is the same as the etdnn architecture in [@JHUMITsre18], except that the context of layer 5 of our system is t-3:t+3 instead of t-3, t, t+3. The x-vector is extracted from layer 12 prior to the ReLU non-linearity. For the loss, we use additive margin softmax with $m=0.15$ instead of traditional softmax loss or angular softmax loss. Additive margin softmax is proposed in [@AMsoftmax18] and then used in speaker verification in our paper [@AMsoftmax19]. It is easier to train and generally performs better than angular softmax.
Layer Layer Type Context Size
------- ----------------------- ----------- ---------------
1 TDNN-ReLU t-2:t+2 512
2 Dense-ReLU t 512
3 TDNN-ReLU t-2,t,t+2 512
4 Dense-ReLU t 512
5 TDNN-ReLU t-3:t+3 512
6 Dense-ReLU t 512
7 TDNN-ReLU t-4,t,t+4 512
8 Dense-ReLU t 512
9 Dense-ReLU t 512
10 Dense-ReLU t 1500
11 Pooling(mean+stddev) Full-seq 2$\times$1500
12 Dense(Embedding)-ReLU 512.
13 Dense-ReLU 512.
14 Dense-Softmax Num. spks.
: Etdnn architecture[]{data-label="tab:etdnn"}
ftdnn/as
--------
Factorized TDNN (ftdnn) architecture is listed in table \[tab:ftdnn\]. It is the same to [@JHUMITsre18] except that we use 1024 nodes instead of 512 nodes in layer 12 and 13. The x-vector is extracted from layer 12 prior to the ReLU non-linearity. So our x-vector is 1024 dimensional. More details about the architecture can be found in [@JHUMITsre18].
---- --------- ---------- ---------- ------- ---------- -------
Layer Context Context conn. Inner
Type factor 1 factor 2 from size
1 TDNN t-2:t+2 512
2 F-TDNN t-2,t t, t+2 1024 256
3 F-TDNN t t 1024 256
4 F-TDNN t-3, t t, t+3 1024 256
5 F-TDNN t t 3 1024 256
6 F-TDNN t-3, t t, t+3 1024 256
7 F-TDNN t-3, t t, t+3 2,4 1024 256
8 F-TDNN t-3, t t, t+3 1024 256
9 F-TDNN t-3, t t, t+3 4,6,8 1024 256
10 Dense t t 2048
11 Pooling full-seq 4096
12 Dense 1024
13 Dense 1024
Dense- N. spks.
Softmax
---- --------- ---------- ---------- ------- ---------- -------
: ftdnn architecture[]{data-label="tab:ftdnn"}
eftdnn/ams
----------
Extended ftdnn (eftdnn) is a combination of etdnn and ftdnn. Its architecture is listed in table \[tab:eftdnn\]. The x-vector is extracted from layer 22 prior to the ReLU non-linearity.
---- --------- ---------- ---------- ---------- --------- ---------- -------
Layer Context Context Context conn. Inner
Type factor 1 factor 2 factor 3 from size
1 TDNN t-2:t+2 512
2 Dense 512
3 F-TDNN t-3,t -1 t-1, t+1 t+1, t+3 1024 256
4 Dense 1024
5 F-TDNN t t t 1024 256
6 Dense 1024
7 F-TDNN t-5, t-2 t-2, t+1 t+1,t+4 1024 256
8 Dense 1024
9 F-TDNN t t t 5 1024 256
10 Dense 1024
11 F-TDNN t-5, t-2 t-2, t+1 t+1,t+4 1024 256
12 Dense 1024
13 F-TDNN t-5, t-2 t-2,t+1 t+1, t+4 3,7 1024 256
14 Dense 1024
15 F-TDNN t-5, t-2 t-2, t+1 t+1,t+4 1024 256
16 Dense 1024
17 F-TDNN t t t 7,11,15 1024 256
18 Dense t 2048
19 Dense t 2048
20 Dense t 2048
21 Pooling full-seq 4096
22 Dense 1024
23 Dense 1024
Dense- N. spks.
Softmax
---- --------- ---------- ---------- ---------- --------- ---------- -------
resnet
------
ResNet architecture is also based on tdnn x-vector [@Snyder2017Deep]. The five frame level tdnn layers in [@Snyder2017Deep] are replaced by ResNet34 (512 nodes) + DNN(512 nodes) + DNN(1000 nodes). Further details about ResNet34 can be found in [@ResNet16]. In our realization, acoustic features are regarded as a single channel picture and feed into the ResNet34. If the dimensions in the residual network don’t match, zeros are added. The statistic pooling and segment level network stay the same. For the loss function, we use angular softmax with $m=4$. The x-vector is extracted from first DNN layer in segment level prior to the ReLU non-linearity. It has 512 dimensions.
multitask {#sec:multitask}
---------
Multitask architecture is proposed in [@cvec2018]. It is a hybrid multi-task learning based on x-vector network and ASR network. It aims to introduce phonetic information by another neural acoustic model in ASR to help speaker recognition task. The architecture is shown in Fig. \[fig:multitask\].
![multitask architecture for the speaker embedding extraction.[]{data-label="fig:multitask"}](multitask){width="10cm"}
The frame-level part of the x-vector network is a 10-layer TDNN. The input of each layer is the sliced output of the previous layer. The slicing parameter is: {t - 2; t - 1; t; t + 1; t + 2}, { t }, { t - 2; t; t + 2 }, {t}, { t - 3; t; t + 3 }, {t }, {t - 4; t; t + 4 }, { t }, { t } , { t }. It has 512 nodes in layer 1 to 9, and the 10-th layer has 1500 nodes. The segment-level part of x-vector network is a 2-layer fully-connected network with 512 nodes per layer. The output is predicted by softmax and the size is the same as the number of speakers.
The ASR network has no statistics pooling component. The frame-level part of the x-vector network is a 7-layer TDNN. The input of each layer is the sliced output of the previous layer. The slicing parameter is: {t - 2; t - 1; t; t + 1; t + 2}, {t - 2; t; t + 2}, {t - 3; t; t + 3}, {t}, {t}, {t}, {t}. It has 512 nodes in layer 1 to 7.
Only the first TDNN layer of the x-vector network is shared with the ASR network. The phonetic classification is done at the frame level, while the speaker labels are classified at the segment level.
To train the multitask network, we need training data with speaker and ASR transcribed. But only Phonetic dataset fits this condition and the data amount is too small to train a neural network. So, we need to train a GMM-HMM speech recognition system to do phonetic alignment for other datasets. The GMM-HMM is trained using Phonetic dataset with features of 20-dimensional MFCCs with delta and delta-delta, totally 60-dimensional. The total number of senones is 3800. After training, forced alignment is applied to the SRE, Switchboard, and Voxceleb datasets using a fMLLR-SAT system.
c-vector
--------
![multitask architecture for the speaker embedding extraction.[]{data-label="fig:cvector"}](cvector){width="10cm"}
C-vector architecture is also one of our proposed systems in paper [@cvec2019]. As shown in Fig. \[fig:cvector\], it is an extension of multitask architecture. It combines multitask architecture with an extra ASR Acoustic Model. The output of ASR Acoustic Model is concatenated with x-vector’s frame-level output as the input of statistics pooling. Refer to [@cvec2019] for more details.
The multitask part of c-vector has the same architecture as in the above section \[sec:multitask\] ASR Acoustic Model of c-vector is a 5-layer TDNN network. The slicing parameter is { t - 2; t - 1; t; t + 1; t + 2 }, { t - 1; t; t + 1 }, { t - 1; t; t + 1 }, { t - 3; t; t + 3}, { t - 6; t - 3; t}. The 5-th layer is the BN layer containing 128 nodes and other layers have 650 nodes.
A GMM-HMM is also trained as like in section \[sec:multitask\] to do phonetic alignment for training datasets.
feature and back-end
====================
23-dimensional MFCC (20-3700Hz) is extracted as feature for etdnn/ams, ftdnn/as, eftdnn/ams, multitask and c-vector subsystems. 23-dimensional Fbank is used as feature for ResNet 16kHz subsystems. A simple energy-based VAD is used based on the C0 component of the MFCC feature [@kaldi2011].
For each neural network, its training data are augmented using the public accessible MUSAN and RIRS\_NOISES as the noise source. Two-fold data augmentation is applied for etdnn/ams, ftdnn/as, resnet, multitask and cvector subsystems. For eftdnn/ams subsystem, five-fold data augmentation is applied.
After the embeddings are extracted, they are then transformed to 150 dimension using LDA. Then, embeddings are projected into unit sphere. At last, adapted PLDA with no dimension reduction is applied.
The execution time is test on Intel Xeon E5-2680 v4. Extracting x-vector cost about 0.087RT. Single trial cost around 0.09RT. The memory cost about 1G for a x-vector extraction and a single trial. In the inference, only CPU is used.
The speed test was performed on Intel Xeon E5-2680 v4 for etdnn\_ams, multitask, c-vector and ResNet system. Test on Intel Xeon Platinum 8168 for ftdnn and eftdnn system. Extracting embedding cost about 0.103RT for etdnn\_ams, 0.089RT for multitask, 0.092RT for c-vector, 0.132RT for eftdnn, 0.0639RT for ftdnn, and 0.112RT for ResNet. Single trial cost around 1.2ms for etdnn\_ams, 0.9ms for multitask, 0.9ms for c-vector, 0.059s for eftdnn, 0.0288s for ftdnn, 1.0ms for ResNet. The memory cost about 1G for an embedding extraction and a single trial. In the inference, we just use CPU.
Fusion
======
Our primary system is the linear fusion of all the above six subsystems by BOSARIS Toolkit on SRE19 dev and eval [@BOSARIS]. Before the fusion, each score is calibrated by PAV method (*[pav\_calibrate\_scores]{}) on our development database. It is evaluated by the primary metric provided by NIST SRE 2019.*
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the nonlinear scattering theory for quantum systems with strong Seebeck and Peltier effects, and consider their use as heat-engines and refrigerators with finite power outputs. This article gives detailed derivations of the results summarized in Phys. Rev. Lett. [**112**]{}, 130601 (2014). It shows how to use the scattering theory to find (i) the quantum thermoelectric with maximum possible power output, and (ii) the quantum thermoelectric with maximum efficiency at given power output. The latter corresponds to a minimal entropy production at that power output. These quantities are of quantum origin since they depend on system size over electronic wavelength, and so have no analogue in classical thermodynamics. The maximal efficiency coincides with Carnot efficiency at zero power output, but decreases with increasing power output. This gives a fundamental lower bound on entropy production, which means that reversibility (in the thermodynamic sense) is impossible for finite power output. The suppression of efficiency by (nonlinear) phonon and photon effects is addressed in detail; when these effects are strong, maximum efficiency coincides with maximum power. Finally, we show in particular limits (typically without magnetic fields) that relaxation within the quantum system does not allow the system to exceed the bounds derived for relaxation-free systems, however, a general proof of this remains elusive.'
author:
- 'Robert S. Whitney'
date: 'March 16, 2015'
title: |
Finding the quantum thermoelectric with maximal efficiency\
and minimal entropy production at given power output
---
Introduction
============
Thermoelectric effects in nanostructures [@Pekola-reviews; @Casati-review; @Sothmann-Sanchez-Jordan-review; @Haupt-review] and molecules [@Paulsson-Datta2003; @Reddy2007] are of great current interest. They might enable efficient electricity generation and refrigeration [@books; @DiSalvo-review; @Shakouri-reviews], and could also lead to new types of sub-Kelvin refrigeration, cooling electrons in solid-state samples to lower temperatures than with conventional cryostats [@Pekola-reviews], or cooling fermionic atomic gases [@Grenier2012; @Brantut-Grenier-et-al2013; @Grenier2014]. However, they are also extremely interesting at the level of fundamental physics, since they allow one to construct the simplest possible quantum machine that converts heat flows into useful work (electrical power in this case) or vice versa. This makes them an ideal case study for [*quantum thermodynamics*]{}, i.e. the thermodynamics of quantum systems [@QuantumThermodyn-book].
![\[Fig:thermocouple\] (a) The simplest heat-engine is a thermocouple circuit made of two thermoelectrics (filled and open circles). The filled and open circles are quantum systems with opposite thermoelectric responses, an example could be that in (b). For a heat-engine, we assume $T_L > T_R$, so heat flows as shown, generating a current $I$, which provides power to a load (battery charger, motor, etc.) that converts the electrical power into some other form of work. The same thermocouple circuit can act as a refrigerator; if one replaces the load with a power supply that generates the current $I$. This induces the heat flow out of Reservoir $L$, which thereby refrigerates Reservoir $L$, so $T_L < T_R$. Note that in both cases the circuit works because the two thermoelectrics are electrically in series but thermally in parallel. In (b), $N$ indicates the number of transverse modes in the narrowest part of the quantum system. ](figure1.pdf){width="\columnwidth"}
The simplest heat-engine is a thermocouple circuit, as shown in Fig. \[Fig:thermocouple\]. It consists of a pair of thermoelectrics with opposite thermoelectric responses (filled and open circles) and a load, connected in a ring. Between each such circuit element is a big reservoir of electrons, the reservoir on the left ($L$) is hotter than the others, $T_L > T_R$, so heat flows from left to right. One thermoelectric’s response causes an electric current to flow in the opposite direction to the heat flow (filled circle), while the other’s causes an electric current to flow in the same direction as the heat flow (open circle). Thus, the two thermoelectrics turn heat energy into electrical work; a current flow $I$ through the load. The load is assumed to be a device that turns the electrical work into some other form of work; it could be a battery-charger (turning electrical work into chemical work) or a motor (turning electrical work into mechanical work).
The same thermocouple circuit can be made into a refrigerator simply by replacing the load with a power supply. The power supply does work to establish the current $I$ around the circuit, and this current through the thermoelectrics can “drag” heat out of reservoir $L$. In other words, the electrical current and heat flow are the same as for the heat-engine, but now the former causes the latter rather than vice versa. Thus, the refrigerator cools reservoir $L$, so $T_L < T_R$.
The laws of classical thermodynamics inform us that entropy production can never be negative, and maximal efficiency occurs when a system operates reversibly (zero entropy production). Thus, it places fundamental bounds on heat-engine and refrigerator efficiencies, known as Carnot efficiencies. In both cases, the efficiency is defined as the power output divided by the power input. For the heat-engine, the power input is the heat current out of the hotter reservoir (reservoir $L$), $J_L$, and the power output is the electrical power generated $P_{\rm gen}$. Thus, the heat-engine (eng) efficiency is $$\begin{aligned}
\eta_{\rm eng} = P_{\rm gen}\big/ J_L.
\label{Eq:eff-eng}\end{aligned}$$ This efficiency can never exceed Carnot’s limit, $$\begin{aligned}
\eta_{\rm eng}^{\rm Carnot} &=& 1-T_R/ T_L,
\label{Eq:Carnot-eng}\end{aligned}$$ where we recall that we have $T_L > T_R$.
For the refrigerator the situation is reversed, the load is replaced by a power supply, and the power input is the electrical power that the circuit absorbs from the power supply, $P_{\rm abs}$. The power output is the heat current out of the colder reservoir (reservoir $L$), $J_L$. This is called the cooling power, because it is the rate at which the circuit removes heat energy from reservoir $L$. Thus, the refrigerator (fri) efficiency is $$\begin{aligned}
\eta_{\rm fri} = J_L \big/ P_{\rm abs}.
\label{Eq:eff-fri}\end{aligned}$$ This efficiency is often called the coefficient of performance or COP. This efficiency can never exceed Carnot’s limit, $$\begin{aligned}
\eta_{\rm fri}^{\rm Carnot} &=& (T_R/T_L -1)^{-1}, \ \
\label{Eq:Carnot-fri}\end{aligned}$$ where we recall that $T_L < T_R$ (opposite of heat-engine).
Strangely, the laws of classical thermodynamics do not appear to place a fundamental bound on the power output associated with reversible (Carnot efficient) operation. Most textbooks say that reversibility requires “small” power output, but rarely define what “small” means. The central objective of Ref. \[\] was to find the meaning of “small”, and find a fundamental upper bound on the efficiency of an irreversible system in which the power output was [*not*]{} small.
Ref. \[\] did this for the class of quantum thermoelectrics that are well modelled by a scattering theory, which enables one to straightforwardly treat quantum and thermodynamic effects on an equal footing. It summarized two principal results absent from classical thermodynamics. Firstly, there is a quantum bound (qb) on the power output, and no quantum system can exceed this bound (open circles in Fig. \[Fig:summary\]). Secondly, there is a upper bound on the efficiency at any given power output less than this bound (thick black curves in Fig. \[Fig:summary\]). The efficiency at given power output can only reach Carnot efficiency when the power output is very small compared to the quantum bound on power output. The upper bound on efficiency then decays monotonically as one increases the power output towards the quantum bound. The objective of this article is to explain in detail the methods used to derive these results, along with the other results that were summarized in Ref. \[\].
![\[Fig:summary\] The thick black curves are qualitative sketches of the maximum efficiency as a function of heat-engine power output (main plot), or refrigerator cooling power (inset), with the shaded regions being forbidden. Precise plot of such curves for different temperature ratios, $T_R/T_L$, are shown in Fig. \[Fig:allpowers\]. The colored loops (red, grey and blue) are typical sketches of the efficiency versus power of [*individual*]{} heat-engines as we increase the load resistance (direction of arrows on loop). The power output $P_{\rm gen}=IV$ vanishes when the load resistance is zero (for which $V=0$) or infinite (for which $I=0$), with a maximum at an intermediate resistance (open square). The curves have a characteristic loop form [@Casati-review], however the exact shape of the loop depends on many system specific details, such as charging effects. The dashed blue loop is for a typical non-optimal system (always well below the upper bound), while the solid red and grey loops are for systems which achieve the upper bound for a particular value of the load. The star marks the Curzon-Ahlborn efficiency. ](figure2.pdf){width="\columnwidth"}
Contents of this article
------------------------
This article provides detailed derivations of the results in Ref. \[\]. The first part of this article is an extended introduction. Section \[Sect:literature\] is a short review of the relevant literature. Section \[Sect:Unique\] discusses how we define temperature, heat and entropy. Section \[Sect:entropy-prod\] recalls the connection between efficiency and entropy production in any thermodynamic machine. Section \[Sect:ScatteringTheory\] reviews the nonlinear scattering theory, which section \[Sect:over-estimates\] uses to make very simple over-estimates of a quantum system’s maximum power output.
The second part of this article considers how to optimize a system which is free of relaxation and has no phonons or photons. Section \[Sect:guess-heat\] gives a hand-waving explanation of the optimal heat engine, while Section \[Sect:eng\] gives the full derivation. Section \[Sect:guess-fri\] gives a hand-waving explanation of the optimal refrigerator, while Section \[Sect:fri\] gives the full derivation. Section \[Sect:chain\] proposes a system which could in principle come arbitrarily close to the optimal properties given in sections \[Sect:eng\] and \[Sect:fri\]. Section \[Sect:in-parallel\] considers many quantum thermoelectrics in parallel.
The third part of this article considers certain effects neglected in the above idealized system. Section \[Sect:ph\] adds the parasitic effect of phonon or photon carrying heat in parallel to the electrons. Section \[Sect:Relax\] treats relaxation within the quantum system.
Comments on existing literature {#Sect:literature}
===============================
There is much interest in using thermoelectric effects to cool fermionic atomic gases [@Grenier2012; @Brantut-Grenier-et-al2013; @Grenier2014], which are hard to cool via other methods. This physics is extremely similar to that in this work, but there is a crucial difference. For the electronic systems that we consider, we can assume the temperatures to be much less than the reservoir’s Fermi energy, and so take all electrons to have the same Fermi wavelength. In contrast, fermionic atomic gases have temperatures of order the Fermi energy, so the high-energy particles in a reservoir have a different wavelength from the low-energy ones. Thus, our results do not apply to atomic gases, although our methodology does[@Grenier2014].
Nonlinear systems and the figure of merit $ZT$ {#Sect:nonlinear+ZT}
----------------------------------------------
Engineers commonly state that wide-ranging applications for thermoelectrics would require them to have a dimensionless figure of merit, $ZT$, greater than three. This dimensionless figure of merit is a dimensionless combination of the linear-response coefficients [@books] $ZT= TGS^2/\Theta$, for temperature $T$, Seebeck coefficient $S$, electrical conductance $G$, and thermal conductance $\Theta$ . Yet for us, $ZT$ is just a way to characterize the efficiency, via $$\begin{aligned}
\eta_{\rm eng} = \eta_{\rm eng}^{\rm carnot} {\sqrt{ZT+1} -1 \over \sqrt{ZT+1} +1},
\nonumber\end{aligned}$$ with a similar relationship for refrigerators. Thus, someone asking for a device with a $ZT > 3$, actually requires one with an efficiency of more than one third of Carnot efficiency. This is crucial, because the efficiency is a physical quantity in linear and nonlinear situations, while $ZT$ is only meaningful in the linear-response regime [@Zebarjadi2007; @Grifoni2011; @2012w-pointcont; @Meair-Jacquod2013; @Michelini2014; @Azema-Lombardo-Dare2014].
Linear-response theory rarely fails for bulk semiconductors, even when $T_L$ and $T_R$ are very different. Yet it is completely [*inadequate*]{} for the quantum systems that we consider here. Linear-response theory requires the temperature drop on the scale of the electron relaxation length $l_{\rm rel}$ (distance travelled before thermalizing) to be much less than the average temperature. For a typical millimetre-thick bulk thermoelectric between a diesel motor’s exhaust system ($T_L\simeq 700$K) and its surroundings ($T_R\simeq 280$K), the relaxation length (inelastic scattering length) is of order the mean free path; typically 1-100nm. The temperature drop on this scale is tens of thousands of times smaller than the temperature drop across the whole thermoelectric. This is absolutely tiny compared with the average temperature, so linear-response [@Mahan-Sofo1996] works well, even though $(T_L-T_R)/T_L$ is of order one.
In contrast, for quantum systems ($L \ll l_{\rm rel}$), the whole temperature drop occurs on the scale of a few nanometres or less, and so linear-response theory is inapplicable whenever $(T_L-T_R)/T_L$ is not small.
Carnot efficiency {#Sect:Carnot}
-----------------
A system must be reversible (create no entropy) to have Carnot efficiency; proposals exist to achieve this in bulk [@Mahan-Sofo1996] or quantum [@Humphrey-Linke2005; @Kim-Datta-Lundstrom2009; @Jordan-Sothmann-Sanchez-Buttiker2013] thermoelectric. It requires that electrons only pass between reservoirs L and R at the energy where the occupation probabilities are identical in the two reservoirs [@Humphrey-Linke2005]. Thus, a thermoelectric requires two things to be reversible. Firstly, it must have a $\delta$-function-like transmission [@Mahan-Sofo1996; @Humphrey-Linke2005; @Kim-Datta-Lundstrom2009; @Jordan-Sothmann-Sanchez-Buttiker2013; @Sothmann-Sanchez-Jordan-Buttiker2013], which only lets electrons through at energy ${\epsilon}_0$. Secondly,[@Humphrey-Linke2005] the load’s resistance must be such that ${ e^{\operatorname{-}} }V = {\epsilon}_0 (1-T_R/T_L)$, so the reservoirs’ occupations are equal at ${\epsilon}_0$, see Fig. \[Fig:Fermi\].
By definition this means the current vanishes, and thus so does the power output, $P_{\rm gen}$. However, one can see how $P_{\rm gen}$ vanishes by considering a quantum system which lets electrons through in a tiny energy window $\Delta$ from ${\epsilon}_0$ to ${\epsilon}_0+\Delta$, see Fig \[Fig:tophat-width\]. When we take $\Delta\big/({k_{\rm B}}T_{L,R}) \to 0$, one has Carnot efficiency, however we will see (leading order term in Eq. (\[Eq:Pgen-eng-lowpower\])) that $$\begin{aligned}
P_{\rm gen} \propto {1 \over \hbar} \Delta^2,
\label{Eq:power-for-Delta-to-zero}\end{aligned}$$ which vanishes as $\Delta\big/({k_{\rm B}}T_{L,R}) \to 0$.
Heat-engine efficiency at finite power output and Curzon-Ahlborn efficiency {#Sect:eff-CA}
---------------------------------------------------------------------------
To increase the power output beyond that of a reversible system, one has to consider irreversible machines which generate a finite amount of entropy per unit of work generated. Curzon and Ahlborn[@Curzon-Ahlborn1975] popularized the idea of studying the efficiency of a heat-engine running at its maximum power output. For classical pumps, this efficiency is $\eta_{\rm eng}^{\rm CA} = 1- \sqrt{T_L/T_R}$, which is now called the Curzon-Ahlborn efficiency, although already given in Refs. \[\]. As refrigerators, these pumps have an efficiency at maximum cooling power of zero, although Refs. \[\] discuss ways around this.
The response of a given heat-engine is typically a “loop” of efficiency versus power (see Fig. \[Fig:summary\]) as one varies the load on the system[@Casati-review]. For a peaked transmission function with width $\Delta$ (see e.g. Fig. \[Fig:tophat-width\]), the loop moves to the left as one reduces $\Delta$. In the limit $\Delta \to 0$, the whole loop is squashed onto the $P_{\rm gen}=0$ axis. In linear-response language, this machine has $ZT \to \infty$. In this limit, the efficiency at maximum power can be very close to that of Curzon and Ahlborn [@Esposito2009-thermoelec] (the star in Fig. \[Fig:summary\]), just as its maximum efficiency can be that of Carnot[@Humphrey-Linke2005] (see previous section). However, its maximum power output is $\propto { e^{\operatorname{-}} }V\Delta/\hbar$ for small $\Delta$ (where $V$ is finite, chosen to ensure maximum power), which vanishes for $\Delta \to 0$, although it is much larger than Eq. (\[Eq:power-for-Delta-to-zero\]). Fig. \[Fig:summary\] shows that a system with larger $\Delta$ (such as the red curve) operating near its maximum efficiency will have both higher efficiency and higher power output than the one with small $\Delta$ (left most grey curve) operating at maximum power.
This article shows how to derive the thick black curve in Fig. \[Fig:summary\], thereby showing that there is a fundamental trade-off between efficiency and power output in optimal thermodynamic machines made from thermoelectrics [@footnote:casati-review]. As such, our work overturns the idea that maximizing efficiency at maximum power is the best route to machines with both high efficiency and high power. It also overturns the idea that systems with the narrowest transmission distributions (the largest $ZT$ in linear-response) are automatically the best thermoelectrics.
At this point we mention that other works[@Nakpathomkun-Xu-Linke2010; @Leijnse2010; @Meden2013; @Hershfield2013] have studied efficiencies for various systems with finite width transmission functions, for which power outputs can be finite. In particular, Ref. \[\] considered a boxcar transmission function, which is the form of transmission function that we have shown can be made optimal [@2014w-prl].
Pendry’s quantum bound on heat-flow {#Sect:Pendry}
-----------------------------------
An essential ingredient in this work is Pendry’s upper bound [@Pendry1983] on the heat-flow through a quantum system between two reservoirs of fermions. He found this bound using a scattering theory of the type discussed in Section \[Sect:ScatteringTheory\] below. It is a concrete example of a general principle due to Bekenstein [@Bekenstein], and the same bound applies in the presence of thermoelectric effects [@2012w-2ndlaw]. The bound on the heat flow out of reservoir $L$ is achieved when all the electrons and holes arriving at the quantum system from reservoir $L$ escape into reservoir $R$ without impediment, while there is no back-flow of electrons or holes from reservoir $R$ to L. The easiest way to achieve this is to couple reservoir $L$ through the quantum system to a reservoir $R$ at zero temperature, and then ensure the quantum system does not reflect any particles. In this case the heat current equals $$\begin{aligned}
J^{\rm qb}_L = {\pi^2 \over 6h} N {k_{\rm B}}^2 T_L^2,
\label{Eq:Jqb}\end{aligned}$$ where $N$ is the number of transverse modes in the quantum system. We refer to this as the quantum bound (qb) on heat flow, because it depends on the quantum wave nature of the electrons; it depends on $N$, which is given by the cross-sectional area of the quantum system divided by $\lambda_{\rm F}^2$, where $\lambda_{\rm F}$ is the electron’s Fermi wavelength. As such $J_L^{\rm qb}$ is ill-defined within classical thermodynamics.
Uniquely defining temperature, heat and entropy {#Sect:Unique}
===============================================
![\[Fig:Unique\] To implement the procedure in Section \[Sect:Unique\], one starts with the circuit unconnected, as in (a), one then connects the circuit, as in (b). After a long time $t_{\rm expt}$, one disconnects the circuit, returning to (a). The circles are the quantum thermoelectrics, as in Fig. \[Fig:thermocouple\]. ](figure3.pdf){width="\columnwidth"}
Works on classical thermodynamics have shown that the definition of heat and entropy flows can be fraught with difficulties. For example, the rate of change of entropy cannot always be uniquely defined in classical continuum thermodynamics[@Kern1975; @Day1977; @book:irreversible-thermodyn]. Here the situation is even more difficult, since the electrons within the quantum systems (circles in Fig. \[Fig:thermocouple\]) are not at equilibrium, and so their temperature cannot be defined. Thus, it is crucial to specify the logic which leads to our definitions of temperature, heat flow and entropy flow.
Our definition of heat flow originated in Refs. \[\], the rate of change of entropy is then found using the Clausius relation [@Footnote:Sivan-Imry] (see below). To explain these quantities and show they are unambiguous, we consider the following three step procedure for a heat engine. An analogue procedure works for a refrigerator.
- [**Step 1.**]{} Reservoir $L$ is initially decoupled from the rest of the circuit (see Fig. \[Fig:Unique\]a), has internal heat energy $Q_L^{(0)}$, and is in internal equilibrium at temperature $T_L^{(0)}$. The rest of the circuit is in equilibrium at temperature $T_R^{(0)}$ with internal heat energy $Q_R^{(0)}$. The internal heat energy is the total energy of the reservoir’s electron gas minus the energy which that gas would have in its ground-state. As such, the internal energy can be written as a sum over electrons and holes, with an electron at energy ${\epsilon}$ above the reservoir’s chemical potential (or a hole at energy ${\epsilon}$ below that chemical potential) contributing ${\epsilon}$ to this internal heat energy. The initial entropies are then $S^{(0)}_i = Q^{(0)}_i \big/ T^{(0)}_i$ for $i=L,R$.
- [**Step 2.**]{} We connect reservoir $L$ to the rest of the circuit ( (see Fig. \[Fig:Unique\]b) and leave it connected for a long time $t_{\rm expt}$. While we assume $t_{\rm expt}$ is long, we also assume that the reservoirs are all large enough that the energy distributions within them change very little during time $t_{\rm expt}$. Upon connecting the circuit elements, we assume a transient response during a time $t_{\rm trans}$, after which the circuit achieves a steady-state. We ensure that $t_{\rm expt}\gg t_{\rm trans}$, so the physics is dominated by this steady-state. Even then the flow will be noisy [@Blanter-Buttiker] due to the fact electrons are discrete with probabilistic dynamics. So we also ensure that $t_{\rm expt}$ is much longer than the noise correlation time, so that the noise in the currents is negligible compared to the average currents.
- [**Step 3.**]{} After the time $t_{\rm expt}$, we disconnect reservoir $L$ from the rest of the circuit. Again, there will be a transient response, however we assume that a weak relaxation mechanism within the reservoirs will cause the two parts of the circuit to each relax to internal equilibrium (see Fig. \[Fig:Unique\]a). After this one can unambiguously identify the temperature, $T_i$, internal energy $Q_i$ and Clausius entropy $S_i=Q_i\big/ T_i$ of the two parts of the circuit (for $i=L,R$). Since the reservoirs are large, we assume $T_i = T_i^{(0)}$.
Thus, we can unambiguously say that the heat-current out of reservoir $i$ [*averaged*]{} over the time $t_{\rm expt}$ is $$\begin{aligned}
\langle J_i \rangle = \big (Q^{(0)}_i - Q_i \big) \big/ t_{\rm expt}.\end{aligned}$$ For the above thermocouple, we treat the currents for each thermoelectric separately, writing the heat current out of reservoir $L$ as $J_L+J_{L'}$, where $J_L$ is the heat current from reservoir $L$ into the lower thermoelectric in Fig. \[Fig:thermocouple\] (the filled circle), and $J_{L'}$ is the heat current from reservoir $L$ into the upper thermoelectric in Fig. \[Fig:thermocouple\] (the open circle). Treating each thermoelectric separately is convenient, and also allows one to generalize the results to “thermopiles”, which contain hundreds of thermoelectrics arranged so that they are electrically in series, but thermally in parallel.
The average rate of change of entropy in the circuit is $\langle \dot S_{\rm circuit} \rangle = \langle \dot S \rangle +\langle \dot S' \rangle$, where $\langle \dot S \rangle$ is the average rate of change of entropy associated with the lower thermoelectric in Fig. (\[Fig:thermocouple\]), while $\langle \dot S' \rangle$ is that for the upper thermoelectric. Then $$\begin{aligned}
\langle \dot S\rangle = \langle \dot S_L \rangle + \langle \dot S_R \rangle
= -{\langle J_{\rm L} \rangle \big/ T_L} \,-\, {\langle J_{\rm R} \rangle \big/ T_R}\,,
\label{Eq:average-dotS-def}\end{aligned}$$ while $\langle \dot S'\rangle$ is the same with $J_L,J_R,T_R$ replaced by $J_{L'},J_{R'},T_{R'}$. We neglect the entropy of the thermoelectrics and load, by assuming their initial and final state are the same. This will be the case if they are small compared to the reservoirs, so their initial and final states a simply given by the temperature $T_R$.
The nonlinear scattering theory in Ref. \[\] captures long-time average currents (usually called the DC response in electronics), such as electrical current $\langle I_i \rangle$ and heat current $\langle J_i \rangle$, see references in Section \[Sect:ScatteringTheory\]. It is believed to be exact for non-interacting particles, and also applies when interactions can be treated in a mean-field approximation (see again section \[Sect:ScatteringTheory\]). A crucial aspect of the scattering theory is that we do not need to describe the non-equilibrium state of the quantum system during step 2. Instead, we need that quantum system’s transmission function, defined in section \[Sect:ScatteringTheory\].
In this article we will [*only*]{} discuss the long-time average of the rates of flows (not the noisy instantaneous flows), and thus will not explicitly indicate the average; so $I_i$, $J_i$ and $\dot S_i$ should be interpreted as $\langle I_i \rangle$, $\langle J_i \rangle$ and $\langle\dot S_i \rangle$.
Entropy production {#Sect:entropy-prod}
==================
There are little known universal relations between efficiency, power and and entropy production, which follow trivially from the laws of thermodynamics [@Cleuren2012]. Consider the lower thermoelectric in Fig. \[Fig:thermocouple\]a (filled circle), with $J_L$ and $J_R$ being steady-state heat currents into it from reservoir $L$ and R. Then the first law of thermodynamics is $$\begin{aligned}
J_R + J_L=P_{\rm gen},
\label{Eq:firstlaw}\end{aligned}$$ where $P_{\rm gen}$ is the electrical power generated. The Clausius relation for the rate of change of total entropy averaged over long times as in Eq. (\[Eq:average-dotS-def\]), is $$\begin{aligned}
\dot S = -{J_L \over T_L} + {J_L - P_{\rm gen} \over T_R},
\label{Eq:secondlaw}\end{aligned}$$ where we have used Eq. (\[Eq:firstlaw\]) to eliminate $J_R$.
For a heat engine, we take $J_L$ to be positive, which means $T_L > T_R$ and $J_R$ is negative. We use Eq. (\[Eq:eff-eng\]) to replace $J_L$ with $P_{\rm gen}/\eta_{\rm eng}$ in Eq. (\[Eq:secondlaw\]). Then, the rate of entropy production by a heat-engine with efficiency $\eta_{\rm eng}(P_{\rm gen})$ at power output $P_{\rm gen}$ is $$\begin{aligned}
\dot S (P_{\rm gen})
&=&
{P_{\rm gen} \over T_R} \left({\eta_{\rm eng}^{\rm carnot} \over \eta_{\rm eng}(P_{\rm gen})} -1 \right),
\label{Eq:dotS-eng}\end{aligned}$$ where the Carnot efficiency, $\eta_{\rm eng}^{\rm carnot}$, is given in Eq. (\[Eq:Carnot-eng\]). Hence, knowing the efficiency at power $P_{\rm gen}$, tells us the entropy production at that power. Maximizing the former minimizes the latter.
For refrigeration, the load in Fig. \[Fig:thermocouple\] is replaced by a power supply, the thermoelectric thus absorbs a power $P_{\rm abs}$ to extract heat from the cold reservoir. We take reservoir $L$ as cold ($T_L < T_R$) , so $J_L$ is positive. We replace $P_{\rm gen}$ by $-P_{\rm abs}$ in Eqs. (\[Eq:firstlaw\],\[Eq:secondlaw\]). We then use Eq. (\[Eq:eff-fri\]) to replace $P_{\rm abs}$ by $J_L/\eta_{\rm fri}$. Then the rate of entropy production by a refrigerator at cooling power $J_L$ is $$\begin{aligned}
\dot S (J_L)
&=&
{J_L \over T_R} \left({1\over \eta_{\rm fri}(J_L)} -{1 \over \eta_{\rm fri}^{\rm carnot}} \right),
\label{Eq:dotS-fri}\end{aligned}$$ where the Carnot efficiency, $\eta_{\rm fri}^{\rm carnot}$, is given in Eq. (\[Eq:Carnot-fri\]). Hence knowing a refrigerator’s efficiency at cooling power $J_L$ gives us its entropy production, and we see that maximizing the former minimizes the latter.
Eqs. (\[Eq:dotS-eng\],\[Eq:dotS-fri\]) hold for systems modelled by scattering theory, because this theory satisfies the laws of thermodynamics [@Bruneau2012]$^,$[@2012w-2ndlaw]. The rate of entropy production is zero when the efficiency is that of Carnot, but becomes increasingly positive as the efficiency reduces. In this article, we calculate the maximum efficiency for given power output, and then use Eqs. (\[Eq:dotS-eng\],\[Eq:dotS-fri\]) to get the minimum rate of entropy production at that power output.
Nonlinear Scattering Theory {#Sect:ScatteringTheory}
===========================
This work uses Christen and Büttiker’s nonlinear scattering theory [@Christen-ButtikerEPL96], which treats electron-electron interactions as mean-field charging effects. Refs. \[\] added thermoelectric effects by following works on linear-response [@Engquist-Anderson1981; @Sivan-Imry1986; @Butcher1990]. Particle and heat flows are given by the transmission function, ${\cal T}_{RL}({\epsilon})$, for electrons to go from left ($L$) to right ($R$) at energy ${\epsilon}$, where ${\cal T}_{RL}({\epsilon})$ is a [*self-consistently*]{} determined function of $T_L$, $T_R$ and $V$. In short, this self-consistency condition originates from the fact that electrons injected from the leads change the charge distribution in the quantum system, which in turn changes the behaviour of those injected electrons (via electron-electron interactions). The transmission function can be determined self-consistently with the charge distribution, if the latter is treated in a time-independent mean-field manner (neglecting single electron effects). We note that the same nonlinear scattering theory was also derived for resonant level models [@Humphrey-Linke2005; @Nakpathomkun-Xu-Linke2010] using functional RG to treat single-electron charging effects [@Meden2013].
The scattering theory for the heat current is based on the observation that an electron leaving reservoir $i$ at energy ${\epsilon}$ is carrying heat ${\epsilon}- \mu_i$ out of that reservoir [@Butcher1990], where $\mu_i$ is the reservoir’s chemical potential. Thus, a reservoir is cooled by removing an electron above the Fermi surface, but heated by removing a electron below the Fermi surface. It is convenient to treat empty states below a reference chemical potential (which we define as ${\epsilon}=0$), as “holes”. Then we do not need to keep track of a full Fermi sea of electrons, but only the holes in that Fermi sea. Then the heat-currents out of reservoirs L and R and into the quantum system are $$\begin{aligned}
J_L \! &=& \!
{1 \over h} \sum_\mu \int_0^\infty {\rm d}{\epsilon}\, ({\epsilon}- \mu{ e^{\operatorname{-}} }V_L) \,
{\cal T}^{\mu\mu}_{RL}({\epsilon}) \, \big[f_L^\mu ({\epsilon}) - f_R^\mu ({\epsilon})\big],
\nonumber \\
\label{Eq:JL}
\\
J_R \! &=& \!
{1 \over h} \sum_\mu \int_0^\infty {\rm d}{\epsilon}\, ({\epsilon}- \mu{ e^{\operatorname{-}} }V_R) \,
{\cal T}^{\mu\mu}_{RL}({\epsilon}) \, \big[f_R^\mu ({\epsilon}) - f_L^\mu ({\epsilon})\big],
\nonumber \\
\label{Eq:JR}\end{aligned}$$ where ${ e^{\operatorname{-}} }$ is the electron charge (${ e^{\operatorname{-}} }<0$), so ${ e^{\operatorname{-}} }V_i$ is the chemical potential of reservoir $i$ measured from the reference chemical potential (${\epsilon}=0$). The sum is over $\mu=1$ for “electron” states (full states above the reference chemical potential), and $\mu=-1$ for “hole” states (empty states below that chemical potential). The Fermi function for particles entering from reservoir $j$, is $$\begin{aligned}
f_j^\mu({\epsilon}) = \left(1+\exp\left[({\epsilon}- \mu { e^{\operatorname{-}} }V_j)\big/ ({k_{\rm B}}T_j) \right] \right)^{-1}.
\label{Eq:Fermi}\end{aligned}$$ The transmission function, ${\cal T}^{\nu\mu}_{ij}({\epsilon})$, is the probability that a particle $\mu$ with energy ${\epsilon}$ entering the quantum system from reservoir $j$ will exit into reservoir $i$ as a particle $\nu$ with energy ${\epsilon}$. We only allow $\nu=\mu$ here, since we do not consider electron to hole scattering within the quantum system (only common when superconductors are present). Interactions mean that ${\cal T}^{\mu\mu}_{RL}({\epsilon})$, is a [*self-consistently*]{} determined function of $T_L$, $T_R$ $V_L$ and $V_R$.
The system generates power $P_{\rm gen} = (V_R-V_L) I_L$, so $$\begin{aligned}
P_{\rm gen} \! &=& \!
{1\over h} \sum_\mu \int_0^\infty {\rm d}{\epsilon}\ \mu{ e^{\operatorname{-}} }(V_R-V_L)\,
{\cal T}^{\mu\mu}_{RL}({\epsilon}) \, \big[f_L^\mu ({\epsilon}) - f_R^\mu ({\epsilon})\big],
\nonumber \\
\label{Eq:Pgen}\end{aligned}$$ It is easy to verify that Eqs. (\[Eq:JL\]-\[Eq:Pgen\]) satisfy the first law of thermodynamics, Eq. (\[Eq:firstlaw\]). This theory assumes the quantum system to be relaxation-free, although decoherence is allowed as it does not change the structure of Eqs. (\[Eq:JL\]-\[Eq:Pgen\]). Relaxation is discussed in Section \[Sect:Relax\].
We define the voltage drop as $V=V_R-V_L$. Without loss of generality we take the reference chemical potential to be that of reservoir $L$, so $$\begin{aligned}
V_L=0, \qquad V_R=V,
\label{Eq:def-V}\end{aligned}$$ then $J_L$ and $P_{\rm gen}$ coincide with Eqs. (8,9) in Ref. \[\].
Numerous works have found the properties of thermoelectric systems from their transmission functions, ${\cal T}_{RL}({\epsilon})$. Linear-response examples include Refs. \[\], while nonlinear responses were considered in Refs. \[\], see Refs. \[\] for recent reviews. However, here we do not ask what is the efficiency of a given system, we ask what is the system that would achieve the highest efficiency, and what is this efficiency? This is similar in spirit to Ref. \[\], except that we maximize the efficiency for given power output.
We need to answer this question in the context of the mean-field treatment of electron-electron interactions[@Christen-ButtikerEPL96], in which the transmission function for any given system is the solution of the above mentioned self-consistency procedure. Despite this complexity, any transmission function (including all mean-field interactions) must obey $$\begin{aligned}
0\leq{\cal T}^{\mu\mu}_{RL}({\epsilon})\leq N \ \ \hbox{ for all }{\epsilon},
\label{Eq:basic-limits-on-transmisson}\end{aligned}$$ where $N$ is the number of transverse modes at the narrowest point in the nanostructure, see Fig. \[Fig:thermocouple\]. Let us assume that this is the [*only*]{} constraint on the transmission function. Let us assume that for any given $T_L$, $T_R$ and $V$, a clever physicist could engineer any desired transmission function, so long as it obeys Eq. (\[Eq:basic-limits-on-transmisson\]). Presumably they could do this either by solving the self-consistency equations for ${\cal T}^{\mu\mu}_{RL}({\epsilon})$, or by experimental trial and error. Thus, in this work, we find the ${\cal T}^{\mu\mu}_{RL}({\epsilon})$ which maximizes the efficiency given solely the constraint in Eq. (\[Eq:basic-limits-on-transmisson\]), and get this maximum efficiency. We then rely on future physicists to find a way to construct a system with this ${\cal T}^{\mu\mu}_{RL}({\epsilon})$ (although some hints are given in Section \[Sect:chain\]).
From thermoelectric optimization to thermocouple optimization {#Sect:transforming-from-full-to-open}
=============================================================
The rest of this article considers optimizing a single thermoelectric. However, an optimal thermocouple heat engine (or refrigerator) consists of two systems with opposite thermoelectric responses (full and open circles in Fig. \[Fig:thermocouple\]). So here we explain how to get the optimal thermocouple from the optimal thermoelectric.
Suppose the optimal system between $L$ and $R$ (the full circle) has a given transmission function ${\cal T}_{RL}^{\mu,\mu} ({\epsilon})$, which we will find in Section \[Sect:eng\]. This system generates an electron flow parallel to heat flow (so electric current is anti-parallel to heat flow, implying a negative Peltier coefficient). The system between $L$ and $R'$ (the open circle) must have the opposite response. For this we interchange the role played by electrons and holes compared with ${\cal T}_{RL}^{\mu,\mu} ({\epsilon})$, so the optimal system between $L$ and $R'$ has $$\begin{aligned}
{\cal T}_{R'L}^{\mu,\mu} ({\epsilon}) &=& {\cal T}_{RL}^{-\mu,-\mu} ({\epsilon}).\end{aligned}$$ If the optimal bias for the system between $L$ and $R$ is $V$ (which we will also find in Section \[Sect:eng\]), then the optimal bias for the system between $L$ and $R'$ is $-V$. Then the heat flow from reservoir $L$ into $R'$ equals that from $L$ into $R$, while the electrical current from $L$ into $R'$ is opposite to that from $L$ into $R$, and so $P_{\rm gen}$ is the same for each thermoelectric. The load across the thermocouple (the two thermoelectrics) must be chosen such that the bias across the thermocouple is $2V$. The condition that the charge current out of $L$ equals that into $L$ will then ensure that both thermoelectrics are at their optimal bias.
In the rest of this article we discuss power output and heat input [*per thermoelectric*]{}. For a thermocouple, one simply needs to multiply these by two, so the efficiency is unchanged but the power output is doubled.
Simple estimate of bounds on power output {#Sect:over-estimates}
=========================================
One of the principal results of Ref. \[\] is the quantum bounds on the power output of heat-engines and refrigerators. The exact derivation of these bounds is given in Sections \[Sect:qb-eng\] and \[Sect:qb-fri\]. Here, we give simple arguments for their basic form based on Pendry’s limit of heat flow discussed in Section \[Sect:Pendry\] above.
For a refrigerator, it is natural to argue that the upper bound on cooling power will be closely related to Pendry’s bound, Eq. (\[Eq:Jqb\]). We will show in Section \[Sect:qb-fri\] that this is the case. A two-lead thermoelectric can extract as much as half of $J^{\rm qb}_L$. In other words, the cooling power of any refrigerator must obey $$\begin{aligned}
J_L &\leq& {1 \over 2} J^{\rm qb}_L \ =\ {\pi^2 \over 12h} N {k_{\rm B}}^2 T_L^2.\end{aligned}$$
Now let us turn to a heat-engine operating between a hot reservoir $L$ and cold reservoir $R$. Following Pendry’s logic, we can expect that the heat current into the quantum system from reservoir $L$ cannot be more than $J_L^{\hbox{\scriptsize over-estimate}} ={\pi^2 \over 6h} N {k_{\rm B}}^2 (T_L^2-T_R^2)$. Similarly, no heat engine can exceed Carnot’s efficiency, Eq. (\[Eq:Carnot-eng\]). Thus, we can safely assume any system’s power output is less than $$\begin{aligned}
P_{\rm gen}^{\hbox{\scriptsize over-estimate}}
&=& \eta_{\rm eng}^{\rm carnot} J_L^{\hbox{\scriptsize over-estimate}}
\nonumber \\
&=& {\pi^2 N {k_{\rm B}}^2 (T_L+T_R) (T_L-T_R)^2 \over 6h \ T_L} .\end{aligned}$$ We know this is a significant over-estimate, because maximal heat flow cannot coincide with Carnot efficiency. Maximum heat flow requires ${\cal T}^{\mu\mu}_{RL}({\epsilon})$ is maximal for all ${\epsilon}$ and $\mu$, while Carnot efficiency requires a ${\cal T}^{\mu\mu}_{RL}({\epsilon})$ with a $\delta$-function-like dependence on ${\epsilon}$ (see Section \[Sect:Carnot\]). None the less, the full calculation in Section \[Sect:qb-eng\] shows that the true quantum bound on power output is such that [@footnote:qb2] $$\begin{aligned}
P_{\rm gen} &\leq& P_{\rm gen}^{\rm qb2} \,\equiv\,
A_0\, {\pi^2 \over h} N {k_{\rm B}}^2 \big(T_L-T_R\big)^2, \quad \quad
$$ where $A_0 \simeq 0.0321$. Thus, the simple over-estimate of the bound, $P_{\rm gen}^{\hbox{\scriptsize over-estimate}}$, differs from the true bound $P_{\rm gen}^{\rm qb2}$ by a factor of $(1+T_R/T_L)/(6A_0)$. In other words it over estimates the quantum bound by a factor between 5.19 and 10.38 (that is 5.19 when $T_R=0$ and 10.38 when $T_R=T_L$). This is not bad for such a simple estimate.
![\[Fig:Fermi\] Sketch of Fermi functions $f_L^\mu({\epsilon})$ and $f_L^\mu({\epsilon})$ in Eq. (\[Eq:Fermi\]), when $\mu { e^{\operatorname{-}} }V$ is positive, and $T_L > T_R$. Eq. (\[Eq:Eps0-guess\]) gives the point where the two curves cross, ${\epsilon}_0$.](figure4.pdf){width="0.9\columnwidth"}
Guessing the optimal transmission for a heat-engine {#Sect:guess-heat}
===================================================
Here we use simple arguments to guess the transmission function which will maximize a heat-engine’s efficiency for a given power output. We consider the flow of electrons from reservoir $L$ to reservoir $R$ (the filled circle Fig. \[Fig:thermocouple\]a, remembering ${ e^{\operatorname{-}} }<0$, so electron flow is in the opposite direction to $I$). To produce power, the electrical current must flow against a bias, so we require ${ e^{\operatorname{-}} }V$ to be positive, with $V$ as in Eq. (\[Eq:def-V\]). Inspection of the integrand of Eq. (\[Eq:Pgen\]) shows that it only gives positive contributions to the power output, $P_{\rm gen}$, when $\mu \big(f^\mu_L({\epsilon}) - f^\mu_R({\epsilon})\big) >0$. From Eq. (\[Eq:Fermi\]), one can show that $f^\mu_L({\epsilon})$ and $f^\mu_R({\epsilon})$ cross at $$\begin{aligned}
{\epsilon}_0 = \mu { e^{\operatorname{-}} }V \big/ (1-T_R/T_L),
\label{Eq:Eps0-guess}\end{aligned}$$ see Fig. \[Fig:Fermi\]. Since ${ e^{\operatorname{-}} }V$ is positive, we maximize the power output by blocking the transmission of those electrons ($\mu=1$) which have ${\epsilon}< {\epsilon}_0$, and blocking the transmission all holes ($\mu=-1$). For $\mu=1$, all energies above ${\epsilon}_0$ add to the power output. Hence, maximizing transmission for all ${\epsilon}> {\epsilon}_0$ will maximize the power output, giving $P_{\rm gen}=P_{\rm gen}^{\rm qb}$. However, a detailed calculation, such as that in Section \[Sect:eng\], is required to find the $V$ which will maximize $P_{\rm gen}$; remembering that $P_{\rm gen}$ depends directly on $V$ as well as indirectly (via the above choice of ${\epsilon}_0$).
Now we consider maximizing the efficiency at a given power output $P_{\rm gen}$, where $P_{\rm gen} < P_{\rm gen}^{\rm qb}$. Comparing the integrands in Eqs. (\[Eq:JL\],\[Eq:Pgen\]), we see that $J_L$ contains an extra factor of energy ${\epsilon}$ compared to $P_{\rm gen}$. As a result, the transmission of electrons ($\mu=1$) with large ${\epsilon}$ enhances the heat current much more than it enhances the power output. This means that the higher an electron’s ${\epsilon}$ is, the less efficiently it contributes to power production. Thus, one would guess that it is optimal to have an upper cut-off on transmission, ${\epsilon}_1$, which would be just high enough to ensure the desired power output $P_{\rm gen}$, but no higher. Then the transmission function will look like a “band-pass filter” (the “boxcar” form in Fig \[Fig:tophat-width\]), with ${\epsilon}_0$ and ${\epsilon}_1$ further apart for higher power outputs. This guess is correct, however the choice of $V$ affects both ${\epsilon}_0$ and ${\epsilon}_1$, so the calculation in Section \[Sect:eng\] is necessary to find the $V$, ${\epsilon}_0$ and ${\epsilon}_1$ which maximize the efficiency for given $P_{\rm gen}$.
Maximizing heat-engine efficiency for given power output {#Sect:eng}
========================================================
![\[Fig:tophat-width\] How the optimal “boxcar” transmission changes with increasing required power output. At maximum power output, a heat engine has ${\epsilon}_1 = \infty$ while ${\epsilon}_0$ remains finite. At maximum cooling power, a refrigerator has ${\epsilon}_1 = \infty$ and ${\epsilon}_0=0$. The qualitative features follow this sketch for all $T_R/T_L$, however the details depend on $T_R/T_L$, see Fig. \[Fig:Delta+V\]. ](figure5.pdf){width="\columnwidth"}
Now we present the central calculations of this article, finding the maximum efficiency of a quantum thermoelectric with [*given*]{} power output. In this section we consider heat-engines, while Section \[Sect:fri\] addresses refrigerators.
For a heat engine, our objective is to find the transmission function, ${\cal T}^{\mu\mu}_{RL}({\epsilon})$, and bias, $V$, that maximize the efficiency $\eta_{\rm eng}(P_{\rm gen})$ for given power output $P_{\rm gen}$. To do this we treat ${\cal T}^{\mu\mu}_{RL}({\epsilon})$ as a set of many slices each of width $\delta \to 0$, see the sketch in Fig. \[Fig:T-functions\]a. We define $\tau^{\mu}_\gamma$ as the height of the $\gamma$th slice, which is at energy ${\epsilon}_\gamma \equiv \gamma\delta$. Our objective is to find the optimal value of $\tau^{\mu}_\gamma$ for each $\mu,\gamma$, and optimal values of the bias, $V$; all under the constraint of fixed $P_{\rm gen}$. Often such optimization problems are formidable, however this one is fairly straightforward.
The efficiency is maximum for a fixed power, $P_{\rm gen}$, if $J_L$ is minimum for that $P_{\rm gen}$. If we make an infinitesimal change of $\tau^{\mu}_\gamma$ and $V$, we note that $$\begin{aligned}
\delta P_{\rm gen} &=& \left. {\partial P_{\rm gen} \over \partial \tau^{\mu}_\gamma} \right|_V \delta\tau^{\mu}_\gamma \ +\ P'_{\rm gen} \,\delta V,
\label{Eq:deltaPgen}
\\
\delta J_L &=& \left. {\partial J_L \over \partial \tau^{\mu}_\gamma} \right|_V \delta\tau^{\mu}_\gamma \ +\ J'_L \,\delta V,
\label{Eq:deltaJL}\end{aligned}$$ where $|_x$ indicates that the derivative is taken at constant $x$, and the primed indicates $\partial/\partial V$ for fixed transmission functions. If we want to fix $P_{\rm gen}$ as we change $\tau^{\mu}_\gamma$, we must change the bias $V$ to compensate. For this, we set $\delta P_{\rm gen}=0$ in Eq. (\[Eq:deltaJL\]) and substitute the result for $\delta V$ into Eq. (\[Eq:deltaPgen\]). Then $J_L$ decreases (increasing efficiency) for an infinitesimal increase of $\tau^{\mu}_\gamma$ at fixed $P_{\rm gen}$, if $$\begin{aligned}
\left.{\partial J_L \over \partial \tau^{\mu}_\gamma} \right|_{P_{\rm gen}}
&=&
\left.{\partial J_L \over \partial \tau^{\mu}_\gamma} \right|_V
- {J'_L \over P'_{\rm gen}}
\left.{\partial P_{\rm gen} \over \partial \tau^{\mu}_\gamma }\right|_V \ <\ 0.
\qquad \label{Eq:eng-condition}\end{aligned}$$ Comparing Eq. (\[Eq:JL\]) and Eq. (\[Eq:Pgen\]), one sees that $$\begin{aligned}
\left.{\partial J_L\over \partial \tau^{\mu}_\gamma }\right|_V &=&
{{\epsilon}_\gamma\over \mu { e^{\operatorname{-}} }V}\,
\left.{\partial P_{\rm gen} \over \partial \tau^{\mu}_\gamma }\right|_V .
\label{Eq:change-JL-to-change-Pgen}\end{aligned}$$ where $V$ is given in Eq. (\[Eq:def-V\]). Thus, the efficiency $\eta_{\rm eng}(P_{\rm gen})$ grows with a small increase of $\tau^{\mu}_\gamma$ if $$\begin{aligned}
\left({\epsilon}_\gamma - \mu { e^{\operatorname{-}} }V {J'_L \over P'_{\rm gen}} \right) \times
\left.{\partial P_{\rm gen} \over \partial \tau^{\mu}_\gamma }\right|_V \ <\ 0,
\label{Eq:eng-condition2}\end{aligned}$$ where $P_{\rm gen}$, $P'_{\rm gen}$, $J_L$, $J'_L$ and ${ e^{\operatorname{-}} }V$ are positive.
![\[Fig:T-functions\] A completely arbitrary transmission function $ {\cal T}_{RL}^{\mu\mu} ({\epsilon})$ (see Section \[Sect:eng\]). We take it to have infinitely many slices of width $\delta \to 0$, so slice $\gamma$ has energy ${\epsilon}_\gamma \equiv \gamma \delta$ and height $\tau^{\mu}_\gamma$. We find the optimal height for each slice. ](figure6.pdf){width="0.85\columnwidth"}
For what follows, let us define two energies $$\begin{aligned}
{\epsilon}_0 &=& { e^{\operatorname{-}} }V \big/ (1-T_R/T_L),
\label{Eq:eng-bounds-eps0}
\\
{\epsilon}_1 &=& { e^{\operatorname{-}} }V \, J'_L / P'_{\rm gen}.
\label{Eq:eng-bounds-eps1}\end{aligned}$$ One can see that $ \left.\left({\partial P_{\rm gen}/\partial\tau^{\mu}_\gamma }\right)\right|_V >0$ when both $\mu=1$ and ${\epsilon}> {\epsilon}_0$, and is negative otherwise. Thus, for $\mu=1$, Eq. (\[Eq:eng-condition2\]) is satisfied when ${\epsilon}_\gamma$ is between ${\epsilon}_0$ and ${\epsilon}_1$. For $\mu=-1$, Eq. (\[Eq:eng-condition2\]) is never satisfied.
A heat-engine is only useful if $P_{\rm gen}>0$, and this is only true for ${\epsilon}_0 <{\epsilon}_1$. Hence, if $\mu=1$ and ${\epsilon}_0 <{\epsilon}<{\epsilon}_1$, then $\eta_{\rm eng}(P_{\rm gen})$ is maximum for $\tau^{\mu}_\gamma$ at its maximum value, $\tau^{\mu}_\gamma=N$. For all other $\mu$ and ${\epsilon}_\gamma$, $\eta_{\rm eng}(P_{\rm gen})$ is maximum for $\tau^{\mu}_\gamma$ at its minimum value, $\tau^{\mu}_\gamma=0$. Since the left-hand-side of Eq. (\[Eq:eng-condition2\]) is not zero for any ${\epsilon}_\gamma\neq {\epsilon}_0,{\epsilon}_1$, there are no stationary points, which is why $\tau^{\mu}_\gamma$ never takes a value between its maximum and minimum values. Thus, the optimal ${\cal T}^{\mu\mu}_{RL}({\epsilon})$ is a “boxcar” or “top-hat” function, $$\begin{aligned}
{\cal T}^{\mu\mu}_{RL}({\epsilon})
\! &=& \! \left\{ \! \begin{array}{cl}
N & \hbox{ for } \mu=1 \ \hbox{ \& } \ \ {\epsilon}_0 \! <\! {\epsilon}\!
<\! {\epsilon}_1 \phantom{\big|}
\\
0 & \hbox{ otherwise } \phantom{\big|} \end{array} \right. \quad
\label{Eq:top-hat}\end{aligned}$$ see Fig. \[Fig:T-functions\]b. It hence acts as a band-pass filter, only allowing flow between L and R for electrons ($\mu=1$) in the energy window between ${\epsilon}_0$ to ${\epsilon}_1$.
Substituting a boxcar transmission function with arbitrary ${\epsilon}_0$ and ${\epsilon}_1$ into Eqs. (\[Eq:JL\],\[Eq:Pgen\]) gives $$\begin{aligned}
J_L &=& N \,
\big[F_L({\epsilon}_0)-F_R({\epsilon}_0)-F_L({\epsilon}_1)+F_R({\epsilon}_1) \big],
\label{Eq:JL-eng}
\\
P_{\rm gen} \!\! &=& \!N{ e^{\operatorname{-}} }V \, \big[G_L({\epsilon}_0)-G_R({\epsilon}_0)
-G_L({\epsilon}_1)+G_R({\epsilon}_1) \big], \qquad
\label{Eq:Pgen-eng}\end{aligned}$$ where we define $$\begin{aligned}
F_j({\epsilon}) = {1 \over h} \int_{\epsilon}^\infty
{ x \ {{\rm d}}x \over
1+ \exp\big[(x-{ e^{\operatorname{-}} }V_j)\big/({k_{\rm B}}T_j)\big] },
\label{Eq:Fintegral}
\\
G_j({\epsilon}) = {1 \over h}\int_{\epsilon}^\infty
{{{\rm d}}x \over
1+ \exp\big[(x-{ e^{\operatorname{-}} }V_j)\big/({k_{\rm B}}T_j)\big] },
\label{Eq:Gintegral}\end{aligned}$$ which are both positive for any ${\epsilon}>0$. Remembering that we took $V_L=0$ and $V_R=V$, these integrals are $$\begin{aligned}
F_L({\epsilon}) &=& {\epsilon}G_L({\epsilon}) -{({k_{\rm B}}T_L)^2 \over h}
{\rm Li}_2\big[-{{\rm e}}^{-{\epsilon}/({k_{\rm B}}T_L)}\big], \qquad \\
F_R({\epsilon}) &=& {\epsilon}G_R({\epsilon}) -{({k_{\rm B}}T_R)^2 \over h}
{\rm Li}_2\big[-{{\rm e}}^{-( {\epsilon}-{ e^{\operatorname{-}} }V)/({k_{\rm B}}T_R)}\big], \qquad \\
G_L({\epsilon}) &=& {{k_{\rm B}}T_L\over h}\ln\big[1+{{\rm e}}^{-{\epsilon}/({k_{\rm B}}T_L)}\big],
\\
G_R({\epsilon}) &=& {{k_{\rm B}}T_R\over h}\ln\big[1+{{\rm e}}^{-({\epsilon}-{ e^{\operatorname{-}} }V)/({k_{\rm B}}T_R)}\big],\end{aligned}$$ for dilogarithm function, ${\rm Li}_2(z)= \int_0^\infty t \, dt \big/({{\rm e}}^t/z -1)$.
We are only interested in cases where ${\epsilon}_0$ fulfills the condition in Eq. (\[Eq:eng-bounds-eps0\]), in this case $({\epsilon}_0-{ e^{\operatorname{-}} }V)/({k_{\rm B}}T_R) = {\epsilon}_0/({k_{\rm B}}T_L)$, which means $G_R({\epsilon}_0)$ and $F_R({\epsilon}_0)$ are related to $G_L({\epsilon}_0)$ and $F_L({\epsilon}_0)$ by $$\begin{aligned}
G_R({\epsilon}_0)&=&{T_R \over T_L}\, G_L({\epsilon}_0),
\label{Eq:G_R}
\\
F_R({\epsilon}_0)-{\epsilon}_0 G_R({\epsilon}_0)&=&{T_R^2\over T_L^2}\, \left( F_L({\epsilon}_0) -{\epsilon}_0 G_L({\epsilon}_0) \right). \quad\end{aligned}$$
![\[Fig:bounds\] Solutions of the transcendental equations giving optimal ${\epsilon}_1$ (heat-engine) or ${\epsilon}_0$ (refrigerator). In (a), the red curve is the optimal ${\epsilon}_1(V)$ for ${\epsilon}_1> {\epsilon}_0$, and the thick black line is ${\epsilon}_0$ in Eq. (\[Eq:eng-bounds-eps0\]). The red circle and red arrow indicate the low and high power limits discussed in the text. In (b), the red curve is the optimal ${\epsilon}_0(V)$ for ${\epsilon}_0<{\epsilon}_1$, and the thick black line is ${\epsilon}_1$ in Eq. (\[Eq:fri-bounds-eps1\]). ](figure7.pdf){width="\columnwidth"}
Eq. (\[Eq:eng-bounds-eps1\]) tells us that ${\epsilon}_1$ depends on $J_L$ and $P_{\rm gen}$, but that these depend in-turn on ${\epsilon}_1$. Hence to find ${\epsilon}_1$, we substitutes Eqs. (\[Eq:JL-eng\],\[Eq:Pgen-eng\]) into Eq. (\[Eq:eng-bounds-eps1\]) to get a transcendental equation for ${\epsilon}_1$ as a function of $V$ for given $T_R/T_L$. This equation is too hard to solve analytically (except in the high and low power limits, discussed in Sections \[Sect:qb-eng\] and \[Sect:eff-at-given-power\] respectively). The red curve in Fig. \[Fig:bounds\]a is a numerical solution for $T_R/T_L=0.2$.
Having found ${\epsilon}_1$ as a function of $V$ for given $T_R/T_L$, we can use Eqs. (\[Eq:JL-eng\],\[Eq:Pgen-eng\]) to get $J_L(V)$ and $P_{\rm gen}(V)$. We can then invert the second relation to get $V(P_{\rm gen})$. At this point we can find $J_L(P_{\rm gen})$, and then use Eq. (\[Eq:eff-eng\]) to get the quantity that we desire — the maximum efficiency at given power output, $\eta_{\rm eng}(P_{\rm gen})$.
In Section \[Sect:qb-eng\], we do this procedure analytically for high power ($P_{\rm gen} =P_{\rm gen}^{\rm qb2}$), and in Section \[Sect:eff-at-given-power\], we do this procedure analytically for low power ($P_{\rm gen} \ll P_{\rm gen}^{\rm qb2}$). For other cases, we only have a numerical solution for the transcendental equation for ${\epsilon}_1$ as a function of $V,T_R/T_L$, so we must do everything numerically.
![\[Fig:Delta+V\] (a) Plots of optimal $\Delta$ (left) and ${ e^{\operatorname{-}} }V$ (right) for a heat-engine with given power output, $P_{\rm gen}$, for $T_R/T_L=$ 0.05, 0.1, 0.2, 0.4, 0.6 and 0.8. We get ${\epsilon}_0$ from ${ e^{\operatorname{-}} }V$ by using Eq. (\[Eq:eng-bounds-eps0\]). (b) Plots of optimal $\Delta$ (left) and ${ e^{\operatorname{-}} }V$ (right) for a refrigerator with a given cooling power output, $J_L$, for $T_R/T_L=$ 1.05, 1.2, 1.5, 2, 4 and 10. We get ${\epsilon}_1$ from ${ e^{\operatorname{-}} }V$ by using Eq. (\[Eq:fri-bounds-eps1\]). ](figure8.pdf){width="\columnwidth"}
![\[Fig:allpowers\] Efficiencies of (a) heat-engines and (b) refrigerators. In (a) the curves are the maximum allowed heat-engine efficiency as a function of power outputs for $T_R/T_L= 0.05,0.2,0.4,0.6,0.8$ (from top to bottom). In (b) the curves are the maximum allowed refrigerator efficiency as a function of cooling power for $T_R/T_L= 1.05,1.2,1.5,2,4$ (from top to bottom). In both (a) and (b) the horizontal black lines indicate Carnot efficiency for each $T_R/T_L$, while the dashed black curves are the analytic theory for small cooling power, given in Eq. (\[Eq:eta-eng-small-Pgen\]) or Eq. (\[Eq:eta-fri-smallJ\]). The circles mark the analytic result for maximum power output. ](figure9.pdf){width="\columnwidth"}
Fig. \[Fig:Delta+V\]a gives the values of $\Delta=({\epsilon}_1-{\epsilon}_0)$ and ${ e^{\operatorname{-}} }V$ which result from solving the transcendental equation numerically for a variety of different $T_R/T_L$. Eq. (\[Eq:eng-bounds-eps0\]) then relates ${\epsilon}_0$ to ${ e^{\operatorname{-}} }V$. The qualitative behaviour of the resulting boxcar transmission function is shown in Fig. \[Fig:tophat-width\]. This numerical evaluation enables us to find the efficiency as a function of $P_{\rm gen}$ and $T_R/T_L$, which we plot in Fig. \[Fig:allpowers\]a.
Quantum bound on heat engine power output {#Sect:qb-eng}
-----------------------------------------
Here we want to find the highest possible power output of the heat-engine. In the previous section, we had the power as a function of two independent parameters, $V$ and ${\epsilon}_1$, with ${\epsilon}_0$ given by Eq. (\[Eq:eng-bounds-eps0\]). However, we know that Eq. (\[Eq:eng-bounds-eps1\]) will then determine a line in this two-dimensional parameter space (Fig. \[Fig:bounds\]a), which we can parametrize by the parameter $V$. The maximum possible power corresponds to $P_{\rm gen}'=0$, where we recall $P_{\rm gen}' \equiv {{\rm d}}P_{\rm gen} \big/ {{\rm d}}V$. This has two consequences, the first is that from Eq. (\[Eq:eng-bounds-eps0\]), we see that $P_{\rm gen}'=0$ means that ${\epsilon}_1 \to \infty$. Thus, the transmission function ${\cal T}_{RL}^{\mu\mu}({\epsilon})$, taking the form of a Heaviside step function, $\theta({\epsilon}-{\epsilon}_0)$, where ${\epsilon}_0$ is given in Eq. (\[Eq:eng-bounds-eps0\]). Taking Eq. (\[Eq:Pgen-eng\]) combined with Eq. (\[Eq:G\_R\]) for ${\epsilon}_1 \to \infty$, gives $$\begin{aligned}
P_{\rm gen}\big({\epsilon}_1\to\infty\big) &=&
N{ e^{\operatorname{-}} }V \, \left(1-{T_R \over T_L}\right)\ G_L\left( {{ e^{\operatorname{-}} }V \over 1-T_R/T_L}\right).
\nonumber\end{aligned}$$ The second consequence of $P_{\rm gen}'=0$, is that the $V$-derivative of this expression must be zero. This gives us the condition that $$\begin{aligned}
(1+B_0)\ln[1+B_0] +B_0\ln[B_0] =0\end{aligned}$$ where we define $B_0= \exp[-{ e^{\operatorname{-}} }V/({k_{\rm B}}T_L-{k_{\rm B}}T_R)] = \exp[-{\epsilon}_0/({k_{\rm B}}T_L)]$. Numerically solving this equation gives $B_0 \simeq 0.318$. Eq. (\[Eq:eng-bounds-eps0\]) means that this corresponds to ${ e^{\operatorname{-}} }V = -{k_{\rm B}}(T_L-T_R) \ln[0.318]= 1.146 \,{k_{\rm B}}(T_L-T_R)$, indicated by the red arrow in Fig. \[Fig:bounds\]a. Substituting this back into $P_{\rm gen}\big({\epsilon}_1\to\infty\big)$ gives the maximum achievable value of $P_{\rm gen}$, $$\begin{aligned}
P_{\rm gen}^{\rm qb2} =
A_0\, {\pi^2 \over h} N {k_{\rm B}}^2 \big(T_L-T_R\big)^2 \quad \quad
\label{Eq:P-qb2}\end{aligned}$$ with $$\begin{aligned}
A_0 \equiv B_0\ln^2[B_0]\big/\big[\pi^2(1+B_0)\big] \simeq 0.0321.\end{aligned}$$ We refer to this as the quantum bound (qb) on power output[@footnote:qb2], because of its origin in the Fermi wavelength of the electrons, $\lambda_{\rm F}$. We see this in the fact that $P_{\rm gen}^{\rm qb2}$ is proportional to the number of transverse modes in the quantum system, $N$, which is given by the cross-sectional area of the quantum system divided by $\lambda_{\rm F}^2$. This quantity has no analogue in classical thermodynamics.
The efficiency at this maximum power, $P_{\rm gen}^{\rm qb2}$, is $$\begin{aligned}
\eta_{\rm eng} (P_{\rm gen}^{\rm qb2})
&=& \eta_{\rm eng}^{\rm Carnot}\big/ \big( 1+C_0 (1+T_R/T_L) \big),
\label{Eq:Eff-at-Pqb2}\end{aligned}$$ with $$\begin{aligned}
C_0=-(1+B_0){\rm Li}_2(-B_0)\big/\big(B_0\ln^2[B_0]\big) \simeq 0.936.\end{aligned}$$ As such, it varies with $T_R/T_L$, but is always more than $0.3\,\eta_{\rm eng}^{\rm Carnot}$. This efficiency is less than Curzon and Ahlborn’s efficiency for all $T_R/T_L$ (although not much less). However, the power output here is infinitely larger than the maximum power output of systems that achieve Curzon and Ahlborn’s efficiency, see Section \[Sect:eff-CA\].
The form of Eq. (\[Eq:Eff-at-Pqb2\]) is very different from Curzon and Ahlborn’s efficiency. However, we note in passing that Eq. (\[Eq:Eff-at-Pqb2\]) can easily be written as $\eta_{\rm eng} (P_{\rm gen}^{\rm qb2})
= \eta_{\rm eng}^{\rm carnot} \big/ \left[(1+2C_0) -C_0\eta_{\rm eng}^{\rm carnot}\right]$, which is reminiscent of the efficiency at maximum power found for very different systems (certain classical stochastic heat-engines) in Eq. (31) of Ref. \[\].
Optimal heat-engine at low power output {#Sect:eff-at-given-power}
---------------------------------------
Now we turn to the opposite limit, that of low power output, $P_{\rm gen}\ll P_{\rm gen}^{\rm qb2}$, where we expect the maximum efficiency to be close to Carnot efficiency. In this limit, ${\epsilon}_1$ is close to ${\epsilon}_0$. Defining $\Delta= {\epsilon}_1- {\epsilon}_0$, we expand Eqs. (\[Eq:JL-eng\],\[Eq:Pgen-eng\]) in small $\Delta$ up to order $\Delta^3$. This gives $$\begin{aligned}
J_L &=& {P_{\rm gen} \over 1-T_R/T_L}
\ +\ {N\,\Delta^3\, (1-T_R/T_L) \over 3h\,{k_{\rm B}}T_R} g\!\left(x_0\right), \qquad
\label{Eq:JL-eng-lowpower}
\\
P_{\rm gen} &=& {N\,{\epsilon}_0 \,\Delta^2 \,(1-T_R/T_L)^2 \over 2h\,{k_{\rm B}}T_R}
\nonumber \\
& & \times \bigg[ g\!\left(x_0\right)
+ {\Delta \, (1+T_R/T_L) \over 3 \, {k_{\rm B}}T_R} \,{{\rm d} g(x_0)\over {\rm d} x_0}
\bigg],
\label{Eq:Pgen-eng-lowpower}\end{aligned}$$ where Eq. (\[Eq:eng-bounds-eps0\]) was used to write ${ e^{\operatorname{-}} }V$ in terms of ${\epsilon}_0$, and we defined $x_0={\epsilon}_0/({k_{\rm B}}T_L)$, and $g(x)={{\rm e}}^x/(1+{{\rm e}}^x)^2$. Thus, for small $\Delta$ we find that, $$\begin{aligned}
\eta_{\rm eng}(\Delta) = \eta_{\rm eng}^{\rm Carnot} \left(1-{2\Delta \over 3x_0{k_{\rm B}}T_L} + \cdots\right).
\label{Eq:Efficiency-in-terms-of-Delta}\end{aligned}$$ Eq. (\[Eq:eng-bounds-eps1\]) gives a transcendental equation for $x_0$ and $\Delta$. However, $\Delta$ drops out when it is small, and the transcendental equation reduces to $$\begin{aligned}
x_0 \tanh[x_0/2]=3,
\label{Eq:transcendental-small-Delta}\end{aligned}$$ for which $x_0 \equiv {\epsilon}_0/({k_{\rm B}}T_L) \simeq 3.24$. Eq. (\[Eq:eng-bounds-eps0\]) means that this corresponds to ${ e^{\operatorname{-}} }V =3.24 \,{k_{\rm B}}(T_L-T_R)$, indicated by the circle in Fig. \[Fig:bounds\]a. Now we can use Eq. (\[Eq:Pgen-eng-lowpower\]) to lowest order in $\Delta$, to rewrite Eq. (\[Eq:Efficiency-in-terms-of-Delta\]) in terms of $P_{\rm gen}$. This gives the efficiency for small $P_{\rm gen}$ as, $$\begin{aligned}
\eta_{\rm eng} \big(P_{\rm gen}\big) = \eta_{\rm eng}^{\rm Carnot}
\left(1- 0.478
\sqrt{ {T_R \over T_L} \ {P_{\rm gen} \over P_{\rm gen}^{\rm qb2}}} \,+ \cdots
\right)\!, \quad
\label{Eq:eta-eng-small-Pgen}\end{aligned}$$ where the dots indicate terms of order $(P_{\rm gen} /P_{\rm gen}^{\rm qb2})$ or higher. Eq. (\[Eq:dotS-eng\]) then gives the minimum rate of entropy production at power output $P_{\rm gen}$, $$\begin{aligned}
\dot S \big(P_{\rm gen}\big) = 0.478{P_{\rm gen}^{\rm qb2} \over \sqrt{T_LT_R} }
\left( {P_{\rm gen} \over P_{\rm gen}^{\rm qb2}}\right)^{3/2} \,+ {\cal O}[P_{\rm gen}^2] , \quad
\label{Eq:dotS-eng-small-Pgen}\end{aligned}$$ Thus, the maximal efficiency at small $P_{\rm gen}$ is that of Carnot minus a term that grows like $P_{\rm gen}^{1/2}$ (dashed curves in Fig. \[Fig:allpowers\]a), and the associated minimal rate of entropy production goes like $P_{\rm gen}^{3/2}$.
Note that Eq. (\[Eq:Efficiency-in-terms-of-Delta\]), shows that Carnot efficiency occurs at any $x_0$ (i.e. any ${\epsilon}_0$) when $\Delta$ is strictly zero (and so $P_{\rm gen}$ is strictly zero). However, for arbitrary $x_0$ the factor 0.478 in Eq. (\[Eq:eta-eng-small-Pgen\]) is replaced by $\sqrt{ 8\pi^2 A_0/[9 x_0^3 g(x_0)]}$. The value of $x_0$ that satisfied Eq. (\[Eq:transcendental-small-Delta\]) is exactly the one which minimizes this prefactor (its minimum being 0.478), and thus maximizes the efficiency for any small but finite $P_{\rm gen}$.
Guessing the optimal transmission for a refrigerator {#Sect:guess-fri}
====================================================
Here we use simple arguments to guess the transmission function which maximizes a refrigerator’s efficiency for given cooling power. The arguments are similar to those for heat-engines (Section \[Sect:guess-heat\]), although some crucial differences will appear.
We consider the flow of electrons from reservoir $L$ to reservoir $R$ (the filled circle in Fig. \[Fig:thermocouple\]a, remembering ${ e^{\operatorname{-}} }<0$ so electrons flow in the opposite direction to $I$). To refrigerate, the thermoelectric must absorb power, so the electrical current must be due to a bias, this requires ${ e^{\operatorname{-}} }V$ to be negative, with $V$ as in Eq. (\[Eq:def-V\]).
Inspection of the integrand of Eq. (\[Eq:JL\]) shows that it only gives positive contributions to the cooling power output, $J_L$, when $\big(f^\mu_L({\epsilon}) - f^\mu_R({\epsilon})\big) >0$. Since $T_L< T_R$ and ${ e^{\operatorname{-}} }V<0$, we can use Eq. (\[Eq:Fermi\]) to show that this is never true for holes ($\mu=-1$), and is only true for electrons ($\mu=1$) with energies ${\epsilon}< {\epsilon}_1$, where $$\begin{aligned}
{\epsilon}_1 = -{ e^{\operatorname{-}} }V \big/ (T_R/T_L-1).
\label{Eq:Eps1-guess}\end{aligned}$$ Thus, it is counter-productive to allow the transmission of electrons with ${\epsilon}> {\epsilon}_1$, or the transmission of any holes. Note that this argument gives us an [*upper*]{} cut-off on electron transmission energies, despite the fact it gave a [*lower*]{} cut-off for the heat engine (see Eq. (\[Eq:Eps0-guess\]) and the text around it). All electron ($\mu=1$) energies from zero to ${\epsilon}_1$ contribute positively to the cooling power $J_L$. To maximize the cooling power, one needs to maximize $\big(f^\mu_L({\epsilon}) - f^\mu_R({\epsilon})\big)$, this is done by taking ${ e^{\operatorname{-}} }V \to -\infty$ , for which ${\epsilon}_1 \to \infty$. This logic gives the maximum cooling power, which Section \[Sect:fri\] will show equals ${{\textstyle{\frac{1}{2}}}}J_L^{\rm qb}$.
Now we consider maximizing the efficiency at a given cooling output $J_L$, when $J_L <{{\textstyle{\frac{1}{2}}}}J_L^{\rm qb}$. Comparing the integrands in Eqs. (\[Eq:JL\],\[Eq:Pgen\]), we see that the extra factor of ${\epsilon}$ in $J_L$, means that allowing the transmission of electrons at low energies has a small effect on cooling power, while costing a similar electrical power as higher energies. Thus, it would seem to be optimal to have a lower cut-off on transmission, ${\epsilon}_0$, which would be just low enough to ensure the desired cooling power $J_L$, but no lower. Then the transmission function will acts as a “band-pass filter” (the “box-car” in Fig \[Fig:tophat-width\]), with ${\epsilon}_0$ and ${\epsilon}_1$ further apart for higher cooling power. This is correct, however the choice of $V$ affects ${\epsilon}_0$ and ${\epsilon}_1$, so the calculation in Section \[Sect:fri\] is necessary to find the $V$, ${\epsilon}_0$ and ${\epsilon}_1$ which maximize the efficiency for cooling power $J_L$.
Maximizing refrigerator efficiency for given cooling power {#Sect:fri}
==========================================================
Here we find the maximum refrigerator efficiency, also called the coefficient of performance (COP), for given cooling power $J_L$. The method is very similar to that for heat-engines, and here we mainly summarize the differences. The refrigerator efficiency increases for a fixed cooling power, $J_L$, if the electrical power absorbed $P_{\rm abs}=-P_{\rm gen}$ decreases for fixed $J_L$. This is so if $$\begin{aligned}
\left.{\partial P_{\rm abs} \over \partial \tau^{\mu}_\gamma }\right|_{J_L}
&=&
\left.{\partial P_{\rm abs} \over \partial \tau^{\mu}_\gamma }\right|_V
- {P'_{\rm abs} \over J'_L}
\left.{\partial J_L \over \partial \tau^{\mu}_\gamma }\right|_V \ <\ 0. \qquad
\label{Eq:fri-condition}\end{aligned}$$ where we recall that the primed means $({{\rm d}}/ {{\rm d}}V)$. This is nothing but Eq. (\[Eq:eng-condition\]) with $J_L \to P_{\rm abs}$ and $P_{\rm gen} \to J_L$. Using Eq. (\[Eq:change-JL-to-change-Pgen\]), we see that $\eta_{\rm fri}(J_L)$ grows with $\tau^{\mu}_\gamma$ for $$\begin{aligned}
\left( {-\mu { e^{\operatorname{-}} }V \over {\epsilon}_\gamma} - {P'_{\rm abs} \over J'_L} \right) \times
\left.{\partial J_L \over \partial \tau^{\mu}_\gamma }\right|_V \ <\ 0,
\label{Eq:fri-condition2}\end{aligned}$$ where $P_{\rm abs}$, $P'_{\rm abs}$, $J_L$, $J'_L$ and $-{ e^{\operatorname{-}} }V$ are all positive.
To proceed we define the following energies $$\begin{aligned}
{\epsilon}_0 &=& -{ e^{\operatorname{-}} }V \,J'_L / P'_{\rm abs},
\label{Eq:fri-bounds-eps0}
\\
{\epsilon}_1 &=& {-{ e^{\operatorname{-}} }V \big/ (T_R/T_L-1)}.
\label{Eq:fri-bounds-eps1}\end{aligned}$$ Then one can see that $ \left.\left({\partial J_L/\partial \tau^{\mu}_\gamma}\right)\right|_V$ is positive when both $\mu=1$ and ${\epsilon}< {\epsilon}_1^{\rm fri}$, and is negative otherwise. Thus, for $\mu=-1$, Eq. (\[Eq:fri-condition2\]) is never satisfied. For $\mu=1$, Eq. (\[Eq:fri-condition2\]) is satisfied when ${\epsilon}_\gamma$ is between ${\epsilon}_0^{\rm fri}$ and ${\epsilon}_1^{\rm fri}$. A refrigerator is only useful if $J_L>0$ (i.e. it removes heat from the cold reservoir), and this is only true for ${\epsilon}_0^{\rm fri} <{\epsilon}_1^{\rm fri}$. Hence, if $\mu=1$ and ${\epsilon}_0^{\rm fri} <{\epsilon}<{\epsilon}_1^{\rm fri}$, then $\eta_{\rm fri}(J_L)$ grows upon increasing $\tau^{\mu}_\gamma$. Thus, the optimum is when such $\tau^{\mu}_\gamma=N$. For all other $\mu$ and ${\epsilon}_\gamma$, $\eta_{\rm fri}(J_L)$ grows upon decreasing $\tau^{\mu}_\gamma$. Thus, the optimum is when such $\tau^{\mu}_\gamma=0$. This gives the boxcar transmission function in Eq. (\[Eq:top-hat\]), with ${\epsilon}_0$ and ${\epsilon}_1$ given by Eqs. (\[Eq:fri-bounds-eps0\],\[Eq:fri-bounds-eps1\]). Comparing with Eqs. (\[Eq:eng-bounds-eps0\],\[Eq:eng-bounds-eps1\]), we see these energies are the opposite way around for a refrigerator compared to a heat-engine (up to a minus sign).
Substituting Eqs. (\[Eq:JL-eng\],\[Eq:Pgen-eng\]) into Eq. (\[Eq:fri-bounds-eps0\]), one gets a transcendental equation for ${\epsilon}_0$ as a function of $V$ for given $T_R/T_L$. This equation is too hard to solve analytically (except in the high and low power limits, discussed in Sections \[Sect:qb-fri\] and \[Sect:lowpower-fri\]). The red curve in Fig. \[Fig:bounds\]b is a numerical solution for $T_R/T_L=1.5$.
Having found ${\epsilon}_0$ as a function of $V$ for given $T_R/T_L$, we can use Eqs. (\[Eq:JL-eng\],\[Eq:Pgen-eng\]) to get $J_L(V)$ and $P_{\rm abs}(V)=-P_{\rm gen}(V)$. We can then invert the first relation to get $V(J_L)$. Now, we can find $P_{\rm abs}(J_L)$, and then use Eq. (\[Eq:eff-fri\]) to get the quantity that we desire — the maximum efficiency (or COP), $\eta_{\rm fri}(J_L)$, at cooling power $J_L$.
Fig. \[Fig:Delta+V\]b gives the values of $\Delta=({\epsilon}_1-{\epsilon}_0)$ and ${ e^{\operatorname{-}} }V$ which result from solving the transcendental equation numerically. As noted, ${\epsilon}_1$ is related to ${ e^{\operatorname{-}} }V$ by Eq. (\[Eq:fri-bounds-eps1\]). The qualitative behaviour of the resulting boxcar transmission function is sketched in Fig. \[Fig:tophat-width\]. This numerical evaluation enables us to find efficiency as a function of $J_L$ and $T_R/T_L$, which we plot in Fig. \[Fig:allpowers\]b.
Quantum bound on refrigerator cooling power {#Sect:qb-fri}
-------------------------------------------
To find the maximum allowed cooling power, $J_L$, we look for the place where $J'_L=0$. From Eq. (\[Eq:fri-bounds-eps0\]) we see that this immediately implies ${\epsilon}_0 =0$. Taking Eq. (\[Eq:JL-eng\]) with ${\epsilon}_0=0$, we note by using Eq. (\[Eq:Fintegral\]) that $F_L(0)-F_R(0)$ grows monotonically as one takes $-{ e^{\operatorname{-}} }V \to \infty$. Similarly, for ${\epsilon}_1$ given by Eq. (\[Eq:eng-bounds-eps1\]), we note by using Eq. (\[Eq:Fintegral\]) and $T_R > T_L$ that $F_R({\epsilon}_1)-F_L({\epsilon}_1)$ grows monotonically as one takes $-{ e^{\operatorname{-}} }V \to \infty$. Thus, we can conclude that $J_L$ is maximal for $-{ e^{\operatorname{-}} }V \to \infty$, which implies ${\epsilon}_1 \to \infty$ via Eq. (\[Eq:fri-bounds-eps1\]). Physically, this corresponds to all electrons arriving at the quantum system from reservoir $L$ being transmitted into reservoir $R$, but all holes arriving from reservoir $L$ being reflected back into reservoir $L$. At the same time, reservoir $R$ is so strongly biased that it has no electrons with ${\epsilon}>0$ (i.e. no electrons above reservoir $L$’s chemical potential) to carry heat from R to L.
In this limit, $F_L({\epsilon}_1)=F_L({\epsilon}_1)=F_R({\epsilon}_0)=0$, so the maximal refrigerator cooling power is $$\begin{aligned}
J_L = {\pi^2 \over 12 h} N {k_{\rm B}}^2 T_L^2 ,
\label{Eq:J-qb-fri}\end{aligned}$$ where we used the fact that ${\rm Li}_2[1] = \pi^2/12$. This is exactly half the quantum bound on heat current that can flow out of reservoir $L$ given in Eq. (\[Eq:Jqb\]). The quantum bound is achieved by coupling reservoir $L$ to another reservoir with a temperature of absolute zero, through an contact with $N$ transverse mode. By definition a refrigerator is cooling reservoir $L$ below the temperature of the other reservoirs around it. In doing so, we show its cooling power is always less than or equal to $J_L^{\rm qb}/2$. However, it is intriguing that the maximum cooling power is independent of the temperature of the environment, $T_R$, of the reservoir being cooled (reservoir $L$). In short, the best refrigerator can remove all electrons (or all holes) that reach it from reservoir $L$, but it cannot remove all electrons [*and*]{} all holes at the same time.
It is easy to see that the efficiency of the refrigerator (COP) at this maximum possible cooling power is zero, simply because $|V| \to \infty$, so the power absorbed $P_{\rm abs} \to \infty$. However, one gets exponentially close to this limit for $-{ e^{\operatorname{-}} }V \gg {k_{\rm B}}T_R$, for which $P_{\rm abs}$ is large but finite, and so $\eta_{\rm fri}(J_L)$ remains finite (see Fig. \[Fig:allpowers\]b).
Optimal refrigerator at low cooling power {#Sect:lowpower-fri}
-----------------------------------------
Now we turn to the opposite limit, that of low cooling power output, $J_L\ll J_L^{\rm qb}$, where we expect the maximum efficiency to be close to Carnot efficiency. In this limit, ${\epsilon}_0$ is close to ${\epsilon}_1$. Defining $\Delta= {\epsilon}_1- {\epsilon}_0$, we expand Eqs. (\[Eq:JL-eng\],\[Eq:Pgen-eng\]) in small $\Delta$ up to order $\Delta^3$. This gives $$\begin{aligned}
J_L &=& {P_{\rm abs} \over T_R/T_L-1}
\ -\ {N\,\Delta^3 \,(T_R/T_L-1) \over 3h\,{k_{\rm B}}T_R} g\!\left(x_1\right), \qquad
\label{Eq:JL-fri-lowpower}
\\
P_{\rm abs}&=& {N\,{\epsilon}_1 \,\Delta^2 \,(T_R/T_L-1)^2 \over 2h\,{k_{\rm B}}T_R}
\nonumber \\
& & \times \bigg[ g\!\left(x_1\right)
- {\Delta \, (T_R/T_L+1) \over 3 \, {k_{\rm B}}T_R} \,{{\rm d} g(x_1)\over {\rm d} x_1}
\bigg],
\label{Eq:Pabs-fri-lowpower}\end{aligned}$$ where Eq. (\[Eq:fri-bounds-eps1\]) was used to write ${ e^{\operatorname{-}} }V$ in terms of ${\epsilon}_1$, and we define $x_1={\epsilon}_1/({k_{\rm B}}T_L)$, and $g(x)={{\rm e}}^x/(1+{{\rm e}}^x)^2$. Thus, for small $\Delta$ we find that the efficiency is $$\begin{aligned}
\eta_{\rm fri}(\Delta) = \eta_{\rm fri}^{\rm Carnot} \left(1-{2\Delta \over 3x_1{k_{\rm B}}T_L} + \cdots\right).
\label{Eq:COP-in-terms-of-Delta}\end{aligned}$$ Note that this is the same Eq. (\[Eq:Efficiency-in-terms-of-Delta\]) for the heat-engine at low power output, except that $x_0$ is replaced by $x_1$, and the Carnot efficiency is that of the refrigerator rather than that of the heat-engine.
Eq. (\[Eq:fri-bounds-eps0\]) gives a transcendental equation for $x_1$ and $\Delta$. However, $\Delta$ drops out when it is small, and the transcendental equation reduces to $$\begin{aligned}
x_1 \tanh [x_1/2]=3,
\label{Eq:condition-fri-smallJ}\end{aligned}$$ for which $x_1\equiv {\epsilon}_1/({k_{\rm B}}T_L) = 3.2436\cdots$. Again this is the same as for a heat-engine, Eq. (\[Eq:transcendental-small-Delta\]), but with $x_1$ replacing $x_0$. Eq. (\[Eq:fri-bounds-eps1\]) means that this corresponds to $-{ e^{\operatorname{-}} }V =3.2436 \,{k_{\rm B}}(T_R-T_L)$, indicated by the circle in Fig. \[Fig:bounds\]b. Now we can use Eq. (\[Eq:JL-fri-lowpower\]) to lowest order in $\Delta$, to rewrite Eq. (\[Eq:COP-in-terms-of-Delta\]) in terms of $J_L$. This gives the efficiency (or coefficient of performance, COP) for small $J_L$ as, $$\begin{aligned}
\eta_{\rm fri}(J_L) = \eta_{\rm fri}^{\rm Carnot}
\left(1- 1.09
\sqrt{
\,{T_R \over T_R-T_L}\ {J_L \over J_L^{\rm qb}} }\, + \cdots \right)\!,
\nonumber \\
\label{Eq:eta-fri-smallJ}\end{aligned}$$ where the dots indicate terms of order $(J_L/J_L^{\rm qb})$ or higher. Eq. (\[Eq:dotS-fri\]) gives the minimum rate of entropy generation at cooling power output $J_L$, as $$\begin{aligned}
\dot S \big(J_L\big) =
1.09
{J^{\rm qb}_L \over T_L}\sqrt{1-{T_L\over T_R}}
\left({J_L \over J_L^{\rm qb}} \right)^{3/2}\, + {\cal O}[J_L^2],
\nonumber \\
\label{Eq:dotS-fri-smallJ}\end{aligned}$$ Thus, we conclude that the maximum efficiency at small $J_L$ is that of Carnot minus a term that grows like $J_L^{1/2}$ (dashed curves in Fig. \[Fig:allpowers\]b), while the associated minimum entropy production goes like $J_L^{3/2}$.
We note that Carnot efficiency occurs at $J_L =0$ at any $x_1={\epsilon}_1/({k_{\rm B}}T_L)$. However, then the 1.09 factor in Eq. (\[Eq:eta-fri-smallJ\]) becomes $\sqrt{ 4\pi^2/[27 x_1^3 g(x_1)]}$. The condition in Eq. (\[Eq:condition-fri-smallJ\]) minimizes this factor (the minimum being 1.09), and thereby maximizes the efficiency for given $J_L$.
Implementation with a chain of quantum systems {#Sect:chain}
==============================================
![\[Fig:band\] (a) A chain of single level quantum dots with their energy levels aligned at energy $E_0$. (b) Transmission function when all hoppings are equal (note the strong oscillations). (c) Transmission function when all hoppings are carefully chosen (see text). To aid comparison all bandwidths in the plots have been normalized. ](figure10.pdf){width="\columnwidth"}
The previous sections have shown that maximum efficiency (at given power output) occurs when the thermoelectric system has a boxcar transmission function with the right position and width. In the limit of maximum power, the boxcar becomes a Heaviside step-function. Here, we give a detailed recipe for engineering such transmission functions for non-interacting electrons, and then discuss how to include mean-field interaction effects.
A Heaviside step-function is easily implemented with point-contact, whose transmission function is [@Buttiker-pointcont], $$\begin{aligned}
{\cal T}_{\rm L,isl}({\epsilon}) = \left(1+ \exp \left[- {{\epsilon}-E(V) \over D_{\rm tunnel} }\right] \right)^{-1}
\label{Eq:transmission-pc}\end{aligned}$$ where $E(V)$ is the height of the energy barrier induced by the point contact, and $D_{\rm tunnel}$ is a measure of tunnelling through the point contact. A sufficiently long point contact exhibits negligible tunnelling, $D_{\rm tunnel} \to 0$, so the transmission function simplifies to the desired Heaviside step-function, $\theta[{\epsilon}-E(V)]$.
For a potential implementation of a boxcar function we consider a chain of sites (quantum dots or molecules) with one level per site, as sketched in Fig. \[Fig:band\]a. The objective is that the hoppings between sites, $\{t_i\}$, will cause the states to hybridize to form a band centred at $E_0$, with a width given by the hopping[@Buttiker-private-comm]. Neglecting electron-electron interactions, the hopping Hamiltonian for five sites in the chain ($k=5$) can be written as $$\begin{aligned}
{\cal H}_{\rm chain} = \left(\begin{array}{ccccc}
-{{\rm i}}a_0 /2 \ & t_1 & 0 & 0 & 0 \\
t_1 & 0 & t_2 & 0 & 0 \\
0 & t_2 & 0 & t_3 & 0 \\
0 & 0 & t_3 & 0 & t_4 \\
0 & 0 & 0 & t_4 & \ -{{\rm i}}a_0/2
\end{array}\right).\end{aligned}$$ This is easily generalized to arbitrary chain length, $k$. Here we treat $a_0$ as a phenomenological parameter, however in reality it would be given by $|t_0|^2$ multiplied by the density of states in the reservoir. The fact that particles escape from the chain into the reservoirs, means the wavefunction for any given particle in the chain will decay with time. To model this, the Hamiltonian must be non-Hermitian, with the non-Hermiticity entering in the matrix elements for coupling to the reservoirs (top-left and bottom right matrix elements). These induce an imaginary contribution to each eigenstate’s energy $E_i$, with the wavefunction of any eigenstate decaying at a rate given by the imaginary part of $E_i$. The non-Hermiticity of $ {\cal H}_{\rm chain}$ also means that its left and right eigenvectors are different, defining $\big | \psi_i^{\rm (r)}\big \rangle$ as the $i$th right eigenvector of the matrix ${\cal H}_{\rm chain}$, and $\big\langle\psi_i^{\rm (l)} \big|$ as the $i$th left eigenvector, we have $\big\langle\psi_i^{\rm (l)} \big | \psi_j^{\rm (r)}\big \rangle = \delta_{ij}$ and $\big\langle\psi_i^{\rm (l)} \big | {\cal H}_{\rm chain} \big | \psi_i^{\rm (r)}\big \rangle = E_i$. The resolution of unity is $\sum_{i} \big | \psi_i^{\rm (r)}\big \rangle \, \big\langle\psi_i^{\rm (l)} \big | = {\bm 1}$, where ${\bm 1}$ is the $k$-by-$k$ unit matrix.
We define $|1\rangle$ as the vector whose first element is one while all its other elements are zero, and $|k\rangle$ as the vector whose last element (the $k$th element) is one while all its other elements are zero. Then the transmission probability at energy ${\epsilon}$ is given by $$\begin{aligned}
{\cal T}_{RL}({\epsilon}) &=&
\left| \big\langle k \big| \ \left[{\epsilon}-{\cal H}_{\rm chain}\right]^{-1} \big| 1 \big\rangle
\right|^2 \ a_0 \, ,\end{aligned}$$ where $[\cdots]^{-1}$ is a matrix inverse. To evaluate this matrix inverse, we introduce a resolution of unity to the left and right of $\left[{\epsilon}-{\cal H}_{\rm chain}\right]^{-1}$. This gives $$\begin{aligned}
{\cal T}_{RL} &=&
\sum_i \
\left|{\big\langle k \big | \psi_i^{\rm (r)}\big\rangle \ \big\langle\psi_i^{\rm (l)} \big| 1 \big\rangle
\over {\epsilon}-E_i} \right|^2 \ a_0.
\label{Eq:T-for-chain}\end{aligned}$$ For any given set of hoppings $a_0, t_1,\cdots t_k$, one can easily use a suitable eigenvector finder (we used Mathematica) to evaluate this equation numerically, while an analytic solution is straight-forward[@Grenier-private] for $k\leq 3$. When all hoppings in the chain are equal, there is a mismatch between the electron’s hopping dynamics in the chain and their free motion in the reservoirs. This causes resonances in the transmission, giving the Fabry-Perot-type oscillations in Fig. \[Fig:band\]b for $k=5$. However, we can carefully tune the hoppings (to be smallest in the middle of the chain and increasing towards the ends) to get the smooth transmission functions in Fig. \[Fig:band\]c. The $k=5$ curve in Fig. \[Fig:band\]c has $t_1=t_4= 0.39a_0$ and $t_2=t_3= 0.28 a_0$, and we choose $a_0= 1.91$ to normalize the band width to 1. As the number of sites in the chain, $k$, increases, the transmission function tends to the desired boxcar function.
The above logic assumes no electron-electron interactions. When we include interaction effects at the mean-field level, things get more complicated. If the states in the chain are all at the same energy $E_0$ when the chain is unbiased, they will not be aligned when there is a bias between the the reservoirs, because the reservoirs also act as gates on the chain states. To engineer a chain where the energies are aligned at the optimal bias, one must adjust the confinement potential of the dots in the chain (or adjust the chemistry of the molecules in the chain) so that their energies are sufficiently out of alignment at zero bias that they all align at optimal bias. In principle, we have the control to do this. However, in practice it would require a great deal of trial-and-error experimental fine tuning. We do not enter further into such practical issues here. Rather, we use the above example to show that there is no [*fundamental*]{} reason that the bound on efficiency cannot be achieved.
Many quantum systems in parallel {#Sect:in-parallel}
================================
To increase the efficiency at given power output, one must increase the number of transverse modes, $N$. This is because the efficiency decays with the power output divided by the quantum bounds in Eqs. (\[Eq:P-qb2\],\[Eq:J-qb-fri\]), and these bounds go like $N$. However, a strong thermoelectric response requires a transmission function that is highly energy dependent, this typically only occurs when the quantum system (point-contact, quantum dot or molecule) has dimensions of about a wavelength, which implies that $N$ is of order one. Crucial exceptions (beyond the scope of this work) are systems containing superconductors, either SNS structures[@Pekola-reviews] or Andreev interferometers [@Chandra98] (see also Ref. \[\] and references therein), where strong thermoelectric effects occur for large $N$.
In the absence of a superconductor, the only way to get large $N$ is to construct a device consisting of many $N=1$ systems in parallel, such as a surface covered with a certain density of such systems [@Jordan-Sothmann-Sanchez-Buttiker2013; @Sothmann-Sanchez-Jordan-Buttiker2013]. In this case $P_{\rm gen}^{\rm qb2}$ and $J_{\rm L}^{\rm qb}$ in Eqs. (\[Eq:P-qb2\],\[Eq:J-qb-fri\]) become bounds on the power per unit area, with $N$ being replaced by the number of transverse modes per unit area. With this one modification, all calculations and results in this article can be applied directly to such a situation. Carnot efficiency is achieved for a large enough surface area that the power per unit area is much less than $P_{\rm gen}^{\rm qb2}$ and $J_{\rm L}^{\rm qb}$.
It is worth noting that the number of modes per unit area cannot exceed $\lambda_{\rm F}^{-2}$, for Fermi wavelength $\lambda_{\rm F}$. From this we can get a feeling for the magnitude of the bounds discussed in this article. Take a typical semiconductor thermoelectric (with $\lambda_{\rm F}\sim 10^{-8}$m), placed between reservoirs at 700 K and 300 K (typical temperatures for a thermoelectric recovering electricity from the heat in the exhaust gases of a diesel engined car). Eq. (\[Eq:P-qb2\]) tells us that to get 100 W of power output from a semiconductor thermoelectric one needs a cross section of at least 4 mm$^2$. Then Eq. (\[Eq:eta-eng-small-Pgen\]) tells us that to get this power at 90% of Carnot efficiency, one needs a cross section of at least 0.4 cm$^2$. Remarkably, it is [*quantum mechanics*]{} which gives these bounds, even though the cross sections in question are macroscopic.
Phonons and photons carrying heat in parallel with electrons {#Sect:ph}
============================================================
![\[Fig:thermocouple-phonons\] The thermocouple heat-engine in Fig. \[Fig:thermocouple\], showing the heat flow due to phonon and photons, which carry heat from hot to cold by all possible routes (in parallel with the heat carried by the electrons). This always reduces the efficiency, so it should be minimized with suitable thermal insulation.](figure11.pdf){width="\columnwidth"}
Any charge-less excitation (such as phonons or photons) will carry heat from hot to cold, irrespective of the thermoelectric properties of the system. While some of the phonons and photons will flow through the thermoelectric quantum system, most will flow via other routes, see Fig. \[Fig:thermocouple-phonons\]. A number of theories for these phonon or photon heat currents take the form $$\begin{aligned}
J_{\rm ph}= \alpha (T_L^\kappa-T_R^\kappa),
\label{Eq:J_ph}\end{aligned}$$ where $J_{\rm ph}$ is the heat flow out of the L reservoir due to phonons or photons. The textbook example of such a theory is that of black-body radiation between the two reservoirs, then $\kappa=4$ and $\alpha$ is the Stefan-Boltzmann constant. An example relevant to suspended sub-Kelvin nanostructures is a situation where a finite number $N_{\rm ph}$ of phonon or photon modes carry heat between the two reservoirs [@Pendry1983; @photons; @phonons; @2012w-pointcont] then $\kappa=2$ and $\alpha \leq N_{\rm ph}\pi^2 {k_{\rm B}}^2/(6h)$.
One of the biggest practical challenges for quantum thermoelectrics is that phonons and photons will often carry much more heat than the electrons. This is simply because the hot reservoir can typically radiate heat in all directions as phonons or photons, while electrons only carry heat through the few nanostructures connected to that reservoir. Thus, in many cases the phonon or photon heat flow will dominate over the electronic one. However, progress is being made in blocking phonon and photon flow, by suspending the nanostructure to minimize phonon flow [@phonons] and engineering the electromagnetic environment to minimize photon flow [@photons], and it can be hoped that phonon and phonon effects will be greatly reduced in the future. Hence, here we consider the full range from weak to strong phonon or photon heat flows.
For compactness in what follows we will only refer to phonon heat flows (usually the dominant parasitic effect). However, strictly one should consider $J_{\rm ph}$ as the sum of the heat flow carried by phonons, photons and any more exotic charge-less excitations that might exist in a given circuit (mechanical oscillations, spin-waves, etc.).
![\[Fig:allpowers-phonons\] Plots of the maximum efficiency allowed when the there is a phonon heat flow, $J_{\rm ph}$, in parallel with the heat carried by the electrons. The curves in (a) are for $T_R/T_L=0.2$, with $J_{\rm ph} = 0, 0.01, 0.1, 1$ (from top to bottom); the curves come from Eq. (\[Eq:eng-e+ph\]) with $\eta_{\rm eng}(P_{\rm gen})$ given in Fig. \[Fig:allpowers\]a. The curves in (b) are for $T_R/T_L=1.5$, with $J_{\rm ph} = 0, 0.02, 0.1, 0.4$ (from top to bottom); the curves come from Eq. (\[Eq:fri-e+ph\]) with $\eta_{\rm eng}(P_{\rm gen})$ given in Fig. \[Fig:allpowers\]b. The maximum cooling power (open circles) is $({{\textstyle{\frac{1}{2}}}}J^{\rm qb}_L-J_{\rm ph})$. ](figure12.pdf){width="\columnwidth"}
Heat-engine with phonons
------------------------
For heat-engines, the phonon heat-flow is in parallel with electronic heat-flow, so the heat-flow for a given $P_{\rm gen}$ is $(J_L+J_{\rm ph})$, rather than just $J_L$ (as it was in the absence of phonons). Thus, the efficiency in the presence of the phonons is $$\begin{aligned}
\eta^{\rm e+ph}_{\rm eng}(P_{\rm gen})={P_{\rm gen} \over J_L(P_{\rm gen})+J_{\rm ph}}.\end{aligned}$$ Writing this in terms of the efficiency, we get $$\begin{aligned}
\eta_{\rm eng}^{\rm e+ph} (P_{\rm gen})
&=& \big[ \eta_{\rm eng}^{-1} (P_{\rm gen}) +J_{\rm ph} / P_{\rm gen} \big]^{-1},
\label{Eq:eng-e+ph}\end{aligned}$$ where $\eta_{\rm eng} (P_{\rm gen})$ is the efficiency for $J_{\rm ph}=0$. Given the maximum efficiency at given power in the absence of phonons, we can use this result to find the maximum efficiency for a given phonon heat flow, $J_{\rm ph}$. An example of this is shown in Fig. \[Fig:allpowers-phonons\]a. It shows that for finite $J_{\rm ph}$, Carnot efficiency is not possible at any power output.
Phonons have a huge effect on the efficiency at small power output. Whenever $J_{\rm ph}$ is non-zero, the efficiency vanishes at zero power output, with $$\begin{aligned}
\eta^{\rm e+ph}_{\rm eng}(P_{\rm gen})=P_{\rm gen}\big/J_{\rm ph} \ \ \ \hbox{ for } \ P_{\rm gen} \ll J_{\rm ph}.
\label{Eq:eta-with-phonons-smallP}\end{aligned}$$ As $J_{\rm ph}$ increases, the range of applicability of this small $P_{\rm gen}$ approximation (shown as dashed lines in Fig. \[Fig:allpowers-phonons\]) grows towards the maximum power $P_{\rm eng}^{\rm qb}$ (open circles). In contrast, phonon heat flows have little effect on the efficiency near the maximum power output, until these flows become strong enough that $J_{\rm ph} \sim P_{\rm gen}$.
For strong phonon flow, where $J_{\rm ph} \gg P_{\rm gen}$, Eq. (\[Eq:eta-with-phonons-smallP\]) applies at all powers up to the maximum, $P_{\rm gen}^{\rm qb2}$. Then, the efficiency is maximal when the power is maximal, where maximal power is the quantum bound given in Eq. (\[Eq:P-qb2\]). Thus, the system with both maximal power and maximal efficiency is that with a Heaviside step transmission function (see section \[Sect:chain\]).
Refrigerator with phonons
-------------------------
For a refrigerator to extract heat from a reservoir at rate $J$ in the presence of phonons carrying a back flow of heat $J_{\rm ph}$, that refrigerator must extract heat at a rate $J_L=J+J_{\rm ph}$. Note that for clarity, in this section we take $J_{\rm ph}$ to be positive when $T_L< T_R$ (opposite sign of that in Eq. (\[Eq:J\_ph\])). Thus, the efficiency, or COP, in the presence of phonons, is the heat current extracted, $J$, divided by the electrical power required to extract heat at the rate $J_L=(J+J_{\rm ph})$. This means that $$\begin{aligned}
\eta_{\rm fri}^{\rm e+ph}(J) &=& {J \, \eta_{\rm fri}(J+J_{\rm ph}) \over J+J_{\rm ph}} ,
\label{Eq:fri-e+ph}\end{aligned}$$ where $\eta_{\rm fri} (J)$ is the efficiency for $J_{\rm ph}=0$. We can use this result to find the maximum efficiency for a given phonon heat flow, $J_{\rm ph}$. An example is shown in Fig. \[Fig:allpowers-phonons\]b.
Eq. (\[Eq:fri-e+ph\]) means that the phonon flow suppresses the maximum cooling power, so $J$ must now obey $$\begin{aligned}
J&\leq& {{\textstyle{\frac{1}{2}}}}J_L^{\rm qb} -J_{\rm ph}
\label{Eq:Jqb-phonons}\end{aligned}$$ with $J_L^{\rm qb}$ given in Eq. (\[Eq:Jqb\]). Thus, the upper bound (open circles) in Fig. \[Fig:allpowers-phonons\]b move to the left as $J_{\rm ph}$ increases.
When the reservoir being refrigerated (reservoir $L$) is at ambient temperature, $T_R$, then $J_{\rm ph}=0$ while $J_L^{\rm qb}$ is finite. However, as reservoir $L$ is refrigerated (reducing $T_L$), $J_{\rm ph}$ grows, while $J_L^{\rm qb}$ shrinks. As a result, at some point (before $T_L$ gets to zero) one arrives at $J_{\rm ph} = {{\textstyle{\frac{1}{2}}}}J_L^{\rm qb}$, and further cooling of reservoir $L$ is impossible. Thus, given the $T_L$ of $J_{\rm ph}$ for a given system, one can easily find the lowest temperature that reservoir $L$ can be refrigerated to, by solving the equation $J_{\rm ph} = {{\textstyle{\frac{1}{2}}}}J_L^{\rm qb}$ for $T_L$ To achieve this temperature, one needs the refrigerator with the maximum cooling power (rather than the most efficient one), this is a system with a Heaviside step transmission function (see section \[Sect:chain\]). Such a system’s refrigeration capacities were discussed in Ref. \[\].
We also note that, as with the heat-engine, phonons have a huge effect on the efficiency at small cooling power, as can be seen in Fig. \[Fig:allpowers-phonons\]b. Whenever $0< J_{\rm ph}<{{\textstyle{\frac{1}{2}}}}J_L^{\rm qb}$, the efficiency vanishes for small cooling power, with $$\begin{aligned}
\eta^{\rm e+ph}_{\rm fri}(J)=J \ {\eta_{\rm fri}(J_{\rm ph}) \over J_{\rm ph}}\ \ \ \ \hbox{ for } \ J \ll J_{\rm ph}.\end{aligned}$$
Relaxation in a quantum system without B-field {#Sect:Relax}
==============================================
Elsewhere in this article, we neglected relaxation in the quantum system. In other words, we assumed that electrons traverse the system in a time much less than the time for inelastic scattering from phonons, photons or other electrons. We now consider systems in which there is such relaxation, and ask if this relaxation could enable a system to exceed the bounds found above for relaxationless systems. To make progress, we restrict our interest to systems with negligible external magnetic field (B-field) [@Footnote:Error-my-PRL]. As yet, we have not been able to consider the rich interplay of relaxation and B-field [@Casati2011; @Sanchez-Serra2011; @Entin-Wohlman2012].
We use the voltage-probe model [@voltage-probe] shown in Fig. \[Fig:relax\]a. A system with relaxation is modeled as a phase-coherent scatterer coupled to a fictitious reservoir $M$ (a region in which relaxation occurs instantaneously). The rate of the relaxation is controlled by the transmission of the lead coupling to reservoir $M$. We then separate the phase-coherent scatterer into scatterers 1,2 and 3, as shown in Fig. \[Fig:relax\]b, each with their own transmission functions ${\cal T}_{ij}({\epsilon})$ with $i,j \in L,M,R$. We assume that the transmission is unchanged under reversal of direction, so ${\cal T}_{ij}({\epsilon})={\cal T}_{ji}({\epsilon})$ for all ${\epsilon}$ and $i,j$. This condition is guaranteed by time-reversal symmetry whenever the B-field has a negligible effect on the electron and hole dynamics. However, it also applies for any B-field when all particles relax as they traverse the quantum system (then ${\cal T}_{LR}({\epsilon})={\cal T}_{RL}({\epsilon})=0$, which is sufficient to force ${\cal T}_{ij}({\epsilon})={\cal T}_{ji}({\epsilon})$ for all $i,j$).
![\[Fig:relax\] (a) A quantum system in which relaxation occurs is modelled phenomenologically by a coherent quantum system coupled to a third fictitious reservoir $M$ in which the relaxation occurs. (b) The same model after we have separated the system’s scattering matrix into three components. The dashed arrows are the exchange of phonons or photons. The arm containing scatterers 1 and 2 is shown in (c) for a heat-engine, and in (d) for a refrigerator. ](figure13.pdf){width="\columnwidth"}
If the relaxation involves electron-phonon or electron-photon interactions (typically any system which is not sub-Kelvin), the phonons or photons with which the electrons interact usually flow easily between the system and the reservoirs. Thus, these phonons or photons can carry heat current between the fictitious reservoir $M$ and reservoirs $L,R$ (dashed arrows in Fig. \[Fig:relax\]). The total electrical and heat currents into reservoir $M$ must be zero, and this constraint determines reservoir $M$’s bias, $V_M$, and temperature, $T_M$.
Method of over-estimation
-------------------------
The optimal choice of ${\cal T}_{ML}$ and ${\cal T}_{RM}$ depends on $T_M$, while $T_M$ depend on the heat current, and thus on ${\cal T}_{ML}$ and ${\cal T}_{RM}$. The solution and optimization of this self-consistency problem has been beyond our ability to resolve, even though we have restricted ourselves to a simple model of relaxation in a system with negligible B-field. Instead, we make a simplification which leads to an [*over-estimate*]{} of the efficiency. We assume $V_M,T_M$ are free parameters (not determined from ${\cal T}_{ML}$ and ${\cal T}_{RM}$), with $T_M$ between $T_L$ and $T_R$. If we find the optimal ${\cal T}_{ML}$ and ${\cal T}_{RM}$ for given $T_M$, and then find the optimal $T_M$ (irrespective of whether it is consistent with ${\cal T}_{ML}$ and ${\cal T}_{RM}$ or not), we have an over-estimate of the maximal efficiency. Even with this simplification, we have only been able to address the low-power and high-power limits. However, we show below that this over-estimate is sufficient to prove the following.
- At low power, relaxation cannot make the system’s efficiency exceed that of the optimal relaxation-free system with $N_{\rm max}$ modes.
- Relaxation cannot make a system’s power exceed that of the maximum possible power of a relaxation-free system with $N_{\rm max}$ modes.
Defining $N_L$ and $N_R$ as the number of transverse modes in the system to the left and right of the region where relaxation occurs, $$\begin{aligned}
N_{\rm max}={\rm max}[N_L,N_R],
\label{Eq:Nmax}\end{aligned}$$
Efficiency of heat-engine with relaxation {#Sect:eng-eff-relax}
-----------------------------------------
To get the efficiency for our model of a quantum system with relaxation, we must find the efficiency for the system in Fig. \[Fig:relax\]b. This system has two “arms”. One arm contains scatterers 1 and 2, and we define its efficiency as $\eta_{\rm eng}^{(1\&2)}$. The other arm contains scatterer 3, and we define its efficiency as $\eta_{\rm eng}^{(3)}$. The efficiency of the full system, $ \eta_{\rm eng}^{\rm total}(P_{\rm gen})$, is given by $$\begin{aligned}
{1 \over \eta_{\rm eng}^{\rm total}(P_{\rm gen}) } =
{p_{\rm rel} \over \eta_{\rm eng}^{(1\&2)} (p_{\rm rel}P_{\rm gen}) }
+{q_{\rm rel} \over \eta_{\rm eng}^{(3)} ( q_{\rm rel}P_{\rm gen}) },
\label{Eq:heatengines-in-parallel}\end{aligned}$$ Here $p_{\rm rel}$ is the proportion of transmitted electrons that have passed through the arm containing scatterers 1 and 2, while $q_{\rm rel}=(1-p_{\rm rel})$ is the proportion that have passed through the arm containing scatterer 3. Physically, $p_{\rm rel}$ is the probability that an electron entering the quantum system relaxes before transmitting, while $q_{\rm rel}$ is the probability that it transmits before relaxing. One sees from Eq. (\[Eq:heatengines-in-parallel\]) that the maximal efficiency for a given $p_{\rm rel}$ occurs when both $\eta_{\rm eng}^{(1\&2)}$ and $\eta_{\rm eng}^{(3)}$ are maximal.
The upper-bound on $\eta_{\rm eng}^{(3)}$ is that given in section \[Sect:eng\] with $q_{\rm rel}N_L$ modes to the left and $q_{\rm rel}N_R$ modes to the right. Our objective now is to find the maximum $\eta_{\rm eng}^{(1\&2)}$ with $N_1=p_{\rm rel}N_L$ modes on the left and $N_2=p_{\rm rel}N_R$ modes on the right. More precisely our objective is to find an [*over-estimate*]{} of this maximum. For the heat flows indicated in Fig. \[Fig:relax\]c, the efficiency is $$\begin{aligned}
\eta_{\rm eng}^{(1\&2)} &\equiv& P_{\rm gen}^{(1\&2)}\big/J
\nonumber \\
&=&\!\! {1 \over J}\left[
P_{\rm gen}^{(1)}(J_1;T_M,T_L) + P_{\rm gen}^{(2)}(J_2;T_R,T_M)\right]\! , \qquad \ \end{aligned}$$ where $J_1=J-J_1^{\rm ph}-J^{\rm ph}$ and $J_2 =J-J_2^{\rm ph}-J^{\rm ph}-P_{\rm gen}^{(1)}$. One sees that $\eta_{\rm eng}^{(1\&2)}$ is maximal for given $T_M$ when $J^{\rm ph}=J_1^{\rm ph}=J_2^{\rm ph}=0$ (these heat currents cannot be negative because $T_L > T_M> T_R$). Thus, to get our over-estimate of the maximal efficiency for given $T_M$, we assume these phonon and photon heat-currents are zero. Then, with a little algebra, one finds that $$\begin{aligned}
1-\eta_{\rm eng}^{(1\&2)}\big(P_{\rm gen}^{(1\&2)}\big) = \left(1-\eta_{\rm eng}^{(1)}\big(P_{\rm gen}^{(1)}\big) \right)\left(1-\eta_{\rm eng}^{(2)}\big(P_{\rm gen}^{(2)}\big) \right),
\nonumber\end{aligned}$$ where $P_{\rm gen}^{(1)}$ and $P_{\rm gen}^{(1)}$ are related to $P_{\rm gen}^{(1\&2)}$ by $$\begin{aligned}
P_{\rm gen}^{(\mu)} = P_{\rm gen}^{(1\&2)} \eta_{\rm eng}^{(\mu)} \big/ \eta_{\rm eng}^{(1\& 2)},
\label{Eq:P1-or-2}\end{aligned}$$ for $\mu=1,2$. For given $T_M$, one maximizes $\eta_{\rm eng}^{(1\&2)}$ by independently maximizing $\eta_{\rm eng}^{(1)}$ and $\eta_{\rm eng}^{(2)}$. For low powers, Eq. (\[Eq:eta-eng-small-Pgen\]) with $P,N,T_R\to P_1, N_1, T_M$ gives $\eta_{\rm eng}^{(1)}$, while with $P,N,T_L\to P_2, N_2, T_M$ gives $\eta_{\rm eng}^{(2)}$. In this limit, we can treat efficiencies in Eq. (\[Eq:P1-or-2\]) to zeroth order in $P_{\rm gen}^{(1\&2)}$, taking them to be Carnot efficiencies, so $$\begin{aligned}
P_{\rm gen}^{(1)} \simeq {T_L-T_M \over T_L-T_R}P_{\rm gen}^{(1\&2)}, \quad
P_{\rm gen}^{(1)} \simeq {T_M-T_R \over T_L-T_R}P_{\rm gen}^{(1\&2)}.
\nonumber\end{aligned}$$ Then some algebra gives the over-estimate of efficiency at low powers for given $T_M$, to be $$\begin{aligned}
\eta_{\rm eng}^{(1\&2)} \leq
\eta_{\rm eng}^{\rm Carnot}
\left(1- 0.478
\sqrt{ {T_R \over T_L} \ {P_{\rm gen} \ K_{\rm rel}\over P_{\rm gen}^{\rm qb2}(N=1)} } \right)\! ,
\quad\end{aligned}$$ with $P_{\rm gen}^{\rm qb2}(N=1)$ given by Eq. (\[Eq:P-qb2\]) with $N=1$, and $$\begin{aligned}
K_{\rm rel} =
\sqrt{{1 \over N_1}\,{T_R(T_L-T_M) \over T_M(T_L-T_R)}}
+ \sqrt{{1 \over N_2}\,{T_L(T_M-T_R) \over T_M(T_L-T_R)}},\quad
\label{Eq:Krelax}\end{aligned}$$ where $N_1= p_{\rm rel} N_L$ and $N_2=p_{\rm rel} N_L$ are respectively the number of transmission modes in scattering matrices 1 and 2. The over-estimate of $\eta_{\rm eng}^{(1\&2)}$ is maximal when $T_M$ is chosen to minimize $K_{\rm rel}$. The two minima of $K_{\rm rel}$ are at $T_M=T_R$ and $T_M=T_L$, for which the values of $K_{\rm rel}$ are $1/\sqrt{N_1}$ and $1/\sqrt{N_2}$ respectively. Thus, we have $$\begin{aligned}
K_{\rm rel} \geq 1/\sqrt{p_{\rm rel}N_{\rm max}}\ ,
\label{Eq:Krelax-limit} \end{aligned}$$ with $N_{\rm max}$ in Eq. (\[Eq:Nmax\]). Thus, whatever $T_M$ may be, $$\begin{aligned}
\eta_{\rm eng}^{(1\&2)} \left(P_{\rm gen}^{(1\&2)}\right) &\leq&
\eta_{\rm eng}^{\rm Carnot}
\nonumber \\
& & \times
\left(\! 1- 0.478
\sqrt{ {T_R \over T_L} {P_{\rm gen}^{(1\&2)} \over P_{\rm gen}^{\rm qb2}(p_{\rm rel}N_{\rm max}) } } \right)\! .
\nonumber \\
\label{Eq:eta1&2-bound}\end{aligned}$$ Since $P_{\rm gen}^{(1\&2)} = p_{\rm rel} P_{\rm gen}$, we can simplify Eq. (\[Eq:eta1&2-bound\]) by noting that $$\begin{aligned}
{P_{\rm gen}^{(1\&2)} \over P_{\rm gen}^{\rm qb2}(p_{\rm rel}N_{\rm max}) } =
{P_{\rm gen} \over P_{\rm gen}^{\rm qb2}(N_{\rm max}) } \end{aligned}$$ where $P_{\rm gen}$ is the total power generated by the combined system made of scatterers 1,2 and 3. Then substituting the result into Eq. (\[Eq:heatengines-in-parallel\]), we get an over-estimate of the efficiency at power output $P_{\rm gen}$ which is equal to the upper bound we found in the absence of relaxation, Eq. (\[Eq:eta-eng-small-Pgen\]).
Thus, we can conclude that for small power outputs, no quantum system with relaxation within it can exceed the upper-bound on efficiency found for a [*relaxation-free*]{} system with $N_{\rm max}$ transverse modes. Since the proof is based on an over-estimate of the efficiency for a system with relaxation, we cannot say if a system with finite relaxation can approach the bound in Eq. (\[Eq:eta-eng-small-Pgen\]). Unlike in the relaxation-free case, we cannot say what properties the quantum system with relaxation (as given in terms of the properties of the effective scatterers 1, 2 and 3) are necessary to maximize the efficiency at given power output. We simply know that it cannot exceed Eq. (\[Eq:eta-eng-small-Pgen\]).
Refrigerator with relaxation
----------------------------
Our objective is to find an over-estimate of the maximal efficiency of a refrigerator that is made of quantum systems in which relaxation occurs. The efficiency of the system with relaxation, $ \eta_{\rm fri}^{\rm total}(P_{\rm gen})$, is given by $$\begin{aligned}
\eta_{\rm fri}^{\rm total}(J_L) =
p_{\rm rel} \eta_{\rm fri}^{(1\&2)} (p_{\rm rel}J_L)
+q_{\rm rel} \eta_{\rm fri}^{(3)} ( q_{\rm rel}J_L),
\label{Eq:fridges-in-parallel}\end{aligned}$$ thus we need to find an upper bound on $\eta_{\rm fri}^{(1\&2)}$. We make an over-estimate of this efficiency by taking $T_M$ to be a free parameter between $T_L$ and $T_R$. For given $T_M$, the efficiency of the combined systems 1 and 2 is $$\begin{aligned}
\eta_{\rm fri}^{(1\&2)} (J)= J \Big/ \big[ P_{\rm abs}^{(1)}(J_1)
+ P_{\rm abs}^{(2)}(J_2) \big], \end{aligned}$$ where $J_1=J+J_1^{\rm ph}+J^{\rm ph}$ and $J_2=J+J_2^{\rm ph} +J^{\rm ph}+P_{\rm abs}^{(1)}$, see Fig. \[Fig:relax\]d. This efficiency is maximized when $J_1^{\rm ph},J_2^{\rm ph},J^{\rm ph} = 0$ (since $T_L<T_M<T_R$ means these currents are not negative). Then a little algebra gives $$\begin{aligned}
1+{1 \over \eta_{\rm fri}^{(1\&2)}(J)}
= \left[
1+{1 \over \eta_{\rm fri}^{(1)}(J)}
\right]\left[
1+{1 \over \eta_{\rm fri}^{(2)}\big(J_2\big)}
\right] \! , \qquad\end{aligned}$$ where $J_2= J+P_{\rm abs}^{(1)}= J\big[1+1/\eta_{\rm fri}^{(1)}(J)\big]$. Thus, to maximize $\eta_{\rm fri}^{(1\&2)}(J)$ for given $T_M$, one must maximize both $\eta_{\rm fri}^{(1)}$ and $\eta_{\rm fri}^{(2)}$. For low power, this can be done using Eq. (\[Eq:eta-fri-smallJ\]) (much as for the heat-engine in Section \[Sect:eng-eff-relax\] above) giving $$\begin{aligned}
\eta_{\rm fri}^{(1\&2)}
\leq \eta_{\rm fri}^{\rm Carnot}
\! \left(\! 1- 1.09
\sqrt{
{T_R \over T_R-T_L}{J_L K_{\rm rel}\over J_L^{\rm qb}(N=1)}}\,\right) \! , \ \ \end{aligned}$$ where $K_{\rm rel}$ is given in Eq. (\[Eq:Krelax\]), and $J_L^{\rm qb}(N=1)$ is given by Eq. (\[Eq:J-qb-fri\]) with $N=1$. The over-estimate of $\eta_{\rm fri}^{(1\&2)}$ is maximal when $K_{\rm rel}$ is minimal, see Eq. (\[Eq:Krelax-limit\]). Substituting this into Eq. (\[Eq:fridges-in-parallel\]), we see that the efficiency with relaxation does not exceed the result in Eq. (\[Eq:eta-fri-smallJ\]) for a [*relaxation-free*]{} system with $N_{\rm max}$ transverse modes.
Quantum bounds on power with relaxation
---------------------------------------
For a heat-engine, the arm with scatterers 1 and 2, has a maximum power, $$\begin{aligned}
P_{\rm gen}^{(1\&2)} \leq A_0\, {\pi^2 \over h} {k_{\rm B}}^2
\left[ N_1\big(T_L-T_M\big)^2+ N_2\big(T_M-T_R\big)^2 \right],
\nonumber\end{aligned}$$ Since $(T_L-T_M)^2 +(T_M-T_R)^2 \leq (T_L-T_R)^2$, the power of the full system cannot exceed the maximum power of a relaxation-less system, Eq. (\[Eq:P-qb2\]), with $N_{\rm max}$ modes.
For a refrigerator, the arm containing scatterers 1 and 2 has a maximum cooling power, $$\begin{aligned}
J \leq \left\{
\begin{array}{l}
\pi^2 N_1 {k_{\rm B}}^2 T_L^2 \big/(12 h) \\
\pi^2 N_2 {k_{\rm B}}^2 T_M^2 \big/(12 h) -P_{\rm abs}^{(1)} \ ,
\end{array}\right. \end{aligned}$$ where $P_{\rm abs}^{(1)}$ is the electrical power absorbed by scatter 1. The upper (lower) term is the limit on the heat-flow into scatterer 1 (scatterer 2), noting that the heat-flow into scatterer 2 is $J+P_{\rm abs}^{(1)}$. Unless $N_2 \gg N_1$, the lower limit is the more restrictive one. In any case, the cooling power of the full system can never exceed the maximum power of a relaxation-less system, Eq. (\[Eq:J-qb-fri\]), with $N_{\rm max} $ modes.
Conclusions {#Sect:conclusions}
===========
The upper bound on efficiency at zero power (i.e. Carnot efficiency) is classical, since it is independent of wavelike nature of the electrons. However, this work on thermoelectrics shows that the upper bound on efficiency at finite power is quantum, depending on the ratio of the thermoelectric’s cross-section to the electrons’ Fermi wavelength. If one thought that electrons were classical (strictly zero wavelength), one would believe that Carnot efficiency was achievable at any power output. Quantum mechanics appears to tell us that this is not so.
However, a crucial point for future work is to discover how universal our bounds on efficiency at finite power are. Our bounds currently rely on the quantum system being (a) well modelled by the nonlinear scattering theory with its mean-field treatment of electron-electron interactions, (b) coupled to only two reservoirs (hot and cold), and (c) relaxation free. Under certain conditions we have also shown that they apply when there is relaxation in the quantum system. We cannot yet prove that our results are as general as Pendry’s bound on heat flow[@Pendry1983], which applies for arbitrary relaxation and for more than two reservoirs [@2012w-2ndlaw], as well as for electronic Luttinger liquids[@Kane-Fisher] and bosons[@Pendry1983]. It also remains to be seen if our bound occurs in systems with strong electron-electron interactions (Coulomb blockade, Kondo physics, etc.). More generally, we wonder whether similar bounds apply to those thermodynamic machines that do not rely on thermoelectric effects, such as Carnot heat engines.
Acknowledgements
================
I am very grateful to M. Büttiker for the suggestion which led to the implementation in Section \[Sect:chain\]. I thank P. Hänggi for questions on entropy flow which led to section \[Sect:Unique\]. I thank L. Correa for questions which led to a great improvement of section \[Sect:eff-CA\]. I thank C. Grenier for an analytic solution of Eq. (\[Eq:T-for-chain\]) for $k=3$.
[1]{}
F. Giazotto, T.T. Heikkila, A. Luukanen, A.M. Savin, J.P. Pekola, Rev. Mod. Phys. [**78**]{}, 217 (2006). J.T. Muhonen, M. Meschke, J.P. Pekola, Rep. Prog. Phys. [**75**]{}, 046501 (2012).
G. Benenti, G. Casati, T. Prosen, and K. Saito, Eprint arXiv:1311.4430.
B. Sothmann, R. Sánchez, and A.N. Jordan, Nanotechnology [**26**]{}, 032001 (2015).
F. Haupt, M. Leijnse, H.L. Calvo, L. Classen, J. Splettstoesser, and M.R. Wegewijs, Physica Status Solidi (b) [**250**]{}, 2315 (2013).
M. Paulsson and S. Datta, Phys. Rev. B [**67**]{}, 241403(R) (2003).
P. Reddy, S.Y. Jang, R.A. Segalman, A. Majumdar, Science, [**315**]{}, 1568 (2007).
H.J. Goldsmid, [*Introduction to Thermoelectricity*]{} (Springer, Heidelberg, 2009). F.J. DiSalvo, Science [**285**]{}, 703 (1999). A. Shakouri and M. Zebarjadi, Chapt 9 of [*Thermal nanosystems and nanomaterials*]{}, S. Volz (Ed.) (Springer, Heidelberg, 2009). A. Shakouri, Annu. Rev. Mater. Res. [**41**]{}, 399 (2011).
C. Grenier, C. Kollath, A. Georges, Eprint arXiv:1209.3942. J.-P. Brantut, C. Grenier, J. Meineke, D. Stadler, S. Krinner, C. Kollath, T. Esslinger, and A. Georges, Science [**342**]{}, 713 (2013). Ch. Grenier, A. Georges, and C. Kollath, Phys. Rev. Lett. [**113**]{}, 200601 (2014).
J. Gemmer, M. Michel, and G Mahler, Quantum Thermodynamics (Springer-Verlag, Berlin, 2010).
R.S. Whitney Phys. Rev. Lett. [**112**]{}, 130601 (2014).
M. Zebarjadi, K. Esfarjani, and A. Shakouri, Appl. Phys. Lett. [**91**]{}, 122104 (2007); Mater. Res. Soc. Proc. [**1044**]{}, U10-04 (2008). B. Muralidharan, and M. Grifoni, Phys. Rev. B [**85**]{}, 155423 (2012). R.S. Whitney, Phys. Rev. B [**88**]{}, 064302 (2013). J. Meair, and Ph. Jacquod, J. Phys.: Condens. Matter [**25**]{} 082201, (2013). A. Crépieux, and F. Michelini, J. Phys.: Condens. Matter [**27**]{}, 015302 (2015).
J. Azema, P. Lombardo, and A.-M. Daré, Phys. Rev. B, [**90**]{}, 205437 (2014).
G.D. Mahan, and J.O. Sofo, Proc. Nat. Acad. Sci. USA, [**93**]{}, 7436 (1996).
T.E. Humphrey, and H. Linke, Phys. Rev. Lett. [**94**]{}, 096601 (2005).
R. Kim, S. Datta, M.S. Lundstrom, J. Appl. Phys. [**105**]{}, 034506 (2009).
A.N. Jordan, B. Sothmann, R. Sanchez, and M. Büttiker, Phys. Rev. B, [**87**]{}, 075312 (2013).
B. Sothmann, R. Sanchez, A.N. Jordan,and M. Büttiker, New J. Phys. [**15**]{} (2013) 095021.
F.L. Curzon, and B. Ahlborn, Am. J. Phys. [**43**]{}, 22 (1975).
J. Yvon, Proceedings of the International Conference on Peaceful Uses of Atomic Energy (Vol. 2), 337 (United Nations, New York, 1956).
P. Chambadal, Les Centrales Nucléaires (Armand Colin, 1957), p. 41. I. I. Novikov, J. Nucl. Energy II 7, 125 (1958) \[Atomnaya Energiya 3, 409 (1957)\].
S. Velasco, J.M.M. Roco, A. Medina, and A.C. Hernández, Phys. Rev. Lett. [**78**]{}, 3241 (1997) C. de Tomás, A.C. Hernández, and J.M.M. Roco, Phys. Rev. E [**85**]{}, 010104(R) (2012) Y. Apertet, H. Ouerdane, A. Michot, C. Goupil, P. Lecoeur, EPL, [**103**]{}, 40001 (2013). L.A. Correa, J.P. Palao, G. Adesso, and D. Alonso, Phys. Rev. E [**90**]{}, 062124 (2014)
M. Esposito, K. Lindenberg and C. Van den Broeck, EPL, [**85**]{}, 60010 (2009).
This point is easily overlooked in plots where power is in units of each system’s maximum power, such as Fig. 4 of Ref. \[\]. It is easy to misread such a plot as implying that the system with $ZT \to \infty$ is the most efficient at all powers.
N. Nakpathomkun, H.Q. Xu, and H. Linke, Phys. Rev. B, [**82**]{}, 235428 (2010). D.M. Kennes, D. Schuricht and V. Meden, EPL [**102**]{} (2013) 57003 (2013).
M. Leijnse, M.R. Wegewijs, and K. Flensberg, Phys. Rev. B [**82**]{}, 045412 (2010).
S. Hershfield, K.A. Muttalib, and B.J. Nartowt, Phys. Rev. B, [**88**]{}, 085426 (2013).
J.B. Pendry, J. Phys. A.: Math. Gen. [**16**]{}, 2161 (1983).
J.D. Bekenstein, Phys. Rev. Lett. [**46**]{}, 623 (1981). J.D. Bekenstein, Phys. Rev. D [**30**]{}, 1669 (1984).
R.S. Whitney, Phys. Rev. B [**87**]{}, 115404 (2013).
W. Kern, Z. Physik B[**20**]{}, 215 (1975). W.A. Day, Acta Mechanica [**27**]{}, 251 (1977). D. Jou, J. Casas-Vázquez, G. Lebon, [*Extended Irreversible Thermodynamics*]{} (Springer-Verlag, Berlin, 2001)
H.-L. Engquist and P.W. Anderson, Phys. Rev. B [**24**]{}, 1151 (1981). U. Sivan and Y. Imry, Phys. Rev. B [**33**]{}, 551 (1986). P.N. Butcher, J. Phys.: Condens. Matter, [**2**]{}, 4869 (1990). Ref. \[\] instead considered the von Neumann entropy.
Ya.M. Blanter, and M. Büttiker, Phys. Rep. [**336**]{}, 1 (2000).
T. Christen and M. Büttiker, Europhys. Lett. [**35**]{}, 523 (1996).
See e.g. Eq. (11) of B. Cleuren, B. Rutten, and C. Van den Broeck, Phys. Rev. Lett. [**108**]{}, 120603 (2012).
L. Bruneau, V. Jakšić, and C.-A. Pillet, Commun. Math. Phys. [**319**]{}, 501 (2013).
D. Sánchez and R. López, Phys. Rev. Lett. [**110**]{}, 026804 (2013). R. López, and D. Sánchez, Phys. Rev. B [**88**]{}, 045129 (2013).
L.W. Molenkamp, Th. Gravier, H. van Houten, O.J.A. Buijk, M.A.A. Mabesoone, and C.T. Foxon, Phys. Rev. Lett. [**68**]{}, 3765 (1992). M.G. Vavilov and A.D. Stone, Phys. Rev. B [**72**]{}, 205107 (2005). D. Nozaki, H. Sevinçli, W. Li, R. Gutiérrez, and G. Cuniberti, Phys. Rev. B [**81**]{}, 235406 (2010). Ph. Jacquod and R.S. Whitney, Europhys. Lett. [**91**]{}, 67009 (2010). K. Saito, G. Benenti, G. Casati, and T. Prosen, Phys. Rev. B [**84**]{}, 201306(R) (2011). K.K. Saha, T. Markussen, K.S. Thygesen, and B.K. Nikolić, Phys. Rev. B [**84**]{}, 041412(R) (2011). D. Sánchez and L. Serra, Phys. Rev. B [**84**]{}, 201307(R) (2011) O. Karlström, H. Linke, G. Karlström, and A. Wacker, Phys. Rev. B [**84**]{}, 113415 (2011). Ph. Jacquod, R.S. Whitney, Jonathan Meair, and M. Büttiker, Phys. Rev. B [**86**]{}, 155118 (2012). S.-Y. Hwang, R. López, M. Lee, and D. Sánchez, Phys. Rev. B [**90**]{}, 115301 (2014) B. Sothmann, R. Sánchez, and A.N. Jordan, EPL [**107**]{} 47003 (2014). J. Matthews, F. Battista, D. Sánchez, P. Samuelsson, and H. Linke, Phys. Rev. B [**90**]{}, 165428 (2014)
M. Galperin, M.A. Ratner, and A. Nitzan, J. Phys.: Condens. Matter [**19**]{}, 103201 (2007). M. Galperin, A. Nitzan, and M.A. Ratner, Mol. Phys. [**106**]{}, 397 (2008). P. Murphy, S. Mukerjee, and J. Moore, Phys. Rev. B [**78**]{}, 161406(R) (2008). S.F. Svensson, E.A. Hoffmann, N. Nakpathomkun, P.M. Wu, H. Xu, H.A. Nilsson, D. Sánchez, V. Kashcheyevs, and H. Linke, New J. Phys. [**15**]{}, 105011 (2013)
F. Battista, F. Haupt, J. Splettstoesser, Phys. Rev. B [**90**]{}, 085418 (2014).
M.A. Sierra, and D. Sánchez, Phys. Rev. B [**90**]{}, 115313 (2014)
The “2” on $P_{\rm gen}^{\rm qb2}$ indicates that this bound is for a quantum system between [*two*]{} reservoirs. We do not have the exact bound for a system between three (or more) reservoirs at different temperatures, however an over-estimate is given in section IX of Ref. \[\].
T. Schmiedl and U. Seifert, EPL, [**81**]{}, 20003 (2008).
M. Büttiker, Phys. Rev. B [**41**]{}, 7906 (1990).
This was an idea proposed by M. Büttiker, private comm.
C. Grenier, private comm.
J. Eom, C.-J. Chien, and V. Chandrasekhar, Phys. Rev. Lett. [**81**]{}, 437 (1998).
J.S. Heron, T. Fournier, N. Mingo, O. Bourgeois, Nano Lett. [**9**]{}, 1861 (2009). J.-S. Heron, C. Bera, T. Fournier, N. Mingo, and O. Bourgeois, Phys. Rev. B 82, 155458 (2010). D.R. Schmidt, R.J. Schoelkopf, and A.N. Cleland, Phys. Rev. Lett. [**93**]{}, 045901 (2004). L.M.A. Pascal, H. Courtois, and F.W.J. Hekking, Phys. Rev. B [**83**]{}, 125113 (2011).
Ref. \[\] does not mention this crucial restriction, because I had overlooked its essential role in the proof.
O. Entin-Wohlman and A. Aharony, Phys. Rev. B [**85**]{}, 085401 (2012).
M. Büttiker, IBM J. Res. Dev. [**32**]{}, 63 (1988).
C.L. Kane and M.P.A. Fisher, Phys. Rev. B [**55**]{}, 15832 (1997).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The aim of this paper is to develop a general theory for the class of skip-free Markov chains on denumerable state space. This encompasses their potential theory via an explicit characterization of their potential kernel expressed in terms of family of fundamental excessive functions, which are defined by means of the theory of Martin boundary. We also describe their fluctuation theory generalizing the celebrated fluctuations identities that were obtained by using the Wiener-Hopf factorization for the specific skip-free random walks. We proceed by resorting to the concept of similarity to identify the class of skip-free Markov chains whose transition operator has only real and simple eigenvalues. We manage to find a set of sufficient and easy-to-check conditions on the one-step transition probability for a Markov chain to belong to this class. We also study several properties of this class including their spectral expansions given in terms of Riesz basis, derive a necessary and sufficient condition for this class to exhibit a separation cutoff, and give a tighter bound on its convergence rate to stationarity than existing results.'
address:
- 'Institute for Data and Decision Analytics, The Chinese University of Hong Kong, Shenzhen, Guangdong, 518172, P.R. China.'
- 'School of Operations Research and Information Engineering, Cornell University, Ithaca, NY 14853.'
author:
- 'Michael C.H. Choi'
- Pierre Patie
bibliography:
- 'skipfree2.bib'
title: 'Skip-free Markov chains'
---
[^1]
[^2]
Introduction {#sec:intro}
============
Let $X=(X_n)_{n\in \mathbb{N}}$ be a Markov chain on the countable state space $E = [[{\mathfrak{l}},{\mathfrak{r}}]] \subseteq \mathbb{Z}$, where we use the notation $[[$ (resp. $]]$) to denote that ${\mathfrak{l}}$ (resp. ${\mathfrak{r}}$) may or may not be in $E$, defined on the filtered probability space $ (\Omega,(\mathcal{F}_n)_{n\in \mathbb{N}},{\mathbb{P}}=({\mathbb{P}})_{x\in {\mathbb{E}}})$. We denote its transition matrix by $P = (p(x,y))_{x,y \in E}$. We assume further that $X$ is irreducible, i.e. for all $x,y \in \dot{E}= ]{\mathfrak{l}},{\mathfrak{r}}[$, $p^{(n)}(x,y)={\mathbb{P}}_x(X_n=y)>0$ for some $n\in {\mathbb{N}}$, and $\textit{upward skip-free}$, i.e. for all $x \in E$, $p(x,x+1)>0$ and $p(x,x+y)=0, y{\geqslant}2$. We denote by ${\mathcal{M}}$ be the set of such upward skip-free Markov chains (or transition operators) on $E$.
The aim of this paper to develop a comprehensive theory for the set ${\mathcal{M}}$, including [*the potential theory, the fluctuation theory*]{} and, resorting to the algebraic concept of similarity, [*the spectral theory*]{}. As a by-product, we also provide, for ergodic chains, a detailed analysis of the speed of convergence to stationarity and investigate the separation cutoff phenomena.
We recall that under the additional condition that $p(x,x-y)=0, y{\geqslant}2$, that is, it is also skip-free to the left, $X \in {\mathcal{M}}$ becomes a birth-death chain. These chains have been and are still the object of intensive and fascinating studies. This probably originates from the seminal work of Karlin and McGregor [@Karlin-McGregor_Spectral], see also Lederman and Reuter [@LR54] and Anderson [@Anderson91], on the diagonalization of their transition operator that provide deep insights into fine distributional properties of these chains. Note that the spectral analysis of these operators has also revealed fascinating links with the theory of orthogonal polynomials and Stieljes moment problem. A review of birth-death chains including their potential theory, is given below in Section \[subs:bd\].
For skip-free Markov chains however, the literature is much more limited than their birth-death counterparts even though skip-free chains appear naturally and frequently in theoretical and applied investigation. On the theoretical side, the (upward) hitting time distribution of (upward) skip-free chains has been studied by [@AW89; @Fill]. Another notable research is the work of Adikahri [@Adikahri86] on excursion theory of such chains. In addition, Mao et al. [@MZZ16] have recently investigated the separation cutoff for skip-free chains. We also mention that Miclo and Patie [@Miclo-Patie18] initiated a new approach to study these chains based on intertwining relationships with some diffusions with jumps. On the other hand, from an application viewpoint, skip-free chains have been widely used in population models, queueing theory and branching processes, see for example [@BGR82; @ASM03; @APW18] and the references therein.
Our contributions to the theory of the class ${\mathcal{M}}$ can be summarized as follows.
a) *Potential theory*: We shall start our program by studying the potential theory of chains in ${\mathcal{M}}$. More specifically, we implement an original approach based on the theory of Martin boundary for Markov chains as developed by Dynkin [@dynkin] to express their $q$-potential kernel. In this vein, we recall that in the specific case of birth-death chains, this $q$-potential is given in terms of the two fundamental solutions (the decreasing and increasing one) of a three-term recurrence equations (discrete analogue of a second order differential equation), see Section \[subs:bd\] for more details regarding this expression. In our context, the situation is more delicate as one has to solve an infinite recurrence equation whose set of solutions does not seem to have been clearly identified in the literature. Although the issue of solving this equation is of algebraic nature, we shall elaborate a strategy based mostly on a combination of techniques from probability theory and potential theory. More specifically, instead of trying to identify directly the convex cone of $q$-excessive functions for the chain $X \in {\mathcal{M}}$, we take an alternative route which consists in characterizing the $q$-potential kernel in terms of the so-called fundamental $q$-excessive (for short F$q$E) functions of three different chains: $(X,{\mathbb{P}})$, $(X,\widehat{{\mathbb{P}}})$ the dual chain as defined in below and $(X,{\mathbb{P}}^{y]}), y \in E$, where $(X,\mathbb{P}^{y]})$ is the Markov chain $(X,{\mathbb{P}})$ killed upon entering the half-line $[[{\mathfrak{l}},y]$, which is plainly an upward skip-free Markov chain on the state space $E^{y]} = (y,{\mathfrak{r}}]]$.
b) *Fluctuation identities*: From this representation of the $q$-potential kernel, we derive the main fluctuations identities for chains in ${\mathcal{M}}$. We recall that the fluctuation theory, which is concerned with the distribution of the first visit of the chain to some (finite or infinite) intervals is of great importance in many applications such as biology, epidemiology and also in ruin theory where the upward skip-free property is required. This theory is well established for the case of birth-death chains (skip-free on both sides), see e.g. Karlin and McGregor [@Karlin-McGregor_Spectral], Keilson [@Ke71]. On the other hand the famous Wiener-Hopf factorization technique has proven to be useful to characterize the law of the first exit times for the class of random walks, that is for Markov chains with stationary and independent increments, see Spitzer [@Spitzer] for further details on these identities. We also mention the interesting work of Fill [@Fill] where he characterizes the upward hitting time distribution of upward skip-free chain via establishing an intertwining with pure-birth chain. Our original approach goes decisively beyond these frameworks as it allows to treat in an unified way, not only the first upward passage time, but also the first downward passage time, including the possibility of undershoot, of the irreducible skip-free Markov chain. To implement our methodology on some specific examples, one merely needs to have access to a (Laplace transform, Fourier transform, moment generating function…) that determines the one-step transition kernel of the chain. We shall illustrate this idea by recovering, in a simple manner, Spitzer’s identities for skip-free random walks, and, by studying the first passage times of branching Galton-Watson processes with immigration, whose details will be provided in a subsequent paper [@Patie-Wang], see also [@APW18]. We also mention that the continuous analogue of these results for skip-free continuous-time Markov processes on the real line has been detailed in [@Patie-Vigon] and applications to generalized Ornstein-Uhlenbeck processes and continuous branching processes with immigration are carried out in [@Lefevre2013] and [@Patie-Wev] respectively.
c) *Spectral theory and its applications*: We also aim at providing some insights into the spectral theory of the (transition operator of) Markov chains that belong to the class ${\mathcal{M}}$. This is a challenging issue as the transition operator $P$ of such a chain is non-self-adjoint (non-reversible) in the weighted Hilbert space $\ell^2(\pi),$ where $\pi$ is the reference measure, implying that there is no spectral theorem available for such bounded linear operator. To overcome this difficulty, we propose an original approach based on the concept of similarity. More specifically, we introduce a subclass of ${\mathcal{M}}$, denoted by ${\mathcal{S}}$, whose each element is related to a (diagonalizable) transition matrix of a birth-death Markov chain via a certain commutation relation. Using this identity, we resort to some techniques from non-harmonic analysis to investigate how to transfer the known spectral information of the reversible birth-death transition operator to the non-reversible one in order to obtain its spectral decomposition. By means of the inverse spectral theorem we manage to show that the class ${\mathcal{S}}$ characterizes completely the family of transition operators in ${\mathcal{M}}$ that have real and distinct eigenvalues. On the other hand, in Theorem \[thm:mcins\], we shall provide a set of sufficient and easy-to-check conditions on the one-step transition probability for a Markov chain to belong to ${\mathcal{S}}$. It is worth mentioning that, to the best of our knowledge, this is the first identification of a class of non-reversible chains with such a spectrum. This is a useful fact which answers an open question raised by Fill [@Fill] on understanding the class of skip-free chains that have real and non-negative eigenvalues.
We believe that this new way of classifying Markov chains based on similarity orbit is powerful enough to tackle many substantial and delicate problems arising in the analysis of such chains. To illustrate this fact, we provide the spectral decomposition of their non-symmetric transition matrices, the law of their first passage times, including the case with possible overshoot. We also study for ergodic chains in ${\mathcal{S}}$ the speed of convergence to equilibrium in both the $\ell^2(\pi)$-topology and the total variation distance. In this line of work, we indicate that Miclo [@Miclo15] has recently introduced a new notion known as Markov similarity and compares the mixing speed of Markov similar generators. As for the speed of convergence to equilibrium, there is a vast literature devoted to this important topic in various settings. For instance, in the case when $P$ is a linear self-adjoint operator in $\ell^2(\pi)$ (reversible), the chain satisfies the spectral gap, which according to Roberts and Rosenthal [@RR], is equivalent to $P$ being geometrically ergodic. In the non-reversible case, to overcome the lack of a spectral theory, many interesting reversibilization techniques have been implemented to obtain a quantitative rate of convergence and we refer the interested readers to Fill [@Fill91] and Saloff-Coste [@LSC97]. We propose an alternative approach based on the concept of similarity which seems to be a natural extension of the spectral gap estimate developed in the reversible case. Moreover, our technique enables us to provide an explicit and a spectral interpretation as a perturbed spectral gap estimate of the (discrete) hypocoercivity phenomena that was introduced by Villani [@V09] for general non-self-adjoint semigroups. We also manage to obtain generalization to the class $\mathcal{S}^M$, a subclass of $\mathcal{S}$ with stochastic monotone time-reversal to be formally introduced in Section \[sec:st\], of the remarkable *spectral gap times mixing time going to infinity* separation cutoff criteria established by Diaconis and Saloff-Coste [@DSC06] for reversible birth-death chains, and recently by Mao et al. [@MZZ16] for continuous-time skip-free chain with stochastic monotone time-reversal.
We point out that these three topics (potential, fluctuation and spectral theory) are intertwined. Indeed, there is of course a fundamental connection between the resolvent operator and the spectral theory of $P$ as the spectrum of $P$ is, by definition, the set of complex numbers $z$ such that the resolvent operator $R_z=(zI - P)^{-1}$ does not exist or is unbounded. Note that since $P$ is a contraction operator, this set is included in the unit disc. Moreover, the development of the fluctuation identities is based on the expression of the $q$-potential kernels whereas the spectral theory allow us to get an explicit representation of the transition kernel (and hence by integration of the $q$-resolvent operator) and, also, of the distribution of the first passage times.
The rest of the paper is organized as follows. In Section \[sec:prelim\], we fix our notations and provide a review of birth-death chains and Martin boundary theory. The definition and study of the fundamental $q$-excessive functions as well as the expression of the $q$-potential kernel are discussed in Section \[sec:pot\_th\]. In Section \[sec:fluc\], we present the fluctuations identities, that is the explicit characterization (via the probability generating function) of the law of the first exit times of skip-free chains, and, following the line of work by Feller [@F66], Kent and Longford [@KL83] and Viskov [@Viskov00], a characterization of these stopping times as discrete infinitely divisible variables in Section \[sec:fpt\]. We proceed in Sections \[sec:st\]-\[sec:sc\] by introducing the subclass of skip-free Markov chains that are similar to a diagonalizable birth-death chain and discuss their spectral properties as well as its applications to the speed of convergence to equilibrium and the study of the cutoff phenomena.
Preliminaries {#sec:prelim}
=============
In this Section, we review some classical concepts on Markov chains that will play a central role throughout the paper. This include some facts of the potential theory and the Martin boundary theory of Markov chains.
Basic facts on Markov chains
----------------------------
We recall that $X$ is said to be *upward skip-free* if the only upward transition is of unit size, yet it can have downward jump of any arbitrary magnitude, i.e. for all $x\in E$ and $y {\geqslant}x+2$, ${\mathbb{P}}_x(X_{1}=y)=0$. We consider an upward skip-free irreducible Markov chain $X = (X_n)_{n \in \mathbb{N}_0}$ on a denumerable state space $E\subseteq \mathbb{Z}$ with left endpoint ${\mathfrak{l}}$ and right endpoint ${\mathfrak{r}}$. We use the convention that ${\mathfrak{r}}\in E$ if the boundary point ${\mathfrak{r}}$ is not absorbing. Otherwise, if ${\mathfrak{r}}$ is absorbing or ${\mathfrak{r}}= \infty$, we say that $X \in {\mathcal{M}}_{\infty}$. We assume through Section 2.1 that $X \in {\mathcal{M}}_{\infty}$ and postpone to Section \[sec:regular\] the study of the Markov chain $X \notin {\mathcal{M}}_{\infty}$. Since $X$ is irreducible, there exists $\pi$ a positive and finite-valued excessive measure for $P$, that is, $\pi P {\leqslant}\pi$, see [@KSK Section $5.2$ and $6.8$], and that will serve as a reference measure. Let $G_q$ be the $q$-potential kernel of $X$ (or its Green function) with respect to the reference measure $\pi$. That is, for $0 < q < 1$, $$\begin{aligned}
\label{eq:defpk}
G_q(x,y) &=& \sum_{n = 0}^\infty q^n \frac{{\mathbb{P}}_x(X_n = y)}{\pi(y)} = \sum_{n = 0}^\infty q^n \frac{p^{(n)}(x,y)}{\pi(y)}, \: x,y \in E.\end{aligned}$$ When $q = 1$, we write $G$, rather than $G_1$, to denote the 1-potential kernel. The $\pi$-dual matrix $\widehat{P} = (\widehat{p}(y,x))_{y,x \in E}$ is defined to be $$\begin{aligned}
\label{eq:dualdef}
\widehat{p}(y,x)\pi(y) &=& p(x,y)\pi(x).\label{def:dual}
\end{aligned}$$ Denote $\zeta$ to be the lifetime of $X$, that is, $\zeta(\omega) = k$ if $k$ is the last time that $X$ is in the state space $E$, and $\zeta(\omega) = \infty$ otherwise.
Let us now define the hitting times associated with $X$. Denote $T_A$ to be the first hitting time of the set A, that is, $$T_A = \inf\{n {\geqslant}0; X_n \in A\},$$ with the usual convention that $\inf \{\emptyset\} = \infty$. If $A = \{a\}$, we write $T_a = T_A$. Similarly, if $A = ({\mathfrak{l}},b]$, we use $T_{b]} = T_A$. Denote the first return time to be $T_a^+ = \inf\{n {\geqslant}1; X_n = a\}$.
We also use, for a real-valued function $f$ and a measure $\mu$ on $E$, the following notation, for any $x,y \in E$, $$\begin{aligned}
P f (x) &= \sum_{y \in E} p(x,y) f(y),\\
\mu P(y) &= \sum_{x \in E} \mu(x) p(x,y).\end{aligned}$$
A review of birth-death chains {#subs:bd}
------------------------------
Let $Y$ be a birth-death chain on $E$ with transition operator $Q \in {\mathcal{B}}$, that is, $Q \in {\mathcal{M}}$ with the additional requirements that $Q(x,x-1)>0$ and $Q(x,x-y)=0$ for $y{\geqslant}2$. $Q$ is a bounded self-adjoint operator in the Hilbert space $\ell^2(\pi_Q)$ where $\pi_Q$ is the reversible measure, i.e. for all $x,y\in E$, $Q(x,y)\pi_Q(x)=Q(y,x)\pi_Q(y)$. The $q$-potential of $Q$, denoted by $U_q$, is well-known to take the form, for any $x,y \in E$, $$\label{eq:gbd}
U_q(x,y)= C_q F_q(x \wedge y) \widehat{F}_q(x \vee y),$$ where $F_q$ (resp. $\widehat{F}_q$) is the unique increasing (resp. decreasing) solution to the equation $q Q F = F$ satisfying appropriate boundary conditions, and $C_q$ is the inverse of their Wronskrian. Note that this equation boils down to a three-term recurrence equations, see [@DM76 Ex. 5.3 p.150 and Section 5.4] for details regarding this expression. This is reminiscent of the expression of the potential kernel of the continuous time-space analogue, whose generator is a second order differential operator. We recall that a systematic and thorough study of one-dimensional diffusion has been undertaken by Feller [@Fe54]. Moreover, the moment generating function of the first hitting time $T^Y_a=\inf\{ n{\geqslant}0;\: Y_n=a\}$ of $Y$ to a fixed level $a \in E$ is given by $$\label{eq:gbd}
\mathbb{E}_x[q^{T^Y_a}]= \frac{F_q(x)}{F_q(a)}{\mathds{1}}_{\{x{\leqslant}a\}} +\frac{\widehat{F}_q(x)}{\widehat{F}_q(a)}{\mathds{1}}_{\{x> a\}}.$$ Let us now describe a link between this expression and the diagonalization of the operator $Q$ in ${\mathcal{B}}$. We assume, for sake of simplicity, that $\mathfrak{l}=0$ and thus $E$ is countable subset of $\mathbb{N}$. It is worth mentioning that the similarity transform $D_{\sqrt{\pi}} Q D_{\frac{1}{\sqrt{\pi}}}$, where $D_a$ is the diagonal matrix of $a$ yields to a symmetric tridiagonal matrix which belongs to the well-studied class of Jacobi matrices. We also recall that a function $\phi$ is a Pick function if $\phi$ admits an analytical continuation to the cut complex plane $\mathbb{C}\setminus [0,\infty)$ such that $\Im \phi(z)\Im(z){\geqslant}0$, and we denote by $\mathcal{P}$ the subset of such Pick functions that admit the representation, for $\Im(z)>0$, $$\label{eq:dp}
\phi(z) = \int_{-1}^{1}\frac{d\Delta(r)}{r-z},$$ where $\Delta$ is a probability measure on $[-1,1]$. From the work of Karlin and McGregor [@Karlin-McGregor_Spectral], we know that there exists a spectral mapping $\mathrm{K}: {\mathcal{B}}\rightarrow \mathcal{P}$ which is one-to-one and $\Delta$ is the spectral measure of $Q$. More specifically, we have for any $f \in \ell^2(\pi_Q)$ and $n\in \mathbb{N}$, $Q^n$ can be diagonalizable as follows $$\label{eq:sbd}
Q^n f= \int_{-1}^{1} r^n \langle f , \mathrm{F}_r \rangle_{\pi_Q} \mathrm{F} _r \: d\Delta(r),$$ where for $r \in supp(\Delta)$, $Q\mathrm{F}_r=r\mathrm{F}_r$. Note that $S(Q)$ the spectrum of $Q$ is such that $S(Q)=supp(\Delta)\subset [-||Q||,||Q||] \subseteq [-1,1]$. Another remarkable and deep result, which was conjectured by Krein, is the onto property of the spectral map $\mathrm{K}$, see [@Gorbachuk1997 Appendix 3] for an historical account. Indeed for any $\phi \in \mathcal{P}$, i.e. a Pick function of the form , there exists a unique $Q \in {\mathcal{B}}$ on $E$ with $\mathfrak{l}$ a non-killing boundary with a spectral representation of the form . The first proof of this inverse spectral result was given in the monograph by H. Dym and H.P. McKean [@DM76 Chapter 6]. It uses the theory of Hilbert spaces of entire functions, and a deep uniqueness theorem due to de Branges. Note also that such a spectral representation reveals that when the spectrum of $Q$ is composed of isolated eigenvalues then they are necessarily simple (see e.g. [@DM76 Section $5.8$]).
Martin boundary theory of denumerable Markov chains
---------------------------------------------------
In this subsection, we review some essential results of Martin boundary theory of denumerable irreducible Markov chains based on [@dynkin; @KSK; @Woess]. First, we recall the definition of $q$-excessive, $q$-harmonic and $q$-potential functions. A non-negative function $f$ on $E$ is $q$-excessive (resp. $q$-harmonic) if $q P f (x) {\leqslant}f(x)\,(\mathrm{resp.}~q P f (x) = f(x))$ for all $x \in E$, where $0 < q {\leqslant}1$. A non-negative function $f$ on $E$ is a $q$-potential if $f$ is $q$-excessive and $\lim_{n \to \infty} (qP)^n f(x) = 0$ for all $x \in E$. We say that $f$ is harmonic (resp. excessive) if $f$ is $1$-harmonic (resp. $1$-excessive). We write $$\mathcal{E}_q = \{f:E\to\mathbb{R}^+;~qPf {\leqslant}f\} \, \mathrm{(resp.}~\mathcal{H}_q, \mathcal{P}_q)$$ for the set of $q$-excessive (resp. $q$-harmonic, $q$-potential) functions on $E$. We simply write $\mathcal{E}$ (resp. $\mathcal{H}$, $\mathcal{P}$) to denote the set of excessive (resp. harmonic, potential) functions. Note that the irreducibility property implies that $f \in \mathcal{E}_q$ is positive, since if there exists $y \in E$ such that $f(y) > 0$, then by irreducibility there exists $n$ such that for $x \in E$ $f(x) {\geqslant}q p^n(x,y) f(y) > 0$. We will also use the notation $\widehat{\mathcal{E}}_q$ (resp. $\widehat{\mathcal{H}}_q, \widehat{\mathcal{P}}_q$) to denote the set of $q$-excessive (resp. $q$-harmonic, $q$-potential) functions associated with $\widehat{P}$, which was defined in . We point out that if $\hat{h} \in \widehat{\mathcal{E}}_q$, then $\hat{h} \pi$ is a $q$-excessive measure for $P$ in the sense that $q \: \hat{h} \pi P f{\leqslant}\hat{h} \pi f$. Another commonly used terminology for excessive (resp. harmonic, potential) function is superharmonic (resp. invariant, purely-excessive) function. We further recall the definition of minimal $q$-harmonic function. A non-zero function $f$ on $E$ is minimal $q$-excessive if $f = f_1 + f_2$ implies $f = c_i f_i$ for $i = 1,2$, where $f_1, f_2 \in \mathcal{E}_q$, $c_1$ and $c_2$ are constants and $0 < q {\leqslant}1$. We say that $f$ is minimal excessive if $f$ is minimal $1$-excessive. We write $\mathcal{E}_q^{min}$ (resp. $\mathcal{H}_q^{min}$) to be the set of minimal $q$-excessive (resp. minimal $q$-harmonic) functions on $E$. For further discussion on minimal excessive and minimal harmonic functions, we refer the interested readers to [@Woess Chapter $7$] and [@dynkin Section $14$]. Next, we state the classical Riesz representation theorem (see e.g. [@dynkin Theorem $6$]), which gives an unique decomposition of excessive function.
\[thm:riesz\] For $0 < q {\leqslant}1$, every $q$-excessive function can be written uniquely as the sum of a $q$-potential and a $q$-harmonic function. That is, if $f \in \mathcal{E}_q$, then $$f(x) = \sum_{y \in E} G_q(x,y) k_q(y) \pi(y) + h_q(x),$$ where $k_q(x) = f(x) - qPf(x)$ and $h_q(x) = \lim_{n \to \infty} (qP)^n f(x) \in \mathcal{H}_q$.
Recall that $X \in \mathcal{M}_{\infty}$ with transition matrix $P$ and reference measure $\pi$ on a denumerable state space $E$. For a measure $\mu$ on $E$, define $E_{\mu} = \{y \in E;~ \mu G_q(y) > 0\}$, where $\mu G_q(y) = \sum_{x \in E} \mu(x) G_q(x,y)$, for a $0 < q <1$. We say that the measure $\mu$ is a *standard measure* if $E_{\mu} = E$. For $x,y \in E$ and $0 < q <1$ (with also $q=1$ when $(X,{\mathbb{P}})$ is a transient), the $q$-Martin kernel associated to a standard measure $\mu$ is defined to be $$\begin{aligned}
\label{eq:defmk}
{}_\mu K_{q}(x,y) &= \dfrac{G_q(x,y)}{\mu G_q(y)},\end{aligned}$$ where the denominator is positive since $\mu$ is standard. Note that ${}_\mu K_{q}(x,y)$ is finite for any $x,y \in E$. Next, define a metric ${}_\mu d_{q}$ on the space $E$ by $$\begin{aligned}
{}_\mu d_{q}(y,z) = \sum_{x \in E} w_x \left(|{}_\mu K_{q}(x,y) - {}_\mu K_{q}(x,z)| + |{\mathds{1}}_{\{x = y\}} - {\mathds{1}}_{\{x = z\}}| \right),\end{aligned}$$ where the weights $(w_x)_{x \in E}$, with $w_x > 0$, are chosen such that $\sum_{x \in E} w_x < \infty$. We can obtain the completion of $E$ with respect to the metric ${}_\mu d_{q}$, namely ${\overline{E}}$, and the boundary of $E$ in ${\overline{E}}$ is denoted as $\partial E = {\overline{E}}- E$. ${\overline{E}}$ is the Martin compactification of $E$ and $\partial E$ is the Martin boundary. The set $$\partial_{{\mathbb{P}}} E = \{y \in \partial E; \: {}_{\mu}K_q(x,y) \, \text{is minimal } q \text{-harmonic in}\, x\}$$ is known as the *minimal Martin boundary*. When there is no ambiguity in the probability measure, we write $\partial_m E = \partial_{{\mathbb{P}}} E$. The inclusion of the indicator terms ${\mathds{1}}_{\{x = y\}}$ and ${\mathds{1}}_{\{x = z\}}$ in the metric ${}_\mu d_{q}$ ensures that $E$ is an open set in the Martin compactification ${\overline{E}}$, and the Martin boundary $\partial E$ is closed. Next, fix a reference point $\mathfrak{o} < {\mathfrak{r}}- 1$, and we write the $q$-Martin kernel $_{\delta_{\mathfrak{o}}}K_q(x,y)$ as $$\begin{aligned}
_{\delta_{\mathfrak{o}}} K_q(x,y) &= \dfrac{\mu G_q (y)}{G_q(\mathfrak{o},y)} {}_{\mu} K_{q}(x,y) = \dfrac{_{\mu} K_{q}(x,y)}{_{\mu} K_q(\mathfrak{o},y)}.\end{aligned}$$ Similarly, we have $$\begin{aligned}
_{\mu} K_q(x,y) &= \dfrac{G_q(\mathfrak{o},y)}{\mu G_q (y)} {}_{\delta_{\mathfrak{o}}} K_q(x,y) = \dfrac{{}_{\delta_{\mathfrak{o}}} K_q(x,y)}{{}_\mu K_q(y)}.\end{aligned}$$ A sequence $(y_n)$ is a Cauchy sequence in the metric space $(E,{}_\mu d_{q})$ if and only if $(_{\mu} K_{q}(x,y_n))$ is a Cauchy sequence of real numbers for every $x$ if and only if $(_{\delta_{\mathfrak{o}}} K_q(x,y_n))$ is a Cauchy sequence of real numbers for every $x$ if and only if $(y_n)$ is a Cauchy sequence in the metric space $(E,{}_{\delta_{\mathfrak{o}}}d_q)$. Thus, up to homeomorphism, ${\overline{E}}$ is independent of the choice of the initial measure and of the reference point $\mathfrak{o}$. It can also be shown that ${\overline{E}}$ is independent of the choice of the weights $(w_x)_{x \in E}$ (see e.g. Proposition 10.13 in [@KSK]). From now on, we fix the reference point $\mathfrak{o}$ and write for all $x,y \in E$, $$\begin{aligned}
\label{def:Kqxy}
K_q(x,y) &= {}_{\delta_{{\mathfrak{o}}}} K_q(x,y) = \dfrac{G_q(x,y)}{G_q({\mathfrak{o}},y)}.\end{aligned}$$ Let $\Omega_{\infty} = \{\omega; \: \text{there is}\, x_\infty \in \partial E \, \text{such that} \, x_n \rightarrow x_\infty \, \text{in the Martin topology} \}.$ $\Omega_{\infty}$ can be interpreted as the set of *non-terminating* trajectories of $X$ that converges to $\partial E$. If $(X,P)$ is transient and $P$ is a stochastic matrix then there is a random variable $X_{\infty}$ taking values in $\partial E$ such that for each $x \in E$, ${\mathbb{P}}_x \left(\lim_{n \rightarrow \infty} X_n = X_{\infty} \right) = 1.$ In terms of trajectory space, we have ${\mathbb{P}}_x \left(\Omega_{\infty}\right) = 1.$ If $P$ is strictly substochastic at some states, we should extend the trajectories to $E \cup \{\nabla\}$, where $\nabla$ is the graveyard state. Define $$\Omega_{\nabla} = \{\omega \in \Omega;\: \, \text{there is} \, k {\geqslant}1 \, \text{such that} \, x_n \in E, \, \forall n {\leqslant}k \, \text{and} \, x_n = \nabla, \forall n {\geqslant}k + 1 \}.$$ $\Omega_{\nabla}$ is the set of trajectories that eventually reach $\nabla$. Denote $\zeta$ to be the lifetime of $X$, that is, $\zeta(\omega) = \mathfrak{K}$ for $\omega \in \Omega_{\nabla}$, where $\mathfrak{K}$ is the last time that $X$ is in the state space $E$ (as defined in $\Omega_{\nabla}$), and $\zeta(\omega) = \infty$ otherwise. Define $\Omega_{\zeta} = \Omega_{\nabla} \cup \Omega_{\infty}$. If $(X,{\mathbb{P}})$ is transient then there is a random variable $X_{\zeta}$ taking values in ${\overline{E}}$ such that for each $x \in E$, ${\mathbb{P}}_x \left(\lim_{n \rightarrow \zeta} X_n = X_{\zeta} \right) = 1.$ In terms of trajectory space, this means that ${\mathbb{P}}_x \left(\Omega_{\zeta}\right) = 1.$
Next, suppose now that $h \in \mathcal{E}_q$. The Doob transform or $h$-process of $X$ is defined to be the Markov chain on $E^h = \{y \in E;~ h(y) > 0 \}= E$, by irreducibility, with the canonical measure ${\mathbb{P}}^h_x$ such that ${\mathbb{P}}^h_x(X_1 = y) = \dfrac{p(x,y)h(y)}{h(x)}.$ Recalling that ${\mathfrak{o}}$ is the fixed reference point, the $q$-potential and $q$-Martin kernels associated with the $h$-process take respectively the form $$\begin{aligned}
G^h_q(x,y) &= \dfrac{G_q(x,y)h(y)}{h(x)}, \\
K^h_q(x,y) &= \dfrac{K_q(x,y)h({\mathfrak{o}})}{h(x)}. \label{eq:Kh}\end{aligned}$$ From and the definition of the metric $_{\mu} d_{q}$, we observe that the Martin compactification ${\overline{E}}$ is homeomorphic to the Martin compactification of the $h$-process. Next, we state the following which are the main claims of Theorem $6$ and $7$ in [@dynkin].
\[thm:uniq\] \[thm:htransp\] Let $q \in (0,1]$. If $h \in \mathcal{E}_q$ such that $h({\mathfrak{o}})=1$ then $h$ has a unique representation of the form $$h(x) = \int_{E \cup \partial_m E} K_q(x,y) \, \mu_h(dy) = K_q\mu_h(x),$$ where $\mu_h(\cdot) = {\mathbb{P}}_1^h(X_{\zeta} \in \cdot)$ defines a probability measure. Conversely, for any finite measure $\mu$, the mapping $x \mapsto \int_{E \cup \partial_m E} K_q(x,y) \, d \mu(y)$ defines an $q$-excessive function, which is $q$-harmonic if and only if $\mu(E) = 0$. Finally, for all $y \in E \cup \partial_m E$, let $h^y(\cdot) = K(\cdot,y)$. Then $y \in E \cup \partial_m E$ if and only if ${\mathbb{P}}^{h^y}(X_{\zeta} = y) = 1$. Moreover we have ${\mathbb{P}}^{h^y}(\zeta = \infty) = 1$ if and only if $y \in \partial_m E$.
The previous claim means that $X$ is forced to terminate at the point $y \in E \cup \partial_m E$ $\,$ ${\mathbb{P}}_x^{h^y}$-almost surely.
\[thm:minE\] We have $\mathcal{E}_q^{min}=\{h_q; \: h_q(x) = C K_q(x,y), C>0 \textrm{ and } y \in E \cup \partial_m E\}$.
Finally, we recall the following useful result whose proof follows readily from [@CW05 Theorem $11.9$].
\[lem:CW\] Suppose that $h_q \in \mathcal{E}_q$, and $T$ is a stopping time with respect to $(\mathcal{F}_n)_{n {\geqslant}1}$, the sigma field generated by $(X_n)$. For $x \in E^{h_q}$, $${\mathbb{P}}_x^{h_q} (T < \infty) = \dfrac{1}{h_q(x)}{\mathbb{E}}_x \left[ q^T h_q(X_T) {\mathds{1}}_{\{T < \infty\}}\right].$$
Potential Theory {#sec:pot_th}
================
In this Section, we provide an explicit representation of the $q$-potential kernel $G_q$, as defined in , of an upward skip-free Markov chain $X \in {\mathcal{M}}$. We recall that in the simpler case when $X$ is a birth-death Markov chain, i.e. skip-free in both directions, then its $q$-potential kernel takes the form $$\label{eq:defpbd}
U_q(x,y)= C_q F_q(x \wedge y) \widehat{F}_q(x \vee y),$$ where $F_q$ and $\widehat{F}_q$ are the two fundamental solutions of a three-term recurrence equations (discrete analogue of a second order differential equation), see Section \[subs:bd\] for more details regarding this expression. In our context, the situation is more delicate as one has to solve an infinite recurrence equation whose set of solutions does not seem to have been clearly identified in the literature. Although the issue of solving this equation is of algebraic nature, we shall elaborate a strategy based mostly on a combination of techniques from probability theory and potential theory. We start by expressing the $q$-potential kernel in terms of the so-called fundamental $q$-excessive (for short F$q$E) functions of the following three processes: $(X,{\mathbb{P}})$, $(X,\widehat{{\mathbb{P}}})$ and $(X,{\mathbb{P}}^{y]})$, where $(X,\mathbb{P}^{y]})$ is the Markov chain $(X,{\mathbb{P}})$ killed upon entering the half-line $[[{\mathfrak{l}},y]$, which is plainly an upward skip-free Markov chain on the state space $E^{y]} = (y,{\mathfrak{r}}]]$, with transition kernel denoted by $P^{y]}$. We are now ready to state the main result of this Section.
\[thm:green2\] Suppose that $X \in \mathcal{M}_{\infty}$.
1. \[it:mb\] \[it:ph\] Writing, for any $x\in E$ and $0<q<1$, $$H_q(x) = K_q\delta_{\mathfrak{r}}(x),$$ (resp. $\widehat{H}_q(x) =
\widehat{K}_q\delta_{\mathfrak{l}}(x)$) with $\delta_{{\mathfrak{r}}}$ is the Dirac mass at ${\mathfrak{r}}$ and $K_q$ defined in , we have, for all $0 < q < 1$, that $$H_q \in \mathcal{E}_q^{min}$$ (resp. $\widehat{H}_q \in \widehat{\mathcal{E}}_q^{min}$) and it is the unique minimal increasing (resp. decreasing) $q$-excessive for $P$ (resp. $\widehat{P}$) such that $H_q({\mathfrak{o}}) = 1$ (resp. $\widehat{H}_q({\mathfrak{o}})=1$). Moreover, if $X \in {\mathcal{M}}_{\infty}$ (resp. $X \in {\mathcal{M}}\setminus {\mathcal{M}}_{\infty}$) then $H_q $ is the unique increasing function in $\mathcal{H}_q$ (resp. $H_q \in \mathcal{P}_q$ with $H_q({\mathfrak{r}})<\infty$).
2. \[it:phk\] For any $y<{\mathfrak{r}}$, $0<\kappa_q^{y]}=\lim\limits_{x\to {\mathfrak{r}}} \frac{K_q\delta_{\mathfrak{r}}(x)}{K^{y]}_q\delta_{\mathfrak{r}}(x)} <\infty $. Then the function ${\mathbf{H}_q^{y]}}$ defined, for any $x \in E^{y]}$, by $${\mathbf{H}_q^{y]}}(x) = \kappa_q^{y]} K^{y]}_q\delta_{\mathfrak{r}}(x)$$ has (with respect to $P^{y]}$) the same property as $H_q$ but with the normalization $$\lim\limits_{x\to {\mathfrak{r}}} \frac{{\mathbf{H}_q^{y]}}(x)}{H_q(x)}=1$$ (recall that by convention ${\mathbf{H}_q^{y]}}(x) =0$ for any $x {\leqslant}y$).
3. \[it:poh\]Finally, set $C_q = G_q({\mathfrak{o}},{\mathfrak{o}})\,>0$. Then, we have, for all $x,y \in E,$ $$\label{eq:bargxysum}
G_q(x,y)=
C_q \: \widehat{H}_q(y)\left(H_q(x) - {\mathbf{H}_q^{y]}}(x)\right).$$
Up to minor modifications all statements hold for $q=1$ when $(X,{\mathbb{P}})$ is transient.
The terminology F$q$E comes from the birth-death case where these functions are usually called the fundamental solutions of the associated three-term recurrence equation and are $q$-excessive.
Recall, from , that the $q$-potential of a birth-death chain is given in terms of two excessive functions. Each of them is associated to one point in the Martin boundary which is, due to its skip-free property in both directions, reduced to two points. In our situation, due to the random size of downwards jumps, the description of the Martin boundary below is more delicate. To overcome this difficulty, we start by taking advantage of the upward skip-free property to identify, as in the case of birth-death chain, the $q$-excessive function $H_q$ attached to the unique point above of the Martin boundary. Then, we use the dual chain which is downward skip-free to identify, in the dual way, the $q$-excessive function $\widehat{H}_q$ associated to the unique point below of the Martin boundary. Finally, instead of trying to describe the Martin boundary below for the original chain, we introduce the Markov chain killed at the first passage time at a point, say $y$, below the starting point. This latter being also an upward skip-free Markov chain, we identify, in a similar way than for the original chain, its $q$-excessive function ${\mathbf{H}_q^{y]}}$ associated to the point above of the Martin boundary. It turns out that these three $q$-excessive functions coming from three different chains enable us to characterize the $q$-potential kernel at a fixed point for the original chain.
\[rk:pk\] The expression of the $q$-potential is expressed in terms of the function $H_q$ which itself is defined in terms of the Martin kernel, which is a normalized version of the $q$-potential. Although this may sound a bit awkward, this representation is very useful when applied to specific (and solvable) instances. Indeed, in such case, the functions $H_q$ can be computed by either solving the equation $q P f = f$ which admits an unique increasing solution when $H_q \in \mathcal{H}_q$, see Section \[sec:sfrw\] for an application of this idea to the class of skip-free random walks. Otherwise, in the other case, i.e. $H_q \notin \mathcal{H}_q$ but is purely $q$-excessive, its computation can be realized through some asymptotic analysis carried out on some transforms determining the transition kernel, such as the moment generating functions. This latter approach has been exploited in [@APW18] to compute the corresponding function $H_q$ for (continuous-time) branching process, see also [@Patie-Wang] for the study of the potential and fluctuation theory of these chains including immigration. We also refer to Remark \[rk:ft\] for further discussion on how to use our approach to obtain the fluctuation identities of skip-free Markov chains.
Additional properties of $H_q$ and its relation with infinite divisibility are studied in Section \[sec:fpt\] Corollary \[it:id\].
We proceed with the proof of these statements which is split into several intermediate results.
Proof of Theorem \[thm:green2\]
-------------------------------
We start with the following result which relates the Martin kernel to the hitting time distribution.
\[thm:hit\] For any $x,y \in E$, $${\mathbb{E}}_x(q^{T_y}) = \dfrac{K_q\delta_y(x)}{K_q\delta_y(y)}.$$
By Theorem \[thm:riesz\], for any $y\in E$, $K_q\delta_y \in \mathcal{P}_q$ which leads, by Theorem \[thm:htransp\], to ${\mathbb{P}}^{K_q\delta_y}(X_{\zeta} = y, \zeta < \infty) = 1$, that is, for any $x,y\in E$, $${\mathbb{P}}_x^{K_q\delta_y}(T_y < +\infty) = 1.$$ Since, on the other hand, by Lemma \[lem:CW\], we have, for any $x,y\in E$, $${\mathbb{P}}_x^{K_q\delta_y}(T_y < +\infty)= {\mathbb{E}}_x(q^{T_y})\dfrac{K_q\delta_y(y)}{K_q\delta_y(x)}$$ we complete the proof.
### Proof of Theorem \[thm:green2\]
Suppose that $x \vee {\mathfrak{o}}{\leqslant}y {\leqslant}a $, where $x \vee {\mathfrak{o}}= \max \{x,{\mathfrak{o}}\}$. By means of Lemma \[thm:hit\], the upward skip-free property and the strong Markov property, we obtain that $$\begin{aligned}
\label{eq:ratiogenTaK}
K_q \delta_a(x)=\frac{\mathbb{E}_x(q^{T_a})}{\mathbb{E}_{\mathfrak{o}}(q^{T_a})} &=\frac{\mathbb{E}_x(q^{T_y})\mathbb{E}_y(q^{T_a})}{\mathbb{E}_{\mathfrak{o}}(q^{T_y})\mathbb{E}_y(q^{T_a})}=
K_q \delta_y(x).
\end{aligned}$$ Thus, for any $y{\geqslant}x \vee {\mathfrak{o}},\: K_q \delta_y(x)=K_q (x, x \vee {\mathfrak{o}})$. Hence, one can trivially define the function $H_q$ as the extended Martin kernel at ${\mathfrak{r}}$, that is, for $x \in E$, $$H_q(x) = \lim_{y \to {\mathfrak{r}}} K_q(x,y)= \int_{E \cup \partial_m E} K_q(x,y) \delta_{{\mathfrak{r}}}(dy).$$ Hence if $X \in {\mathcal{M}}_{\infty}$ (resp. $X \in {\mathcal{M}}\setminus {\mathcal{M}}_{\infty}$) then ${\mathfrak{r}}\in \partial_{{\mathbb{P}}} E $ (resp. ${\mathfrak{r}}\in E$) and thus by theorems \[thm:uniq\] and \[thm:minE\] (resp. and Theorem \[thm:riesz\]), we get that $H_q \in \mathcal{H}_q \cap \mathcal{E}_q^{min}$ (resp. $H_q \in \mathcal{P}_q \cap \mathcal{E}_q^{min}$). Next note, from the first identity in , that $H_q({\mathfrak{o}}) = 1$, and, for $x {\leqslant}y$, $H_q(x)=K_q \delta_y(x)$. Hence, by Lemma \[thm:hit\] we have, for any $x {\leqslant}y$, $$\label{eq:uphit}
{\mathbb{E}}_x(q^{T_y}) = \dfrac{H_q(x)}{H_q(y)},$$ which entails, by the irreducibility of $X$, that $H_q$ is positive everywhere since for any $x \in E$, the ratio $H_q(x)=\frac{\mathbb{E}_x(q^{T_a})}{\mathbb{E}_{\mathfrak{o}}(q^{T_a})} > 0$. To see that the mapping $x \mapsto H_q(x)$ is increasing, one observes from again the strong Markov property and the upward skip-free property that (recall that $x {\leqslant}y {\leqslant}a$) $$H_q(x) = \frac{{\mathbb{E}}_x(q^{T_y}) {\mathbb{E}}_y(q^{T_a})}{{\mathbb{E}}_{\mathfrak{o}}(q^{T_a})} = {\mathbb{E}}_x(q^{T_y}) H_q(y) < H_q(y).$$ To prove the uniqueness, we proceed by contradiction and thus assume that there exists a positive function $h_q \in \mathcal{E}_q^{min}$ (resp. in $\mathcal{H}_q$ when $X\in {\mathcal{M}}_{\infty}$) which differs from $H_q$ and which is also an increasing function on $E$. Then, according to Theorem \[thm:minE\], there exists $y_0 \in E$ (resp. $y_0={\mathfrak{l}}$ or $y_0={\mathfrak{r}}$) such that for all $x\in E$, $h_q(x)=\frac{K_q(x,y_0)}{K_q(o,y_0)}$. Thus, on the one hand, from Lemma \[thm:hit\] we deduce that for any $x{\leqslant}y_0$, $${\mathbb{E}}_x(q^{T_y}) = \dfrac{K_q\delta_{y_0}(x)}{K_q\delta_{y_0}(y_0)}=\frac{h_q(x)}{h_q(y_0)}.$$ This combines with yield that $h_q(x)=H_q(x)$ for any $x{\leqslant}y_0$, which proves the claim when $H_q\in \mathcal{H}_q$ and $y_0={\mathfrak{r}}$. In the other cases, choose $x> y_0$ such that $h_q(x)\neq H_q(x)$. Then, observe from Theorem \[thm:uniq\] that $ {\mathbb{P}}_x^{h_q}(T_{y_0} < +\infty) =1$. As $ {\mathbb{P}}_x^{h_q}(T_{y_0} < +\infty)=\frac{h_q(y_0)}{h_q(x)}{\mathbb{E}}_x(q^{T_{y_0}})$ and $h_q$ is increasing and $x>y_0$, we get that ${\mathbb{E}}_x(q^{T_{y_0}})>1$ which is impossible. This completes the uniqueness property of $H_q$. To complete the proof of Theorem \[thm:green2\], we use similar arguments for deriving the stated properties of $\widehat{H}_q$ after recalling that the dual chain $(X,\widehat{{\mathbb{P}}})$ is downward skip-free.
### Proof of Theorem \[thm:green2\]
We start with the following claim which is a straightforward reformulation of Theorem \[thm:green2\] for the killed chains.
\[lem:minEb\] Let $b \in E$ and choose a reference point ${\mathfrak{o}}^{b]} \in E^{b]}$. Then, for all $0 < q <1$, the function $H_q^{b]}(x) = K^{b]}_q\delta_{\mathfrak{r}}(x)$ defined on $E^{b]}$ is positive on $E^{b]}$, minimal, increasing $q$-harmonic for $P^{b]}$ with $H_q^{b]}({\mathfrak{o}}^{b]}) = 1$. Moreover, for any $b{\leqslant}x {\leqslant}a$, $$\mathbb{E}_x^{b]}(q^{T_a}) = \mathbb{E}_x(q^{T_a} {\mathds{1}}_{\{T_a < T_{b]}\}}) = \dfrac{H_q^{b]}(x)}{H_q^{b]}(a)}.$$
Under ${\mathbb{P}}^{b]}$, $X$ is an upward skip-free Markov chain on $E^{b]}$, the results follows from Theorem \[thm:green2\] and the identity .
The following lemma expresses the pgf of the *downward* hitting times $(T_b, {\mathbb{P}}^{[a}_x)$, where $a> x > b$, in terms of the F$q$E functions of $(X,{\mathbb{P}})$ and $(X,{\mathbb{P}}^{b]})$.
\[lem: below\] For any $b < x < a$ and $0<q<1$, we have $$\mathbb{E}_x(q^{T_b} {\mathds{1}}_{\{T_b < T_a\}}) = \dfrac{H_q(x)}{H_q(b)} - \dfrac{H_q(a)}{H_q(b)} \dfrac{H_q^{b]}(x)}{H_q^{b]}(a)}.$$
Since by definition $H_q = K_q\delta_{\mathfrak{r}}$, we have, from Theorem \[thm:htransp\] that, $$\label{eq:h1}
{\mathbb{P}}^{H_q}(X_{\zeta} = {\mathfrak{r}}) = 1.$$ This combines with the upward skip-free property, see Lemma \[lem:CW\], yields that under $\mathbb{P}_x^{H_q}$ the sample paths of $X$ that drop below $b$ before hitting $a$ must reaches $b$ before reaching $a$. That is $$\label{eq:trick}
\mathbb{P}_x^{H_q}(T_b < T_a)=\mathbb{P}_x^{H_q}(T_{b]} < T_a) = 1 - \mathbb{P}_x^{H_q}(T_a < T_{b]}),$$ where we used again for the second identity. Hence, an application of Lemma \[lem:CW\] gives $$\begin{aligned}
\mathbb{P}_x^{H_q}(T_b < T_a)=\dfrac{H_q(b)}{H_q(x)} \mathbb{E}_x(q^{T_b} {\mathds{1}}_{\{T_b < T_a\}}) &= 1 - \dfrac{H_q(a)}{H_q(x)} \mathbb{E}_x(q^{T_a}{\mathds{1}}_{\{T_a < T_{b]}\}}) = 1 - \dfrac{H_q(a)}{H_q(x)}\dfrac{H_q^{b]}(x)}{H_q^{b]}(a)},
\end{aligned}$$ where the last equality follows from Lemma \[lem:minEb\]. Rearranging the terms provides the desired result.
The proof of Theorem \[thm:green2\] follows readily after the following claim.
\[lem:Kqb\] For $x \in E^{b]}$, define $$\kappa_q^{b]}(x) = \dfrac{H_q(x)}{H_q^{b]}(x)}.$$ Then the mapping $x \mapsto \kappa_q^{b]}(x)$ is non-increasing on $E^{b]}$ with $0 < \kappa_q^{b]}(x) < \infty$. Furthermore, $0<\kappa_q^{b]} = \lim_{x \to {\mathfrak{r}}} \kappa_q^{b]}(x) < \infty.$
It is clear that, for all $x \in E^{b]}$, $0 < \kappa_q^{b]}(x) < \infty$, since both $H_q$ and $H_q^{b]}$ are positive and finite on $E^{b]}$. Next, for any $x \in E^{b]}$, $$qP^{b]}H_q(x) {\leqslant}qPH_q(x) {\leqslant}H_q(x),$$ where the first inequality follows from the fact that $P^{b]}$ is the restriction of $P$ to $E^{b]}$, and we use that $H_q \in \mathcal{E}_q$ in the second inequality. Therefore, $H_q$ (restricted on $E^{b]}$) is $q$-excessive for $P^{b]}$. Thus, one may define the $H_q$-transform of $qP^{b]}$ by $$^{H_q}P^{b]}(x,y) = \dfrac{H_q(y)}{H_q(x)} q P^{b]}(x,y),$$ where $x,y \in \{x \in E^{b]}:\: H_q(x) > 0\} = E^{b]}$ by Theorem \[thm:hit\]. Using Lemma \[lem:CW\] and Lemma \[lem:minEb\], we have, for any $x {\leqslant}a$, $$\begin{aligned}
\label{eq:HqPb}
^{H_q}{\mathbb{P}}_x^{b]}(T_a < \infty) &= \dfrac{H_q(a)}{H_q(x)} \mathbb{E}_x^{b]}(q^{T_a}) = \dfrac{H_q^{b]}(x)}{H_q(x)} \dfrac{H_q(a)}{H_q^{b]}(a)} = \dfrac{\kappa_q^{b]}(a)}{\kappa_q^{b]}(x)} .
\end{aligned}$$ Since $(X,^{H_q}{\mathbb{P}}_x^{b]})$ is a transient upward skip-free Markov chain, using Lemma \[thm:hit\] and Theorem \[thm:green2\] for $q=1$, one easily deduces, with the obvious notation, that $\frac{1}{C \kappa_q^{b]}(x)}= \:\: \:^{H_q}K^{b]}\delta_{\mathfrak{r}}(x), $ for some $C>0$. Thus the mapping $x \mapsto \kappa_q^{b]}(x)$ is non-increasing. Henceforth $\kappa_q^{b]} = \lim_{x \to {\mathfrak{r}}} \kappa_q^{b]}(x) {\leqslant}\kappa_q^{b]}(b+1) < \infty$. Observing that both $H_q({\mathfrak{r}}) $ and $H_q^{b]}({\mathfrak{r}})$ are finite when $X \in {\mathcal{M}}\setminus {\mathcal{M}}_{\infty}$, we readily get that $\kappa_q^{b]}>0$ which completes the proof in this case. It remains to show that $$\begin{aligned}
\label{eq:Kqbfinite}
\lim_{a \rightarrow {\mathfrak{r}}} \dfrac{\kappa_q^{b]}(a)}{\kappa_q^{b]}(x)} &= {\lim_{a \rightarrow {\mathfrak{r}}}} ^{H_q}\mathbb{P}_x^{b]}(T_a < \infty) > 0,
\end{aligned}$$ when $X \in {\mathcal{M}}_{\infty}$. To this end, we assume the contrary, that is, $\lim_{a \rightarrow {\mathfrak{r}}} {}^{H_q}\mathbb{P}_x^{b]}(T_a < \infty) = 0$. Since ${}^{H_q}\mathbb{P}_x^{b]}(T_a < \infty) = \mathbb{P}_x^{H_q}(T_a < T_{b]})$, the assumption becomes $\lim_{a \rightarrow {\mathfrak{r}}} \mathbb{P}_x^{H_q}(T_a < T_{b]}) = 0$ and leads to $$\begin{aligned}
\label{eq:assump}
\lim_{a \rightarrow {\mathfrak{r}}} \mathbb{P}_x^{H_q}(T_a < T_b) = 0.
\end{aligned}$$ Next, by means of a first step analysis and the upward skip-free property, we obtain $$\begin{aligned}
\label{eq:pk}
\mathbb{P}_b^{H_q}(T_b^+ < T_a) &= p^{H_q}(b,b+1) \mathbb{P}_{b+1}^{H_q}(T_b^+ < T_a) + p^{H_q}(b,b) + \sum_{y < b} p^{H_q}(b,y) \mathbb{P}_{y}^{H_q}(T_b^+ < T_a).
\end{aligned}$$ Taking $a \to {\mathfrak{r}}$, the left-hand side converges to $\mathbb{P}_b^{H_q}(T_b^+ < \infty)$ (recall that $T_b^+ = \inf\{n {\geqslant}1;\: X_n = b\}$) due to the monotone convergence theorem, the upward skip-free property and the fact that $X \in {\mathcal{M}}_{\infty}$. On the right-hand side of , the first term converges to $p^{H_q}(b,b+1)$ as a result of , while the third term converges to $\sum_{y < b} p^{H_q}(b,y) \mathbb{P}_{y}^{H_q}(T_b^+ < \infty)$ by invoking the dominated convergence theorem. Therefore, we arrive at $$\begin{aligned}
\mathbb{P}_b^{H_q}(T_b^+ < \infty) &= p^{H_q}(b,b+1) + p^{H_q}(b,b) + \sum_{y < b} p^{H_q}(b,y) \mathbb{P}_{y}^{H_q}(T_b^+ < \infty) \\
&= p^{H_q}(b,b+1) + \sum_{y {\leqslant}b} p^{H_q}(b,y) = \mathbb{P}_{b}^{H_q}(\zeta>1)
= 1,
\end{aligned}$$ where the second equality comes from the identity $\mathbb{P}_{y}^{H_q}(T_b^+ < \infty) = 1$ which holds since ${\mathbb{P}}_y^{H_q}(X_{\zeta} = {\mathfrak{r}}) = 1$ and $y< b$, while the last equality is due to Theorem \[thm:htransp\] with the fact that $H_q \in \mathcal{H}_q$ since $X \in {\mathcal{M}}_{\infty}$. This is not possible since $X$ is transient. Therefore, we conclude that $\kappa_q^{b]} > 0$.
Proof of Theorem \[thm:green2\]
-------------------------------
We start with the following extension of Lemma \[lem: below\].
\[lem: belowf\] For any $x,y \in E$ and $0<q<1$, we have $$\label{eq:hitg}
\mathbb{E}_x\left(q^{T_y}\right)= \dfrac{H_q(x) - {\mathbf{H}_q^{y]}}(x)}{H_q(y)}.$$
The case when $x{\leqslant}y$ is proved in . Next assume that $x>y$. Thanks to the upward skip-free property of $X$, for any $b \in E$ the mapping ${\mathds{1}}_{\{T_b < T_a\}}$ is increasing in $a$ large enough. Then, the monotone convergence theorem and the fact that $X \in {\mathcal{M}}_{\infty}$ give $ \lim_{a \rightarrow {\mathfrak{r}}}\mathbb{E}_x(q^{T_b} {\mathds{1}}_{\{T_b < T_a\}}) = \mathbb{E}_x(q^{T_b})$. The sought result follows immediately from Lemma \[lem: below\] and Lemma \[lem:Kqb\].
We are now ready to prove the expression . First using , Lemma \[thm:hit\] and the definition of the Martin kernel in , we obtain, for any $x {\leqslant}{\mathfrak{o}}$, $$\begin{aligned}
\label{eq:ho}
\mathbb{E}_x(q^{T_{\mathfrak{o}}}) = H_q(x) = \dfrac{G_q(x,{\mathfrak{o}})}{G_q({\mathfrak{o}},{\mathfrak{o}})}= \dfrac{G_q(x,{\mathfrak{o}})}{C_q}.\end{aligned}$$ Next, for sake of clarity we state the analogue of the identity for the dual chain $(X,\widehat{{\mathbb{P}}})$.
\[lem:dualminEexnar\] For all $0 < q < 1$ and any $x {\leqslant}y$, $$\begin{aligned}
\label{eq:hatHq}
\widehat{\mathbb{E}}_y(q^{T_x}) &= \dfrac{\widehat{H}_q(y)}{\widehat{H}_q(x)}.
\end{aligned}$$
A specific application of the previous result yields, for any $x {\leqslant}{\mathfrak{o}}$, that $$\begin{aligned}
\label{eq:hatHq2}
\widehat{\mathbb{E}}_{\mathfrak{o}}(q^{T_x}) &= \dfrac{1}{\widehat{H}_q(x)} = \dfrac{\widehat{G}_q({\mathfrak{o}},x)}{\widehat{G}_q(x,x)} = \dfrac{ G_q(x,{\mathfrak{o}})}{G_q(x,x)},\end{aligned}$$ where we use the identity $\widehat{H}_q({\mathfrak{o}}) = 1$ in the first equality, the dual version of for the second one and the integrated version of the dual identity for the last one. Combining and , we get, for any $x {\leqslant}{\mathfrak{o}}$, $$\begin{aligned}
\label{eq:gqyy}
G_q(x,x) = C_q H_q(x) \widehat{H}_q(x).\end{aligned}$$ For any $x {\geqslant}{\mathfrak{o}}$, we reverse the role of $x$ and ${\mathfrak{o}}$ to obtain, respectively, $$\begin{aligned}
\mathbb{E}_{\mathfrak{o}}(q^{T_x}) &= \dfrac{1}{H_q(x)} = \dfrac{G_q({\mathfrak{o}},x)}{G_q(x,x)}, \label{eq:hatHqrev}\\
\widehat{\mathbb{E}}_x(q^{T_{\mathfrak{o}}}) &= \widehat{H}_q(x) = \dfrac{\widehat{G}_q(x,{\mathfrak{o}})}{\widehat{G}_q({\mathfrak{o}},{\mathfrak{o}})} = \dfrac{G_q({\mathfrak{o}},x)}{G_q({\mathfrak{o}},{\mathfrak{o}})} \label{eq:hatHqrev2}.\end{aligned}$$ Combining and , we again arrive at , which shows that holds for all $x \in E$. Note that holds regardless of the boundary condition at ${\mathfrak{r}}$. In particular, when $X \in {\mathcal{M}}_{\infty}$, , (replacing $y$ by ${\mathfrak{o}}$), and Lemma \[thm:hit\] give which complete the proof of Theorem \[thm:green2\].
$X \in {\mathcal{M}}\setminus {\mathcal{M}}_{\infty}$ {#sec:regular}
-----------------------------------------------------
Throughout this section we work under the following hypothesis that $X \in {\mathcal{M}}\setminus {\mathcal{M}}_{\infty}$, that is, ${\mathfrak{r}}$ is a regular boundary. The state space $E$ includes the point ${\mathfrak{r}}$, that is, $E = [[{\mathfrak{l}},{\mathfrak{r}}]$. Under $\mathbb{P}^{[{\mathfrak{r}}}$, $X$ is a skip-free Markov chain on the state space $E^{[{\mathfrak{r}}} = [[{\mathfrak{l}},{\mathfrak{r}})$, which is killed whenever the process hits the state ${\mathfrak{r}}$. We are ready to state the following.
\[prop:regular\] Suppose ${\mathfrak{r}}$ is a regular boundary. Then for any $x > b$, $$\mathbb{E}_x(q^{T_b}) = \begin{cases}
\dfrac{H_q(x) - \bar{K}_q^{b]} H_q^{b]}(x)}{H_q(b)} &\text{if ${\mathfrak{r}}> x > b,$}\\
\dfrac{G_q({\mathfrak{r}},b)}{G_q(b,b)} &\text{if ${\mathfrak{r}}= x > b,$}
\end{cases}$$ where $\bar{K}_q^{b]} = \dfrac{G_q^{[{\mathfrak{r}}}({\mathfrak{o}},{\mathfrak{o}}) \widehat{H}_q^{[{\mathfrak{r}}}(b) }{G_q({\mathfrak{o}},{\mathfrak{o}}) \widehat{H}_q(b)}$ and $G_q({\mathfrak{r}},b) = \dfrac{\eta_b(q)}{c + (1-q)p + \sum_{j \in E} (1-q) \eta_j(q)},$ where $c {\geqslant}0$, $p > 0$, and $\eta(q) = (\eta_j(q))_{j \in E}$ is a family of non-negative numbers that satisfies, for any $0<s,q<1$, $$s \eta(s) - q \eta(q) = (s - q) \eta(q) G_q^{[{\mathfrak{r}}},$$ where $G_q^{[{\mathfrak{r}}} = (G_q^{[{\mathfrak{r}}}(i,j))_{i,j \in E}$.
In order to prove this Proposition, we need the following classical result that enables to connect the $q$-potential of $G_q$ and $G_q^{[a}$ (resp. $\widehat{G}_q$ and $\widehat{G}_q^{b]}$).
\[lem:excursion\] We have, for any $0 < q < 1$, $$\begin{aligned}
G_q(x,y) &= G_q^{[a}(x,y) + {\mathbb{E}}_x(q^{T_a}) G_q(a,y) \label{eq:killedabv} , \quad x,y \in E^{[a}, \\
\widehat{G}_q(x,y) &= \widehat{G}_q^{b]}(x,y) + \widehat{{\mathbb{E}}}_x(q^{T_b}) \widehat{G}_q(b,y), \quad x,y \in E^{b]} \label{eq:killedbel}.
\end{aligned}$$
Since $X$ is upward skip-free, $T_{[a} = T_a$ and for any $n \in \mathbb{N}$, $a \in E$, $x,y \in E^{[a}$, we have $$\begin{aligned}
\mathbb{P}_x(X_n = y) &= \mathbb{P}_x^{[a}(X_n = y) + \sum_{k = 1}^{n-1} \mathbb{P}_x(X_n = y | T_a = k) \mathbb{P}_x(T_a = k) \\
&= \mathbb{P}_x^{[a}(X_n = y) + \sum_{k = 1}^{n-1} \mathbb{P}_a(X_{n-k} = y) \mathbb{P}_x(T_a = k),
\end{aligned}$$ where the second equality follows from strong Markov property. Next, multiplying by $q^n$, dividing by $\pi(y)$, which is positive by irreducibility, and summing over $n$, we obtain $$\begin{aligned}
G_q(x,y) &= G_q^{[a}(x,y) + \sum_{n=1}^{\infty} \sum_{k = 1}^{n-1} q^{n-k} \frac{\mathbb{P}_a(X_{n-k} = y)}{\pi(y)} q^k \mathbb{P}_x(T_a = k) \\
&= G_q^{[a}(x,y) + \sum_{k=1}^{\infty} \sum_{n = k+1}^{\infty} q^{n-k} \frac{\mathbb{P}_a(X_{n-k} = y)}{\pi(y)} q^k \mathbb{P}_x(T_a = k) \\
&= G_q^{[a}(x,y) + {\mathbb{E}}_x(q^{T_a}) G_q(a,y)
\end{aligned}$$ which proves . is the dual statement of .
We are now ready to complete the proof of Proposition \[prop:regular\]. First, observe that $(X,{\mathbb{P}}^{[{\mathfrak{r}}}) \in {\mathcal{M}}_{\infty}$ and thus by means of Lemma \[lem:excursion\] and , one gets $$\begin{aligned}
\label{eq:gqbb2} G_q(b,b) &= G_q^{[{\mathfrak{r}}}({\mathfrak{o}},{\mathfrak{o}}) H_q^{[{\mathfrak{r}}}(b) \widehat{H}_q^{[{\mathfrak{r}}}(b) + \dfrac{H_q(b)}{H_q({\mathfrak{r}})} G_q({\mathfrak{r}},b).\end{aligned}$$ If one chooses the same reference point ${\mathfrak{o}}$ for $(X,{\mathbb{P}})$ and $(X,{\mathbb{P}}^{[{\mathfrak{r}}})$, then plainly, for any $x \in E^{[{\mathfrak{r}}}$, $H_q(x) = H_q^{[{\mathfrak{r}}}(x)$, this leads, using again , to $$\begin{aligned}
\label{eq:rreg}
G_q({\mathfrak{o}},{\mathfrak{o}}) \widehat{H}_q(b) = G_q^{[{\mathfrak{r}}}({\mathfrak{o}},{\mathfrak{o}}) \widehat{H}_q^{[{\mathfrak{r}}}(b) + \dfrac{G_q({\mathfrak{r}},b)}{H_q({\mathfrak{r}})}.\end{aligned}$$ For ${\mathfrak{r}}> x > b$, using the lemmas \[thm:hit\] and \[lem:excursion\], we have $$\begin{aligned}
\mathbb{E}_x(q^{T_b}) &= \dfrac{G_q(x,b)}{G_q(b,b)}
= \dfrac{G_q^{[{\mathfrak{r}}}(x,b) + {\mathbb{E}}_x(q^{T_{{\mathfrak{r}}}}) G_q({\mathfrak{r}},b)}{G_q({\mathfrak{o}},{\mathfrak{o}}) H_q(b) \widehat{H}_q(b)} \\
&= \dfrac{G_q^{[{\mathfrak{r}}}({\mathfrak{o}},{\mathfrak{o}})(H_q^{[{\mathfrak{r}}}(x) - K_q^{b]} H_q^{b]}(x)) \widehat{H}_q^{[{\mathfrak{r}}}(b) + \frac{H_q(x)}{H_q({\mathfrak{r}})} G_q({\mathfrak{r}},b)}{G_q({\mathfrak{o}},{\mathfrak{o}}) H_q(b) \widehat{H}_q(b)} \\
&= \dfrac{G_q({\mathfrak{o}},{\mathfrak{o}}) H_q(x) \widehat{H}_q(b) - G_q^{[{\mathfrak{r}}}({\mathfrak{o}},{\mathfrak{o}}) H_q^{b]}(x) \widehat{H}_q^{[{\mathfrak{r}}}(b) }{G_q({\mathfrak{o}},{\mathfrak{o}}) H_q(b) \widehat{H}_q(b)} \\
&= \dfrac{H_q(x) - \bar{K}_q^{b]} H_q^{b]}(x)}{H_q(b)},\end{aligned}$$ where we used in the fourth equality. For $x={\mathfrak{r}}> b$, it follows from that $$\begin{aligned}
\mathbb{E}_{{\mathfrak{r}}}(q^{T_b}) &= \dfrac{G_q({\mathfrak{r}},b)}{G_q(b,b)}
= \dfrac{G_q({\mathfrak{r}},b)}{G_q({\mathfrak{o}},{\mathfrak{o}}) H_q(b) \widehat{H}_q(b)}.\end{aligned}$$ To complete the proof one appeals to Theorem 3.1 of [@R59] that enables the determination of $G_q({\mathfrak{r}},b)$.
Fluctuation identities {#sec:fluc}
======================
We pursue our program by exploiting the potential theoretic results of the previous Section to derive the fluctuation identities of a general skip-free Markov chain. This consists in determining the law, through the pgf, of the first exit time of $X$ to a (in)finite interval. These quantities are critical in many applications in various settings including insurance mathematics, biology and epidemiology to name but a few. For instance, the time $T_{0]}$ corresponds to the time of ruin for a risk process having only negative jumps, which is a natural assumption in risk theory as they (the jumps) model the size of claims. We emphasize that our methodology enables the determination of the pgf of this time of ruin and hence allows the solution of this problem for general Markovian risk processes. We also recall that for a birth-death chain the pgf of exit times is given as a linear combination of the two F$q$E functions, the one for $(X,{\mathbb{P}})$ and its dual $(X,\widehat{{\mathbb{P}}})$. There exists also a theoretical characterization of the laws of these variables for random walks through the celebrated Wiener-Hopf factorization. This technique, which finds its root in complex analysis, has been nicely exploited by pure probabilistic arguments, see Feller [@Fe71] and Greenwood and Pitman [@GP80]. Our original approach appears to be more comprehensive for this issue in the context of general skip-free Markov chains. We have already mentioned that in order to illustrate its applicability, we will provide in the next subsection an alternative way to recover the fluctuation identities for skip-free random walks. The analogue of these results for skip-free continuous-time Markov processes on the real line can be found in [@Patie-Vigon] and applications to generalized Ornstein-Uhlenbeck processes and continuous branching processes with immigration are carried out in [@Lefevre2013] and [@Patie-Wev] respectively. We focus below on the case $X \in {\mathcal{M}}_{\infty}$, the other case can be dealt with similarly by means of Proposition \[prop:regular\].
\[thm:minEexnad\] Suppose that $X \in {\mathcal{M}}_{\infty}$. Then for any $b \in E$, $x \in E^{b]}$ and $0<q{\leqslant}1$, we have $$\label{eq:fptb}
\mathbb{E}_x(q^{T_{b]}}) = 1 + (q-1) C_q \sum_{y \in E^{b]}} \widehat{H}_q(y) \left({\mathbf{H}_q^{b]}}(x) - {\mathbf{H}_q^{y]}}(x) \right) \pi(y), $$ and, for any $a>b$, $x \in E^{(b,a)^c}$, $$\label{eq:fptab}
\mathbb{E}_x(q^{T_{(b,a)^c}}) = 1 + (q-1) C_q \sum_{y \in E^{(b,a)^c}} \widehat{H}_q(y) \left( \dfrac{ {\mathbf{H}_q^{y]}}(a)}{\mathbf{H}_q^{b]}(a)} {\mathbf{H}_q^{b]}}(x) - {\mathbf{H}_q^{y]}}(x) \right) \pi(y),$$ where we recall that $C_q = G_q({\mathfrak{o}},{\mathfrak{o}})$ as in Theorem \[thm:green2\] and for birth-death chains $C_q$ is the inverse of the Wronskian.
\[rk:ft\] We emphasize that the expressions and are comprehensive and are very useful for solving the first exit times problems when applied to some specific instances. Indeed they reveal that the characterization of the (probability generating function of the) distribution of first passage times of skip-free Markov chains boils down to determining the positive constant $C_q$ and the functions $\widehat{H}_q$ and ${\mathbf{H}_q^{b]}}$. In practice, it turns out that the knowledge of a transformation, such as the Laplace transform, Fourier transform or moment generating function, of the one-step transition (and hence by integration of the $q$-potential) of the chain enables the determination of $H_q$ and $\widehat{H}_q$. The constant $C_q$ can be determined by an argument of analytical continuation applied to the transform of the $q$-resolvent. Finally, (a transformation of) the function ${\mathbf{H}_q^{b]}}$ can be obtained from the previous identifications combined with the following relations $${\mathbf{H}_q^{y]}}(x)=\frac{1}{C_q \: \widehat{H}_q(y)}
\left(H_q(x) - G_q(x,y)\right).$$ that can be easily derived from . This procedure will be illustrated in Section \[sec:sfrw\] below to the class of skip-free random walks and, in the subsequent paper [@Patie-Wang], to the (continuous-time) branching Galton-Watson processes with immigration.
The analysis in Theorem \[thm:minEexnad\] can be extended to study the joint law of $X_{T_{b]} - 1}$ and $X_{T_{b]}}$. Indeed, for $x,y \in E$ and $k {\leqslant}b$, $$\begin{aligned}
{\mathbb{E}}_x(q^{T_{b]}} {\mathds{1}}_{\{X_{T_{b]} - 1} = y, X_{T_{b]}} = k\}}) &= \sum_{n=1}^\infty q^n {\mathbb{P}}_x(X_{n-1} = y, X_n = k, T_{b]} = n) \\
&= q \: G^{b]}_q(x,y) p(y,k),
\end{aligned}$$ where the last equality follows from the Markov property. Taking $q \rightarrow 1^-$ and using the monotone convergence theorem, one gets $$\begin{aligned}
{\mathbb{P}}_x(X_{T_{b]} - 1} = y, X_{T_{b]}} = k) &= G^{b]}(x,y)p(y,k).
\end{aligned}$$ Summing over $y \in E$ (or $y \in E^{b]}$, since $G^{b]}(x,y) = 0$ for $y {\leqslant}b$), we have $${\mathbb{P}}_x(X_{T_{b]}} = k) = \sum_{y \in E} G^{b]}(x,y)p(y,k).$$
We proceed with the proof of Theorem \[thm:minEexnad\]. First, let $B \subset E$ and denote $G_q^{B}$ (resp. $\widehat{G}_q^{B}$) to be the $q$-potential of the $X$ (resp. its dual $\widehat{X}$) killed upon entering into the set $B$. We recall the Hunt’s switching identity for Markov chains, which can be found in [@KSK page $140$], and says that, for any $x,y \in E\backslash B$, $$\label{eq:huntdual}
G_q^B(x,y) = \widehat{G}_q^{B}(y,x) .$$ For sake of simplicity, we simply write $G_q^{B} = G_q^{b]}$ (resp. $G_q^{A} = G_q^{[a}$) if $B = ({\mathfrak{l}},b]$ (resp. if $B = [a,{\mathfrak{r}})$). With this notation in mind, we express the $q$-potential kernels of $(X,{\mathbb{P}}^{[a})$ and $(X,{\mathbb{P}}^{b]})$ in terms of F$q$E functions of the three processes $(X,{\mathbb{P}})$, $(X,\widehat{{\mathbb{P}}})$ and $(X,{\mathbb{P}}^{y]})$.
\[lem:green2\] Suppose that $X \in {\mathcal{M}}_{\infty}$. $$\begin{aligned}
G_q^{[a}(x,y) &= C_q \widehat{H}_q(y) \left(\dfrac{ {\mathbf{H}_q^{y]}}(a)}{H_q(a)} H_q(x) - {\mathbf{H}_q^{y]}}(x) \right), \quad x,y \in E^{[a}, \label{eq:killedabvH} \\
G_q^{b]}(x,y) &= C_q\widehat{H}_q(y) \left({\mathbf{H}_q^{b]}}(x) - {\mathbf{H}_q^{y]}}(x) \right), \quad x,y \in E^{b]}, \label{eq:killedbelH} \\
G_q^{(b,a)^c}(x,y) &= C_q \widehat{H}_q(y) \left( \dfrac{ {\mathbf{H}_q^{y]}}(a)}{{\mathbf{H}_q^{b]}}(a)} {\mathbf{H}_q^{b]}}(x) - {\mathbf{H}_q^{y]}}(x) \right), \quad x,y \in E^{(b,a)^c}. \label{eq:killed2H}
\end{aligned}$$
To show , we use and to get $$\begin{aligned}
G_q^{[a}(x,y) &= G_q(x,y) - {\mathbb{E}}_x(q^{T_a}) G_q(a,y) \\
&= C_q (H_q(x) - {\mathbf{H}_q^{y]}}(x) ) \widehat{H}_q(y) - \dfrac{H_q(x)}{H_q(a)} C_q (H_q(a) - {\mathbf{H}_q^{y]}}(a))\widehat{H}_q(y) \\
&= C_q \widehat{H}_q(y) \left(\dfrac{H_q(x) {\mathbf{H}_q^{y]}}(a)}{H_q(a)} - {\mathbf{H}_q^{y]}}(x) \right).\end{aligned}$$ Next, using the Hunt’s switching identity and , we have $$\begin{aligned}
G_q^{b]}(x,y) &= \widehat{G}_q^{b]}(y,x) = \widehat{G}_q(y,x) - \widehat{{\mathbb{E}}}_y(q^{T_b}) \widehat{G}_q(b,x) = G_q(x,y) - \dfrac{\widehat{H}_q(y)}{\widehat{H}_q(b)} G_q(x,b) \\
&= C_q (H_q(x) - {\mathbf{H}_q^{y]}}(x) ) \widehat{H}_q(y) - \dfrac{\widehat{H}_q(y)}{\widehat{H}_q(b)} C_q (H_q(x) - {\mathbf{H}_q^{b]}}(x) )\widehat{H}_q(b) \\
&= C_q\widehat{H}_q(y) \left({\mathbf{H}_q^{b]}}(x) - {\mathbf{H}_q^{y]}}(x) \right),\end{aligned}$$ which proves . Finally, to get , we use the identity $G_q^{b]}(x,y) = G_q^{(b,a)^c}(x,y) + {\mathbb{E}}_x^{b]}(q^{T_a}) G_q^{b]}(a,y)$ and combined with Lemma \[lem:minEb\].
With Lemma \[lem:green2\] in mind, the proof of Theorem \[thm:minEexnad\] follows readily by applying the second claim stated in the following classical results which we prove for sake of completeness.
For any $b<x$ and $n\in {\mathbb{N}}$, we have $${\mathbb{P}}_x( T_{b]} > n)=P_n^{b]}{\mathds{1}}(x)=\frac{\pi \widehat{P}_n^{b]}\delta_x}{\pi(x)}$$ and thus, for any $0<q<1$, $$\mathbb{E}_x(q^{T_{b]}})= 1 + (q-1){\bf{G}}_q^{b]}{\mathds{1}}(x) = 1 + (q-1) \frac{\pi \widehat{\bf{G}}_q^{b]}\delta_x}{\pi(x)},$$ where ${\bf{G}}_q^{b]}f(x) = \sum_{y \in E^{b]}} f(y) G^{b]}_q(x,y)\pi(y)$.
First, an application of Fubini’s theorem yields, that for any $b<x$ and $0<q{\leqslant}1$, $$\begin{aligned}
{\bf{G}}_q^{b]}{\mathds{1}}(x) &= \sum_{y \in E^{b]}} G^{b]}_q(x,y)\pi(y) = \sum_{y \in E^{b]}} \sum_{n = 0}^{\infty} q^n {\mathbb{P}}_x(X_n = y, n < T_{b]}) \\ &= \sum_{n = 0}^{\infty} \sum_{y \in E^{b]}} q^n {\mathbb{P}}_x(X_n = y, n < T_{b]}) = \sum_{n = 0}^{\infty} q^n {\mathbb{P}}_x( T_{b]} > n ).\end{aligned}$$ Moreover, from the Hunt’s switching identity , we observe that $$\begin{aligned}
{\bf{G}}_q^{b]}{\mathds{1}}(x) &= \sum_{y \in E^{b]}} G^{b]}_q(x,y)\pi(y) = \sum_{y \in E^{b]}} \pi(y) \widehat{G}^{b]}_q(y,x) = \frac{\pi \widehat{{\bf{G}}}_q^{b]}\delta_x}{\pi(x)}\end{aligned}$$ which completes the proof of the first claim. Finally, the lemma is proved after observing that, for any $x \in E^{b]}$, $$\begin{aligned}
\mathbb{E}_x(q^{T_{b]}}) &= \sum_{n = 1}^\infty q^n {\mathbb{P}}_x(T_{b]} = n) = \sum_{n = 1}^\infty q^n \left( {\mathbb{P}}_x(T_{b]} > n-1) - {\mathbb{P}}_x(T_{b]} > n) \right) \\
&= q \sum_{n = 0}^{\infty} q^n {\mathbb{P}}_x( T_{b]} > n ) - \sum_{n = 1}^\infty q^n {\mathbb{P}}_x(T_{b]} > n) \\
&= 1 + (q-1) \sum_{n = 0}^{\infty} q^n {\mathbb{P}}_x( T_{b]} > n ). $$
Another method to study the first passage time of $b$ is by collapsing and combining all the states at or below $b$, and consider the hitting time to the glued state for the chain. Precisely, let $\widetilde{P}$ be the transition matrix of the glued chain with an absorbing boundary at $b$, that is, $$\widetilde{P} = \kbordermatrix{ & b & b+1 & b+2 & \ldots\\
b & 1 & 0 & \ldots & \ldots\\
b+1 & \sum_{j=0}^{b} p(b+1,j) & p(b+1,b+1) & \ldots & \ldots\\
b+2 & \sum_{j=0}^{b} p(b+2,j) & p(b+2,b+1) & \ldots & \ldots\\
\vdots & \vdots & \vdots & \vdots & \vdots
}$$ By construction, $${\mathbb{P}}_x(T_{b]} {\leqslant}n) = \widetilde{{\mathbb{P}}}_x(T_b {\leqslant}n).$$ Therefore, by , using the obvious notation, $$\begin{aligned}
\mathbb{E}_x(q^{T_{b]}}) = \widetilde{\mathbb{E}}_x(q^{T_{b}}) = \dfrac{\widetilde{H_q}(x) - \widetilde{H}_q^{b]}(x)}{\widetilde{H}_q(b)}.
\end{aligned}$$
Skip-free random walk on $\mathbb{Z}$ revisited {#sec:sfrw}
-----------------------------------------------
We illustrate how the comprehensive result stated in Theorem \[thm:minEexnad\] can be used to obtain explicit representations for the pgf of the first exit time of a skip-free Markov chain when applied to some specific instances. Below, we consider an upward skip-free random walk $(X,{\mathbb{P}})$ on the state space $\mathbb{Z}$. Note that for these chains, these problems are of course classic and explicitly treated by Feller in [@Fe71]. However, we provide here a new methodology to recover the well known fluctuation identities of these chains. To emphasize the comprehensive aspect of our approach, we mention that it has also been successfully used in [@APW18] and [@Patie-Wang] to the study the potential and fluctuation theory of Galton-Watson processes with immigration which are (continuous-time) skip-free to the left Markov chains. Suppose now that $X = (X_n)_{n \in \mathbb{N}_0}$ is a upward skip-free random walk given by ${\mathbb{P}}_x(X_0 = x)=1$, $x\in \mathbb Z$, and $X_n = X_0+ \sum_{i=1}^n S_i$, where $(S_i)_{n \in \mathbb{N}_0}$ are i.i.d. random variables with common distribution $F$ which is supported on $\{n \in \mathbb{Z}; \: n {\leqslant}1\}$. We write $\mathbf{F}$ for the pgf of $F$. Thus, writing $p_{s}(x)=s^{x}$, we have $$\label{eq:mgsmc}
Pp_{s}(x)= \mathbf{F}(s) p_{s}(x). $$ Since $\lim_{s \to \infty } \mathbf{F}(s)=\infty$, one also sees that the mapping $s \mapsto \mathbf{F}(s) $ is continuous, increasing and convex on $(1/h(1),\infty)$ where $h(1){\geqslant}1$ is the largest root of the equation $\mathbf{F}(s)=1$. Note that $1$ is always a root and $h(1)>1$ when $\mathbf{F}'(1^+)<0$ (the right-derivative at $1$). Therefore, $\frac{1}{\mathbf{F}}$ is continuous, decreasing on $(0, h(1))$ and thus has a well-defined inverse $h : (0,1) \to (h(1),\infty)$. Recall also that, in this case, the dual, with respect to the reference measure $\pi \equiv 1$, is $(X,\widehat{{\mathbb{P}}}) \overset{d}{=} (-X, {\mathbb{P}})$.
We take the reference point to be ${\mathfrak{o}}= 0$ and let $0<q<1$.
1. $H_q \in \mathcal{H}_q$ and $\widehat{H}_q \in \widehat{\mathcal{H}}_q$, where for $x \in \mathbb{Z}$, $$\begin{aligned}
\label{eq:hq}
H_q(x) =h(q)^x, \quad \widehat{H}_q(x) = h(q)^{-x}.
\end{aligned}$$ In addition, $$\begin{aligned}
\label{eq:Cq}
C_q = -q \dfrac{ h'(q)}{h(q)}. \end{aligned}$$
2. For any $x {\geqslant}y+1$, we have ${\mathbf{H}_q^{y]}}(x) = h(q)^y\mathbf{H}_q^{0]}(x-y)$ where, for $s \in \mathbb{R}$ such that $\mathbf{F}(\frac{1}{s}) > \frac{1}{q}$, $$\begin{aligned}
\sum_{x \in \mathbb{N}} p_{s}(x) \mathbf{H}_q^{0]}(x) = \dfrac{1}{C_q(q \mathbf{F}(1/s) - 1)}.
\end{aligned}$$
3. \[it:qp\] For any $x,y \in \mathbb{Z}$, we have for any $0<q<1$, $G_q(x,y)=\overline{G}_q(x-y)$ where $$\overline{G}_q(x)=-q \dfrac{ h'(q)}{h(q)} \left(h(q)^x-\mathbf{H}_q^{0]}(x)\right).$$
4. For any $x >0$, $$\begin{aligned}
\mathbb{E}_x(q^{T_{0]}}) &= 1 + (1-q) C_q\sum_{y=1}^{x-1} \mathbf{H}_q^{0]}(y) + q h'(q)\dfrac{q-1}{h(q)-1} \mathbf{H}_q^{0]}(x).
\end{aligned}$$
First, we deduce easily from and the definition of $h$, that, with $H_q(x) = h(q)^x=p_{h(q)}(x)$, $0<q<1$, $$q P H_q (x) = H_q(x)$$ that is $H_q \in \mathcal{H}_q$. Since $\ln h$ is nonnegative, clearly $H_q$ is increasing. As $(X,\widehat{{\mathbb{P}}}) \overset{d}{=} (-X, {\mathbb{P}})$, the first claims follow readily from Theorem \[thm:minEexnad\]. Next, by means of Tonelli’s theorem, on the one hand for $s \in \mathbb{R}$ such that $\left|q \mathbf{F}(1/s)\right| < 1$, we have $$\sum_{x \in \mathbb{Z}} s^{-x} G_q(0,x) = \sum_{n {\geqslant}0} q^n {\mathbb{E}}_0 (s^{-X_n}) = \sum_{n {\geqslant}0} \left( q \mathbf{F}(1/s) \right)^n = \dfrac{1}{1 - q \mathbf{F}(1/s)}.$$ On the other hand, using the translation invariant property of $X$, we have $G_q(-x,0) = G_q(0,x)$, which leads to $$\sum_{x \in \mathbb{Z}} s^{-x} G_q(-x,0) = \sum_{x > 0} s^x G_q(x,0) + \sum_{x {\geqslant}0} s^{-x} G_q(-x,0) .$$ Rearranging the terms and using and with $\sum_{x > 0} s^x H_q(x) = - \frac{s h(q)}{s h(q)-1}$ yields $$\begin{aligned}
\label{eq:fluct}
\sum_{x > 0} s^x (H_q(x) - \mathbf{H}_q^{0]}(x)) = \dfrac{1}{C_q(1 - q \mathbf{F}(1/s))} - \dfrac{s h(q)}{s h(q)-1}.
\end{aligned}$$ As the left-hand side can be treated as $\sum_{x > 0} s^x {\mathbb{E}}_x(q^{T_0}) {\leqslant}\sum_{x > 0} s^x < \infty$ for $0 < s < 1$, it is analytical on the unit disc and by the principle of analytical continuation, one gets that $$\begin{aligned}
C_q &= \lim_{s \to \frac{1}{h(q)}} \dfrac{1 - 1/s h(q)}{1-q\mathbf{F}(1/s)} = \dfrac{-q h'(q)}{h(q)},
\end{aligned}$$ which shows . Next, following and using again $\sum_{x > 0} s^x H_q(x) = - \frac{s h(q)}{s h(q)-1}$, we get $$\sum_{x \in \mathbb{N}} s^x \mathbf{H}_q^{0]}(x) = \dfrac{1}{C_q \left(q \mathbf{F}(1/s) - 1\right)}.$$ Then, note that by the translation invariance property of $X$, $G_q(x,y)=G_q(x-y,0)$ for any $x,y \in E$ which after some easy algebra yields ${\mathbf{H}_q^{y]}}(x) = h(q)^y\mathbf{H}_q^{0]}(x-y)$ for any $x {\geqslant}y$. The proof of item follows readily from the previous items and the identity . Finally, using the first claim of Theorem \[thm:minEexnad\], we have $$\begin{aligned}
\mathbb{E}_x(q^{T_{0]}}) &= 1 + (q-1)C_q \sum_{y \in E^{0]}} \widehat{H}_q(y) \left(\mathbf{H}_q^{0]}(x) - {\mathbf{H}_q^{y]}}(x) \right) \\
&= 1 + (q-1)C_q \sum_{y \in E^{0]}} h(q)^{-y} \left( \mathbf{H}_q^{0]}(x) - h(q)^y\mathbf{H}_q^{0]}(x-y) \right) \\
&= 1 + (1-q) C_q\sum_{y=1}^{x-1} \mathbf{H}_q^{0]}(y) + q h'(q)\dfrac{q-1}{h(q)-1}\mathbf{H}_q^{0]}(x)
\end{aligned}$$ which completes the proof of the Theorem.
Classes of first passage time distributions {#sec:fpt}
===========================================
The aim of this part is two-fold. First, we provide a characterization of the first passage times distribution, both from above and below the starting point, in which we built upon the results of Fill [@Fill]. This follows from the line of work by Kent and Longford [@KL83] who characterize the class of upward and downward hitting time of birth-death Markov chains on non-negative integers. More specifically, they define and introduce a new class $\mathcal{K}(b,\tau,M)$ of discrete infinitely divisible distributions with support on non-negative integers and pgf given by $$\phi(z) = \exp\bigg\{b(z-1) - \tau + z(z-1) \int_{-1}^{1} (1-pz)^{-1} M(dp)\bigg\}$$ that contain both the upward and downward hitting time of such chain, where $b, \tau$ are non-negative parameters and $M$ is a finite measure on $(-1,1)$ such that $\int_{-1}^{1} (1+p)^{-1} M(dp) < \infty.$ In particular, using the interlacing property of the eigenvalues for birth-death process, Kent and Longford [@KL83] show that the measure $M$ that corresponds to the hitting times are non-negative. However, this interlacing property is lost in general when we move to the realm of skip-free chains. This motivates us to define a more general class of distribution that we call $\mathbb{G}_p$ in Definition \[def:classdist\] which contains both upward and downward hitting times in Theorem \[thm:hitting\]. This will be further demonstrated in Remark \[rk:KL83\] below. Second, we derive an explicit representation of the F$q$E functions associated to $(X,{\mathbb{P}})$ and $(X,{\mathbb{P}}^{b]})$ in Section \[subsec:fqe\]. As we will elaborate in Section \[subsec:id\], this allows us to investigate the infinite divisibility of the upward hitting time for skip-free Markov chains and characterizes the associated $R$-function by means of the eigenvalues. In this vein we emphasize that Feller [@F66] studied the infinite divisibility of the hitting times of birth-death random walk and Viskov [@Viskov00] extended his work to skip-free random walk using the classical Lagrange inversion formula. We focus on ${\mathcal{M}}_{\mathfrak{l}}$ the subclass of ${\mathcal{M}}$ that has a finite left boundary point, i.e. ${\mathfrak{l}}<\infty$. We shall also need the following definition where we use the notation, for $d\in \mathbb{N}$, $\mathbb{D}^d=\{\boldsymbol{\lambda} = (\lambda_k)_{k=1}^d \in {\mathbb{C}}^d;\: |\lambda_k| \in [0,1], \:k=1,\ldots,d \textrm{ and } \lambda_i \neq \lambda_k, i \neq k\}$.
\[def:classdist\] Let $p,d\in {\mathbb{N}}$ with $d {\leqslant}p$, $\boldsymbol{\lambda} = (\lambda_k) \in \mathbb{D}^d$ and $\boldsymbol{m} = (m_k) \in {\mathbb{N}}^d$ with $\sum_{k=1}^d m_k = p$, and for each $k \in [1,d]$, the multiplicity of $\lambda_k$ is $m_k$, $\mathbf{c}_k = (c_{k,i})_{i=1}^{m_k} \in {\mathbb{C}}^{m_k}$ and we write $\mathbf{c} = (\mathbf{c}_{k,i})$. We say that a non-negative discrete random variable $X \in \mathbb{G}_p(\boldsymbol{c};\boldsymbol{\lambda};\boldsymbol{m})$ if its probability mass function can be written, for any $n\in {\mathbb{N}}_0$, as $$0{\leqslant}{\mathbb{P}}(X=n) = \sum_{k=1}^d \sum_{i=1}^{m_k} c_{k,i} \binom{n+i-1}{n} \lambda_k^{n} (1-\lambda_k)^{i}{\leqslant}1,$$ where $\sum_{k=1}^d \sum_{i=1}^{m_k} c_{k,i} {\leqslant}1.$ In particular, if $\boldsymbol{\theta}=\boldsymbol{\lambda} \in [0,1]^d$, then we write $\mathcal{G}_p(\boldsymbol{c};\boldsymbol{\theta};\boldsymbol{m})=\mathbb{G}_p(\boldsymbol{c};\boldsymbol{\theta};\boldsymbol{m})$. Note that it can be interpreted as a (signed) mixture of geometric random variables when $m_k = 1$ and $\lambda_k \in [0,1]$ for all $k$.
Before stating the main result of this part, we introduce the notation $\boldsymbol{\lambda}^{[x,y]} = (\lambda_i^{[x,y]})_{i=0}^{y-x}$ for the (non-unit) eigenvalues of the transition matrix $P$ restricted to $[x,y]$ with $x {\leqslant}y$ and we use $\mathcal{ID}$ to denote the class of infinitely divisible laws, see Section \[subsec:id\] for formal definition and further discussion on related topics. In Fill [@Fill], under the condition that the transition matrix of an upward skip-free chain has only real and non-negative eigenvalues, it is shown that its upward first hitting time is a convolution of geometric distributions. We also refer to Miclo [@Miclo] for similar results in the context of reversible Markov chains.
\[thm:hitting\] Suppose that $(X,{\mathbb{P}})$ is irreducible. For any ${\mathfrak{l}}{\leqslant}b {\leqslant}x {\leqslant}a {\leqslant}{\mathfrak{r}}$, we have, under ${\mathbb{P}}_x$, $$T_a - (a - x) \in \mathbb{G}_{a}(\mathbf{c}(x,a);\boldsymbol{\lambda};\boldsymbol{m})\cap \mathcal{ID} \quad \textrm{and} \quad T_b \:(\textrm{resp.~}T_{b]}) \in \mathbb{G}_{{\mathfrak{r}}-1}(\boldsymbol{\tilde{c}}(x,b);\boldsymbol{\tilde{\lambda}};\boldsymbol{\tilde{m}}),$$ where $\boldsymbol{\lambda}$ are the distinct eigenvalues of $\boldsymbol{\lambda}^{[0,a-1]}$, $\boldsymbol{\tilde{\lambda}}$ are the distinct eigenvalues of $\boldsymbol{\lambda}^{[0,b-1]}$ and $\boldsymbol{\lambda}^{[b+1,{\mathfrak{r}}-1]}$, $\boldsymbol{m}$ (resp. $\boldsymbol{\tilde{m}}$) are the multiplicities of $\boldsymbol{\lambda}$ (resp. $\boldsymbol{\tilde{\lambda}}$). In particular, when $\boldsymbol{\lambda}, \boldsymbol{\tilde{\lambda}}$ are all real and non-negative, then, under ${\mathbb{P}}_x$, $$T_a - (a - x) \in \mathcal{G}_{a}(\mathbf{c}(x,a);\boldsymbol{\lambda};\boldsymbol{m})\cap \mathcal{ID} \quad \textrm{and} \quad T_b \:(\textrm{resp.~}T_{b]}) \in \mathcal{G}_{{\mathfrak{r}}-1}(\boldsymbol{\tilde{c}}(x,b);\boldsymbol{\tilde{\lambda}};\boldsymbol{\tilde{m}}).$$
Note that for class $\mathcal{S}$ to be introduced in Section \[sec:st\], the passage time distributions are in the particular cases of Theorem \[thm:hitting\].
Complex $c_{k,i}$ can occur as a result of partial fraction, and therefore it is necessary for us to consider such cases in Definition \[def:classdist\]. For example, if the $\alpha_j'$ s in are complex (here $\alpha_j$ is usually $\lambda_j^{[0,x-1]}$, so this will occur when $P$ restricted to $[0,x-1]$ has complex eigenvalues), then $c_{k,i}$ are complex in general.
Characterizations of the class $\mathbb{G}_p$
---------------------------------------------
We first provide a characterization of the class $\mathbb{G}_p$, which will simplify our proof of Theorem \[thm:hitting\].
\[prop:fptchar\] $X \in \mathbb{G}_p(\boldsymbol{c};\boldsymbol{\lambda};\boldsymbol{m})$ if and only if its probability generating function can be written as $$\label{eq:fptchar}
{\mathbb{E}}[q^X] = \sum_{k=1}^{d} \sum_{i=1}^{m_k} \dfrac{C_{k,i}}{(1 - \lambda_k q)^{i}},$$ where $(C_{k,i})$ depends on $\boldsymbol{c},\boldsymbol{\lambda}$. Moreover, if the probability generating function of $X$ can be written as $$\label{eq:fptchar2}
{\mathbb{E}}[q^X] =
\prod_{j=0}^{b-1} \frac{(1 - \beta_j) q}{1 - \beta_j q} \prod_{j=0}^{x-1} \frac{1 - \alpha_j q}{(1 - \alpha_j) q},$$ where $\boldsymbol{\beta} = (\beta_j) \in \mathbb{C}^b$, $\boldsymbol{\alpha} = (\alpha_j) \in \mathbb{C}^x$ with $|\lambda_j|, |\alpha_j| \in [0,1]$ for all $j$, $\boldsymbol{\lambda}$ are the distinct elements of $\boldsymbol{\beta}$ and $\boldsymbol{m}$ are the multiplicities, then $X \in \mathbb{G}_b(\boldsymbol{c};\boldsymbol{\lambda};\boldsymbol{m})$.
\[rk:KL83\] We remark that the class $\mathbb{G}_p$ is more general than the class $\mathcal{K}(b,\tau,M)$ studied by Kent and Longford [@KL83]. Indeed, using the expression in [@KL83 Section $10$], we see that can be rewritten as $$\begin{aligned}
q^{x-b} {\mathbb{E}}[q^X] &= \exp \bigg\{ (q-1) \left(\sum_{j=0}^{b-1} |\log(1-\beta_j)| - \sum_{j=0}^{x-1} |\log(1-\alpha_j)|\right) + q(q-1)\int_{-1}^{1} (1-pz)^{-1} M(dp)\bigg\}, \end{aligned}$$ where $M(dp) = \left(\sum_{j=0}^{b-1} {\mathds{1}}_{(0,\beta_j)} - \sum_{j=0}^{x-1} {\mathds{1}}_{(0,\alpha_j)} \right)|p|(1-p)dp$. In the particular case of birth-death chains, we can see that the measure $M$ that corresponds to $(T_{a+1},{\mathbb{P}}_{a})$ is, by the reality and the interlacing property of eigenvalues, non-negative. In general, the measure $M$ is a signed measure.
We first show that if $X \in \mathbb{G}_p(\boldsymbol{c};\boldsymbol{\lambda};\boldsymbol{m})$ then holds. By writing $C_{k,i} = c_{k,i} (1-\lambda_k)^{i}$, we see that $$\begin{aligned}
{\mathbb{E}}[q^X] &= \sum_{k=1}^{d} \sum_{i=1}^{m_k} C_{k,i} \sum_{n {\geqslant}0} \binom{n+i-1}{n} (q\lambda_k)^{n} = \sum_{k=1}^{d} \sum_{i=1}^{m_k} \dfrac{C_{k,i}}{(1 - \lambda_k q)^{i}}.
\end{aligned}$$ The opposite direction can be shown by differentiating the pgf $n$ times followed by dividing $n!$. Next, we show that pgf of the form belongs to the class $\mathbb{G}_p$. By means of partial fraction, we note that can be written as $$\begin{aligned}
{\mathbb{E}}[q^X] &= \sum_{k=0}^{d} \sum_{i=1}^{m_k} \dfrac{C_{k,i}}{(1 - \beta_k q)^{i}},
\end{aligned}$$ so by , $X \in \mathbb{G}_b(\boldsymbol{c};\boldsymbol{\lambda};\boldsymbol{m})$.
F$q$E functions of $(X,{\mathbb{P}})$ and $(X,{\mathbb{P}}^{b]})$ for skip-free processes {#subsec:fqe}
-----------------------------------------------------------------------------------------
In this section, we give explicit formulas on the F$q$E functions of $(X,{\mathbb{P}})$ and $(X,{\mathbb{P}}^{b]})$. Note that the former case follows from Fill [@Fill], whereas the latter one is a slight generalization of the first one by allowing substochastic transition matrices.
\[prop:sfmc\] Suppose that $(X,{\mathbb{P}})$ is an irreducible upward skip-free Markov chain on $E$ with $|{\mathfrak{l}}| < \infty$. For $b \in E$, we consider the killed process $(X,{\mathbb{P}}^{b]})$ on $E^{b]} = [b+1,{\mathfrak{r}})$. Let $\lambda_{i} = \lambda_i^{[b+1,x-1]}$ be the eigenvalues of $P$ when restricted to $[b+1,x-1]$. If we take $b+1$ to be the reference point (i.e. ${\mathfrak{o}}^{b]} = b+1$), then the F$q$E function of $(X,{\mathbb{P}}^{b]})$ is $$H_q^{b]}(x) = \begin{cases}
1, &\text{if $x = b+1,$}\\
\displaystyle \frac{1}{{\mathbb{P}}_{b+1}(T_x < T_{b]})} \prod_{i=0}^{x-b-2} \frac{1 - \lambda_{i} q}{(1 - \lambda_{i}) q}, &\text{if $x {\geqslant}b+2,$}
\end{cases}$$ where ${\mathbb{P}}_{b+1}(T_x < T_{b]}) = \dfrac{p(b+1,b+2)\ldots p(x-1,x)}{(1-\lambda_{0})\ldots(1-\lambda_{x-b-2})}$.
Suppose that $X$ is an upward skip-free Markov chain on $[-1,a]$, where $a$ and $-1$ are absorbing states. Let $P$ be the substochastic transition matrix on $[0,a]$, with $(\lambda_i)_{i=0}^{a-1}$ to be the (non-unit) eigenvalues, where $\lambda_i = \lambda_i^{[0,a-1]}$ for $i = 0,\ldots,a-1$. We define $\widetilde{P} = (\widetilde{p}(i,j))$ for $i,j = 0,1,\ldots,a$ $$\widetilde{p}(i,j) = \begin{cases} \lambda_i, &\quad j = i, \\
1 - \lambda_i , &\quad j = i + 1, \\
0 , &\quad\mbox{otherwise,} \end{cases}$$ where $\lambda_a = 1$. Next, the so-called spectral polynomial is defined to be $Q_0 = I$ and $$Q_k = \dfrac{(P - \lambda_0 I)\ldots(P - \lambda_{k-1}I)}{(1-\lambda_0) \ldots (1-\lambda_{k-1})},\quad k = 1,\ldots,a.$$ Define the link matrix $\Lambda = (\Lambda(i,j))$ to be $$\Lambda(i,j) = Q_i(0,j),\quad i,j = 0,\ldots,a.$$ Then $\Lambda$ satisfies the following properties:
(i) \[it:l1\] The sum of each row of $\Lambda$ is less than or equal to $1$.
(ii) \[it:l2\] $\Lambda$ is lower-triangular with $\Lambda(0,0) = 1$ and $$\Lambda(k,k) = \dfrac{p(0,1)\ldots p(k-1,k)}{(1-\lambda_0)\ldots(1-\lambda_{k-1})} \neq 0, \quad k =1,\ldots,a,$$ and hence $\Lambda$ is invertible.
(iii) \[it:l3\] $\Lambda P = \widetilde{P} \Lambda$.
The proof of $\eqref{it:l1},\eqref{it:l2}$ and $\eqref{it:l3}$ above are identical to that of Lemma $2.1$ in [@Fill], except that the sum of each row of $\Lambda$ is now less than or equal to $1$, since each row of $(1-\lambda_i)^{-1}(P - \lambda_i I)$ has that property. The next two results and are analogous to [@Fill Lemma $2.3$ and Theorem $1.2$], and we omit the proofs here as they are almost identical, except now we have ${\mathbb{P}}_0(T_a < T_{-1}) = \Lambda(a,a)$. For $t \in \mathbb{N}_0$, $$\label{eq:tsupht2}
{\mathbb{P}}_0(T_a {\leqslant}t) = {\mathbb{P}}_0(T_a < T_{-1}) \widetilde{P^t}(0,a),$$ where $${\mathbb{P}}_0(T_a < T_{-1}) = \Lambda(a,a) = \dfrac{p(0,1)\ldots p(a-1,a)}{(1-\lambda_0)\ldots(1-\lambda_{a-1})}.$$ For $q \in [0,1]$, $$\label{eq:tsupht}
{\mathbb{E}}^{-1]}_0(q^{T_a}) = {\mathbb{P}}_0(T_a < T_{-1}) \prod_{i=0}^{a-1} \frac{(1 - \lambda_i) q}{1 - \lambda_i q}.$$ The desired result follows immediately from Theorem \[thm:green2\], Lemma \[lem:minEb\] and .
With the two previous results in hand, we are now ready to complete the proof of Theorem \[thm:hitting\].
Proof of Theorem \[thm:hitting\] {#sec:pfc}
--------------------------------
Using the result in [@SVH04 Chapter VII Theorem 2.1], $(T_a - (a-x),{\mathbb{P}}_x)$ is infinitely divisible. It follows from that $$\begin{aligned}
\mathbb{E}_x(q^{T_a - (a-x)}) &= q^{- (a-x)}\dfrac{H_q(x)}{H_q(a)},
\end{aligned}$$ and by Proposition \[prop:sfmc\], the ratio $q^{- (a-x)}\dfrac{H_q(x)}{H_q(a)}$ is of the form . Therefore, by Proposition \[prop:fptchar\], $T_a - (a - x) \in \mathbb{G}_{a}(\mathbf{c}(x,a);\boldsymbol{\lambda};\boldsymbol{m})\cap \mathcal{ID}$. Next, using Lemma \[lem: below\], we have $$\mathbb{E}_x(q^{T_b} {\mathds{1}}_{\{T_b < T_{{\mathfrak{r}}}\}}) = \dfrac{H_q(x)}{H_q(b)} - \dfrac{H_q({\mathfrak{r}})}{H_q(b)} \dfrac{H_q^{b]}(x)}{H_q^{b]}({\mathfrak{r}})}.$$ We substitute the F$q$E functions from Proposition \[prop:sfmc\] and again recognize that it is of the form . By Proposition \[prop:fptchar\], $T_b \:(\textrm{resp.~}T_{b]}) \in \mathbb{G}_{{\mathfrak{r}}-1}(\boldsymbol{\tilde{c}}(x,b);\boldsymbol{\tilde{\lambda}};\boldsymbol{\tilde{m}})$.
Infinite divisibility and R-functions {#subsec:id}
-------------------------------------
In this subsection, we first review several main results in the study of infinitely divisible distribution, which will be used in analyzing the upward hitting times of $(X, {\mathbb{P}})$. We refer interested readers to [@SVH04] for a formal reference in the literature. $\phi$ is the probability generating function of an infinitely divisible distribution on $\mathbb{Z}$ if and only if for every $n \in \mathbb{N}$ there is a pgf $\phi_n$ such that for $q \in (0,1)$, $$\phi(q) = \phi_n(q)^n.$$ We first characterize infinite divisibility via canonical sequence: $\phi$ is the pgf of an infinitely divisible distribution on $\mathbb{N}_0$ with $\phi(0) > 0$ and $\phi(1) = 1$ if and only if $\phi$ has the form $$\phi(q) = \exp\bigg\lbrace-\sum_{k=0}^\infty \dfrac{r_k}{k+1}(1-q^{k+1})\bigg\rbrace,\quad 0 {\leqslant}q {\leqslant}1,$$ where $r_k {\geqslant}0$ for all $k {\geqslant}0$. The canonical sequence $(r_k)_{k \in \mathbb{N}_0}$ is unique. Apart from the canonical representation, an alternative way to characterize an infinitely divisible law is by means of the R-function. $\phi$ is the pgf of a possibly defective infinitely divisible distribution on $\mathbb{N}_0$ with $\phi(0) > 0$ and $\phi(1) = \exp\{-b\} \in (0,1], b{\geqslant}0$ ($b=0$ if the distribution is non-defective) if and only if $\phi$ has the form $$\begin{aligned}
\label{eq:rfunctdefect}
\phi(q) = \exp\bigg\lbrace -b - \int_q^1 \mathrm{R}(s) \, ds \bigg\rbrace , \quad 0 {\leqslant}q < 1,\end{aligned}$$ where R is an absolutely monotone function on $[0,1)$, that is, $R$ has non-negative derivatives of all order. R is unique and is the generating function of the canonical sequence $(r_k)_{k \in \mathbb{N}_0}$ of $\phi$.
Our first result in this section is a by-product of Theorem \[thm:green2\], where we obtain the interesting connection between absolute monotonicity and the F$q$E function $H_q$.
\[it:id\] The mapping $q \mapsto \dfrac{d}{dq} \log \left(q^{-(a-x)} \dfrac{H_q(x)}{H_q(a)}\right)$ is absolutely monotone if and only if $x {\leqslant}a$.
Assume that $x {\leqslant}a$. Then, from [@SVH04 Chapter VII Theorem 2.1] we have that $(T_a - (a-x),{\mathbb{P}}_x)$ is a positive infinitely divisible variable. Combining this fact with the identity and , we conclude that $\dfrac{d}{dq} \log \left(q^{-(a-x)} \dfrac{H_q(x)}{H_q(a)}\right)$ is an R-function. Conversely, if $x > a$, then $\dfrac{d}{dq} \log \left(q^{-(x-a)} \dfrac{H_q(a)}{H_q(x)}\right) {\geqslant}0$ is an R-function, so $\dfrac{d}{dq} \log \left(q^{-(a-x)} \dfrac{H_q(x)}{H_q(a)}\right) = - \dfrac{d}{dq} \log \left(q^{-(x-a)} \dfrac{H_q(a)}{H_q(x)}\right) < 0$ cannot be an R-function which completes the proof of the Corollary.
In the next two propositions, we proceed by offering explicit spectral formulas of R-functions and canonical sequences associated with the infinitely divisible variables $(T_a - a,{\mathbb{P}}_0)$ and $(T_{a}-(a-x),{\mathbb{P}}_x)$ for $0 {\leqslant}x {\leqslant}a$, building upon the results in Section \[subsec:fqe\], and thereby extending the work by Feller [@F66] for the case of birth-death random walk and Viskov [@Viskov00] for skip-free random walk. We start by defining a few notations. For $0 {\leqslant}j {\leqslant}a-1$, let $${\textrm{R}}_j(s) = \frac{\lambda_j}{1 - \lambda_j s}.$$ Denote the number of real (resp. complex) eigenvalues by $N_r^{0 \to a} = |\{j : \lambda_j \,\text{is real} \}|$ (resp. $N_c^{0 \to a} = |\{j : \lambda_j \,\text{is complex} \}|$).
Suppose that $(X,{\mathbb{P}})$ is irreducible, $a \in \mathbb{N}_0$, and let $\lambda_j = \lambda_j^{[0,a-1]}$. The ${\mathrm{R}}$-function of $(T_a - a,{\mathbb{P}}_0)$ is $${\mathrm{R}}^{0 \to a}(s) = \sum_{j=0}^{a-1} {\mathrm{R}}_j(s)$$ with canonical sequence $$r^{0 \to a}_k = \sum_{j=0}^{a-1} \lambda_j^{k+1} = \sum_{j = 1}^{N_r^{0 \to a}} \lambda_j^{k+1} + \sum_{j = 1}^{N_c^{0 \to a}} |\lambda_j|^{k+1} \cos{({(k+1)\mathrm{Arg}\lambda_j})} , \quad k \in \mathbb{N}_0,$$ and constant $b^{0 \to a} = - \ln {\mathbb{P}}_{0}(T_a < \infty)$.
Using the result in [@SVH04 Chapter VII Theorem 2.1], $(T_a - a,{\mathbb{P}}_0)$ is infinitely divisible. By Proposition \[prop:sfmc\], $$\begin{aligned}
\mathbb{E}_0(q^{T_a - a}) &= {\mathbb{P}}_{0}(T_a < \infty) \prod_{j = 0}^{a-1} \dfrac{(1 - \lambda_j)}{1 - \lambda_j q}
= {\mathbb{P}}_{0}(T_a < \infty) \exp\bigg \lbrace- \int_q^1 \sum_{j=0}^{a-1} {\mathrm{R}}_j(s) \, ds \bigg \rbrace,
\end{aligned}$$ where the second equality follows from identity $$\dfrac{1 - \lambda}{1 - \lambda q} = \exp \bigg \lbrace - \int_q^1 \dfrac{\lambda}{1 - \lambda s} \, ds \bigg \rbrace,$$ valid for $|\lambda|< 1/q$. Since complex eigenvalues occur in conjugate pair, we can check that $\sum_{j=0}^{a-1} {\mathrm{R}}_j(s) \in \mathbb{R}$.
\[prop:Rfunctxa\] Suppose that $0 {\leqslant}x {\leqslant}a$. Following , the ${\mathrm{R}}$-function of $(T_a - (a-x),{\mathbb{P}}_x)$ is $${\mathrm{R}}^{x \to a}(s) ={\mathrm{R}}^{0 \to a}(s) - {\mathrm{R}}^{0 \to x}(s)$$ with canonical sequence $$r^{x \to a}_k = r^{0 \to a}_k - r^{0 \to x}_k, \quad k \in \mathbb{N}_0,$$ and constant $b^{x \to a} = - \ln {\mathbb{P}}_{x}(T_a < \infty)$.
Using the result in [@SVH04 Chapter VII Theorem 2.1], $(T_a - (a-x),{\mathbb{P}}_x)$ is infinitely divisible. Let $\beta_j = \lambda_j^{[0,x-1]}$ for $j = 0,\ldots,x-1$. By Proposition \[prop:sfmc\], $$\begin{aligned}
\mathbb{E}_x(q^{T_a - (a-x)}) &= {\mathbb{P}}_{x}(T_a < \infty) \left(\prod_{j = 0}^{x-1} \dfrac{\frac{(1 - \lambda_j)}{1 - \lambda_j q}}{\frac{(1 - \beta_j)}{1 - \beta_j q}} \right) \left( \prod_{j = x}^{a-1} \frac{(1 - \lambda_j)}{1 - \lambda_j q} \right) \\
&= {\mathbb{P}}_{x}(T_a < \infty) \exp\bigg \lbrace- \int_q^1 {\mathrm{R}}^{0 \to a}(s) - {\mathrm{R}}^{0 \to x}(s) \, ds \bigg \rbrace.
\end{aligned}$$
Since $r^{x \to a}_k {\geqslant}0$ for $k \in {\mathbb{N}}_0$, we immediately obtain the following ordering of eigenvalues.
\[cor:Rfunctxa\] Suppose that $0 {\leqslant}x {\leqslant}a$, and let $\beta_j = \lambda_j^{[0,x-1]}$ for $j = 0,\ldots,x-1$. Then $$\sum_{j=0}^{a-1} \lambda_j^{k+1} {\geqslant}\sum_{j=0}^{x-1} \beta_j^{k+1} {\geqslant}0, \quad k \in \mathbb{N}_0.$$
We can rewrite the results in Proposition \[prop:Rfunctxa\] and Corollary \[cor:Rfunctxa\] in terms of trace. Define $P^{[0,a]}$ to be the restriction of $P$ from state $0$ to $a$. For $k \in \mathbb{N}_0$, we observe that $r_k^{0 \to a} = \mathrm{Tr}((P^{[0,a]})^{k+1})$, and so Corollary \[cor:Rfunctxa\] can be rewritten as $$\mathrm{Tr}((P^{a]})^{k+1}) {\geqslant}\mathrm{Tr}((P^{x]})^{k+1}).$$
### Upward hitting times
In this part, we characterize a class of discrete distribution that contains all upward hitting times within the class ${\mathcal{M}}$.
We say that a (possibly defective) distribution lies in class $\mathcal{U}$ if it is discrete with support on $\mathbb{N}_0$, infinitely divisible and each term of the canonical sequence can be written as $$\begin{aligned}
\label{eq:uphitclass}
\dfrac{r_k}{k+1} = \int_{-1}^{1} x^k w(x) \, dx, k \in \mathbb{N}_0,
\end{aligned}$$ where the mapping $w$ satisfies the integrability condition $$\int_{-1}^{1} |w(x)|\, dx < \infty.$$
Suppose that $0 {\leqslant}x {\leqslant}a$. Then, $(T_a - (a-x),{\mathbb{P}}_x)$ belongs to the class $\mathcal{U}$.
It suffices to show for $r_k^{x \to a}$. Define $$w^{0 \to a} = \sum_{j=1}^{N_r^{0 \to a}} {\mathds{1}}_{(0,\lambda_j)} + \sum_{j=1}^{N_c^{0 \to a}} {\mathds{1}}_{(0,|\lambda_j| \cos{({(k+1)\mathrm{Arg}\lambda_j})}^{1/(k+1)})},$$ then $$\begin{aligned}
\int_{-1}^{1} y^k w^{0 \to a}(y) \, dy &= \dfrac{\sum_{j=0}^{a-1} \lambda_j^{k+1}}{k+1} = \dfrac{r^{0 \to a}_k}{k+1}.
\end{aligned}$$ In addition, by the triangle’s inequality, $$\int_{-1}^{1} |w^{0 \to a}(y)|\, dy {\leqslant}\sum_{j=0}^{a-1} \lambda_j < \infty.$$ Let $w^{x \to a} = w^{0 \to a} - w^{0 \to x}$, we obtain $$\begin{aligned}
\int_{-1}^{1} y^k w^{x \to a}(y) \, dy &= \dfrac{r^{0 \to a}_k - r^{0 \to x}_k}{k+1} = \dfrac{r^{x \to a}_k}{k+1}, \\
\int_{-1}^{1} |w^{x \to a}(y)|\, dy &{\leqslant}r_0^{0 \to a} + r_0^{0 \to x} < \infty,
\end{aligned}$$ which completes the proof.
The class of skip-free Markov chains similar to birth-death chains {#sec:st}
==================================================================
In this Section, we develop an original methodology to obtain the spectral decomposition in Hilbert space of the (transition operator of) Markov chains that belong to the class ${\mathcal{S}}$ a subclass of ${\mathcal{M}}$ (the class of upward skip-free Markov chains) which is defined in Definition \[def:Sclass\] below. We recall that as the transition operator $P$ of a chain in ${\mathcal{M}}$ is usually non-self-adjoint (non-reversible) in the weighted Hilbert space $$\ell^2(\pi)=\bigg\{ f: E \mapsto \mathbb{R}; \:||f||^2_{\pi}=\sum_{x\in E} f^2(x)\pi(x)<\infty\bigg\},$$ where $\pi$ is the reference measure, there is no spectral theorem available for such bounded linear operator except for normal operators. We already point out that the two subsequent Sections contain interesting and substantial applications of this spectral decomposition namely the study of the speed of convergence to equilibrium and the separation cutoff phenomenon. We proceed by defining an equivalence relation on the set of transition matrices in ${\mathcal{M}}$.
\[def:sim\] We say that the transition matrix $P$ of a Markov chain $X \in {\mathcal{M}}$ is similar to the transition matrix of a Markov chain $Q$ on $E$, and we write $P \sim Q$, if there exits a bounded linear operator $\Lambda:\ell^2(\pi_Q) \to \ell^2(\pi)$ ($\pi_Q$ being the reference measure for $Q$) with bounded inverse such that $$P \Lambda = \Lambda Q.$$ When needed we may write $P \stackrel{\Lambda}{\sim} Q$ to specify the intertwining kernel. Note that $\sim$ is an equivalence relationship on the set of transition matrices. In the context of Markov chains theory, intertwining between birth-death processes has first been studied in [@FD].
With Definition \[def:sim\] in mind, we are now ready to define the ${\mathcal{S}}$ class.
\[def:Sclass\] Suppose that $Q \in \mathcal{B}$, the set of transition matrix $Q$ on $E$ of an irreducible (or with at most one absorbing or entrance state) birth-death chain. The similarity orbit of $Q$ (in ${\mathcal{M}}$) is $${\mathcal{S}}(Q) = \{P \in \mathcal{M};\: P \sim Q \},$$ and the ${\mathcal{S}}$ class is the union over all possible orbits $${\mathcal{S}}= \bigcup_{Q \in \mathcal{B}} {\mathcal{S}}(Q).$$ Similarly, we define $\widehat{{\mathcal{S}}}(Q)$ and $\widehat{{\mathcal{S}}}$ by replacing $P$ with $\widehat{P}$ above. Finally, we say that $X \in {\mathcal{S}}^{M}$ if $X \in {\mathcal{S}}$ and $(X,\widehat{{\mathbb{P}}})$ is stochastically monotone, i.e. $y\mapsto \widehat{{\mathbb{P}}}_y(X_1 {\leqslant}x)$ is non-increasing for every fixed $x$.
It is in general hard to find examples for the ${\mathcal{S}}$ class since $Q$ is restricted to $\mathcal{B}$ and $P$ is restricted to $\mathcal{M}$. However, if we only require $Q$ to be a reversible Markov chain, there are a few examples that have been studied in the thesis of Choi [@ChoiThesis Section $3.6$].
We remark that $\widehat{\Lambda}:\ell^2(\pi) \to \ell^2(\pi_Q)$, the adjoint of $\Lambda$, is a bounded operator with a bounded inverse as well (see e.g. [@Con90 Proposition $2.6$]). We write ${\left\lVert\cdot\right\rVert}_{op}$ to be the operator norm, i.e. ${\left\lVertP\right\rVert}_{op}=\sup_{||f||_{\pi}=1}||Pf||_{\pi}$.
Before stating the main result of this Section, we introduce the following class.
\[def:Mcclass\] We say that, for some ${\mathfrak{r}}{\geqslant}3$, $X \in {\mathcal{MC}}_{{\mathfrak{r}}}$ if $(X,{\mathbb{P}}) \in {\mathcal{M}}$ with $E=\{0,1\ldots,{\mathfrak{r}}\}$ and for every $x \in [0,{\mathfrak{r}}-1]$, its time-reversal $(X,\widehat{{\mathbb{P}}})$ satisfies
1. (stochastic monotonicity) $\widehat{{\mathbb{P}}}_{x+1}(X_1 {\leqslant}x){\leqslant}\widehat{{\mathbb{P}}}_{x}(X_1 {\leqslant}x) $,
2. (strict stochastic monotonicity) $\widehat{{\mathbb{P}}}_{x+1}(X_1 {\leqslant}x+1) < \widehat{{\mathbb{P}}}_{x}(X_1 {\leqslant}x+1), \quad x\neq {\mathfrak{r}}-1, $
3. (restricted upward jump) $\widehat{{\mathbb{P}}}_{x+1}(X_1 {\leqslant}x+k) = \widehat{{\mathbb{P}}}_x(X_1 {\leqslant}x+k), \quad x\neq {\mathfrak{r}}-1, \quad k \in [2,{\mathfrak{r}}-1-x]$.
Moreover, we say $X \in \mathcal{MC}^+_{{\mathfrak{r}}}$ if $X \in {\mathcal{MC}}_{{\mathfrak{r}}}$ and
1. (lazy Siegmund dual) $\widehat{{\mathbb{P}}}_{x}(X_1 {\geqslant}x+1)+\widehat{{\mathbb{P}}}_{x+1}(X_1 = x){\leqslant}\frac{1}{2}, \quad x=0,\ldots,{\mathfrak{r}}-1.$
When there is no ambiguity of the state space, we write ${\mathcal{MC}}= {\mathcal{MC}}_{{\mathfrak{r}}}$ (resp. ${\mathcal{MC}}^+ = {\mathcal{MC}}_{{\mathfrak{r}}}^+$).
For a numerical example illustrating the ${\mathcal{MC}}$ class, we refer the interested readers to Section \[subsec:numericalex\].
We proceed by recalling that $P$ has an $\pi$-dual or time-reversal $\widehat{P}$, that is, for $x,y \in E$, $$\pi(x) \widehat{p}(x,y) = \pi(y) p(y,x),$$ where $\pi$ is a reference measure for $P$. We equip the Hilbert space $\ell^2(\pi)$ with the usual inner product $\langle\cdot,\cdot\rangle_{\pi}$ defined by $$\langle f,g \rangle_{\pi} = \sum_{x \in E} f(x) g(x) \pi(x), \quad f,g \in \ell^{2}(\pi).$$ We also recall that a basis $(f_k)$ of a Hilbert space $\mathcal{H}$ is a Riesz basis if it is obtained from an orthonormal basis $(e_k)$ under a bounded invertible operator $T$, that is, $T e_k = f_k$ for all $k$. It can be shown, see e.g. [@Y01 Theorem $9$], that the sequence $(f_k)$ forms a Riesz basis if and only if $(f_k)$ is complete in $\mathcal{H}$ and there exist positive constants $A,B$ such that for arbitrary $n \in {\mathbb{N}}$ and scalars $c_1,\ldots,c_n$, we have $$\label{eq:Rieszb}
A \sum_{k=1}^n |c_k|^2 {\leqslant}{\left\lVert\sum_{k=1}^n c_k f_k\right\rVert}^2 {\leqslant}B \sum_{k=1}^n |c_k|^2.$$ If $(g_k)$ is a biorthogonal sequence to $(f_k)$, that is, $\langle f_k, g_m \rangle_{\pi} = \delta_{k,m}$, $k,m \in {\mathbb{N}}$ and $\delta_{k,m}$ is the Kronecker symbol, then $(g_k)$ also forms a Riesz basis.
\[thm:mcins\]
1. \[it:eig\] Assume that ${\mathfrak{r}}< \infty$ and $Q$ is the transition kernel of an irreducible birth-death process, then $P \stackrel{\Lambda}{\sim} Q$ if and only if $P$ has real and distinct eigenvalues.
2. \[it:sbd\] $(X, {\mathbb{P}}) \in {\mathcal{S}}$ with $P \stackrel{\Lambda}{\sim} Q$ if and only if $(X, \widehat{{\mathbb{P}}}) \in \widehat{\mathcal{S}}$ with $Q \stackrel{\widehat{\Lambda}}{\sim} \widehat{P}$, where $\widehat{\Lambda}$ is the adjoint operator of $\Lambda$.
Let us assume that $(X, {\mathbb{P}}) \in {\mathcal{S}}$ with $P \stackrel{\Lambda}{\sim} Q$. Then the following holds.
(a) \[it:h\] For any $h \in \mathcal{E}$ then $(X, {\mathbb{P}}^h) \in {\mathcal{S}}$ with $P^h \stackrel{\Lambda^
h}{\sim} Q$ and $\Lambda^h = D_{h}^{-1}\Lambda $, where $D_{h}$ is a diagonal matrix of $h$.
(b) \[it:normal\] If $P$ is normal, i.e. $P\widehat{P}=\widehat{P}P$, then $P$ is self-adjoint, i.e. $P=\widehat{P}$.
(c) \[it:comp\] $P$ is compact (resp. trace class) if and only if $Q$ is.
(d) \[it:spec\] Assume that $P$ is compact then for any $f \in \ell^{2}(\pi)$ and $n \in \mathbb{N}$, $$P^n f = \sum_{k=0}^{{\mathfrak{r}}} \lambda_k^n \langle f, f_k^* \rangle_{\pi} f_k,$$ where the set $(f_k)_{k=0}^{\mathfrak{r}}$ are real eigenfunctions of $P$ associated to the real eigenvalues $(\lambda_k)_{k=0}^{\mathfrak{r}}$ and form a Riesz basis of $\ell^2(\pi)$, and the set $(f_k^*)_{k=0}^{\mathfrak{r}}$ is the unique Riesz basis biorthogonal to $(f_k)_{k=0}^{\mathfrak{r}}$. In particular, for any $x,y \in E$ and $n \in \mathbb{N}$, the spectral expansion of $P$ is given by $$P^n(x,y) = \sum_{k=0}^{{\mathfrak{r}}} \lambda_k^n f_k(x) f_k^*(y)\pi(y).$$
(e) \[it:Mc\] ${\mathcal{MC}}\subseteq {\mathcal{S}}^M$.
(f) \[it:spech\] Assume that the condition of the item holds. If ${\mathfrak{r}}<\infty$ is absorbing and ${\mathfrak{l}}$ is regular, then with the same notation as above, we have $${\mathbb{P}}_x( T_{{\mathfrak{r}}} = n) = \sum_{k=0}^{{\mathfrak{r}}} \lambda_k^n(1-\lambda_k) \langle {\mathds{1}}, f_k^* \rangle_{\pi} f_k(x),$$ and assuming that ${\mathfrak{l}}<\infty$ is absorbing and ${\mathfrak{r}}<\infty$ is regular then $${\mathbb{P}}_x( T_{{\mathfrak{l}})} = n) = \sum_{k=0}^{{\mathfrak{r}}} \lambda_k^n(1-\lambda_k) \langle {\mathds{1}}, f^*_k \rangle_{\pi} f_k(x).$$
First, if $P \stackrel{\Lambda}{\sim} Q$, then $P$ has real and distinct eigenvalues since $Q$ has real and distinct eigenvalues. Conversely, if $P$ has real and distinct eigenvalues, $P$ is diagonalizable, so there exists an invertible $\Lambda$ such that $$P = \Lambda D \Lambda^{-1}.$$ where $D$ is the diagonal matrix storing the eigenvalues of $P$. Given the spectral data $D$, by the inverse spectral theorem, see e.g. [@DM76 Section 5.8], one can always construct an ergodic Markov chain with transition matrix $Q$ such that $$Q = VDV^{-1}.$$ Next, we show item . If $P \stackrel{\Lambda}{\sim} Q$, then for $f \in \ell^2(\pi_{Q})$ and $g \in \ell^2(\pi)$, $$\langle f, \widehat{\Lambda} \widehat{P} g \rangle_{\pi_{Q}} = \langle P \Lambda f, g \rangle_{\pi} = \langle \Lambda Q f, g \rangle_{\pi} = \langle f, Q \widehat{\Lambda} g \rangle_{\pi_{Q}},$$ which shows that $Q \stackrel{\widehat{\Lambda}}{\sim} \widehat{P}$. The opposite direction can be shown similarly. Item follows directly from $$P \Lambda = D_h P^h D_h^{-1} \Lambda = D_h P^h \Lambda^h = \Lambda Q.$$ For the item , we recall, from the spectral theorem, that if two normal matrices are similar then they are unitary equivalent that is $\Lambda^{-1}=\widehat{\Lambda}$. Then, the proof of this claim is completed since we easily deduce, from the item , that $$\label{eq:interadj}
P=P\Lambda \widehat{\Lambda}=\Lambda \widehat{\Lambda}\widehat{P}=\widehat{P}.$$ Next, we turn to the proof of the item . If $P$ is compact (resp. trace class), then $Q = \Lambda^{-1} P \Lambda$ is compact (resp. trace class) since the product of bounded and compact (resp. trace class) operator is a compact (resp. trace class) operator, see [@Con90 Proposition $4.2$] (resp. see [@RS80 Page $218$]). To show, in the item , that $(f_k)$ and $(f_m^*)$ are biorthogonal, we note that the fact that $P$ has distinct eigenvalues yields that $\langle f_k , f_m^* \rangle_{\pi} = \delta_{{k, m}}$ for any $k,m$. Next, denote $(g_k)$ to be the (orthogonal) eigenfunctions of $Q$, see e.g. [@LR54]. Since $f_k = \Lambda g_k$ and $\Lambda$ is bounded, $(f_k)$ is complete as $(g_k)$ is a basis. As $\Lambda$ is bounded from above and below, for any $n \in {\mathbb{N}}$ and arbitrary sequence $(c_k)_{k=1}^{n}$, we have $$\begin{aligned}
A \sum_{k=1}^n |c_k|^2 {\leqslant}{\left\lVert\sum_{k=1}^n c_k f_k\right\rVert}^2_{\pi} = {\left\lVert\Lambda \sum_{k=1}^n c_k g_k\right\rVert}^2_{\pi} {\leqslant}B \sum_{k=1}^n |c_k|^2,
\end{aligned}$$ where we can take $A = {\left\lVert\Lambda^{-1}\right\rVert}^{-2}$ and $B = {\left\lVert\Lambda\right\rVert}^2$, so is satisfied. It follows from [@Y01 Theorem $9$] that there exists the sequence $(f_k^*)$ being the unique Riesz basis biorthogonal to $(f_k)_{k=0}^{\mathfrak{r}}$, and, any $f \in \ell^2(\pi)$ can be written as $$f = \sum_{k=0}^{{\mathfrak{r}}} c_k f_k,$$ where $c_k = \langle f , f_k^* \rangle_{\pi}$. Desired result follows by applying $P^n$ to $f$ and using $P^n f_k = \lambda_k^n f_k$. In particular, if we take $f = \delta_y$, the Dirac mass at $y$, and evaluate the resulting expression at $x$, we obtain the spectral expansion of $P$. To show , we write $\widetilde{P}$ the so-called Siegmund dual (or $H_S$-dual) of $\widehat{P}$. That is, $\widetilde{P}^T = H_S^{-1} \widehat{P} H_S$ where $H_S = (H_S(x,y))_{x,y \in E}$ is defined to be $H_S(x,y) = {\mathds{1}}_{\{x {\leqslant}y\}}$ and its inverse $H_S^{-1}=(H^{-1}_S(x,y))_{x,y \in E}$ is $H_S^{-1}(x,y) = {\mathds{1}}_{\{x = y\}} - {\mathds{1}}_{\{x = y - 1\}}$, see [@Sieg76]. Since $X \in {\mathcal{MC}}$, then $\widehat{P}$ is stochastically monotone, hence from [@ASM03 Proposition 4.1], we have that $\widetilde{P}$ is a sub-Markovian kernel. For $x \in [1,{\mathfrak{r}}-1]$, $\widetilde{p}(x,x+1) = \widehat{p}(x+1,x) > 0$ since $\widehat{P}$ is irreducible and downward skip-free, and $\widetilde{p}(x,y) = 0 \quad \forall y {\geqslant}x+2$. We also have $\widetilde{p}(0,1) = \widehat{p}(1,0) > 0$. For $x \in [0,{\mathfrak{r}}-2]$, condition $(2)$ in ${\mathcal{MC}}$ gives that $\widetilde{p}(x,x-1) > 0$, while condition $(3)$ in ${\mathcal{MC}}$ guarantees that $\widetilde{p}(x,y) = 0$ for each $x \in [0,{\mathfrak{r}}-1]$ and $y \in [0,x-2]$. That is, $\widetilde{P}$ is a (strictly substochastic) irreducible birth-death chain when restricted to the state space $[0,{\mathfrak{r}}-1]$. Denote $\widetilde{P}^{bd}$ the restriction of $\widetilde{P}$ to $[0,{\mathfrak{r}}-1]$. By breaking off the last row and last column of $\widetilde{P}$, we can write $$\begin{aligned}
\label{eq:hatPbd}
\widetilde{P} = \left( \begin{array}{cc}
\widetilde{P}^{bd} & \mathbf{v} \\
\mathbf{0} & 1 \\
\end{array} \right) = (H_S^{-1} \widehat{P} H_S)^T ,
\end{aligned}$$ where $\mathbf{0}$ is a row vector of zero, and $\mathbf{v}$ is a column vector storing $\widetilde{p}(x,{\mathfrak{r}})$ for $x \in [0,{\mathfrak{r}}-1]$. Observing that the last row of $\widetilde{P}$ is zero except the last entry, we have $$M^{[ 0,{\mathfrak{r}}-1 ]} = \Lambda^{[0,{\mathfrak{r}}-1]} \widetilde{P}^{bd}.$$ Note that $\widetilde{P}^{bd}$ is a strictly substochastic matrix with ${\mathfrak{r}}$ being a killing boundary. Denote $\widetilde{T}^{bd}$ to be the lifetime of Markov chain with transition kernel $\widetilde{P}^{bd}$. However, defining, with the obvious notation, for any $x \in [ 0,{\mathfrak{r}}-1 ]$, $$\widetilde h(x) = \mathbb{P}_x(\widetilde{T}^{bd}_{{\mathfrak{r}}-1} < \widetilde{T}^{bd}),$$ we have, according to Theorem \[thm:green2\], that $\widetilde h$ is an harmonic function for $\widetilde{P}^{bd}$, i.e. $\widetilde{P}^{bd} \widetilde h =\widetilde h$. Hence, a standard result in Martin boundary theory, see e.g. Theorem \[thm:htransp\], entails that the Markov chain with transition kernel $Q$, defined on $[ 0,{\mathfrak{r}}-1 ]\times [ 0,{\mathfrak{r}}-1 ]$ by $Q(x,y) = \frac{\widetilde h(y)}{\widetilde h(x)}\widetilde{P}^{bd}(x,y)$, is an ergodic birth-death chain, which completes the proof. Finally to show item , after observing that, for any $n \in \mathbb{N} $ and $x \in E$, $${\mathbb{P}}_x(T_{\mathfrak{r}}>n ) = \sum_{y \in E}{\mathbb{P}}_x(X_n=y,T_{\mathfrak{r}}>n )=P{\mathds{1}}(x)$$ the first representation in with $f\equiv 1$, $n=1$ and easy algebra yields the desired result. The last claim follows by similar means.
Convergence to equilibrium
==========================
As a first application of the spectral decomposition stated in Theorem \[thm:mcins\], we derive accurate information regarding the speed of convergence to stationarity for ergodic chains in ${\mathcal{S}}$ in both the Hilbert space topology and in total variation distance. There have been a rich literature devoted to the study of convergence to equilibrium for non-reversible chains, see e.g. [@Fill91; @LSC97] and the references therein. In these papers, to overcome the lack of a spectral theory, the authors resort to reversibilization procedures to extract bounds for the distance to stationarity. Another popular approach for studying the rate of convergence is by coupling techniques [@Thorisson00]. Our approach reveals a natural extension to the non-reversible case of the classical spectral gap that appears in the study of reversible chains, see e.g. [@LSC97]. To state our result we need to introduce some notation. We denote the second largest eigenvalue in modulus (SLEM) or the spectral radius of $P$ in the Hilbert space $\ell^2_0(\pi) = \{ f \in \ell^2(\pi); \:~\langle {\mathds{1}},f \rangle_{\pi} = 0 \}$, by $\lambda_{*} = \lambda_{*}(P) = \sup\{|\lambda_i|;~ \lambda_i \neq 1\}$, then the *absolute spectral gap* is $\gamma_* = 1 - \lambda_*$. For any two probability measures $\mu, \nu$ on $E$, the total variation distance between $\mu$ and $\nu$ is given by $$|| \mu - \nu ||_{TV} = \dfrac{1}{2} \sum_{x \in E} |\mu(x) - \nu(x)|.$$ For $n \in \mathbb{N}$, the total variation distance from stationarity of $X$ is $$d(n) = \max_{x \in E} || \delta_x P^n - \pi ||_{TV}.$$ For $g \in \ell^2(\pi)$, the mean of $g$ with respect to $\pi$ can be written as ${\mathbb{E}}_{\pi}(g) = \langle g,{\mathds{1}}\rangle_{\pi}$. Similarly, the variance of $g$ with respect to $\pi$ is $\mathrm{Var}_{\pi}(g) = \langle g,g \rangle_{\pi} - {\mathbb{E}}_{\pi}^2(g) $. Finally, we recall that Fill in [@Fill91 Theorem $2.1$] obtained when ${\mathfrak{r}}<\infty$ the following bound valid for all $n \in \mathbb{N}_0$ $$\begin{aligned}
\label{eq:tvfill}
d(n) {\leqslant}\dfrac{\sigma_*(P)^n}{2} \sqrt{\dfrac{1-\pi_{min}}{\pi_{min}}},\end{aligned}$$ where $\pi_{min} = \min\limits_{x \in E} \pi(x)$ and $\sigma_*(P) = \sqrt{\lambda_*(P \widehat{P})}$ is the second largest singular value of $P$. We obtain the following refinement for Markov chains in the class ${\mathcal{S}}$.
\[thm:spectralexp\] Let $X \in {\mathcal{S}}$ and $X$ is ergodic with stationary distribution $\pi$ and $P \stackrel{\Lambda}{\sim} Q$.
1. \[it:varbd\] For any $n \in \mathbb{N}_0$, we have $$\begin{aligned}
\label{eq:hyp}
\lambda_*^n{\leqslant}{\left\lVertP^n - \pi\right\rVert}_{\ell^2(\pi) \to \ell^2(\pi)} {\leqslant}\sigma^{n}_*(P){\mathds{1}}_{\{n<n^*\}} +\kappa(\Lambda) \lambda_*^n{\mathds{1}}_{\{n{\geqslant}n^*\}}
\end{aligned}$$ where $n^* = \lceil \frac{\ln \kappa(\Lambda)}{\ln \sigma_*(P)-\ln\lambda_*} \rceil$ and $\kappa(\Lambda)= {\left\lVert\Lambda\right\rVert}_{op} \: {\left\lVert\Lambda^{-1}\right\rVert}_{op} {\geqslant}1$ is the condition number of $\Lambda$.
2. \[it:v\] If ${\mathfrak{r}}<\infty$ then $P$ is non-reversible if and only if $\kappa(\Lambda)>1$.
3. \[it:st\] A sufficient condition for which $\lambda_* < \sigma_*(P)$ is given by $\max_{i \in E} p(i,i) > \lambda_*$. In such case, for $n$ large enough, the convergence rate $\lambda_*$ given in item is strictly better than the reversibilization rate $\sigma_*(P)$.
4. \[it:totalvar\] Suppose now that ${\mathfrak{r}}< \infty$. Then, for any $n \in \mathbb{N}_0$, $$\begin{aligned}
d(n) {\leqslant}\dfrac{\min\left(\sigma^{n}_*(P),{\kappa(\Lambda)} \lambda_*^n \right)}{2} \sqrt{\dfrac{1-\pi_{min}}{\pi_{min}}},
\end{aligned}$$ where $\lambda_{*}{\leqslant}\sigma_*(P)$.
It is interesting to recall that when $P$ is reversible and compact then the sequence of eigenfunctions is orthonormal and thus an application of the Parseval identity yields ${\left\lVertP^n - \pi\right\rVert}_{\ell^2(\pi) \to \ell^2(\pi)} = \lambda_*^n$ and $\kappa(\Lambda) = 1$ which is a specific instance of item and .
We also recall the discrete analogue of the notion of hypocoercivity introduced in [@V09], i.e. there exists a constant $C < \infty$ and $\rho \in (0,1)$ such that, for all $ n \in \mathbb{N}$, $${\left\lVertP^n - \pi\right\rVert}_{\ell^2(\pi) \to \ell^2(\pi)} {\leqslant}C \rho^n.$$ Note that, in general, these constants are not known explicitly. We observe that the upper bound in reveals that the ergodic chains in ${\mathcal{S}}$ satisfy this hypocoercivity phenomena. More interestingly, our approach based on the similarity concept enables us to get on the one hand an explicit and on the other hand a spectral interpretation of this rate of convergence. Indeed, it can be understood as a modified spectral gap where the perturbation from the classical spectral gap is given by the condition number $\kappa(\Lambda)$ which may be interpreted as a measure of deviation from symmetry. In this vein, we mention the recent work [@Patie-Savov] where a similar spectral interpretation of the hypocoercivity phenomena is given for a class of non-self-adjoint Markov semigroups.
We first show the upper bound in . Define the synthesis operator $T^* : \ell^2 \to \ell^2(\pi)$ by $\alpha = (\alpha_i) \mapsto T^*(\alpha) = \sum_{i = 0}^{{\mathfrak{r}}} \alpha_i f_i$, where $(f_i)$ are the eigenfunctions of $P$ and $(f_i^*)$ are the unique biorthogonal basis of $(f_i)$ as in Theorem \[thm:mcins\]. For $1 {\leqslant}i {\leqslant}{\mathfrak{r}}$, we take $\alpha_i = \lambda_i^n \langle g,f_i^* \rangle_{\pi}$, and denote $(q_i)$ to be the orthonormal eigenfunctions of $Q$, where $f_i = \Lambda q_i$. Note that $||T^*||_{op} {\leqslant}||\Lambda||_{op} < \infty$, since $$||T^*(\alpha)|| = {\left\lVert\sum_{i = 0}^{{\mathfrak{r}}} \alpha_i \Lambda q_i\right\rVert} {\leqslant}{\left\lVert\Lambda\right\rVert}_{op} {\left\lVert\sum_{i = 0}^{{\mathfrak{r}}} \alpha_i q_i\right\rVert}_{\pi_Q} {\leqslant}||\Lambda||_{op} ||\alpha||_{\ell^2}.$$ For $g \in \ell^2(\pi)$, we also have $$\begin{aligned}
\sum_{i=0}^{{\mathfrak{r}}} |\langle g,f_i^* \rangle_{\pi}|^2 &= \sum_{i=0}^{{\mathfrak{r}}} |\langle g, (\Lambda^{*})^{-1}q_i \rangle_{\pi}|^2
= \sum_{i=0}^{{\mathfrak{r}}} |\langle \Lambda^{-1}g, q_i \rangle_{\pi_Q}|^2
= ||\Lambda^{-1}g||^2_{\pi_Q}
{\leqslant}||\Lambda^{-1}||^2_{op} ||g||^2_{\pi},
\end{aligned}$$ where the third equality follows from Parseval’s identity, which leads to $$\begin{aligned}
\label{eq:l2}
|| P^ng - \pi g||_{\pi}^2 = ||T^*(\alpha)||^2 {\leqslant}||\Lambda||_{op}^2 ||\alpha||_{l^2}^2 {\leqslant}||\Lambda||_{op}^2 ||\Lambda^{-1}||_{op}^2 \lambda_*^{2n} ||g||^2_{\pi}.
\end{aligned}$$ Desired upper bound follows from and $${\left\lVertP^n - \pi\right\rVert}_{\ell^2(\pi) \to \ell^2(\pi)} {\leqslant}\lambda_*(\widehat{P}P)^{n/2} = \lambda_*(P\widehat{P})^{n/2},$$ see e.g. [@LSC97]. The lower bound in follows readily from the well-known result that the $n^{th}$ power of the spectral radius $\lambda_*^n$ is less than or equal to the norm of $P^n$ on the reduced space $\ell^2_0(\pi)$. Next, we show that $P$ is non-self-adjoint if and only if $\kappa(\Lambda) > 1$. We recall from that $PA=A\widehat{P}$ where $A=\Lambda \widehat{\Lambda}$ is a positive self-adjoint matrix. Thus, by the spectral theorem there exists $T \in \mathrm{GL}_{\mathfrak{r}}$, the general linear group of dimension ${\mathfrak{r}}$, such that $A=T D_A \widehat{T}$ with $D_A$ the diagonal matrix of its eigenvalues where since $\Lambda$ is defined up to a non-zero multiplicative constant we can assume, without loss of generality, that its largest eigenvalue $\lambda_1(A)$=1. Next, we recall from [@HJ13 p. 382] that $$\label{eq:defcond}
\kappa(\Lambda)=\frac{\sigma_{1}(\Lambda)}{\sigma_{{\mathfrak{r}}}(\Lambda)}=\frac{\sqrt{\lambda_1(A)}}{\sqrt{\lambda_{\mathfrak{r}}(A)}},$$ where $\sigma_{1}(\Lambda)$ (resp. $\sigma_{{\mathfrak{r}}}(\Lambda)$) is the largest (resp. smallest) singular value of $\Lambda$ and $\lambda_{\mathfrak{r}}(A)$ is the smallest eigenvalue which is positive as $\Lambda \in \mathrm{GL}_{\mathfrak{r}}$ and hence $A \in \mathrm{GL}_{\mathfrak{r}}$. Thus, $\kappa(\Lambda)=1$ implies that $D_A=I_{{\mathfrak{r}}}$ where $I_{\mathfrak{r}}$ is the identity matrix, that is $\Lambda \widehat{\Lambda} =A=I_{\mathfrak{r}}$ and, from , we deduce that $P$ is self-adjoint. Conversely, if $P$ is self-adjoint then by means of the same argument used for the proof of Theorem \[thm:mcins\], we have that $\Lambda$ is unitary and hence $\Lambda \widehat{\Lambda} =I_{\mathfrak{r}}$, that is $\kappa(\Lambda)=1$, which completes the proof of this statement. The claim in is a straightforward consequence of the Sing-Thompson theorem [@Thompson77]. Next, using , we get $$\begin{aligned}
\label{eq:vareig}
\mathrm{Var}_{\pi} \left({\widehat{P}}^n g \right) &{\leqslant}\kappa(\widehat{\Lambda})^2 \lambda_*^{2n} \mathrm{Var}_{\pi}(g) = \kappa(\Lambda)^2\lambda_*^{2n} \mathrm{Var}_{\pi}(g), \quad n \in \mathbb{N}_0,\end{aligned}$$ where we used the obvious identity $\kappa(\Lambda) = \kappa(\widehat{\Lambda})$ in the equality. This leads to $$\begin{aligned}
|| \delta_x P^n - \pi ||_{TV}^2 &= \dfrac{1}{4} {\mathbb{E}}_{\pi}^2 \left| \dfrac{\delta_x P^n}{\pi} - 1 \right|
{\leqslant}\dfrac{1}{4} \mathrm{Var}_{\pi} \left( \dfrac{\delta_x P^n}{\pi} \right)
= \dfrac{1}{4} \mathrm{Var}_{\pi} \left({\widehat{P}}^n \frac{\delta_x}{\pi} \right)
{\leqslant}\dfrac{1}{4} \kappa(\Lambda)^2 \lambda_{*}^{2n} \mathrm{Var}_{\pi} \left( \frac{\delta_x}{\pi} \right) \\
&= \dfrac{1}{4} \kappa(\Lambda)^2 \lambda_{*}^{2n} \dfrac{1 - \pi(x)}{\pi(x)} {\leqslant}\dfrac{1}{4} \kappa(\Lambda)^2 \lambda_{*}^{2n} \dfrac{1 - \pi_{min}}{\pi_{min}},\end{aligned}$$ where the first inequality follows from Cauchy-Schwartz inequality. The proof is completed by combining the above bound with .
A numerical example {#subsec:numericalex}
-------------------
To illustrate the previous result, we consider the following example in the ${\mathcal{MC}}$ class where the dual transition matrix is given by $$\label{eq:example}
\widehat{P} = \begin{pmatrix}
0.275 & 0.7 & 0.005 & 0.02 \\
0.17 & 0.8 & 0.01 & 0.02 \\
0 & 0.94 & 0.02 & 0.04 \\
0 & 0 & 0.95 & 0.05 \\
\end{pmatrix}.$$ Table \[tab:example\] shows the rate of convergence of $P$ towards $\pi=(0.18, 0.77,0.03,0.02)$ with $\widehat{P}$ given in . We observe that for $n = 1,2$, the reversibilization bound $\lambda_{*}(P\widehat{P})^{n/2}$ is smaller while for $n {\geqslant}3$, our upper bound in Theorem \[thm:spectralexp\] $\kappa_{\Lambda} \lambda_*^n$ is smaller. Also, we point out that since $\max_i p(i,i) = 0.8 > \lambda_* = 0.17$, the Sing-Thompson condition in item of Theorem \[thm:spectralexp\] holds.
$n$ $||P^n - \pi||_{\ell^2(\pi) \to \ell^2(\pi)}$ $\kappa(\Lambda) \lambda_*^n$ $\lambda_{*}(P\widehat{P})^{n/2}$
----- ----------------------------------------------- ------------------------------- -----------------------------------
1 0.81 8.64 0.81
2 0.08 1.42 0.65
3 0.02 0.24 0.52
4 0.003 0.04 0.42
5 0.0005 0.006 0.34
: $\ell^2(\pi)$-rate of convergence of $P$
\[tab:example\]
Separation cutoff {#sec:sc}
=================
As another illustration of the similarity concept, we aim at generalizing, to the non-reversible chains in the class $\mathcal{S}^M$, the separation cutoff criteria established by Diaconis and Saloff-Coste in [@DSC06] and Chen and Saloff-Coste in [@CSC15] for reversible birth-death chains. We also offer an alternative necessary and sufficient condition as obtained by Mao et al. [@MZZ16] recently for continuous-time upward skip-free chain with stochastic monotone time-reversal. The term “cutoff phenomenon" was first formally introduced by Aldous and Diaconis in [@AldousDiaconis86], and total variation cutoff for birth-death chains have been studied by Ding et al. [@DLP10]. To this end, we recall the definition of separation distance of Markov chains, which is used as a standard measure for convergence to equilibrium. For $n \in {\mathbb{N}}$, the maximum separation distance $s(n)$ is defined by $$s(n) = \max_{x,y \in E} \left[ 1 - \dfrac{P^n(x,y)}{\pi(y)}\right] = \max_{x \in E}\, \mathrm{sep}(P^n(x,\cdot),\pi) = \max_{x \in E} s_x(n).$$
Note that separation distance is not a metric. One of its nice feature is its connection to strong stationary times that we now describe. We say that a strong stationary time $T$ for a Markov chain $X$ with stationary distribution $\pi$ is a randomized stopping time $T$, possibly depending on the initial starting position $x$, if, for all $x,y \in E$, $${\mathbb{P}}_x(T = n, X_T = y) = {\mathbb{P}}_x(T = n) \pi(y).$$ The fastest strong stationary time is a strong stationary time such that for all $n \in {\mathbb{N}}$, $s_x(n) = {\mathbb{P}}(T > n)$. We now provide a description of the cutoff phenomenon for Markov chains. Recall that the separation mixing times are defined, for any $x \in E$ and $\epsilon>0$, as $$T^s(x,\epsilon) = \min \{ n {\geqslant}0;~\mathrm{sep}(P^n(x,\cdot),\pi) {\leqslant}\epsilon\}$$ and $$T^s(\epsilon) = \min \{ n {\geqslant}0;~s(n) {\leqslant}\epsilon\}.$$ A family, indexed by $n \in {\mathbb{N}}$, of ergodic chains $X^{(n)}$ defined on $E_{{\mathfrak{r}}_n} = \{0,\ldots,{\mathfrak{r}}_n\}$ with transition matrix $P_n$, stationary distribution $\pi_n$ and separation mixing times ${\rm{T}}_n(\epsilon)=T^s_n(\epsilon)$ or $T^s_n(x,\epsilon)$, for some $x\in E$, is said to present a separation cutoff if there is a positive sequence $(t_n)$ such that for all $\epsilon \in (0,1)$, $$\lim_{n \to \infty} \dfrac{\rm{T}_n(\epsilon)}{t_n} = 1.$$ The family has a $(t_n, b_n)$ separation cutoff if the sequences $(t_n)$ and $(b_n)$ are positive, $b_n/t_n \to 0$ and for all $\epsilon \in (0,1)$, $$\limsup_{n \to \infty} \dfrac{|{\rm{T}}_n(\epsilon) - t_n|}{b_n} < \infty.$$ Let us now write $\mathrm{R}_n=(I - P_n)^{-1}_{\ell^2_0}$ for the centered resolvent, that is, the resolvent operator restricted to $l^2_0(\pi)$. The main result of this section is the following.
\[thm:sepcutoffsf\] Suppose that, for each $n {\geqslant}1$, $X^{(n)} \in {\mathcal{S}}_{{\mathfrak{r}}_n}^{M}$ and let $(\theta_{n,i})_{i=1}^{{\mathfrak{r}}_n}$ be the non-zero eigenvalues of $I - P_n$. Define $$\underline{\theta}_n = \min_{1 {\leqslant}i {\leqslant}{\mathfrak{r}}_n} \theta_{n,i}, \quad \rho_n^2 = \sum_{i=1}^{{\mathfrak{r}}_n} \dfrac{1-\theta_{n,i}}{\theta_{n,i}^2}.$$ Then the family of chains $(X^{(n)})$ with transition kernel $(P_n)$, all started from $0$, has a separation cutoff if and only if $\mathrm{Tr}(\mathrm{R}_n) \underline{\theta}_n \to \infty$ if and only if $T_n^s(0,\epsilon) \underline{\theta}_n \to \infty$. In this case there is a $(\mathrm{Tr}(\mathrm{R}_n), \max\{\rho_n,1\})$ separation cutoff.
We begin the proof of Theorem \[thm:sepcutoffsf\] by giving an important lemma that gives a lower bound on the mixing time in terms of the eigenvalues of $I - P$. The corresponding result for reversible and ergodic Markov chain can be found in [@LPW09 Theorem $12.4$].
\[lem:lowerbdmix\] For an ergodic chain $X \in {\mathcal{S}}$ and $\epsilon \in (0,1)$, denote by $(\theta_i)_{i = 0}^{{\mathfrak{r}}}$ the eigenvalues of $I - P$ arranged in ascending order and $\underline{\theta} = \min_{i \neq 0} \theta_i$. We have $$T^s(\epsilon) {\geqslant}\left(\underline{\theta}^{-1}-1 \right) \log \left( \dfrac{1}{2\epsilon}\right).$$
Take $f$ to be an eigenfunction of $P$ associated to an arbitrary eigenvalue $\lambda \neq 1$, where $\lambda \in (\lambda_i)_{i=0}^{{\mathfrak{r}}}$ which is the set of (real) eigenvalues of $P$ arranged in descending order. Note that $f$ is orthogonal to ${\mathds{1}}$, since $$\langle Pf,{\mathds{1}}\rangle_{\pi} = \lambda \langle f,{\mathds{1}}\rangle_{\pi} = \langle f,\widehat{P}{\mathds{1}}\rangle_{\pi} = \langle f,{\mathds{1}}\rangle_{\pi},$$ where the last equality follows from stochasticity of $\widehat{P}$. By writing $||f||_{\infty} = \max_{x \in E} |f(x)| = f(x^*)$, we have $$|\lambda^n f(x)| = |P^n f(x)| = \left| \sum_{y \in E} P^n(x,y)f(y) - \pi(y)f(y) \right| {\leqslant}2 ||f||_{\infty} ||P^n(x,\cdot) - \pi||_{\mathrm{TV}} {\leqslant}2 ||f||_{\infty} s(n),$$ where the first inequality follows from the definition of total variation distance $||P^n(x,\cdot) - \pi||_{\mathrm{TV}}$, and the second inequality comes from the result that total variation distance is less than or equal to separation distance, see e.g. [@LPW09 Lemma $6.13$]. Taking $n = T^s(\epsilon)$ and $x = x^*$, the above yields $|\lambda|^{T^s(\epsilon)} {\leqslant}2 \epsilon$, which leads to $$\frac{1-|\lambda|}{|\lambda|} T^s(\epsilon){\geqslant}- \log(|\lambda|) T^s(\epsilon) {\geqslant}- \log( 2\epsilon ).$$ The proof is completed by specializing to $|\lambda| = \max\{|\lambda_1|,|\lambda_{{\mathfrak{r}}}|\}$.
The next lemma gives a necessary condition for separation cutoff in terms of the spectral information.
\[lem:sepcutoffnec\] Suppose that, for each $n {\geqslant}1$, $X^{(n)} \in {\mathcal{M}}_{{\mathfrak{r}}_n}$ and let $(\theta_{n,i})_{i=1}^{{\mathfrak{r}}_n}$ be the eigenvalues of $I - P_n$. For the $n^{th}$ chain in the family, define $\underline{\theta}_n = \min_{1 {\leqslant}i {\leqslant}{\mathfrak{r}}_n} \theta_{n,i}$ and $T_n^s(\epsilon)$ to be the separation mixing time. If the family of chain with transition kernel $(P_n)$ exhibits a separation cutoff, then $T_n^s(\epsilon) \underline{\theta}_n \to \infty$.
Let $T_n^s = T_n^s(0.25)$, that is, choosing $\epsilon = 0.25$. If $T_n^s(\epsilon) \underline{\theta}_n$ is bounded above in $n$ by $c>0$, then by Lemma \[lem:lowerbdmix\] we have $$\dfrac{T_n^s(\epsilon)}{T_n^s} {\geqslant}\dfrac{\underline{\theta}_n^{-1}-1 }{T_n^s}\log \left( \dfrac{1}{2\epsilon}\right) {\geqslant}c \log \left( \dfrac{1}{2\epsilon}\right),$$ so $\dfrac{T_n^s(\epsilon)}{T_n^s} \to \infty$ as $\epsilon \to 0$, which implies that there is no separation cutoff.
We are now ready to complete the proof of Theorem \[thm:sepcutoffsf\]. Denote by $P_n^k$ the distribution of the $n^{th}$ chain at time $k$, and by $\pi_n$ the stationary measure of the $n^{th}$ chain. We also write $t_n = \mathrm{Tr}(\mathrm{R}_n)$. It is known, see e.g. [@Fill Theorem $1.4$], that $$\mathrm{sep}(P_n^k,\pi_n) = {\mathbb{P}}(T_n > k),$$ where $T_n$ is the fastest strong stationary time of the $n^{th}$ chain, which is equal in distribution to a ${\mathfrak{r}}_n$-fold convolution of geometric random variables each with success probability $\theta_{n,i}$ for $i = 1,\ldots,{\mathfrak{r}}_n$, mean $t_n$ and variance $\rho_n^2$. The key to establish the proof is the following. $$\begin{aligned}
\label{eq:eigcompare}
\rho_n^2 = \underline{\theta}_n^{-2} \sum_{i=1}^{{\mathfrak{r}}_n} \dfrac{\left(1-\theta_{n,i}\right) \underline{\theta}_n^2}{\theta_{n,i}^2} {\leqslant}\underline{\theta}_n^{-2}\sum_{i=1}^{{\mathfrak{r}}_n} \dfrac{\underline{\theta}_n}{\theta_{n,i}} = \underline{\theta}_n^{-1} t_n,
\end{aligned}$$ where we use the facts that $\theta_{n,i} {\geqslant}0$ and $\underline{\theta}_n/\theta_{n,i} {\leqslant}1$ in the inequality. Assume that $t_n \underline{\theta}_n \to \infty$, which together with yields $\rho_n/t_n \to 0$. The rest of the argument are similar to the ones developed in the proof of [@DSC06 Theorem $5.1$]. For sake of completeness we now provide its main ingredients. First, by means of Chebyshev’s inequality, we have $$\label{eq:cheby}
t_n - (\epsilon^{-1} - 1)^{1/2} \rho_n {\leqslant}T^s_n(0,\epsilon) {\leqslant}t_n + (\epsilon^{-1} - 1)^{1/2} \rho_n.$$ This shows that the family of chain exhibits a separation cutoff if we divide by $t_n$ and take $n \to \infty$. On the other hand, separation cutoff implies $T_n^s(0,\epsilon) \underline{\theta}_n \to \infty$ by Lemma \[lem:sepcutoffnec\], so it remains to show $T_n^s(0,\epsilon) \underline{\theta}_n \to \infty$ implies $t_n \underline{\theta}_n \to \infty$. Using $t_n \underline{\theta}_n {\geqslant}1$, yields $\rho_n {\leqslant}t_n$, and together with the upper bound of leads to $$T^s_n(0,\epsilon) {\leqslant}t_n + (\epsilon^{-1} - 1)^{1/2} \rho_n {\leqslant}t_n (1 + (\epsilon^{-1} - 1)^{1/2}),$$ so $t_n \underline{\theta}_n \to \infty$ holds if and only if $T_n^s(0,\epsilon) \underline{\theta}_n \to \infty$. It follows from [@CSC15 Remark $1.1$] that there is a $(t_n, \max\{\rho_n,1\})$ separation cutoff. Precisely, gives $$|T_n^s(0,\epsilon) - t_n| {\leqslant}(\epsilon^{-1} - 1)^{1/2} \rho_n + 1,$$ and a $(t_n, \max\{\rho_n,1\})$ cutoff is observed by noting that $\theta_{n,i} {\leqslant}2$, $t_n {\geqslant}n/2$ and $\rho_n/t_n \to 0 $.
$\ell^p$-cutoff
---------------
We proceed by investigating the $\ell^p$-cutoff for fixed $p \in (1,\infty)$ for the class ${\mathcal{S}}$. Recall that Chen and Saloff-Coste [@CSC07 Theorem $4.2,4.3$] have shown that for a family of *normal* ergodic transition kernel $P_n$, the max-$\ell^p$ cutoff is equivalent to the [*spectral gap times mixing time*]{} going to infinity. We can extend their result to the case of the non-normal chains in $ {\mathcal{S}}$ as follows.
Suppose that, for each $n {\geqslant}1$, $X^{(n)} \in \mathcal{S}_{{\mathfrak{r}}_n}$ with compact transition kernel $P_n \stackrel{\Lambda_n}{\sim} Q_n$ and stationary measure $\pi_n$, and let $\lambda_{n,*}$ be the second largest eigenvalue in modulus of $P_n$. Assume that $$\sup_{n {\geqslant}1} ||\Lambda_n||_{op} \: ||\Lambda^{-1}_n||_{op} < \infty.$$ Fix $p \in (1,\infty)$ and $\epsilon > 0$. Consider the max-$\ell^p$ distance to stationarity $$f_n(t) = \sup_{x \in E_{{\mathfrak{r}}_n}} {\left\lVert\dfrac{p^t_n(x,\cdot)}{\pi_n} - 1\right\rVert}_{\ell^p(\pi)}$$ and define $$t_n = \inf\{t > 0;\: f_n(t) {\leqslant}\epsilon\}, \quad \theta_{n,*} = - \log \lambda_{n,*} \quad \textrm{and } \mathcal{F} = \{f_n;\: n=1,2,\ldots\}.$$ Assume that each $n$, $f_n(t) \to 0$ as $t \to \infty$ and $t_n \to \infty$. Then the family $\mathcal{F}$ has a max-$\ell^p$ cutoff if and only if $t_n \theta_{n,*} \to \infty$. In this case there is a $(t_n, \max\{1,\theta_{n,*}^{-1}\})$ cutoff.
The proof in [@CSC07 Theorem $4.2,4.3$] works nicely as long as we have Lemma \[lem:l2lp\] below, which gives a two-sided control on the $\ell^p(\pi)$ norm of $P^n - \pi$. The following lemma is then the key to the proof.
\[lem:l2lp\] Suppose that $X \in {\mathcal{S}}$ on $E$ with transition kernel $P \stackrel{\Lambda}{\sim} Q$. Fix $p \in (1,\infty)$. Then, for any $n \in \mathbb{N}$, we have $$\begin{aligned}
2^{-1+\theta_p} \lambda_*(P)^{n\theta_p} &{\leqslant}{\left\lVertP^n - \pi\right\rVert}_{\ell^p(\pi) \to \ell^p(\pi)} {\leqslant}2^{|1-2/p|} (\kappa(\Lambda) \lambda_*(P)^n)^{1-|1-2/p|}, \label{eq:lp}
\end{aligned}$$ where $\theta_p \in [1/2,1]$ and $\kappa(\Lambda) = {\left\lVert\Lambda\right\rVert}_{op} \: ||\Lambda^{-1}||_{op}$.
By the Riesz-Thorin interpolation theorem, see e.g. [@CSC07 equation $3.4$], we have $${\left\lVertP^n - \pi\right\rVert}_{\ell^p(\pi) \to \ell^p(\pi)} {\leqslant}2^{|1-2/p|} {\left\lVertP^n - \pi\right\rVert}_{\ell^2(\pi) \to \ell^2(\pi)}^{1-|1-2/p|},$$ which when combined with Theorem \[thm:spectralexp\] gives the upper bound of . Next, to show the lower bound in , we use another version of the Riesz-Thorin interpolation theorem, see e.g. [@CSC07 Lemma $4.1$], to get $${\left\lVertP^n - \pi\right\rVert}_{\ell^p(\pi) \to \ell^p(\pi)} {\geqslant}2^{-1+\theta_p} {\left\lVertP^n - \pi\right\rVert}_{\ell^2(\pi) \to \ell^2(\pi)}^{\theta_p} {\geqslant}2^{-1+\theta_p} \lambda_*(P)^{n\theta_p},$$ where we use Theorem \[thm:spectralexp\] in the second inequality. This completes the proof.
[^1]: The authors would like to thank an anonymous referee for his constructive comments and suggestions that improve substantially the quality of the paper. The authors are also grateful to Jim Dai, Daniel Jerison, Laurent Miclo, Jim Pitman and Laurent Saloff-Coste for stimulating discussions on several aspects of this work. This work was partially supported by NSF Grant DMS-1406599 and ARC IAPAS, a fund of the Communautée francaise de Belgique. The second author is grateful for the hospitality of the LMA at the UPPA, where part of this work was completed.
[^2]: [*Keywords:* Markov chains, Potential theory, Martin boundary, Fluctuation theory, Spectral theory, Non-self-adjoint operator, rate of convergence, cutoff.]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The LHC collaborations ATLAS and CMS recently reported on the excess of the events in the diphoton final states at the invariant mass of about $750~\text{GeV}$. In this article we speculate on the possibility that the excess arises from the neutral CP-even component $\phi $ of the scalar triplet $\Phi $ of the $SU(3)_{c}\times SU(3)_{L}\times U(1)_{X}$ $(3\text{-}3\text{-}1)$ model that has a $U(1)_{X}$ charge equal to $X=-1/3$ and acquires a vacuum expectation value larger than the electroweak symmetry breaking scale. The interactions of the scalar field $\phi$ to the photon- and gluon-pairs are mediated by the virtual vector-like fermions which appear as components of the anomaly-free chiral fermion representations of the $3\text{-}3\text{-}1$ gauge group.'
author:
- 'A. E. Cárcamo Hernández'
- 'Ivan Nišandži'' c'
title: 'LHC diphoton resonance at $750\,\text{GeV}$ as an indication of $SU(3)_L\times U(1)_X$ electroweak symmetry'
---
Introduction
============
The experimental collaborations ATLAS and CMS recently presented the results of the analysis of the early data obtained from the second LHC run of the proton-proton collisions at the center-of-mass energy $\sqrt{s}=13$ TeV [ATLAS2015,CMS:2015dxe]{}. Interestingly, both experiments observed the excess of the events with respect to the background in the diphoton final states at the invariant mass of around $750\,\text{GeV}$. The local statistical significance of the ATLAS (CMS) excess is about $3.9\,\sigma$ ($2.6\,\sigma$). The ATLAS found the signal in more than a single bin, preferring the large width of the resonance that corresponds to about $6\%$ of its mass $(\simeq 45\, \text{GeV})$. This feature has not yet been confirmed by the CMS collaboration. The available data from the second run did not reveal additional excess of the leptons or jets at this invariant mass. While it is possible that the reported excess is a random statistical fluctuation, if confirmed, it would provide the first direct evidence for the physics beyond the Standard Model (SM).
The results of many theoretical studies of the excess have been presented in the literature in the months following the announcement. General analyses of the excess, including surveys of several different specific model realisations can be found in [@Franceschini:2015kwy; @Buttazzo:2015txu; @Gupta:2015zzs; @Ellis:2015oso; @Agrawal:2015dbf; @Aloni:2015mxa; @Falkowski:2015swt; @Csaki:2015vek]. Variety of possibilities to accommodate the excess within the new physics models was presented in e.g. [Angelescu:2015uiz, DiChiara:2015vdm, Becirevic:2015fmu, Han:2015qqj, McDermott:2015sck, Chang:2015sdy, Cao:2015pto, Dutta:2015wqh, Alves:2015jgx, Han:2015dlp, Antipin:2015kgh, Fichet:2015vvy, Murphy:2015kag, Bauer:2015boy, Wang:2015kuj, Petersson:2015mkr, Bellazzini:2015nxw, Demidov:2015zqn, Bardhan:2015hcr, Ahmed:2015uqt, Pilaftsis:2015ycr, Harigaya:2015ezk, Molinaro:2015cwg, Arun:2015ubr, Kobakhidze:2015ldh, Bian:2015kjt, Curtin:2015jcv, No:2015bsn, Matsuzaki:2015che, Kim:2015ron, Knapen:2015dap, Nakai:2015ptz, Backovic:2015fnp, Mambrini:2015wyu, Martinez:2015kmn, Barducci:2015gtd, Bi:2015uqd, Heckman:2015kqk]{}.
Authors of the several articles [Franceschini:2015kwy,Buttazzo:2015txu,Angelescu:2015uiz,Ellis:2015oso,Falkowski:2015swt,Benbrik:2015fyz,Dhuria:2015ufo,Wang:2015kuj]{} noted the possibility that the electrically charged and colored vector-like fermions can be invoked for the mediation of the scalar boson interactions to the photon and gluon pairs. In this article we identify the excess with the scalar boson within the extended electroweak gauge group $SU(3)_L\times
U(1)_X$, that is component of the $SU(3)_L$ triplet with $U(1)_X$ charge $X=-1/3$. The anomaly free assignment of the fermion fields to the representations of the $3\text{-}3\text{-}1$ group [^1] leads to the appearance of the non-standard leptons and quarks that are vector-like under the SM gauge group. These fermions mediate the interactions of the scalar boson to the gluon- and photon pairs at the loop(s) level.
The Model
=========
The $3\text{-}3\text{-}1$ extension of the SM was first proposed in the late seventies [@Georgi:1978bv]. Several versions of the model have been subsequently studied, see e.g. [Valle:1983dk,Pisano:1991ee,Montero:1992jk,Frampton:1992wt,Ng:1992st]{}. Minimal versions do not include additional chiral fermion multiplets under the $SU(3)_L\times U(1)_X$ group, beyond those that contain three generations of the standard leptons and quarks. Many phenomenological aspects of the model have been investigated so far. As an example, the model can include the Peccei-Quinn symmetry, which leads to the possible solution of the strong-CP problem [@Pal:1994ba; @Dias:2002gg; @Dias:2003zt; @Dias:2003iq]. The studies of the models that contain sterile neutrinos in connection with weakly interacting massive fermionic dark matter candidates were reported in Refs. [@Mizukoshi:2010ky; @Dias:2010vt; @Alvares:2012qv; @Cogollo:2014jia], as well as the explorations of the fermion mass and mixing patterns [@CarcamoHernandez:2005ka; @Dias:2005yh; @Dias:2005jm; @Dias:2010vt; @Dong:2010zu; @Dong:2011vb; @Dong:2011pn; @Dias:2012xp; @Hernandez:2013hea; @Hernandez:2013mcf; @Boucenna:2014ela; @Boucenna:2014dia; @Hernandez:2014vta; @Hernandez:2014lpa; @Vien:2014gza; @Vien:2014pta; @Hernandez:2015tna; @Hernandez:2015cra; @Boucenna:2015zwa; @Vien:2016tmh; @Hernandez:2016eod].
We now briefly review the field content of the model and the interactions relevant for the present discussion. The electric charge generator can be expressed as the following linear combination $$Q=T_{3}+\beta T_{8}+XI, \label{eq1}$$where the $T_{i}$ are the generators of the $SU(3)_{L}$ group, which act on the triplet representation via the usual Gell-Mann matrices $\lambda _{i}$, i.e. $T_{i}=1/2\lambda _{i}$. The $X$ is the charge of the given representation under the $U(1)_{X}$ group factor, the $I$ stands for an identity matrix, while $\beta $ is an arbitrary real parameter.
Several versions of the $3\text{-}3\text{-}1$ models differ in the choice of the $\beta$ parameter. The most studied versions correspond to $\beta=\pm1/\sqrt{3}$ [@Georgi:1978bv] and $\beta=\pm\sqrt{3}$ [@Pisano:1991ee; @Frampton:1992wt]. The standard left (right) handed quarks and leptons are embedded into the chiral representations of the $SU(3)_L\times U(1)_X$, i.e. as triplets (singlets) of the $SU(3)_L$ group with the corresponding non-anomalous assignments of the $X$ charges. These representations contain non standard fermions, which residue in the vector-like representations of the SM gauge group. We denote the new quarks by the letter $J$ and new leptons by the symbol $\tilde{e}$. It then follows that the cancellation of the chiral anomalies requires that one of the quark generations residues in different representation of the gauge group than the remaining two. As a consequence, one obtains that the number of chiral fermion generations is a positive integer multiple of the number of colors, which provides the theoretical support to the observation of the existence of three generations of leptons and quarks. For concreteness, we assign the first two generations of left-handed quarks to the triplets of $SU(3)_L$ and the third generation to the antitriplet representation. The assignments of the $X$-charges are easily determined using the formula and requirement that the standard leptons and quarks have correct electric charges. It turns out that the $X$-charge of the first two generations of the left-handed triplets is given by $X_{Q_L^{1,2}}=1/6-\beta/(2\sqrt{3})$, while for the third generation antitriplet $X_{Q_L^{3}}=1/6+\beta/(2\sqrt{3}) $. The corresponding $X$-charges of the right handed quarks are equal to their electric charges, and are given as $X_{u_R^{1,2},d_R^{1,2},J_R^{1,2}}=2/3,-1/3,1/6-\beta \sqrt{3}/2$. The non-standard right handed quark of the third generation carries $X_{J_R^{3}}=1/6+\beta \sqrt{3}/2$. All three generations of the left-handed leptons are assigned to $SU(3)_L$ antitriplets with $X_{L_L}=-1/2-\beta/(2\sqrt{3})$, while the right-handed leptons are corresponding $SU(3)_L$ singlets and carry $X_{e_R,\tilde{e}_R}=-1,-1/2+\beta \sqrt{3}/2.$ Note that the exotic fermions residue in vector-like representations of the SM gauge group and are singlets under the $SU(2)_L$.
The scenarios with $\beta =\pm 1/\sqrt{3}$ introduce the non-standard fermions with the non-exotic electric charges, i.e equal to the electric charge of some standard model fermion. The options with $\beta =\pm \sqrt{3}$ involve large exotic electric charges of the new fermions, which makes these possibilities suitable for the enhancement of the branching fraction of the scalar resonance to the photon pairs. However, this scenario requires the departure from the perturbative description at the scale of several $\text{TeV}$‘s in order to remain in agreement with the measured value of the weak mixing angle at low energies, see e.g. [@Martinez:2006gb]. Other possibilities, like $\beta =0,\pm 2/\sqrt{3}$, involve new particles with the exotic (rational) electric charges. The electric charge conservation forbids the decay of the lightest such particle state. The phenomenological viability of such models would then require the detailed analysis of the abundance of the stable exotic charged particles in the Universe’s history.
We choose the value of the parameter $\beta =-1/\sqrt{3}$. The electric charges of the vector-like quarks are $Q(J^{1,2})=2/3$ and $Q(J^{3})=-1/3$, while the electric charges of vector-like leptons are $Q(\tilde{e}^{i})=-1$.
There are several possible choices of the scalar representations responsible for the spontaneous symmetry breaking of the $3\text{-}3\text{-}1$ group to the unbroken $SU(3)_{c}\times U(1)_{\text{Q}}$, see e.g. [@Diaz:2003dk] for the detailed review. The spontaneous symmetry breaking (SSB) proceeds in two steps. For the first step of breaking down to the SM gauge group we choose the scalar field $\Sigma ^{ij}$ that residues in the symmetric (sextet) representation of the $SU(3)_{L}$ and carries $X_{\Sigma }=-1/3$. The sextet develops the non-vanishing vacuum expectation value (vev) in the direction $\langle \Sigma ^{33}\rangle =w$, such that $w\gg v_{\text{ew}}$, where $v_{ew}\simeq 246\,\text{GeV}$ is the vev of the standard Higgs doublet. It turns out that this sextet does not contribute to the masses of the fermions, since $SU(3)_{L}$ invariant Yukawa term, $\bar{\psi}_{L}\psi
_{L}^{c}\Sigma $, also requires $2X_{\psi _{L}}=X_{\Sigma }$, which is not satisfied for any of the quark or lepton representations in the model. The spectrum of the massive gauge bosons can be obtained from the kinetic term $Tr\left[ (D_{\mu }\Sigma )^{\dagger }(D^{\mu }\Sigma )\right] $ using the expression for the covariant derivative for the sextet representation $$D_{\mu }\Sigma ^{ij}=\partial _{\mu }\Sigma ^{ij}-ig_{L}\bigg(\left( W_{\mu
}\right)^{ik}\Sigma ^{kj}+\left( W_{\mu }\right) ^{jl}\Sigma ^{li}\bigg)-ig_{X}X_{\mu }\delta ^{im}\Sigma ^{mj},$$where $W_{\mu }=W_\mu^a T^a$ denote the gauge boson field matrix, while $X_{\mu
} $ denotes the $X$ gauge boson field. The $SU(2)_{L}\times U(1)_{Y}$ symmetry is further broken to the $U(1)_{\text{Q}}$ by two triplet representations of the scalars, $\rho $ with $X_{\rho }=2/3$, and $\eta $ with $X_{\eta }=-1/3$. These triplets then generate the masses of the SM fermions and $W^{\pm }$ and $Z$ gauge bosons.
We introduce the triplet $\Phi$ with the $U(1)_X$ charge $X=-1/3$ and the vev pattern $\langle \Phi \rangle =(0,0,v_{\phi })$ to provide masses for the exotic fermions through the Yukawa interactions $${\small -\mathcal{L}_{\text{Y}}\supset \sum_{i=1}^{2}y_{Q}^{(i)}\overline{Q_{L}^{i}}\Phi J_{R}^{i}+y_{Q}^{(3)}\overline{Q_{L}^{3}}\Phi ^{\ast
}J_{R}^{3}+\sum_{i=1}^{3}y_{L}^{(i)}\overline{L_{L}^{i}}\Phi ^{\ast }\tilde{e}_{R}^{i}+h.c.} \label{Lyq}$$We identify the electrically neutral CP-even scalar component $\phi $ as a candidate for the resonance at the mass equal to $750\,\text{GeV}$. The coupling of the $\phi $ component of the triplet $\Phi$ to the vector-like fermions is then found from the above Yukawa terms after expanding around the vacuum, $\phi (x)\rightarrow \phi (x)+v_{\phi }$. The scalar potential which includes the interactions among the three $SU(3)_{L}$ scalar triplets contains a large number of unknown couplings and is given for completeness in the Appendix \[Scalar potential\]. After the SSB there remain three physical charged scalar bosons with masses around the $\text{TeV}$ scale and a doubly charged scalar boson that arises from the sextet and whose mass is expected to be of the order of the scale $w$. The contributions to the decay rate $\phi
\rightarrow \gamma \gamma$ from the loops involving charged scalars stem from the trilinear couplings denoted by $C_{\phi S_{i}^{+(+)}S_{i}^{-(-)}}$, where $S_{i}$ labels the physical charged scalar bosons. For example, the trilinear $C_{\phi \sigma _{1}^{++}\sigma
_{1}^{--}} $ coupling is given by $C_{\phi \sigma _{1}^{++}\sigma
_{1}^{--}}=\lambda _{14}v_{\phi }$.
The resonance at $750~\text{GeV}$
=================================
The $\phi$ boson interacts with the photon and gluon pairs via the loops of vector-like quarks to which it couples through the Yukawa terms in the Lagrangian (\[Lyq\]). The resonance is produced via gluon-gluon fusion, so that the cross section for the proton-proton scattering into the two-photon final state via the intermediate scalar boson $\phi $ is given in the narrow width approximation by the formula $$\sigma(pp\to\phi\to\gamma\gamma) =\frac{\pi ^{2}}{8}\frac{\Gamma (\phi
\rightarrow gg)\frac{1}{s}\int_{m_{\phi }^{2}/s}^{1}\frac{dx}{x}f_{g}(x)f_{g}\big(m_{\phi }^{2}/(sx)\big)\Gamma (\phi \rightarrow \gamma \gamma )}{m_{\phi }\Gamma _{\phi }}, \label{formula}$$where $m_{\phi }\simeq 750\,\text{GeV}$ is the mass of the resonance, $\Gamma _{\phi } $ its total decay width and $f_{g}(x)$ denotes the parton distribution function (pdf) of the gluon inside of the proton. We evaluate the partial decay widths in the above formula at the leading order in QCD and include the higher order QCD corrections by correcting the formula with the multiplicative factor $K^{gg}\sim 1.5$, as is costumary.
The corresponding decay widths of the resonance are given at leading order in QCD by $$\Gamma (\phi \rightarrow \gamma \gamma )=\frac{\alpha _{\text{em}}^{2}m_{\phi }^{3}}{512\pi ^{3}}\bigg\vert\sum_{i=1}^{3}\frac{N_{c}}{m_{J_{i}}}Q_{J_{i}}^{2}y_{Q}^{(i)}F(x_{J_{i}})+\sum_{i}\frac{1}{m_{\tilde{e}_{i}}}Q_{\tilde{e}_{i}}^{2}y_{L}^{(i)}F(x_{\tilde{e}_{i}})+\sum_{i}\frac{2\sqrt{2}C_{S_{i}}Q_{S_{i}}^{2}}{m_{\phi}^{2}}S(x_{S_{i}})\bigg\vert^{2}$$ and $$\quad \Gamma (\phi \rightarrow gg)=\frac{\alpha _{s}^{2}m_{\phi }^{3}}{256\pi ^{3}}\bigg\vert\sum_{i=1,2}\frac{1}{m_{J_{i}}}y^{(i)}F(x_{i})\bigg\vert^{2}, \label{Gammaphoton}$$where $x_{i}=4m_{i}^{2}/m_{\phi }^{2}$. The loop functions for fermion contributions $F(x)$ and the charged scalar contribution $S(x)$ are given by the expressions $$F(x)=2\,x\big(1+(1-x)f(x)\big),\,\quad S(x)=\left(-1+x f(x)\right) \quad
\text{where}\quad f(x)=\big(\arcsin \sqrt{1/x}\big)^{2},
\label{loopfunctions}$$valid for $x\geq1$. We use the value of the strong coupling constant $\alpha
_{s}(m_{\phi }/2)\simeq 0.1$ and the next-to-leading-order (NLO) set of pdfs from [@Martin:2009iq] (MSTW2008) at the factorization scale $\mu
=m_{\phi }$.
Given that we have $v_{\text{ew}}<v_{\phi }\ll w$ and since the couplings of the $126\,$GeV Higgs boson are consistent the SM expectations, we consider a benchmark scenario characterised by the absence of mixings between the $\phi $ resonance and the remaining neutral physical scalar fields. In addition, we assume that the $\phi $ is kinematically forbidden to decay into charged scalar bosons. Note also that the $\phi $ boson does not couple at the tree level to $W$ and $Z$ gauge bosons, which acquire their masses from the $\eta $ and $\rho $ triplets. We have explicitly checked that the contributions of the charged scalars to the diphoton rate is subleading, so that the only relevant contribution arises from the vectorlike fermions. For an illustration we assume that these fermions are degenerate and show in the Fig. \[Figure1\] the total cross section for the production of the $750\,\text{GeV}$ diphoton resonance at the LHC center of mass energy $\sqrt{s}=13\text{TeV}$, as a function of the charged exotic fermion masses $m_{F}$, and for several values of the exotic fermion Yukawa couplings, set to be equal to $2.5,2$ and $1.5$. Keeping all the Yukawa couplings equal and fixed to the value $1.5$, we note that the charged exotic fermion masses cannot be higher than about $800\,\text{GeV}$, in order to provide large enough signal cross section. For charged exotic Yukawa couplings equal to $1.5$ and charged exotic fermion masses of $700$ GeV, we find a total cross section of $4.7$ fb and total width for the $\phi
$ resonance of $45\,$MeV. In case that the large width of the resonance is confirmed, the present model would be immediately excluded as the explanation of the observed signal. This is the difficulty shared by all (the most) weakly coupled models that aim at explaining the excess. Since the vector-like fermions are singlets under the $SU(2)_{L}$ the decay rate $\Gamma (\phi \rightarrow WW)$ is also absent at one loop level. The rate $\Gamma (\phi \rightarrow Z\gamma )$ is suppressed with respect to the diphoton rate by the factor $2\tan \theta _{W}=0.60$. This factor is easily found by noticing that only the vector couplings of the $Z$ to the fermions contributes to the corresponding amplitude, and in the limit of the heavy scalar boson the amplitude is to a good approximation given by the amplitude of the decay to two photons, albeit with different couplings that involve the weak mixing angle. Furthermore, the rate $\Gamma (\phi \rightarrow ZZ)$ is even more suppressed than the rate $\Gamma (\phi \rightarrow Z\gamma )$, since it is suppresed with respect to the diphoton rate by the factor $\tan
^{4}\theta _{W}=0.08$.
Note that the vector-like fermions may have the couplings to the standard fermions, i.e. terms of the type $\tilde{y}^{ij}_Q\bar{Q}_L^{i}\Phi u^j_R$. This applies also to the standard down-type quarks and charged leptons. After the $\Phi$ develops the vev, these terms contribute to the quark (charged lepton) mass matrices. The mixing then causes the deviations from the unitarity of the standard Cabibbo-Kobayashi-Maskawa (CKM) matrix, and the observable effects in the $Z$-pole and electroweak precision observables. These effects are very tightly constrained from the available measurements [@Fajfer:2013wca; @Aguilar-Saavedra:2013qpa; @Allanach:2001sd]. In order to avoid these constraints we need to set the Yukawa couplings in the corresponding mixing terms to some small values. This can be achieved, at the formal level, by imposing discrete symmetry as shown in Refs. [Hernandez:2013hea,Hernandez:2015tna,Hernandez:2015cra,Vien:2016tmh,Hernandez:2016eod]{}. Although technically natural, setting these couplings to small values would constitute the new flavor hierarchy problem, especially if we keep in mind that the couplings that induce the $\phi\to\gamma\gamma$ need to be rather large. The absence of mixings between the SM and exotic quarks will imply that the exotic fermions will not exhibit flavor changing decays into SM quarks and gauge (or Higgs) bosons. After being pair produced they will decay into the standard fermions and the intermediate states of heavy gauge bosons, which in turn decay into the pairs of the standard fermions, see e.g. [@Cabarcas:2008ys]. The precise signature of the decays of the vector-like fermions depends on details of the spectrum and other parameters of the model. The present lower bounds from the LHC on the masses of the $Z^{\prime }$ gauge bosons in the $3\text{-}3\text{-}1$ models reach around $2.5\text{TeV}$ [@Salazar:2015gxa]. One can translate these bounds on the order of magnitude of the scale $w$. The suppression of the decay rates involving SM gauge bosons and the large masses of the nonstandard gauge bosons then imply long-lived vector-like fermions. We plan to study the details of the corresponding collider signatures in the future.
Summary
=======
To summarise, we point out that the diphoton signal recently observed by the ATLAS and CMS collaborations at the invariant mass $\sim 750\,\text{GeV}$ could arise from the $\phi$, electrically neutral CP-even component of one of the scalar triplets representation of the $3\text{-}3\text{-}1$ model. Its couplings to photons and gluons are mediated by the loops that involve exotic vector-like fermions. Such fermions appear as components of the anomaly-free fermion representations. In order to reproduce the observed signal, the vector like fermions need to be light (around $1\text{TeV}$) and couple to the $\phi$ boson rather strongly. On the other hand the mixings of the vector-like fermions to the standard chiral fermions needs to be highly suppressed in order to remain in accordance with the precision experiments.
**Acknowledgments**. A.E.C.H was supported by DGIP internal Grant No. 111458. I. N. is supported in part by the Bundesministerium für Bildung und Forschung (BMBF).
Scalar potential {#Scalar potential}
================
The scalar potential which includes the interactions among the three $SU\left( 3\right) _{L}$ scalar triplets and with the scalar sextet is given by: $$\begin{aligned}
V_{H} &=&\mu _{\chi }^{2}({\small \Phi }^{\dagger }{\small \Phi })+\mu
_{\eta }^{2}(\eta ^{\dagger }\eta )+\mu _{\rho }^{2}(\rho ^{\dagger }\rho
)+f_{1}\left( \eta _{i}{\small \Phi }_{j}\rho _{k}\varepsilon
^{ijk}+H.c\right) +\lambda _{1}({\small \Phi }^{\dagger }{\small \Phi })({\small \Phi }^{\dagger }{\small \Phi }) \notag \\
&&+\lambda _{2}(\rho ^{\dagger }\rho )(\rho ^{\dagger }\rho )+\lambda
_{3}(\eta ^{\dagger }\eta )(\eta ^{\dagger }\eta )+\lambda _{4}({\small \Phi
}^{\dagger }{\small \Phi })(\rho ^{\dagger }\rho )+\lambda _{5}({\small \Phi
}^{\dagger }{\small \Phi })(\eta ^{\dagger }\eta ) \notag \\
&&+\lambda _{6}(\rho ^{\dagger }\rho )(\eta ^{\dagger }\eta )+\lambda _{7}({\small \Phi }^{\dagger }\eta )(\eta ^{\dagger }{\small \Phi })+\lambda _{8}({\small \Phi }^{\dagger }\rho )(\rho ^{\dagger }{\small \Phi })+\lambda
_{9}(\rho ^{\dagger }\eta )(\eta ^{\dagger }\rho ) \notag \\
&&+\mu _{\Sigma }^{2}\left( \Sigma _{ij}\Sigma ^{ij}\right) +f_{2}\left(
\eta _{i}\rho _{j}\Sigma ^{ij}+H.c\right) +f_{3}\left( {\small \Phi }_{i}\rho _{j}\Sigma ^{ij}+H.c\right) \notag \\
&&+\lambda _{10}\left( \Sigma _{ij}\Sigma ^{ij}\right) \left( \Sigma
_{kl}\Sigma ^{kl}\right) +\lambda _{11}\left( \Sigma _{ij}\Sigma
^{il}\right) \left( \Sigma _{kl}\Sigma ^{jk}\right) +\lambda _{12}(\eta
^{\dagger }\eta )\left( \Sigma _{kl}\Sigma ^{kl}\right) \notag \\
&&+\lambda _{13}(\rho ^{\dagger }\rho )\left( \Sigma _{kl}\Sigma
^{kl}\right) +\lambda _{14}({\small \Phi }^{\dagger }{\small \Phi })\left(
\Sigma _{kl}\Sigma ^{kl}\right) +\lambda _{15}\left[ ({\small \Phi }^{\dagger }\eta )\left( \Sigma _{kl}\Sigma ^{kl}\right) +H.c\right] \notag
\\
&&+\lambda _{16}\eta ^{i}\Sigma _{ij}\Sigma ^{jk}\eta _{k}+\lambda _{17}\rho
^{i}\Sigma _{ij}\Sigma ^{jk}\rho _{k}+\lambda _{18}{\small \Phi }^{i}\Sigma
_{ij}\Sigma ^{jk}{\small \Phi }_{k}+\lambda _{19}\eta ^{i}\Sigma _{ij}\Sigma
^{jk}{\small \Phi }_{k} \label{scalarpotential}\end{aligned}$$
where the three scalar triplets and the sextet are given in terms of the components: $$\begin{aligned}
{\small \Phi }& =\begin{pmatrix}
\phi _{1}^{0} \\
\phi _{2}^{-} \\
\frac{1}{\sqrt{2}}(\upsilon _{\phi }+\phi \pm i\zeta _{\phi })\end{pmatrix},\hspace{1cm}\rho =\begin{pmatrix}
\rho _{1}^{+} \\
\frac{1}{\sqrt{2}}(\upsilon _{\rho }+\xi _{\rho }\pm i\zeta _{\rho }) \\
\rho _{3}^{+}\end{pmatrix}, \notag \\
\eta & =\begin{pmatrix}
\frac{1}{\sqrt{2}}(\upsilon _{\eta }+\xi _{\eta }\pm i\zeta _{\eta }) \\
\eta _{2}^{-} \\
\eta _{3}^{0}\end{pmatrix},\hspace{1cm}\Sigma =\left(
\begin{array}{ccc}
\sigma _{1}^{0} & \sigma _{1}^{-} & \sigma _{2}^{0} \\
\sigma _{1}^{-} & \sigma _{1}^{--} & \sigma _{2}^{-} \\
\sigma _{2}^{0} & \sigma _{2}^{-} & w+\sigma _{3}^{0}\end{array}\right) . \label{331-scalar}\end{aligned}$$
[9]{} The ATLAS collaboration, ATLAS-CONF-2015-081.
[^1]: In the following text we refer to the models that are based on this gauge group as $3\text{-}3\text{-}1$ models, as the $SU(3)_c$ group factor of the QCD remains intact.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The longitudinal proton-proton femtoscopy (HBT) correlation function, based on the idea that in a heavy ion collision at $\sqrt{s}\lesssim20\gev$ stopped protons are likely to be separated in configuration space, is evaluated. It shows a characteristic oscillation which appears sufficiently pronounced to be accessible in experiment. The proposed measurement is essential for estimating the baryon density in the central rapidity region, and can be also viewed as an (almost) direct verification of the Lorentz contraction of the fast-moving nucleus.'
author:
- Andrzej Bialas
- Adam Bzdak
- Volker Koch
title: Femtoscopy of stopped protons
---
\#1[\#1 ]{} \#1[| \#1 |]{} \#1
Introduction
============
The search for a possible phase structure of QCD has been a focus point in strong interaction research. Lattice QCD calculations have established that for vanishing and small net-baryon density the transition from hadrons to quarks and gluons is an analytic cross-over [@Aoki:2006we]. The situation at larger baryon density, on the other hand, is less clear, since at present lattice methods cannot access this region because of the fermion sign problem. Here one has to rely on model calculations and a large class of these models do indeed predict a first-order phase coexistence region which ends in a critical point (see, e.g., Ref. [@Stephanov:2007fk] for an overview).
In order to explore the region of large net-baryon density experimentally one studies heavy ion collisions at moderate beam energies $\sqrt{s}\lesssim20\gev$, where a sufficient amount of the incoming nucleons are stopped at mid-rapidity in order to achieve the necessary baryon density. Indeed, since produced baryons always come as baryon – anti-baryon pairs, the only means of producing a finite net baryon density is by stopping the nucleons of the colliding nuclei. Thus, in order to explore the QCD phase diagram at large baryon density, the question of baryon stopping is essential to understand. In fact, stopping the baryons is only a necessary condition. In addition to being at mid-rapidity in momentum space they also need to overlap in configuration space.
The mechanism by which the incoming nucleons are stopped is indeed a very interesting question [@Busza:1983rj; @Busza:1989px; @Kharzeev:1996sq; @Capella:1996th; @Anishetty:1980zp; @Li:2016wzh]. However, independent of the specific mechanism, it seems rather unphysical that the nucleons are stopped instantaneously. Instead, it will take time and space for the nucleons to decelerate. Therefore, it is rather unlikely that the stopped nucleons will end up at $z\simeq0$, i.e., at the point of the collision of the two nuclei. Instead, one would expect that the nucleons from the right-going nucleus will end up at positions in configuration space with $z>0$ and the left-going ones at $z<0$ so that the stopped nucleons may actually be distributed bi-modally in configuration space. This observation was recently pointed out in Ref. [@Bialas:2016epd]. Based on a simple string model, Ref. [@Bialas:2016epd] found that for collision energies $\sqrt{s}\gtrsim10\gev$ the stopped nucleons actually will not overlap significantly in configuration space. Of course this observation was based on a rather simple model and it would be much better if this observation could be verified or ruled out in experiment. This is the purpose of this paper, where we propose to measure [*longitudinal*]{} Hanbury Brown Twiss (HBT) type correlations (also known as femtoscopy [@Lisa:2005dd]) of the stopped protons, i.e. protons at $y_{cm}\approx 0$ with transverse momentum not exceeding, say, 1 GeV. Since femtoscopy does not a priori distinguish between stopped and produced protons, it is important to choose a collision energy which is small enough for proton production to be negligible but sufficiently high so that the deceleration length is large enough for the stopped protons to be separated in configuration space. Thus an energy of $\sqrt{s}\simeq20\gev$ appears to be a good choice since at this energy the anti-proton to proton ratio is still very small, $\bar{p}/p\simeq0.1$ [@Adamczyk:2017iwn; @Anticic:2010mp].
This paper is organized as follows: In the next section we present and discuss the source function based on the same simple string model used in Ref. [@Bialas:2016epd] (corrected, however, for the Fermi motion inside a nucleus). Next we calculate the resulting femtoscopy correlation function before we close with a discussion of the various issues and limitations of this study.
The source function
===================
The essential ingredient for femtoscopy is the underlying source of the the emitted particles, protons in our case. The source is the phase-space distribution of emission points, which are typically the points of the last interaction of the protons before they fly to the detector. Clearly a quantitative calculation of such a source function would require a sophisticated simulation. However, we believe that certain semi-quantitative aspects can be discussed without such a treatment, and it is this approach we will take in the following.
As already eluded to in the Introduction, once two nucleons collide it is very unlikely or possibly even unphysical for them to come to a stop right at the collision point. Instead they will only come to a stop after a certain distance and time.
The distance $\Delta z$ and time $\Delta t$ between the collision and the final space-time point $(z,t)$ where and when a nucleon acquires its final rapidity $y$, depends on the mechanism of deceleration and thus on the model used for its description. It is in general a function of the initial and final rapidity, $Y_{i}$ and $y$ as well as the typical transverse mass, $M_{\perp}$ the nucleon acquires after the collision.
For a given collision space-time point $\left(z_{c},t_{c}\right)$ in the center-of-mass frame of the two nucleons we thus have z & =z\_[c]{}z(Y\_[i]{},y,M\_),t & =t\_[c]{}+t(Y\_[i]{},y,M\_), \[ztf\] where the plus sign refers to the right-going particles and the minus sign to the left-going ones.
One sees from (\[ztf\]) that, in order to construct the source needed for femtoscopy, we need a distribution of the collision points in space and time and a model or theory which determines $\Delta z$ and $\Delta t$.
Distribution of collision points
--------------------------------
Let us start with the collision point distribution. Here we follow Ref. [@Bialas:2016epd] and assume that the distribution of nucleons inside the target and projectile nuclei can be reasonably described by a Gaussian. In this case the longitudinal ($z$-direction) and transverse components of the collision point distribution factorise and subsequently we will concentrate on the collision point distribution in the z-direction, which, following Ref. [@Bialas:2016epd], we assume to be proportional to the overlap of the distribution of the nucleons in the left- and right- moving nuclei.
We thus have W\_c(z\_c,t\_c)\~e\^[-\^2\[z\_c-\_L(t\_c)\]\^2/R\_L\^2]{}e\^[-\^2\[z\_c-\_R(t\_c)\]\^2/R\_R\^2]{} (t\_c), where \_L(t)=-\_R(t)=\_0-Vt; \_0R\_[L,R]{}/, are the positions of the centers of the nuclei at the time $t$. $\zeta_0$ and $-\zeta_0$ are positions of the centres of left-moving and right-moving nuclei at $t=0$ before the nuclei have any contact with each other. This implies $\zeta_0\gg R_{L,R}/\gamma$ and $t_c\geq 0$. Also, $\gamma=\cosh(Y_{cm})$ denotes the Lorentz contraction factor for the incoming nuclei in the center-of-mass frame, which we are working in.
Distribution of nucleon emission points $z$ and $t$
---------------------------------------------------
Consider first the right-movers. For the distribution of $z$ and $t$, we have W\_R(z,t)&=&dz\_c dt\_c W\_c(z\_c,t\_c) ( z-z\_c-z) (t-t\_c-t) & \~& e\^[-\^2\[z-z-Z\]\^2/R\_L\^2]{}e\^[-\^2\[z-z+Z\]\^2/R\_R\^2]{}( t -t) \[right\] with $Z\equiv \zeta_0-V(t-\Delta t )$.
For left-movers, the formula differs by the sign of $\Delta z$: W\_L(z,t) \~e\^[-\^2\[z+z-Z\]\^2/R\_L\^2]{}e\^[-\^2\[z+z+Z\]\^2/R\_R\^2]{}( t -t).
For identical nuclei we have W(z,t;P\_i,P\_f) &=& W\_L(z,t)+W\_R(z,t) [\
]{}& \~& ( e\^[-\[z+ z\]\^2/\_c\^2]{}+e\^[-\[z- z\]\^2/\_c\^2]{})e\^[-Z\^2/\_c\^2]{}( t-t), where $\zeta_0\gg R/\gamma$, $\Gamma_c^2=R^2/2\gamma^2$, and the dependence on the initial and final momenta is implicit via $\Delta z$ and $\Delta t$.
What remains then is to determine $\Delta z$ and $\Delta t$. For nucleons with small transverse velocities, the simplest model is that of linear energy loss, as for instance used in the Lund model [@Lund_model] or the Bremsstrahlung model [@Stodolsky:1971qf]. Using the conditions[^1] dE/dz=,dP/dt =, \[fund\] where $\sigma$ denotes the energy loss per unit length or string tension one obtains [@Bialas:2016epd] $$\begin{aligned}
\Delta z\left(P_{i},\,P_{f},\,\sigma\right) & =\frac{E_{i}-E_{f}}{\sigma}\label{eq:delta_z_general}; \qquad E_f=M_\perp \cosh y ,\\
\Delta t\left(P_{i},\,P_{f},\,\sigma\right) & =\frac{P_{i}-P_{f}}{\sigma}\label{eq:delta_t_general}; \qquad P_f=M_\perp \sinh y .\end{aligned}$$ Here $P_{i}=M\sinh(Y_{i})$, $E_{i}= \sqrt{M^{2}+P_{i}^{2}}$, and $P_{f}=M_{\perp}\sinh(y)$, $E_{f}= \sqrt{M_{\perp}^{2}+P_{f}^{2}}$ are the initial and final longitudinal momenta and energies
These equations determine $\Delta z$ and $\Delta t$ for initial and final longitudinal momenta, $P_{i}$ and $P_{f}$, transverse mass of the final proton, $M_\perp$, and for a given rate of energy loss $\sigma$ (string tension). In reality the string tension is not a constant but may fluctuate from collision to collision (e.g., depending on number of constituent quarks wounded in a given collision). Since it is unlikely, however, that a nucleon with only one or two wounded quarks may fully stop, the sample of the nucleons with the final rapidity $y\approx 0$ is expected to be largely dominated by those with three wounded quarks. Thus we shall ignore fluctuations due to the string tension and take $\sigma = 3
\sigma_0$= 3 GeV/fm. For the transverse mass we subsequently will chose a value of $M_{\perp}=1.2 \gev$ (we verified that the results are not sensitive to the actual value of $M_\perp$).
Fermi motion
------------
In the case of a nucleus-nucleus collision, the nucleons inside the target and projectile nuclei experience Fermi motion. Consequently, the initial momentum of the colliding nucleons is distributed around the nominal (mean) value of the nucleus-nucleus collision. This broadens the emission source in the longitudinal spatial direction, and thus affects the femtoscopy signal. We have W\_F(z,t,P\_i,P\_f)= dP\_i G\_F(P\_i- P\_i ) W(z,t; P\_i,P\_[f]{} ), \[wf\] where $G_F(P_i - \langle P_i \rangle)$ is the distribution of the actual initial momentum $P_i$ of the nucleon around the average $\langle P_i \rangle$. We shall take it in the form G\_F(P\_i-P\_i ) \~e\^[-\[P\_i-P\_i \]\^2/[\_F ]{}\^2]{}; \_F= k\_[F]{} 165 , where $k_F$ is the Fermi momentum.[^2] Note that due to the Lorentz boost the width of the distribution ${\Gamma_F }$ scales with the Lorentz factor $\gamma$, ${\Gamma_F }\sim \gamma$. This increases substantially the width of this distribution in the energy region of interest.
The HBT correlation function
============================
The femtoscopic longitudinal correlation function we are seeking is given by [@Pratt:1984su; @Lisa:2005dd] C(q\_z;q\_0) - 1 =- 12 , \[c\] where $\delta q_z$ is the difference of the longitudinal momenta of the two protons, $\delta q_0$ is the difference of their energies and (q\_z,q\_0;P\_i,P\_f )=\_[-]{}\^dz e\^[i zq\_z]{} \_[t]{}\^dt e\^[ -i tq\_0]{} W(z,t;P\_i, P\_f) is the Fourier transform of the density.
Since we are working with Gaussians, the Fourier transforms are straightforward. We have (q\_z,q\_0;P\_i,P\_f)\~e\^[-(q\_z [\_c]{})\^2/4]{} e\^[- iq\_0(t +\_0/V)]{}e\^[-(q\_0 [\_c]{})\^2/(4V\^2)]{} , \[phi\] where for the Fourier transform in $t$ it was essential to use the condition $\zeta_0\gg R/\gamma$ which allowed us to integrate over time from $-\infty$.
Thus, the final result for the correlation function, including Fermi motion, is C\_F(q\_z;q\_0) - 1 =- 12 , \[cf\] where \_[F]{}(q\_[z]{},q\_[0]{};P\_[i]{},P\_[f]{})= dP\_[i]{}G\_[F]{}(P\_[i]{}-P\_i )(q\_[z]{};q\_[0]{};P\_[i]{},P\_[f]{}) \~e\^[-( q\_[z]{}\^[2]{}+q\_[0]{}\^[2]{}) ]{} , \[PhiF\] where we have omitted a common phase which does not play any role in the correlation function. Here we have assumed that the momentum dependence of the shift $\Delta
z$, Eq. , may be approximated by $$\begin{aligned}
\Delta z = \frac{E_{i}-E_{f}}{\sigma} = \frac{E_{i}-\langle E_{i} \rangle}{\sigma} +
\frac{\langle E_{i} \rangle-E_{f}}{\sigma} \simeq \frac{P_{i}-\langle P_{i} \rangle}{\sigma} +
\Delta Z;\;\;\;\;\Delta Z \equiv (\langle E_{i} \rangle - E_{f})/ \sigma ,
\label{eq:deltaZ_approx}\end{aligned}$$ where $\langle E_{i} \rangle$ is the mean (nominal) energy of the incident nuclei, and $\Delta Z$ denotes the shift in space for the mean energy. We further assumed that the velocity of the nuclei in the c.m. frame is close to unity, $V\simeq 1$. For the energies under consideration, $\sqrt{s}> 10 \gev $, these should be very good approximations and indeed we find that the resulting correlation function agrees with the full result within a few per mille.
Results
=======
The correlation function (\[cf\]) evaluated for $\delta q_0=0$ (corresponding to protons with equal and opposite rapidities) is shown in Fig. \[fig:corr\_result\]. Here we used for the radius $R=7 \, \mathrm{fm}$ and $M_{\perp,1}=M_{\perp,2} = 1.2$ GeV. In panel (a) we show the predicted femtoscopy correlation function for a collision energy of $\sqrt{s}=20\gev$ and in panel (b) for $\sqrt{s}=14\gev$. The black dashed lines represent the result for the model discussed above. The blue solid lines are the results, where we doubled the width $\Gamma_c$ of the collision point distribution in order to allow for additional smearing (induced, e.g., by a non-zero proton radius not taken into account in our model). One sees characteristic oscillations of the correlation function, reflecting the two maximum structure of the source density.
To obtain experimental predictions, the correlation function seen in Fig. 1 must be corrected for the final state interactions [@Lednicky:1981su; @Adam:2015vja]. They are shown in Fig. 2 with strong and Coulomb interaction effects taken into account. One sees that although these corrections strongly affect the very small region of $\delta q_z$, the region of $\delta q_z$ where the oscillations are observed remains qualitatively unchanged.
![Same as Fig. \[fig:corr\_result\] but with strong and Coulomb interaction effects included.[]{data-label="fig:corr_result_strong"}](h_20.pdf "fig:"){width="45.00000%"} ![Same as Fig. \[fig:corr\_result\] but with strong and Coulomb interaction effects included.[]{data-label="fig:corr_result_strong"}](h_14.pdf "fig:"){width="45.00000%"}
![Same as Fig. \[fig:corr\_result\] but with strong and Coulomb interaction effects included.[]{data-label="fig:corr_result_strong"}](h_20_strong.pdf "fig:"){width="45.00000%"} ![Same as Fig. \[fig:corr\_result\] but with strong and Coulomb interaction effects included.[]{data-label="fig:corr_result_strong"}](h_14_strong.pdf "fig:"){width="45.00000%"}
In Fig. \[fig:source\_result\] we show the corresponding time-integrated source distribution, $$\begin{aligned}
\bar{W}(z;P_{i},P_{f}) &= \int dt W_{F}(z,t; P_i,P_{f}) {\nonumber\\}&\sim \exp\left( {-\frac{(z-\Delta Z)^{2}}{{\Gamma_F }^{2}/\sigma^{2} + {\Gamma_c}^{2}}} \right)
+ \exp\left(-{\frac{(z+\Delta Z)^{2}}{{\Gamma_F }^{2}/\sigma^{2} + {\Gamma_c}^{2}}}\right) .
\label{eq:source_integrated}\end{aligned}$$ Here we used the same approximation as before, Eq. .
We see that the separation of the stopped protons exhibited in the source distribution manifests itself as extra oscillation in the femtoscopy correlation function. For a collision energy of $\sqrt{s}=20\gev$ the signal is clearly visible for both the model result as well the more conservative result, where we doubled the width ${\Gamma_c}$. At $\sqrt{s}=14\gev$ the signal is much weaker, however.
![Time integrated source function for stopped protons as a function of $z$ for (a) $\sqrt{s}=20\protect\gev$ and (b) $\sqrt{s}=14\protect\gev$. The black dashed lines represent the result of our model calculation while the blue solid lines are obtained by doubling the value of width of the collision point distribution, ${\Gamma_c}$. The source functions shown are normalized to unity. \[fig:source\_result\]](s_20.pdf "fig:"){width="45.00000%"} ![Time integrated source function for stopped protons as a function of $z$ for (a) $\sqrt{s}=20\protect\gev$ and (b) $\sqrt{s}=14\protect\gev$. The black dashed lines represent the result of our model calculation while the blue solid lines are obtained by doubling the value of width of the collision point distribution, ${\Gamma_c}$. The source functions shown are normalized to unity. \[fig:source\_result\]](s_14.pdf "fig:"){width="45.00000%"}
Conclusion and Remarks
======================
In conclusion, we have presented a calculation of the longitudinal femtoscopy correlation function of stopped protons based on the observation that in a heavy ion collision at $10 \gev \lesssim \sqrt{s}\lesssim20\gev$ such protons are likely to be separated in configuration space. The resulting correlation function shows extra oscillations which appear sufficiently pronounced to be accessible in experiment. Clearly such a measurement, if feasible, would be most desirable. It will provide useful information about the longitudinal configuration space distribution of the nucleons in a heavy ion collision, and, more importantly, it will provide essential constraints on the mechanism by which baryon number is transported to mid-rapidity.
Some remarks are in order.
(i) The observation of the suggested extra oscillations will not only confirm the idea that the nucleons do not stop immediately after collision. It should also allow to measure the effective distance at which the energy is deposited in the produced particles. Indeed, as seen from Eq. (\[PhiF\]), $\Phi_F$ (and thus also $C_F$) explicitly depends on $\Delta Z$, the average distance required to stop a proton.
(ii) Even if the oscillations are not seen, the measurement will determine the (longitudinal) size of the volume from which the protons at $y_{cm}\approx 0$ are emitted. This should allow to estimate the actual density of protons in configuration space, the quantity essential for the studies of this system. One also obtains the upper limit on the distance the nucleons travel before attaining the rapidity $y\approx 0$, thus improving our understanding of the process of the energy loss by the leading particles in a high energy collision.
(iii) The definition of the longitudinal correlation function requires that the vector $\delta \vec{q}$ points in the $z$-direction, i.e. $\delta q_\perp = 0$. In our approximation of the nuclear densities as Gaussians this restriction is not important, as the longitudinal and transverse degrees of freedom factorise. To increase statistics, one may thus integrate over transverse momenta. Since the Lund model is best justified at small transverse velocities, and since the Gaussian form is only an approximation, it seems reasonable, however, to restrict measurements to protons with transverse momenta not exceeding, say, 1 GeV.
(iv) It turns out that the corrections due to the Coulomb and strong interactions do not change qualitatively the possibility of observation of the expected oscillations of the correlation function.
(v) Our calculation ignored entirely possible correlations between the outgoing protons due to quark mixing at very short distances [@bz]. Introducing such correlations may result in the correlation function being positive in some region of $\delta q_z$. As shown in Ref. [@bz], however, this effect is small and should not modify our conclusions.
(vi) Finally, let us add that our results rely strongly on the idea that the longitudinal distribution of nucleons inside moving nucleus are Lorentz contracted and that this contraction survives during the collision. The proposed measurement should thus provide an interesting test of this effect (for the recent discussion of the measurements of Lorentz contraction, see Ref. [@Rafelski:2017yob]).
We thank Scott Pratt for providing his code to calculate the effects of strong and Coulomb interaction effects and useful correspondence. Thanks are due to Mike Lisa for interesting discussions. This work was partially supported by the Faculty of Physics and Applied Computer Science AGH UST statutory tasks No. 11.11.220.01/1 within subsidy of Ministry of Science and Higher Education, and by the National Science Centre, Grants No. 2014/15/B/ST2/00175 and No. 2013/09/B/ST2/00497, and the U.S. Department of Energy, Office of Science, Office of Nuclear Physics, under contract number DE-AC02-05CH11231.
[99]{}
Y. Aoki, G. Endrodi, Z. Fodor, S. D. Katz and K. K. Szabo, Nature [**443**]{}, 675 (2006). M. A. Stephanov, PoS LAT [**2006**]{}, 024 (2006). R. Anishetty, P. Koehler and L. D. McLerran, Phys. Rev. D [**22**]{}, 2793 (1980). W. Busza and A. S. Goldhaber, Phys. Lett. [**139B**]{}, 235 (1984). W. Busza and R. Ledoux, Ann. Rev. Nucl. Part. Sci. [**38**]{}, 119 (1988). D. Kharzeev, Phys. Lett. B [**378**]{}, 238 (1996). A. Capella and B. Z. Kopeliovich, Phys. Lett. B [**381**]{}, 325 (1996). M. Li and J. I. Kapusta, Phys. Rev. C [**95**]{}, no. 1, 011901 (2017). A. Bialas, A. Bzdak and V. Koch, Acta Phys. Polon. B [**49**]{}, 103 (2018). M. A. Lisa, S. Pratt, R. Soltz and U. Wiedemann, Ann. Rev. Nucl. Part. Sci. [**55**]{}, 357 (2005). L. Adamczyk [*et al.*]{} \[STAR Collaboration\], Phys. Rev. C [**96**]{}, no. 4, 044904 (2017). T. Anticic [*et al.*]{} \[NA49 Collaboration\], Phys. Rev. C [**83**]{}, 014901 (2011). B. Andersson, [*The Lund Model*]{} (Cambridge University Press, 1998).
L. Stodolsky, Phys. Rev. Lett. [**28**]{}, 60 (1972). S. Pratt, Phys. Rev. Lett. [**53**]{}, 1219 (1984). R. Lednicky and V. L. Lyuboshits, Sov. J. Nucl. Phys. [**35**]{}, 770 (1982); \[Yad. Fiz. [**35**]{}, 1316 (1981)\]. J. Adam [*et al.*]{} \[ALICE Collaboration\], Phys. Rev. C [**92**]{}, no. 5, 054908 (2015). A. Bialas and K. Zalewski, Phys. Lett. B [**727**]{}, 182 (2013); A. Bialas, W. Florkowski and K. Zalewski, Phys. Lett. B [**748**]{}, 9 (2015).
J. Rafelski, Eur. Phys. J. A [**54**]{}, no. 2, 29 (2018).
[^1]: The second condition expresses the equation of motion with the force equal to $\sigma$. It is exact when the transverse velocity of the nucleon vanishes. We have verified that for the nucleons of transverse momenta not exceeding 1 GeV the corrections are negligible.
[^2]: The value of $\Gamma_F$ follows from the demand that the distribution $G_F$ exhibits the same variance as the Fermi gas with the Fermi momentum $k_F$. Also, we ignore the small effect of the binding energy of the nucleons and instead assume that we can treat the nucleons as free particles.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a solid state refrigeration technique based on repeated adiabatic magnetization/demagnetization cycles of a superconductor which acts as the working substance. The gradual cooling down of a substrate (normal metal) in contact with the working substance is demonstrated for different initial temperatures of the substrate. Excess heat is given to a hot large-gap superconductor. The on-chip refrigerator works in a cyclic manner because of an effective thermal switching mechanism: Heat transport between N/N versus N/S junctions is asymmetric because of the appearance of the energy gap. This switch permits selective cooling of the metal. We find that this refrigeration technique can cool down a 0.3cm$^{3}$ block of Cu by almost two orders of magnitude starting from 200mK, and down to about 1mK starting from the base temperature of a dilution fridge (10mK). The corresponding cooling power at 200mK and 10mK for a 1cm$\times$1cm interface are 25 nW and 0.06 nW respectively, which scales with the area of the interface.'
author:
- 'Sreenath K. Manikandan'
- Francesco Giazotto
- 'Andrew N. Jordan'
bibliography:
- 'ref.bib'
title: 'Superconducting quantum refrigerator: Breaking and rejoining Cooper pairs with magnetic field cycles'
---
The goal of building solid state refrigerators and heat engines working on quantum principles is an outstanding need for next generation quantum technologies [@sothmann2014thermoelectric]. It is known since the earlier days of superconductivity that the process of magnetizing a superconducting material quasistatically and adiabatically can reduce the temperature of the material substantially as it transitions to the normal state [@keesom1934further; @mendelssohn1934magneto; @yaqub1960cooling]. This is because a material in its superconducting state has more order, and therefore, entropy equal to that of a normal metal at a lower temperature. Hence when driven to the normal state adiabatically by an applied magnetic field, the achieved final state is much colder than the initial superconducting state as depicted in the $T-\mathcal{S}$ diagram in Fig. 1(b). There were attempts in the past to try and implement adiabatic magnetization of a superconductor as an effective cooling technique, notably the early proposals by Mendelssohn and Moore [@mendelssohn1934magneto], and by Keesom and Kok [@keesom1934further]. Recently Dolcini and Giazotto had studied the adiabatic magnetization of a superconductor by including dynamical dissipative effects such as eddy current losses, and suggested that this mechanism can still be used to achieve significant cooling for micro-refrigeration purposes [@dolcini2009adiabatic].
![(a) Steps of the refrigeration cycle. In step A, the central region (working substance) is thermally isolated from its neighbors, and undergoes adiabatic magnetization from a superconductor to a normal metal, $S_2 \rightarrow N_2$, and cools to a much colder temperature, $T_1$. In step B, thermal contact with the normal metal $N_1$ is made, resulting in heat transfer from $N_1$ to $N_2$, eventually coming to equilibrium at temperature $T_C$. In step C, the working substance is thermally isolated again (black walls), and adiabatically demagnetized from $N_2 \rightarrow S_2$, heating up the system to its hottest temperature $T_4$. In step D, thermal contact with superconductor $S_3$ is made, allowing heat to escape from $S_2 \rightarrow S_3$, reducing the temperature to temperature $T_H$. The cycle closes by closing off thermal contact to $S_3$ with the black wall, and returning to step A. (b) Entropy of the normal metal and a superconductor, showing different stages of the refrigeration cycle.\[fig1\]](Fig1.pdf){width="\linewidth"}
{width="\linewidth"}
Here we propose a cyclic superconducting refrigerator based on adiabatic magnetization of a superconductor, with a working mechanism similar to that of a domestic refrigerator. A conventional refrigerator operates by cyclically moving a working fluid between hot and cold reservoirs. Work is done by compressing a fluid, and letting it freely expand to a gas in a phase transition where it cools down and absorbs heat from the cold reservoir. The now hot gas is then re-compressed, liquifying it, and then dumps the excess heat to a hot reservoir, which is usually the environment that allows the fluid to thermalize and reset to its initial temperature. The cycle repeats many times such that a stable low final temperature is achieved in the cold reservoir. In comparison, the working fluid in our example is the electron gas in the working superconductor. The cold reservoir is a normal metal, and the hot reservoir is another superconductor having a larger gap. The superconducting state of electrons in the working substance is analogous to a compressed fluid. With an applied magnetic field, the electron fluid expands in a phase transition into the unpaired normal state at a lower temperature. Heat is then absorbed from the cold reservoir, and the electron fluid is re-compressed by reducing the applied magnetic field. The working substance, which is now hotter than the hot reservoir, has reduced electronic entropy in the paired state. The entropy of phonons has increased in the working substance in proportion, effectively holding the excess heat. Note that here the phonon entropy changes complementary to the electron entropy (in both steps, A and C), such that the sum of the entropy of electrons and phonons remains constant, and the process is adiabatic. Electron-phonon interactions in the working substance and a tunneling contact with the hot reservoir selectively removes hot electrons from the working substance, and facilitate reaching thermal equilibrium in the hot junction. This cycle repeats, establishing a low temperature steady state in the cold reservoir.
*The cyclic superconducting refrigerator* – Adiabatic magnetization of a superconductor preserves the total entropy of the material such that the entropies of the two phases are equal, $\mathcal{S}^{N}(T_{f}, H=H_{c}) = \mathcal{S}^{S}(T_{i}, H=0)$, where $H$ is the applied magnetic field. This results in cooling of the material to a final temperature $T_{f}$ that is approximately equal to $T_{i}^{3}/T_{*}^{2}$. Here $T_{*}^{2} = \frac{\gamma_{2}}{\alpha_{2}}$, is a characteristic temperature of the working substance [@yaqub1960cooling; @dolcini2009adiabatic]. We consider Tantalum as the working substance for which $T_{*}=11.6$K.
For the cyclic superconducting refrigerator presented here, the crucial point is that the magnetic field inducing the phase transition, when applied quasi-statically, can be reversed quasi-statically to its initial value and therefore reversing the superconducting to normal phase transition of the working substance. This cycle can be performed repeatedly, where the working substance is driven between two different temperatures (hot and cold), envisaging a refrigeration cycle. The energy transfer is asymmetric. That is, energy flow has a preferred direction that is different for the different phases, as a consequence of the energy-structure of the N/S materials [@martinez2015rectification; @giazotto2013thermal; @martinez2013efficient]. The proposed refrigerator is sketched in Fig. \[fig1\] and Fig. \[fridge\] (a).
We assume that the initial temperature of the working substance $T< 0.1~T_{c}$ of the working substance, such that its specific heat in the superconducting state can be approximated by [@yaqub1960cooling], $$C_{S} = 3\alpha T^3 + a \gamma T_{c}\exp (-b T_{c}/T),$$ where $b=1.44$, and $a=9.14$. Here $\alpha$ and $\gamma$ are parameters specifying its specific heat at the normal state, $ C_{N} = 3\alpha T^3 + \gamma T.$ Here the common $T^3$ term is the phononic (Debye) contribution to the specific heat [^1]. In the superconducting case, the exponential behavior of the electronic specific heat of the superconductor at low temperatures can be associated to the presence of a superconducting gap. The critical field as a function of temperature can be found from free energy differences [@dolcini2009adiabatic], which agrees reasonably well with the empirical formula, $H_{c}(T) = H_{0}(1-\frac{T^{2}}{T_{c}^{2}})$ for $T<T_{c}$, where $H_{0}$ is the zero-temperature critical field of the working substance. We consider $B_{0}=\mu_{0}H_{0} \simeq 0.08$T as the critical magnetic field at zero temperature for our working substance, Tantalum [@milne1961superconducting].
We can calculate some ideal thermodynamic properties of the cyclic refrigerator. The temperature of the hot reservoir is $T_{H}$. The working substance is in thermal equilibrium with the hot reservoir initially and the following cycle occurs (See Fig. \[fig1\] and Fig. \[fridge\]):
- [Step A: A quasistatically applied magnetic field drives the working substance to the normal state. The transformation is iso-entropic and the working substance cools down to $T_{1}= T_{H}^{3}/T_{*}^{2}$. Magnetic work $W_{3,1}$ is done.]{}
- [Step B: The working substance is put in contact with the cold reservoir where it absorbs heat $Q_{C}$. Since the electronic contribution to the entropy and specific heat dominates in the normal state, the transferred heat per unit volume can be approximated,]{} $ Q_C = \int T d\mathcal{S} = \frac{\gamma_{2}}{2} (T_C^2 - T_1^2).$
The temperature $T_{C}$ can be identified as the equilibrium final temperature between the cold reservoir and the working substance, approximated as $T_{C} \simeq \sqrt{\frac{\gamma_{2}T_{1}^{2}+\gamma_{1}T_{i}^2}{\gamma_{1}+\gamma_{2}}} > T_{1}$, where $T_{i}$ is the initial temperature of the substrate prior to the cycle. Maximum cooling power is obtained when $T_{i} = T_{H}$, and the cooling power tends to zero when $T_{i}\rightarrow T_{1}$. Here $\gamma_{1}T_{i}$ is the electronic entropy of the substrate.
- [Step C: The electron fluid in the working substance is re-compressed by reducing the magnetic field quasi-statically and adiabatically, where it returns to the superconducting state at temperature $T_{4} = (T_{C}T_{*}^{2})^{1/3}$. Magnetic work $W_{2,4}$ is done.]{}
- [Step D: The working substance is put in contact with the hot reservoir. Since the reservoir has a high specific heat and bandgap, the final temperature achieved can be approximated to the temperature of the hot reservoir $T_{H}$. In this process, the amount of heat transferred to the hot reservoir per unit volume is given by, $ Q_H = \int T dS = \frac{3 \alpha_{2}}{4} (T_4^4 - T_H^4).$ We have approximated the entropy lines for the superconducting state to be only phononic, since the electronic contribution goes to zero exponentially at low temperatures.]{}
Fig. \[fig1\] (a) illustrates this ideal process. By the first law, we have $W_{3,1}+Q_{1,2}+W_{2,4}+Q_{4,3} = 0$. Defining $W=W_{3,1}+W_{2,4}$, we have $W=-Q_{4,3}-Q_{1,2} = Q_{H}-Q_{C}$. The Coefficient of Performance (COP) is the ratio of heat taken from the cold reservoir $Q_{C}$ to work $W$ given by, $$\begin{aligned}
\text{COP} =\frac{Q_{C}}{W} = \frac{t_{C}^{2}-t_{H}^{6}}{\frac{3}{2}(t_{C}^{4/3}-t_{H}^{4})-(t_{C}^{2}-t_{H}^{6})},\end{aligned}$$ where $t_H = T_H/T_*,~t_C = T_C/T_*$. Please see Fig. \[fridge\](b), where we plot the coefficient of performance as a function of $t_{H}=\frac{T_{H}}{T_{*}}$.
In the above idealized analysis, we assume an on/off type energy exchange, so heat transfer to either a hot reservoir or a cold reservoir can be made on demand, like a piston operating a heat-transfer switch. While liquid-gas refrigerators can make a good approximation to this idealized description because of their ability to be freely moved around, solid state systems do not have such freedom. Instead, we must design appropriate physics to effectively turn on and off a switch of exchanging heat with either a hot reservoir or a cold reservoir in order to make an effective solid-state refrigerator. Below, we show that the asymmetry of heat transport between normal metals and superconductors has such a “switch” built in [@martinez2015rectification], which permits selectively cooling down the cold reservoir, due to the presence of an energy gap in the superconductor.
For efficient cooling, it is desired that the working substance, when it is cold, is as much thermally isolated from the hot reservoir as possible, so that significant amount of heat is absorbed from the cold reservoir which we want to cool down. To achieve this, we consider a reservoir superconductor having a larger gap. As a result, there is significantly less back-flow of quasi-particles from the hot reservoir to the working substance when the working substance is colder (in its normal state), and there is more in-flow of heat from the cold reservoir to the working substance, since both are in their normal state. The population of quasi-particle excitations in the reservoir superconductor, which could potentially tunnel back to the working substance when it is colder, are exponentially suppressed by the presence of a large superconducting energy gap in the reservoir. Similarly, the reverse flow of heat from the the working substance to the cold reservoir, when the working substance is hot (superconducting state) is also exponentially suppressed due to the appearance of the superconducting gap. Therefore, in each cycle, there is more heat absorbed from the cold reservoir, than the reverse flow of heat. This is further facilitated by maintaining a high magnetic field for most of the time in each cycle \[see Fig. \[fig3\](c)\] such that the working substance spends most of its time in each cycle in the cold (normal) state. Majority of the excess heat is distributed in the phonon modes of the working substance. By increasing the volume of the working substance (and therefore its specific heat) relative to the volume of the cold reservoir, we also ensure that the temperature of the working substance increase at a relatively slow rate with the amount of heat absorbed, compared to the decrease in temperature of the cold reservoir in each cycle, adding to efficient cooling. Electron-phonon scattering and contact with the large gap superconductor further facilitates achieving thermal equilibrium in the hot junction, by selective removal of high energy quasi-particles from the working substance. In practice, the reservoir superconductor can also come in direct contact with rest of the internal environment of a dilution refrigerator which sets the initial equilibrium temperature. Here the reservoir superconductor also provides additional thermal isolation between the working substance and the base contact, owing to the presence of a large superconducting gap in the reservoir.
A large gap superconductor as the reservoir is also desired, if we need to produce the magnetizing B field by running a super-current in the reservoir superconductor \[see Fig. \[fridge\](c)\], without breaking the reservoir’s own superconductivity. Our choice, Niobium, as the large gap superconductor has a higher critical field ($B_{0}\sim 0.82$ T) compared to Tantalum ($B_{0}\sim 0.08$ T) [@stromberg1965superconducting; @eisenstein1954superconducting; @milne1961superconducting]. Therefore Niobium can sustain the supercurrent that produce the magnetizing B field without breaking its own superconductivity. Niobium is also type II, and therefore it can enter a mixed state with normal vortices. It is still acceptable as the lower critical field above which Niobium enters a mixed state has been measured around 0.19 T [@stromberg1965superconducting; @eisenstein1954superconducting], which is still higher than the critical field of Tantalum. We also assume that the Kapitza coupling [@pollack1969kapitza; @rajauria2007electron; @elo2017thermal] across the tunnel junctions can be avoided by carefully choosing the disordered tunnel barriers such that it causes phonon mismatch, and prevents phonon mediated heat transport. This is another desired feature for the experimental implementation of the refrigeration scheme presented below. Two alternate experimental implementations for our scheme that reduce phonon mediated heat transport between the junctions using suspended membranes are presented in the appendix.
*Continuous adiabatic cooling* – Here we provide a dynamical description for the gradual cooling of a substrate $N_{1}$ in contact with the working substance $S_{2}/N_{2}$, which is subsequently in contact with a hot reservoir, $S_{3}$. The quasiparticle tunneling across the interface and the dissipative effects determine the temperature evolution of the three regions, 1: substrate ($T_{L}$), 2: working substance ($T_{w}$), 3: hot reservoir ($T_{R}$). The adiabatic description for cooling of the working substance with dissipative effects is governed by the relation $\frac{d\mathcal{S}_{w}}{dt} = \frac{P_{w}(t)}{T_{w}(t)}$, where $$\mathcal{S}_{w}(T_{w},t) = x_{N}(T_{w},t)\mathcal{S}_{w}^{N}(T_{w})+(1-x_{N}(T_{w},t))\mathcal{S}_{w}^{S}(T_{w}),$$ and $P_{w}$ is the net dissipative power per unit volume in the working substance, due to thermal contacts and eddy currents, and $T_{w}$ is the temperature of the working substance. Here $x_{N}(T_{w},t)$ is the fraction of normal metal in the working substance at time $t$ given by, $$x_{N}(T_{w},t) = 1-n^{-1}\bigg(1-\frac{H(t)}{H_{c}(T_{w})}\bigg),$$ where $H(t)$ is the applied magnetic field and $n$ is the demagnetization factor of the material. We set $n=5\times 10^{-4}$ for the working substance [@dolcini2009adiabatic]. Variation of $x_{N}$ for our refrigeration protocol in shown in Fig. \[fig3\](d), which shows that the fraction increases from zero to one, and then falls back to zero in the proposed magnetization cycle. The dynamics of the refrigerator is described by the following set of simultaneous differential equations (assuming unit volume): $$\begin{aligned}
C_{N_{1}}(T_{L})\dot{T_{L}} &=& -x_{N}(T_{w},t)P_{N_{1},N_{2}}^{qp}+P_{load}\nonumber\\&-&(1-x_{N}(T_{w},t))P_{N_{1},S_{2}}^{qp}\nonumber\\
C_{w}(H,T_{w})\dot{T_{w}} &=& x_{N}(T_{w},t)(P_{N_{1},N_{2}}^{qp}-P_{N_{2},S_{3}}^{qp})\nonumber\\&+&(1-x_{N}(T_{w},t))(P_{N_{1},S_{2}}^{qp}-P_{S_{2},S_{3}})\nonumber\\&+&P_{mag}+P_{eddy}.
\label{eqcont}\end{aligned}$$ A similar dynamical equation exists for $T_{R}$, but for a large volume of the hot reservoir, and coupling to a support at fixed initial temperature, we can safely assume that $\dot{T_{R}} = 0$. The specific heat $C_{w}$ is the specific heat of the intermediate state, given by [@dolcini2009adiabatic], $$C_{w}(H,T_{w}) = x_{N} C_{N}(T_{w})+(1-x_{N})C_{S}(T_{w})+C^{Lat}_{V}(H,T_{w}),$$ where $$\begin{aligned}
C^{Lat}_{V}(H,T_{w}) &=& (
T_{w} H/\mu_{0}n H_{c}^{3}(T_{w}))\nonumber\\&\times&(\mathcal{S}_{w}^{N}(T_{w},0)-S^{S}_{w}(T_{w},0))^2,\end{aligned}$$ corresponds to the latent heat of the phase transition. The competing cooling power is, $$P_{mag} = \frac{\mu_{0}}{n}T_{w}\frac{dH_{c}(T_{w})}{dT_{w}}\dot{H}.$$
We treat the electron and phonon temperatures identical in Eq. , since electron-phonon relaxation occurs much faster compared to adiabatic magnetization, which is a slow process. Under this assumption, here $P_{load}$ accounts for a small heating contribution from thermal contacts by treating them as hot-spots, where lattice temperature is approximately constant in the immediate neighborhood of the contact [@giazotto2006opportunities]. The heating power at each contact varies as $$P_{\text{ct}}\simeq \Sigma V_{\text{ct}}(T_{i}^{q}-T_{L}^{q}),$$ where to a good approximation, the dissipation is caused due to electron-phonon scattering at the contact (which sets $q=5$), with the phonon /lattice temperature in the neighborhood of the contact held fixed at the initial equilibrium temperature $T_{i}$. The volume $V_{\text{ct}}$ of the thermal hot-spot at the contact is modeled as a sphere of radius $r_{\text{ct}}$. Here $\Sigma = 2\times10^{9}~\text{WK}^{-5}\text{m}^{-3}$ is the electron phonon coupling constant for Cu. In simulations, we consider two such hot-spots with $r_{\text{ct}}\sim$ 600 nm each, accounted by $P_{load}$. This adds only a maximum heating contribution of nearly $\text{1 pW}$ at $T_{i}=200$mK, $\sim$ 0.03 pW at $T_{i}=100$mK and $\sim$ 1 fW at 50mK, which are much smaller compared to the respective cooling powers in the nW range (see Fig. \[fig3\]). The heating contribution further drops down with the ambient temperature $T_{i}$, set by the lowest temperature achievable in a dilution fridge. In each of the cooling curves in Fig. \[fig3\](b), we assume that our refrigerator begins to function from different ambient cold temperatures $T_{i}$ achieved in a dilution fridge, and the contacts are in thermal equilibrium at this ambient temperature, $T_{i}$.
The working substance can heat up due to eddy currents introduced by the magnetic field $B=\mu_{0}x_{N}(T_{w},t)H_{c}(T_{w})$ varies as $P_{eddy}(t)= \frac{A^{2}\dot{B}^{2}}{R_{w}}$, where $A$ is the area which the normal component of the field is passing through, and $R_{w}$ is the bulk resistance of the working substance. Eddy current effects can be reduced by a factor $\propto \frac{1}{N_{w}^{2}}$, by subdividing the bulk into $N_{w}$ thin sheets. We assume $N_{w}\sim 10^{2}$ in the simulations. Note that such a laminar formation occurs naturally in the effective description of the intermediate state where the metal and superconducting phases coexist with alternating thin strips of metal and superconducting phases, and the magnetic field lines pass only through the normal phase [@rose2012introduction].
The quasiparticle power (energy exchange per unit time) transported between two normal metals is, $$\begin{aligned}
P_{N_{1},N_{2}}^{qp} &=&\frac{2}{e^{2}\mathcal{R}} \int_{0}^{\infty} E~dE(\mathcal{F}_{1}(T_{L})-\mathcal{F}_{2}(T_{w}))\nonumber\\
&=&\frac{1}{e^{2}\mathcal{R}} \frac{\pi^{2}k_{B}^{2}}{6}(T_{L}^{2}-T_{w}^{2}).\label{pow}\end{aligned}$$ Here $\mathcal{F}(T)$ is the Fermi-Dirac distribution at temperature $T$, $k_{B}$ is the Boltzmann constant, and $\mathcal{R}=\frac{\mathcal{R}_{s}}{\mathcal{A}}$ is the normal state resistance of the junction. The specific resistance $\mathcal{R}_{s}$ is assumed to be $2\text{M}\Omega\mu \text{m}^{2}$, and identical for both the junctions. As expected, good energy transfer is found (going as a power law of the temperature difference) because of the density of states-matching of the two normal metals. The maximum cooling power provided by the junction can be calculated from Eq. . As noted previously, maximum cooling power is obtained when $T_{L}=T_{i}=T_{H}$, and $T_{w}=\frac{T_{H}^{3}}{T_{*}^2}$. Substituting, we obtain $$\begin{aligned}
P^{\text{c}}_{\text{max}}
&=& \frac{\gamma_{2}}{e^{2}\mathcal{R}~\alpha_{2}} \frac{\pi^{2}k_{B}^{2}}{6}t_{H}^{2}(1-t_{H}^{4}).\end{aligned}$$ Using the parameters mentioned in the caption of Fig. \[fridge\] (c), we obtain $T_{*}=\sqrt{\frac{\gamma_{2}}{\alpha_{2}}} = 11.6$K for Tantalum. For a specific resistance $\mathcal{R}_{s}=2\text{M}\Omega\mu\text{m}^{2}$, a 10cm$\times$10cm contact has resistance $\mathcal{R}=2\times10^{-4}\Omega$, yielding the cooling power at 10mK nearly equal to 6 nW. Further maximizing $P^{\text{c}}_{\text{max}}$ over $t_{H}$ we obtain the optimal point of operation $t_{H}^{\text{max}} = \frac{1}{3^{1/4}}\simeq 0.76$. Since our refrigerator operates below the critical temperature $T_{c_{w}}$ of the working substance, the optimal point of operation is achievable if $t_{c_{w}}=\frac{T_{c_{w}}}{T_{*}}>t_{H}^{\text{max}}$, i.e., when $$0.746\frac{\Delta_{2}}{ k_{B}}\bigg(\frac{\alpha_{2}}{\gamma_{2}}\bigg)^{\frac{1}{2}}>1, \hspace{0.5cm}\text{using}\hspace{0.5cm}k_{B}T_{c} = \frac{\Delta}{1.764}\label{con}$$ from the BCS theory [@bardeen1957theory]. The ideal refrigerator sketched in Fig \[fridge\](c) has a $\text{COP}=1.65$ at this optimal point.
\], starting from 100mK. Temperature of the working substance (blue, oscillating), the temperature variation of the substrate (orange, steady decrease with oscillations $<$1mK), and temperature of the hot reservoir (green) are shown in the figure. (b) Refrigeration action starting from different initial temperatures 0.2K, 0.1K, 0.05K and 10mK, which is the base temperature of a dilution fridge. The operating power of the refrigerator for these initial temperatures are 25 nW, 6 nW, 1.5 nW and 0.06 nW respectively, for a 1cm$\times$1cm interface. For each case, we assume that the hot and cold reservoirs, and the working substance are in thermal equilibrium such that their temperatures are identical before the refrigerator is turned on. (c) The applied magnetic field profile \[same form for different realizations, $H(t) = (1-n)H_{c}(T_{i})+(H_{c}(0)-(1-n)H_{c}(T_{i})) \tanh(t/\tau)$ until $t=4\tau$, and then reduced symmetrically\], and (d) variation of the fraction of normal metal in the working substance during repeated cooling cycles of a junction refrigerator, for $T_{i}=100$mK. \[fig3\]](Fig3.pdf){width="\linewidth"}
Similarly, the quasiparticle power exchange between a normal metal and a superconductor is given by, $$\begin{aligned}
&&P_{N_{1},S_{2}}^{qp} =\frac{2}{e^{2}\mathcal{R}} \int_{\Delta_{2}}^{\infty} E~dE \frac{E}{\sqrt{E^2-\Delta_{2}^2}}(\mathcal{F}_{1}(T_{L})-\mathcal{F}_{2}(T_{w}))\nonumber\\
&&\simeq\frac{2}{e^{2}\mathcal{R}} \bigg[ \bigg(\Delta_{2}^2 K_0\left[\frac{\Delta_{2}}{k_{B}T_{L}} \right] + \Delta_{2} k_B T_L
K_1\left[
\frac{\Delta_{2}}{k_{B} T_{L}} \right]\bigg)\nonumber\\&&- \left(\Delta_{2}^2 K_0\left[\frac{\Delta_{2}}{k_{B}T_{w}}\right] + \Delta_2 k_B T_w K_1\left[
\frac{\Delta_{2}}{k_{B} T_{w}} \right]\right)\bigg].\end{aligned}$$ Here $\Delta_{2}$ is the energy gap of $S_{2}$, and $K_{0,1}(x)$ are modified Bessel functions of order 0 and 1. A similar relation can be found for $P_{N_{2},S_{3}}^{qp}$. In pursuing the integrals, we have assumed low temperatures such that the integrals are effectively approximated using Laplace transformations. For large $\Delta_{2}/k_B T_{L,w}$, the asymptotic expansion of Bessel function, $K_n(x) \sim e^{-x} \sqrt{\pi/2x}$, insures an exponential cut-off of the transport between the N/S junction, acting as the desired switch.
The heat exchange between the two superconducting elements has two contributions, the quasiparticle power exchange, $P_{S_2,S_3}^{qp}$ and a term depending on the Josephson phase, $\phi=\phi_{R}-\phi_{w}$, $P_{S_2,S_3}^{\phi}$ [@maki1965entropy; @giazotto2012josephson]. Here $\phi_{R},~\phi_{w}$ are respectively the phase of the supercondcuting BCS wavefunctions of $R$ and $w$. The quasiparticle tunneling power across the $S_2/S_3$ junction is approximated [@golubev2013heat], $$\begin{aligned}
&&P_{S_{2},S_{3}}^{qp}
\simeq\frac{1}{e^{2}\mathcal{R}} \frac{\sqrt{2 \pi} \Delta_{3}^{5/2}}
{\sqrt{\Delta_{3}^{2} - \Delta_{2}^{2}}} \bigg(\sqrt{k_{B} T_{w}} e^{-\Delta_{3}/(k_B T_{w})} \nonumber\\&&\times\cosh\bigg(\frac{\hbar\dot{\phi}}{2k_{B}T_{w}}\bigg)-
\sqrt{k_{B} T_{R}} e^{-\Delta_{3}/(k_{B} T_{R})}\bigg),\end{aligned}$$ where we have assumed that the difference $\Delta_{3} - \Delta_{2}$ is much bigger than the thermal energies of quasiparticles which help reduce the back-flow of heat from the reservoir to the working substance. The magnitude of the $\phi$ dependent term is always smaller than $P_{S_{2},S_{3}}^{qp}$ and is given by $P_{S_{2},S_{3}}^{\phi}=-\Delta_{2}/\Delta_{3}P_{S_{2},S_{3}}^{qp}\cos{\phi}$ [@golubev2013heat]. In the examples considered, we have taken $T_{c_{w}} = 4.48$K for Tantalum, $T_{c_{R}} = 9.29$K for Niobium. The refrigerator operates below $0.1~T_{c_{w}}$ in the examples presented, where the superconducting gap remains constant at its zero temperature value. In Fig. \[fig3\], we have set $\phi = 0$, and $\dot{\phi}=0$. In general, the relative phase between the superconductors provides another control knob in the problem, and is significant in determining the cooling power when the magnetizing cycles are applied faster than the thermal relaxation time of the Josephson junction [@fornieri2017towards; @martinez2014coherent].
*Discussions* – We proposed a cyclic superconducting refrigerator using the principle of adiabatic magnetization of a superconductor. The refrigerator action is similar to a conventional kitchen refrigerator. Here, the working fluid is the electron gas in a superconductor switching between normal (expanded) and superconducting (compressed) states in an applied magnetic field. Substantial cooling down of a substrate is predicted, as depicted in Fig. \[fig3\](b) for different equilibrium initial temperatures of the refrigerator. Although we discussed a refrigeration scheme in which conventional s-wave superconductivity and the BCS description holds, we note that similar adiabatic cooling effects can be achieved in high-temperature superconductors as well. For example, cooling by adiabatically increasing the super-current in a high temperature superconductor has been studied by Svidzinsky in Ref. [@svidzinsky2002possible], which could be an alternate way of achieving single shot adiabatic cooling with superconductors [@dolcini2009adiabatic]. It should be noted that a high temperature superconductor may increase the operating temperature of the refrigerator, but cooling down a normal metal may still be more efficient in the low temperature regime, as the phonon entropy ($\propto T^{3}$), and electron phonon scattering ($\propto T^{5}$) starts to dominate at high temperatures, effectively nullifying any cooling effect in the normal metal from electron-mediated transport phenomenon at the interface.
We conclude by noting that many variations of our proposal are possible. For example, if another set of metal/superconductor junctions is placed on the other side of the metal to be cooled, then out-of-phase double-action refrigeration is possible, where one side continues to cool the metal down while the other side is heating up and ejecting its excess heat. Our solid state refrigeration technique can be very effective for achieving significant cooling in superconducting circuits, and for applications such as superconducting single photon detectors [@gol2001picosecond] and sensors [@giazotto2008ultrasensitive].
*Acknowledgements* – Work by SKM and ANJ was supported by the U.S. Department of Energy (DOE), Office of Science, Basic Energy Sciences (BES), under Award No. DE-SC0017890. Work by FG was supported by the European Research Council under the European Union’s Seventh Framework Programme (FP7/20072013)/ERC Grant Agreement No. 615187-COMANCHE. This research was supported in part by the National Science Foundation under Grant No. NSF PHY-1748958 (KITP program QTHERMO18). We thank Jukka Pekola, Björn Sothmann, Rafael Sánchez, Matthew LaHaye, and Britton Plourde for discussions, and also Elia Strambini for suggesting the possible method of injecting a super-current in the reservoir superconductor to realize the magnetizing B field.
Alternative implementation schemes for the refrigerator
=======================================================
{width="\linewidth"}
Here we propose two alternative implementation schemes for the refrigeration protocol discussed in the letter. Both use a suspended membrane to reduce the Kapitza coupling in order to inhibit phonon thermal transport between interfaces (See Fig. \[figSM\]).
The Kapitza coupling for an interface between materials $j$ and $k$, with phonon temperatures $T_{ph_{j}}$ and $T_{ph_{k}}$ is given by [@pollack1969kapitza; @rajauria2007electron; @elo2017thermal], $$P_{ph_{j},ph_{k}} = KA(T_{ph_{j}}^{4}-T_{ph_{k}}^{4}),$$ where $A$ is the area of the interface, and $K$ is the coupling$~\sim 200~$W m$^{-2}$ K$^{-4}$, for typical metal interfaces. We stress that in general, different temperatures for electrons and phonons can be investigated in this scheme, as marked in the figures. Negligible Kapitza coupling, and fast electron-phonon interaction relative to the adiabatic magnetization process ensures that we can essentially treat the electron/phonon temperatures the same during the quasi-static operation of the refrigerator, as described in Eq. of the main text.
[^1]: The phononic contribution to specific heat is related to the Debye temperature $\theta_{D}$ by the relation, $C_{ph}(T)=1944\frac{T^{3}}{\theta_{D}^{3}}$J/(mol K) [@pobell1996matter]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A new resistance bridge has been built at the Laboratoire national de métrologie et d’essais (LNE) to improve the ohm realization in the *Système International* (SI) of units from the quantum Hall effect. We describe the instrument, the performance of which relies on the development of two synchronized and noise filtered current sources, an accurate and stable current divider and a cryogenic current comparator (CCC) having a low noise of $\mathrm{80~pA.t/Hz^{1/2}}$. As targeted, the uncertainty budget for the measurement of the 100 $\Omega/(R_\mathrm{K}/2)$ ratio, where $R_\mathrm{K}$ is the von Klitzing constant, amounts to a few parts in $10^{10}$ only.'
author:
- 'Wilfrid Poirier, Dominique Leprat and Félicien Schopfer[^1]'
title: 'A resistance bridge based on a cryogenic current comparator targeting sub-$10^{-9}$ measurement uncertainties'
---
Metrology, resistance, quantum Hall effect, bridge, cryogenic current comparator, SQUID.
Introduction
============
the SI[@BIPM], the ohm can be realized from the von Klitzing constant $R_\mathrm{K}=h/e^2$[@AmpereBIPM2019], where $h$ is the Planck constant and $e$ is the elementary charge, using the quantum Hall effect[@Klitzing1980]. More precisely, the quantized Hall resistance value of a quantum resistance standard (QHR), $R_\mathrm{K}/i$, where $i$ is an integer, is used as a primary reference of resistance[@Poirier2019]. The resistance unit is then disseminated by means of comparisons with this universal and reproducible reference using a resistance bridge. Comparing a resistance with the quantized Hall resistance with the lowest measurement uncertainties is challenging since the measurement current of QHR devices must remain small, i.e. a few tens of $\mu$A if based on GaAs/AlGaAs heterostructure[@Poirier2009] and a few hundreds of $\mu$A if based on graphene[@Ribeiro2015]. Thus, performing a resistance calibration with a relative uncertainty of a few parts in $10^{9}$ requires a resistance bridge very sensitive in current. The most accurate and sensitive bridge able to disseminate the resistance unit[@Poirier2009] is based on the performance of a cryogenic current comparator (CCC). The CCC[@Harvey1972] is basically a perfect transformer operating in direct current regime (dc) able to measure the ratio of the currents circulating through the two resistances to compare with a relative uncertainty below $10^{-10}$. Made of superconducting windings embedded in a superconducting shielding, its accuracy relies on the Messner effect. Its high-sensitivity comes from the flux detector equipping it, which is based on a Superconducting QUantum Interference Device (SQUID)[@Gallop2006]. The development of resistance bridge equipped with a CCC started in several national metrology institutes (NMI), including the French institute, right after the discovery of the QHE. First ones were operating in dc[@Delayahe1985; @Williams1991; @Hartland92; @Dziuba1993]. Then, other bridges adapted to measurements in the low-frequency (below 1 Hz) alternating current regime (ac) were proposed[@Delahaye1991; @Seppa1997]. Accurate operation at higher frequencies was achieved by replacing the CCC with a room-temperature current comparator using high-permeability magnetic cores[@DelahayeAC1993; @Satrapinski2017] but at the expense of larger measurement uncertainties. Since the nineties, international bilateral resistance comparisons[@BIPMEMK12], organized by the *Bureau International des Poids et Mesures* (BIPM), have demonstrated the equivalence of several NMIs using CCC-based resistance bridge (RB) for the realization of the ohm from the QHE with a relative uncertainty of a few parts in $10^9$.
The LNE has been using such a dc bridge[@Delayahe1985; @Piquemal1995] for more than thirty years to perform calibrations of wire resistors with an accuracy of a few parts in $10^9$, as about twenty NMIs do at the present time. This old instrument now suffers several limitations with regards to the current needs not only for calibrations but also for research works: few resistance ratios can be measured (1, 10, 12.9, 50, 64.5, 100, 129), the optimal resistance range extends from 1 $\Omega$ to $R_\mathrm{K}/2$ only, the type B uncertainty (of about 7 parts in $10^{10}$ in relative value for the 100 $\Omega$ calibration from the QHE) cannot be further reduced and the type A uncertainty is limited to one part in $10^{9}$ (for one hour measurement time) by the outdated performance of the radio-frequency SQUID. The improvement of digital and analog electronic components and the availability of dc SQUIDs allow the development of a resistance bridge with better performance in terms of sensitivity, accuracy and automation. Several NMIs have thus recently developed more modern and automated RB[@Sanchez2009; @Drung2009; @Gotz2009; @Williams2010].
Here, we report on a semi-automated RB designed to perform comparisons of resistances, of values ranging from $1~\Omega$ to $\mathrm{1.29~M\Omega}$, in ratios from 1 up to 1290, based on current sources delivering currents from $1~\mu$A up to 100 mA. Compared to the older LNE bridge, a strong performance upgrade is expected due to improvements not only in the technical design but also in the instruments making up the bridge. Our target combined standard uncertainty for the calibration of a $100~\Omega$ wire resistor in terms of $R_\mathrm{K}/2$ is a few parts in $10^{10}$ (k=1). The paper is organized as follows: the principle of the RCB is described in Section II, the design and the performance of the CCC are presented in Section III, Section IV describes the electronics of current sources, Section V describes current dividers used to set the fine tuning of the current ratio, Section VI presents the shielding techniques implemented, measurements of SQUID noise and of resistance ratio are presented in Section VII and finally a conclusion is made in Section VIII.
Principle of the RB
===================
![Principle of the new LNE resistance bridge based on a CCC. The figure shows the two interlocked current sources, the CCC equipped with a DC SQUID and the feedback control on the secondary current source, the current dividers injecting the in-phase, $\epsilon I_\mathrm{S}$, and the in-quadrature, $j\epsilon_{q}I_\mathrm{S}$, current fractions respectively, the null detector and the two resistances $R_\mathrm{S}$ and $R_\mathrm{P}$ to compare. The ground can be connected in position A (low potential of the secondary resistor) or B (low potential of the secondary winding).[]{data-label="fig1"}](Fig-PrincipleSchematic.pdf){width="3.5in"}
The principle of the new RB, described in fig.\[fig1\], is close to that of the older one. It is based on two synchronized sources, primary (P) and secondary (S) sources, that deliver currents $I_\mathrm{P}$ and $I_\mathrm{S}$ respectively. The primary (secondary) source supplies the resistance $R_\mathrm{P}$ ($R_\mathrm{S}$) in series with a superconducting winding of a CCC of number of turns $N_\mathrm{P}$ ($N_\mathrm{S}$). The number of turns $N_\mathrm{S}$ and $N_\mathrm{P}$ are chosen so that the ratio $N_\mathrm{S}/N_\mathrm{P}$ is close to the resistance ratio $R_\mathrm{S}/R_\mathrm{P}$. A standard current divider (SCD) is used to deviate an in-phase calibrated fraction $\epsilon$ of the current $I_\mathrm{S}$ into an auxiliary winding of number of turns $N_\mathrm{A}$. The windings of the CCC are wound according to a toroidal geometry and embedded in a superconducting shielding. Application of the Ampere’s theorem to a circulation along a cross-section of the shielding, where the magnetic flux density is zero, leads to the relationship $N_\mathrm{P}I_\mathrm{P}-(N_\mathrm{S}+\epsilon N_\mathrm{A})I_\mathrm{S}=I_\mathrm{CCC}$, where $I_\mathrm{CCC}$ is a screening current. Because the CCC shielding overlaps itself on two or three turns without electrical contact, this superconducting current circulates from the inner to the outer side of the shielding. It can be therefore detected by a pick-up coil coupled to the outer side and connected to the entry inductance of the SQUID. The secondary current source is servo-controlled by the output of the CCC SQUID electronics so that the screening current $I_\mathrm{CCC}$ (i.e. the total ampere.turn) is nulled. It results that: $$N_\mathrm{P}I_\mathrm{P}-(N_\mathrm{S}+\epsilon N_\mathrm{A})I_\mathrm{S}=0.
\label{Equation:AmpereTurns}$$ From the fraction $\epsilon_0$ setting the voltage unbalance, $R_\mathrm{S}I_\mathrm{S}=R_\mathrm{P}I_\mathrm{P}$, one obtains: $$R_\mathrm{S}/R_\mathrm{P}=(N_\mathrm{S}+\epsilon_0 N_\mathrm{A})/N_\mathrm{P}.
\label{Equation:resistance}$$ The SCD can also be inserted in the primary circuit to deviate a fraction of the current $I_\mathrm{P}$. In this case, the previous equations remain valid by simply exchanging S and P index. This is the operating mode planned for measurements involving a low resistance $R_\mathrm{S}$ (for example 1 $\Omega$) supplied by a large current $I_\mathrm{S}$ (for example 50 mA) which would lead to a too strong dissipation in the SCD if placed in the secondary circuit. Instead, the SCD inserted in the primary circuit is biased by the lower current $I_\mathrm{P}$ which is usually below 10 mA. Improvements of the bridge mainly concern i) the two current sources able to operate in DC and at very low frequencies which are both servo-controlled by a unique external voltage source, ii) the accurate and stable standard current divider used to adjust the current ratio to the resistance ratio, and iii) the new DC SQUID-based CCC.
The new bridge also includes a second current divider able to deviate an in-quadrature current fraction $j\epsilon_{q}I_\mathrm{S}$ in a fourth winding of number of turns $N_\mathrm{A}^q$. It is used to cancel the voltage overshoots occurring during current reversals at the entry of the null detector (ND) that are caused by the capacitances $C_\mathrm{P}$ and $C_\mathrm{S}$ in parallel to the resistors $R_\mathrm{P}$ and $R_\mathrm{S}$ respectively. More precisely, the master equation for ampere.turns becomes: $$N_\mathrm{P}I_\mathrm{P}-(N_\mathrm{S}+\epsilon N_\mathrm{A}+j\epsilon_{q}N_\mathrm{A}^q)I_\mathrm{S}=0.
\label{Equation:AmpereTurns}$$ Assuming first-order approximation in angular frequency $\omega$, the voltage balance condition leads to the first equation \[Equation:resistance\] and a second equation involving capacitances according to: $$(R_\mathrm{P}C_\mathrm{P}-R_\mathrm{S}C_\mathrm{S})\omega=\epsilon_{q}\frac{R_\mathrm{P}N_\mathrm{A}^q}{R_\mathrm{S}N_\mathrm{P}}.
\label{Equation:Phase}$$ The quadrature current divider (QCD) is therefore used to compensate quadrature signals caused by capacitances. By cancelling the voltage overshoots that can saturate the null detector, the reversal frequency of the current can be increased. This reduces the impact of voltage offset drift and of the $1/f$ SQUID noise on measurements. Moreover, it allows optimizing the ratio between the acquisition time and the total experience time. The calibration of the QCD fractions is not required.
Finally, careful shielding of cables and guarding of circuits are implemented to ensure the equality of the currents circulating through the resistor and the winding for better accuracy.
the CCC
=======
Design and fabrication
----------------------
![Pictures of the CCC at different stages of shielding assembly. a) The CCC alone. b) The CCC on the probe with all shields removed. c) First Pb/Brass shield of the CCC in place. d) Second Pb/Brass shield of the SQUID in place. e) External cryoperm shield in place.[]{data-label="fig2"}](Fig2.pdf){width="3.5in"}
** The cryogenic current comparator is made of 15 windings of 1, 1, 2, 4, 16, 16, 32, 64, 128, 160, 160, 1600, 1600, 2065 and 2065 turns which hold together with epoxy glue[@Soukiassian2010]. Each winding is made of superconducing and insulated $\mathrm{60~\mu m}$ diameter NbTi/Cu wire. We used optically checked 0.1 mm thick Pb sheets and Pb/Sn/Cd superconducting solder at a temperature lower than $150^{\circ}$C to realize the toroidal shielding around the windings. Our shield overlaps twice (3 layers) to prevent flux leakage. Each layer is covered with PTFE (poly-tetra-fluoro-ethylene) tape for electrical insulation. The CCC is fixed to the cryogenic probe with a piece in MACOR$^{\scriptsize\textregistered}$ material. The DC-SQUID (Quantum Design, Inc) has an input inductance of $L_i=1.8~\mu H$ and a nominal flux noise in flux-lock feedback mode of $\mathrm{3~\mu\phi_0/Hz^{1/2}}$ above 0.3 Hz. It is coupled to the CCC via a superconducting flux transformer made of a NbTi wire inserted in a lead tube. The sensitivity of the system $S_\mathrm{CCC}$ depends on the number of turns $N_\mathrm{PC}$ of the pick-up coil coupled to the CCC through the relationship:\
$$S_\mathrm{CCC}=(2/k)N_\mathrm{PC}S_\mathrm{SQ},$$ where $k$ is the coupling constant between the CCC and the pickup coil, and $S_\mathrm{SQ}$ the SQUID sensitivity (in $\mathrm{\mu A/\phi_0}$). The best sensitivity $S_\mathrm{CCC}^\mathrm{opt}$ is obtained for $N_\mathrm{PC}^\mathrm{opt}$ given by:\
$$N_\mathrm{PC}^\mathrm{opt}=\sqrt{L_i/L^\mathrm{eff}_\mathrm{CCC}},$$\
where $L^\mathrm{eff}_\mathrm{CCC}$ is the effective self inductance of the CCC taking account of the superconducting screen that isolates the SQUID from external magnetic fields (Earth’s field for instance). $L^\mathrm{eff}_\mathrm{CCC}$ and therefore $S_\mathrm{CCC}$ can be determined in a given geometry using the analytical calculation of Sesé and co-authors[@Sese1999; @Sese2003]. In our case, one calculates $L^\mathrm{eff}_\mathrm{CCC}\sim 14$ nH, $N_\mathrm{PC}^\mathrm{opt}=12$ and $S_\mathrm{CCC}^\mathrm{opt}=\mathrm{5~\mu A.t/\phi_0}$. Due to geometrical constraints, the number of turns of the pick-up coil was reduced to $N_\mathrm{PC}=6$ leading to an experimental sensitivity $S_\mathrm{CCC}=\mathrm{8~\mu A.t/\phi_0}$ close to the calculated value of $\mathrm{6~\mu A.t/\phi_0}$. Our magnetic screen is made up of 5 concentric cylinders: two Pb ones, each one embedded in a brass one, and an outer Cryoperm® cylinder. Each cylinder is closed at the top with the same material. The cryogenic probe body is made of three rods to stabilize it mechanically and prevent from vibrations.
Noise performance
-----------------
![Noise spectral density measured by the SQUID *versus* frequency *f* for the CCC alone and not connected.[]{data-label="Fig-NoiseCCCAlone"}](Fig-NoiseCCCAlone.pdf){width="3.5in"}
The CCC was firstly tested with all windings disconnected. Fig.\[Fig-NoiseCCCAlone\] shows the noise spectral density in $\mathrm{\phi_0/Hz^{1/2}}$ as a function of the frequency. The main frequency resonance due to the coupling of the large inductance of windings and the capacitance between wires is around 14 kHz. Between 6 Hz and about 2 kHz, the noise spectral density is dominated by sharp peaks with an amplitude lower than $1~\mathrm{m\phi_0/Hz^{1/2}}$ which are caused by mechanical resonances. At lower frequencies down to about 0.1 Hz, there exists a white noise regime with a constant noise spectral density of about $10~\mathrm{\mu \phi_0/Hz^{1/2}}$. Considering the CCC sensitivity of $\mathrm{8~\mu A.t/\phi_0}$, this leads to a current sensitivity of about $\mathrm{80~pA/Hz^{1/2}}$. One can guess *1/f* noise above $10~\mathrm{\mu \phi_0/Hz^{1/2}}$ at frequencies lower than 0.1 Hz what is expected if considering the 0.3 Hz corner frequency and the $3~\mathrm{\mu \phi_0/Hz^{1/2}}$ white noise level of the Quantum Design DC SQUID. We therefore conclude that the current reversal frequency of the resistance bridge should be higher than 0.1 Hz to obtain the lowest measurement noise.
Accuracy
--------
The accuracy of the CCC was tested by realising winding opposition experiments. Windings with same nominal number of turns are connected in series-opposition and supplied by a large current. The voltage at the SQUID output $V_\mathrm{Output}$ operating in internal feedback mode (MODE 5) is converted in magnetic flux $\delta_{\phi_0}$ using the SQUID gain in $V/\phi_0$. By dividing $\delta_{\phi_0}$ by the total magnetic flux $NI/S_\mathrm{CCC}$ generated by one winding, one obtains the relative error $\Delta N/N= \delta_{\phi_0} S_\mathrm{CCC}/(NI)$. We have used the very low noise current source of the RB to carry out these experiments in order to prevent from noise rectification by the SQUID. The measurement current is reversed from 100 mA to -100 mA to remove offset voltages from recorded data. For turn numbers *N* equal to 16 and 2065 which are used in the calibration of a $100~\Omega$ resistor in terms of $R_\mathrm{K}/2$, $\Delta N/N$ is found equal to $(1.9 \pm 1.2)\times 10 ^{-11}$ and $(2.5 \pm 0.04)\times 10 ^{-11}$ respectively. For all other winding opposition, turn errors are found smaller than $6\times10^{-11}$, except for 1-1 and 2-2 combinations which seem let us conclude to significant errors of $\sim1\times10^{-9}$ and $\sim5\times10^{-10}$. But, a magnetic flux leakage caused by a hole in the toroidal shield or by a imperfect chimney would manifest by large errors for all winding opposition. Our interpretation is that these apparent errors are only caused by spurious signals coming from residual noise rectification which manifest themselves all the more as the total ampere.turn number is small, i.e. for 1-1 and 2-2 winding opposition. Further reduction of the noise emitted by the current source in the 100 mA range (which is the most noisy range) is required to refine the determination of winding errors for small turn numbers.
\[1\][>m[\#1]{}]{}
[|c|M[5cm]{}|]{} **Winding combination**& **$\Delta N/N$** 1-1&$(1.29\pm 0.38)\times10^{-9}$ 2-2&$(4.64\pm 0.63)\times10^{-10}$ 16-16&$(1.9\pm 1.2)\times10^{-11}$ 16+16-32&$(2.5\pm 1.1)\times10^{-11}$ 16+16+32-64&$(0.22\pm 0.76)\times10^{-11}$ 16+16+32+64-128&$(1.0\pm 1.1)\times10^{-11}$ 160-160&$(2.5\pm 0.14)\times10^{-11}$ 160-128-32&$(0.5\pm 0.4)\times10^{-11}$ 1600-1600&$(5.6\pm 0.032)\times10^{-11}$ 2065-2065&$(2.54\pm 0.035)\times10^{-11}$\
Our conclusion is that the CCC error in the measurements of usual resistance ratios, which exploited mainly windings of number of turns 2065, 1600, 160, and 16, is of a few $10^{-11}$. The CCC contribution to the type-B uncertainty is therefore below $10^{-10}$, in relative value.
The current source electronics
==============================
Design and fabrication
----------------------
![Schemes of the electronic circuits of primary and secondary current sources. Both current sources are controlled by a unique external voltage source.[]{data-label="Fig-Circuits"}](Fig-Circuits.pdf){width="3.5in"}
The RB is based on two current sources generating the currents $I_\mathrm{P}$ and $I_\mathrm{S}$ which supply the resistors $R_\mathrm{P}$ and $R_\mathrm{S}$ respectively. In some recently developed resistance bridges, current sources are based on digital electronics connected by fiber optics to a internal micro-controller[@Drung2009] or an external PXI computer. This provides strong electrical insulation and easy automation but requires the implementation of efficient noise filtering techniques to protect the SQUID from the radio-frequency noise emitted by digital circuits.
On contrary, current sources of the LNE RB are based on linear analog circuits to avoid high-frequency noise[@Soukiassian2010]. Fig.\[Fig-Circuits\] shows the schematic of the two electronic circuits, the primary and the secondary ones. They are controlled and synchronized by a single external voltage source allowing automation of measurements. The latter can be either a dc voltage generator, like a Yogogawa 7651 for usual dc measurements or the oscillator of a lock-in detector for low-frequency ac measurements. The external reference voltage supplies the primary circuit through a high-impedance differential amplifier. A low-pass filter is then used to limit the signal bandwidth with an adjustable cutoff frequency ranging from 1 mHz to 1 kHz. After a stage summing additional voltage corrections and a division stage allowing the setting of a decimal fraction, the voltage is converted into a current in ranges extending from 1 $\mu$A up to 100 mA. This conversion is done using an amplifier inverter circuit boosted by a buffer amplifier (BUF634T) and a dividing resistor. The secondary current source, controlled by the output signal of the low-pass filter of the primary current source, is similar but the current range selected can also be multiplied by a factor 1.2906 or 1/1.2906 to adapt to measurements involving the QHR connected either to the primary or to the secondary current sources. Several additional circuits are implemented to finely adjust the current ratio $r_I=I_\mathrm{S}/I_\mathrm{P}$ to within a few parts in $10^{6}$. This is necessary to limit the ampere.turn unbalance in the CCC, not only to avoid unlocking of the SQUID feedback notably during current switching, but also to achieve the best accuracy in the $I_\mathrm{S}/I_\mathrm{P}$ ratio adjustment. Offset, in-phase and in-quadrature correction circuits are used to tune the secondary current while an asymmetry correction circuit injecting a fraction of the voltage absolute value is used in the primary circuit to compensate, to some extent, the asymmetry behaviour of operational amplifiers. Finally, the SQUID feedback voltage, after insulation by an differential amplifier, is converted into a feedback current at the last stage of the secondary circuit so that the closed-loop feedback gain remains the same as in internal feedback mode, i.e. $\sim0.75~V/\phi_0$.
Electronic circuits of each current source are integrated, but electrically isolated with PTFE material, into their own metallic box connected to ground, as shown in pictures of fig.\[FigAppendix-Sources\]. The electronic components are powered by stabilized voltages provided by a circuit itself energized by rechargeable batteries. These are also electrically isolated from the grounded metallic box in which they are placed. The only electrical link between the electronic circuits and the ground comes from the high-impedance operational amplifiers (OPA128LM) which ensure a high-isolation (in principle $\sim10^{15} ~\Omega$ resistance) from the piloting external voltage source and the SQUID feedback electronics. All these precautions aim at cancelling leakage currents.
Test of the current ratio adjustability
---------------------------------------
![Illustration of the adjustability of the current ratio $I_\mathrm{P}/I_\mathrm{S}$ using resistances $R_\mathrm{P}=\mathrm{10~k\Omega}$ and $R_\mathrm{S}=\mathrm{100~\Omega}$. The balance voltage $\Delta V$, measured by the null detector (ND) is recorded as a function of time while the current is periodically reversed (red dashed line) for different settings of the corrections circuits : adjustment of the offset correction only (blue line), adjustment of the in-quadrature correction (green line), adjustment of in-phase, in-quadrature and asymmetry corrections (deep blue line).[]{data-label="Fig-CurrentAdjustment"}](Fig-CurrentAdjustment.pdf){width="3.5in"}
Fig.\[Fig-CurrentAdjustment\] shows the experiment carried out to test the adjustability of the ratio of the two current sources[@Soukiassian2010]. Two resistors of resistance $\mathrm{10~k\Omega}$ and $\mathrm{100~\Omega}$ are fed by currents $I_\mathrm{P}$ and $I_\mathrm{S}$ respectively. The potential drop difference $\Delta V$ at the terminals of the two resistors is recorded by a null detector (nanovoltmeter EMN11). For a nominal voltage reversing from 1 V to -1 V every 20 secondes, Fig.\[Fig-CurrentAdjustment\] shows that it is possible to reduce the peak to peak $\Delta V$ amplitude to less than $\mathrm{2~\mu V}$ by optimizing the in-phase, the in-quadrature, the offset and the asymmetry corrections. Let us notably remark the effect of the in-quadrature correction in cancelling the voltage overshot caused by the fast current reversal. The current ratio can therefore be adjusted with an accuracy of 2 parts in $10^6$ even during fast current reversal which is an advantage to avoid any SQUID unlocking.
Noise optimization and filtering {#Noise optimization and filtering}
--------------------------------
![Electronic scheme of the two last stages (A and B) of the secondary current source (the primary current source does not include the SQUID feedback circuit) describing the noise filtering techniques. Picture of the common mode torus used to block the circulation of current noise towards ground. The dotted line represents the limit of the case (at ground) of the current source.[]{data-label="Fig-Filtering"}](Fig-Filtering.pdf){width="3.5in"}
Noise filtering is crucial particularly from the resonance frequency of the CCC (14 kHz) up to the operating frequency of the modulation circuit (500 kHz) not only to ensure a good working of the SQUID but also to avoid noise rectification that would alter measurement accuracy. Fig.\[Fig-Filtering\] shows the last stage of the electronic circuit of the secondary current source. The primary current source is based on a similar stage but differs by the absence of the SQUID feedback electronic circuit. In practice, the frequency bandwidth of the primary and secondary current sources was reduced to 160 Hz at the stage A and 1 kHz at the stage B of the electronic circuit using simple low-pass filters based on resistors and capacitors. By this way, the frequency bandwidth of the SQUID feedback circuit servo-controlling the secondary current is set to 1 kHz. Table.\[tableau:Range\] summarizes the capacitance $C_\mathrm{F}$ and resistance $R_\mathrm{F}$ values chosen to set the 1 kHz cut-off frequency that damps the CCC resonance for each current range (defined by the value of the resistor $R_\mathrm{C}$), considering the nominal value of the resistance in measurement $R_\mathrm{E}$.
It is also essential to avoid the circulation of the current noise coming from the capacitive coupling of the electronics circuit with ground. To cancel this noise source which renders the SQUID inoperative, a common mode torus (CMT) was introduced in the current circuit of each source (see fig.\[Fig-Filtering\]). This CMT is made of a PTFE insulated wire pair wounded about 60 times around an APERAM Nano magnetic torus (magnetic permeability of about 80000 up to a 100 kHz frequency) with a return spire. The differential inductance is around $\mathrm{3~\mu H}$ while the common mode inductance is around 0.6 H. The common mode impedance, which increases from about $200~\Omega$ at 50 Hz up to $\mathrm{150~k\Omega}$ at 1 MHz, drastically reduces the circulation of the common mode current noise. This protects the SQUID and makes it operating quite ideally whether it is a radio-frequency SQUID or a DC SQUID.
\[1\][>m[\#1]{}]{}
**Range** **$R_\mathrm{C}$** **$R_\mathrm{E}$** **$C_\mathrm{F}$** **$R_\mathrm{F}$**
---------------------- -------------------- -------------------- -------------------- --------------------
1 $\mathrm{\mu A}$ 5 M$\Omega$ 1 M$\Omega$ 300 pF 4 M$\Omega$
10 $\mathrm{\mu A}$ 500 k$\Omega$ 100 k$\Omega$ 300 pF 400 k$\Omega$
100 $\mathrm{\mu A}$ 50 k$\Omega$ 10 k$\Omega$ 3 nF 40 k$\Omega$
1 mA 5 k$\Omega$ 1 k$\Omega$ 30 nF 4 k$\Omega$
10 mA 500 $\Omega$ 100 $\Omega$ 300 nF 400 $\Omega$
100 mA 50 $\Omega$ 10 $\Omega$ 3 $\mathrm{\mu F}$ 40 $\Omega$
: Resistance and capacitance values in the last stage of the secondary current source for the different current ranges.[]{data-label="tableau:Range"}
The SQUID feedback circuit
--------------------------
As shown in fig.\[Fig-Filtering\], the SQUID feedback voltage $V_\mathrm{SQUID}$ probed in the SQUID Quantum Design Preamplifier is sent to the secondary current source of the bridge after decoupling by a high impedance differential amplifier. A resistor is biased by the $V_\mathrm{SQUID}$ voltage to inject the feedback current in the secondary current circuit supplying the secondary winding ($N_\mathrm{S}$). The resistor value, $R_\mathrm{FB}=\mathrm{1.5~M\Omega}$, is chosen so that the closed-loop feedback gain $G_\mathrm{CLG}=R_\mathrm{FB}\times S_\mathrm{CCC}/N_\mathrm{S}$ for for $N_\mathrm{S}=16$ is the same as in the most sensitive internal feedback mode of the SQUID (mode 5 and mode 5s) equal to $0.75~V/\phi_0$. A circuit made of a small capacitance of 200 pF in series with a $\mathrm{20~k\Omega}$ resistor is connected in parallel to the $\mathrm{1.5~M\Omega}$ resistor to partially compensate the dephasing caused by the 1 kHz low-pass filter implemented in the stage B of the electronic circuit and therefore optimize the SQUID operation.
Current dividers
================
The standard (or in-phase) current divider
------------------------------------------
The standard current divider (SCD) is used to balance the voltages measured at the terminals of the two resistors by deviating a fraction of current towards the auxiliary winding. This is a key element of the RB, the accuracy of which directly impacts the uncertainty budget of the RB. An alternative technique to null the voltage measured by the detector consists in using an auxiliary current source servo-controlled by the null detector voltage output[@Williams1991; @Hartland92]. This relies on an active component, a second feedback electronics and an accurate measurement of the current delivered. On contrary, the SCD developed is a passive component that, once calibrated, can inject a known current. Moreover, it avoids the use of a second feedback electronics. However, one requirement is a good stability of the current fractions defined by the SCD. This is achieved with a design that limits the number of electrical commutations required to select the current fraction. The counterpart of these technical choices aiming at better reproducibility is that the calibration of the LNE SCD is not fully-automated contrary to that of binary compensation units[@Drung2013; @Gotz2017].
![Electrical scheme of the standard current divider (SCD). $N$, $P$, $Q$, $P'$, $Q'$ are integers defining the the setting of the SCD. A CMT is inserted between the SCD and the auxiliary winding to reduce the current noise circulation through ground.[]{data-label="fig-SCD"}](Fig-SCD.pdf){width="3.5in"}
The SCD, described in fig.\[fig-SCD\], is built to inject in the auxiliary winding fractions of a main current (lower than 10 mA) ranging from 0 to $5\times10^{-5}$ by minimal step of $5\times10^{-8}$. Made of three main series resistor networks ($10\times 20~\Omega$, $10\times 2~\Omega$, $10\times 200~\Omega$) and a large $\mathrm{4~M\Omega}$ division resistor, the SCD ratio can be adjusted using three mechanical IEC MONACO commutators with goal-coated silver contacts (see picture in fig.\[FigAppendix-SCD\] of appendix section). The current fraction is given by: $\epsilon_{(N,P,Q)}= N\times5\times10^{-6}+P\times5\times10^{-7}+Q\times5\times10^{-8}$ where *N*, *P*, *Q* are the integer values between 0 and 10 indexing the three commutator positions, respectively.
To achieve this ideal behavior, it was necessary to implement two additional compensations resistor networks in order to circumvent the non-linearities that result from the variation of the division resistance varying *P* and *Q*. Indeed, the triangle formed by resistors ($2~\Omega$, $Q\times 200~\Omega$, $(10-Q)\times 200~\Omega$) leads to the addition of a *Q* dependent resistance to the $\mathrm{4~M\Omega}$ resistance. To solve this non-linearity, a compensation resistance in series with the $\mathrm{4~M\Omega}$ resistance is selected for each *Q* value from the resistor network ($0~\Omega$, $20~\Omega$, $80~\Omega$, $180~\Omega$, $320~\Omega$, $500~\Omega$, $320~\Omega$, $180~\Omega$, $80~\Omega$, $20~\Omega$, $0~\Omega$). The variation of the division resistance changing the *P*-dependent position of the potentiometer is compensated by selecting a fraction of the resistor network ($10\times 2~\Omega$) using the index *P’*=10-*P*. Finally, the fraction of the resistor network ($10\times 20~\Omega$) selected by *N’* index adds to keep constant the total resistance of the SCD independently of the *N* index change: *N’*=10-*N*. This is useful to set the frequency bandwidth of the current source independently of the SCD setting. To achieve the best stability of the SCD fractions with time and under load, resistor networks are constituted of high stability (drift lower than $10^{-5}$/year, in relative value), low temperature coefficient ($<0.6\times10^{-6}/^{\circ}$C) and hermetic Vishay resistors (VH516-4 $20~\Omega$, VHS102 $2~\Omega$, VH518-10 $\mathrm{1~M\Omega}$, VH102K $200~\Omega$) connected in series with low electromotive force solder. Finally, a CMT is introduced that drastically reduces the current noise signals circulating from ground to the auxiliary winding through the SCD circuit.
The fractions corresponding to the three following sets of commutator positions: ($N$, $P=0$, $Q=0$) with $N$ varying from zero to ten, ($N=0$, $P$, $Q=0$) with $P$ varying from zero to ten, ($N=0$, $P=0$, $Q$) with $Q$ varying from zero to ten, were calibrated by measuring the ratio of the voltage $V_\mathrm{AB}$ to a reference bias voltage applied in place of the auxiliary winding. Calibrations performed over ten years showed that fractions have remained close to their nominal value within one part in $10^9$, drifting each by no more than 5 parts in $10^{10}$. More generally, the fraction $\epsilon_{(N,P,Q)}$ corresponding to the commutator position ($N$, $P$, $Q$) is given by : $$\begin{aligned}
\epsilon_{(N,P,Q)}&=\epsilon_{(N,P=0,Q=0)}+\epsilon_{(N=0,P,Q=0)}\\
&+\epsilon_{(N=0,P=0,Q)}-2\epsilon_{(0,0,0)}.\end{aligned}$$
The quadrature current divider
------------------------------
Capacitive leakages to ground short-circuiting the secondary resistor lead to voltage overshoots during current variations that can possibly saturate the null detector. This limits the maximum current reversing speed. To circumvent this difficulty, a second current divider (QCD) is connected in series with the SCD to cancel these voltage overshoots (see fig.\[fig1\]) by injecting an in-quadrature fraction $j\epsilon_q$ of the main current in a CCC auxiliary winding of number of turns ${N_\mathrm{A}^q}$, as highlighted by equations \[Equation:AmpereTurns\] and \[Equation:Phase\]. More precisely, the main current is flowing through a $R_\mathrm{q}$ resistor of $1~\Omega$ or $10~\Omega$ resistance value. A $100~\Omega$ potentiometer connected in parallel allows the adjustment of the voltage fraction $\alpha$ biasing a $C_\mathrm{q}=235$ nF PTFE capacitor (its parallel resistance is higher than $2\times 10^{13}~\Omega$) in series with the CCC auxiliary winding (typically ${N_\mathrm{A}^q}=1600$). The current fraction injected by the QCD is therefore $j\epsilon_q\simeq j\alpha R_\mathrm{q}C_\mathrm{q}\omega$, where $\alpha\in[0:1]$ is the potentiometer fraction. It is crucial that this current remains negligible during the data acquisition. Considering a current setting time constant of 0.5 s, one can calculate that the current fraction drops down to no more than a few parts in $10^{13}$ of the main current after a waiting time of only 8 s. Let us note that the quadrature current divider is also equipped with a CMT to reduce current noise circulation (see picture in fig.\[FigAppendix-SCD\] of appendix section).
Shielding and guarding
======================
![Left: Picture of the resistance bridge. Right: Picture of the CCC winding switching box at the top of the cryostat.[]{data-label="fig4"}](Fig4.pdf){width="3.5in"}
As shown in fig. \[fig4\], each sensitive element of the RB is carefully shielded against noise. The null detector, the current sources, the current dividers, the power supply are in grounded metallic boxes. The QHR and the CCC are each in an independent cryostat connected at ground. The continuity of the shielding between the different elements is ensured by the connection cables, the metallic sheath of which is also connected at ground as schematized in fig.\[fig1\]. This directs any leakage current between wires at different potentials towards the ground. In normal operation, the ground is connected both to the low potential of the resistance $R_\mathrm{S}$ (position A in fig.\[fig1\]) and to the case of the EM detector (there is no common mode voltage). The leakage current, $I_g$ therefore short-circuits the lowest resistance (black arrow) which reduces its impact. This grounding is usually efficient to limit the leakage current contribution to the type B uncertainty below $10^{-9}$ for the measurement of the $100~\Omega/(R_\mathrm{K}/2)$ ratio.
The ground can also be connected at the low potential of the secondary winding (position B in fig.\[fig1\]). The leakage current, $I_g$, short-circuiting both the resistor and the winding of the secondary circuit (grey arrow) in this case, the measurement accuracy of the resistance ratio is not altered at all by leakage currents. But, this is at the expense of a common mode voltage, although weak, existing between the ground and the low potential of the null detector due to the small resistance (about 2 $\Omega$) of the winding. To minimize its effect on the measurement accuracy, the case and the low potential of the null detector are short-circuited. Besides, capacitance hand-effects are cancelled because the null detector is itself placed in a grounded metallic box.
Resistance ratio Measurements
=============================
Noise spectrum and SQUID feedback stability
-------------------------------------------
![Noise spectral density measured by the SQUID *versus* frequency *f* for the measurement of the 100$\Omega$/10 k$\Omega$ ratio. SQUID operating in internal feedback mode 5 (black), in external feedback mode 5s (red), and in external feedback mode 500 (blue).[]{data-label="Fig-NoiseCCC10k100"}](Fig-NoiseCCC10k100.pdf){width="3.5in"}
The operation stability of the SQUID was demonstrated for several ratio measurements: 10 k$\Omega$/1 M$\Omega$, 100 $\Omega$/10 k$\Omega$ and 1 $\Omega$/100 $\Omega$. The following settings of the bridge were used: $N_\mathrm{P}=1600$, $N_\mathrm{S}=16$, $N_\mathrm{A}=16$ and a current divider connected in series with the secondary current source for the two first ratios, $N_\mathrm{P}=1600$, $N_\mathrm{S}=16$, $N_\mathrm{A}=1600$ and a current divider connected in series with the primary current source for the ratio 100 $\Omega$/1 $\Omega$. The quadrature divider was not used for these tests.
Fig.\[Fig-NoiseCCC10k100\] shows the noise spectral density, expressed in $\mathrm{\phi_0/Hz^{1/2}}$, determined by the Quantum Design SQUID operating in different feedback modes for the measurement of the 100 $\Omega$/10 k$\Omega$ ratio using current ranges $10$ mA/100 $\mu$A. In closed feedback mode operation, the output noise corresponds to the difference between the noises emitted by the primary and the secondary current sources since the current ratio $I_\mathrm{S}/I_\mathrm{P}$ is adjusted within $10^{-6}$ to cancel to ampere.turn unbalance of the CCC, i.e. the magnetic flux in the SQUID. So there remains only the uncorrelated noise contributions of both current sources. Let us note that the residual magnetic flux noise crossing the SQUID itself has a much lower level because of the real-time compensation by the feedback signal. It is given by the combination of the intrinsic SQUID noise (3 $\mathrm{\mu\phi_0/Hz^{1/2}}$), the environmental noise directly captured by the SQUID and the current source noise divided by the open-loop amplification gain. This latter contribution is negligible. The two others, which manifest in the CCC alone and disconnected, give a contribution of about 10 $\mathrm{\mu\phi_0/Hz^{1/2}}$ as observed in fig.\[Fig-NoiseCCCAlone\].
The internal (through the modulation coil of the SQUID) and the external (through the CCC winding) feedback mode operations mainly differ by their bandwidth as one can observe in fig.\[Fig-NoiseCCC10k100\]. In internal feedback mode 5, the bandwidth of about 20 kHz allows measuring the noise level up to the frequency resonances of the CCC. Fig.\[Fig-NoiseCCC10k100\] shows that the noise amplitude is above a noise floor level of about 140 $\mathrm{\mu\phi_0/Hz^{1/2}}$. This bottom level is notably explained by the Johnson-Nyquist noise, of 120 $\mathrm{\mu\phi_0/Hz^{1/2}}$, generated by the $R_\mathrm{C}=50$ k$\Omega$ resistor defining the 100 $\mu$A current range of the primary current source. Below 10 Hz, the noise increase is mainly caused by the $1/f$ voltage noise of the operational amplifiers (OPA111BM) which polarizes the 50 k$\Omega$ dividing resistor. The operation in external feedback mode 5s is very stable. The noise spectrum is similar but is characterized by a lower frequency bandwidth limited to about 500 Hz by the SQUID electronics. The cutoff frequency manifests itself by a peak, above which the signal then decreases of 20 dB by decade. The operation in external feedback mode 500 is also stable but with a higher cutoff frequency of about 1 kHz defined by the secondary current source filters (in fact the 200 pF capacitance of the feedback circuit (fig.\[Fig-Filtering\]) slightly extends the frequency bandwidth above 1 kHz set by the $C_\mathrm{F}$ capacitance). In some configurations, this larger frequency bandwidth can ensure a better stability of the bridge operation against higher-frequencies acoustic noises.
![Noise spectral density measured by the SQUID *versus* frequency *f* for the measurement of the 1 $\Omega$/100 $\Omega$ ratio a) and 10 k$\Omega$/1 M$\Omega$ ratio b). SQUID operating in internal feedback mode 5 (black) and in external feedback mode 5s (red).[]{data-label="Fig-NoiseCCCAutres"}](Fig-NoiseCCCAutres.pdf){width="3.5in"}
Fig.\[Fig-NoiseCCCAutres\]a) and b) demonstrate stability of operation of the external SQUID feedback in the measurements of ratios 1$\Omega$/100 $\Omega$ and 10 k$\Omega$/1 M$\Omega$ respectively. The base noise level is larger for the measurement of the 1 $\Omega$/100 $\Omega$ ratio (above $\mathrm{{2~m\phi_0/Hz^{1/2}}}$). This comes from the reduction to $R_\mathrm{C}=5$ k$\Omega$ of the resistor defining the 1 mA range used to supply the 100 $\Omega$ resistor. Conversely, the current sources are feebly noisy in the measurement configuration of the 10 k$\Omega$/1 M$\Omega$ ratio because of the $R_\mathrm{C}=5$ M$\Omega$ resistor defining the 1 $\mu$A range. One can observe in fig.\[Fig-NoiseCCCAutres\]b), a white noise level of no more than $\mathrm{20~\mu\phi_0/Hz^{1/2}}$ between 0.2 Hz and 6 Hz. This low base noise level allows observing the manifestation of moderate mechanical resonances in the range from 10 Hz and 1 kHz.
Measurement protocol and type A uncertainty
-------------------------------------------
![Measurement of the resistance ratio $r_R=100~\Omega/(R_\mathrm{K}/2)$ using the old a) and the new b) LNE bridge: relative voltage deviation $\Delta V/V$ as a function of time for several ($\mathrm{I^+}$, 0, $\mathrm{I^-}$) sequences. The signal period is about 200 s. The following settings were used: $N_\mathrm{P}=1936$, $N_\mathrm{S}=15$ and $N_\mathrm{A}=15$ for the older bridge and $N_\mathrm{P}=2065$, $N_\mathrm{S}=16$, $N_\mathrm{A}=16$ and $N_\mathrm{A}^q=1600$ for the new bridge. $\epsilon$ fractions are chosen to obtain similar deviation amplitude, $\Delta V/V$, for both bridges. $\epsilon_q$ setting of the new bridge QCD is optimized to cancel overshoots at current reversals.[]{data-label="Fig-StabilitySteps"}](Fig-StabilitySteps.pdf){width="3.5in"}
Measurements of the resistance ratio $r_R=100~\Omega/(R_\mathrm{K}/2)$ were performed using the old and the new LNE bridges. The primary current circulating through the GaAs/AlGaAs-based quantum resistance standard is set to $I_\mathrm{P}=70~\mu$A. For a $\epsilon$ fraction of the SCD which differs from $\epsilon_0$, a finite voltage $\Delta V$ can be detected by the null detector. The relative voltage $\Delta V/V$, where $V=R_\mathrm{P}I_\mathrm{P}$, is related, at the first order, to the deviation $(\epsilon-\epsilon_0)$ by: $$\Delta V/V=(\epsilon-\epsilon_0)\frac{N_\mathrm{A}}{N_\mathrm{S}}.$$ Let us note that, reversely, $\Delta V/V$ could be interpreted as a relative deviation of the ratio $r_R$ to the value $r_{R_0}$ giving $\Delta V=0$ for the fraction $\epsilon$. $\Delta V/V$ measured as a function of time during several ($\mathrm{I^+}$, 0, $\mathrm{I^-}$) sequences is reported in Fig.\[Fig-StabilitySteps\]a) and b) for the old and the new LNE bridge respectively. The signal period, of about 200 s, is imposed by the low-speed capability of the older bridge. The comparison of both data first shows the lower noise level and better stability achieved in measurements performed with the new bridge. This comes not only from the better performance of the current source electronics but also from the lower noise level of the CCC. Second, it demonstrates the efficiency and interest of the quadrature current divider which allows the cancellation of any voltage overshoot during the current switchings. This is useful to speed up the current reversal which reduces the impact of the voltage offset drift and of the $1/f$ SQUID noise.
![Measurement of the resistance ratio $r_R=100~\Omega/(R_\mathrm{K}/2)$ using the new LNE bridge with a primary current $I_\mathrm{P}=70~\mu$A: relative deviation $\Delta V/V$ as a function of time for 16 ($\mathrm{I^+}$, $\mathrm{I^-}$, $\mathrm{I^+}$) sequences and two successive settings, $\epsilon^+=-12.1\times10^{-6}$ and $\epsilon^-=+12.6\times10^{-6}$, of the SCD fractions.[]{data-label="Fig-QuickSequence"}](Fig-QuickSequence.pdf){width="3.5in"}
Fig.\[Fig-QuickSequence\] shows the typical data record for the measurement of the resistance ratio $r_R=100~\Omega/(R_\mathrm{K}/2)$ with the new bridge. It consists of two successive acquisitions of voltage measurements, $V^{\epsilon^+}$ and $V^{\epsilon^-}$, that are obtained for the two settings of the SCD fractions $\epsilon^+=-12.1\times10^{-6}$ and $\epsilon^-=+12.6\times10^{-6}$ respectively. Each acquisition is made of 16 ($\mathrm{I^+}$, $\mathrm{I^-}$, $\mathrm{I^+}$) sequences of current reversal that are used to remove voltage offsets. A mean voltage value is calculated from the average of the 16 values $[V(I^+)_1+V(I^+)_3-2V(I^-)_2]/4$, where 1,2,3 index the current state of each sequence. The resistance ratio is then obtained from the $\epsilon_0$ value calculated from the two mean voltages, $<V^{\epsilon^-}>$ and $<V^{\epsilon^+}>$, according to $\epsilon_0=\epsilon^- + (\epsilon^+-\epsilon^-)\times |<V^{\epsilon^-}>|/(|<V^{\epsilon^-}>|+|<V^{\epsilon^+}>|)$. Owing to the new bridge performances, the period of the signal was therefore reduced to about 70 s and the zero crossing step was removed. It results that the ratio between the acquisition time and the total experiment time is increased from about 50 percents with the older bridge to 75 percents with the new bridge, which is favorable to a reduction of the type A uncertainty.
![Measurement of the resistance ratio $r_R=100~\Omega/(R_\mathrm{K}/2)$ using the new LNE bridge with a primary current $I_\mathrm{P}=50~\mu$A. From an experiment described in supplementary of [@Ribeiro2015]. Standard Allan deviation of $r_R$, expressed in relative value, as a function of the acquisition time $t$ (red square). $t^{-1/2}$ adjustment of data (black dashed line). Standard deviation of the mean (blue open hexagon). Inset: ($\mathrm{I^+}$, $\mathrm{I^-}$, $\mathrm{I^+}$) sequence of $\Delta V$ measurements characterized by a period of 70 s.[]{data-label="Fig-Allan"}](Fig-Allan.pdf){width="3.5in"}
The noise performance of the new LNE bridge is demonstrated by the calculation of the Allan standard deviation[@Allan1987; @Witt2005] of $r_R=100~\Omega/(R_\mathrm{K}/2)$ from the statistical analysis of the voltage measurements performed using a primary current $I_\mathrm{P}=50~\mu$A. The evolution of this quantity, expressed in relative value, is reported in fig.\[Fig-Allan\] as a function of the experience time $t$. It follows a $t^{-1/2}$ law over measuring times longer than one hour. This shows that the white noise is dominant and that the standard deviation of the mean can be used to estimate the type A uncertainty. It follows that a type A standard uncertainty of $1.5\times10^{-10}$ can be achieved for the measurement of the ratio $r_R=100~\Omega/(R_\mathrm{K}/2)$ using a current of $50~\mu$A after an experience time of one hour. This is five time less than the best uncertainty achievable with the older bridge. This improvement relies not only on the quicker measurement protocol but also on the lower noise of the current source electronics and of the CCC. Let us remark that the contribution of the CCC to the voltage noise at the terminals of the null detector is no more than $\mathrm{0.5~nV/Hz^{1/2}}$. This is ten times lower than the EMN11 nanovoltmeter contribution, of about $\mathrm{5~nV/Hz^{1/2}}$, which limits the bridge type A uncertainty.
Preliminary uncertainty budget
------------------------------
A preliminary uncertainty budget, reported in table \[tableau:Uncertainty\], was established for the measurement of the ratio $100~\Omega/(R_\mathrm{K}/2)$. As demonstrated in the previous section, a type A uncertainty of $0.15\times10^{-9}$ can indeed be achieved for a 1 hour measurement experiment and a $I_\mathrm{P}=50~\mu$A measurement current. Further improvement would require the implementation of a lower-noise null detector.
Several components contributing to the Type B uncertainty were estimated as reported in Table \[tableau:Uncertainty\]. From the error measurements of the number of turns of windings, one can deduce a contribution of the CCC accuracy below $0.1\times10^{-9}$. However, lower-noise oppositions of windings of one and two turns are required in future to confirm the accuracy in measurements using such small turn numbers.
The accuracy of the resistance ratio measurement depends also on the SQUID feedback accuracy in cancelling the total ampere.turns number, i.e. in setting the current ratio $r_I=I_\mathrm{S}/I_\mathrm{P}$ to the target ratio given by the equation \[Equation:AmpereTurns\]. A setpoint error comes from the finite value of the open loop gain $G_\mathrm{OLG}$ of the SQUID electronics. Although the SQUID amplifier is based on an integrator which leads to an infinite gain at DC, measurements are in practice carried out with a finite time periodicity of the current reversal, typically of 70 s. The relative current ratio error is equal to $(\Delta^{adj}r_I/r_I)\times(G_\mathrm{CLG}/G_\mathrm{OLG})$, where $\Delta^{adj}r_I/r_I$ is the relative deviation between the target ratio and the preliminary adjusted current ratio. Measurements of the $100~\Omega/(R_\mathrm{K}/2)$ resistance ratio were performed for $\Delta^{adj}r_I/r_I$ values as large as a few $10^{-4}$. A relative error in the measurement of the resistance ratio lower than $10^{-11}$ is deduced in normal adjustment of the current ratio, i.e. for $\Delta^{adj}r_I/r_I\sim 10^{-6}$. This corresponds to $G_\mathrm{OLG}>7.5\times10^{4}$ V/$\phi_0$. This value is compatible with that yet determined in quantized current experiment[@Brun-Picard2016].
A main contribution comes from the calibration and stability of the standard current divider fractions. Characterizations performed over 10 years have demonstrated a very low drift of the fraction values, less than $5\times10^{-11}$/year in average value. At the present time, the calibration method of a fraction $\epsilon$ gives an uncertainty below $0.5\times10^{-9}$. A new calibration method, under development, aims at decreasing the uncertainty down to $0.3\times10^{-9}$ in order to benefit from the high-stability of the new standard current divider.
\[1\][>m[\#1]{}]{}
**Uncertainty components (k=1)** **Contributions ($10^{-9}$)**
---------------------------------- -------------------------------
**Type A (1 hour)** **0.15**
**Type B** **0.7 (A), 0.5 (B)**
CCC accuracy $<0.1$
SQUID feedback accuracy $<0.01$
Current divider calibration $<0.5$
Leakage to ground $\sim 0.5$ (A), 0.1 (B)
**Combined Uncertainty** **0.7 (A), 0.6 (B)**
: Preliminary relative uncertainty budget for the $100~\Omega/(R_\mathrm{K}/2)$ ratio considering a $I_\mathrm{P}=50~\mu$A measurement current.[]{data-label="tableau:Uncertainty"}
The impact of leakage current to ground on the measurement accuracy was also estimated through several experiments. At first, no significant deviation was found within a relative uncertainty of about 0.8 part in $10^{9}$ between measurements of the resistance ratios $100~\Omega/(R_\mathrm{K}/2)$ and $200~\Omega/(R_\mathrm{K}/2)$, performed with the ground either connected in position A ($I_g$ parallel to $R_\mathrm{S}$) or position B ($I_g$ fully deviated). To obtain a better knowledge of leakage currents, comparisons were repeated with a larger secondary resistance $R_\mathrm{S}$=1 k$\Omega$ to amplify their effect. From the comparison of the two measurements of the ratio 1 k$\Omega/$1 k$\Omega$ performed with the resistors interchanged, a relative deviation of $(-3\pm 0.4)\times10^{-9}$ is found for the ground in position A which reduces to $(-0.04\pm 0.3)\times10^{-9}$ for the ground in position B. A similar discrepancy of a few parts in $10^{9}$ is also found by comparing the measurements of the ratio 1 k$\Omega/(R_\mathrm{K}/2)$ obtained for both ground positions. Moreover, it is found that the value obtained with the ground in position B agrees within 1.2 parts in $10^{9}$ with that deduced by combining the measurements of the 100 $\Omega/(R_\mathrm{K}/2)$ and 100 $\Omega/1$ k$\Omega$ ratios. One concludes that a significant discrepancy of a few parts in $10^9$ caused by leakage currents exists but can nevertheless be fully cancelled moving the ground in position B.
From these characterizations, a leakage current effect of about 0.5 part in $10^{9}$, i.e. ten times lower, can therefore be deduced for the measurement of the 100 $\Omega/(R_\mathrm{K}/2)$ ratio with the ground in position A. The contribution to the type B uncertainty budget falls below 0.1 part in $10^{9}$ by connecting the ground in position B.
Finally, no significant effect of the current reversal duration (from $I^+$ to $I^-$ and reversely) was found within a relative uncertainty of 0.35 part in $10^9$ by varying its value from 12 s to 24 s while keeping the same acquisition time.
To conclude, the total type B relative uncertainty is estimated to be either 0.7 part or 0.5 part in $10^{9}$ depending on whether the ground is connected to position A or B. The type A uncertainty being lower, the combined uncertainty is below one part in $10^{9}$. Further reduction of the measurement uncertainty will come from the improvement of the current divider calibration.
Validation of measurement accuracy
----------------------------------
The new LNE bridge was used to perform accurate universality tests of the QHE[@Lafont2015; @Ribeiro2015]. The agreement of the quantized Hall resistance, $R_\mathrm{H}$, measured in GaAs and graphene devices was demonstrated with a record[@Ribeiro2015] relative uncertainty of $8\times10^{-11}$. This result was obtained by comparing the two measurements of the ratio $100~\Omega/(R_\mathrm{H}/2)$ carried out using a 100 $\Omega$ transfer resistor. This performance therefore emphasizes the low-noise level and the reproducibility of the measurement bridge, rather than its accuracy. Besides, the capability of the resistance bridge to perform measurements at low frequency (2 Hz) allowed the determination of the temperature evolution of the quantized Hall resistance in graphene during dynamic temperature drift[@Lafont2015]. Many elements of the resistance bridge, i.e. the CCC, the current source and the current divider, were also used to build the programmable quantum current generator that allowed a practical realization of the ampere from the elementary charge with a $10^{-8}$ relative uncertainty[@Brun-Picard2016].
Table \[tableau:Comparisons\] reports on the deviations between the measurements of the ratios $100~\Omega/(R_\mathrm{K}/2)$ and $100~\Omega/10~k\Omega$ performed using the new and the old bridges. It shows that there is no significant discrepancy within a combined uncertainty below 1.5 part in $10^{9}$. Let us note that the comparison uncertainty is limited by the larger type A uncertainty of the older bridge. This agreement between the two resistance bridges, which differ not only by their electronics but also by their CCC and standard current divider, make us very confident in our measurements of these resistance ratios. It also consolidate the Type B uncertainty budget described previously in table \[tableau:Uncertainty\].
\[1\][>m[\#1]{}]{}
**Ratio** **$100~\Omega/(R_\mathrm{K}/2)$** **$100~\Omega/10~k\Omega$**
------------------------ ----------------------------------- -----------------------------
**Relative deviation** $(-1.4\pm1.5)\times10^{-9}$ $(-0.7\pm1.4)\times10^{-9}$
: Relative deviations with combined standard uncertainties (k=1) between the measurements of the resistance ratio performed by the new and the older bridges.[]{data-label="tableau:Comparisons"}
Conclusion
==========
A new comparison resistance bridge based on a CCC was built at LNE. It is based on low-noise synchronized current sources that are carefully shielded and electrically isolated from ground, a new CCC with a very low-noise level of $\mathrm{80~pA.t/Hz^{1/2}}$ and a very stable standard current divider characterized by a drift of less than 0.5 part in $10^{9}$ over ten years. Stable operation of the resistance bridge, i.e. of the SQUID in external closed feedback mode, was yet demonstrated in the measurements of ratios $100~\Omega/(R_\mathrm{K}/2)$, $\mathrm{100~\Omega/10~k\Omega}$, $\mathrm{10~k\Omega/1~M\Omega}$ and $1~\Omega/100~\Omega$. The ratio $100~\Omega/(R_\mathrm{K}/2)$ ratio can be determined with a relative type A uncertainty below $0.15\times 10^{-9}$ within one hour measurement time. This performance results not only from the lower noise of the bridge and particularly of the new CCC but also from the optimization of the data acquisition thanks to the quadrature current divider which cancels voltage overshoots. Main contributions to the type B uncertainty budget have yet been estimated. They concern the standard current divider calibration, the CCC accuracy, the finite open-loop feedback gain and the electrical leakage current. The total type B uncertainty is estimated to be below 0.7 part in $10^{9}$. A further reduction to about 0.4 part in $10^{9}$ is expected using both a new calibration method of the standard current divider and the cancellation method of the leakage currents (position B). Thus, a combined standard uncertainty of 0.5 part in $10^{9}$ is expected at term.
Next development steps will consist in characterizations of the measurement of the $1~\Omega/100~\Omega$ and $\mathrm{10~k\Omega/1~M\Omega}$ resistance ratios. Preliminary experiments have yet demonstrated that the measurement of the $1~\Omega/100~\Omega$ resistance ratio with the current dividers inserted in the primary circuit works. Measurement of the $\mathrm{10~k\Omega/1~M\Omega}$ resistance ratio will require a more adapted null detector than the EMN 11 or EMN 31. The bias current noise of these devices is indeed too large. A battery-powered higher-impedance amplifier is therefore under development. Moreover, an efficient rejection of leakage currents is essential. We therefore plan to test not only the connection of the ground in position B but also the implementation of a virtual ground, as described in [@Delahaye1991].
Pictures of resistance bridge components
========================================
![Pictures (front a) and b), top c)) of the primary and secondary current sources, each one being placed in an independent box.[]{data-label="FigAppendix-Sources"}](FigAppendix-Sources.pdf){width="3.5in"}
![Pictures of the standard current divider (top) and of the quadrature current divider (bottom), each one being placed in an independent box.[]{data-label="FigAppendix-SCD"}](FigAppendix-SCD-reduc.pdf){width="3.5in"}
Acknowledgment {#acknowledgment .unnumbered}
==============
The authors would like to thank Guillaume Spengler and Laetitia Soukiassian for their work during the first stage of the development of the resistance measurement bridge that started ten years ago as well as François Piquemal for useful comments on the manuscript. The authors would like also to thank Carlos Sanchez from NRC for useful discussions about the resistance bridge operation with a ground connected to the secondary winding.
[10]{} \[1\][\#1]{} url@samestyle \[2\][\#2]{} \[2\][[l@\#1=l@\#1\#2]{}]{}
*The International System of units (SI)*.1em plus 0.5em minus 0.4emhttp://www.bipm.org/en/si/, Sèvres.
*Mise en pratique for the definition of the ampere and other electric units in the SI*.1em plus 0.5em minus 0.4emSI Brochure-9th edition (2019)-Appendix 2, https://www.bipm.org/utils/en/pdf/si-mep/SI-App2-ampere.pdf.
K. von Klitzing, G. Dorda, and M. Pepper, “New method for high-accuracy determination of the fine-structure constant based on quantized hall resistance,” *Phys. Rev. Lett.*, vol. 45, p. 494, 1980.
W. Poirier, S. Djordjevic, F. Schopfer, and O. Thévenot, “The ampere and the electrical units in the quantum era,” *C. R. Physique*, vol. 20, p. 92, 2019.
W. Poirier and F. Schopfer, “Resistance metrology based on the quantum hall effect,” *Eur. Phys. J. Spec. Top.*, vol. 172, p. 207, 2009.
R. Ribeiro-Palau, F. Lafont, J. Brun-Picard, D. Kazazis, A. Michon, F. Cheynis, O. Couturaud, C. Consejo, B. Jouault, W. Poirier, and F. Schopfer, “Quantum hall resistance standard in graphene devices under relaxed experimental conditions,” *Nature Nano.*, vol. 10, pp. 965–971, 2015.
I. K. Harvey, “A precise low temperature dc ratio transformer,” *Rev. Sci. Instrum.*, vol. 43, no. 11, pp. 1626–1629, 1972.
J. Gallop and F. Piquemal, “Squid handbook vol. ii,” pp. 95–137, 2006.
F. Delahaye and D. Reymann, “Progress in resistance ratio measurements using a cryogenic current comparator at lcie,” *IEEE Trans. Instrum. Meas.*, vol. 34, p. 316, 1985.
J. M. Williams and A. Hartland, “An automated cryogenic current comparator resistance ratio bridge,” *IEEE Trans. Instrum. Meas.*, vol. 40, p. 267, 1991.
A. Hartland, “The quantum hall effect and resistance standards,” *Metrologia*, vol. 29, pp. 175–190, 1992.
R. F. Dziuba and R. E. Elmquist, “Improvements in resistance scaling at nist using cryogenic current comparators,” *IEEE Trans. Instrum. Meas.*, vol. 42, p. 126, 1993.
F. Delahaye, “An ac-bridge for low-frequency measurements of the quantized hall resistance,” *IEEE Trans. Instrum. Meas.*, vol. 40, p. 883, 1991.
H. Seppa and A. Satrapinski, “Ac resistance bridge based on the cryogenic current comparator,” *IEEE Trans. Instrum. Meas.*, vol. 46, p. 463, 1997.
F. Delahaye and D. Bournaud, “Accurate ac measurements of standard resistors between 1 and 20 hz,” *IEEE Trans. Instrum. Meas.*, vol. 42, p. 287, 1993.
A. Satrapinski, M. Gotz, E. Pesel, N. Fletcher, P. Gournay, and B. Rolland, “New generation of low-frequency current comparators operated at room temperature,” *IEEE Trans. Instrum. Meas.*, vol. 66, p. 1417, 2017.
BIPM, “The bipm key comparison database (kcdb), key and supplementary comparisons (appendix b), comparison bipm.em-k12. https://kcdb.bipm.org.”
F. Delahaye, T. Witt, F. Piquemal, and G. Genevès, “Comparison of quantum hall effect resistance standards of the bnm/lcie and the bipm,” *IEEE Trans. Instrum. Meas.*, vol. 44, p. 258, 1995.
C. A. Sanchez, B. M. Wood, and A. D. Inglis, “Ccc bridge with digitally controlled current sources,” *IEEE Trans. Instrum. Meas.*, vol. 58, p. 1202, 2009.
D. Drung, M. G[ö]{}tz, E. Pesel, J.-H. Storm, C. Assmann, M. Peters, and T. Schurig, “Improving the stability of cryogenic current comparator setups,” *Supercond. Sci. Technol.*, vol. 22, p. 114004, 2009.
M. G[ö]{}tz, D. Drung, E. Pesel, H. Barthelmess, C. Hinnrichs, C. Assmann, M. Peters, H. Scherer, B. Schumacher, and T. Schurig, “Improved cryogenic current comparator setup with digital current sources,” *IEEE Trans. Instrum. Meas.*, vol. 58, p. 1176, 2009.
J. M. Williams, T. J. B. M. Janssen, G. Rietveld, and E. Houtzager, “An automated cryogenic current comparator resistance ratio bridge for routine resistance measurements,” *Metrologia*, vol. 47, p. 167, 2010.
L. Soukiassian, G. Spengler, D. Leprat, F. Schopfer, and W. Poirier, “New cryogenic current comparator-based resistance comparison bridge at lne,” in *2010 CPEM Digest*, Y. S. Song and J.-S. Kang, Eds.1em plus 0.5em minus 0.4emIEEE, June 2010, p. 761, dOI:10.1109/CPEM.2010.5544176.
J. Sesé, F. Lera, A. Camon, and C. Rillo, “Calculation of effective inductances of superconducting devices. application to the cryogenic current comparator,” *IEEE Trans. Appl. Supercon.*, vol. 9, no. 1, pp. 58–62, 1999.
J. Sesé, E. Bartolomé, J. Flokstra, G. Rietveld, A. Camon, and C. Rillo, “Simplified calculus for the design of a cryogenic current comparator,” *IEEE Trans. Instrum. Meas.*, vol. 52, p. 612, 2003.
D. Drung, M. G[ö]{}tz, E. Pesel, H. J. Barthelmess, and C. Hinnrichs, “Aspects of application and calibration of a binary compensation unit for cryogenic current comparator setups,” *IEEE Trans. Instrum. Meas.*, vol. 62, p. 2820, 2013.
M. G[ö]{}tz and D. Drung, “Stability and performance of the binary compensation unit for cryogenic current comparator bridges,” *IEEE Trans. Instrum. Meas.*, vol. 66, p. 1467, 2017.
D. Allan, “Should the classical variance be used as a basic measure in standards metrology ?” *IEEE Trans. Instrum. Meas.*, vol. 36, no. 2, pp. 646–654, 1987.
T. Witt, “Allan variances and spectral densities for dc voltage measurements with polarity reversals,” *IEEE Trans. Instrum. Meas.*, vol. 54, p. 550, 2005.
J. Brun-Picard, S. Djordjevic, D. Leprat, F. Schopfer, and W. Poirier, “Practical quantum realization of the ampere from the elementary charge,” *Phys. Rev. X*, vol. 6, p. 041051, 2016.
F. Lafont, R. Ribeiro-Palau, D. Kazazis, A. Michon, O. Couturaud, C. Consejo, T. Chassagne, M. Zielinski, M. Portail, B. Jouault, F. Schopfer, and W. Poirier, “Quantum hall resistance standards from graphene grown by chemical vapour deposition on silicon carbide,” *Nature Communications*, vol. 6, p. 6806, 2015.
[Wilfrid Poirier]{} is a graduate of the Ecole Supérieure de Physique et de Chimie Industrielles de Paris (ESPCI). He then completed his thesis on quantum electronic transport at the CEA-SPEC and received his doctorate in solid state physics in 1997. He joined LCIE in 1998 and then LNE in 2001 as head of studies on quantum resistance standards. Since then, he has devoted his research to quantum electrical metrology. These included the development of a graphene quantum resistance standard, of GaAs-based quantum Hall arrays and the realization of universality tests of the quantum Hall effect. He has also been involved in the development of precision quantum instrumentation based on SQUID technology. More recently, he has proposed and developed a quantum current generator to achieve the new definition of ampere. He obtained his Habilitation to Direct Research from the University of Paris-SUD in 2017.
[Dominique Leprat]{} worked in LNE from 2001 to 2017 as a technician in the electrical metrology department. He was first involved in the activity of traceability of alternating voltage based on thermal transfer. In 2007, he joined the quantum Hall effect team to develop metrology instrumentation.
[Félicien Schopfer]{} obtained an engineer’s degree from École Nationale Supérieure de Physique - Grenoble INP in 2001, a master’s degree in condensed matter physics from the University of Grenoble the same year. He received a PhD in Physics from the University of Grenoble in 2005, for quantum electronic transport experiments in nanostructures carried out at CNRS. He was appointed at Laboratoire national de métrologie et d’essais – LNE in 2005 to advance research in quantum electrical metrology. His research has mainly focused on the quantum Hall effect (QHE) for applications in fundamental metrology. He worked on quantum Hall arrays in GaAs/AlGaAs, co-authored reproducibility and universality tests of the quantum Hall effect with record uncertainties, and is strongly involved in graphene research, notably with important results for the development of the quantum Hall resistance standard operating under relaxed experimental conditions.
[^1]: Authors are with the Department of Fundamental Electrical Metrology, Laboratoire national de métrologie et d’essais, 78197 Trappes, France; e-mail: [email protected].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Demands on the disaster response capacity of the European Union are likely to increase, as the impacts of disasters continue to grow both in size and frequency. This has resulted in intensive research on issues concerning spatially-explicit information and modelling and their multiple sources of uncertainty. Geospatial support is one of the forms of assistance frequently required by emergency response centres along with hazard forecast and event management assessment. Robust modelling of natural hazards requires dynamic simulations under an array of multiple inputs from different sources. Uncertainty is associated with meteorological forecast and calibration of the model parameters. Software uncertainty also derives from the data transformation models (D-TM) needed for predicting hazard behaviour and its consequences. On the other hand, social contributions have recently been recognized as valuable in raw-data collection and mapping efforts traditionally dominated by professional organizations. Here an architecture overview is proposed for adaptive and robust modelling of natural hazards, following the *Semantic Array Programming* paradigm to also include the distributed array of social contributors called *Citizen Sensor* in a semantically-enhanced strategy for D-TM modelling. The modelling architecture proposes a multi-criteria approach for assessing the array of potential impacts with qualitative rapid assessment methods [based on a *Partial Open Loop Feedback Control* (POLFC) schema and complementing more traditional and]{} accurate a-posteriori assessment. We discuss the computational aspect of environmental risk modelling using [array-based]{} parallel paradigms on *High Performance Computing* (HPC) platforms, in order for the implications of urgency to be introduced into the systems (Urgent-HPC).
[[**Keywords**]{}: Geospatial, Integrated Natural Resources Modelling and Management, Semantic Array Programming, Warning System, Remote Sensing, Parallel Application, High Performance Computing, Partial Open Loop Feedback Control ]{}
title: 'Architecture of Environmental Risk Modelling: for a faster and more robust response to natural disasters'
---
[Dario Rodriguez-Aseretto $^{a,}$[^1], Christian Schaerer $^{b}$, Daniele de Rigo $^{a,c}$ ]{}
[$^a$ European Commission, Joint Research Centre, Institute for Environment and Sustainability, Ispra (VA), Italy.\
$^b$ Polytechnic School, National University of Asuncion, San Lorenzo, Central, Paraguay.\
$^c$ Politecnico di Milano, Dipartimento di Elettronica, Informazione e Bioingegneria, Milano, Italy. ]{}
### 1. INTRODUCTION: CONTEXT, PITFALLS AND THE SCIENCE-POLICY INTERFACE {#introduction-context-pitfalls-and-the-science-policy-interface .unnumbered}
Europe experienced a series of particularly severe disasters in the [recent years [@Sippel_Otto_2014; @Cirella_etal_2014], with worrying potential impacts of similar disasters under future projected scenarios of economy, society and climate change [@Ciscar_etal_2013; @Ciscar_etal_2014]. They range]{} from flash floods [[@Dankers_Feyen_2008; @Gaume_etal_2009; @Marchi_etal_2010]]{} and severe storms in Western Europe [with an expected increasing intensity trend [@Feser_etal_2014]]{}, large-scale floods in Central Europe[ [@Jongman_etal_2014],]{} volcanic ash cloud[s]{} [@Self_2006; @Swindles_2011; @Gramling_2014] [(e.g. ]{}after the Eyjafjallajökull eruption[)]{}, large forest fires in Portugal and Mediterranean countries [[@Allard_etal_2013; @Schmuck_etal_2014]. Biological invasions such as emerging plant pests and diseases have the potential to further interact e.g. with wildfires [@Nijhuis_2012] and to impact on ecosystem services [@Boyd_etal_2013] and economy with substantial uncertainties [@Venette_etal_2012]]{}.
It should be underlined that these recent highlights are set in the context of systemic changes in key sectors [@Maes_etal_2013; @EC_2013; @EC_2013b] which overall may be expected to at least persist in the next decades. As a general trend, demands on the EU’s resilience in preparedness and disaster response capacity are likely to increase, as the impacts of disasters continue to grow both in size and frequency, even considering only the growing exposure (societal factors) [@Barredo_2007; @Barredo_2010]. The aforementioned examples of disturbances are often characterised by non-local system feedbacks and off-site impacts which may connect multiple natural resources (system of systems) [@Evans_etal_2012; @Phillis_Kouikoglou; @Langmann_2014]. In this particular multifaceted context [@Gottret_2001; @Hagmann_2001; @Zhang_2004], landscape [@Estreguil_etal_2013] and ecosystem dynamics show intense interactions with disturbances [@Turner2010].
As a consequence, classical disciplinary and domain-specific approaches which might be perfectly suitable at local-scale may easily result in unacceptable simplifications within a broader context. A broad perspective is also vital for investigating future natural-hazard patterns at regional/continental scale and adapting preparedness planning [@Van_Westen_2013; @Urban_etal_2012; @Baklanov_2007]. The complexity and uncertainty associated with these interactions – along with the severity and variety of the involved impacts [@Steffen_2011] – urge robust, holistic coordinated [@White_etal_2012] and transparent approaches [@deRigo2013; @deRigoSubm]. At the same time, the very complexity itself of the control-system problems involved [@Lempert_2002; @Rammel_2007; @van_der_Sluijs_2012] may force the analysis to enter into the region of deep-uncertainty [@de_Rigo_etal_IFIP2013].
The mathematization of systems in this context as a formal control problem should be able to establish an effective science-policy interface, which is *not* a trivial aspect. This is easily recognised even just considering the peculiarities – which have been well known for a long time – of geospatially-aware environmental data [@Guariso] and decision support systems [@GuarisoEDSS; @Guariso_1985; @Soncini_Sessa_2007], their entanglement with growingly complex ICT aspects [@Guariso_1994; @Casagrandi] and their not infrequent cross-sectoral characterisation. Several pitfalls may degrade the real-world usefulness of the mathematization/implementation process. While it is relatively intuitive how a poor mathematization with a too simplistic approach might result in a failure, subtle pitfalls may lie even where an “appropriately advanced” theoretical approach is proposed. Mathematization should resist *silo thinking* [@Cole_2010; @Sterman_2002] temptations such as academic solution-driven pressures [@deRigoSubm; @Weichselgartner_2010] to force the problem into fashionable “hot topics” of control theory: robust approximations of the real-world broad complexity may serve egregiously instead of state-of-art solutions of oversimplified problems.
Other long-lasting academic claims are “towards” fully automated scientific workflows in computational science, maybe including self-healing and self-adapting capabilities of the computational models implementing the mathematization. These kinds of claims might easily prompt some irony [@Bainbridge_1983] among experienced practitioners in wide-scale transdisciplinary modelling for environment (WSTMe, [@GeoSemAP_2013]) as a never-ending research Pandora’s box with doubtful net advantages [@Stensson_2013]. Complex, highly uncertain and sensitive problems for policy and society, as WSTMe problems typically are, will possibly never be suitable for full automation: even in this family of problems, “humans will always be part of the computational process” [@Russell_2003] also for vital accountability aspects [@Anderson_2003].
While a certain level of autonomic computing [@Kephart_2003] capabilities might be essential for the evolvability and robustness of WSTMe (in particular, perhaps, a higher level of semantic awareness in computational models and a self-adapting ability to scale up to the multiple dimensions of the arrays of data/parameters; see next section), here the potential pitfall is the illusion of *fully automating* WSTMe. The domain of applicability of this puristic academic *silo* – although promising for relatively simple, well-defined (and not too policy-sensitive) case studies – might be intrinsically too narrow for climbing up to deal with the *wicked problems* typical of complex environmental systems [@van_der_Sluijs_2005; @Frame_2008; @McGuire_2010].
The discussed pitfalls might deserve a brief summary. First, perhaps, is the risk of “solving the wrong problem precisely” [@Bea_etal_2009] by neglecting key sources of uncertainty – e.g. unsuitable to be modelled within the “warmly supported” solution of a given research group. During emergency operations, the risks of providing a “myopic decision support” should be emphasised; i.e. suggesting inappropriate actions [@Adams_Hester_2012] – e.g. inaction or missing precaution – due to the potential overwhelming lack of information [@Larsson_etal_2010] or the oversimplification/underestimation of potential chains of impacts due to the lack of computational resources for a decent (perhaps even qualitative and approximate) rapid assessment of them.
Overcoming these pitfalls is still an open issue. Here, we would like to contribute to the debate by proposing the integrated use of some mitigation approaches. We focus on some general aspects of the *modelling architecture* for the computational science support, in order for emergency-operators, decision-makers, stakeholders and citizens to be involved in a participatory [@Innocenti_Albrito_2011] information and decision support system which assimilates uncertainty and precaution [@Ravetz_2004; @van_der_Sluijs_2005]. Since no silver bullet seems to be available for mitigating the intrinsic wide-extent of complexity and uncertainty in environmental risk modelling, an array of approaches is integrated and the computational aspects are explicitly connected with the supervision and distributed interaction of human expertise. This follows the idea that the boundary between classical control-theory management strategies for natural resources and hazards (driven by automatic control problem formulations – “minimize the risk score function”) and scenario modelling under deep-uncertainty (by e.g. merely supporting emergency-operators, decision-makers and risk-assessors with understandable information – “sorry, no such thing as a risk score function can be precisely defined”) is fuzzy. Both modelling and management aspects may be computationally intensive and their integration is a transdisciplinary problem (*integrated natural resources modelling and management*, INRMM [@INRMM]).
### 2. ENVIRONMENTAL RISK MODELLING - ARCHITECTURE {#environmental-risk-modelling---architecture .unnumbered}
Figure 1 illustrates a general modelling conceptualization [where]{} the interactions among natural hazard behaviour, related transdisciplinary impacts, risk management and control strategies are taken into account. The special focus on the many sources of uncertainty [@deRigo2013] leads to a robust semantically-enhanced modelling architecture based on the paradigm of Semantic Array Programming (SemAP) [@deRigo2012; @deRigo2012b], with an emphasis on the array of input, intermediate and output data/parameters and the array of data-transformation modules (D-TM) dealing with them.
Arrays of hazard models $h_j^{\zeta_f}(\cdot)$, dynamic information forecasts $X^{\,\zeta_X}$ (i.e. meteorology) and static parametrisation $\theta^{\,\zeta_\theta}$ (i.e. spatial distribution of land cover) are considered. Their multiplicity derives from the many sources on uncertainty $\zeta = \{\zeta_f, \zeta_X, \zeta_\theta\}$ which affect their estimation (or implementation, for the D-TM software modules $f_i^{\zeta_f}(\cdot)$ which are the building blocks of the hazard models $h_j^{\zeta_f}(\cdot)$).
Furthermore, during emergency modelling support the lack of timely and accurate monitoring systems over large spatial extents (e.g. at the continental scale) may imply a noticeable level of uncertainty to affect possibly even the location of natural hazards (*geoparsing* [@Corti_etal_2012] uncertainty). This peculiar information gap may be mitigated by integrating remote sensing (e.g. satellite imagery) with a distributed array of social contributors (Citizen Sensor [@Sheth_2009; @Zhang_etal_2011; @Adam_etal_2012]), exploiting mobile applications (Apps) and online social networks [@Fraternali_2012]. Remote sensing and the Citizen Sensor are here designed to cooperate by complementing accurate (but often less timely) geospatial information with distributed alert notifications from citizens, which might be timely but not necessarily accurate. Their safe integration implies the supervision of human expertise, even if the task may be supported by automatic tools [@Aseretto_submitted]. Assessing the evolution in the timespan $\mathcal{U}^{\,t} = [t_\text{begin},t_\text{end}]$ of a certain hazard event for the associated array of impacts $C^{k,t}$ may be also complex (e.g. [@de_Rigo_etal_IFIP2013; @Bosco_etal_2013; @DiLeo_etal_2013]). In particular, the array of impacts is often irreducible to a unidimensional quantity (e.g. monetary cost) [@Ackerman_Heinzerling_2002; @Gasparatos_2010].
{width="\textwidth"}
[**Figure 1** - Modular architecture for environmental risk modelling. Based on Urgent HPC, it follows the Semantic Array Programming paradigm (image adapted from [@deRigo2013; @GeoSemAP_2013]) integrating as inputs remote sensing, meteo data and the Citizen Sensor.]{}
The analysis of non-trivial systems subject to environmental risk and natural resources management may naturally lead to multi-objective (multi criteria) control problems, which might benefit from advanced machine learning techniques for mitigating the involved huge computational costs [@de_Rigo_2001]. Indeed, the multiplicity of modelling dimensions (states; controls; uncertainty-driven arrays of parameters and scenarios; arrays of D-TM modules to account for software uncertainty) may easily lead to an exponential increase of the required computational processes (the so called “curse of dimensionality”). A viable mitigation strategy might be offered by HPC tools (such as Urgent HPC [@RodriguezAseretto_etal_2009; @Cencerrado_etal_2009; @Yoshimoto2012]) in order to sample [high-dimensional modelling space with a proper method]{}.
[**Box 1**]{} – In a nutshell.
Context
: Demands on the EU’s resilience in preparedness and disaster response capacity are likely to increase, as the impacts of disasters continue to grow.
[$\bullet\;$]{} Classical disciplinary and domain-specific approaches which might be perfectly suitable at local-scale may result in unacceptable simplifications in a broader context.
Pitfalls
: Mathematization of systems in this contex as a formal control problem should be able to establish an *effective* science-policy interface. Academic *silo thinking* should stop advertising solution-driven oversimplification to fit control theory “hot topics”.
[$\bullet\;$]{} Although in this family of problems “humans will always be part of the computational process” (despite any academic potential illusion of fashionable *full automation*),
[$\bullet\;$]{} evolvability (for adapting models to new emerging needs and knowledge) and robustness (for supporting uncertainty-aware decision processes) would still need
[$\qquad\triangleright\;$]{} a higher level of semantic awareness in computational models and
[$\qquad\triangleright\;$]{} a self-adapting ability to scale up to the multiple dimensions of the arrays of\
data/parameters.
Multiplicity: uncertainty and complexity
: In this context, the boundary between classical control-theory management strategies for natural resources and hazards and scenario modelling under deep-uncertainty is fuzzy (INRMM).
[$\bullet\;$]{} A key aspect of soundness relies on explicitly considering the multiple dimensions of the problem and the array of uncertainties involved.
[$\bullet\;$]{} As no silver bullet seems to be available for reliably attacking this amount of uncertainty and complexity, an integration of methods is proposed.
Mitigating with an integrated approach
: Array programming is well-suited for easily managing a multiplicity of arrays of hazard models, dynamic input information, static parametrisation and the distribute array of social contributions (Citizen Sensor).
[$\bullet\;$]{} *Array-based abstract* – thus better scalable – *modularisation* of the data-transformations (D-TM), and a *semantically-enhanced* design of the D-TM structure and interactions (Semantic Array Programming) is proposed to consider also the array of uncertainties (data, modelling, geoparsing, software uncertainty) and the array of criteria to assess the potential impacts associated with the hazard scenarios.
[$\bullet\;$]{} The unevenly available information during an emergency event may be efficiently exploited by means of a POLFC schema.
[$\bullet\;$]{} Its demanding computations may become affordable during an emergency event with an appropriate array-based parallelisation strategy within Urgent-HPC.
[SemAP can simplify WSTMe modelling of nontrivial static [@de_Rigo_Bosco_2011; @de_Rigo_etal_EZ2013] and dynamic [@de_Rigo_etal_IFIP2013; @DiLeo_etal_2013; @RodriguezAseretto_etal_2013] geospatial quantities. Under the SemAP paradigm, the generic $i$-th D-TM module $Y_i = f_i(\theta_i,X_i)$]{} is subject to the semantic checks $\text{\textit{sem}}_i$ as pre-, post-conditions and invariants on [the]{} inputs [$\theta_i,X_i$]{}, outputs [$Y_i$]{} and the D-TM itself [$f_i(\cdot)$]{}. The control problem is associated with the unevenly available dynamic updates of field measurements and other data related to an on-going hazard emergency. An *Emergency Manager* may thus be interested in assessing the best control strategy $u^t(\cdot)$ given a set of impacts and their associated costs as they can be approximately estimated (rapid assessment) with the currently available data. This data-driven approach can be implemented as Partial Open Loop Feedback Control (POLFC) approach [@Castelletti2008] for minimizing the overall costs associated with the natural hazard event, from the time $t \in \mathcal{U}^{\,t}$ onwards:
$$\label{eq:multicriteria_policy}
u^t(\cdot) = {\operatorname{arg\,min}}_{u\, \in \,{ U }^{ \, u }_{ t\, ,\, t_{ { end } } }} \left[ \mathcal{C}^{\,1,\,t} \mathcal{C}^{\,2,\,t} \cdots \mathcal{C}^{\,k,\,t} \cdots \mathcal{C}^{\,n,\,t} \right]$$
where [the $k$-th]{} cost $\mathcal{C}^{\,k,\,t}$ is linked to the corresponding impact assessment criterion. [This POLFC schema within the SemAP paradigm may be considered a semantically-enhanced dynamic data-driven application system (DDDAS) [@de_Rigo_etal_IFIP2013; @DiLeo_etal_2013; @RodriguezAseretto_etal_2013].]{} Finally, the Emergency Manager may communicate the updated scenarios of the emergency evolution (by means of geospatial maps and other executive summary information) in order for decision-makers and stakeholders to be able to assess the updated multi-criteria pattern of costs and the preferred control options. This critical communication constitutes the science-policy interface and must be as supportive as possible. It is designed to exploit web map services (WMS) [@McInerney2012; @Bastin2012] (on top of the underpinning free software for WSTMe, e.g. [@RodriguezAseretto_EFDAC2013]) which may be accessed in a normal browser or with specific Apps for smart-phones [@Aseretto_submitted].
### 3. CONCLU[DING]{} REMARKS {#concluding-remarks .unnumbered}
NSF Cyberinfrastructure Council report reads: *While hardware performance has been growing exponentially - with gate density doubling every 18 months, storage capacity every 12 months, and network capability every 9 months - it has become clear that increasingly capable hardware is not the only requirement for computation-enabled discovery. Sophisticated software, visualization tools, middleware and scientific applications created and used by interdisciplinary teams are critical to turning flops, bytes and bits into scientific breakthroughs* [@NSFC2007]. Transdisciplinary environmental problems such as the ones dealing with complexity and deep-uncertainty in supporting natural-hazard emergency might appear as seemingly intractable [@Altay_Green_2006]. Nevertheless, approximate rapid-assessment based on computationally intensive modelling may offer a new perspective at least able to support emergency operations and decision-making with qualitative or semi-quantitative scenarios. Even a partial approximate but timely investigation on the potential interactions of the many sources of uncertainty might help emergency managers and decision-makers to base control strategies on the best available – although typically incomplete – sound scientific information. In this context, a key aspect of soundness relies on explicitly considering the multiple dimensions of the problem and the array of uncertainties involved. As no silver bullet seems to be available for reliably attacking this amount of uncertainty and complexity, an integration of methods is proposed, inspired by their promising synergy. Array programming is perfectly suited for easily managing a multiplicity of arrays of hazard models, dynamic input information, static parametrisation and the distribute array of social contributions (Citizen Sensor). The transdisciplinary nature of complex natural hazards – their need for an unpredictably broad and multifaceted readiness to robust scalability – may benefit (1) from a disciplined *abstract modularisation* of the data-transformations which compose the models (D-TM), and (2) from a *semantically-enhanced* design of the D-TM structure and interactions. These two aspects define the Semantic Array Programming (SemAP, [@deRigo2012; @deRigo2012b]) paradigm whose application – extended to geospatial aspects [@GeoSemAP_2013] – is proposed to consider also the array of uncertainties (data, modelling, geoparsing, software uncertainty) and the array of criteria to assess the potential impacts associated with the hazard scenarios. The unevenly available information during an emergency event may be efficiently exploited by means of a partial open loop feedback control (POLFC, [@Castelletti2008]) schema, already successfully tested in this integrated approach [@de_Rigo_etal_IFIP2013; @DiLeo_etal_2013; @RodriguezAseretto_etal_2013] as a promising evolution of adaptive data-driven strategies [@RodriguezAseretto_etal_2008]. Its demanding computations may become affordable during an emergency event with an appropriate array-based parallelisation strategy within Urgent-HPC.
[0.85]{}
[43]{}
SIPPEL, S., OTTO, F.E.L., 2014. [[Beyond climatological extremes - assessing how the odds of hydrometeorological extreme events in South-East Europe change in a warming climate](http://dx.doi.org/10.1007/s10584-014-1153-9)]{}. Climatic Change: 1-18. CIRELLA, G.T., et al, 2014. [[Natural hazard risk assessment and management methodologies review: Europe](http://dx.doi.org/10.1007/978-94-007-7161-1_16)]{}. In: Linkov, I. (Ed.), Sustainable Cities and Military Installations. NATO Science for Peace and Security Series C: Environmental Security. Springer Netherlands, pp. 329-358. CISCAR, J. C., et al, 2013. [[Climate impacts in Europe: an integrated economic assessment](http://www.climate-impacts-2013.org/files/cwi_ciscar.pdf)]{}. In: Impacts World 2013 - International Conference on Climate Change Effects. Potsdam Institute for Climate Impact Research (PIK) e. V., pp. 87-96.
CISCAR, J. C., et al, 2014. [[Climate Impacts in Europe - The JRC PESETA II project](http://dx.doi.org/10.2791/7409)]{}. Vol. 26586 of EUR - Scientific and Technical Research. Publ. Off. Eur. Union, 155 pp. DANKERS, R., FEYEN, L., 2008. [[Climate change impact on flood hazard in Europe: An assessment based on high-resolution climate simulations](http://dx.doi.org/10.1029/2007jd009719)]{}. J. Geophys. Res. 113(D19): D19105+. GAUME, E., et al, 2009. [[A compilation of data on European flash floods](http://dx.doi.org/10.1016/j.jhydrol.2008.12.028)]{}. Journal of Hydrology 367(1-2): 70-78. MARCHI, L., et al, 2010. [[Characterisation of selected extreme flash floods in Europe and implications for flood risk management](http://dx.doi.org/10.1016/j.jhydrol.2010.07.017)]{}. Journal of Hydrology 394(1-2): 118-133. FESER, F., et al, 2014. [[Storminess over the North Atlantic and Northwestern Europe - a review](http://dx.doi.org/10.1002/qj.2364)]{}. Quarterly Journal of the Royal Meteorological Society. JONGMAN, B., et al, 2014. [[Increasing stress on disaster-risk finance due to large floods](http://dx.doi.org/10.1038/nclimate2124)]{}. Nature Climate Change 4 (4): 264-268. SELF, S., 2006. [[The effects and consequences of very large explosive volcanic eruptions](http://dx.doi.org/10.1098/rsta.2006.1814)]{}. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 364(1845): 2073-2097. SWINDLES, G. T., et al, 2011. [[A 7000 yr perspective on volcanic ash clouds affecting northern Europe](http://dx.doi.org/10.1130/g32146.1)]{}. Geology 39(9): 887-890. GRAMLING, C., 2014. [[As volcano rumbles, scientists plan for aviation alerts](http://dx.doi.org/10.1126/science.345.6200.990)]{}. Science 345(6200): 990. ALLARD, G., et al, 2013. [[State of Mediterranean forests 2013](http://www.fao.org/docrep/017/i3226e/i3226e.pdf)]{}. FAO, 177 pp. SCHMUCK, G., et al, 2014. [[Forest Fires in Europe, Middle East and North Africa 2013](http://dx.doi.org/10.2788/99870)]{}. Publications Office of the European Union. 107 pp. NIJHUIS, M., 2012. [[Forest fires: Burn out](http://dx.doi.org/10.1038/489352a)]{}. Nature 489 (7416), 352-354. BOYD, I.L., et al, 2013. [[The consequence of tree pests and diseases for ecosystem services](http://dx.doi.org/10.1126/science.1235773)]{}. Science 342 (6160): 1235773+. VENETTE, R.C., et al, 2012. [[Summary of the international pest risk mapping workgroup meeting sponsored by the cooperative research program on biological resource management for sustainable agricultural systems](http://www.webcitation.org/6BOcoB2eZ)]{}. In: 6th International Pest Risk Mapping Workgroup Meeting: “Advancing risk assessment models for invasive alien species in the food chain: contending with climate change, economics and uncertainty”. Organisation for Economic Co-operation and Development (OECD), pp. 1-2.
MAES, J., et al, 2013. [[Mapping and Assessment of Ecosystems and their Services - An analytical framework for ecosystem assessments under action 5 of the EU biodiversity strategy to 2020](http://dx.doi.org/10.2779/12398)]{}. Publications office of the European Union, 57 pp. EUROPEAN COMMISSION, 2013. [[Communication from the commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions - A new EU forest strategy: for forests and the forest-based sector](http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2013:0659:FIN:EN:PDF)]{}. No. COM(2013) 659 final. Communication from the Commission to the Council and the European Parliament. EUROPEAN COMMISSION, 2013. [[Commission staff working document accompanying the document: Communication from the commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions - a new EU forest strategy: for forests and the forest-based sector](http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=SWD:2013:0342:FIN:EN:PDF)]{}. Commission Staff Working Document 2013 (SWD/2013/0342 final), 98pp. BARREDO, J. I., 2007. [[Major flood disasters in Europe: 1950-2005](http://dx.doi.org/10.1007/s11069-006-9065-2)]{}. Natural Hazards 42(1): 125-148. BARREDO, J. I., 2010. [[No upward trend in normalised windstorm losses in Europe: 1970-2008](http://dx.doi.org/10.5194/nhess-10-97-2010)]{}. Natural Hazards and Earth System Science 10(1): 97-104. EVANS, M. R., et al, 2012. [[Predictive ecology: systems approaches](http://dx.doi.org/10.1098/rstb.2011.0191)]{}. Philosophical transactions of the Royal Society of London. Series B, Biological sciences 367(1586): 163-169. PHILLIS, Y. A., KOUIKOGLOU, V. S., 2012. [[System-of-Systems hierarchy of biodiversity conservation problems](http://dx.doi.org/10.1016/j.ecolmodel.2012.03.032)]{}. Ecological Modelling 235-236: 36-48. LANGMANN, B., 2014. [[On the role of climate forcing by volcanic sulphate and volcanic ash](http://dx.doi.org/10.1155/2014/340123)]{}. Advances in Meteorology 2014: 1-17. GOTTRET, M. V., WHITE, D., 2001. [[Assessing the impact of integrated natural resource management: Challenges and experiences](http://www.ecologyandsociety.org/vol5/iss2/art17/)]{}. Ecology and Society 5 (2): 17+ HAGMANN, J., et al, 2001. [[Success factors in integrated natural resource management R&D: Lessons from practice](http://www.ecologyandsociety.org/vol5/iss2/art29/)]{}. Ecology and Society 5 (2), 29+. ZHANG, X., et al, 2004. [[Scaling issues in environmental modelling](http://www.citeulike.org/group/15400/article/11477904)]{}. In: Wainwright, J., Mulligan, M. (Eds.), Environmental modelling : finding simplicity in complexity. Wiley.
ESTREGUIL, C., et al, 2013. [[Forest landscape in Europe: Pattern, fragmentation and connectivity](http://dx.doi.org/10.2788/77842)]{}. EUR - Scientific and Technical Research 25717 (JRC 77295), 18 pp. TURNER, R., 2010, [[Disturbance and landscape dynamics in a changing world](http://dx.doi.org/10.1890/10-0097.1)]{}. Ecology, 91 10: 2833-2849. Van WESTEN, C.J., 2013. [[Remote sensing and GIS for natural hazards assessment and disaster risk management](http://dx.doi.org/10.1016/b978-0-12-374739-6.00051-8)]{}. In: Bishop, M. P. (Ed.), Remote Sensing and GIScience in Geomorphology. Vol. 3 of Treatise on Geomorphology. Elsevier, pp. 259-298. URBAN, M.C., et al, 2012. [[A crucial step toward realism: responses to climate change from an evolving metacommunity perspective](http://dx.doi.org/10.1111/j.1752-4571.2011.00208.x)]{}. Evolutionary Applications 5 (2): 154-167. BAKLANOV, A., 2007. [[Environmental risk and assessment modelling - scientific needs and expected advancements](http://dx.doi.org/10.1007/978-1-4020-5877-6_3)]{}. In: Ebel, A., Davitashvili, T. (Eds.), Air, Water and Soil Quality Modelling for Risk and Impact Assessment. NATO Security Through Science Series. Springer Netherlands, pp. 29-44. STEFFEN, W., et al, 2011. [[The anthropocene: From global change to planetary stewardship](http://dx.doi.org/10.1007/s13280-011-0185-x)]{}. AMBIO 40 (7): 739-761. WHITE, C., et al, 2012. [[The value of coordinated management of interacting ecosystem services](http://dx.doi.org/10.1111/j.1461-0248.2012.01773.x)]{}. Ecology Letters 15 (6): 509-519. de RIGO, D., 2013, [[Software Uncertainty in Integrated Environmental Modelling: the role of Semantics and Open Science](http://arxiv.org/abs/1311.4762)]{}. Geophys. Res. Abstr. 15: 13292+. de RIGO, D., (exp.) 2014. Behind the horizon of reproducible integrated environmental modelling at European scale: ethics and practice of scientific knowledge freedom. F1000 Research, submitted.
LEMPERT, R. J., May 2002. [[A new decision sciences for complex systems](http://dx.doi.org/10.1073/pnas.082081699)]{}. Proceedings of the National Academy of Sciences 99 (suppl 3): 7309-7313. RAMMEL, C., et al, 2007. [[Managing complex adaptive systems - a co-evolutionary perspective on natural resource management](http://dx.doi.org/10.1016/j.ecolecon.2006.12.014)]{}. Ecological Economics 63 (1): 9-21. van der SLUIJS 2012, J. P., 2012. [[Uncertainty and dissent in climate risk assessment: A Post-Normal perspective](http://dx.doi.org/10.3167/nc.2012.070204)]{}. Nature and Culture 7 (2): 174-195. de RIGO, D., et al, 2013. [[An architecture for adaptive robust modelling of wildfire behaviour under deep uncertainty](http://dx.doi.org/10.1007/978-3-642-41151-9_35)]{}. IFIP Adv. Inf. Commun. Technol. 413: 367-380. GUARISO, G., et al, 2000. A map-based web server for the collection and distribution of environmental data. In: Kosmatin Fras, M., Mussio, L., Crosilla, F., Podobnikar, T. (Eds.), Bridging the gap: ISPRS WG VI/3 and IV/3 Workshop, Ljubljana, February 2-5, 2000 : Collection of Abstracts. Ljubljana.
GUARISO, G., WERTHNER, H., 1989. Environmental decision support systems. E. Horwood; Halsted Press.
GUARISO, G., et al, 1985. [[Decision support systems for water management: The lake Como case study](http://dx.doi.org/10.1016/0377-2217(85)90150-X)]{}. European Journal of Operational Research 21(3): 295-306. SONCINI-SESSA, R., et al, 2007. Integrated and participatory water resources theory. Elsevier.
GUARISO, G., PAGE, B. (Eds), 1994. Computer support for environmental impact : proceedings of the IFIP TC5/WG5.11 Working Conference on Computer for Environmental Impact Assessment, CSEIA 93, Como, Italy, 6-8 October, 1993. North-Holland.
CASAGRANDI, R., GUARISO, G., 2009. [[Impact of ICT in environmental sciences: A citation analysis 1990-2007](http://dx.doi.org/10.1016/j.envsoft.2008.11.013)]{}. Environmental Modelling & Software 24 (7): 865-871. COLE, J., 2010. [[Interoperability in a crisis 2: Human factors and organisational processes](http://www.rusi.org/publications/whitehallreports/ref:O4C2CC38D725EE/)]{}. Tech. rep., Royal United Services Institute. STERMAN, J.D., 2002. [[All models are wrong: reflections on becoming a systems scientist](http://dx.doi.org/10.1002/sdr.261)]{}. System Dynamics Review 18 (4): 501-531. WEICHSELGARTNER, J., KASPERSON, R., 2010. [[Barriers in the science-policy-practice interface: Toward a knowledge-action-system in global environmental change research](http://dx.doi.org/10.1016/j.gloenvcha.2009.11.006)]{}. Global Environmental Change 20 (2): 266-277. BAINBRIDGE, L., 1983. [[Ironies of automation](http://dx.doi.org/10.1016/0005-1098(83)90046-8)]{}. Automatica 19(6): 775-779. de RIGO, D., et al, 2013. [[Toward open science at the European scale: Array Programming for integrated environmental modelling](http://dx.doi.org/10.6084/m9.figshare.155703)]{}. Geophysical 15: 13245+. STENSSON, P., JANSSON, A., 2013. [[Autonomous technology - sources of confusion: a model for explanation and prediction of conceptual shifts](http://dx.doi.org/10.1080/00140139.2013.858777)]{}. Ergonomics 57 (3): 455-470. RUSSELL, D. M., et al, 2003. [[Dealing with ghosts: Managing the user experience of autonomic computing](http://dx.doi.org/10.1147/sj.421.0177)]{}. IBM Systems Journal 42 (1): 177-188. ANDERSON, S., et al, 2003. [[Making autonomic computing systems accountable: the problem of human computer interaction](http://dx.doi.org/10.1109/dexa.2003.1232106)]{}. In: Database and Expert Systems Applications, 2003. Proceedings. 14th International Workshop on. IEEE, pp. 718-724. KEPHART, J. O., CHESS, D. M. 2003. [[The vision of autonomic computing](http://dx.doi.org/10.1109/mc.2003.1160055)]{}. Computer 36 (1): 41-50. van der SLUIJS, J.P., 2005. [[Uncertainty as a monster in the science-policy interface: four coping strategies](http://www.iwaponline.com/wst/05206/wst052060087.htm)]{}. Water Science & Technology 52 (6): 87-92.
FRAME, B., 2008. [[’Wicked’, ’messy’, and ’clumsy’: long-term frameworks for sustainability](http://dx.doi.org/10.1068/c0790s)]{}. Environment and Planning C: Government and Policy 26 (6): 1113-1128. MCGUIRE, M., SILVIA, C., 2010. [[The effect of problem severity, managerial and organizational capacity, and agency structure on intergovernmental collaboration: Evidence from local emergency management](http://dx.doi.org/10.1111/j.1540-6210.2010.02134.x)]{}. Public Administration Review 70(2): 279-288. BEA, R., et al, 2009. [[A new approach to risk: The implications of e3](http://dx.doi.org/10.1057/rm.2008.12)]{}. Risk Management 11 (1): 30-43. ADAMS, K.M., HESTER, P.T., 2012. [[Errors in systems approaches](http://dx.doi.org/10.1504/ijsse.2012.052683)]{}. International Journal of System of Systems Engineering 3 (3/4): 233+. LARSSON, A., et al, 2010. [[Decision evaluation of response strategies in emergency management using imprecise assessments](http://dx.doi.org/10.2202/1547-7355.1646)]{}. Journal of Homeland Security and Emergency Management 7 (1). INNOCENTI, D., ALBRITO, P., 2011. [[Reducing the risks posed by natural hazards and climate change: the need for a participatory dialogue between the scientific community and policy makers](http://dx.doi.org/10.1016/j.envsci.2010.12.010)]{}. Environmental Science & Policy 14 (7): 730-733. RAVETZ, J., 2004. [[The post-normal science of precaution](http://dx.doi.org/10.1016/S0016-3287(03)00160-5)]{}. Futures 36 (3): 347-357. de RIGO, D., 2012. Integrated Natural Resources Modelling and Management: minimal redefinition of a known challenge for environmental modelling. Excerpt from the Call for a shared research agenda toward scientific knowledge freedom, Maieutike Research Initiative
de RIGO, D., 2012, [[Semantic Array Programming for Environmental Modelling: of the Mastrave Library](http://www.iemss.org/iemss2012/proceedings/D3_1_0715_deRigo.pdf)]{}. Int. Congress on Environmental Modelling and . Managing Resources of a Limited Plant, 1167-1176. de RIGO, D., 2012. [[Semantic Array Programming with Mastrave - Introduction to Semantic Computational Modelling](http://mastrave.org/doc/MTV-1.012-1/)]{}. CORTI, P., et al, 2012. [[Fire news management in the context of the european forest fire information system (EFFIS)](http://dx.doi.org/10.6084/m9.figshare.101918)]{}. In: proceedings of “Quinta conferenza italiana sul software geografico e sui dati geografici liberi” (GFOSS DAY 2012). SHETH, A., 2009. [[Citizen sensing, social signals, and enriching human experience](http://dx.doi.org/10.1109/mic.2009.77)]{}. Internet Computing, IEEE 13 (4): 87-92. ZHANG, D., et al, 2011. [[The emergence of social and community intelligence](http://dx.doi.org/10.1109/mc.2011.65)]{}. Computer 44 (7): 21-28. ADAM, N.R., et al, 2012. [[Spatial computing and social media in the context of disaster management](http://dx.doi.org/10.1109/mis.2012.113)]{}. Intelligent Systems, IEEE 27 (6): 90-96. FRATERNALI, P., et al, 2012. [[Putting humans in the loop: Social computing for water resources management](http://dx.doi.org/10.1016/j.envsoft.2012.03.002)]{}. Environmental Modelling & Software 37: 68-77. RODRIGUEZ-ASERETTO, D., et al, (exp.) 2014. Image geometry correction of daily forest fire progression map using MODIS active fire observation and Citizens sensor. IEEE Earthzine 7(2). Submitted. BOSCO, C., et al, 2013. [[Multi-Scale Robust Modelling of Landslide Susceptibility: Rapid Assessment and Catchment Robust Fuzzy Ensemble](http://dx.doi.org/10.1007/978-3-642-41151-9_31)]{}. IFIP Adv. Inf. Commun. . 413: 321-335.
DI LEO, M., et al, 2013. [[Dynamic data driven ensemble for wildfire behaviour assessment: a case study](http://dx.doi.org/10.1007/978-3-642-41151-9_2)]{}. IFIP Adv. Inf. Commun. Technol. 413: 11-22. ACKERMAN, F., HEINZERLING, L., 2002. [[Pricing the priceless: Cost-Benefit analysis of environmental protection](http://dx.doi.org/10.2307/3312947)]{}. University of Pennsylvania Law Review 150 (5): 1553-1584. GASPARATOS, A., 2010. [[Embedded value systems in sustainability assessment tools and their implications](http://dx.doi.org/10.1016/j.jenvman.2010.03.014)]{}. Journal of Environmental Management 91 (8): 1613-1622. de RIGO, D., et al, 2001. [[Neuro-dynamic programming for the efficient management of reservoir networks](http://dx.doi.org/10.5281/zenodo.7481)]{}. In: Proceedings of MODSIM 2001, International Congress on Modelling and Simulation. Vol. 4. Model. Simul. Soc. Australia and New Zealand, pp. 1949-1954. RODRIGUEZ-ASERETTO, D., et al, 2009. [[Injecting dynamic Real-Time data into a DDDAS for forest fire behavior prediction](http://dx.doi.org/10.1007/978-3-642-01973-9_55)]{}. Lecture Notes in Computer Science 5545: 489-499. CENCERRADO, A., et al, 2009. [[Support for urgent computing based on resource virtualization](http://dx.doi.org/10.1007/978-3-642-01970-8_23)]{}. Lecture Notes in Computer Science 5544: 227-236. JOSHIMOTO, K, K, et al, 2012. [[t]({}{h})tp://dx.doi.org/10.1016%2Fj.procs.2012.04.186]{}[Implementations of Urgent Computing on Production HPC Systems]{}. Procedia Computer Science 9: 1687-1693. de RIGO, D., BOSCO, C., 2011. [[Architecture of a Pan-European framework for soil water erosion assessment](http://dx.doi.org/10.1007/978-3-642-22285-6_34)]{}. IFIP Adv. Inf. Commun. . 359: 310-318. de RIGO, D., et al, 2013. [[Continental-Scale Living Forest Biomass and Carbon Stock: A Robust Fuzzy Ensemble of IPCC Tier 1 Maps for Europe](http://dx.doi.org/10.1007/978-3-642-41151-9_26)]{}. IFIP Adv. Inf. Commun. . 359: 271-284. RODRIGUEZ-ASERETTO, D., et al, 2013. [[A data-driven model for large wildfire prediction in Europe](http://dx.doi.org/10.1016/j.procs.2013.05.355)]{}. Procedia Computer Science 18: 1861-1870. CASTELLETTI, A., et al, [[On-Line design of water reservoir policies based on ](http://www.nt.ntnu.no/users/skoge/prost/proceedings/ifac2008/data/papers/1685.pdf)]{}. IFAC-PapersOnLine 17: 14540-14545.
MCINERNEY, D., et al, 2012. [[Developing a forest data portal to support Multi-Scale decision making](http://dx.doi.org/10.1109/JSTARS.2012.2194136)]{}. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 5(6): 1692-1699. BASTIN, et al, 2012. [[Web services for forest data, analysis and monitoring: Developments from EuroGEOSS](http://www.earthzine.org/?p=389531)]{}. IEEE Earthzine 5(2): 389531+. RODRIGUEZ-ASERETTO, D., et al, 2013. [[Free and open source software underpinning the European Forest Data Centre](http://dx.doi.org/10.6084/m9.figshare.155700)]{}. Geophysical Research Abstracts 15: 12101+. NATIONAL SCIENCE FOUNDATION, CYBERINFRASTRUCTURE COUNCIL, 2007. [[Cyberinfrastructure vision for 21st century discovery](http://www.nsf.gov/pubs/2007/nsf0728/index.jsp?org=EF)]{}. Tech. Rep. NSF 07-28, National Science Foundation. ALTAY, N., GREEN, W.G., 2006. [[OR/MS research in disaster operations management](http://dx.doi.org/10.1016/j.ejor.2005.05.016)]{}. European Journal of Operational Research 175 (1): 475-493. RODRIGUEZ-ASERETTO, D., et al, 2008. [[An adaptive system for forest fire behavior prediction](http://dx.doi.org/10.1109/CSE.2008.15)]{}. In: Computational Science and Engineering, 2008. CSE ’08. 11th IEEE International Conference on. IEEE, pp. 275-282.
[^1]: E-mail Corresponding Author: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a canonical tranformation approach to the effective interaction $W_{eff}$ between two holes, based on the three-band Hubbard model but ready to include extra interactions as well. An effective two-body Hamiltonian can in principle be obtained including any kind of virtual intermediate states. We derive the closed-form analytic expression of the effective interaction including 4-body virtual states, describing the exchange of an electron-hole pair to all orders. The resulting integral equation, valid for the full plane, leads to a Cooper-like instability of the Fermi liquid. The two-hole bound states generalize those reported earlier in cluster calculations by exact diagonalisation methods.'
address: |
INFM, Dipartimento di Fisica,Università di Roma Tor Vergata, Via\
della Ricerca Scientifica 1- 00133 Roma, Italy
author:
- 'Michele Cini, Gianluca Stefanucci and Adalberto Balzarotti'
title: 'Canonical Transformation of the Three-Band Hubbard Model and Hole Pairing '
---
=10000
The three-band Hubbard model Hamiltonian has often been used to characterize the electronic properties of high-$T_{C}$ superconductors[@kn:bm] as observed in electron spectroscopy experiments. The model is: $$H=H_{0}+W \label{h0}$$ where the independent hole hamiltonian reads, in the site representation
$$H_{0}={\sum_{Cu}}\varepsilon _{d}n_{d}+{\sum_{O}}\varepsilon _{p}n_{p}+{\
t\sum_{n.n.}}\left[ c_{p}^{+}c_{d}+h.c.\right] \label{1}$$
where n.n. stands for nearest neighbors. The on-site repulsion Hamiltonian will be denoted by $$W={\sum_{i}}U_{i}n_{i+}n_{i-}, \label{w}$$
where $U_{i}=U_{d}$ for a Cu site, $U_{i}=U_{p}$ for an Oxygen. We also considered first-neighbor O-O hopping and off-site interactions[@kn:cb1], but they were dropped when it was clear that they were not really important. The point symmetry Group of the Cu-O plane is $C_{4v}$, and its characters are listed in Table I. We wish to show that holes are paired in a Cooper-like fashion in the ground state of this popular model. Preliminarly, however, we must introduce the W=0 pairs. Consider a determinantal 2-hole eigenstate of $H_{0}$ built from degenerate Bloch states. Can we superpose such wave functions to obtain simultaneous eigenstates of $H$,$H_{0}$ and $W$? For a general system, of course, we can’t, but the symmetry of the problem enables us to achieve the result, and in addition to choose the eigenvalue W=0 of the on-site interaction. Indeed, omitting the band indices,let $$d[k]=\left\| k_{+},-k_{-}\right\| =c_{k,+}^{\dagger}c_{-k,-}^{\dagger}|vac>
\label{3}$$ be a two-hole determinantal state derived from the Bloch eigenfunctions of $H_{0}$. We introduce the determinants $Rd[k]=d[Rk]=d[k_{R}],R\in
C_{4v}$ , and the projected states $$\Phi _{\eta }\left[ k\right] =\frac{1}{\sqrt{8}}{\sum_{R\in C_{4v}}}\chi
^{\left( \eta \right) }\left( R\right) Rd[k] \label{4}$$ where $\chi ^{\left( \eta \right) }(R)$ is the character of the operation $R$ in the Irreducible Representation (Irrep) $\eta $. In the non-degenerate Irreps, the operations that produce opposite $k_{R}$ have the same character, and the corresponding projections lead to singlets. The explicit W=0 singlet pair states are: $$\begin{aligned}
\Phi _{^{1}B_{2}}\left[ k,r_{1},r_{2}\right] =\frac{\chi _{0}}{\sqrt{2}}
\{
\cos\left[ k(r_{1}-r_{2})\right] \phi \left( k,r_{1}\right) \phi \left(
k,r_{2}\right) \nonumber \\
-\cos\left[ k_{C_{4}}(r_{1}-r_{2})\right] \phi \left( k_{C_{4}},r_{1}\right) \phi
\left( k_{C_{4}},r_{2}\right)\nonumber \\
-\cos\left[ k_{\sigma}(r_{1}-r_{2})\right] \phi
\left( k_{\sigma },r_{1}\right) \phi \left( k_{\sigma },r_{2}\right)
\nonumber \\
+\cos\left[ k_{\sigma^{\prime} }(r_{1}-r_{2})\right] \phi
\left( k_{\sigma^{\prime} },r_{1}\right) \phi \left( k_{\sigma^{\prime}
},r_{2}\right)
\}\label{b2e}\end{aligned}$$
and $$\begin{aligned}
\Phi _{^{1}A_{2}}\left[ k,r_{1},r_{2}\right] =\frac{\chi _{0}}{\sqrt{2}}
\{
\cos\left[ k(r_{1}-r_{2})\right] \phi \left( k,r_{1}\right) \phi \left(
k,r_{2}\right) \nonumber \\
+\cos\left[ k_{C_{4}}(r_{1}-r_{2})\right] \phi \left( k_{C_{4}},r_{1}\right) \phi
\left( k_{C_{4}},r_{2}\right)\nonumber \\
-\cos\left[ k_{\sigma}(r_{1}-r_{2})\right] \phi
\left( k_{\sigma },r_{1}\right) \phi \left( k_{\sigma },r_{2}\right)
\nonumber \\
-\cos\left[ k_{\sigma^{\prime} }(r_{1}-r_{2})\right] \phi
\left( k_{\sigma^{\prime} },r_{1}\right) \phi \left( k_{\sigma^{\prime}
},r_{2}\right)
\},\label{a2e}\end{aligned}$$ where $\chi_{0}$ is a singlet spin function and $\phi$ is the periodic part of the Bloch function. Using the explicit Bloch states we can verify that both vanish for $r_{1}=r_{2}$ . Therefore they have the W=0 property. Turning now to the many-body problem, suppose the Cu-O plane is in its ground state with chemical potential $\mu\equiv E_{F}$ and a couple of extra holes are added. We can consider the W=0 pairs as bare quasiparticles for which no repulsion barrier needs to be overcome for pairing. Do the dressed quasiparticles look like Cooper pairs?
In Cooper theory[@kn:coop], an effective interaction involving phonons is introduced via an approximate canonical transformation. Our problem is similar, except that the holes can exchange electron-hole pairs instead of phonons. Now the vacuum is the filled Fermi [*sphere*]{} and W connects the determinants $d[k]$ to the 4-body (3 hole-1 electron) determinants that carry no quasi-momentum; they are of the form
$$|\alpha >=|\left\| \left( k^{\prime }+q+k_{2}\right) _{+},\bar{k}%
_{2-},-q_{-},-k_{-}^{\prime }\right\| > \label{alfas}$$
where $\bar{k}_{2}$ is the electron state and pedices refer to the spin direction; those with opposite spin indices contribute similarly and yield a factor of 2 at the end. These are eigenstates of $H_{0}$: $$H_{0}\left| \alpha \right\rangle =E_{\alpha }\left| \alpha \right\rangle$$
The interaction matrix element reads:
$$\begin{aligned}
<\left\| \left( k^{\prime }+q+k_{2}\right) _{+},\bar{k}
_{2-},-q_{-},-k_{-}^{\prime }\right\| |W|d[s]>= \nonumber \\
\delta \left( q-s\right) U\left( q+k^{\prime }+k_{2},-k^{\prime
},s,k_{2}\right) \nonumber \\
-\delta \left( k^{\prime }-s\right) U\left( q+k^{\prime
}+k_{2},-q,s,k_{2}\right) \label{me}\end{aligned}$$
where the $U$ matrix is computed on the Bloch basis. If we want to keep a close analogy with the Cooper theory, we temporarily consider $W$ as a small perturbation and look for an approximate canonical transformation such that the new Hamiltonian
$$\tilde{H}=e^{-\Lambda}He^{\Lambda} \label{cano}$$
contains no first-order terms. Here, $ \tilde{H}$ operates on the space of pairs, since the 3 holes-1 electron intermediate states have been decoupled by the canonical transformation. This is accomplished if
$$W+\left[H_{0}, \Lambda\right]=0 \label{hs2}$$
because then, to second order, $$\tilde{H}=H_{0}+\frac{1}{2}\left[W,\Lambda\right].\label{hs3}$$
Assuming that the denominators do not vanish (more about that later) we obtain
$$\left\langle \alpha \right| \Lambda\left| s\right\rangle =\frac{\left\langle
\alpha \right| W\left| s\right\rangle }{E_{s}-E_{\alpha }}$$
We may write $$\tilde{H}=H_{0}+F+\tilde{W}_{eff}\label{hs4}$$ where $F$ is diagonal in the pair space, like $H_{0}$, and corresponds to the unlinked self-energy diagrams, while the effective interaction operator is $ \tilde{W}_{eff} $. Like in Cooper theory, F will be dropped. We obtain
$$\begin{aligned}
2\left\langle p \right| F+\tilde{W}_{eff}\left| s\right\rangle
\nonumber\\
=\sum_{\alpha}W_{p,\alpha}W_{\alpha,s}\left[
\frac{1 }{E_{p}-E_{\alpha }}+\frac{1 }{E_{s}-E_{\alpha }}
\right]\label{u}\end{aligned}$$
Using the interaction matrix element (\[me\]), the product in (\[u\]) yields 4 terms; two are proportional to $\delta\left(p-s\right)$ and belong to $F$, while the cross terms yield identical contributions to the effective interaction.
After long algebra, we write the interaction between symmetry projected states (with $s$ and $p$ empty):
$$\left\langle \Phi_{\alpha}\left[ p\right] \right| \tilde{W}_{eff}\left|
\Phi_{\alpha}\left[ s\right] \right\rangle =
\sum_{R}\chi^{\left(\alpha\right)}\left(R\right)
\left\langle d\left[ p\right] \right| \tilde{W}_{eff}\left|
Rd\left[ s\right] \right\rangle
\label{wfi}$$
where, explicitly, $$\begin{aligned}
\left\langle d\left[ p\right] \right| \tilde{W}_{eff}\left| d\left[
s\right] \right\rangle =2
\sum_{k}^{occ}\Theta \left( \varepsilon \left( s+p+k\right) -E_{F}\right)
\nonumber\\
\times U\left( s+p+k,-p,s,k\right) U\left( p,k,s+p+k,-s\right)
\nonumber\\
\times [\frac{1}{\varepsilon \left( s+p+k\right) -\epsilon \left( k\right)
-\epsilon \left( s\right) +\epsilon \left( p\right) }
\nonumber\\
+\frac{1}{\varepsilon \left( s+p+k\right) -\epsilon \left( k\right)
-\epsilon \left( p\right) +\epsilon \left( s\right) }] \label{wd}\end{aligned}$$
In the continuum limit, the integral must be understood as a principal part, because the denominators vanish in a domain of zero measure within the integration domain. However, this complication is an artifact of the perturbation approach, as will soon become clear. The approximate canonical transformation is open to question because it assumes that $W$ be weak. Therefore, it is important to put the theory on a clearer and firmer basis, by removing the coupling to the $\alpha$ states to all orders.
The transformation (\[cano\],\[hs2\]) corresponds to a second-order diagram for the two-hole amplitude, and if one of the hole lines is closed on itself one gets the corresponding self-energy. The expansion of $\Lambda$ can be continued systematically to produce the perturbation series. Including all the diagrams of a generalized RPA would lead to something like the well-known FLEX approximation [@bsw] whose implications for the superconductivity in the three-band Hubbard model have been explored recently in a series of papers[@eb]. A related self-consistent and conserving T-matrix approximation has been proposed by Dahm and Tewordt[@dt] for the excitation spectra in the 2D Hubbard model; we mention incidentally that recently diagrammatic methods have been successfully applied to the photoelectron spectra of the Cuprates in other contexts too, like the spin-fermion model[@sps].
Realizing the key rôle of symmetry and W=0 pairs in this problem enhances intuition besides simplifying the perturbation formalism considerably. Here, we wish to take advantage of the W=0 pairs to propose a non-perturbative approach to pairing based on a different way of performing the canonical transformation. We write the ground state wave function with two added holes as a superposition of two-body states (roman indices) and 4-body ones:
$$|\Psi _{0}>={\sum_{m}}a_{m}|m>+{\sum_{\alpha }}b_{\alpha }|\alpha
>. \label{psi00}$$
Schrödinger’s equation then yields $$\left( E_{m}-E_{0}\right) a_{m}+{\sum_{m^{\prime }}}a_{m^{\prime
}}V_{m,m^{\prime }}+{\sum_{\alpha }}b_{\alpha }W_{m,\alpha }=0 \label{sys1}$$
$$\left( E_{\alpha }-E_{0}\right) b_{\alpha }+{\sum_{m^{\prime }}}a_{m^{\prime
}}W_{\alpha ,m^{\prime }}=0 \label{sys2}$$
where $E_{0}$ is the ground state energy. $V_{m^{\prime },m}$ vanishes for W=0 pairs in our model; however we keep it for generality, since it allows to introduce the effect of phonons[@kn:pietro; @kn:iad] and any other indirect interaction[@kn:bob; @kn:pw; @kn:sus] that we are not considering. Solving for $b_{\alpha }$ and substituting in (\[sys1\]) we exactly decouple the 4-body states. The eigenvalue problem is now $$\left( E_{0}-E_{m}\right) a_{m}=\sum_{m^{\prime}} a_{m^{\prime
}}\left\{V_{m,m^{\prime }}+\left\langle m|S[E_{0}]|m^{\prime }\right\rangle \right\}
, \label{dec}$$ where $$\left\langle m|S\left[ E_{0}\right] |m^{\prime }\right\rangle ={\sum_{\alpha
}}\frac{<m|W|\alpha ><\alpha |W|m^{\prime }>}{E_{0}-E_{\alpha }}. \label{wt1}$$ This is of the form of a Schrödinger equation with eigenvalue $E_{0}$ for pairs with an effective interaction $V+S$. Then we interpret $a_{m}$ as the wave function of the dressed pair, which is acted upon by an effective hamiltonian $\tilde{H}$. This is the canonical transformation we were looking for. However, the scattering operator $S$ is of the form $S=W_{eff}+F,$ where $W_{eff}$ is the effective interaction between dressed holes, while $F$ is a forward scattering operator, diagonal in the pair indices $m$ ,$m^{\prime }$ which accounts for the self-energy corrections of the one-body propagators: it is evident that it just redefines the dispersion law $E_{m}$, and, essentially, renormalizes the chemical potential. Therefore $F$ must be dropped, as in Cooper theory[@kn:coop] and above. Therefore the effective Schrödinger equation for the pair reads
$$\left( H_{0}+V+W_{eff}\right) |a>=E_{0}|a> \label{cang}$$
and we are interested in the possibility that $E_{0}=2E_{F}-\left| \Delta \right|
$, with a positive binding energy $\left| \Delta \right| $ of the pair. The $V$ interaction just adds to $W_{eff}$, and this feature allows to include in our model the effects of other pairing mechanisms, like off-site interactions, inter-planar coupling and phonons. Again, the product in the numerator of (\[wt1\]) yields 4 terms; two are proportional to $\delta (p-s)$ and belong to F, while the cross terms yield identical contributions to $W_{eff}$. Hence we obtain the effective interaction between W=0 pairs:
$$\begin{aligned}
\left\langle \Phi _{\eta }\left[ p\right] \right| W_{eff}\left| \Phi _{\eta
}\left[ s\right] \right\rangle =4{\sum_{R\in C_{4v}}}\chi
^{\left( \eta \right) }\left( R\right) {\sum_{k}^{occ}}
\Theta \left( \varepsilon \left( Rs+p+k\right) -E_{F}\right)
\nonumber\\
\times \frac{U\left( Rs+p+k,-p,Rs,k\right)U\left( p,k,Rs+p+k,-Rs\right)}
{\varepsilon \left( Rs+p+k\right) -\epsilon \left( k\right)
+\epsilon \left( s\right) +\epsilon \left( p\right)-E_{0} }
\label{weffective}\end{aligned}$$
The sum is over occupied $k$ with empty $Rs+p+k$. There are no vanishing denominators in the new expression, if $E_{0}$ is off the continuum. The $p$ and $s$ indices run over 1/8 of the Brillouin Zone. We denote such a set of empty states $e/8$, and cast the result in the form of a (Cooper-like) Schrödinger equation
$$2\varepsilon \left( k\right) a\left( k\right) +\stackrel{e/8}{
\sum_{k^{\prime }}}W_{eff}\left( k,k^{\prime }\right) a\left( k^{\prime
}\right) =E_{0}a\left( k\right) \label{inteq}$$
for a self-consistent calculation of $E_{0}$ (since $W_{eff}$ depends on the solution). Let $N_{C}$ be the number of cells in the crystal. The $U$ matrix elements scale as $N_{C}^{-1}$ and therefore $W_{eff}$ scales in the same way. For an infinite system, $N_{C}\rightarrow \infty $ , this is a well defined integral equation. In principle, we can do better. By a canonical transformation one can obtain an effective Hamiltonian which describes the propagation of a pair of holes, and includes all many-body effects. The exact many-body ground state with two added holes may be expanded in terms of excitations over the vacuum (the non-interacting Fermi [*sphere*]{}) by a configuration interaction: $$|\Psi _{0}>={\sum_{m}}a_{m}|m>+{\sum_{\alpha }}b_{\alpha }|\alpha >+{\
\sum_{\beta }}c_{\beta }|\beta >+.... \label{psi0}$$ here m runs over pair states, $\alpha $ over 4-body states ($2$ holes and $1$ e-h pair), $\beta $ over 6-body ones ($2$ holes and $2$ e-h pairs), and so on. To set up the Schrödinger equation, we consider the effects of the operators on the terms of $|\Psi _{0}>$. We write:
$$H_{0}|m>=E_{m}|m>,\;H_{0}|\alpha >=E_{\alpha }|\alpha >,... \label{h0m}$$
and since W can create or destroy up to 2 e-h pairs,
$$\begin{aligned}
W|m>={\sum_{m^{\prime }}}V_{m^{\prime },m}|m^{\prime }>+{\sum_{\alpha }}%
|\alpha >W_{\alpha ,m} \nonumber \\
+{\ \sum_{\beta }}|\beta >W_{\beta ,m}, \label{wm}\end{aligned}$$
and the like. The Schrödinger equation now gives an infinite system for the amplitudes of 2n-body states; however we can show [@kn:sub] that these amplitudes can be successively decoupled, producing a renormalization of W matrix elements and energy eigenvalues $E_{\alpha},E_{\beta}$ and so on. In principle, the method applies to all the higher order interactions, and we can recast our problem as if only 2 and 4-body states existed. If one calculates $W_{eff}$ neglecting 6-body and higher excitations, at least the structure of the solution is exact when expressed in terms of renormalized matrix elements.
The Hubbard model, with V=0, leads to pairing, and the mechanism was first discovered in cluster studies[@kn:cb1]. Now we can better understand those results, since the above theory applies not only to the plane but also to clusters, provided that they are fully symmetric and allow W=0 pairs. The symmetry of the cluster is essential, because only fully symmetric clusters allow such solutions. The planar lattice structure is also essential, because no W=0 pairs occur in 3D or in a continuous model. We studied[@kn:cb1; @kn:cb2; @kn:cb3; @kn:cb4] the fully symmetric clusters with up to 21 atoms by exact diagonalisation. The main difference is that in the clusters the symmetries of W=0 pairs were found[@kn:cb5; @kn:cbs] to be $^{1}B_{2}$ and $^{1}A_{1}$. The reason for having $^{1}A_{1}$ instead of $^{1}A_{2}$ is a twofold size effect. On one hand,$^{1}A_{1}$ pairs have the W=0 property only in the small clusters, having the topology of a cross, and belonging to the $S_{4}$ Group, but do not generalize as such to the full plane, when the symmetry is lowered to $C_{4v}$; on the other hand, those small clusters admit no solutions of $^{1}A_{2}$ symmetry. To monitor pairing, initially we used[@kn:cb1] a definition of $\Delta$ taken from earlier cluster calculations[@kn:hirsch; @kn:bal] (where, however, only unsymmetric clusters were considered, and this mechanism could not operate):
$$\Delta=E(N+2)+E(N)-2E(N+1), \label{delta}$$
where $E(N)$ is the ground state energy of the cluster with N holes, as obtained by exact diagonalisation. Pairing, that is, $\Delta<0$, was found when (and only when) the least bound holes formed a W=0 pair. In small clusters this conditions the occupation number, because the holes must partly occupy degenerate (x,y) orbitals, while in the full plane the W=0 pairs (\[b2e\],\[a2e\]) exist at the Fermi level for any filling. The second-order approximation $\Delta^{(2)}$ was obtained by ground state energy diagrams (modified for degenerate ground states, when appropriate); the resulting expression[@kn:cb5] is clearly just the second-order approximation to the diagonal terms of (\[weffective\]). Then, we computed to second-order the two-hole amplitude for holes of opposite spins in the degenerate (x,y) orbitals. We demonstrated that this produced an effective interaction, which pushes down the singlet and up the triplet by $\left| \Delta ^{\left( 2\right) }\right| $. Good agreement between the second-order calculation and the numerical exact diagonalisation results supported the interpretation. Thus, we have shown that in the symmetric Cu-O clusters genuine pairing takes place, due to an effective interaction which is attractive for singlets and repulsive for triplets. The cluster calculations[@kn:cb2; @kn:cb3; @kn:cb4] showed that W=0 pairs are the $^{\prime\prime }bare^{\prime \prime }$ quasiparticles that, when $^{\prime\prime }dressed^{\prime \prime }$, become a bound state. That approach is inherently limited by the small size of solvable clusters, but allows a very explicit display of paired hole properties, that even show superconducting flux-quantization[@kn:cb5; @kn:cbs].
The equations (\[weffective\],\[inteq\]) allow to extend the study of pairs to the full plane. The integrand is very discontinuous because of Umklapp terms and of the limitations to occupied or empty states; moreover, its angular dependences are involved. The numerical solution is far from trivial, and the method that we developed will be presented elsewhere, with the numerical results. The main finding is the instability of the Fermi liquid in the model at hand. For both $^{1}B_{2}$ and $^{1}A_{2}$ we find pairing with a doping-dependent $\Delta$. The three-band Hubbard model might be too idealized to allow a detailed comparison with experiments; however we stress that contributions from phonons and other mechanisms can be included by a non zero $V$. The approach presented here is more general than the model we are using, and can be applied to realistic Hamiltonians.
This work has been supported by the Istituto Nazionale di Fisica della Materia. We gratefully acknowledge A. Sagnotti, Università di Roma Tor Vergata, for useful and stimulating discussions.
J.G. Bednorz and K.A. Muller, Z. Phys. B[**64**]{} (1986) 189. M. Cini and A. Balzarotti, Il Nuovo Cimento D [**18**]{}, 89 (1996). L.Cooper, Phys. Rev. [**104**]{},1189 (1956) . N.E.Bickers, D. J. Scalapino and S.R. White, Phys. Rev. Letters [**62**]{}, 961 (1989); N.E.Bickers and S.R. White, Phys. Rev. B[**43**]{}, 8044 (1991). G. Esirgen and N. E. Bickers, Phys. Rev. B[bf 57]{}, 5376 (1998) and references therein. T. Dahm and L. Tewordt, Phys. Rev. Letters [**74**]{}, 793 (1995). J. Schmalian, D. Pines and B. Stojković, Phys. Rev. Letters [**80**]{}, 3839 (1998). L. Pietronero and S. Strässer, Europhys. Lett[**18**]{}, 627 (1992).
G. Iadonisi,M. Chiofalo, V. Cataudella and D. Ninno, Phys. Rev. B[**48**]{}, (1993) 12966 and references therein.
J.R.Schrieffer, X.C.Wen and S.C.Zhang, Phys. Rev. Letters,[** 60**]{}, 944 (1988).
P.W. Anderson, Science,235,(1987),1196; see also [*The Hubbard Model*]{}, Ed. D. Baeriswyl et al., Plenum Press, New York, 1995, page 217.
O.P. Sushkov, Phys. Rev. B[**54**]{}, 9988 (1996) and reference therein. M. Cini, G. Stefanucci and A. Balzarotti, submitted for publication
M. Cini and A. Balzarotti, J. Phys. Condens. Matter[**8**]{} , L265 (1996).
M. Cini and A. Balzarotti, Solid State Commun.[**101**]{}, 671 (1997).
M. Cini, A. Balzarotti, J. Tinka Gammel and A. R. Bishop, Nuovo Cimento [**19 D**]{}, (1997) 1329.
M. Cini and A. Balzarotti, Phys. Rev. B[**56**]{}, 1, 14711 (1997). M. Cini, A. Balzarotti and G. Stefanucci, submitted for publication. J.E.Hirsch,S.Tang, E.Loh Jr, and D.J.Scalapino, Phys. Rev.Letters[** 60**]{}, 1668 (1988); Phys. Rev. B[**39**]{}, 243 (1989);
C.A.Balseiro, A.G.Rojo, E.R.Gagliano, and B.Alascio, Phys.Rev.B [**38**]{}, 9315 (1988).
---------- --- --------- ---------- ------------ ---------------------- ---------------------- -- -- -- -- -- --
C$_{4v}$ E C$_{2}$ 2C$_{4}$ 2$\sigma $ 2$\sigma ^{\prime }$
A$_{1}$ 1 1 1 1 1
A$_{2}$ 1 1 1 -1 -1 $R_{z}$
B$_{1}$ 1 1 -1 1 -1 $x^{2}-y^{2}$
B$_{2}$ 1 1 -1 -1 1 $xy$
E 2 -2 0 0 0 $\left( x,y\right) $
---------- --- --------- ---------- ------------ ---------------------- ---------------------- -- -- -- -- -- --
: The Character Table of the $C_{4v}$ Group[]{data-label="Table I"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We characterize sandwiched singularities in terms of their link in two different settings. We first prove that such singularities are precisely the normal surface singularities having self-similar non-archimedean links. We describe this self-similarity both in terms of Berkovich analytic geometry and of the combinatorics of weighted dual graphs. We then show that a complex surface singularity is sandwiched if and only if its complex link can be embedded in a Kato surface in such a way that its complement remains connected.'
address:
- 'Institut Mathématique de Jussieu, Université Pierre et Marie Curie, 75252 Paris Cedex, France.'
- 'CNRS - Centre de Mathématiques Laurent Schwartz, École polytechnique, 91128 Palaiseau Cedex, France.'
- 'Institut Mathématique de Jussieu, Université Paris 7, Bâtiment Sophie Germain, Case 7012, 75205 Paris Cedex 13, France.'
author:
- Lorenzo Fantini
- Charles Favre
- Matteo Ruggiero
bibliography:
- 'biblio.bib'
title: 'Links of sandwiched surface singularities and self-similarity'
---
Introduction
============
Let $X$ be any complex algebraic variety of dimension $d$ and let $0\in X$ be an isolated singular point. A classical way to analyze the geometry of $X$ near its singular point is to consider its (archimedean) *link*, which is defined by embedding the complex analytic germ $(X,0)$ in the germ of a complex affine space $(\C^n,0)$ and taking the intersection with the boundary of a small ball around the origin. More precisely, if $z_1,...,z_n$ are coordinates for $\C^n$ at zero, the intersection of $X$ with any sphere centered at $0$ of small enough radius $\varepsilon>0$ is transversal, so that $
\text{L}_\C^\varepsilon(X,0) = \big\{ x\in X(\C) \text{ s.t. } \sum_{i=1}^n|z_i(x)|_\C^2 = \varepsilon \big\}
$ is a smooth manifold of real dimension $2d-1$. Its diffeomorphism type does not depend on the embedding nor on $\varepsilon$, provided that $\varepsilon$ is small enough, and we define the link of $(X,0)$ to be this diffeomorphism type. Note that the topology of a neighborhood of $0$ in $X$ is completely determined by its link, since one can show that the intersection of $X$ with a small ball is homeomorphic to the cone over $\text{L}_\C^\varepsilon(X,0)$. The complex structure on $X$ also induces a canonical contact structure on the link which has attracted a lot of attention recently, see for example [@CaubelNemethiPopescu-Pampu2006; @McLean2015].
When the algebraic variety $X$ is defined over an algebraically closed field $k$, then a non-archimedean version of the link can be defined as follows. Endow $k$ with the trivial absolute value $|\cdot|$, that is the one such that $|k^\times|=1$, and denote by $X^\mathrm{an}$ the non-archimedean analytic space associated with $X$, in the sense of Berkovich [@berkovich:book]. Then, the space $\operatorname{NL}^\varepsilon(X,0) = \big\{ x\in X^\mathrm{an} \text{ such that} \max_{i}|z_i(x)| = \varepsilon \big\}$, with the topology induced from the one of $X^\mathrm{an}$, does not depend on the embedding nor on $\varepsilon\in\left]0,1\right[$. We will call it the *non-archimedean link* of $(X,0)$ and we will simply denote it by $\operatorname{NL}(X,0)$. Observe that, thanks to the non-archimedean triangular inequality, the equation $\max_{i}|z_i(x)| = \varepsilon$ defines the boundary of the ball of radius $\varepsilon$ in the non-archimedean analytification of $\C^n$, making the definition of the non-archimedean link completely analogous to the classical one. Concretely, $\operatorname{NL}(X,0)$ is the set of semi-valuations $v$ on the complete local ring $\widehat{\mathcal O_{X,0}}$ of $X$ at $0$ that are normalized by the condition $\min_{f\in{\mathfrak M}}v(f)=1$, where $\mathfrak M$ is the maximal ideal of $\widehat{\mathcal O_{X,0}}$, endowed with the pointwise convergence topology.
The homotopy type of the non-archimedean link $\operatorname{NL}(X,0)$ is well understood in terms of the resolutions of singularities of $(X,0)$, whenever those exist. Recall that any resolution of singularities $\pi\colon X_\pi\to X$ of $(X,0)$ whose exceptional divisor $\pi^{-1}(0)$ has simple normal crossing singularities gives rise to a *dual simplicial complex* $\Delta_\pi$, which is a finite simplicial complex encoding the incidence relations between the components of $\pi^{-1}(0)$. It follows from the work of Thuillier [@thuillier:geometrietoroidale] that $\Delta_\pi$ can be embedded in $\operatorname{NL}(X,0)$ and that there is a deformation retraction of the latter onto the former. Since every connected finite simplicial complex is the dual complex of an isolated normal singularity by Kollár [@kollar:links], the homotopy type of $\operatorname{NL}(X,0)$ can be arbitrarily complicated. However, de Fernex–Kollár–Xu [@defernex-kollar-xu:dualcomplex] have proved that $\Delta_\pi$ is contractible for isolated log terminal singularities.
On the other hand, the topology of non-archimedean links is poorly understood and has been analyzed in depth only in the case of surfaces. One can show that $\operatorname{NL}(\A^2_k,0)$ is a compact real tree, that is a union of segments which does not contain any non-trivial loop, see [@berkovich:book; @jonsson:berkovich; @favre-jonsson:valtree]. Its structure is however quite intricate since it has a dense set of ramification points (corresponding to points of *type 2* as in [@berkovich:book]), and the set of branches at such a point is naturally parameterized by $\P^1(k)$, which may be uncountable. When $k$ is a countable field, $\operatorname{NL}(\A^2_k,0)$ is metrizable and homeomorphic to the Ważewski universal dentrite by [@HrushovskiLoeserPoonen2014].
The non-archimedean link of a surface singularity $(X,0)$ can be obtained by gluing copies of $\operatorname{NL}(\A^2_k,0)$ to a finite graph. This picture has enabled de Felipe [@deFelipe2017] to completely describe the homeomorphism types of non-archimedean links of surface singularities, but her result shows that the topology of $\operatorname{NL}(X,0)$ fails to encode much information about the singularity. For example, $\operatorname{NL}(X,0)$ is homeomorphic to $\operatorname{NL}\big(\A^2_k,0\big)$ when $(X,0)$ is a rational singularity, or when $X$ admits a good resolution whose exceptional locus is irreducible. The topology of $\operatorname{NL}(X,0)$ thus forgets the function fields of exceptional components of a resolution. In order to characterize interesting classes of singularities one needs to retain some of this information.
To do so, we consider the sheaf on $\operatorname{NL}(X,0)$ which is induced by the sheaf of analytic functions on $X^\mathrm{an}$. The resulting ringed space was studied by the first author in [@fantini:normspaces]. Note that we cannot expect $\operatorname{NL}(X,0)$ with this additional analytic structure to be isomorphic to a proper subspace of itself, as any such isomorphism would have to send the endpoints of $\operatorname{NL}(X,0)$ to endpoints, forcing it to be surjective. In particular, already in the case of a smooth point in a surface the non-archimedean link is isomorphic to a proper subspace of itself only after removing finitely many endpoints (more precisely, finitely many points of *type 1*, that are the endpoints corresponding to semi-valuations with nontrivial kernel). Whenever such an isomorphism exists we will say that the non-archimedean link $\operatorname{NL}(X,0)$ is *self-similar* (see the condition \[condition\_prime\] on page ).
In this paper we show that the non-archimedean link $\operatorname{NL}(X,0)$ of a normal surface singularity $(X,0)$ is self-similar if and only if $(X,0)$ is a *sandwiched singularity*. Over the complex numbers, sandwiched singularities were defined by Spivakovisky in [@spivakovsky:sandsingdesingsurfNashtransf] as those normal surface singularities whose complex analytic germs dominate bimeromorphically a smooth germ; in *loc. cit.* they play a crucial role in the proof of the desingularization of surfaces via Nash transformations. Several authors have further contributed to the study of sandwiched singularities, for example their deformation theory has been investigated by T. de Jong and van Straten [@deJongvanStraten1998], while their Milnor fibers have been described by Némethi and Popescu-Pampu [@NemethiPopescu-Pampu2010]. In order to work over an algebraically closed field $k$ of arbitrary characteristic we will need to replace complex analytic germs with formal germs, that is we will work with complete local rings; a precise definition will be given in Section \[section\_preliminariessandwiched\].
Our main result is the following theorem which give several characterizations of sandwiched singularities.
\[mainthm\] Let $(X,0)$ be a normal surface singularity. The following are equivalent:
1. \[condition\_sandwiched\] $(X,0)$ is sandwiched;
2. \[condition\_strongly\_selfsim\] there exists a finite set $T$ of type 1 points of $\operatorname{NL}(X,0)$ such that every point of $\operatorname{NL}(X,0)$ that is not of type 2 has a basis of neighborhoods each isomorphic to $\operatorname{NL}(X,0)\setminus T$;
3. \[condition\_valtree\] there exists a finite set $T$ of type 1 points of $\operatorname{NL}(X,0)$ such that $\operatorname{NL}(X,0)\setminus T$ is isomorphic to an open subspace of $\operatorname{NL}(\A^2_k,0)$;
4. \[condition\_selfsim\] there exists a finite set $T$ of type 1 points of $\operatorname{NL}(X,0)$ such that every open subset of $\operatorname{NL}(X,0)$ contains an open subset isomorphic to $\operatorname{NL}(X,0) \setminus T$;
5. \[condition\_graph\] there exists a good resolution of $(X,0)$ whose weighted dual graph is self-similar;
6. \[condition\_kato\] there exists a proper birational morphism of algebraic $k$-surfaces $\pi\colon X'\to X$ which is not an isomorphism above $0$, together with a point $p\in \pi^{-1}(0)$ and an isomorphism of complete local rings $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$.
Both \[condition\_strongly\_selfsim\] and \[condition\_selfsim\] can be interpreted as self-similarity properties for $\operatorname{NL}(X,0)$, and imply the condition \[condition\_prime\]. In \[condition\_graph\], a vertex of a dual graph has as weights the genus and the self intersection of the corresponding component; such a weighted graph is said to be *self-similar* if it is isomorphic to a graph modification of itself, see Section \[section\_graphs\] for more details. A datum like the one of \[condition\_kato\] will be called a *Kato datum* for $(X,0)$.
Let us now illustrate the main ingredients of the proof of Theorem \[mainthm\], which requires a combination of methods from resolution of singularities, non-archimedean analytic geometry, formal geometry, and combinatorics.
Begin by observing that any sandwiched singularity can be obtained by performing a composition of point blowups $(Y,D) \to (\A^2_k,0)$, followed by the contraction of a connected divisor $E$ on $Y$ that is supported on $D$. This procedure yields two maps $Y\to X \to \A^2_k$ that shows that the contracted surface $X$, whose singular point $0$ is the image of $E$ through the contraction map, is “sandwiched” between two smooth surfaces, justifying the terminology. By picking any point $y$ in $E$ and performing on it the same sequence of blowups and contraction (see the subsection \[sec:164\] for a detailed explanation of how this can be done), one obtains a surface $X'$ above $Y$ with a singular point $p$ such that $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$, that is a Kato datum. This proves the implication \[condition\_sandwiched\] $\implies$ \[condition\_kato\].
In the sections \[section\_preliminarieslink\] and \[section\_links\_surfaces\] we study in detail the structure of non-archimedean links. In particular, with any modification $(Y',D')$ of $(X,0)$ is associated a *center map* $\operatorname{c}_{Y'}\colon \operatorname{NL}(X,0)\to D'$, and if $y'$ is a closed point of $D'$ then its inverse image $\operatorname{c}_{Y'}^{-1}(y')$ is an open subspace of $\operatorname{NL}(X,0)$ that is isomorphic to the complement of finitely many points of type 1 in the non-archimedean link $\operatorname{NL}(Y',y')$ of $y'$ in $Y'$ (see Proposition \[proposition\_propertiesNL\]). When applied to the Kato datum $X'\to X$ above, this shows that $\operatorname{NL}(X,0)$ contains a strict open subspace that is isomorphic to the complement of finitely many points of type 1 in $\operatorname{NL}(X,0)$ itself, since $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$ implies that $\operatorname{NL}(X',p)$ and $\operatorname{NL}(X,0)$ are isomorphic, that is the condition \[condition\_prime\] holds. Similar arguments based on the study of the structure of non-archimedean links permit to obtain the implications \[condition\_sandwiched\] $\implies$ \[condition\_valtree\] $\implies$ \[condition\_strongly\_selfsim\] $\implies$ \[condition\_selfsim\] of Theorem \[mainthm\], as is explained in Section \[section\_preliminariessandwiched\].
Showing that singularities having self-similar links have also self-similar weighted dual graphs, that is the implication \[condition\_prime\] $\implies$ \[condition\_graph\] of Theorem \[mainthm\], is a slightly more delicate matter, undertaken in Section \[section\_proof\_main\]. We prove an extension result for morphisms of a punctured disc into $\operatorname{NL}(X,0)$, Proposition \[lem:keyboundary\], and deduce that we can assume that the boundary $\partial U$ of an open subset $U$ of $\operatorname{NL}(X,0)$ isomorphic to the complement of finitely many points of type 1 in $\operatorname{NL}(X,0)$ consists only of finitely many points of type 2. We then use other results on the structure of $\operatorname{NL}(X,0)$, proven in subsection \[subsection\_existencemodifications\], to produce a (formal) modification $(Y',D')$ of $(X,0)$ together with a closed point $y'$ of $D'$ such that $U\cong \operatorname{c}_{Y'}^{-1}(y)$. Finally, we show that the dual graph associated with $(X,0)$ is self-similar by carefully choosing compatible resolutions of $(Y',y)$ and $(X,0)$.
The proof of the remaining implication, that is \[condition\_graph\] $\implies$ \[condition\_sandwiched\], requires two distinct steps that we believe to be of independent interest. First of all, we prove a purely combinatorial result, Theorem \[thm:graph-sdw\], showing that every self-similar weighted graph is *sandwiched*, which means that it can be embedded in a graph modification of the trivial graph (the dual graph of the blowup of $\A^2_k$ at $0$). We then show in Theorem \[thm:extend-spiv\] that, if $(X,0)$ is a normal surface singularity admitting a good resolution whose associated weighted dual graph is sandwiched, then $(X,0)$ is a sandwiched singularity. Over the complex numbers this result was originally proven by Spivakovsky in [@spivakovsky:sandsingdesingsurfNashtransf], but his proof relies on plumbing techniques for complex analytic spaces; we proceed in a similar way, using an analogue of plumbing in formal geometry.
$\diamond$
Since sandwiched singularities can be characterized in terms of their non-archimedean links, it is also natural to look for a characterization of them in terms of their archimedean links. We have not been able to find a self-similar property reminiscent of Theorem \[mainthm\], \[condition\_strongly\_selfsim\] or \[condition\_selfsim\]. However, building on Theorem \[mainthm\], \[condition\_kato\], in Section \[section\_complexanalytic\] we observe that links of sandwiched singularities are exactly those arising on a specific class of smooth compact complex surfaces.
To state precisely our results, we need to introduce some terminology. A compact complex surface $S$ contains *a global spherical shell* if it admits a biholomorphic copy of a neighborhood of the $3$-sphere in $\C^2$ that does not disconnect $S$. Surfaces containing a global spherical shell have been completely described by Kato [@kato:cptcplxmanifoldsGSS] (see also the subsequent work of G. Dloussky [@dloussky:phdthesis]). They are non-kähler compact surfaces of Kodaira dimension equal to $-\infty$ that play a special role in the Kodaira classification of compact complex surfaces, see the introduction of [@MR2726099]. Primary Hopf surfaces are the most emblematic examples of such surfaces: they are obtained as the orbit space of a contracting germ of biholomorphism of $(\C^2,0)$, and are diffeomorphic to the product of spheres $\mathbb S^3\times \mathbb S^1$. Any surface containing a global spherical shell is a deformation of a modification of a primary Hopf surface.
\[thm3\] Let $(X,0)$ be a complex sandwiched singularity, and choose a local embedding $X \subset \C^n$. Then, for any $\eps$ small enough, there exist a smooth compact complex surface $S$ having a global spherical shell and a holomorphic embedding $$\imath : X\cap \{z\in \C^n,\, \eps/2 < \|z\| < 2 \eps\} \longrightarrow S$$ such that $S \setminus \operatorname{{L}_\C^\eps}(X,0)$ is connected.
Observe that the link $\operatorname{{L}_\C^\eps}(X,0)$ is included in the domain $X\cap \{z\in \C^n,\, \eps/2 < \|z\| < 2 \eps\}$ so that one can phrase the property in the previous theorem as an embedding property for the archimedean link of a sandwiched singularity. Following the terminology of M. Kato, this says that $\operatorname{{L}_\C^\eps}(X,0)$ can be realized as a real-analytic global strongly pseudoconvex $3$-fold in a surface containing a global spherical shell, see Section \[section\_complexanalytic\] for a discussion of these notions.
We prove that this property characterizes sandwiched singularities.
\[thm2\] Let $(X,0)$ be a complex normal surface singularity, and choose a local embedding $X \subset \C^n$. Suppose that for some small enough $\eps>0$, the archimedean link $\operatorname{{L}_\C^\eps}(X,0)$ can be realized as a real-analytic global strongly pseudoconvex $3$-fold in a compact complex surface $S$. Then we are in exactly one of the following situations:
1. \[item:thm2a\] $(X,0)$ is a weighted homogeneous singularity which is not rational, and $S$ is an elliptic surface of Kodaira dimension either $0$ or $1$;
2. \[item:thm2b\] $(X,0)$ is a quotient singularity, and $S$ is a secondary Hopf surface;
3. \[item:thm2c\] $(X,0)$ is a sandwiched singularity, and $S$ carries a global spherical shell.
An elliptic surface is a compact complex surface carrying an elliptic fibration [@barth-hulek-peters-vanderven:compactcomplexsurfaces p.200]. A Hopf surface is a compact complex surface whose universal cover is biholomorphic to $\C^2\setminus\{0\}$. When its fundamental group is cyclic, then it is a primary Hopf surface in the previous sense. Otherwise it admits a non-trivial finite cyclic cover by a primary Hopf surface, in which case it is called secondary Hopf surface.
A secondary Hopf surface is never elliptic and does not contain any global spherical shell, so that the three classes of surfaces arising in Theorem \[thm2\] are really disjoint.
Observe that Theorems \[thm3\] and \[thm2\] imply immediately the following result.
\[cor2\] A normal surface singularity is sandwiched if and only if its archimedean link can be realized as a real-analytic global strongly pseudoconvex $3$-fold in a compact complex surface containing a global spherical shell.
[**Acknowledgements.**]{} We would like to warmly thank B. Teissier, who asked us about a characterization of singularities having self-similar Riemann-Zariski spaces. This paper is a tentative answer to his question in the framework of normalized Berkovich analytic spaces.
Preliminaries on non-archimedean links {#section_preliminarieslink}
======================================
In this section we recall the construction of the non-archimedean link $\operatorname{NL}(X,Z)$ of a subscheme $Z$ in an algebraic variety $X$ from [@fantini:normspaces] (where it is called normalized non-archimedean link).
Berkovich analytifications
--------------------------
We begin by recalling the definition of the Berkovich analytification of an algebraic variety, following [@berkovich:book]. Let $K$ be a field complete with respect to a non-archimedean absolute value $|\cdot|$, that is an absolute value such that $|a+b|\leq \max\{|a|,|b|\}$ for every $a$ and $b$ in $K$. We denote by $K^\circ =\{ |a| \le 1\}$ the valuation ring of $K$. In the sequel $K$ will either be the field of Laurent series $k((t))$ with a $t$-adic absolute value for some algebraically closed field $k$, or any field endowed with the trivial absolute value, that is the absolute value such that $|K^\times|=1$.
Let $X=\operatorname{Spec}(A)$ be an affine algebraic variety over $K$. The *analytification* $X^\mathrm{an}$ of $X$ is defined as the following set of multiplicative semi-norms: $$X^{\mathrm{an}}=\Big\{x\colon A\to \R_{\geq0} \,\big|\, x(ab)=x(a)x(b), x(a+b)\leq x(a)+x(b), x(c)=|c| \, \forall a,b\in A, c\in K\Big\},$$ with the topology of the pointwise convergence, that is the topology induced by the product topology of $(\R_{\geq0})^A$. The definition extends by gluing to any algebraic variety over $K$ (and more generally to any $K$-scheme of finite type).
If $x$ is a point of $X^\mathrm{an}$ then its kernel $\ker x$ is a prime ideal of $A$ and $x$ induces an absolute value on the quotient $A/\ker x$, which is the residue field of $X$ at $\ker x$. The completion of $A/\ker x$ with respect to this absolute value is a complete valued field extension of $K$ which will be denoted by $\mathscr H(x)$ and called the *complete residue field* of $X^\mathrm{an}$ at $x$. If $f$ is an algebraic function on $X$, we will denote by $f(x)$ its image in $\mathscr H(x)$, and by $|f(x)|\in\R_{\geq0}$ the image of $f(x)$ through the absolute value of $\mathscr H(x)$.
More generally, $X^\mathrm{an}$ comes equipped with a sheaf of $K$-algebras of *analytic functions*, consisting of the functions which can be written locally as a uniform limit of rational functions without poles. If $U$ is an open subspace of $X^\mathrm{an}$, then an analytic function $f\in\mathcal O_{X^\mathrm{an}}(U)$ can be evaluated in a point $x$ of $U$, yielding an element $f(x)\in \mathscr H(x)$, and therefore a positive real number $|f(x)|$. An analytic funtion $f$ on $U$ is said to be *bounded* if $|f(x)|\leq1$ for every $x$ in $U$. Bounded analytic functions form a subsheaf $\mathcal O^\circ_{X^\mathrm{an}}$ of $K^\circ$-modules of $\mathcal O_{X^\mathrm{an}}$.
Moreover, $X^\mathrm{an}$ can be endowed with an additional Grothendieck topology. We will not discuss this last aspect in the rest of the paper since we will only be considering open subspaces of analytic spaces.
The center map
--------------
For the remaining of this section we work with an algebraic variety $X$ over a trivially valued and algebraically closed field $k$.
Any point $x$ of $X^\mathrm{an}$ comes together with a morphism $\alpha\colon \operatorname{Spec}\big(\mathscr H (x)\big)\to X$. We say that $x$ *has center* on $X$ if $\alpha$ extends to a morphism $\overline{\alpha} \colon \operatorname{Spec}\big(\mathscr H (x)^\circ\big)\to X$, where $\mathscr H (x)^\circ$ is the valuation ring of $\mathscr H (x)$. The *center* of $x$ is then defined as the point $\mathrm{c}_X(x)=\overline{\alpha}(s)$ of $X$, where $s$ is the closed point of $\operatorname{Spec}\big(\mathscr H (x)^\circ\big)$. This coincides with the notion, classical in valuation theory, of the center of the valuation ring $\mathscr H(x)^\circ$. Observe that, since $X$ is separated by hypothesis, whenever $x$ has center on $X$ then its center $\mathrm{c}_X(x)$ is a well defined point of $X$. If moreover $X$ is proper over $k$, then by the valuative criterion of properness every point of $X^\mathrm{an}$ has center on $X$, but this is not true in general.
The *center map* $\mathrm c_X\colon \big\{x\in X^\mathrm{an} \text{ having center on } X\}\to X$ is anticontinuous, which means that $\mathrm c_X^{-1}(Z)$ is open whenever $Z$ is a closed subvariety of $X$. The notion of center describes the property for a point of $X^\mathrm{an}$ to be close to a point of $X$, so whenever $Z$ is a closed subvariety of $X$ the subset $\mathrm c_X^{-1}(Z)$ of $X^\mathrm{an}$ can be thought of as a tubular neighborhood of $Z$ in $X^\mathrm{an}$. The complement $\mathrm c_X^{-1}(Z)\setminus Z^\mathrm{an}$ of $Z^\mathrm{an}$ in the tubular neighborhood can then be thought of as a *punctured tubular neighborhood* of $Z$ in $X^\mathrm{an}$.
\[rmk:thu\] In the terminology of [@thuillier:geometrietoroidale], the punctured tubular neighborhood $\mathrm c_X^{-1}(Z)\setminus Z^\mathrm{an}$ coincides with the analytic space $\big(\widehat{X/Z}\big)_\eta$ associated with the formal completion $\widehat{X/Z}$ of $X$ along $Z$.
Non-archimedean links
---------------------
We fix an algebraic variety $X$ over $k$, and a nonempty and nowhere dense closed subscheme $Z$ of $X$.
An element $\lambda$ of $\R_{>0}$ acts on a point $x$ of $X^\mathrm{an}$ by raising the semi-norm $x$ to the power $\lambda$. Indeed, the condition for $x^\lambda$ to be in $X^\mathrm{an}$, that is the fact that it is trivial on $k$, is satisfied since the trivial absolute value is invariant under exponentiation by elements of $\R_{>0}$. Observe that as an abstract field the completed residue field $\mathscr H(x)$ is isomorphic to $\mathscr H(x^\lambda)$, but the absolute value of the latter is obtained by raising to the power $\lambda$ the one of the former. Therefore, neither the abstract valuation ring $\mathscr H(x)^\circ$ nor the morphism of schemes $\operatorname{Spec}\big(\mathscr H(x)\big)\to X$ associated with $x$ change when replacing $x$ by $x^\lambda$. It follows that the action of $\R_{>0}$ on $X^\mathrm{an}$ induces an action on the punctured tubular neighborhood $\mathrm c_X^{-1}(Z)\setminus Z^\mathrm{an}$.
We define the *non-archimedean link* $\operatorname{NL}(X,Z)$ of $Z$ in $X$ as the quotient of the punctured tubular neighborhood of $Z$ in $X^\mathrm{an}$ by this action: $$\operatorname{NL}(X,Z) = \big( \mathrm{c}_X^{-1}(Z)\setminus Z^{\mathrm{an}} \big) \big/ \R_{>0}.$$
We endow $\operatorname{NL}(X,Z)$ with the quotient topology, and see it as a ringed space by endowing it with the following two sheaves. The *sheaf of analytic functions* $\mathcal O_{\operatorname{NL}(X,Z)}$ on $\operatorname{NL}(X,Z)$ is by definition the push-forward to $\operatorname{NL}(X,Z)$ of the sheaf of analytic functions $\mathcal O_{\mathrm c_X^{-1}(Z)\setminus Z^\mathrm{an}}$ on the Berkovich analytic space $\mathrm c_X^{-1}(Z)\setminus Z^\mathrm{an}$ via the quotient map. Analogously, the *sheaf of bounded analytic functions* $\mathcal O_{\operatorname{NL}(X,Z)}^\circ$ on $\operatorname{NL}(X,Z)$ is the push-forward of the sheaf of bounded analytic functions $\mathcal O_{\operatorname{c}_X^{-1}(Z)\setminus Z^\mathrm{an}}^\circ$. Both are local sheaves of $k$-algebras and $\mathcal O_{\operatorname{NL}(X,Z)}^\circ$ is a subsheaf of $\mathcal O_{\operatorname{NL}(X,Z)}$. We will say more about the analytic structure of $\operatorname{NL}(X,Z)$ in the subsequent subsections. Observe that $\operatorname{NL}(X,Z)$ is a compact topological space by [@fantini:normspaces Proposition 5.9].
It is worth noticing that by Remark \[rmk:thu\] the space $\mathrm c_X^{-1}(Z)\setminus Z^{\mathrm{an}} $ only depends on the formal completion of $X$ along $Z$, so the same is true for $\operatorname{NL}(X,Z)$. In particular, if $0$ and $0'$ are closed points of algebraic $k$-varieties $X$ and $X'$ respectively, then the non-archimedean links $\operatorname{NL}(X,0)$ and $\operatorname{NL}(X',0')$ are isomorphic (as locally ringed spaces) if and only if the corresponding complete local rings $\widehat{\mathcal O_{X,0}}$ and $\widehat{\mathcal O_{X',0'}}$ are isomorphic. It follows that if $k=\C$ then analytically isomorphic singularities have isomorphic non-archimedean links.
One important feature of non-archimedean links is their invariance under modifications. We use the following non-conventional terminology: we define a *modification* of $(X,Z)$ to be a pair $(Y,D)$, where $Y$ is a normal algebraic $k$-variety and $D$ is a Cartier divisor of $Y$, together with a proper morphism $\pi\colon Y\to X$ which is an isomorphism out of $D$ and such that $D=\pi^{-1}(Z)$ is the (schematic) inverse image of $Z$ through $\pi$.
The following result is then a consequence of the valuative criterion of properness.
\[proposition\_invariance\_modifications\] If $\pi\colon(Y,D)\to(X,Z)$ is a modification of $(X,Z)$, then $\pi$ induces an isomorphism of locally ringed spaces $\operatorname{NL}(Y,D)\stackrel{\sim}{\to}\operatorname{NL}(X,Z)$.
The map $\mathrm{c}_X$ induces an anticontinuous map $\mathrm{c}_X\colon \operatorname{NL}(X,Z)\to Z$ which we still call the *center map*. Thanks to the result above, we also have a center map $\mathrm c_Y\colon \operatorname{NL}(X,Z)\to D$ associated with any modification $(Y,D)$ of $(X,Z)$.
Analytic structure of non-archimedean links
-------------------------------------------
We fix an algebraic variety $X$ over $k$, and a nonempty and nowhere dense closed subscheme $Z$ of $X$. We now explain some properties of the ringed space structure of the non-archimedean link $\operatorname{NL}(X,Z)$.
Some caution is needed when working with analytic functions on $\operatorname{NL}(X,Z)$, as they cannot be evaluated at points of $\operatorname{NL}(X,Z)$, given that such a point is only a $\R_{>0}$-equivalence class of semi-norms.
However, it is possible to say whether the value of a function $f\in\mathcal O_{\operatorname{NL}(X,Z)}(U)$ at a point $x\in U$ lies in $\{0\}$, $]0,1[$, $\{1\}$, or $]1,+\infty[$, as those are the orbits of $\R_{\geq0}$ under the action of $\R_{>0}$ by exponentiation. In particular it makes sense to ask whether a function vanishes at a point. Moreover, for any open set $U$ one can interpret $\mathcal O_{\operatorname{NL}(X,Z)}^\circ(U)$ as the ring of those functions on $U$ which are bounded by $1$.
As follows from the definition, with any point $x$ of $\operatorname{NL}(X,Z)$ is associated a field extension $\mathscr H(x)$ of $k$, endowed with a rank $1$ valuation (but not with an absolute value) trivial on $k$, with respect to which $\mathscr H(x)$ is complete, and therefore a rank $1$ valuation ring which we still denote by $\mathscr H(x)^\circ$. Conversely, every complete (rank 1) valuation ring on the residue field of a scheme-theoretic point of $X\setminus Z$ such that its center on $X$ is in $Z$ comes from a point of $\operatorname{NL}(X,Z)$. Observe that a function $f\in\mathcal O_{\operatorname{NL}(X,Z)}(U)$ is bounded by $1$ on $U$ if and only if $f(x)\in \mathscr H(x)^\circ$ for every $x$ in $U$. The valued field $\mathscr H(x)$ can be defined more intrinsically, in a way that only depends on the ringed space structure of $\operatorname{NL}(X,Z)$, as the completion of the residue field of the local ring $\mathcal O_{\operatorname{NL}(X,Z),x}$ with respect to the valuation induced by $\mathcal O_{\operatorname{NL}(X,Z),x}^\circ$.
The analytic structure of $\operatorname{NL}(X,0)$ contains abundant information about the pair $(X,0)$, as is clear from the following result of [@fantini:normspaces], which we recall for the reader’s convenience.
\[proposition\_propertiesNL\_1\] Let $0$ be any closed point in an algebraic variety $X$ and assume that $X$ is normal at $0$. Then the canonical morphism $$\widehat{\mathcal O_{X,0}}
\to
\mathcal O_{\operatorname{NL}(X,0)}^\circ \big( \operatorname{NL}(X,0) \big)$$ is an isomorphism.
Non-archimedean links and $k((t))$-analytic spaces
--------------------------------------------------
A crucial property of non-archimedean links is that they are locally isomorphic to analytic spaces over a field of Laurent series $k((t))$ with $t$-adic absolute value, in the following sense. Choose $\varepsilon\in]0,1[$ and endow $k((t))$ with the $t$-adic absolute value such that $|t|=\varepsilon$. As we have seen, any Berkovich analytic space $\mathfrak X$ over $k((t))$, for example an open or closed subspace of the analytification of an algebraic $k((t))$-variety, comes equipped both with a sheaf of $k((t))$-algebras $\mathcal O_{\mathfrak X}$ and with a sheaf of $k[[t]]$-algebras $\mathcal O_{\mathfrak X}^\circ$. We can see these two sheafs only as sheaves in $k$-algebras, yielding a triple which we denote by $\operatorname{For}(\mathfrak X)=\big({\mathfrak X},\mathcal O_{\mathfrak X},\mathcal O^\circ_{\mathfrak X}\big)$. It then makes sense to ask whether a non-archimedean link is isomorphic as such a triple to a triple of the form $\operatorname{For}(\mathfrak{X})$. In general, this is true only locally, in the following sense. If $X$ is affine, then $\operatorname{NL}(X,Z)$ can be covered by finitely many open subspaces which, as ringed spaces in $k$-algebras, are of the form $\operatorname{For}(\mathfrak X)$ for some $k((t))$-analytic space $\mathfrak X$. Observe that this is also the case if $Z$ is a single point of $X$, since $X$ can then be replaced by an affine neighborhood of $Z$. If $X$ is not affine, then $\operatorname{NL}(X,Z)$ is covered by the compact domains $\mathrm c_X(U\cap Z)$, for $U$ ranging among an open affine cover of $X$, and each $\mathrm c_X(U\cap Z)$ is locally isomorphic to a $k((t))$-analytic space in the sense above. Mind that the datum consisting of such a covering and $k((t))$-analytic structures is non canonical.
A proof of this fact can be found in [@fantini:normspaces Corollary 4.10], but to help the reader familiarize with the structure of non-archimedean links we illustrate here what happens in the case when $Z=\{0\}$ is a closed point of $X$. Let $f$ be an element of the completed local ring $\widehat{\mathcal O_{X,0}}$ of $X$ at $0$. This defines a $k$-analytic map from $\mathrm{c}_X^{-1}(0)$ into the open unit ball in ${\mathbb A}^{1,\mathrm{an}}_k$. The latter is canonically homemorphic to the interval $[0,1[$, and under this homeomorphism this analytic map is given by the absolute value $\mathrm{c}_X^{-1}(0) \stackrel{|f|}{\longrightarrow} \left[0,1\right[$. The fiber $|f|^{-1}(\varepsilon)$ of $|f|$ at $\varepsilon \in \left]0,1\right[$ is then an analytic space over $k((t))$, since the completed residue field of $\mathbb A^{1,\mathrm{an}}_k$ at the point corresponding to $\varepsilon$ is the field $k((t))$ with the $t$-adic absolute value such that $|t|=\varepsilon$.
Then the projection $\pi\colon \mathrm{c}_X^{-1}(0)\setminus\{0\}^\mathrm{an} \to \operatorname{NL}(X,0)$ defining $\operatorname{NL}(X,0)$ identifies $\operatorname{For}\big(|f|^{-1}(\varepsilon)\big)$ with its image in $\operatorname{NL}(X,0)$, which is the complement $\operatorname{NL}(X,0)\setminus V(f)$ of the zero locus $V(f)$ of $f$ in $\operatorname{NL}(X,0)$. Finally, by having $f$ range among a finite set of generators of the maximal ideal of $\widehat{\mathcal O_{X,0}}$, we obtain a cover $\operatorname{NL}(X,0)$ with finitely many open subspaces, each of which is isomorphic to a $k((t))$-analytic space.
Observe that the $k((t))$-analytic space $|f|^{-1}(\varepsilon)$ is the analytic Milnor fiber $\mathscr F_{f,0}$ of $f$ at $0$, an object defined and studied in [@nicaise-sebag:analyticmilnorfiber]. The non-archimedean link $\operatorname{NL}(X,0)$ can be thought of as a global version of $\mathscr F_{f,0}$, dependent only on the germ of $X$ at $0$ and not on $f$.
Discs and annuli {#sec:DSK}
----------------
Some specific subspaces of non-archimedean links of surfaces are particularly important and deserve to be studied in depth. Let $T$ denote a coordinate function on the $k((t))$-analytic affine line $\A^{1,\mathrm{an}}_{k((t))}$. We say that a subspace $U$ of a non-archimedean link $\operatorname{NL}(X,Z)$ is a *disc* if, as a ringed space in $k$-algebras, $U$ is isomorphic to $\operatorname{For}(D)$, where $D=\big\{x\in \A^{1,\mathrm{an}}_{k((t))} \,\big|\, |T(x)|<1\big\}$ is an open unit $k((t))$-analytic disc. We say that $U$ is an *annulus* if, as a ringed space in $k$-algebras, $U$ is isomorphic to $\operatorname{For}(A)$, where $A=\big\{x\in \A^{1,\mathrm{an}}_{k((t))} \,\big|\, |t|<|T(x)|<1\big\}$ is an open $k((t))$-analytic annulus of modulus one. We collect in the following statement some well known results about the topology of discs and annuli.
\[thm:R-tree\] Let $D=\big\{x\in \A^{1,\mathrm{an}}_{k((t))} \,\big|\, |T(x)|<1\big\}$ and $A=\big\{x\in \A^{1,\mathrm{an}}_{k((t))} \,\big|\, |t|<|T(x)|<1\big\}$ be a $k((t))$-analytic disc and a $k((t))$-analytic annulus respectively. For any $r\in (0,1)$ (respectively for $r\in (|t|,1)$), denote by $x_r$ the point of $D$ (respectively of $A$) defined by $|P(x_r) | = \sup_{|z|\le r} |P(z)|$ for any $P\in k((t))[T]$. \[def of xr\]
1. $D$ and $A$ are uniquely arcwise connected: any two distint points $x, y$ are included in a unique closed subset $I$ which is homeomorphic to the closed interval $[0,1]$ and such that $I \setminus \{x,y\}$ is homeomorphic to the open interval $(0,1)$.
2. $D$ has a single endpoint. Given a continuous, proper, and injective map $\gamma \colon \R_+ \to D$, there exists a constant $C >0$ such that $\gamma \big([C, +\infty)\big) = \{ x_r \textrm{, } 1 - \varepsilon \le r < 1\}$ for some $0<\varepsilon <1$.
3. $A$ has two endpoints. Given a continuous, proper, and injective map $\gamma\colon \R_+ \to D$, there exist constants $C>0$ and $|t|<\varepsilon <1$ such that $\gamma \big([C, +\infty)\big)$ is either equal to $\{ x_r \textrm{, } 1 - \varepsilon \le r < 1\}$, or to $\{ x_r\textrm{, } |t| < r\le |t| + \varepsilon\}$.
Now let $U$ be a subset of a non-archimedean link $\operatorname{NL}(X,Z)$ and assume that $U$ is an annulus. Fix a $k((t))$-analytic annulus $A$ such that $U \cong \operatorname{For}(A)$, and denote by $\Sigma(U)$ the subset of $U$ corresponding to the subset $\{ x_r \textrm{ s.t. } |t| < r < 1\}$ of $A$. It consists of points of type 2 or 3 of $A$. The fact that the subset $\Sigma(U)$ of $U$ does not depend on the choice of a $k((t))$-analytic annulus $A$ such that $U \cong \operatorname{For}(A)$ is a consequence of the following proposition.
\[proposition skeleton annulus\] The subset $\Sigma(U)$ of $U$ coincides with the set of points of $U$ which have no neighborhood isomorphic to a disc.
Any neighborhood of a point of $\Sigma(U)$ in $U$ has at least two endpoints, and thus can’t be a disc. Conversely, let as above $A$ be a $k((t))$-analytic annulus such that $U \cong \operatorname{For}(A)$. The complement of $\Sigma(U)$ in $A$ is the union of the open balls $D(z,|z|)$, for $z$ ranging among the rigid points of $A$. If $D=D(z,|z|)$ is such a ball, $z$ is a defined over a finite extension $k((s))$ of $k((t))$, and the analytic function $s$ is globally defined on $D$. Having a rational point as center and a rational radius, $D$ can be seen as an open $k((s))$-analytic disc, and therefore $\operatorname{For}(D)$ is a disc.
The following lemma, which will be used in section \[section\_proof\_main\], shows how discs and annuli appear as open subspaces of $\operatorname{NL}(\A^2_k,0)$.
\[lemma\_disc\_annulus\] Let $T$ and $t$ denote two coordinates for $\mathbb A^2_k$ at $0$. Then:
1. $\operatorname{NL}(\A^2_k,0) \setminus V(t)$ is a disc;
2. $\operatorname{NL}(\A^2_k,0) \setminus V(tT)$ is an annulus.
Note that $\mathrm{c}_{\A^{2}_{k}}^{-1}(0) = \big\{x\in \A^{2,\mathrm{an}}_{k} \,\big|\, |T(x)|<1, |t(x)|<1 \big\}$. Therefore we have $$\begin{aligned}
\operatorname{NL}(\A^{2}_{k},0) \setminus V(t)
& \cong \big\{ x \in \mathrm{c}_{\A^{2}_{k}}^{-1}(0) \,\big|\, |t(x)|=\varepsilon\big\} \\
& = \big\{x\in \A^{2,\mathrm{an}}_{k} \,\big|\, |T(x)|<1, |t(x)|=\varepsilon\big\} \\
& \cong \operatorname{For}\big(\big\{x\in \A^{1,\mathrm{an}}_{k((t))} \,\big|\, |T(x)|<1\big\}\big) = \operatorname{For}(D).\end{aligned}$$ The isomorphisms above respect the ringed space structure, proving $(i)$.
Set now $t'=tT$. Then we have $$\begin{aligned}
\operatorname{NL}(\A^2_k,0)\setminus V(t')
& \cong \big\{\mathrm{c}_{\A^{2}_{k}}^{-1}(0) \,\big|\, |t(x)|=\varepsilon\big\}
\\
& = \big\{x\in \A^{2,\mathrm{an}}_{k} \,\big|\, |T(x)|<1, |t(x)|<1, |t'(x)|=\varepsilon\big\}
\\
& = \big\{x\in \A^{2,\mathrm{an}}_{k} \,\big|\, |t'(x)|<|T(x)|<1, |t'(x)|=\varepsilon\big\}
\\
& \cong \operatorname{For}\big(\big\{x\in \A^{1,\mathrm{an}}_{k((t'))} \,\big|\, |t'|<|T(x)|<1\big\}\big)
= \operatorname{For}(A),\end{aligned}$$ which concludes the proof of $(ii)$.
\[remark\_disc\_minus\_point\] Note that $V(t)$ and $V(T)$ are single points of $\operatorname{NL}(\A^{2}_{k},0)$. Observe that the lemma also shows that a punctured open $k((t))$-analytic disc $D\setminus\{0\}$ and an open $k((t))$-analytic annulus of modulus one, which are not isomorphic as analytic spaces, have isomorphic underlying ringed spaces in $k$-algebras.
Morphisms
---------
A morphism $\operatorname{NL}(X,Z) \to \operatorname{NL}(X',Z')$ between two non-archimedean links is a morphism of the underlying ringed spaces $$\Big(\operatorname{NL}(X,Z),\mathcal O_{\operatorname{NL}(X,Z)},\mathcal O^\circ_{\operatorname{NL}(X,Z)}\Big)
\longrightarrow
\Big(\operatorname{NL}(X',Z'),\mathcal O_{\operatorname{NL}(X',Z')},\mathcal O^\circ_{\operatorname{NL}(X',Z')}\Big)$$ which can be locally lifted to a morphism of $k((t))$-analytic spaces. Rather than giving the precise definition from [@fantini:normspaces 6.1], we content ourselves with giving a list of example of morphism of non-archimedean links, including all the morphisms which will be considered in this paper.
1. An *isomorphism* of non-archimedean links $\operatorname{NL}(X,Z) \to \operatorname{NL}(X',Z')$ is an isomorphism of the underlying ringed spaces.
2. \[item:bloqups\_induce\_isomorphism\] As noted in Proposition \[proposition\_invariance\_modifications\], if $ f \colon (X,Z) \to (X',Z')$ is a modification then the induced morphism $f \colon \operatorname{NL}(X,Z) \to \operatorname{NL}(X',Z')$ is an isomorphism.
3. A morphism of $k$-varieties $f \colon X \to X'$ such that $f^{-1}(Z') = Z$ induces a morphism of non-archimedean links $f \colon \operatorname{NL}(X,Z) \to \operatorname{NL}(X',Z')$.
4. In particular, if $Y \subsetneq X$ is a subvariety such that $Y\cap Z$ is nowhere dense in $Y$, then the induced morphism $\operatorname{NL}(Y, Y\cap Z) \to \operatorname{NL}(X,Z)$ is a closed immersion, that is a map which lifts to closed immersions of $k((t))$-analytic spaces in the sense of Berkovich. Its image is the closed subspace of $\operatorname{NL}(X,Z)$ consisting of the elements coming from elements of $Y^\mathrm{an}$; note that each such point is a semi-norm with nontrivial kernel.
5. If $(X,Z)$ are as above and $F$ is a closed subvariety of $Z$, then $\operatorname{c}_X^{-1}(F)$, which is isomorphic to $\operatorname{NL}(X,F)\setminus Z^\mathrm{an}$, is an open subspace of $\operatorname{NL}(X,Z)$.
Observe that the converse to \[item:bloqups\_induce\_isomorphism\] is partially true, in the following sense. By [@fantini:normspaces Theorem 6.4], if $\operatorname{NL}(X,Z)$ is isomorphic to $\operatorname{NL}(X',Z')$ then there exist modifications $(Y,D)\to(X,Z)$ and $(Y',D')\to(X',Z')$ such that the corresponding formal completions $\widehat{Y/D}$ and $\widehat{Y'/D'}$ are isomorphic as formal schemes.
Non-archimedean links of surfaces {#section_links_surfaces}
=================================
In this section we assume that $X$ is a *normal algebraic surface* over an algebraically closed field $k$ and that $0$ is a closed point of $X$. Our approach follows the one of [@fantini:normspaces Sections 7, 9, 10], but for some results we will provide more concrete proofs.
Topology of $\operatorname{NL}(X,0)$ and center maps
----------------------------------------------------
In dimension two the topology of $\operatorname{NL}(X,0)$ can be described in terms of its center maps as follows.
\[prop:topo\_NL\] Let $(X,0)$ be a normal surface singularity, and let $x$ be a point of $\operatorname{NL}(X,0)$. Then the family of all sets of the form $\mathrm c_Y^{-1}\big(\overline{\mathrm c_Y(x)}\big)$, for $(Y,D)$ ranging over all the modifications of $(X,0)$, is a basis of neighborhoods of $x$ in $\operatorname{NL}(X,0)$.
Let $\mathfrak{m}$ be the maximal ideal of $\cO_{X,0}$, which consists of the functions vanishing at $0$, let $V$ be any affine neighborhood of $0$ in $X$, and denote by $\mathrm{L}(X,\mathfrak{m})$ the subset of $U^{\mathrm{an}}$ consisting of the multiplicative semi-norms $x$ on $\mathcal O_X(U)$ whose restriction to $k$ is trivial and such that $\min_{f\in \mathfrak{m}} -\log|f(x)| = 1$. It is a compact subset of $\big( \mathrm{c}_X^{-1}(0)\setminus \{0\}^{\mathrm{an}} \big)$ and the natural projection $\mathrm{L}(X,\mathfrak{m})\to \operatorname{NL}(X,0)$, is bijective hence a homeomorphism. It is then sufficient to prove that the family of all subsets of $\mathrm L(X,\mathfrak{m})$ of the form $\mathrm c_Y^{-1}\big(\overline{\mathrm c_Y(x)}\big)$, for $(Y,D)$ ranging over all the modifications of $(X,0)$, is a basis of neighborhoods of any given point $x$ of $\mathrm L(X,\mathfrak{m})$. Any set of the form $\mathrm c_Y^{-1}\big(\overline{\mathrm c_Y(x)}\big)$ contains $x$, and it is open because the center map is anti-continuous. We have to prove that, given any open subset $U$ of $\mathrm L(X,\mathfrak{m})$ containing $x$, there exists a modification $(Y,D) \to (X,0)$ such that $\mathrm c_Y^{-1}\big(\overline{\mathrm c_Y(x)}\big) \subset U$. Since $X^{\mathrm{an}}$ is endowed with the weakest topology making all evaluations maps $y \mapsto |f(y)|$ continuous, it is sufficient to prove it assuming that $U$ is a finite intersection of sets of the form $U_<(f,p,q)= \{y, |f(y)| <e^{-p/q}\}$ or $U_>= \{y, |f(y)| >e^{-p/q}\}$, where $f$ is an element of $\mathcal O_X(U)$ and $p,q$ are coprime integers. Moreover, since $|f| =1$ on $\mathrm{L}(X,\mathfrak{m})$ whenever $f\notin \mathfrak{m}$, we can also assume that $f\in \mathfrak{m}$. It is sufficient to find a modification $(Y,D) \to (X,0)$ such that $U_<(f,p,q)$ (resp. $U_>(f,p,q)$) is included in a set of the form $ c_Y^{-1} (\cE_<)$ (resp. $ c_Y^{-1} (\cE_>)$), for some closed subschemes $\cE_<$ and $\cE_>$ of $D$. In order to do so we proceed as follows. Let $g_1, \ldots, g_N$ be a finite set of generators of $\mathfrak{m}$, and choose a modification $\pi\colon (Y,D) \to (X,0)$ such that $g_i^p/f^q \circ \pi $ defines a regular map $Y \to \mathbb{P}^1_k$ for all $i=1, \ldots, N$. Let $\cE_>$ be the union of all (scheme-theoretic) points $\xi$ of $D$ such that $$\label{eq:>}
\operatorname{ord}_\xi( f^q\circ \pi)> \operatorname{ord}_\xi(\mathfrak{m}^p) := \mathrm{min}_{g\in\mathfrak M^p}\{\operatorname{ord}_\xi(g\circ\pi)\}
= \mathrm{min}_{g\in\mathfrak M}\{\operatorname{ord}_\xi(g^p\circ\pi)\}~.$$ Observe that $z\in \cE_>$ if and only if the value of the rational function $g_i^p/f^q (\pi(z))$ is equal to $\infty \in \mathbb P^1_k$ for at least one index $i$, which implies that $\cE_>$ is a finite union of closed points in $D$ together with those irreducible components $E$ of $D$ whose generic point $\xi_E$ satisfies . In particular $\cE_>$ is closed in $D$. One defines in the same way $\cE_<$ as the set of points $\xi\in D$ satisfying $$\operatorname{ord}_E( f^q\circ \pi)< \operatorname{ord}_E(\mathfrak{m}^p),$$ and similarly $\cE_<$ is closed (it is in fact a finite union of irreducible components of $D$).
Since $g_i^p/f^q \circ \pi $ is regular, we have that $\cE_> \cap \cE_< =\emptyset$, and moreover $$\begin{aligned}
U_<(f,p,q) & = \{y, \, -\log |f^q(y)| > p\} = c_Y^{-1} (\cE_<)
\\
U_>(f,p,q) & = \{y,\, -\log |f^q(y)| < p\} = c_Y^{-1} (\cE_>)~,
\end{aligned}$$ which concludes the proof.
Types of points
---------------
Observe that a point $x$ of $\operatorname{NL}(X,0)$ corresponds to an equivalence class of semi-norms on $X$; being centered in $0$, those induce semi-norms on the completed local ring $\widehat {\mathcal O_{X,0}}$. The points of $\operatorname{NL}(X,0)$ can then be divided into four different types by looking at the associated valued field ${\mathscr H(x)}$, its trascendence degree over $k$, and its valuative invariants.
We say that a point $x$ of $\operatorname{NL}(X,0)$ is a *rigid point* (or of *type 1*) if the transcendence degree $\mathrm{trdeg}_{k}{\mathscr H(x)}$ of $\mathscr H(x)$ over $k$ is equal to $1$. Equivalently, $x$ is a rigid point if it corresponds to an equivalence class of semi-norms on $\widehat {\mathcal O_{X,0}}$ with nontrivial kernel. Moreover, this kernel is generated by an irreducible element of $\widehat {\mathcal O_{X,0}}$, so that $x$ corresponds to an irreducible germ of a formal curve on $(X,0)$; then $x$ can be seen as the equivalence class of the order of vanishing along this germ. When this is the case, the rational rank $$\mathrm{rank}_\Q \big( |\mathscr H(x)^\times|/|k^\times|\otimes_\Z\Q \big)$$ of $\mathscr H(x)$ is equal to $1$ and its residue field $\widetilde{\mathscr H(x)}$ is equal to $k$.
In every other case we have $\mathrm{trdeg}_{k}{\mathscr H(x)}=2$. We say that $x$ is a *divisorial point* (or of *type 2*) if the residue field $\widetilde{\mathscr H(x)}$ is a nontrivial extension of $k$, while we say that $x$ is of *type 3* if the rational rank of $\mathscr H(x)$ is equal to $2$. Finally, we say that $x$ is of *type 4* if it satisfies $\mathrm{rank}_\Q \big( |\mathscr H(x)^\times|/|k^\times|\otimes_\Z\Q \big)=1$ and $\widetilde{\mathscr H(x)}=k$ but it is not of type 1 (that is, the corresponding semi-norms on $\widehat{\cO_{X,0}}$ are norms).
Every point of $\operatorname{NL}(X,0)$ is of one (and only one) of the four types above, since by Abhyankar’s inequality [@abhyankar:valuationscentered] we have $$\mathrm{rank}_\Q \big( |\mathscr H(x)^\times|/|k^\times|\otimes_\Z\Q \big)+\mathrm{trdeg}_{k}\widetilde{\mathscr H(x)}\leq \mathrm{trdeg}_{k}{\mathscr H(x)} \leq 2,$$ and $\mathrm{rank}_\Q \big( |\mathscr H(x)^\times|/|k^\times|\otimes_\Z\Q \big)\geq1$ because the valuation associated with $x$ is nontrivial. Observe that an isomorphism of non-archimedean links respects the complete residue fields, and therefore must send a point to a point of the same type.
Recall that $\operatorname{NL}(X,0)$ is locally isomorphic to a $k((t))$-analytic curve. Under any such isomorphism the type of a point as defined above coincides with its type as defined by Berkovich. For a definition of types of points in Berkovich curves in terms of their valuative invariants see e.g. [@ducros:structurecourbesanalytiques 3.3.2].
Let $(Y,D)$ be a modification of $(X,0)$ and let $x$ be a point of $\operatorname{NL}(X,0)$. The existence of a morphism $\overline\alpha\colon \operatorname{Spec}\big( \mathscr H(x)^\circ \big) \to Y$ associated with $x$ implies that the residue field $\widetilde{\mathscr H(x)}$ of $\mathscr H(x)$ is a field extension of the schematic residue field of $Y$ at $\mathrm{c}_Y(x)$. It follows that $\operatorname{c}_Y(x)$ is a closed point of $Y$ if $x$ is not divisorial. Proposition \[prop:topo\_NL\] implies then that such a point $x$ has a basis of neighborhoods of the form $\operatorname{c}_Y^{-1}\big({\mathrm c_Y(x)}\big)$, for $(Y,D)$ ranging over all the modifications of $(X,0)$.
\[rmk:divisorial\] Let $(Y,D)$ be a modification of $(X,0)$ and let $x$ be a point of $\operatorname{NL}(X,0)$. If $x$ is divisorial, then $\mathrm{c}_Y(x)$ is either a closed point of $Y$ or the generic point of an irreducible curve in $Y$. Moreover, it is always possible to find a modification $(Y',D')$ of $(X,0)$ that dominates $(Y,D)$ and such that $\mathrm{c}_{Y'}(x)$ is the generic point of an irreducible curve $E$ in $Y'$, which explains our terminology. Furthermore, the residue field $\widetilde{\mathscr H(x)}$ of ${\mathscr H(x)}$ is then isomorphic to the function field $k(E)$ of $E$.
Formal modifications and fibers of the center maps
--------------------------------------------------
Let $X$ be a normal $k$-surface and let $0$ be a closed point of $X$. We start by fixing some notation.
A modification $(Y,D)$ of $(X,0)$ is said to be a *resolution* of $(X,0)$ if $Y$ is regular. If moreover the irreducible components of $D$ are all non-singular, intersect transversally and no three of them meet at a point, then $(Y,D)$ is said to be a *good resolution* of $(X,0)$. Whenever $C$ is a germ of (formal) curve in $(X,0)$, then a good resolution $(Y,D)$ of $(X,0)$ is also said to be a good resolution of $C$ if the strict transform of $C$ in $Y$ meets $D$ transversally.
We mentioned that the non-archimedean link of a pair $(Y,D)$ only depends on an infinitesimal neighborhood of $D$ in $Y$. The notions above can then be slightly generalized by working in a suitable category of formal $k$-schemes. A formal $k$-scheme $\Y$ is called *special* if it is covered by formal subschemes $\Y_i$ such that each $\Y_i$ can be written as the formal completion of a $k$-scheme of finite type $Y_i$ along a closed subscheme $D_i$. We can then define the non-archimedean link $\operatorname{NL}(\Y)$ of $\Y$ by gluing the non-archimedean links $\operatorname{NL}(Y_i,D_i)$. Observe that when $\Y=\widehat{Y/D}$ is the formal completion of $Y$ along $D$, then $\operatorname{NL}(\Y)=\operatorname{NL}(Y,D)\cong\operatorname{NL}(X,0)$. We also obtain a center map $\operatorname{c}_\Y\colon \operatorname{NL}(\Y) \to \Y_0$, where the reduction $\Y_0$ of $\Y$ is a scheme of finite type over $k$ covered by the reduced schemes associated to the $D_i$.
A special formal $k$-scheme $\Y$ is called a *formal modification* of the pair $(X,0)$ if it is normal and it comes endowed with an adic morphism $f\colon\Y\to\X=\widehat{X/0}$ that induces an isomorphism of non-archimedean links $\operatorname{NL}(\Y)\stackrel{\sim}{\longrightarrow}\operatorname{NL}(X,0)$ and such that the fiber product $\Y\times_\X \{0\}$ is a Cartier divisor of $\Y$. If $(Y,D)\to(X,0)$ is a modification, then the formal completion $\Y=\widehat{Y/D}\to\X$ of $Y$ along $D$ is a formal modification of $(X,0)$. Such a formal modification $\Y$ of $(X,0)$ is said to be *algebraizable*, and a modification $(Y,D)$ of $(X,0)$ such that $\Y\to\X$ is isomorphic to $\widehat{Y/D}\to\X$ is called an *algebraization* of $\Y$. If $\Y$ is a formal modification of $(X,0)$ such that $\Y$ is regular, then by [@fantini:normspaces Proposition 7.6] $\Y$ is algebraized by a resolution of $(X,0)$. Observe that a formal modification $\Y$ of $(X,0)$ induces an isomorphism of non-archimedean links $\operatorname{NL}(\Y)\cong\operatorname{NL}(X,0)$, and therefore also a center map $\operatorname{c}_\Y\colon \operatorname{NL}(X,0) \to \Y_0$.
If $\Y$ is a formal modification of $(X,0)$, we denote by $\operatorname{Div}(\Y)$ the finite nonempty subset of $\operatorname{NL}(X,0)$ consisting of the divisorial points associated with the components of $\Y_0$. Whenever $\Y$ is algebraized by a modification $(Y,D)$, we will also denote $\operatorname{Div}(\Y)$ by $\operatorname{Div}(Y)$.
The following proposition is a simple consequence of Lemmas 7.14 and 9.3 of [@fantini:normspaces].
\[proposition\_propertiesNL\] Let $(X,0)$ be a normal surface singularity and let $\Y$ be a formal modification of $(X,0)$. Then the following properties hold:
1. \[proposition\_propertiesNL\_2\] The map $\operatorname{c}_{\Y}^{-1}$ gives a bijection between the set of closed points of $\Y_0$ and the set of connected components of $\operatorname{NL}(X,0)\setminus \operatorname{Div}(\Y)$.
2. \[proposition\_propertiesNL\_3\] Let $W$ be a connected component of $\operatorname{NL}(X,0)\setminus \operatorname{Div}(\Y)$, let $p$ be the corresponding closed point of $\Y_0$, let $\Y_p=\operatorname{Spf}\big(\widehat{\cO_{\Y,p}} \big)$ be the formal completion of $\Y$ along $p$, let ${{\mathcal{I}}}_p$ be the ideal of $\widehat{\cO_{\Y,p}}$ which defines $\Y_0$ locally around $p$, and denote by $\varphi\colon\operatorname{NL}(\Y_p)\to \operatorname{NL}(X,0)$ the map induced by the composition $\Y_p\to\Y\to\X$. Then $\varphi$ maps the zero locus $V({{\mathcal{I}}}_p)$ of ${{\mathcal{I}}}_p$ in $\operatorname{NL}(\Y,p)$ to a finite set of type 1 points of $\operatorname{NL}(X,0)$, and it induces an isomorphism $\operatorname{NL}(\Y_p)\setminus V({{\mathcal{I}}}_p) \cong W \subset \operatorname{NL}(X,0)$.
\[remark\_closure\_component\] Let $\Y$ and $W$ be as above. It follows from the first part of the proposition that the closure $\overline W$ of $W$ in $\operatorname{NL}(X,0)$ is obtained by adding to $W$ a subset of $\operatorname{Div}(\Y)$. Indeed, the union of all the connected components of $\operatorname{NL}(X,0)\setminus \operatorname{Div}(\Y)$ different from $W$, that is the complement of $W\cup \operatorname{Div}(Y)$ in $\operatorname{NL}(X,0)$, is an open subset of $\operatorname{NL}(X,0)$.
Existence of formal modifications and resolutions {#subsection_existencemodifications}
-------------------------------------------------
We will now explain how to produce formal modifications of $(X,0)$ with prescribed exceptional divisors and how to detect when such a modification is a good resolution of $(X,0)$.
\[theorem\_existence\_modifications\] Let $S$ be a finite subset of divisorial points of $\operatorname{NL}(X,0)$. Then there exists a formal modification $\Y$ of $(X,0)$ such that $\operatorname{Div}(\Y)=S$.
As it readily follows from the observation made in Remark \[rmk:divisorial\], there exists a resolution $(Y,D)$ of $(X,0)$ such that $S\subset \operatorname{Div}(Y)$. The contractibility criterion of Grauert-Artin [@artin:contractibilityalgebraicspaces] guarantees that we can contract every component of $D$ which does not correspond to an element of $S$, yielding a normal algebraic space over $k$ with a proper morphism $f$ to $X$. Indeed, the intersection matrix of the divisor that we want to contract is negative definite because the entire exceptional divisor of $Y$ can be contracted to the point $0$ in $X$. By taking the formal completion of this algebraic space along $f^{-1}(0)$ we obtain the formal modification $\Y$ that we wanted.
If we are working over the field of complex numbers, we can apply Grauert contractibility criterion [@grauert:ubermodifikationen] instead of Artin’s and obtain $\Y$ as a complex analytic space. Of course this is the same as analytifying the algebraic space given by Artin’s criterion. On the other hand, observe that if $k$ is the algebraic closure of $\mathbb F_p$ for some prime $p$ or if $(X,0)$ is a rational singularity, then the contractibility results [@artin:contractibilitycriterion 2.3, 2.9] grant that $\Y$ is an algebraic variety.
Recall that each type 1 point can be seen as the equivalence class of the order of vanishing along an irreducible germ of a formal curve on $(X,0)$. In particular, with a finite set of type 1 points of $\operatorname{NL}(X,0)$ is associated the germ of a curve on $(X,0)$.
\[theorem\_characterization\_resolutions\] Let $\Y$ be a formal modification of $(X,0)$, let $T$ be a finite set of type 1 points of $\operatorname{NL}(X,0)$ and let $C$ be the germ of curve on $(X,0)$ associated with $T$. Then $\Y$ can be algebraized by a good resolution of both $(X,0)$ and the germ $C$ if and only if each connected component $V$ of $\operatorname{NL}(X,0)\setminus \operatorname{Div}(\Y)$ has one of the following three forms:
1. $V$ is a disc and $V\cap T=\emptyset$;
2. $V$ is an annulus and $V\cap T=\emptyset$;
3. $V$ is a disc and $V\cap T$ can be taken to be its origin.
Observe that in the statement can take $T$ to be empty, so that there is no curve $C$ and we simply obtain a good resolution of $(X,0)$. The following proof follows the lines of the proof of Proposition 10.2 of [@fantini:normspaces], where more details are given.
By Proposition \[proposition\_propertiesNL\_2\] the connected components of $\operatorname{NL}(X,0)\setminus \operatorname{Div}(X)$ are the inverse images through the center map of the closed points of $\Y_0$. Let $W$ be such a component, let $p$ be a closed point of $\Y_0$, and let $\mathcal I_p$ be the image of the ideal defining $\Y_0$ in $\mathcal O_{\Y,p}$. By Cohen theorem, $\Y$ is regular at $p$ if and only if $\widehat{\mathcal O_{\Y,p}}\cong k[[x,y]]$, that is if and only if $\operatorname{NL}(\Y_p)\cong \operatorname{NL}(\mathbb A^2_k,0)$. Moreover, $p$ is a smooth point of $\Y_0$ if and only we can take $\mathcal I_p=(x)$ in the isomorphism above, while $p$ is an ordinary double point of $\Y_0$ if and only if we can take $\mathcal I_p=(xy)$. Since by Proposition \[proposition\_propertiesNL\_3\] we have a canonical isomorphism $\operatorname{c}_\Y^{-1}(p)\cong\operatorname{NL}({\Y_p})\setminus V(\mathcal I_p)$, it follows from Proposition \[proposition\_propertiesNL\_1\] that $\Y$ is a (formal) good modification of $(X,0)$ if and only if every connected component of $\operatorname{NL}(X,0)\setminus \operatorname{Div}(\Y)$ is either a disc or an annulus. Whenever this is the case, $\Y$ is algebraized by a good resolution $(Y,D)$ of $(X,0)$ by [@fantini:normspaces Proposition 7.6]. Finally, by definition $(Y,D)$ is also a good resolution of the germ $C$ if and only if the strict transform of $C$ in $Y$ meets $D$ transversally. This means that if $p$ is a point of $D$ contained in the strict transform of $C$ we can find an isomorphism $\widehat{\mathcal O_{Y,p}} \cong k[[x,y]]$ as above with $\mathcal I_p=(x)$, and the strict transform locally defined by $y$ at $p$. This corresponds precisely to the conditions on $T$ in the statement.
Structure of $\operatorname{NL}(X,0)$
-------------------------------------
The structure of the non-archimedean link $\operatorname{NL}(X,0)$ can be described using a resolution of the singularities of the pair $(X,0)$ and the results of the previous section.
\[corollary\_global\_topology\] Let $(Y,D)$ be a good resolution of $(X,0)$ and let $y$ be a closed point of $D$. If $D$ is smooth (respectively singular) at $y$ then $c_Y^{-1}(y)$ is a disc (resp. an annulus), and the boundary of $c_Y^{-1}(y)$ in $\operatorname{NL}(X,0)$ consists of one type 2 point (resp. two type 2 points). In particular, $\operatorname{NL}(X,0) \setminus \operatorname{Div}(Y)$ is a disjoint union of discs and finitely many annuli.
As above, the space $Y$ is smooth at $y$, so by Cohen theorem we can find an isomorphism $\widehat{\mathcal O_{Y,y}} \cong k[[U,V]] \cong \widehat{\mathcal O_{\mathbb A^2_k,0}}$ such that a local equation for $D$ at $y$ is either $U$ (if $y$ is a smooth point of $D$) or $UV$ (if $y$ is a double point of $D$). Since $\operatorname{NL}\big(\Y_{y}\big)\cong\operatorname{NL}\big(\mathbb A^2_k,0\big)$, Proposition \[proposition\_propertiesNL\_3\] and Lemma \[lemma\_disc\_annulus\] imply that $\operatorname{c}_Y^{-1}\big(y \big)$ is isomorphic to an open $k((t))$-analytic disc in the first case and an annulus in the second case, where the $k((t))$-analytic structure is defined by sending $t$ to a local equation of $D$ at $y$. The fact that the boundary $\operatorname{c}_Y^{-1}(y)$ consists of one or two points follows from Remark \[remark\_closure\_component\]. The last statement is now a consequence of Proposition \[proposition\_propertiesNL\_2\].
We can now deduce some results about the type 1 points of $\operatorname{NL}(X,0)$.
\[corollary\_topology\] Any type 1 point $x$ of $\operatorname{NL}(X,0)$ admits a basis of neighborhoods consisting of discs centered in $x$. Moreover, the set of type 1 points is dense in $\operatorname{NL}(X,0)$.
By Proposition \[prop:topo\_NL\] a basis of neighborhoods of $x$ consists of the open subsets of $\operatorname{NL}(X,0)$ of the form $\big\{\operatorname{c}_Y^{-1}\big({\overline{\operatorname{c}_Y(x)}}\big)\big\}$, for $(Y,D)$ ranging among the good resolutions of $(X,0)$, since the family of good resolutions is cofinal among the partially ordered set of modifications of $(X,0)$. Since $x$ is not of type $2$, $\operatorname{c}_Y(x)$ is a closed point of $D$, and we can find an isomorphism $\widehat{\mathcal O_{Y,y}} \cong k[[U,V]] \cong \widehat{\mathcal O_{\mathbb A^2_k,0}}$ such that $U$ is a local equation for $D$ at $y$ and $V$ defines the germ of curve at $y$ associated to the type one point $x$. This shows that $\operatorname{c}_Y^{-1}\big(\overline{\operatorname{c}_Y(x)}\big)=\operatorname{c}_Y^{-1}\big(\operatorname{c}_Y(x)\big)$ is an open $k((t))$-analytic disc centered in $x$.
It remains to prove the density of the set of type 1 points of $\operatorname{NL}(X,0)$. As noted in Remark \[remark\_closure\_component\], each of the divisorial points associated with a good resolution of $(X,0)$ is not isolated in $\operatorname{NL}(X,0)$. The result then follows from Corollary \[corollary\_global\_topology\], Lemma \[lemma\_disc\_annulus\], and the fact that the set of its points of type 1 is dense in a $k((t))$-analytic annulus $A=\big\{x\in \A^{1,\mathrm{an}}_{k((t))} \,\big|\, |t|_{k((t))}<|T(x)|<1\big\}$. The latter is a classical fact that can be proven directly by exhibiting suitable type 1 points; as an example, the semi-norm that sends an element $P$ of $k((t))[T]$ to $|P(x)| = |P(t^{1/2})|_{k((t))}$ is a type 1 point $x$ of $A$.
It is also true that each of the sets consisting of type 2, type 3 or type 4 points is dense in $\operatorname{NL}(X,0)$, but we will not need this fact.
Any bounded analytic function on the complement of finitely many type 1 points extends to $\operatorname{NL}(X,0)$. This follows from the fact that $\operatorname{NL}(X,0)$ is locally a $k((t))$-analytic space combined with the more general result [@berkovich:book Proposition 3.3.14]. However, we include a proof here since it is simpler to deduce the result in our very special case from Corollary \[corollary\_topology\].
\[lemma\_extension\_to\_type1\] Let $U$ be an open subset of $\operatorname{NL}(X,0)$ and let $S$ be a finite subset of $U$ consisting of type $1$ points. Then the inclusion $U\setminus S\hookrightarrow U$ induces an isomorphism $$\mathcal O_{\operatorname{NL}(X,0)}^\circ \big( U \big) \cong \mathcal O_{\operatorname{NL}(X,0)}^\circ \big( U\setminus S \big).$$
Since $\mathcal O^\circ_{NL(X,0)}$ is a sheaf it is enough to prove the result for $T=\{x\}$ a single type $1$ point, and $U$ some neighborhood of $x$ in $\operatorname{NL}(X,0)$. Hence, by Corollary \[corollary\_topology\] we can assume without loss of generality that $U$ is a disc centered in $x$. Then the ring of bounded analytic functions on $U\setminus\{x\}$ is isomorphic to $k((t))[[T]]$, where $T$ is a coordinate function on $U$, and all such functions extend to $x=V(T)$.
The results of this section can also be used to deduce a characterization of the *essential valuations* of a surface singularities, as appearing in the Nash problem (see [@fantini:normspaces Theorems 10.4 and 10.8]), and to give another proof of the existence of resolutions for surfaces (see [@FantiniTurchetti2017 Theorem 8.6]).
Self-similarity of sandwiched singularities {#section_preliminariessandwiched}
===========================================
In this section we introduce sandwiched singularities and prove the implications \[condition\_sandwiched\] $\implies$ \[condition\_strongly\_selfsim\] $\implies$ \[condition\_valtree\] $\implies$ \[condition\_selfsim\] of Theorem \[mainthm\].
Sandwiched singularities {#ssec:sandwich}
------------------------
\[def-sandwich\] Let $\mathcal O$ be a normal 2-dimensional complete local ring with algebraically closed residue field $k$. We say that $\mathcal O$ is *sandwiched* if there exist two algebraic surfaces $X_0$ and $Y$ over $k$, with $X_0$ smooth over $k$ and $Y$ normal, a proper birational morphism $Y\to X_0$, and a point $y$ in $Y$ such that $\mathcal O\cong\widehat{\mathcal O_{Y,y}}$. If $X$ is an algebraic surface over $k$ and $0$ is a normal point of $X$, we say that $(X,0)$ is a sandwiched singularity if the complete local ring $\widehat{\mathcal O_{X,0}}$ of $X$ in $0$ is sandwiched.
Note that when working over $\C$ our definition is equivalent to [@spivakovsky:sandsingdesingsurfNashtransf Definition II.1.1] thanks to Artin’s approximation theorem [@artin:solaneq Corollary 1.6].
Fix any smooth algebraic surface $X_0$ over $k$, such as for example $X_0 = \A^2_k$. Let $p$ be a point of $X_0$, let $\varphi\colon Y\to X_0$ be a sequence of point blowups centered above $p$, and let $E$ be a connected divisor on $Y$ obtained by removing some irreducible components from $\varphi^{-1}(p)$. Since the divisor $\varphi^{-1}(p)$ can be contracted to $(X_0,p)$ and $X_0$ is smooth, Artin’s contractibility criterion [@artin:contractibilitycriterion Theorem 2.3] ensures that the divisor $E$ can be contracted to a point $0$ in a normal algebraic variety $X$. The normal singularity $(X,0)$ is sandwiched, and any sandwiched surface singularities is isomorphic étale locally (or analytically, if $k=\C$) to such a singularity.
\[rmk:sandwichoverC\] Suppose $(X,0)$ is sandwiched. Then in the category of analytic spaces over $k$, or in the complex analytic category when working over $\mathbb C$, one can always find a morphism $X\to X_0$ to a smooth surface. If $\mu\colon Y\to X$ is a resolution of $(X,0)$, we get back $(X,0)$ by contracting the divisor $\mu^{-1}(0)$ of $Y$. This explains the terminology, as $X$ is sandwiched between the smooth surfaces $X_0$ and $Y$.
We start by proving a stronger form of the implication \[condition\_sandwiched\] $\implies$ \[condition\_valtree\].
\[lem:structure sandw\] Suppose $(X,p)$ is a sandwiched singularity and $y$ is a point of $\operatorname{NL}(\A^2_k,0)$. Then there exists a finite set $T$ of type 1 points in $\operatorname{NL}(X,p)$ such that $\operatorname{NL}(X,p)\setminus T$ is isomorphic to an open subset of $\operatorname{NL}(\A^2_k,0)$ containing $y$.
We may assume that $\cO_{X,p}$ is not a regular ring.
Recall that $\operatorname{NL}(X,p)$ only depends on the formal completion of its local ring. We may thus assume that there exist a proper birational map from a smooth algebraic variety $Y$ to the affine plane $\varphi \colon Y \to \A^2_k$ which is an isomorphism outside $\varphi^{-1}(0)$, and a connected divisor $E \subset \varphi^{-1}(0)$ such that $X$ is the surface obtained from $Y$ by contracting $E$ to a point. We denote by $\mu \colon Y \to X$ the contraction map so that $p= \mu(E)$. Note that $\varphi$ factors through $\mu$, so there exists a regular birational map $\pi\colon X \to \A^2_k$ mapping $p$ to $0$.
It follows from [@spivakovsky:sandsingdesingsurfNashtransf Corollary 1.14] that we may further impose that any irreducible component of $\varphi^{-1}(0)$ that is not included in $E$ has self-intersection $-1$. Since $Y$ is regular, $\varphi$ is a sequence of point blow-ups hence factors through the blow-up of the origin $\varpi\colon Y_1 \to \A^2_k$. The map $Y \to Y_1$ is not an isomorphism, therefore the strict transform $E_0$ of $\varpi^{-1}(0)$ in $Y$ has self-intersection at most $-2$, hence is included in $E$.
Let $T$ be the zero locus in $\operatorname{NL}(X,0)$ of the ideal defining $\pi^{-1}(0)$ around $p$ and let $\Y$ be the formal completion of $X$ along $\pi^{-1}(0)$. By Proposition \[proposition\_propertiesNL\_3\] $T$ is a finite set of type 1 points of $\operatorname{NL}(X,0)$, and $\pi$ induces an isomorphism between $\operatorname{NL}(X,p) \setminus T$ and the open subset $U$ of $\operatorname{NL}(\A^2_k,0)$ consisting of the points whose center in $X$ is equal to $p$. Note that, by the commutativity of the center maps, $U$ is also equal to the set of points of $NL(X,p)$ whose center on $Y$ is contained in $E=\mu^{-1}(p)$.
If $U$ contains $y$ there is nothing left to prove. If this is not the case, observe that the linear group $\operatorname{GL}(2,k)$ acts naturally on $\operatorname{NL}(\A^2_k,0)$ in such a way that any element $g\in \operatorname{GL}(2,k)$ defines an isomorphism of non-archimedean links $g_\bullet \colon \operatorname{NL}(\A^2_k,0) \to \operatorname{NL}(\A^2_k,0) $. The linear map $g$ also lifts to an automorphism $L_g \colon Y_1\to Y_1 $ whose action on the exceptional divisor $\varpi^{-1}(0)$ is given by the projectivization of $g$, and satisfies $\mathrm{c}_{Y_1} (g_\bullet (y)) = L_g( \mathrm{c}_{Y_1}(y))$. We may thus find $g\in \operatorname{GL}(2,k)$ such that $ L_g( \mathrm{c}_{Y_1}(y))$, and hence $\mathrm{c}_{Y_1} (g_\bullet (y))$, does not belong to the indeterminacy locus of the birational map $ \varphi^{-1} \circ \varpi \colon Y_1 \to Y$. In particular, the single irreducible component of $\varphi^{-1}(0)$ containing the point $\mathrm{c}_{Y} (g_\bullet (y))$ is $E_0$, so that $\mathrm{c}_{Y}(g_\bullet (y))\in E$, and $g_\bullet (y)$ belongs to $U$.
It follows that the open subset $g_\bullet^{-1} (U)$ in $\operatorname{NL}(\A^2_k,0)$, which is isomorphic to $\operatorname{NL}(X,p) \setminus T$ because $U$ is and $g_\bullet$ is an isomorphism, contains $y$ as required. This concludes the proof of the Lemma.
The implication \[condition\_sandwiched\] $\implies$ \[condition\_strongly\_selfsim\]
-------------------------------------------------------------------------------------
Pick any point $x\in \operatorname{NL}(X,0)$ which is not of type 2 and an open set $U$ containing $x$. By Proposition \[prop:topo\_NL\] one may choose a good resolution $Y \to X$ and a point $p$ on its exceptional divisor such that $x\in \mathrm{c}_Y^{-1}(p)\subset U$.
Proposition \[proposition\_propertiesNL\] (ii) implies that the good resolution $Y \to X$ induces an isomorphism $\mathrm{c}_Y^{-1}(p) \cong \operatorname{NL}(\Y_p) \setminus V(\mathcal{J}_p)$. Observe that the latter non-archimedean link is isomorphic to $\operatorname{NL}(\A^2_k,0) \setminus S$ where $S$ is a set of type 1 points of cardinality $1$ or $2$ depending on whether $p$ is a smooth point of the exceptional divisor or not. Denote by $y$ the image of $x$ under this isomorphism.
By Lemma \[lem:structure sandw\] there exists a finite set $T$ of type 1 points and an isomorphism of non-archimedean links $\psi$ between $\operatorname{NL}(X,0)\setminus T$ and an open subset of $\operatorname{NL}(\A^2_k,0)$ containing $y$. Adding $\psi^{-1}(S)$ to $T$ if necessary we may suppose that $\psi(\operatorname{NL}(X,0)\setminus T) \subset \operatorname{NL}(\A^2_k,0)\setminus S$.
We conclude observing that $\varphi^{-1} \circ \psi$ is an isomorphism mapping $\operatorname{NL}(X,0)\setminus T$ to $\mathrm{c}_Y^{-1}(p)$ and containing $x$. $\qed$
The implication \[condition\_strongly\_selfsim\] $\implies$ \[condition\_valtree\]
----------------------------------------------------------------------------------
By Corollary \[corollary\_topology\], one can find a disc $D$ in $\operatorname{NL}(X,0)$. By \[condition\_strongly\_selfsim\], there exists a finite set $T$ of type 1 points and an open subset of $D$ which is isomorphic to $\operatorname{NL}(X,0)\setminus T$. By Lemma \[lemma\_disc\_annulus\], $D$ can be realized as an open subset of $\operatorname{NL}(\A^2_k,0)$. $\qed$
The implication \[condition\_valtree\] $\implies$ \[condition\_selfsim\]
------------------------------------------------------------------------
Let $V$ be any open subset of $\operatorname{NL}(X,0)$. By Corollary \[corollary\_topology\], $V$ contains a point of type 1, and therefore it contains a disc $D$ by the same corollary. Now fix a finite set $T$ of type 1 points in $\operatorname{NL}(X,0)$ such that condition \[condition\_valtree\] holds. Adding one more point to $T$ if necessary, one may assume that $\operatorname{NL}(X,0)\setminus T$ is isomorphic to an open subset of $\operatorname{NL}(\A^2_k,0)\setminus V(t)$, which is a disc by Lemma \[lemma\_disc\_annulus\]. We conclude that $\operatorname{NL}(X,0)\setminus T$ can be realized as an open subset of $D\subset V$. $\qed$
Self-similar graphs {#section_graphs}
===================
In this section we introduce the notions of self-similar and sandwiched graphs and prove that these notions are in fact equivalent.
Modifications of graphs
-----------------------
Let us introduce some terminology first. A *graph* $\Gamma$ is the data of a finite set of vertices $V= V(\Gamma)$ and a subset $E= E(\Gamma)$ of pairs in $V$ (the set of edges). In particular our graph has no loop. Two vertices are said to be connected (or joined) by an edge when ${{v_1\!-\!\!\!-v_2}}\in E$. In that case, we write $v_1 \sim v_2$. A graph is connected if for any two vertices $v_0, v_1$ there exists a sequence of edges ${{v_0\!-\!\!\!-w_1}}, {{w_1\!-\!\!\!-w_2}}, \ldots, {{w_n\!-\!\!\!-v_1}}$ joining them. A *weighted graph* is a graph together with a function $V\to \N\times \N^*$. We shall write this function as $v\mapsto (g(v),e(v))$.
The *link* of a vertex $v$ of a graph $\Gamma$ is by definition the set $L(v)$ of vertices $w$ such that ${{v\!-\!\!\!-w}}$ is an edge. The cardinality of $L(v)$ is the *valency* of $v$.
In order to motivate our further definitions, let us indicate in which situation graphs will arise in the sequel. We shall consider the dual graph of a resolution of a normal surface singularity, so that vertices are in bijection with exceptional components of this resolution. The weight function is then given by genus and the opposite of the self-intersection of a component.
If $\Gamma$ is a graph, a *simple modification of* $\Gamma$ *centered at a vertex* $v$ is a graph $\Gamma'$ such that its set of vertices is $V'= V \cup\{v'\}$, its set of edges is $E' = E \cup \big\{ {{v\!-\!\!\!-v'}} \big\}$, and its weight function $(g',e')$ satisfies $g'(w) = g(w)$ and $e'(w) = e(w)$ for all $w\in V\setminus\{v\}$, $g'(v) = g(v)$, $g'(v') =0$, $e'(v) = e(v) +1$, and $e(v')=1$.
A *simple modification of* $\Gamma$ *centered at an edge* ${{v_0\!-\!\!\!-v_1}}\in E$ is a graph $\Gamma'$ such that its set of vertices is $V'= V \cup\{v'\}$, its set of edges is $ E' = \big(E \setminus \big\{ {{v_0\!-\!\!\!-v_1}}\big\}\big) \cup \big\{{{v_0\!-\!\!\!-v'}}, {{v_1\!-\!\!\!-v'}}\big\}$, and its weight function $(g',e')$ satisfies $g'(w) = g(w)$ for all $w\in V$, $g(v') =0$, $e'(w) = e(w)$ for all $w\in V \setminus \{v_0,v_1\}$, $e'(v_0) = e(v_0) +1$, $e'(v_1) = e(v_1) +1$, and $e(v')=1$.
A *modification* of a graph $\Gamma$ is a graph $\Gamma'$ which is obtained from $\Gamma$ by a finite sequence of simple modifications (centered either at a vertex or at an edge). A modification is nontrivial whenever it is not an isomorphism. We shall write $\Gamma' \rightsquigarrow \Gamma$ to say that $\Gamma'$ is a modification of $\Gamma$. Note the following transitivity property: if $\Gamma''\rightsquigarrow \Gamma'$ and $\Gamma' \rightsquigarrow \Gamma$ are modifications, then $\Gamma'' \rightsquigarrow \Gamma$ is also a modification.
Observe that a modification $\Gamma' \rightsquigarrow \Gamma$ yields a canonical inclusion $\imath\colon V(\Gamma) \to V(\Gamma')$ of the set of vertices of $\Gamma$ into $\Gamma'$. The image of a vertex $v$ by $\imath$ is called its *strict transform*. Note that in general two vertices in $V$ joined by an edge need not have strict transforms joined by an edge in $\Gamma'$.
We can also define the *total transform* of a connected subgraph $\Delta$ of $\Gamma$ by a modification $\Gamma' \rightsquigarrow \Gamma$. It is sufficient to explain the construction of the total transform in the case of a simple modification. If the center of the modification is a vertex not belonging to $\Delta$ or an edge ${{v\!-\!\!\!-w}}$ with $v,w\notin \Delta$, then the total transform of $\Delta$ is the copy of $\Delta$ in $\Gamma'$ whose vertices are the strict transform of the vertices of $\Delta$. If the center is a vertex of $\Delta$, then its total transform is obtained from $\Delta$ by adding the new edge and the new vertex. If the center is an edge ${{v\!-\!\!\!-w}}$ of $\Delta$, then its total transform is the graph whose vertices are the vertices of $\Delta$ together with the new vertex, and edges are edges of $\Delta$ different from ${{v\!-\!\!\!-w}}$ together with the two new edges. Finally, if the center is an edge ${{v\!-\!\!\!-w}}$ with $v\in \Delta$ and $w\notin \Delta$, then the vertices of the total transform are the vertices of $\Delta$ together with the new vertex, and edges are edges of $\Delta$ together with the only new edge that contains $v$. It is not difficult to see that the total transform of a connected subgraph of $\Gamma$ is a connected subgraph of $\Gamma'$.
We also introduce the notion of *embedding* of a (weighted) graph of $\Gamma$ into another one $\Gamma'$. This is an injective map $\varphi\colon V(\Gamma) \to V(\Gamma')$ such that $v \sim w$ implies $\varphi(v) \sim \varphi(w)$, $g'(\varphi(v))= g(v)$ and $e'(\varphi(v)) =e(v)$. We shall write $\varphi \colon \Gamma \hookrightarrow \Gamma'$, and denote by $\varphi(\Gamma)$ the subgraph of $\Gamma'$ whose vertices are $\varphi(v)$ with $v\in V(\Gamma)$ and edges ${{\varphi(v)\!-\!\!\!-\varphi(w)}}$ with ${{v\!-\!\!\!-w}} \in E(\Gamma)$. An *isomorphism* is an embedding such that the induced maps on vertices and edges are bijective.
Suppose we are given two modifications $\Gamma' \rightsquigarrow \Gamma$ and $\Delta' \rightsquigarrow \Delta$ and two embeddings $\varphi \colon \Delta \hookrightarrow \Gamma$, $\varphi' \colon \Delta' \hookrightarrow \Gamma'$. Then we say that the pair $(\Gamma' \rightsquigarrow \Gamma, \Delta' \rightsquigarrow \Delta)$ is $(\varphi', \varphi)$-*compatible* whenever there exist simple modifications $\Gamma_{i+1} \rightsquigarrow \Gamma_i$ and embeddings $\varphi_i: \Delta_i \hookrightarrow \Gamma_i$ such that $\Gamma_0 = \Gamma$, $\Gamma_l = \Gamma'$, $\Delta_0 = \Delta$, $\Delta_l = \Delta'$, $\varphi_0 = \varphi$, $\varphi_l = \varphi'$, and the following holds.
1. \[item:compatible\_rules\_vin\] When $\Gamma_{i+1} \rightsquigarrow \Gamma_i$ is centered at a vertex $\varphi_i(v)$ for some $v\in \Delta_i$, then $\Delta_{i+1}$ is obtained from $\Delta_i$ by the simple modification centered at $v$.
2. When $\Gamma_{i+1} \rightsquigarrow \Gamma_i$ is centered at a vertex $v\notin \varphi_i(\Delta_i)$, then $\Delta_{i+1} = \Delta_i$.
3. When $\Gamma_{i+1} \rightsquigarrow \Gamma_i$ is centered at an edge ${{\varphi_i(v)\!-\!\!\!-\varphi_i(w)}}$ for some edge ${{v\!-\!\!\!-w}}$ in $\Delta_i$, then $\Delta_{i+1}$ is obtained from $\Delta_i$ by the simple modification centered at ${{v\!-\!\!\!-w}}$.
4. When $\Gamma_{i+1} \rightsquigarrow \Gamma_i$ is centered at an edge ${{v\!-\!\!\!-\varphi_i(w)}}$ for some $v \notin \Delta_i$ and $w\in\Delta_i$, then $\Delta_{i+1}$ is obtained from $\Delta_i$ by the simple modification centered at $w$.
5. \[item:compatible\_rules\_eout\] When $\Gamma_{i+1} \rightsquigarrow \Gamma_i$ is centered at an edge ${{v\!-\!\!\!-w}}$ for some $v,w \notin \Delta_i$, then $\Delta_{i+1} = \Delta_i$.
\[prop-modif\] Suppose the pair $(\Gamma' \rightsquigarrow \Gamma, \Delta' \rightsquigarrow \Delta)$ is $(\varphi', \varphi)$-compatible, where $\Gamma' \rightsquigarrow \Gamma$ and $\Delta' \rightsquigarrow \Delta$ are modifications, and $\varphi \colon \Delta \hookrightarrow \Gamma$, $\varphi' \colon \Delta' \hookrightarrow \Gamma'$ are embeddings.
1. \[item:prop-modif-1\] The image by $\varphi'$ of the total transform in $\Delta'$ of any subgraph $G\subset \Delta$ is the total transform of $\varphi(G)$. In particular, the total transform of $\varphi(\Delta)$ is $\varphi'(\Delta')$.
2. \[item:prop-modif-2\] If $\imath_\Delta$ and $\imath_{\Gamma}$ denote the strict transform maps induced by the modifications, then $ \varphi' \circ \imath_\Delta = \imath_{\Gamma} \circ \varphi $.
It is only necessary to check these properties when $\Gamma' \rightsquigarrow \Gamma$ is a simple modification in which case $\Delta' \rightsquigarrow \Delta$ is described by one of the rules \[item:compatible\_rules\_vin\]-\[item:compatible\_rules\_eout\] above. It is a routine argument to verify the proposition in each of these cases.
\[prop-induction\] Suppose $\Delta' \rightsquigarrow \Delta$ is a modification, and let $\varphi \colon\Delta\hookrightarrow\Gamma$ be any embedding.
- Then there exists a modification $\Gamma' \rightsquigarrow \Gamma$, called the *induced modification*, and an embedding $\varphi' \colon\Delta'\hookrightarrow\Gamma'$ such that the pair $(\Gamma' \rightsquigarrow \Gamma, \Delta' \rightsquigarrow \Delta)$ is $(\varphi', \varphi)$-compatible.
- Suppose we are given another embedding $\psi\colon \Gamma \hookrightarrow \widetilde{\Gamma}$. Then the two modifications of $ \widetilde{\Gamma}$ induced by $\psi$ or $\psi \circ \varphi$ are isomorphic. We denote either of them by $\widetilde{\Gamma}'\rightsquigarrow\widetilde{\Gamma}$. The embedding $\Delta'\hookrightarrow\widetilde{\Gamma}'$ is the composition of the embeddings $\Delta'\hookrightarrow\Gamma'$ and $\Gamma'\hookrightarrow\widetilde{\Gamma}'$.
Decompose $\Delta' \rightsquigarrow \Delta$ into simple modifications $\Delta_{i+1} \to \Delta_i$, such that $\Delta_0 = \Delta$, $\Delta_l = \Delta'$. To simplify notation we assume $\Delta_0 \subset \Gamma_0 := \Gamma$. We define by induction a graph $\Gamma_i$ that contains $\Delta_i$ as follows. If $\Delta_{i+1}$ is the modification centered at an edge ${{v\!-\!\!\!-w}}$ with $v,w \in \Delta_i$, then set $\Gamma_i$ to be the modification centered at ${{v\!-\!\!\!-w}}$. If $\Delta_{i+1}$ is the modification centered at a vertex $v\in \Delta_i$, then we set $\Gamma_i$ to be the modification centered at $v$. It is clear from the definitions that we obtain compatible modifications in this way. The second point is proven in the same way; details are left to the reader.
A graph $\Gamma'$ satisfying the first point of Proposition \[prop-induction\] is in general not unique. When $\Delta_{i+1}$ is the modification centered at a vertex $v\in \Delta_i$ that is connected to a vertex $w\notin \Delta_i$, we could also have defined $\Gamma_{i+1}$ to be the modification centered at the edge ${{v\!-\!\!\!-w}}$. Note however that the construction given in the proof of the previous proposition yields a graph $\Gamma'$ that is minimal in the sense that the sum of the weights at all its vertices is minimal (among all possible graphs).
Characterization of self-similar graphs
---------------------------------------
\[def:graph-ss\] A connected graph is said to be *regular* if it is a modification of the graph with one vertex and weight $(0,1)$. It is said to be *sandwiched* if it can be embedded into a regular graph.
A connected graph $\Gamma$ is said to be *self-similar* if it admits a nontrivial modification $\Gamma'$ that contains a subgraph $\Gamma_0$ that is isomorphic (as a weighted graph) to $\Gamma$.
\[thm:graph-sdw\] Suppose $\Gamma$ is a connected weighted graph. If $\Gamma$ is self-similar then it is sandwiched.
In fact the reverse implication is also true. We leave it as an exercise to the reader since we shall not need it in the sequel.
We fix a modification $\Gamma'$ of $\Gamma$, and an embedding $\varphi\colon \Gamma \hookrightarrow \Gamma'$. Recall that the strict transform yields an inclusion $\imath\colon V(\Gamma) \to V(\Gamma')$.
[**Step 1**]{}. There exists a sequence of graphs $\Gamma_n$ with $\Gamma_0 := \Gamma$, and $\Gamma_1 := \Gamma'$, such that
1. $\Gamma_{n+1}$ is a modification of $\Gamma_n$;
2. there exist embeddings $\varphi_n\colon \Gamma_n \hookrightarrow \Gamma_{n+1}$;
3. \[item:step1compatibility\] the pair $(\Gamma_{n+1}\rightsquigarrow \Gamma_n,\Gamma_{n}\rightsquigarrow \Gamma_{n-1})$ is $(\varphi_n, \varphi_{n-1})$-compatible.
Since $\Gamma'\rightsquigarrow \Gamma$ is a modification and $\Gamma$ is a subgraph of $\Gamma'$ via the embedding $\varphi$, Proposition \[prop-induction\] gives a modification $\Gamma_2\rightsquigarrow \Gamma'$ and an embedding $\varphi_1\colon \Gamma' \hookrightarrow \Gamma_2$ with the right compatibility properties.
To build the sequence $\Gamma_n$ we proceed in the same way and use repeteadly Proposition \[prop-induction\]. More precisely suppose $\Gamma_{n}$ has been defined. For each $n$ there exists an embedding $\Gamma \hookrightarrow \Gamma_n$ obtained by the map $\Phi_n := \varphi_{n-1} \circ \ldots \circ \varphi_0$ (with $\varphi = \varphi_0$). Then we construct $\Gamma_{n+1}$ from this embedding and the modification $\Gamma'\rightsquigarrow \Gamma$ using Proposition \[prop-induction\].
Observe that $\Gamma_{n}$ is obtained by the modification $\Gamma'\rightsquigarrow \Gamma$ using $\Phi_{n-1}$ whereas $\Gamma_{n+1}$ is obtained by the modification $\Gamma'\rightsquigarrow \Gamma$ using $\varphi_{n-1} \circ \Phi_{n-1}$. This implies the existence of an embedding $\varphi_n \colon \Gamma_n \to \Gamma_{n+1}$ satisfying the compatibility property (c) (see the second point of Proposition \[prop-induction\]).
[**Step 2**]{}. Suppose that there exists a vertex $v\in V(\Gamma)$ such that $\imath(v) = \varphi(v)$. We claim that the modification is trivial.
To see this define the edge distance on $V(\Gamma)$ by setting $d(x, y)$ to be the least integer $n \in \N$ such that there exists $v_0, \ldots, v_n \in V(\Gamma)$ with the property $v_0 =x$, $v_n=y$ and ${{v_0\!-\!\!\!-v_1}}, \ldots, {{v_{n-1}\!-\!\!\!-v_n}} \in E(\Gamma)$.
We first prove by induction on $n:= d(w, v)$, that there exists an integer $k\ge1$ such that $I_k(w) = \Phi_k(w)$ for any $w\in V(\Gamma)$ at distance at most $n$ to $v$ where $I_k$ is the strict transform map induced by the modification $\Gamma_k \rightsquigarrow \Gamma$, and $\Phi_k := \varphi_{k-1} \circ \ldots \circ \varphi_0$.
Our claim for $n=0$ is our standing assumption (with $k=1$). Assume the induction hypothesis for vertices at distance $n-1$ to $v$. Pick $w$ at distance $n$ to $v$. By definition there exists (at least one) $w_0\in V$ at distance $n-1$ to $v$ and $1$ to $w$. Replacing $\Gamma_k$ by $\Gamma'$ we may suppose that $\imath(w_0) = \varphi(w_0)$. In particular we have $e(\imath(w_0)) = e(w_0)$.
Decompose the modification from $\Gamma' \rightsquigarrow\Gamma'$ into a sequence of simple modifications $\Delta_{i+1} \to \Delta_i$ with $\Delta_0:=\Gamma$, and $\Gamma':= \Delta_l$. The equality of weights $e(\imath(w_0)) = e(w_0)$ implies that the center of all simple modifications $\Delta_{i+1} \to \Delta_i$ cannot be an edge that contains the strict transform of $w_0$ in $\Delta_i$ or a vertex that is joined to this strict transform by an edge. It follows that the strict transform of any vertex at distance $1$ from $ w_0$ in $\Gamma'$ is again joined to $\imath(w_0)$ by an edge, and no new edge starting from $\imath(w_0)$ are created. In particular, $\imath$ induces a bijection from the link $L(w_0)$ to $L(\imath(w_0))$, and the valency of $w_0$ is equal to the valency of $\imath(w_0)$. Since $\varphi$ is an isomorphism from $\Gamma$ onto its image, $\varphi(L(w_0))$ is included in $L(\varphi(w_0))$. The two sets having the same cardinality, we conclude that $\varphi$ is a bijection (hence an isomorphism) from $L(w_0)$ to $L(\varphi(w_0))=\varphi(L(w_0))$. Denote by $f := \imath^{-1} \circ \varphi$ the bijection on $L(w_0)$.
Observe that the quantity $\sum_{v\in L(w_0)} e(v)$ is equal to $\sum_{v\in L(\varphi(w_0))} e(v)$. Since this quantity can only increase along the sequence of simple modifications $\Delta_{i+1} \to \Delta_i$ it remains constant. This implies that the total transform of $L(w_0)$ is equal to $L(\varphi(w_0))$.
By Proposition \[prop-modif\] \[item:prop-modif-1\] and Property \[item:step1compatibility\] of Step 1, it follows that the total transform of $L(\varphi(w_0)) = \varphi(L(w_0))$ by $\Gamma_2\rightsquigarrow \Gamma_1$ is also equal to $\varphi_1( \varphi(L(w_0)))$. In other words, the total transform of $L(w_0)$ by the composite modification $\Gamma_2 \rightsquigarrow \Gamma$ is isomorphic to itself. In particular the injective map $I_2$ induced by this strict transform defines a bijection from $L(w_0)$ to $\varphi_1( \varphi(L(w_0)))$, and we get as before a bijection $f_2\colon L(w_0) \to L(w_0)$.
Denote by $\imath_1$ the strict transform map induced by $\Gamma_2 \rightsquigarrow \Gamma$ so that $f_2 = \imath^{-1} \circ \imath_1^{-1}\circ \varphi_1 \circ \varphi$. By Proposition \[prop-modif\] \[item:prop-modif-2\], we have $\varphi_1 \circ \imath= \imath_1\circ \varphi$, whence $f_2 = \imath^{-1} \circ \varphi\circ \imath^{-1} \circ \varphi = f \circ f$.
By repeating this argument, we see that the strict tranform of $L(w_0)$ by the modification $\Gamma_k \rightsquigarrow \Gamma$ is equal to $\Phi_k(L(w_0))$ and that $I_k^{-1} \circ \Phi_k = f^{\circ k}$. Since $L(w_0)$ is finite, it follows that $f^{\circ k} = \operatorname{id}$ for some $k$. Observe that we may choose $k$ to be the least common multiple of all integers less than the cardinality of $\Gamma$, which is then independent of $w_0$. This proves our claim.
We now explain how this claim implies Step 2. Replacing $\Gamma'$ by some $\Gamma_k$ for $k$ sufficiently large, we have $\imath(w) = \varphi(w)$ for all $w$. Since $\varphi$ preserves the weights, we get $e(\imath(w)) = e(w)$ for all $w$ which is only possible when the modification is trivial.
[**Step 3**]{}. If $\Gamma$ is self-similar, then there exists an integer $k\ge 1$ such that for all $v\in V(\Gamma)$ one has $I_k(v) \notin \Phi_k(V(\Gamma))$.
We first prove that if $I_k(v) \notin \Phi_k(V(\Gamma))$ then $I_{k+l}(v) \notin \Phi_{k+l}(V(\Gamma))$ for all $l\ge 0$. Indeed suppose $I_k(v) \notin \Phi_k(V(\Gamma))$, and recall from Step 1 that $\Gamma_{k+1}$ is built from the embedding $\Phi_k \colon \Gamma \hookrightarrow \Gamma_k$ and the modification $\Gamma' \rightsquigarrow \Gamma$. It follows that the strict tranform $I_{k+1}(v)$ of $I_k(v)$ in $\Gamma_{k+1}$ is not contained in the image of $\Gamma'$ in $\Gamma_{k+1}$ which contains the image of $\Gamma$ in $\Gamma_{k+1}$. Therefore $I_{k+1}(v) \notin \Phi_{k+1}(V(\Gamma))$ which proves $I_{k+l}(v) \notin \Phi_{k+l}(V(\Gamma))$ for all $l\ge 0$ by induction as required.
It is thus sufficient to find for any vertex $v$ of $\Gamma$ an integer $k$ such that $I_k(v) \notin \Phi_k(V(\Gamma))$. To prove this we proceed by contradiction, and pick a vertex $v$ of $\Gamma$ for which $I_k(v) \in \Phi_k(V(\Gamma))$ for all $k$. Since $\Phi_k$ is an embedding, for each $k$ there exists a unique vertex $v_k$ of $\Gamma$ such that $I_k(v) = \Phi_k(v_k)$. Choose $k, l >0$ such that $v_{k+l} = v_k$.
Observe that we have a modification $\Gamma_{k+l}\rightsquigarrow \Gamma_k$ and an embedding $\Phi \colon\Gamma_k \hookrightarrow\Gamma_{k+l} $ obtained by composing $\varphi_{k+l-1} \circ \ldots \circ \varphi_k$. Denote by $I$ the strict transform map induced by $\Gamma_{k+l}\rightsquigarrow \Gamma_k$, and let $w := \Phi_k(v_k) = \Phi_k(v_{k+l}) = I_k(v)$. Then we get $$I(w) = \imath_{k+l-1} \circ \ldots \circ \imath_k (I_k(v)) = I_{k+l}(v) = \Phi_{k+l}(v_{k+l}) = \Phi (\Phi_k(v_{k+l})) = \Phi(w)~.$$ Step 2 then implies that $\Gamma_{k+l}\rightsquigarrow \Gamma_k$ is an isomorphism which is only possible if $\Gamma'\rightsquigarrow \Gamma$ is also an isomorphism. This is a contradiction.
[**Step 4**]{}. Finally we prove that $\Gamma$ being self-similar implies the graph to be sandwiched.
By Step 3, replacing $\Gamma'$ by $\Gamma_k$ for a sufficiently large $k$ we may assume that $\imath(v) \notin \varphi(V(\Gamma))$ for all vertices $v$ of $\Gamma$. It remains to prove that this implies the graph to be sandwiched.
This is a consequence of the following general fact. Suppose $\Gamma' \rightsquigarrow \Gamma$ is a modification. Look at the (not necessarily connected) graph $\widehat{\Gamma'}$ obtained by removing the strict transforms of all vertices in $\Gamma$ and all edges connected to any of these.
\[lem:whaou\] The graph $\widehat{\Gamma'}$ is a union of regular graphs.
Now $\varphi(\Gamma)$ is a subgraph that does not contain any of the strict transform $\imath(v)$ for $v\in \Gamma$, hence is a subgraph of $\widehat{\Gamma'}$. In particular, $\varphi(\Gamma)$ (hence $\Gamma$) is sandwiched.
We argue by induction on the number $k$ of simple modifications needed to define $\Gamma' \rightsquigarrow \Gamma$. When $k=1$ then $\widehat{\Gamma'}$ is reduced to the sole vertex added to $\Gamma$ and by definition its weight is $(0,1)$, hence is regular. Suppose the lemma is proved for $k-1$, and suppose $\Gamma' \rightsquigarrow \Gamma$ is decomposed into $k$ simple modifications $\Delta_i \rightsquigarrow \Delta_{i-1}$.
Consider the modification $\Delta_{k-1} \rightsquigarrow \Gamma$. By the inductive assumption, the corresponding graph $\widehat{\Delta}_{k-1}$ is a disjoint union of regular graphs. When $\Delta_k$ is a modification centered at a vertex that does not lie in $\widehat{\Delta}_{k-1}$ then $\widehat{\Gamma'}$ is the disjoint union of $\widehat{\Delta}_{k-1}$ and a single vertex with weight $(0,1)$. When $\Delta_k$ is a modification centered at a vertex $v\in \widehat{\Delta}_{k-1}$ then $\widehat{\Gamma'}$ is obtained by a modification of $\widehat{\Delta}_{k-1}$ centered at $v$ and it still regular. When $\Delta_k$ is a modification centered at an edge ${{v\!-\!\!\!-w}}$ with $v,w\notin \widehat{\Delta}_{k-1}$, then $\widehat{\Gamma'}$ is the disjoint union of $\widehat{\Delta}_{k-1}$ and a single vertex with weight $(0,1)$. When $\Delta_k$ is a modification centered at an edge ${{v\!-\!\!\!-w}}$ with $v\in \widehat{\Delta}_{k-1}$ and $w\notin \widehat{\Delta}_{k-1}$, then $\widehat{\Gamma'}$ is obtained by a modification of $\widehat{\Delta}_{k-1}$ centered at $v$. Finally, when $\Delta_k$ is a modification centered at an edge ${{v\!-\!\!\!-w}}$ with $v,w\in \widehat{\Delta}_{k-1}$, then $\widehat{\Gamma'}$ is obtained by a modification of $\widehat{\Delta}_{k-1}$ centered at ${{v\!-\!\!\!-w}}$. In all cases, all connected components of $\widehat{\Gamma'}$ remain regular which completes the proof.
Self-similar links have self-similar skeletons {#section_proof_main}
==============================================
In this section we prove the implication \[condition\_selfsim\] $\implies$ \[condition\_graph\] of Theorem \[mainthm\]. For technical reasons we introduce the following condition.
1. \[condition\_prime\] There exists a finite set $T$ of type 1 points of $\operatorname{NL}(X,0)$ such that $\operatorname{NL}(X,0)\setminus T$ is isomorphic to an open subset $U$ of $\operatorname{NL}(X,0)$ satisfying $\overline{U} \subsetneq \operatorname{NL}(X,0)$.
\[conditiondag\]
Observe that \[condition\_selfsim\] implies \[condition\_prime\], so that we are reduced to prove \[condition\_prime\] $\implies$ \[condition\_graph\]. To do so, we rely on the following result.
Extending morphisms from the punctured disc
-------------------------------------------
\[lem:keyboundary\] Let $D$ be an open $k((t))$-analytic disc and let $0$ denote its origin. Let $\varphi \colon D\setminus\{0\}\to\operatorname{NL}(X,0)$ be any map that induces an isomorphism of locally ringed spaces between $D\setminus\{0\}$ and its image. Then the map $\varphi$ extends continuously to $0$, and $\varphi(0)$ is either a divisorial point or a rigid point of $\operatorname{NL}(X,0)$. In the latter case $\varphi$ extends to an isomorphism of locally ringed spaces from $D$ to $\varphi(D)$.
Let us first take for granted the following lemma.
\[lem:extension\] The map $\varphi \colon D\setminus\{0\}\to\operatorname{NL}(X,0)$ extends uniquely to a continuous map $D\to\operatorname{NL}(X,0)$.
We continue to denote by $\varphi$ the continuous map on $D$ obtained via Lemma \[lem:extension\]. We will prove by contradiction that $\varphi(0)$ is either a divisorial point or a rigid point of $\operatorname{NL}(X,0)$.
Suppose that $\varphi(0)$ is a point of type 3 of $\operatorname{NL}(X,0)$. Then there exist an open subspace $U$ of $\varphi(D\setminus\{0\})$ with $\varphi(0)\in\overline U$ and two non constant functions $f$ and $g$ on $U$ such that $$\lim_{y\in U,y\to \varphi(0)}\frac{\log|f(y)|}{\log|g(y)|}$$ exists and is an irrational number. Therefore the same holds for the ratio of the logarithms of the absolute values of the two nonconstant functions $\varphi^*f$ and $\varphi^*g$ on $\varphi^{-1}(U)$, tending towards $0$. This yields a contradiction since $0$ is a rigid point of $\operatorname{NL}(X,0)$, therefore its valuation ring has rational rank $1$ over $k$.
Now suppose that the point $\varphi(0)$ is of type 4. Then we can find $U$ as above, bounded functions $\{f_n\}_{n\in\N}$, and $t$ on $U$ such that the group $$\Gamma =
\bigg\langle
\lim_{y\in U,y\to \varphi(0)}\frac{\log|f_n(y)|}{\log|t(y)|}
\bigg\rangle_{n\in\N}$$ is not a finitely generated subgroup of $\Q$. Now consider the bounded functions $F_n=\varphi^*f_n$ and $T=\varphi^*t$ on $\varphi^{-1}(U)$. Since $\varphi(U)\cup\{0\}$ is an open neighborhood of $0$ in $D$ we can find a smaller disc $D'$ centered in $0$ such that all the $F_n$ and $T$ are defined on $D'\setminus\{0\}$. The functions $F_n$ and $T$ extend to bounded functions on $D'$, therefore they can be expressed as power series in a variable $X$. Observe now that we have $$\lim_{y\to \varphi(0),y\in U}\frac{\log|f_n(y)|}{\log|t(y)|}=
\lim_{y\to 0,y\in D}\frac{\log|F_n(y)|}{\log|T(y)|}= \frac{\mathrm{val}_X (F_n(X))}{\mathrm{val}_X (T(X))}.$$ This implies that $\Gamma$ is a subgroup of $\frac{1}{\mathrm{val}_X (T(X)} \Z$, hence it is finitely generated, which is a contradiction. It follows that $\varphi(0)$ can only be a rigid point or a divisorial point of $\operatorname{NL}(X,0)$.
If $\varphi(0)$ is a rigid point then it has a neighborhood $U$ isomorphic to a disc thanks to Corollary \[corollary\_topology\], and without loss of generality we can take a smaller disc $D$ if needed and assume that $\varphi(D)$ is contained in $U$. We obtain an injective analytic map from a punctured disc to disc, thus this map necessarily extends as an isomorphism from $D$ onto its image.
By Corollary \[corollary\_global\_topology\], the compact set $\operatorname{NL}(X,0)$ is a disjoint union of discs, finitely many annuli and finitely many (type 2) points. Recall the definition of $x_r$ for $r\in (0,1)$ from Theorem \[thm:R-tree\] and define the injective continuous map $\gamma(r) = \varphi(x_r)$ from $(0,1)$ to $\operatorname{NL}(X,0)$. Since $\gamma$ is injective, there exists a subset $A$ of $\operatorname{NL}(X,0)$ that is either a disc or an annulus and such that $\gamma(r) \in A$ for all $r$ sufficiently small. Recall that $A$ is uniquely arcwise connected by Theorem \[thm:R-tree\] (i), and has either one or two endpoints by Theorem \[thm:R-tree\] (ii). Moreover, the complement of $A$ in its closure $\overline{A}$ in $\operatorname{NL}(X,0)$ contains either one or two points by Corollary \[corollary\_global\_topology\]. This implies that $\overline{A}$ is both uniquely arcwise connected and compact. It follows that $\gamma$ extends continuously as a map from $[0,1)$ to $\overline{A}$ by setting $\gamma(0) = \lim_{r\to 0} \gamma(x_r)$. We set $\varphi(0) = \gamma(0)$, and claim that the resulting map $\varphi\colon D \to \overline{A} \subset \operatorname{NL}(X,0)$ is continuous. Since $\overline{A}$ and $D$ are uniquely arcwise connected, one may define the segment joining any two points $x, y$, and we denote it by $[x,y]$. For any $0< r < 1$, consider the set $U_r \subset \overline{A}$ (resp. $V_r \subset D$) consisting of those points $x$ such that $\varphi(x_r) \notin [x, \varphi(0)]$ (resp. $x_r \notin [x, 0]$). These sets form bases of neighborhoods of $\varphi(0)$ in $\overline{A}$ and $0$ in $D$ respectively. Since $\varphi$ is continuous and injective on $D\setminus \{0 \}$, one has $\varphi(V_r) \subset U_r$, which implies the continuity of $\varphi$ at $0$.
The implication \[condition\_prime\] $\implies$ \[condition\_graph\]
--------------------------------------------------------------------
Let $T$ be a finite and nonempty set of rigid points of $\operatorname{NL}(X,0)$, let $U$ be an open subset of $\operatorname{NL}(X,0)$ whose closure $\overline U$ in $\operatorname{NL}(X,0)$ is strictly contained in $\operatorname{NL}(X,0)$, and let $\varphi\colon \operatorname{NL}(X,0) \setminus T \to U$ be an isomorphism. We will show that $(X,0)$ has a self-similar dual graph.
Since every point $x$ of $T$ has a neighborhood in $\operatorname{NL}(X,0)$ that is a disc with origin $x$, by repeatedly applying Lemma \[lem:keyboundary\] we deduce that $\partial U=\overline U\setminus U$ is a finite subset of $\operatorname{NL}(X,0)$ consisting of rigid and divisorial points. Moreover, we can extend $\varphi$ to each point of $T$ whose image is a rigid point and therefore assume without loss of generality that $\partial U$ only consists of divisorial points. Observe also that $\partial U$ is nonempty since $\overline U$ is strictly contained in $\operatorname{NL}(X,0)$.
Let $C$ be a germ of curve on $(X,0)$ whose components correspond to the points of $T$, let $\pi \colon X' \to X$ be the good resolution of $(X,0)$ that also resolves the germ $C$ and that is minimal with respect to this property (see e.g. [@laufer:normal2dimsing Theorem 5.12]), and let $Z$ be another good resolution of $(X,0)$ resolving the germ $C$ and whose divisorial set $\operatorname{Div}(Z)$ contains $\partial U$.
Set $S=\big(\operatorname{Div}(Z) \setminus U \big) \cup \varphi (\operatorname{Div}(X'))$ and let $\Y'$ be the formal modification whose divisorial set is $S$, as given by Theorem \[theorem\_existence\_modifications\]. We claim that $\Y'$ is algebraized by a good resolution of $(X,0)$ which resolves the germ $C$. This boils down to verifying the conditions of Theorem \[theorem\_characterization\_resolutions\].
In order to do so, let $V$ be any connected component of $\operatorname{NL}(X,0) \setminus S$. Recall that $S$ contains $\operatorname{Div}(Z)$ hence $\partial{U}$, so that $V$ is either disjoint from ${U}$, or contained in it. In the first case, $V$ is a connected component of $\operatorname{NL}(X,0) \setminus \operatorname{Div}(Z)$. Since $Z$ is a good resolution of $(X,0)$ and $C$, it follows that $V$ has the form we want by Theorem \[theorem\_characterization\_resolutions\]. In the second case, $V$ is isomorphic through $\varphi$ to a connected component of $\operatorname{NL}(X,0) \setminus (\operatorname{Div}(X') \cup T)$. As $X'$ is a resolution of $C$, such a component is a disc, an annulus, or a disc with the origin removed (which is itself isomorphic to an annulus, as observed in Remark \[remark\_disc\_minus\_point\]). This completes the proof of the fact that $\Y'$ is algebraized by a good resolution $Y'$ of $(X,0)$ and $C$.
Now let $\Y$ be the formal modification of $(X,0)$ whose divisorial set is $\operatorname{Div}(Z) \setminus U$ and denote by $y$ the point of the reduction $\Y_0$ of $\Y$ corresponding as in Proposition \[proposition\_propertiesNL\_3\] to the connected component $U$ of $\operatorname{NL}(X,0) \setminus \operatorname{Div}(\Y)$. Let $\cJ_y$ be the ideal of $\widehat{\cO_{\Y,y}}$ which defines $\Y_0$ locally around $y$. Observe now that we have a sequence of isomorphisms of complete local rings $$\begin{gathered}
\widehat{\cO_{\Y,y}}
\mathop{\cong}\limits^{\mathrm{Prop.}\ref{proposition_propertiesNL_1}}
\mathcal O_{\operatorname{NL}(\Y,y)}^\circ \big( \operatorname{NL}(\Y,y) \big)
\mathop{\cong}\limits^{\mathrm{Lem.}\ref{lemma_extension_to_type1}}
\mathcal O_{\operatorname{NL}(\Y,y)}^\circ \big( \operatorname{NL}(\Y,y) \setminus V(\cJ_y)\big)\\
\mathop{\cong}\limits^{\mathrm{Prop.}\ref{proposition_propertiesNL}}
\mathcal O_{\operatorname{NL}(X,0)}^\circ \big( U \big)
\mathop{\cong}\limits^{\varphi}
\mathcal O_{\operatorname{NL}(X,0)}^\circ \big(\operatorname{NL}(X,0)\setminus T)
\mathop{\cong}\limits^{\mathrm{Lem.}\ref{lemma_extension_to_type1}}
\mathcal O_{\operatorname{NL}(X,0)}^\circ \big(\operatorname{NL}(X,0) )
\mathop{\cong}\limits^{\mathrm{Prop.}\ref{proposition_propertiesNL_1}}
\widehat{\cO_{X,0}},
\end{gathered}$$ which tells us that $\varphi$ induces an isomorphism of formal schemes $\widehat{\Y/y}\cong \operatorname{Spf}\big(\widehat{\cO_{X,0}}\big)$.
The inclusion $\operatorname{Div}(\Y) \subset \operatorname{Div}(\Y')$ yields a morphism of formal schemes $\psi\colon \Y' \to \Y$ (geometrically this morphism is the contraction of the exceptional components of $\Y'$ corresponding to divisorial points in $U$). Then $\varphi$ induces an isomorphism of formal schemes between the formal completion $\Y''=\widehat{\Y'/\psi^{-1}(y)}$ of $\Y'$ along $\psi^{-1}(y)$ and the formal completion $\X'=\widehat{X'/\pi^{-1}(0)}$ of $X'$ along its exceptional divisor $\pi^{-1}(0)$.
Since $X'$ is the minimal good resolution of $(X,0)$ and $C$, the resolution $\mu\colon Y'\to X$ factors through a morphism $\rho \colon Y' \to X'$. Now observe that since $\widehat{\Y'/\psi^{-1}(y)}$ and $\widehat{X'/\pi^{-1}(0)}$ are isomorphic as formal schemes, then $\operatorname{Dual}\big(\psi^{-1}(y)\big)$ and $\operatorname{Dual}\big(\pi^{-1}(0)\big)$ are isomorphic as weighted graphs (see section \[ssec:dualgraphs\] for the definition of the weighted graph associated to a good resolution). Indeed, let $x$ be a point of $\operatorname{Div}(X')$ corresponding to a component $E$ of $\pi^{-1}(0)$ and denote by $\varphi(E)$ the component of $\psi^{-1}(y)$ corresponding to the point $\varphi(x)\in \operatorname{Div}(Y')$. Then we have the following sequence of field isomorphisms $$k(E)\cong\widetilde{\mathscr H(x)}\cong \widetilde{\mathscr H\big(\varphi(x)\big)} \cong k\big(\varphi(E)\big)$$ since $\varphi$ is an isomorphism of locally ringed spaces, hence $E$ and $\varphi(E)$ have the same genus. Moreover, we have $\widehat{Y'/\varphi(E)}=\widehat{\Y''/\varphi(E)} \cong \widehat{\X'/E}=\widehat{X'/E}$, which implies that $(E\cdot E)=(\varphi(E)\cdot\varphi(E))$ because the degree of the normal bundle of $E$ in $X'$ is the same as the degree of the formal normal bundle of $E$ in $\X'$, and the same holds for the normal bundle of $\varphi(E)$ in $Y'$.
We now observe that the weighted dual graph $\operatorname{Dual}\big(\psi^{-1}(y)\big)$ is a subgraph of $\operatorname{Dual}\big((\pi\circ\rho)^{-1}(0)\big)$, and the latter is obtained by a sequence of simple modifications from $\operatorname{Dual}\big(\pi^{-1}(0)\big)$ since $\rho \colon Y' \to X'$ is a birational morphism between smooth surfaces. This shows that $(X,0)$ has a self-similar dual graph, which is what we wanted to prove.
For the reader convenience, the following diagram summarizes the constructions made in the course of the proof above.
$$\xymatrix@C=1pc@R=1.3pc@M=3pt@L=3pt{
& Y' \ar@{->}[d]^{\rho} \ar@{->}[dl]_{\psi} && \widehat{{Y'}/{(\pi\circ\rho)^{-1}(0)}} \ar@{->}[ll] \ar@{->}[d]^{\cong} && \widehat{Y'/\psi^{-1}(y)} \ar@{_{(}->}[ll] \ar@{->}[dll]^{\cong}_{\varphi}\\
Y \ar@{->}[dr] & X' \ar@{->}[d]^{\pi} && \widehat{X'/\pi^{-1}(0)} \ar@{->}[ll]\\
& X
}$$
Sandwiched singularities are determined by their dual graphs {#section_plumbing}
============================================================
The goal of this section is two-fold. We first prove that a normal surface singularity is sandwiched if its (weighted) dual graph is sandwiched. This is an extension of [@spivakovsky:sandsingdesingsurfNashtransf Proposition 1.11] to arbitrary characteristic. We then explain how this allows to prove the implication \[condition\_graph\] $\implies$ \[condition\_sandwiched\] of Theorem \[mainthm\].
Dual graphs of sandwiched singularities {#ssec:dualgraphs}
---------------------------------------
Let $(X,0)$ be any normal surface singularity, and let $(Y,D) \to (X,0)$ be a good resolution of the singularities of $(X,0)$. We define the weighted dual graph $\operatorname{Dual}(D)$ as follows. Its set of vertices is the set of the irreducible components of $D$, and there is an edge connecting two vertices if and only if the corresponding components intersect. The weight of a vertex is the pair of positive integers $\big(g(E),-E^2\big)$ consisting of the genus and of the opposite of the self-intersection of the corresponding component $E$ of $D$.
\[thm:extend-spiv\] Let $(X,0)$ be a normal surface singularity and assume there exists a good resolution of $(X,0)$ whose associated weighted dual graph is sandwiched. Then $(X,0)$ is sandwiched.
Over the complex numbers, this result is due to Spivakovsky. His proof is complex analytic in nature, relying on plumbing constructions in an essential way. We proceed in very much the same way, using an analogue of plumbing in formal geometry.
Suppose that $(Y,D)$ is a good resolution of singularities of $(X,0)$, that the associated weighted dual graph $\operatorname{Dual}(D)$ is sandwiched, and choose an embedding of $\operatorname{Dual}(D)$ in a regular weighted graph $\Gamma$. We can assume without loss of generality that each of the vertices of $\Gamma\setminus\operatorname{Dual}(D)$ has weight $(0,1)$ and valence 1, so that it is only adjacent to a vertex of $\operatorname{Dual}(D)$. Indeed, each connected component of $\Gamma\setminus\operatorname{Dual}(D)$ is itself regular and can be contracted to a single vertex adjacent to $\operatorname{Dual}(D)$ and of weight $(0,1)$, see [@spivakovsky:sandsingdesingsurfNashtransf Proposition 1.13]).
We will now build a smooth formal $k$-scheme $\Y'$ whose reduction $\Y'_0$ is a curve with simple normal crossings such that $\operatorname{Dual}(\Y'_0)\cong\Gamma$, together with a closed immersion of formal schemes $\widehat{Y/D} \to \Y'$. We will do so by means of a patching procedure, using as elementary building blocks the following smooth formal $k$-schemes of dimension 2. For $n>0$, let $\mathbb{F}_n = \mathbb{P}_{\mathbb P^1_k} \big( \cO_{\mathbb P^1_k} \oplus \cO_{\mathbb P^1_k}(n) \big)$ be the Hirzebruch surface of order $n$ over $k$ and let $F$ denote the rational curve of self-intersection $-n$ in $\mathbb F_n$. Pick a point $p$ in $\mathbb F_n$ that does not lie in $F$, call $F'$ the curve in the canonical rational fibration passing through $p$, and denote by $\widetilde{\mathbb F_n}$ the blowup of $\mathbb F_n$ at $p$. Define $\X_{n}$ to be the formal completion of $\mathbb F_n'$ along the union of $F$ with the strict transform of $F'$ in $\widetilde{\mathbb F_n}$, so that the reduction of $\X_n$ consists of two smooth and rational curves that intersect transversally and have self-intersection $-n$ and $-1$ respectively. Now take a vertex $v$ in $\Gamma\setminus \operatorname{Dual}(D)$ and let $E$ be the component of $D$ corresponding to the vertex of $\operatorname{Dual}(D)$ adjacent to $v$. Since $\Gamma$ is regular, $E$ is rational and has negative self-intersection, say $E^2=-m<0$. Consider the formal $k$-scheme $\X_m$ constructed above and denote by $F$ its rational curve of self-intersection $-m$ and by $q$ the point of $F$ that intersects the other component of $(\X_m)_0$. The formal completions $\widehat{Y/E}$ and $\widehat{\X_m/F}$ are isomorphic by [@LeeNakayama2013 Theorem 2.11] (in characteristic zero this result was obtained earlier in [@Brieskorn1967 Satz 2.10]), and, after composing with an automorphism of $\widehat{\X_m/F}$ if necessary (for instance one induced by a suitable automorphism of the Hirzeburch surface), we can choose such an isomorphism $\widehat{\X_m/F}\cong\widehat{Y/E}$ sending $q$ to a point of $E$ that is smooth in $D$. Therefore we can glue a copy of $\X_m$ to $\widehat{Y/D}$, obtaining a smooth formal $k$-scheme whose associated dual graph is the weighted graph spanned by $\operatorname{Dual}(D)$ and the vertex $v$. By repeating this procedure for every vertex in $\Gamma\setminus \operatorname{Dual}(D)$ we obtain the formal scheme $\Y'$ as we wanted. Observe that by construction the formal scheme $\widehat{Y/D}$ is isomorphic to the closed formal subscheme of $\Y'$ obtained by formally completing the latter along $D$.
Since $\Gamma\cong\operatorname{Dual}(\Y_0)$ is regular, Grauert-Artin contractibility criterion [@artin:contractibilityalgebraicspaces] gives a formal modification $\pi\colon \Y'\to \mathcal{Z}$ contracting $\Y_0$ to a point $\{z\}=\mathcal Z_0$ in a smooth two-dimensional formal $k$-scheme $\mathcal Z$. Having only one point, $\mathcal Z$ is affine, therefore it can be algebraized by a smooth surface $Z$ over $k$. It follows that $\Y'$, being a formal good resolution of $(Z,z)$, is itself algebraized by a good resolution $\pi\colon Y'\to Z$ by [@fantini:normspaces Proposition 7.6].
Using the other contractibility criterion by Artin, [@artin:contractibilitycriterion Theorem 2.3], one can then also contract the divisor $D$ of $Y'$ to a (possibly singular) point $z'$ of a normal surface $Z'$ over $k$, which yields a modification $\varpi\colon Y'\to Z'$ such that $\pi$ factors as $Y' \stackrel{\varpi}{\longrightarrow} Z' \stackrel{\tau}{\longrightarrow} {Z}$. Observe that the local ring $\widehat{\cO_{Z',z'}}$ is sandwiched since $Z$ is smooth.
Finally, observe that we have $$\widehat{\cO_{X,0}}
\cong
\cO^\circ\big(\operatorname{NL}(X,0)\big)
\cong
\cO^\circ\big(\operatorname{NL}(Y,D)\big)
\cong
\cO^\circ\big(\operatorname{NL}(Y',D)\big)
\cong
\cO^\circ\big(\operatorname{NL}(Z',z')\big)
\cong
\widehat{\cO_{Z',z'}},$$ where the first and the last isomorphisms come from Proposition \[proposition\_propertiesNL\_1\] and the others follow from the invariance of non-archimedean links under modifications. This proves that $\widehat{\cO_{X,0}}$ is sandwiched, which is what we wanted to show.
\[rem-equivalence\] The converse to the theorem is also true: the dual graph $\operatorname{Dual}(D)$ is sandwiched if $(X,0)$ is sandwiched. A direct proof of this fact will be given in Remark \[rem:64\].
The implication \[condition\_graph\] $\implies$ \[condition\_sandwiched\]
-------------------------------------------------------------------------
Let $(X,0)$ be any normal surface singularity such that $\operatorname{Dual}(D)$ is self-similar for some good resolution of singularities $(Y,D) \to (X,0)$. By Theorem \[thm:graph-sdw\], the weighted graph $\operatorname{Dual}(D)$ is sandwiched, which implies that $(X,0)$ is sandwiched by Theorem \[thm:extend-spiv\]. $\qed$
End of the proof of Theorem \[mainthm\]: Kato data {#sec-Kato data}
==================================================
In this section, we prove the two implications \[condition\_sandwiched\] $\implies$ \[condition\_kato\], and \[condition\_kato\] implies the condition \[condition\_prime\] from §\[section\_proof\_main\], concluding the proof of our main theorem.
The implications \[condition\_sandwiched\] $\implies$ \[condition\_kato\] {#sec:164}
-------------------------------------------------------------------------
A *Kato datum* \[def:kato datum\] for a normal surface singularity $(X,0)$ is a modification $\pi \colon (X',D) \to (X,0)$, together with an isomorphism of complete local rings $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$ for some $p\in D$. Note that in particular $\pi$ is not an isomorphism.
Suppose $(X,0)$ is a sandwiched singularity. Proving \[condition\_sandwiched\] $\implies$ \[condition\_kato\] amounts to constructing a Kato datum for $(X,0)$. To do so, recall that we may find a proper birational morphism $h \colon Z \to \A^2_k$ and a point $z \in h^{-1}(0)$ such that $\widehat{\mathcal O_{Z,z}} \cong \widehat{\mathcal O_{X,0}}$. Choose a good resolution of singularities $g \colon Y \to X$ of $(X,0)$, and pick an arbitrary point $q\in g^{-1}(0)$. Since $Y$ is smooth at $q$ we can find an étale map $\varphi \colon Y \to \A^2_k$ mapping $q$ to the origin. We consider the fibered product $X' = Y \times_{\A^2_k} Z$, so that the following diagram is commutative $$\xymatrix{
X' = Y \times_{\A^2_k} Z
\ar[r]^-\varpi \ar[d]^\psi & Y \ar[d]^\varphi \ar[r]^g & X\\
Z \ar[r]^h & \A^2_k &
}$$ Observe that the projection map $\varpi\colon X' \to Y$ is birational since $h$ is, and $\psi \colon X' \to Z$ is étale since étale morphisms are preserved by base change. The image of $X'$ in $Z$ contains $h^{-1}(0)$ since $\varphi(q) =0$. We may thus find a point $p\in X'$ over $q$ such that $\psi(p) = z$. Since $\psi$ is étale it induces an isomorphism of complete local rings $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{Z,z}}$, and the latter is isomorphic to $\widehat{\mathcal O_{X,0}}$. It follows that the composition $g \circ \varpi \colon X' \to X$, which is a modification (that is, $(g \circ \varpi)^{-1}(0)$ is a Cartier divisor) because it factors through the resolution $Y$, defines a Kato datum for $(X,0)$. $\qed$
The implication \[condition\_kato\] $\implies$ \[condition\_prime\] {#sec:165}
-------------------------------------------------------------------
Suppose that $(X,0)$ is a normal surface singularity and that $\pi \colon (X',D) \to (X,0)$ is a Kato datum for $(X,0)$, with $p\in D$ a point such that $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$. By Proposition \[proposition\_propertiesNL\_3\] there exists a finite set $T'$ of type 1 points of $\operatorname{NL}(X',p)$ such that the open subspace $U=\mathrm{c}_{X'}^{-1}(p)$ of $\operatorname{NL}(X,0)$ is isomorphic to $\operatorname{NL}(X',p)\setminus T'$. Since $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$, it follows that $U\cong \operatorname{NL}(X,0) \setminus T$ for some finite set $T$ of type 1 points of $\operatorname{NL}(X,0)$. Finally, the closure $\overline{U}$ of $U$ in $\operatorname{NL}(X,0)$ is strictly contained in the latter, since it is contained in $U\cup \operatorname{Div}(X')$ as observed in Remark \[remark\_closure\_component\]. $\qed$
\[rem:64\] Let us sketch a proof of the implication \[condition\_kato\] $\implies$ \[condition\_graph\]. Although not necessary to complete the proof of Theorem A, we obtain in this way a direct argument showing the implication \[condition\_sandwiched\] $\implies$ \[condition\_graph\], which is equivalent to saying that every sandwiched singularity has a sandwiched dual graph, as mentioned in Remark \[rem-equivalence\]. Suppose that $(X,0)$ is a normal surface singularity and that $\pi \colon (X',D) \to (X,0)$ is a Kato datum for $(X,0)$, with $p\in D$ a point such that $\widehat{\mathcal O_{X',p}} \cong \widehat{\mathcal O_{X,0}}$. Without loss of generality we may assume that $D$ has simple normal crossings away from $p$, so that in particular $X'$ is smooth away from $p$. Let $\mu \colon Y \to X$ and $\mu' \colon Y' \to X'$ be the minimal good resolutions of the singularities of $(X,0)$ and $(X',p)$ respectively. Since $Y'$ is a good resolution of $(X',p')$ it is also a good resolution of $(X,0)$ so that there exists a birational morphism $\pi' \colon Y' \to Y$ satisfying $\mu \circ \pi' = \pi \circ \mu'$. This morphism is not an isomorphism, or $\pi$ would be an isomorphism as well. Moreover, since $Y'$ and $Y$ are smooth, $\pi'$ is a composition of point blow-ups. The graph $\operatorname{Dual}\big((\pi\circ\mu')^{-1}(0)\big)$ is thus a nontrivial modification of $\operatorname{Dual}\big(\mu^{-1}(0)\big)$ and contains $\operatorname{Dual}\big((\mu')^{-1}(p)\big)$ which is isomorphic (as a weighted graph) to $\operatorname{Dual}\big(\mu^{-1}(0)\big)$. This proves that the latter graph is sandwiched as required.
Complex analytic sandwiched singularities {#section_complexanalytic}
=========================================
In this section we work with *complex analytic varieties*. In this context, a sandwiched singularity $(X,0)$ is a germ of normal complex analytic surface such that the completion of its analytic local ring is sandwiched in the sense of Definition \[def-sandwich\]. Our aim is to characterize such singularities in terms of their complex link, proving Theorems \[thm3\] and \[thm2\].
Pseudoconvex $3$-folds {#sec:pdcvx3folds}
----------------------
Let us recall some definitions and terminology from [@kato:cpctcplxsurfwithGSPH]. A *real-analytic strongly pseudoconvex $3$-fold* $\Sigma$ is a smooth (real-analytic) hypersurface in a smooth complex surface $S$ such that for any $p\in \Sigma$ there exists an open subset $U$of $S$ containing $p$ and a real-analytic strictly plurisubharmonic function $\rho \colon U \to \R$ such that $\Sigma \cap U = \rho^{-1}(0)$. A real-analytic strongly pseudoconvex $3$-fold *bounds a Stein domain* if $\Sigma$ is compact and there exists an embedding of a tubular neighborhood of $\Sigma$ in $S$ into a normal (but possibly singular) complex surface $X$ such that $X \setminus \Sigma$ has one component which is Stein.
We shall need one more piece of terminology. A compact real $3$-fold $\Sigma$ in a compact complex surface $S$ is said to be *global* when $S\setminus \Sigma$ is connected.
Archimedean links are the main example of real-analytic strongly pseudoconvex $3$-folds bounding a Stein domain. Let $(X,0)$ be a normal surface singularity, and fix an embedding of $(X,0)$ inside the unit ball in some complex affine space $\C^n$ such that $0$ is sent to the origin. The function $\rho(z) = \rho(z_1, \ldots , z_n) := \sum_{i=1}^n |z_i|^2$ is real-analytic and strictly plurisubharmonic. There exists $\eps_0$ small enough so that for any $0 < \eps < \eps_0$ the intersection of the sphere of center $0$ and radius $\eps$ with $X$ is transversal, so that ${{L}_\C^{\eps}}(X,0) := X \cap \{ \rho = \eps\}$ is a real-analytic strongly pseudoconvex $3$-fold, and bounds the Stein domain $X_\eps := X \cap \{\rho < \eps\}$. Such $\eps_0$ may be taken maximal satisfying the above property, and any $\eps$ strictly smaller than the threshold $\eps_0$ above is said to be *admissible*.
Observe that the diffeomorphism type of ${{L}_\C^{\eps}}(X,0)$ does not depend on the choice of an admissible $\eps$, but its embedding as a real-analytic $3$-fold in $X$ does. In fact the diffeomorphism type of ${{L}_\C^{\eps}}(X,0)$ is also independent on the choice of an embedding in $\C^n$, see [@MR3112993 Proposition 2.5].
Kato surfaces {#ssec:Katosurfaces}
-------------
A special case of real-analytic strongly pseudoconvex $3$-folds is given by *spherical shells*, corresponding to the boundary of the unit ball $B$ in $\C^2$, i.e., to the link of a regular point $(X,0)$. In [@kato:cptcplxmanifoldsGSS], M. Kato considers the following construction to produce compact complex surfaces admitting a global spherical shell. Let $Y$ be any connected open neighborhood of $\overline{B}$ in $\C^2$, and let $\pi\colon Y' \to Y$ be a proper bimeromorphic map which is an isomophism above $Y \setminus \{0\}$. Let $y$ be a point in $\pi^{-1}(0)$, and pick a relatively compact neighborhood $U$ of $y$ in $\pi^{-1}(B)$ such that there exists a biholomorphism $\sigma\colon Y \to U$. We call the pair $(\pi,\sigma)$ a *regular geometric Kato datum*, to distinguish this definition from the one given in §\[sec:164\]. One can define a compact complex surface $S=S(\pi,\sigma)$, called *Kato surface*, obtained from $Y' \setminus \sigma(\overline{B})$ by gluing $Y'\setminus \pi^{-1}(\overline{B})$ and $\sigma(Y\setminus \overline{B})$ using $\sigma \circ \pi$
Kato surfaces have been studied intensively in the literature, see for example the monograph of G. Dloussky [@dloussky:phdthesis]. They are compact complex surfaces with negative Kodaira dimension, $b_1 =1$, $b_2 >0$, and they admit a global spherical shell. In the Kodaira classification of compact complex surfaces, they belong to the VII$_0$ class, and it is believed that they are the only examples of surfaces in this class having $b_2>0$, see [@MR2726099].
Proof of Theorem \[thm3\] {#ssec:proofthm3}
-------------------------
In this section we fix a sandwiched singularity $(X,x_0)$. Our aim is to realize its archimedean link as a global real-analytic strongly pseudoconvex $3$-fold in a compact complex surface $S$ that has a global spherical shell.
Our first objective is to construct a complex analytic version of a Kato datum attached to $(X,x_0)$. Let $Y$ be a connected neighborhood of the closed unit ball in $\C^2$. By Artin’s approximation theorem and §\[ssec:sandwich\], we may find a proper bimeromorphism $\pi \colon Y' \to Y$ that is an isomorphism over $Y\setminus \{0 \}$, and a non-empty connected divisor $D \subset \pi^{-1}(0)$ such that the normal complex analytic germ obtained by contracting $D$ to a point is isomorphic to $(X,x_0)$. We write $\mu \colon Y' \to X$ for the contraction morphism. It is a proper bimeromorphism that is a local isomorphism at any point of $D$ and contracts $D$ to the point $x_0$.
Observe that we may (and shall) assume that the support of $D$ is the union of all rational curves in the exceptional divisor $\pi^{-1}(0)$ of self-intersection $\le -2$ by [@spivakovsky:sandsingdesingsurfNashtransf Corollary 1.14]. In that case the map $\mu$ is the minimal good resolution of $(X,x_0)$. Define the proper bimeromorphism $\eta \colon X \to Y$ by setting $\eta := \pi \circ \mu^{-1}$.
Consider any embedding of a local neighborhood of $x_0$ in $X$ into the unit ball in $\C^n$ that sends $x_0$ to the origin, so that we can talk about its complex link ${{L}_\C^{\eps}}(X,x_0)$ for any admissible $\eps \ll 1$. Recall the definition of the Stein domain $X_\eps$ from §\[sec:pdcvx3folds\].
Now pick any point $y \in \mu^{-1}(x_0)$, and choose a neighborhood $U$ of $y$ such that $\mu(U)$ is relatively compact in $X_\eps$ and there exists a biholomorphism $\sigma\colon Y \to U$ (we possibly have to shrink $Y$ in order to do that). We construct a complex surface $X'$ by patching together $Y' \setminus \sigma(\overline{B})$ and $X$ using $\sigma \circ \eta \colon X \setminus \eta^{-1}(\overline{B}) \to \sigma(Y\setminus\overline{B})$. The identification of $X$ with its copy inside $X'$ induces a holomorphic map $\tilde{\sigma} \colon X \to X'$ that is a biholomorphism onto its image, and such that $\tilde{\sigma}(X)$ is relatively compact inside $X'$. We also have a proper bimeromorphic map $\tilde{\pi} \colon X' \to X$ defined by $\tilde{\pi} = \mu$ on $Y' \setminus \sigma(\overline{B})$ and by $\tilde{\pi}= \mu \circ \sigma \circ \eta$ on $X$.
We proceed to construct a compact complex surface following Kato’s construction as in the previous section. Pick three admissible positive real numbers $ \eps_- < \eps< \eps_+$, and set $X'_t = \tilde{\pi}^{-1}(X_t)$ for every $t$ in $\{\eps_-, \eps, \eps_+\}$. Define the surface $\tilde{S}$ considering $X'_{\eps_+} \setminus \tilde{\sigma}(\overline{X_{\eps_-}})$ and gluing together $X'_{\eps_+} \setminus \overline{X'_{\eps_-}}$ and $\tilde{\sigma}(X_{\eps_+} \setminus \overline{X_{\eps_-}})$ via the map $\tilde{\sigma} \circ \tilde{\pi}$. Observe that the canonical map $\overline{X'_\eps} \setminus \tilde{\sigma}(X_\eps) \to \tilde{S}$ is surjective hence $\tilde{S}$ is compact. It is a smooth surface, since the only singularity of $X'_{\eps_+}$ lies inside $\tilde{\sigma}(\overline{X_{\eps_-}})$.
By construction, $\tilde{S}$ contains a neighborhood of the archimedean link ${{L}_\C^{\eps}}(X,x_0)$, that is hence realized as a real-analytic strongly pseudo-convex $3$-fold bounding a Stein domain in $\tilde{S}$.
To see that $\tilde{S} \setminus{{L}_\C^{\eps}}(X,x_0)$ is connected, it is enough to see that $X'_{\eps} \setminus \overline{\tilde{\sigma}(X_{\eps})}$ is connected. But $\tilde{\sigma}({{L}_\C^{\eps}}(X,x_0))$ is a compact and connected real $3$-fold in $X'_{\eps}$, hence $X'_{\eps} \setminus \tilde{\sigma}({{L}_\C^{\eps}})$ has at most two components, one of which is $\tilde{\sigma}(X_{\eps})$. This implies that $X'_{\eps} \setminus \overline{\tilde{\sigma}(X_{\eps})}$ is connected.
It remains to see that $\tilde{S}$ contains a global spherical shell. This follows from [@kato:cpctcplxsurfwithGSPH Proposition 2], whose proof is given on pp. 541–546. The proof should be read by replacing $Z(\delta)$ (resp. $A$, $0^*$, and $g$) by $X'_\eps$ (resp. by $\tilde{\pi}^{-1}(x_0)$, $\sigma(x_0)$ and $\tilde{\sigma} \circ \tilde{\pi}$).
One can also see it directly, by showing that $\tilde{S}$ is biholomorphic to the Kato surface $S=S(\pi,\sigma)$ associated with the regular Kato datum $(\pi,\sigma)$. Define a map $\eta'\colon X'\to Y'$ by setting $\eta'=\operatorname{id}$ on $Y' \setminus \sigma (\overline{B})\subset X'$, and $\eta'=\sigma \circ \eta$ on $X\subset X'$. Denote by $S^*$ the surface obtained from $X' \setminus \tilde{\sigma}\big(\eta^{-1} (\overline{B})\big)$ by gluing $X'\setminus \tilde{\pi}^{-1}\big(\eta^{-1} (\overline{B})\big)$ and $\tilde{\sigma}\big(Y\setminus \eta^{-1} (\overline{B})\big)$ using $\tilde{\sigma} \circ \tilde{\pi}$. The map $\eta'$ then induces a biholomorphism between $S^*$ and $S$. We show that $S^*$ and $\tilde{S}$ are biholomorphic.
To that end, consider the surface $S'$ obtained from $X' \setminus \tilde{\sigma}(\overline{X_{\eps_-}})$ by gluing $X' \setminus \overline{X'_{\eps_-}}$ to $\tilde{\sigma}(X\setminus \overline{X_{\eps_-}})$ via the map $\tilde{\sigma} \circ \tilde{\pi}$. The inclusions $\imath^*\colon X' \setminus \tilde{\sigma}(\eta^{-1} (\overline{B})) \hookrightarrow X' \setminus \tilde{\sigma}(\overline{X_{\eps_-}})$ and $\tilde{\imath}\colon X'_{\eps_+} \setminus \tilde{\sigma}(\overline{X_{\eps_-}}) \hookrightarrow X' \setminus \tilde{\sigma}(\overline{X_{\eps_-}})$ induce biholomorphisms $\Phi^*\colon S^* \to S'$ and $\tilde{\Phi}\colon\tilde{S} \to S'$.$\qed$
Proof of Theorem \[thm2\]
-------------------------
As above, we fix a closed embedding of $X$ in the open unit ball in $\C^n$, define $\rho (z) := \sum_{i=1}^n |z_i|^2$, and we write ${{L}_\C^{\eps}}(X,0):= X \cap \{\rho = \eps\}$ for any admissible $\eps>0$, $X_\eps:= X \cap \{ \rho < \eps\}$, and $A_{\eps_-,\eps_+} := X_{\eps_+} \setminus \overline{X_{\eps_-}}$.
We suppose that for some admissible $\eps$ the manifold ${{L}_\C^{\eps}}(X,0)$ can be realized as a global strongly pseudoconvex $3$-fold in a smooth compact complex surface $S$. By assumption, there exist $\eps_- < \eps < \eps_+$ and a holomorphic embedding $\imath \colon A_{\eps_-,\eps_+} \to S$ such that $S\setminus \imath(A_{\eps_-,\eps_+})$ is connected. Let $X'$ be the surface obtained by gluing $S\setminus \imath({{L}_\C^{\eps}}(X,0))$ and $X_{\eps_+}$ together along the open subsets $\imath (A_{\eps,\eps_+})$ and $A_{\eps,\eps_+}$ via the map $\imath^{-1}$. This surface has a single singularity at the point $0'$ corresponding to the point $0$ of $X_{\eps_+}$, and $(X',0')$ is a normal singularity which is analytically isomorphic to $(X,0)$. Denote by $\sigma$ the map induced by the identification of $X_{\eps_+}$ inside $X'$. It is a holomorphic embedding mapping the singular point $0$ to $0'$.
Observe that there is a biholomorphic copy $A$ of $A_{\eps_-,\eps}$ that is included in $X'$ and whose complement in $X'$ is compact. Since $X_{\eps}$ is Stein, the natural biholomorphism $A \to A_{\eps_-,\eps}$ extends to a proper bimeromorphic map $\pi \colon X' \to X_{\eps}$ by a classical theorem of Hartogs. Note that $\pi$ could be a local isomorphism at $0$, and it does not necessarily send $0'$ to $0$.
If $\pi(0')\neq 0$, then $\pi$ is a proper bimeromorphic map from $(X',0')$ onto a smooth point in $X_{\eps}$, therefore $(X',0')$ (and hence $(X,0)$) is a sandwiched singularity as was to be shown. We may thus assume that $\pi(0') = 0$ so that $0'$ is fixed by the map $f:=\sigma\circ \pi \colon X' \to X'$.
First suppose that $\pi$ does not induce a local biholomorphism from $(X',0')$ to $(X,0)$. Then the maps $\pi \colon X' \to X_\eps$ and $\sigma\colon X_\eps \to X'$ define a Kato datum in the sense of page . We deduce that $(X,0)$ is sandwiched by the implication \[condition\_kato\]$\Rightarrow$\[condition\_sandwiched\] of Theorem \[mainthm\]. The fact that $\tilde{S}$ contains a global spherical shell follows similarly as before from [@kato:cpctcplxsurfwithGSPH Proposition 2].
Now suppose that $\pi$ induces a local biholomorphism from $(X',0')$ to $(X,0)$. Then the map $f$ induces a local biholomorphism from $(X',0')$ to itself and we have $K:=\bigcap_{n\in\N} f^n(\sigma(X_\eps)) = \{ 0'\}$. Indeed, $K$ is a compact subset of $\sigma(X_\eps)$, and, since the latter can be realized as a bounded set in $\C^n$, Montel’s theorem applies and all limits of the sequence of iterates $\{f^m\}_{m\ge0}$ should be constant (this argument is due to M. Kato, see the proof of [@kato:cptcplxmanifoldsGSS Lemma 2]). But $f$ is fixing $0'$, hence $K = \{ 0'\}$. In other words, $f \colon (X',0')\to (X',0')$ is a contracting automorphism in the terminology of [@MR3270424].
Let $S(f)$ be the space of orbits of $f$, that is the quotient of $X'\setminus\{0'\}$ by the equivalence relation defined by $x \simeq x'$ if and only if $f^n(x) = f^m(x')$ for some positive $n,m\in\N$. This space is naturally a compact complex surface. Observe that one has a natural holomorphic map $S\setminus \imath\big({{L}_\C^{\eps}}(X,0)\big) \subset X'$ to $S(f)$, and that this map actually descends to a map $S \to S(f)$. It follows that $S$ is a modification of $S(f)$.
The singularity $(X,0)$ and the geometry of $S(f)$ are completely described in [@MR3270424 Theorem 7.5]. We are thus in one of the following situations.
1. The singularity $(X,0)$ is a cyclic quotient singularity. In this case [@MR3270424 Corollary B] shows that $S(f)$ is a Hopf surface. More precisely the universal cover of $S(f)$ is isomorphic to $\C^2\setminus\{0\}$ and its fundamental group is the subgroup of polynomial automorphisms generated by $\gamma$ and $\tilde{f}$ in the notations of [@MR3270424 Example 7.1]. In particular, this fundamental group is not cyclic, so that $S(f)$ is a secondary Hopf surface. We are in case \[item:thm2b\] of the theorem.
2. \[item:2\] There exist a Riemann surface $C$, a line bundle $L\to C$ of negative degree, and a finite group $G$ of automorphisms of $C$ that acts linearly on $L$. Let $X'$ be the surface obtained by contracting the zero section in the total space of $L$, and let $0'$ be the image of the zero section in $X'$. Then the germ $(X,0)$ is isomorphic to a neighborhood of the image of $0'$ in the quotient space $X'/G$. Moreover, one can find a positive integer $N\ge 1$ and a complex number $\alpha$ of norm smaller than $1$ such that $f^N$ lifts to a linear map acting by multiplication by $\alpha$ on the fibers of $L$. By [@MR3270424 Lemma 8.1], the natural map $S(f^N)\to S(f)$ is a Galois cyclic (unramified) holomorphic cover.
1. Suppose first that the genus of $C$ is positive. Then $(X',0')$ is not rational, hence $(X,0)$ is not rational either by [@MR1368632 Claim 6.11]. In particular $(X,0)$ is not a quotient singularity. By [@MR3270424 Theorem A], it follows that $(X,0)$ is weighted homogeneous. Finally the proof of [@MR3270424 Corollary B] shows that $S(f^N)$ is a principal elliptic fiber bundle of Kodaira dimension $0$ or $1$ so that we are in case \[item:thm2a\] of our theorem.
2. Suppose that $C=\P^1$ is the Riemann sphere. Then $(X',0')$ is a cyclic singularity and $(X,0)$ is a quotient singularity. Again the proof of [@MR3270424 Corollary B] shows that $S(f)$ is a Hopf surface. If the group $G$ acting on $(X',0')$ is trivial, then $(X,0)$ is a cyclic quotient singularity, and the surface is either a secondary Hopf or a primary Hopf. In the latter case it contains a global spherical shell and we are in case \[item:thm2c\]. In the former case we are in case \[item:thm2b\] of our theorem. When $G$ is non-trivial, then the fundamental group of $S(f)$ is not cyclic hence it is a secondary Hopf surface; and we are in case \[item:thm2b\].
This completes the proof of Theorem \[thm2\]. $\qed$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The well-known bluer-when-brighter trend observed in quasar variability is a signature of the complex processes in the accretion disk, and can be a probe of the quasar variability mechanism. Using a sample of 604 variable quasars with repeat spectra in SDSS-I/II, we construct difference spectra to investigate the physical causes of this bluer-when-brighter trend. The continuum of our composite difference spectrum is well-fit by a power-law, with a spectral index in excellent agreement with previous results. We measure the spectral variability relative to the underlying spectra of the quasars, which is independent of any extinction, and compare to model predictions. We show that our SDSS spectral variability results cannot be produced by global accretion rate fluctuations in a thin disk alone. However, we find that a simple model of a inhomogeneous disk with localized temperature fluctuations will produce power-law spectral variability over optical wavelengths. We show that the inhomogeneous disk will provide good fits to our observed spectral variability if the disk has large temperature fluctuations in many independently varying zones, in excellent agreement with independent constraints from quasar microlensing disk sizes, their strong UV spectral continuum, and single-band variability amplitudes. Our results provide an independent constraint on quasar variability models, and add to the mounting evidence that quasar accretion disks have large localized temperature fluctuations.'
author:
- 'John J. Ruan, Scott F. Anderson, Jason Dexter, Eric Agol'
bibliography:
- 'bibref.bib'
title: Evidence for Large Temperature Fluctuations in Quasar Accretion Disks From Spectral Variability
---
Introduction
============
A well-known characteristic of the quasar phenomena is their strong flux variability in many wavelength regimes, including the radio, optical, X-ray, and $\gamma$-rays [@ul07]. In particular, the rise of optical large-scale time-domain imaging surveys has led to many recent investigations of broadband quasar optical variability properties using large numbers of well-sampled light curves, especially for use in quasar selection [@ke09; @ko10; @ma10; @schmidt10; @bu11; @ki11; @ma11; @ru12; @an13; @zu13]. These studies have generally revealed that quasars are stochastically variable on the $\sim$10-20% level in flux on long time-scales, and show weaker, correlated variability on timescales $\lesssim$ 1 year in the rest-frame. The physical cause of quasar variability is still unclear, but since the optical continuum is likely to be dominated by emission from the accretion disk, some studies have suggested that changes in the global accretion rate in the disk may be able to produce such effects [@pe06; @li08; @zu12]. These claims appear to be supported (although not implied) by various observed trends between optical variability amplitude, black hole mass, and luminosity in different quasars [@ho94; @ga99; @va04; @wi08; @ba09; @ke09; @ma10; @zu12], in turn suggesting that the differences in variability across a sample of quasars may be driven at least in part by their Eddington ratio.
Accretion rate fluctuations are expected in individual quasars due to processes in the disk such as the magnetorotational instability [MRI, @ba91], which is now generally accepted to operate in a wide range of accretion flows. Results from non-radiative global simulations of thin magnetized disks have also shown that such accretion flows are almost certain to be highly turbulent [e.g. @am01; @ar03; @no09a; @no09b; @pe10], but the exact characteristics of a MRI flow in a global radiative MHD simulation are currently unclear, especially their stability in the radiation pressure-dominated regime [see discussions in @hi09; @ja12; @ji13]. Localized temperature fluctuations in highly turbulent disks will also cause flux variations, and such a scenario may be expected since quasar accretion disks are too large to vary coherently in flux over the short variability timescales observed. The characteristic timescales of quasar flux variability have also been shown to be consistent with the thermal timescale [@ke09], independently motivating accretion disk models involving localized temperature fluctuations. Furthermore, it has been shown that such models of inhomogeneous accretion disks can also simultaneously explain quasar microlensing disk sizes, their strong UV spectral continuum, and single-band optical variability properties [@de11].
Although the relative roles of global accretion rate fluctuations and localized temperature fluctuations in accounting for the observed flux variability are unclear, an additional probe is provided by the characteristic bluer-when-brighter trend observed in studies of quasar spectral variability [@cu85; @gi99; @tr01; @tr02; @va04; @wi05; @me11; @sa11; @sc12; @me13]. This trend is almost certainly a direct consequence of the underlying quasar variability mechanism, and thus provides an independent test of quasar variability models. Indeed, both global accretion rate changes and localized temperature fluctuations in accretion disks will generically produce bluer-when-brighter trends, but the details of the predicted trend are dependent on the details of the model. Intriguingly, previous investigations of quasar spectral variability have sometimes resulted in disparate conclusions.
@wi05 used a sample of 315 pairs of repeat spectra of variable quasars from the Sloan Digital Sky Survey [SDSS, @yo00], and constructed ‘difference spectra’, by taking the spectrum of each quasar at the higher-flux epoch and subtracting the spectrum at the lower-flux epoch; this effectively isolates the variable part of the spectrum. After applying a wavelength-dependent spectrophotometric recalibration on each pair of repeat spectra, they find that the resulting composite quasar difference spectrum has a steeper power-law index than the composite of the individual spectra, showing that quasars are indeed bluer-when-brighter. Based on the composite difference spectrum from that study, @pe06 fitted synthetic difference spectra generated from a simple Shakura-Sunayev thin disk model [@sh73], and showed that the composite difference spectrum can be produced from a simple thin disk in which the global accretion rate has changed [also see @sa11].
In contrast, @sc12 used a sample of 9093 multi-band quasar light curves from SDSS Stripe 82 to study the bluer-when-brighter trend using many epochs of broadband photometry. After correcting for the effects of broad emission lines (which are well-known to be less variable than the continuum) in each filter, they compare their results to spectral variability predictions from accretion rate fluctuations in a simple thin disk, as well as more detailed static disk models. They find that accretion rate changes in these disk models *cannot* reproduce the strong bluer-when-brighter trend, and instead suggest that ephemeral hot spots on the accretion disk may be needed [also see @tr02; @me13].
The conclusions of @wi05 [hereafter WI05] and @sc12 [hereafter SC12] appear at first glance to be at odds: can the bluer-when-brighter trend observed in quasar be explained by fluctuations in the global accretion rate in a simple thin disk, or are localized temperature fluctuations needed? In this paper, we revisit the spectral variability study of WI05 using a larger sample of repeat quasar spectra culled from the full SDSS-I/II data set, to investigate these apparently discordant results. We will also compare our spectral variability results to the recently-developed time-dependent model of inhomogeneous accretion disks by @de11. The structure of this paper is as follows: In Section 2, we describe the construction of the sample of repeat quasar spectra used in this study, and our spectrophotometric recalibration. In Section 3, we discuss the properties of the quasars’ difference spectra, as well as the construction of composite spectra and composite difference spectra. In Section 4, we compare our results to the previous studies of WI05 and SC12 using global accretion rate fluctuations in a thin disk. In Section 5, we discuss the quasar spectral variability predicted from a time-dependent model with temperature fluctuations, and show that our observations are well-fit by such inhomogeneous disks. In Section 6, we discuss the connection between disk properties and the resultant spectral variability, as well as other variability mechanisms that might match the observations. We summarize and conclude in Section 7.
Data Selection and Reduction
============================
SDSS-I/II Repeat Spectra
------------------------
All spectroscopic data used in our paper is from the SDSS-I/II, which is publicly available in its entirety as part of SDSS Data Release 7 [DR7, @ab09]. The SDSS-I/II obtained follow-up spectra of approximately $1.6\times10^6$ objects, including more than $1.1\times10^5$ quasars [@schneider10], primarily selected by optical color from the imaging portion of the survey [@ri02]. The two fiber-fed SDSS spectrographs utilize a total of 640 fibers plugged into holes drilled onto plates, which are placed at the telescope focal plane. During the normal course of operations, multiple 15-minute exposures of each plate are taken, and spectra from exposures within approximately a month are typically coadded together. The spectral reduction and calibration using the SDSS Spectro2d pipeline are described in @st02. Occasionally, entire plates may be reobserved and coadded separately as a second epoch of spectra. This may occur if the first epoch did not reach sufficiently high signal-to-noise ratios (SNR), or in some cases by design as part of the survey plan. For these multiply-observed plates, no attempt was made to ensure that the same fiber was plugged into the same hole on the plate, and so spectra of the same object may have different fiber numbers in the different epochs, even though the plate number is identical. For more details about these repeatedly observed plates, we refer to discussions in @wi05.
Although multiple epochs of SDSS spectra are also possible due to spatial overlaps in the sky between adjacent plates, @va04 showed that additional calibrations based on non-variable stars on the same plates enhance sensitivity to the wavelength-dependent variability properties of quasars. Overlapping regions on adjacent plates are generally small, and will not have many non-variable stars in the overlapping regions to accurately recalibrate the quasar spectra in the same regions. Thus, we focus only on multi-epoch spectra from plates that have been reobserved in their entirety, which ensures that multi-epoch spectra of many calibration stars are available in addition to the quasars. We also note that more epochs of spectra for many SDSS DR7 quasars are now publicly available as part of the SDSS-III Baryon Oscillation Spectroscopic Survey [@da13]. However, SDSS-III utilizes a newer spectrograph, fiber system, and spectral reduction pipeline; robust comparison of continuum properties of spectra between SDSS-I/II and SDSS-III are difficult and thus not considered here.
Using the plate list of 2880 observations of all 2698 unique plates in DR7 from SDSS-I/II, we select only those which are part of the main SDSS survey (and its primary reduction pipeline) by requiring the flag SURVEY = ‘sdss’. A data quality cut is then made by requiring the flag PLATEQUALITY = ‘good’ or ‘marginal’; plates which pass this quality cut have SNR $>$ 9 and less than 13% problematic pixels. From the remaining plates, we select those which have multiple observations with time-lag $>$30 days between each pair of epochs in the observed frame. This time-lag cut is physically motivated from photometric studies of quasar light curves, which have shown that quasars are generally not variable above $\sim$1% in flux on such short timescales [@ke09; @ko10; @ma11]. We note that a few plates had three or more observations; in such cases, we use all unique pairs of observations of each plate that pass all the above criterion. There are a total of 71 unique pairs of plate observations in SDSS-I/II, which are listed in Table 1. Figure 1 shows the distribution of the timelags for these 71 pairs of plate observations, which range from 30 days to about 3 years in the observed-frame.
[ccccc]{} 291 & 51660 & 51928 & 21 & 9\
293 & 51994 & 51689 & 54 & 18\
296 & 51578 & 51984 & 26 & 7\
297 & 51663 & 51959 & 23 & 15\
300 & 51666 & 51943 & 41 & 6\
301 & 51641 & 51942 & 45 & 17\
304 & 51957 & 51609 & 26 & 12\
306 & 51690 & 51637 & 35 & 1\
309 & 51666 & 51994 & 47 & 14\
340 & 51691 & 51990 & 24 & 3\
351 & 51780 & 51695 & 36 & 3\
352 & 51789 & 51694 & 29 & 6\
360 & 51780 & 51816 & 69 & 6\
385 & 51783 & 51877 & 49 & 5\
390 & 51816 & 51900 & 30 & 7\
394 & 51876 & 51812 & 31 & 6\
394 & 51812 & 51913 & 28 & 5\
394 & 51876 & 51913 & 31 & 5\
404 & 51877 & 51812 & 14 & 1\
406 & 51817 & 51869 & 48 & 6\
406 & 51876 & 51817 & 50 & 4\
406 & 51817 & 51900 & 49 & 5\
406 & 52238 & 51817 & 48 & 19\
406 & 51900 & 51869 & 54 & 6\
406 & 52238 & 51869 & 47 & 19\
406 & 52238 & 51876 & 51 & 19\
406 & 52238 & 51900 & 49 & 20\
410 & 51816 & 51877 & 83 & 22\
411 & 51873 & 51817 & 28 & 8\
412 & 51871 & 51931 & 30 & 3\
412 & 51871 & 52235 & 29 & 8\
412 & 51871 & 52250 & 31 & 8\
412 & 51871 & 52254 & 30 & 13\
412 & 51871 & 52258 & 29 & 11\
412 & 52235 & 51931 & 32 & 12\
412 & 52250 & 51931 & 35 & 13\
412 & 52254 & 51931 & 32 & 11\
412 & 51931 & 52258 & 32 & 11\
413 & 51821 & 51929 & 46 & 3\
414 & 51869 & 51901 & 39 & 3\
415 & 51879 & 51810 & 39 & 4\
416 & 51885 & 51811 & 68 & 31\
418 & 51884 & 51817 & 72 & 16\
419 & 51812 & 51868 & 69 & 7\
419 & 51812 & 51879 & 64 & 24\
422 & 51878 & 51811 & 26 & 1\
476 & 52027 & 52314 & 80 & 20\
483 & 51942 & 51902 & 78 & 8\
525 & 52029 & 52295 & 52 & 19\
547 & 51959 & 52207 & 66 & 19\
662 & 52178 & 52147 & 37 & 5\
803 & 52264 & 52318 & 4 & 1\
810 & 52326 & 52672 & 4 & 0\
814 & 52370 & 52443 & 47 & 7\
820 & 52405 & 52438 & 80 & 3\
960 & 52466 & 52425 & 31 & 1\
1028 & 52562 & 52884 & 3 & 2\
1034 & 52525 & 52813 & 2 & 1\
1037 & 52826 & 52878 & 1 & 0\
1512 & 53035 & 53742 & 19 & 7\
1670 & 53438 & 54553 & 41 & 19\
1782 & 53383 & 53299 & 31 & 3\
1905 & 53613 & 53706 & 21 & 5\
1907 & 53265 & 53315 & 27 & 3\
2009 & 53857 & 53904 & 44 & 3\
2061 & 53405 & 53711 & 17 & 8\
2252 & 53565 & 53613 & 0 & 0\
2294 & 54524 & 53733 & 63 & 37\
2394 & 54518 & 54551 & 0 & 0\
2474 & 54333 & 54564 & 7 & 1\
2858 & 54498 & 54464 & 1 & 1\
{width="49.00000%"}
Spectrophotometric Recalibration
--------------------------------
We perform a wavelength-dependent spectrophotometric recalibration on all pairs of repeat quasar spectra on each plate, by first producing a ‘recalibration spectrum’ for each pair of observations based on the non-variable stars on each plate. This is done following WI05 with only minor modifications, to allow for faithful comparison to previous work. We note here that to facilitate our difference spectra analysis, we have resampled all spectra and their uncertainties onto a common wavelength grid of the form log$_{10}\lambda = 2.602 + 0.001a$, for integers $a$ from 0 to 1400 ($\lambda$ in Å units). This is approximately a factor of 10 coarser than the actual SDSS spectral resolution, but appropriate for our investigation of the continuum properties of quasars. The resampling is done using a simple linear interpolation, and the resulting common wavelength grid covers 400 to 10,046Å, wide enough for all rest-frame spectra of the quasars in our sample. As part of the interpolation to the common wavelength grid, we mask out problematic pixels in the each spectrum which had SDSS pipeline flags set for NOPLUG, BADTRACE, BADFLAG, BADARC, MANYBADCOLUMNS, MANYREJECTED, NEARBADPIXEL, LOWFLAT, FULLREJECT, SCATTEREDLIGHT, NOSKY, BRIGHTSKY, COMBINEREJ, REDMONSTER [for details on these flags, see @st02]. We consider only pairs of repeat spectra of objects for which $<$20% of pixels are rejected in both epochs.
For each pair of plate observations, all pairs of stellar spectra are selected by requiring the SDSS Spectro1d pipeline classification of both spectra to have CLASS = ‘STAR’ and their SUBCLASS classification to be identical between the two epochs. To remove the stars that have significantly varied between the two epochs, we integrate each pair of stellar spectra and calculate the relative change in flux of the star $\Delta f/f = |(f_1 - f_2)| / (0.5f_1 + 0.5f_2)$, where $f_1$ and $f_2$ are the integrated fluxes in the two epochs. Stars with large $\Delta f/f$ are unsuitable for use in the spectrophotometric recalibration due to their variability. Since $\Delta f/f$ is dependent on the SNR of the spectra, we follow the procedure of WI05 to include the flux uncertainties on the spectra in the variability selection by placing a variability cut on $\Delta f/f$ as a function of the SNR; Figure 2 shows the distribution of $\Delta f/f$ against the SNR of high-SNR epoch spectra for all stars on the 71 plates. We bin the stars in Figure 2 into 13 equally-sized bins of SNR, and calculate the 90th percentile in $\Delta f/f$ in each bin. We then fit an envelope with these 13 points of 90th percentile $\Delta f/f$ to an exponential function of the form $\Delta f/f$ = 0.53exp(SNR/$-0.57)+0.10$. Stars with $\Delta f/f$ below this envelope are considered ‘non-variable’ (Figure 2), and are used in the subsequent spectrophotometric recalibration. There were 6327 stars in total over the 71 plates, 5615 of which were deemed to be non-variable. On a typical plate, a median of 47 non-variable stars were used in the spectrophotometric recalibration.
{width="49.00000%"} {width="49.00000%"}
For each pair of non-variable stellar repeat spectra, we take the ratio of the lower-SNR epoch spectrum to the higher-SNR epoch spectrum; for a non-variable star for which the two epochs of spectra are perfectly calibrated, this results in a flat ratio spectrum with ratio 1. However, plate-wide wavelength-dependent systematic calibration differences between the two epochs may be present. We take the median ratio spectrum of all non-variable stars on each plate and interpolate a 5th-order polynomial to reduce the effects of noise. Prior to the interpolation, we clip the top and bottom 3 percentile of pixels in the ratio spectrum to avoid skewing the interpolation from outlying pixels. The low-SNR epoch spectra of all quasars on each plate will be multiplied by this interpolated median ratio spectrum to match the calibration of the high-SNR epoch spectra. The interpolated median ratio spectra used in the spectrophotometric recalibration are generally a $<$5% correction at all wavelengths, and are almost all $<$10%, consistent with the findings of WI05.
To select all quasars in these 71 pairs of repeat plate observations, we match all spectra to the SDSS DR7 quasar catalog [@schneider10] to find a total of 2,626 quasars, and apply the wavelength-dependent spectrophotometric recalibration for all quasars on each plate. Comparison of repeat spectra for objects that did not significantly vary between the two epochs will be dominated by noise, and so we place a SNR-dependent variability cut on the $\Delta f/f$ for each quasar, similar to the stars, but now to select variable quasars. After binning the quasars into 13 equal bins of SNR, we fit an exponential to the 75th percentile $\Delta f/f$ in each bin of the form $\Delta f/f$ = 0.81exp(SNR/$-0.22)+0.10$. We have used the 75th percentile $\Delta f/f$ in the variability cut rather than the 90th percentile used in the stellar case because quasars are known to be more strongly variable than stars in general [@se07]. Out of 2,626 quasars on the 71 plates, 626 are selected as spectroscopic variables. We note that by design, this sample of quasars we use to study spectral variability are those exhibiting the strongest variability; this is desirable for the present study to ensure high SNR of the spectral variability results.
{width="49.00000%"} {width="49.00000%"} {width="49.00000%"} {width="49.00000%"} {width="49.00000%"} {width="49.00000%"}
Quasar Difference Spectra
=========================
Difference Spectra and Their Properties
---------------------------------------
Before construction of difference spectra using the 626 pairs of variable quasar spectra selected in Section 2, we shift each spectrum to the rest-frame using visually-inspected redshifts from @schneider10. To ensure that the resulting difference spectra are ‘positive’, we subtract the spectral epoch with the lower integrated flux from the higher, with uncertainties added in quadrature. The continuum of each difference spectrum should thus be bluer than either of the individual epochs if these quasars exhibit a bluer-when-brighter trend. We visually inspect all 626 pairs of spectra along with their difference spectra, and find that the continuum of the difference spectra are well-fit by power-laws, and indeed show the bluer-when-brighter trend in the vast majority of cases. In the visual inspection, we identify 11 pairs of spectra which show evidence for strong broad absorption lines (BALs). BAL quasars are known to have atypical continuum properties [e.g., @re03; @gi09], and the BALs are know to exhibit intrinsic variability in their absorption line strengths over long timescales [e.g., @gi08; @gi10; @ca11; @ca12; @fi12; @fi13]. To avoid contamination, we remove these 11 BAL quasars from our sample. We also remove 11 additional quasars identified in the visual inspection for which the variability was clearly dominated by noise. The remaining sample of 604 quasar is the sample for which all further results from our analysis are reported.
We fit the continuum of each difference spectrum to a power-law using a simple $\chi^2$ fit, incorporating the uncertainties in the difference spectra. Although broad emission lines are well-known to be less variable than the continuum (WI05), there is still evidence of emission line variability in our difference spectra. Thus, we mask out the following wavelength regimes dominated by broad emission lines in the continuum fitting: 1360-1446Å (Si IV, O IV\]), 1494-1620Å (C IV), 1830-1976Å (C III\], Fe III), 2686-2913Å (Mg II), 4285-4412Å (H$\gamma$), 4435-4762Å (Fe II), 4760-4980Å (H$\beta$), 4945-4972Å (\[O III\]), 4982-5035Å (\[O III\]), and 5100-5477Å (Fe II), as well as wavelengths $<$1300Å to avoid Ly-$\alpha$ emission and absorption. The choice of these masked regions are informed by the composite SDSS quasar spectrum of @va04; the numerous other lines and line complexes present in quasar optical spectra which we do not mask out tend to be less prominent, and we do not find evidence that these other emission lines significantly affect the continuum fitting in our visual inspections. We clip the top and bottom 1 percentile of pixels in each spectrum after applying these mask (but before the fitting) to avoid strong outliers. After the first power-law fit, we again clip the top and bottom 1 percentile of pixels away from the best-fit power-law, before refitting the final time.
Figure 3 shows a few examples of pairs of quasar spectra and their difference spectra from our sample, for a range in fitted power-law spectral indices $\alpha_\lambda$ (where $F_\lambda$ $\propto$ $\lambda^{\alpha_\lambda}$) of the difference spectra continua. The difference spectra show excellent fits to a simple power-law, and show a strong bluer-when-brighter trend. Figure 4 shows the distribution of the spectral indices of the difference spectra for all 604 quasars, with and without corrections for Galactic extinction using the extinction maps of @sc98 and the reddening law of @ca89. To compare the difference spectra to the underlying spectra, we also calculate power-law spectral indices of the continua from the high-SN epoch in each pair of repeat quasar spectra; this provides a relatively unbiased view of the general spectral properties of each quasar. The fitting of the continua of the spectra to a power-law is performed similar to the difference spectra, but with an additional wavelength region mask of all wavelengths $>$5800Å to avoid contamination from host-galaxy emission [@va04]. As expected, Figure 4 shows that quasar difference spectra are generally bluer than single-epoch quasar spectra, and the addition of corrections for Galactic extinction causes the continuum spectral indices to become even bluer.
{width="49.00000%"}
Composite Spectrum and Difference Spectrum
------------------------------------------
We construct a geometric mean composite spectrum for both the difference spectra as well as the high-SNR single-epoch spectra of the 602 quasar in our sample with corrections for Galactic extinction; the use of a geometric mean to construct the composite spectrum preserves the arithmetic mean power-law spectral indices and extinction of a sample of power-law spectra [@re03], well-suited for our investigation of quasar continuum properties. The composite spectra are created by using our power-law fits to the high-SNR epoch and difference spectra for each pair of spectra, and scaling the flux density of the fitted power-laws at 3062Å (a relatively line-free wavelength and covered by nearly all of the spectra in our sample) to a fixed arbitrary flux density. The spectra and difference spectra themselves are then each rescaled by the same scaling factor as their fitted power-laws, and a geometric mean of all rescaled spectra and difference spectra in each wavelength bin is taken to produce the composites. The geometric mean composite spectrum and composite difference spectrum are shown in Figure 5. The 1$\sigma$ uncertainties on the composite spectrum and difference spectrum is estimated by resampling all the pixel flux densities in each spectrum from a Gaussian with center at the measured flux density and width set to the uncertainty in the flux density. All 604 pairs of spectra are resampled 10$^3$ times, and 10$^3$ composite spectra and composite difference spectra are produced. The 1$\sigma$ uncertainties shown are the 1$\sigma$ spreads in these 10$^3$ resampled composites around the mean.
We fit the continuum of the composite spectrum and composite difference spectrum using the same broad emission line wavelength masks as before to find a power-law spectral index of $\alpha_{\lambda,comp} = -1.56 \pm 0.01$ for the composite spectrum, and $\alpha_{\lambda,diff} =
-2.12 \pm 0.02$ for the composite difference spectrum. Without corrections to each spectrum for Galactic extinction (not shown), the spectral indices are $\alpha_{\lambda,comp} = -1.38 \pm 0.01$ and $\alpha_{\lambda,diff} = -1.94 \pm 0.02$. We note that although the host-galaxy emission that dominates the composite quasar spectrum redward of $\sim$6000Å should not be time-variable, the composite difference spectrum seems to show some host-galaxy emission residuals. This is predominately due to noisy data and poorly-subtracted masked pixels at the red edge of the spectra, as well as the fact that the number of spectra contributing to the composite at wavelengths $>$6000Å is only $\sim$50 out of a total sample of 604. The composite difference spectrum is thus unreliable at $>$6000Å.
Comparison to Previous Studies
==============================
Our analysis thus far has closely followed the work of WI05, and in this section we compare our results to those from previous studies. For the composite quasar spectrum continuum, our calculated spectral index $\alpha_{\lambda,comp} = -1.56 \pm 0.01$ with corrections for Galactic extinction is an exact match to the results of @va04, while our $\alpha_{\lambda,comp} = -1.38 \pm 0.01$ without corrections for Galactic extinction is an excellent match to the $\alpha_{\lambda,comp} = -1.35$ calculated by WI05, all based on SDSS-I/II spectra in a similar range of wavelengths. For the composite quasar difference spectrum continuum, our calculated $\alpha_{\lambda,diff} = -1.94 \pm 0.02$ without corrections for Galactic extinction is also an excellent match to the $\alpha_{\lambda,diff}
= -2.00$ from WI05. The minor discrepancy is likely due to differences in sample size, as our sample of 602 quasars is approximately a factor of two larger than the sample used in WI05 (although of course many objects are in common).
@pe06 utilized the composite difference spectrum generated by WI05 and showed that it is well-fit by synthetic difference spectra generated from a thin disk model which has undergone some change in its global accretion rate. However, difference spectra are subject to both Galactic and intrinsic (host-galaxy) extinction; the composite difference spectrum of WI05 was not corrected for any extinction, and thus should not be directly compared to models. Although we have corrected each spectrum for Galactic extinction in our work, the intrinsic extinction of each quasar is much more difficult to take into account. Thus, our composite difference spectrum is subject to unknown amounts of intrinsic extinction from each of the individual quasars in the sample.
{width=".49\textwidth"}
{width=".49\textwidth"}
To avoid issues with extinction and robustly compare our observed spectral variability to models, we instead consider the spectral variability *relative* to the underlying spectra of the quasars, by dividing the geometric mean composite difference spectrum by the geometric mean composite spectrum from Figure 5. The result, which we call the ‘relative variability spectrum’, is shown in Figure 6 and has power-law spectral index of $\Delta\alpha_\lambda$ $\equiv$ $\alpha_{\lambda,diff}$ $-$ $\alpha_{\lambda,comp}$ = $-0.56 \pm 0.02$. The result of this division of geometric mean composite spectra is independent of any extinction with the reasonable assumption that the extinction does not significantly change between each pair of observations. We note that because broad emission lines are less variable than the continuum, the relative variability spectrum will have inverted emission lines; we have flipped the relative variability spectrum in Figure 6 in the vertical direction, centered on the best-fit continuum to aid in identifying continuum and emission features visually. The spectral variability of quasars relative to their underlying spectra was also investigated in WI05 using the ratio between composite difference spectra to composite spectra of their quasar sample, leading WI05 to conclude that quasars exhibited spectral variability only at wavelengths $<$2500Å. However, this was done by WI05 using arithmetic mean composites, which do not preserve the mean power-law indices of the spectra (making interpretation difficult), and is subject to the effects of extinction. In contrast to WI05, our relative variability spectrum avoids both these issues by using geometric mean composite spectra, facilitating robust comparison to models.
In Figure 6 (middle panel), we compare our observed relative variability spectrum to synthetic relative variability spectra generated from thin disk models in which the global accretion rate has increased by 5%, for a range in characteristic disk temperatures $$\label{equ_tstar}
T^*
\equiv
\left\{ {3 \dot M G M_{BH} \over 8 \pi r_i^3 \sigma_s}
\right\}^{1/4}$$ [@fr02]. The range in $T^*$ in Figure 6 (middle panel) is chosen to span the full range generated for thin disks with log$_{10}L/L_{\rm{Edd}} = [-1.1, -0.8]$, and log$_{10}M_{\rm{BH}} = [8.5, 9.5]$. These ranges in log$_{10}L/L_{\rm{Edd}}$ and log$_{10}M_{\rm{BH}}$ are representative of these values in our sample of 604 quasars, which have median log$_{10}L/L_{\rm{Edd}} = -0.89$ and median log$_{10}M_{\rm{BH}} = 8.83$ M$_\odot$ from the catalog of @sh11. From the spectral variability shown in Figure 6 (middle panel), it is clear that a scenario in which a thin disk changes its global accretion rate cannot account for the strong bluer-when-brighter trend observed in quasars, and it does not produce our observed power-law relative variability spectrum. Our spectral variability evidence against global accretion rate fluctuations as the cause of quasar flux variability is independently in agreement with the argument that quasar accretion disks are too large to vary coherently. We note that although we have shown in Figure 6 (middle panel) difference spectra of a thin disk in which the accretion rate changed by 5%, the shapes of difference spectra from thin disk models with accretion rate fluctuations are not particularly sensitive to the exact change in accretion rate (i.e. the fit will not significantly improve by increasing the change in accretion rate). Instead, the difference spectra from thin disk models depend mainly on the disk’s characteristic temperature, for which we have shown a wide range in Figure 6 (middle panel). A similar conclusion was reached by @pe06.
Aside from our results, the photometric quasar spectral variability study of SC12 also argued against global accretion rate changes as the sole driver in quasar spectral variability. We suggest that the source of the discrepancy between the conclusions of SC12 and @pe06 may be due to their subtly different parameterizations of quasar spectral variability and its effects on the extinction. SC12 fitted linear relations to multi-epoch photometry of quasars in several filters in magnitude-magnitude space, and transformed the relation into color-magnitude space to investigate the variability in different filters. The spectral variability parameter that SC12 compared to models was the slope of the fit in color-magnitude space after the transformation; this slope was fit after all magnitudes for each quasar were normalized to its mean magnitudes for the different filters (see Equation 3 in SC12), and thus parameterizes the spectral variability of each quasar *relative* to its underlying color. This parameterization is similar to the power-law spectral index of the relative variability spectrum ($\Delta\alpha_\lambda$) we calculate, but for photometric colors, and is thus also independent of extinction. Although this subtlety was not discussed in SC12, this may be the source of the conflicting results between SC12 and @pe06 in comparing their observations to global accretion rate fluctuations in a thin disk.
Aside from a simple thin disk, SC12 also compared their spectral variability results to more sophisticated models presented in @da07, finding that their data cannot be explained by accretion rate changes in any of these disk models. SC12 suggested that ephemeral hotspots may be needed to match their observations; in the next section, we compare our observed results to one such time-dependent model of a simple inhomogeneous disk.
Disk Models with Localized Temperature Fluctuations
===================================================
@de11 presented a simple analytic model of a time-dependent inhomogeneous disk, based on a thin disk radiating with independent zones undergoing temperature fluctuations and emitting locally as a blackbody. Aside from predictions of MHD turbulence in simulations of accretion disks, there is now also observational evidence for disks with time-dependent temperature fluctuations from microlensing disk-size measurements, the strong UV spectral continuum, and single-band variability characteristics. @de11 find that to satisfy these observational constrains, accretion disks must be $strongly$ inhomogeneous, with large localized temperature fluctuations. These large temperature fluctuations in the disk inhomogeneities will likely cause the spectrum to be highly variable at short wavelengths, and produce distinct spectral variability with which we will compare our observations.
We set up the inhomogeneous accretion disk model of @de11, starting with a standard thin disk with an inner edge at the innermost stable circular orbit of a non-spinning black hole, and dividing its surface into $n$ zones per factor of two in radius. The zones are roughly equally divided radially and azimuthally, although the exact setup does not noticeably affect the results in our tests. The logarithmic temperature log$_{10}T$ in each zone independently fluctuates as a first-order continuous autoregressive (CAR(1)) process, motivated by studies of single-band variability characteristics [e.g., @ke09]. The mean temperature in each zone is set to the log$_{10}T$ of the thin disk model at that radius, and the constant driving the log$_{10}T$ fluctuations in the CAR(1) process is $\sigma_T$. The characteristic decay timescale of the temperature fluctuations is set to 200 days in the rest-frame, motivated by the observed timescales in @ke09 and @ma10. Our spectral variability results are not sensitive to the choice of this timescale, although we note that if the decay timescale is significantly longer than the time-lag between repeat observations, the quasar will not appear to be significantly variable. All regions in the inhomogeneous disk are assumed to emit locally as a blackbody, and no relativistic effects are considered in this simple model.
We run the inhomogeneous disk model, sampling its spectrum in the wavelength range of 1300-5800Å at 50 day intervals in the rest-frame, after an initial ‘burn in’ time of 500 days to allow the disk to become inhomogeneous. To faithfully compare to our observations, we calculate the change in observable flux between successive time-steps by integrating the spectrum, and calculate the difference spectrum between any two successive time-steps in which the total flux changed by more than 10%, similar to the variability cut placed on the observed spectra in Figure 2. We run the model until we produce $5\times10^3$ synthetic difference spectra, and produce a synthetic geometric mean composite spectrum and composite difference spectrum, by renormalizing each spectrum in the same way as our SDSS spectra.
Figure 7 shows synthetic relative difference spectra generated from the inhomogeneous disk model over a range in independent zones $n$ and temperature fluctuations $\sigma_T$, for the same disk temperatures as shown in Figure 6. The inhomogeneous disk produces relative variability spectra with a characteristic shape that is a power-law at optical wavelengths, which flattens at shorter wavelengths in the UV. For increasing $n$ and $\sigma_T$ in Figure 7, the relative variability spectra remains power-law-like further into the UV before flattening (the physical cause of this is discussed in Section 6.1). Thus, to produce the observed power-law spectral variability in Figure 6 (top panel), the inhomogeneous disk needs to have many independent zones (large $n$) with large temperature fluctuations (large $\sigma_T$).
An example comparison of our observed relative variability spectrum to that from the inhomogeneous disk, using parameters $n = 10^{2.7}$ and and $\sigma_T = 0.45$, is shown in Figure 6 (bottom panel). The large temperature fluctuations and number of fluctuating zones required of inhomogeneous disks to provide such good fits to our observations is in excellent agreement with the $n = 10^{2.5-3}$ and $\sigma_T = 0.35-0.5$ range found by @de11 to simultaneously satisfy independent observational constraints from microlensing disk-size measurements, their strong UV spectral continuum, and single-band variability characteristics of quasars. This independent result based on spectral variability adds to the mounting evidence for large temperature fluctuations in strongly inhomogeneous quasar accretion disks. We note that because the relative variability spectrum is constructed from composite spectra and composite difference spectra of many quasars, each with different Eddington ratios and disk temperatures, the fit of the inhomogeneous disk relative variability spectrum is likely to improve with a more complete consideration of these variations.
{width=".96\textwidth"}
Discussion
==========
The Connection Between Disk Properties and Resultant Spectral Variability
-------------------------------------------------------------------------
Our success of modeling quasar spectral variability in the previous section can be understood as the result of the confluence of large temperature fluctuations in a strongly inhomogeneous disk emitting locally as a blackbody. A flare in a portion of the disk causes the variable part of its optical spectrum to be due to the blackbody emission from the flaring region. For very high-temperature flares, the blackbody spectrum of the flaring region peaks well into the UV, and thus the difference spectrum is dominated by its power-law Rayleigh-Jeans tail. This is the cause of the trend in Figure 7 where the power-law portion of the model relative variability spectrum increasingly extends into the UV as $\sigma_T$ increases. We note that for small $\sigma_T$, the turnover in the UV is a flattening rather than a sharp peak because it is the superposition of the blackbody peaks of numerous flaring regions.
Aside from the large temperature fluctuations, the disk must also be strongly inhomogeneous, with a large number of independently fluctuating regions (i.e. a disk with large $n$). This is due to the fact that the total flux variability amplitude of the disk scales as $N^{-1}$, where $N$ is number of independently varying zones [@de11]; as $n$ increases and the disk becomes more strongly inhomogeneous, the total flux variability decreases. In the case of a disk with small $n$, the difference spectra will be dominated by smaller, lower-temperature flares with blackbody peaks in the optical rather than the UV, and thus the spectral variability will not be dominated by the Rayleigh-Jeans spectrum. Conversely, for large $n$, the disk does not exhibit strong overall flux variability, and only very large, high-temperature flares are actually observable. Thus, the observed flux variability in strongly inhomogeneous disks will be dominated by the Rayleigh-Jeans spectrum, naturally producing power-law spectral variability. This effect causes the power-law portion of the model relative variability spectrum in Figure 7 to extend further into the UV with increasing $n$. We emphasize that this line of evidence for large temperature fluctuations in strongly inhomogeneous disks is independently in excellent agreement with other observational constraints from microlensing disk sizes, their strong UV spectral continuum, and single-band variability properties.
We note that because the temperatures in each independently fluctuating zone is damped in the CAR(1) process and thus cannot increase infinitely, the inhomogeneous disk predicts a flattening in the relative variability spectrum in the UV (e.g. as seen in Figure 7). This flattening is not seen in our difference spectra, which appears to be well-fit by a power-law down to $\sim$1300Å, although the SNR decreases dramatically below $\sim$1500Å. Broadband studies of quasar UV variability have shown that quasars are indeed generally more variable in the UV than optical [@ge13]. However, a more careful investigation of quasar spectral variability from optical to the UV will require contemporaneous optical-UV observations, and will be a fruitful test of the inhomogeneous disk model. We also note that in the simple inhomogeneous disk model of @de11, the global accretion rate is assumed to be constant. This may not be entirely justified, as processes causing the large temperature fluctuations such as the MRI are likely to also induce fluctuations in the accretion rate (e.g. by causing the viscosity to change locally). Although we have ruled out accretion rate fluctuations as the sole driver of quasar spectral variability in Section 4, the observed difference spectra in Figure 5 (middle panel) may still be affected by changes in the accretion rate. In particular, because Figure 6 (middle panel) shows that accretion rate changes produce spectral variability that is particularly strong in the UV, the addition of accretion rate changes to the inhomogeneous disk model may cause difference spectra to become power-law like well into the far UV, further improving the match to the observed UV spectral variability.
Other Possibilities for Difference Spectra
------------------------------------------
The disk models we have considered all assume that the disk emits locally as a blackbody; while this may be a good approximation for the underlying disk spectrum, it is not entirely justified for difference spectra, which instead isolate the variable part of the spectrum. It is possible that the observed power-law difference spectrum is instead at least partially due to non-thermal emission, which could dominate the variable portion of quasar optical/UV spectra. For example, disk inhomogeneities from the photon bubble instability in the magnetized atmosphere of a radiation pressure-dominated disk [@ar92; @ga98; @tu05] may cause the spectrum during flaring epochs to become non-thermal, due the shorter paths in the low gas-density bubbles for photons to diffuse to the photosphere [@ga98], or due to an increase in free-free emission in the high photon-density bubbles [@da09]. Radio-loud quasars have long been suspected to harbor weak or unresolved jets, and the highly-variable jet synchrotron emission can also produce power-law difference spectra in the optical/UV. Future observations of quasar spectral variability across multiple wavelength regimes contemporaneously can help constrain these possibilities.
Variability Correlations with M$_{BH}$ and Luminosity
-----------------------------------------------------
Finally, we note that our findings that quasar spectral variability cannot be driven purely by global accretion rate changes in a thin disk is not at odds with the trends between variability amplitude, black hole mass, and luminosity that many studies have found, and which suggest that the Eddington ratio may be driving these trends. It is still possible that differences in the mean Eddington ratio among different quasars drive their variability properties, a conclusion also reached by SC12. Notably, @ma10 also found that the scaling relation between these quantities is much shallower than that predicted from the Eddington ratio, suggesting additional physics may be necessary to explain the observed trend. Modeling of these trends using inhomogeneous disk models awaits future investigation.
Conclusions
===========
The characteristic flux variability of quasars reflects complex processes in the accretion disk, yet the cause of the variability is still unknown. Aside from single-band variability characteristics, the spectral variability of quasars provides an additional, independent constraint on models of the variability mechanism. Using repeat spectra of quasars in SDSS-I/II, we investigate the optical spectral variability of quasars, which are known to show a bluer-when-brighter trend. After a wavelength-dependent spectrophotometric recalibration of the quasar spectra using non-variable stars observed on the same plates, we construct difference spectra of 602 variable quasars in our sample, thus isolating the variable part of the spectrum. We compare our observations to synthetic difference spectra generated from a thin disk model in which the accretion rate has varied, as well as a simple inhomogeneous disk model with localized temperature fluctuations. In particular, we find the following:
1. Quasar difference spectra appear to be power-laws with spectral indices steeper than their single-epoch spectra, indicating that the vast majority of quasars show a bluer-when-brighter trend. We measure quasar spectral variability using the relative variability spectrum, which is independent of any extinction. We find that accretion rate fluctuations in a thin disk model cannot produce the strong bluer-when-brighter trend observed. This is contrary to the results of some previous investigations, and may be due to the effects of intrinsic and Galactic extinction that were not accounted for in those previous studies.
2. A time-dependent inhomogeneous disk model can produce spectral variability that provides a good match to our observations over optical wavelengths if the disk is strongly inhomogeneous, with large temperature fluctuations. The difference spectra produced by such inhomogeneous disks are dominated by the Rayleigh-Jeans spectrum from the hot ‘flaring’ regions in the optical, and are thus naturally power-laws. The large temperature fluctuations and large number of zones in the disk required to match our observed spectral variability is in excellent agreement with independent observational constraints from quasar microlensing disk-sizes, their strong UV spectral continuum, and single-band flux variability characteristics.
Our spectral variability constraints suggests that future quasar disk models should be time-dependent, and include large temperature fluctuations. Improved global GRMHD simulations of radiative disks will help inform more sophisticated inhomogeneous disk models to compare to observations. Observationally, future time-domain photometric and spectroscopic observations from the Large Synoptic Survey Telescope [@iv08] and the Time-Domain Spectroscopic Survey portion of SDSS-IV will help better constrain these models.
JJR thanks Yusra AlSayyad and James R. A. Davenport for helpful discussions about processing SDSS spectra. Support for JJR was provided by NASA through Chandra Award Numbers AR9-0015X, AR0-11014X, and AR2-13007X, issued by the Chandra X-ray Observatory Center, which is operated by the Smithsonian Astrophysical Observatory for and on behalf of NASA under contract NAS8-03060. Support for EA was provided by NASA 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 NAS 5-26555
Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web Site is http://www.sdss.org/.
The SDSS is managed by the Astrophysical Research Consortium for the Participating Institutions. The Participating Institutions are the American Museum of Natural History, Astrophysical Institute Potsdam, University of Basel, University of Cambridge, Case Western Reserve University, University of Chicago, Drexel University, Fermilab, the Institute for Advanced Study, the Japan Participation Group, Johns Hopkins University, the Joint Institute for Nuclear Astrophysics, the Kavli Institute for Particle Astrophysics and Cosmology, the Korean Scientist Group, the Chinese Academy of Sciences (LAMOST), Los Alamos National Laboratory, the Max-Planck-Institute for Astronomy (MPIA), the Max-Planck-Institute for Astrophysics (MPA), New Mexico State University, Ohio State University, University of Pittsburgh, University of Portsmouth, Princeton University, the United States Naval Observatory, and the University of Washington.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'A. Stroeer'
- 'U. Heber'
- 'T. Lisker'
- 'R. Napiwotzki'
- 'S. Dreizler'
- 'N. Christlieb'
- 'D. Reimers'
bibliography:
- '5564.bib'
date: 'Received date / Accepted date'
subtitle: 'II. Atmospheric parameters of subdwarf O stars'
title: 'Hot subdwarfs from the ESO Supernova Ia Progenitor Survey[^1] '
---
Introduction \[sec:int\]
========================
Hot subluminous stars are an important population of faint blue stars at high galactic latitudes closely related to the horizontal branch. A proper spectral classification of hot subluminous stars is rendered difficult by the diversity of the helium line spectra. They can roughly be grouped into the cooler sdB stars, whose spectra typically display no or only weak helium lines, and the hotter sdO stars, which have a higher helium abundance on average and can even be dominated by helium. The former have recently been studied extensively because they are sufficiently common to account for the UV excess observed in early-type galaxies. Pulsating sdB stars are important tools for asteroseismology [@char04] and sdB stars in close binaries may qualify as Supernova Ia progenitors [@maxt00; @geie06].
Subluminous B stars have been identified as Extreme Horizontal Branch (EHB) stars [@heb86], i.e. they are core helium burning stars with hydrogen envelopes too thin to sustain hydrogen burning (unlike normal HB stars). Therefore they evolve directly to the white dwarf cooling sequence avoiding the Asymptotic Giant Branch (AGB). While the sdB stars spectroscopically form a homogenous class, a large variety of spectra is observed among sdO stars [@heb92; @heb06]. Most subluminous B stars are helium poor, whereas only a relatively small fraction of sdO stars are.
Ever since the pioneering work by @green74, the helium-rich sdO stars were believed to be linked to the evolution of the hydrogen-rich subluminous B stars. Any evolutionary link between subluminous B and O stars, however, is difficult to explain since the physical processes driving a transformation of a hydrogen-rich star into a helium-rich one remain obscure. The convective transformation has been explored by @wese81 as well as by @grot85. While the former found helium convection to occur even at subsolar helium abundances which mixes helium from deeper layers into the photosphere, the latter concluded that a helium driven convection zone develops only in helium-rich atmospheres. If the latter is true, convective transformation would not work. Non-standard evolutionary models were introduced to explain the formation of sdO stars [e.g. @swei97; @bro01; @moe04]. In particular, the [*late hot flasher scenario*]{} predicts that the core helium flash may occur when the star has already left the red giant branch (RGB) and is approaching the white dwarf cooling sequence (delayed He core flash). During the flash, He and C are dredged-up to the surface. Hydrogen is mixed into deeper layers and burnt. The remnant is found to lie close to the helium main sequence, i.e. at the very end of the theoretical extreme horizontal branch.
The fraction of sdB stars in short period binaries (periods less than ten days) is high. @max01 found 2/3 of their sdB sample (drawn from the Palomar Green survey) to be such binaries, whereas somewhat lower fractions of 40% and 48% were found recently for the SPY sample [@napi04] and for a sample drawn from the Edinburgh-Cape catalog [@mora06], respectively. Obviously, binary evolution plays an important role for the formation of sdB stars and possibly also for that of the sdO stars. A recent population synthesis study [@han03] identified three channels to form sdB stars: (i) one or two phases of common envelope evolution, (ii) stable Roche lobe overflow, and (iii) the merger of two helium-core white dwarfs. The latter could explain the population of single stars. The simulations by @han03 cover the observed parameter range of sdB stars [see @lis05 henceforth paper I].
The ESO Supernova Ia Progenitor SurveY [SPY, @spy0] has obtained VLT/UVES spectra for over 1000 white dwarf candidates to test possible scenarios for type Ia supernovae by searching for double degenerate white dwarf binary systems close to the Chandrasekhar mass limit. Many of the target stars of SPY came from the Hamburg ESO survey [@wiso96]. SPY also observed 137 hot subluminous stars that entered the target sample because they were previously classified as white dwarfs. 76 of these stars were now classified as sdB/sdOB, and 58 as O-type subdwarfs [@chri01; @lis03].
The data for our hot subdwarf sample are of unprecedented quality and homogeneity. Spectral analyses of all sdB stars from that sample have already been presented in paper I in order to test evolutionary models, in particular the binary population models of @han03. Two diagnostic tools – the effective temperature (${T_\mathrm{eff}}$) vs. surface gravity (${\log\left(g\right) }$) diagram and the cumulative luminosity function – yielded conflicting results. Moreover, the models of @han03 predicted some EHB stars to be hotter than the sdB stars contained in the sample of paperI. This led us in paper I to the conclusion that their sample of hot subdwarfs may not be sufficiently complete to describe the whole parameter range covered by the simulations, and that it needs to be complemented by a similar analysis of subdwarf O stars.
In this paper we focus on the subdwarf O spectra from the SPY sample and present the spectral analysis of the high resolution spectra using state-of-the-art NLTE model atmospheres. This enables us to address the still open question about the evolutionary status of hot subdwarfs. We combine our results for the SPY sdO stars with those for the SPY sdB stars from paperI to cover the entire stellar atmospheric parameter range with the same high accuracy. This allows for the first time a detailed comparison with evolutionary model predictions, particularly in the regime of hotter temperatures and higher helium abundances than in previous studies of sdB stars alone.
The paper is organised as follows. In section \[sec:data\] we outline the available data and explain our criteria for spectral classification. Section \[sec:nlte\] briefly describes the construction of a new NLTE model atmosphere grid. In section \[sec:results\] we detail the derivation of stellar parameters by fitting model atmospheres to observed spectral lines and present the results in section \[sec:result2\]. Various evolutionary scenarios are tested in section \[sec:obstheo\], leading to a discussion and summary in section \[sec:discuss\].
Observation, data reduction and spectral classification \[sec:data\]
====================================================================
Observation and data reduction
------------------------------
Observations were obtained at the ESO Very Large Telescope with UT2 (Kueyen) equipped with the UV-Visual Echelle Spectrograph (UVES). A slit width of $2\farcs1$ was used, resulting in a resolving power of $18\,500$ (spectral resolution of 0.36Å at ${\mbox{H}\alpha}$) or better. Wavelength coverage of 3300-6650Å is achieved, with gaps at 4500-4600Å and 5600-5700Å [@spy0; @koe01]. For most of the stars, two exposures in different nights were taken, since SPY was originally intended to search for radial velocity (RV) variable objects. The spectra were then reduced with a procedure developed by C. Karl using the ESO MIDAS software package, partly based on the UVES pipeline developed at ESO.
Spectral classification {#sec:class}
-----------------------
For many years it has been attempted to establish a consistent classification of hot subdwarfs [@green86; @jef97; @dril03] that applies to lower resolution spectra. A transfer of these schemes to spectra of higher resolution would be desirable. However, in case of our UVES spectra, we lack the [He [ii]{}]{} 4542Å line due to a wavelength gap between the blue and the red arm of the spectrograph. Since this line plays a crucial role in the classification scheme of @dril03, we cannot apply this scheme to our data. Instead we aim for a less detailed classification strategy.
In the process of classifying all SPY spectra [@lis03], we therefore just separated the spectra into hot white dwarfs (of various subtypes), sdB, He-sdB, sdO, and He-sdO stars by visual inspection and comparison with spectra of prototypical stars. The terms He-sdB/He-sdO were introduced to mark extremely helium-rich sdB/sdO stars [@ahm03; @ahm05]. Note that we did not use the intermediate class of sdOB stars introduced by @bas75 – we subsumed them in the term sdB. For our programme stars in the present paper we continue to use only the general term sdO , in order to prevent any bias from qualitative visual classification entering our quantitative study of atmospheric parameters.
It is worthwhile to note that our classification and the one of @green86 do not agree as becomes apparent from several stars that we have in common (see Table \[tab\_newgrid\] of this paper and Table 1 of paper I). Seven programme stars (four sdB and three sdO) have been misclassified as white dwarfs by @green86. While none of the sdB stars in paper I has been classified as sdO star in the PG survey, six of our sdO stars are classified as sdB in the PG catalog. Five of them turn out to be helium deficient (see section \[sec:result2\]). It is likely that the helium deficient sdO stars are subsumed in the sdB class of the PG catalog. Hence the fraction of sdO stars is underestimated by @green86 . The high spectral resolution of the UVES spectra does not only allow to identify and measure hydrogen and helium lines, but also the metal line spectrum can be investigated. Lines from highly ionized carbon and nitrogen are found in normal O-type stars. For a refined spectral classification we thus used the presence or absence of characteristic absorption lines of carbon and nitrogen to divide our sample into several subclasses (’CN-scheme’). For carbon we used the strongest lines of [C [iii]{}]{} (4650Å, multiplet1, 4070Å, mult.16, and 4186.9Å of mult.18) and of [C [iv]{}]{} (5801.33Å & 5811.98Å, mult.1, and 4658.3Å, mult.8). For nitrogen the strongest lines of [N [iii]{}]{} are used (4640Å, mult.2 and 4858Å, mult.9, 4379.11Å, mult.18,)
We adopt the following classification scheme:
- Stars with [C [iii]{}]{} and/or [C [iv]{}]{} visible in their spectra but with no [N [iii]{}]{} are classified as ’C’.
- Stars with [N [iii]{}]{} and [C [iii]{}]{} and/or [C [iv]{}]{} in their spectra were classified as ’CN’.
- The absence of [C [iii]{}]{} and [C [iv]{}]{} and the presence of [N [iii]{}]{} leads to the classification ’N’.
- With no lines of [C [iii]{}]{}, [C [iv]{}]{}, and [N [iii]{}]{} visible a star was classified as ’0’.
This is not an optimized, quantitative classification scheme, but it is intended to demonstrate that even such a qualitative subdivision already shows a correlation with the atmospheric parameters of our stars, as we shall outline below. The resulting classifications are presented in Table \[tab\_newgrid\]. We checked whether the presence or absence of CN might be significantly influenced by the signal-to-noise ratio (S/N), and found that the S/N distribution of our spectra does not differ for different CN-classes at low and intermediate S/N values, for which it could be crucial. Hence the CN classification scheme is not significantly affected by noise.
Search for companions of sdO stars
----------------------------------
In paper I we found a significant fraction of sdB stars with cool companions (24 out of 76 sdB stars). Therefore we searched for spectroscopic and photometric evidence for such companions amongst our sdO stars.
Only for one subluminous O star, [**HE 1502$-$1019**]{} (alias PG 1502$-$103; EC 15026$-$1019), spectroscopic signatures for a cool companion were found, like the the G-band, and the Mg-triplet. @fer84 classified the companion star as spectral type K0.5. The weakness of the helium lines implies a very low helium abundance of about He/H=0.001. We estimate the effective temperature from a comparison to synthetic spectra (see section \[sec:nlte\]) to be near 45000K .
[**HE 0301$-$3039**]{} turns out to be radial velocity variable and its spectrum also is composite; it indicates that HE 0301$-$3039 is a binary consisting of two sdO stars with spectra dominated by helium lines, the first such system found [@lis04].
[**HE 1200$-$1924**]{} (alias EC 12001$-$1924; Feige 54) shows a helium dominated absorption line spectrum typical of a sdO star. However, Balmer emission lines are superimposed – strong and broad H$\alpha$, strong but narrow H$\beta$ and weak H$\gamma$ and H$\delta$ emission. These emissions do not originate from the photosphere of the sdO star but may stem from a companion star. No spectral absorptions typical for a cool companion can be found. The $B-J$ and $J-K$ colours, however, are unusual for a sdO star (see below).
For the above objects too few spectra are at hand to disentangle the individual spectra and the quantitative spectral analysis of these objects has to be postponed until they become available.
We also searched for photometric evidence for companions by complementing the B-band fluxes with IR measurements from the 2MASS [@2MASS] and DENIS[^2] catalogs. Photoelectric B magnitudes (accurate to better than $\pm 0\fm05$) were taken from the subdwarf data base [@oest06]. If the former were unavailable, B magnitudes are from the Hamburg-ESO survey [@wiso00 accurate to $\pm 0\fm2$].
J magnitudes and corresponding errors are provided by 2MASS for 51 stars, and by DENIS for one star. Only the brightest stars were detected in the K band. A color-magnitude diagram is displayed in Fig. \[fig:bmj\]. Interstellar extinction and reddening have been corrected for using the maps of @schl98. As most of the stars are located at high galactic latitudes, $E(\rm {B-V})$ is small ($\le 0\fm12$) except for the low latitude star HZ 1. For the latter, the reddening as inferred from the Schlegel maps is overestimated, since the star lies near the Galactic plane (z$\approx$80pc). Instead, we compare the measured $B-V$ to model predictions, yielding $E(\rm {B-V})=0\fm25$.
In paper I we found that sdB stars with spectroscopic signatures of cool companions are redder than $B-J \approx -0\fm4$. Similarly, eight sdO stars in Fig. 1 lie at $B-J > -0\fm5$ and clearly separate from the bulk of sdOs. We number those objects in the figure. Among them is HE 1200$-$1924, whose emission lines provide additional support for the existence of a cool companion to that sdO star.
Only for 12 stars, K band magnitudes to better than $\pm 0\fm2$ are available. [@stark03] studied the distributions of $J-K$ colours of sdB and sdO stars and find both to be double-peaked, the sdO stars at $J-K=-0\fm2$ and $+0\fm2$, respectively, indicating that the latter sdO stars have cool companions. $J-K$ measurements are available for five of the eight stars highlighted in Fig. \[fig:bmj\]. All five have positive $J-K$ indicating the presence of cool companions.
Judged from our experience with sdB stars, we can assume the blue spectra of these stars not to be affected by light from the cool companion. Those stars are marked in Table \[tab\_newgrid\] and the quantitative spectral analysis is performed in the same way as for the other stars (see section \[sec:results\]).
Stellar atmosphere modeling {#sec:nlte}
===========================
High quality optical spectra like those from SPY demand accurate theoretical counterparts as input for fitting routines.
Only few detailed quantitative spectral analyses have been carried out for the sdO class because of the complexity of their spectra. In addition deviations from local thermodynamic equilibrium (NLTE) have to be taken into account, because of their high effective temperatures, whereas the LTE assumption is reasonable for sdB stars. Therefore, only few quantitative spectral analyses have been published [@dre90; @the94], with some conflicting results becoming apparent. The reliability of the NLTE calculation depends strongly on the quality and sophistication of the model atom and the atomic data used. The helium model atoms are of utmost importance for the modeling of sdO atmospheres. We used more detailed model atoms for [He [i]{}]{} and [He [ii]{}]{} than in @dre90 and constructed a new grid of atmospheric models and synthetic spectra using a state-of-the-art NLTE model atmosphere code.
An extensive grid of NLTE atmosphere models was calculated using the latest version of the PRO2 code [@wern99] that employs a new temperature correction technique [@drei03]. A new detailed model atom for helium appropriate for the sdO temperature regime was constructed. 2700 partially line blanketed NLTE model atmospheres consisting of hydrogen and helium were calculated resulting in a grid of unprecedented coverage and resolution, extending from 30000 K to 100000 K in T$_\mathrm{eff}$. The gravity ranges from $\log{g}$=4.8 to 6.4 and the helium abundance from $\log{N_{He}/N_H}$=$-$4 to $+$3 in order to match the diversity of observed spectra. The step sizes are 2000 K from 30000 K to 52000 K and 5000 K from 55000 K to 100000 K; 0.2 and $\sim$0.5 dex, respectively.
As a test we carried out the spectral analyses of several programme stars using both the old grid of @dre90 and the new one. The synthetic spectra from the new models match the observed profiles much better in all cases.
The differences of the current model grid compared to the one used by [@dre90] can be traced back to two major improvements: The resonance lines of could only be treated in detailed balance in the previous model grid. This eliminates the transition from the statistical equilibrium equations resulting in a more stable numerical behavior. With an improved numerical stability of the code this approximation is no longer necessary. Since the resonance lines of have a strong influence on the structure of the atmosphere of sdO stars, this improvement is the dominant effect. The second improvement since [@dre90] is the treatment of the level dissociation according to [@humi88]. This mainly affects higher line series members and is therefore important for a precise gravity determination.
Spectral analysis: techniques {#sec:results}
=============================
Line profile fitting \[subsec:fit\]
-----------------------------------
Atmospheric parameters (${T_\mathrm{eff}}$, ${\log\left(g\right) }$, ${\log\left(y\right) }$, y=He/H by number) were determined simultaneously by fitting the synthetic spectra to observed hydrogen and helium line spectra using a $\chi^2$-procedure [@napfit].
For each line, the continuum level is determined and normalized to 1, in order to compare it to the synthetic spectrum. For sdO stars with strong hydrogen Balmer lines, the [H$\alpha$]{} was excluded in the parameter determination because peculiarities in the [H$\alpha$]{} line core were often reported, perhaps caused by stellar winds [@hebwind]. Nevertheless [H$\alpha$]{} was kept in the final plot for visually examining any significant deviations from the model profiles. Those sdO stars showing no evidence for hydrogen from the visual inspection of the [He [ii]{}]{} Pickering decrement required a modified fitting strategy. If hydrogen were present as a trace element it could be best detectable as a contribution of [H$\alpha$]{} to the [He [ii]{}]{} Pickering line 6560Å but may be too weak to be visible to the eye. In a first step ${T_\mathrm{eff}}$ and ${\log\left(g\right) }$ are derived by a fit using all lines including 6560Å. Then a second fit is carried out to derive the hydrogen abundance keeping ${T_\mathrm{eff}}$ and ${\log\left(g\right) }$ fixed from the first run. In a third step new values for ${T_\mathrm{eff}}$ and ${\log\left(g\right) }$ are derived while keeping the hydrogen abundance fixed. If the new values differ by more than 0.5%, we repeat the procedure until convergence is achieved.
Two stars in our sample (DeHt 2 and K2-2) have already been analysed by @napatm with similar NLTE model atmosphere techniques as presented here, and are known to be central stars of planetary nebulae (CSPN). DeHt 2 was classified as a high luminosity CSPN and ${T_\mathrm{eff}}$=117000 K, ${\log\left(g\right) }$=5.64 and ${\log\left(y\right) }$=$-$0.79 were derived, whereas K 2-2 was classified as a peculiar CSPN and ${T_\mathrm{eff}}$=67000 K, ${\log\left(g\right) }$=6.09 and ${\log\left(y\right) }$=$-$1.55 resulted. These stars will not be discussed further in this paper.
For 46 out of the remaining 53 stars the fit procedure converged to convincing line profile fits (as displayed in Fig. \[fit\] for two typical cases). As can be seen the agreement between the best fit synthetic spectra and the observed ones is excellent.
However, for seven stars the fit procedure failed due to various reasons. We shall discuss those stars in some detail below.
[**HE 1349$-$2320**]{} (alias EC 13494$-$2320) and [**HE 1355$-$0622**]{} (PG 1355$-$064): weak He I lines and strong [He [ii]{}]{} lines indicate high temperatures. However, Balmer and [He [ii]{}]{} lines cannot be matched simultaneously. The H$\alpha$ lines display weak emission cores not reproduced by our models.
[**HE 1512$-$0331**]{} [see @chri01] displays very broad Balmer and [He [ii]{}]{} lines with emission cores of [He [ii]{}]{}, 6560Å and 4686Å. Its gravity is too high for our model grid. Extrapolation suggests that its temperature may be near 80000 K and ${\log\left(g\right) }$ slightly below 7. We regard the star as a DAO white dwarf.
[**HE 1518$-$0948**]{} [alias PG 1518$-$098, see @moe90]: Weak He I lines indicate high temperature, while Balmer and [He [ii]{}]{} lines are narrow but strong indicating that it is a hot, helium-rich sdO star of low gravity. In fact, the attempts to fit its spectrum showed that the gravity must be lower than the lower limit of the grid, i.e. ${\log\left(g\right) }$=4.8 at a ${T_\mathrm{eff}}$ near 60000 K.
[**HE 2305$-$1155**]{} displays strong and broad He I lines indicating a low ${T_\mathrm{eff}}$. Trials to match its spectrum indicated a very high gravity beyond the limit of our model grid (i.e. ${\log\left(g\right) }>$6.4).
[**HE 1008$-$179**]{} [see @chri01] is a very hot, high gravity sdO star showing a H$\alpha$ emission core. Individual Balmer lines can not be matched simultaneously [the so-called Balmer line problem, see @napatm]. Attempts to fit its spectrum indicated that its parameters are probably beyond the high temperature, high gravity limit of the model grid, i.e. exceeding ${T_\mathrm{eff}}$=100000 K and ${\log\left(g\right) }$=6.4, and the star should be classified as an extremely hot white dwarf.
[**EC 11481$-$2303**]{} has already been been analysed by @stys00 who derived ${T_\mathrm{eff}}$=41790 K, ${\log\left(g\right) }$=5.84, ${\log\left(y\right) }$=$-$1.85. The authors were unable to match the UV spectral energy distribution. The Balmer line problem is obvious in our optical spectra from the fit, which indicated a higher temperature ($\approx$ 50000 K), lower helium content ($\approx -$2.8) but similar gravity than found by @stys00. The Balmer line problem in very hot white dwarfs was traced back to metal line blanketing by @wern96. @hamm03 analysed the UV spectrum of EC 11481$-$2303 and derived very large iron and nickel abundances of 10 to 100 times solar, causing a strong line blanketing effect. Hence, the neglect of metal line blanketing is the likely cause of our failure to match the optical spectra of this star.
Because of the peculiarities described and problems encountered in the fit procedure, we do not include these stars in the further discussion and are left with a working sample of 46 stars.
Since the SPY project was originally intended to search for radial velocity variable stars, two or more exposures of each star, if available. Thus the final values were calculated as the average of the individual fit results, weighted with the $S/N$ of the corresponding spectra. The final values are presented in Table \[tab\_newgrid\]. We note, that for seven stars of the working sample only a single (useful) exposure was available and denote them by ’S’ in Table \[tab\_newgrid\]. Effective temperatures range from 36000K to 78000K, gravities from ${\log\left(g\right) }$=4.9 to 6.4, and helium-to-hydrogen ratios from 0.0003 to more than 1000.
By adopting a canonical mass of 0.47 ${\mbox{M}_{\odot}}$ for our programme stars, we can further derive the luminosity in solar units (Table \[tab\_newgrid\]).
[llllllllll]{} Object&ICRS 2000.0&B&M$_V$&${T_\mathrm{eff}}$ & ${\log\left(g\right) }$ & ${\log\left(y\right) }$ & $\log(L)$ &CN & Rem.\
& &mag&mag &\[K\] & $[\mathrm{cm\ s^{-2}}]$ & & $[\mathrm{L_{\odot}}]$ &&\
\
HE 1059$-$2735 &11:01:24.95 $-$27:51:42.9&15.21$^{\rm a}$ &3.11& 40966 & 5.38 & $-$3.62 & 2.16 &0& 2\
HE 1130$-$0620 &11:32:41.60 $-$06:36:54.4&15.76$^{\rm a}$ &4.05& 48122 & 5.84 & $-$3.07 & 1.98 &0& 1\
HE 1237$-$1408 &12:39:56.64 $-$14:24:48.4&15.97$^{\rm a}$ &3.02& 39683 &5.33 & $-$2.99 & 2.15 &0& 1,2,5,8\
HE 1318$-$2111 &13:21:15.66 $-$21:27:18.5&14.48$^{\rm a}$ &3.31 & 36254$^{\ast}$& 5.42 & $-$2.91 & 1.92 &0& 2,5\
HE 1115$-$0631 &11:18:11.69 $-$06:47:33.2&14.77$^{\rm a}$ &4.17 & 40443 & 5.80 & $-$2.59 & 1.72 &0& 1\
PG 0026$+$136 &00:28:52.33 +13:54:45.8&15.74$^{\rm 1}$ &3.17 & 38830 & 5.38 & $-$2.39 & 2.07 &0& 1\
EC 09445$-$0905 &09:47:03.39 $-$09:19:50.5&15.69$^{\rm a}$ &4.59 & 73862 & 6.22 & $-$2.08 & 2.28 &0& 2,4\
HE 1423$-$0119 &14:25:51.29 $-$01:33:17.4&16.56$^{\rm a}$ &3.12 & 52662 & 5.50 & $-$1.61 & 2.48 &0& 1\
HE 1513$-$0432 &15:16:19.17 $-$04:43:58.0&16.04$^{\rm a}$ &4.41 & 42699 &5.92 & $-$1.36$^{\ast}$ & 1.69 &0& 1,7\
HE 0040$-$4838 &00:42:31.08 $-$48:22:16.2&16.07&3.79 & 41823 & 5.67 & $-$1.35 & 1.91 &0&\
HE 1047$-$0637 &10:50:28.79 $-$06:53:25.9&14.42$^{\rm a}$ &1.49 & 60650 & 5.03 & $-$1.34 & 3.39 &0& 1,5\
HE 1356$-$1613 &13:59:12.52 $-$16:28:01.8&16.18&4.11 & 55925$^{\ast}$ & 5.93 & $-$1.30 & 2.15 &0& 5\
HE 1106$-$0942 &11:09:08.22 $-$09:58:48.6&16.34&4.92 & 79742$^{\ast}$ & 6.40 & $-$1.03 & 2.27 &0&1\
\
HE 1238$-$1745 &12:41:01.16 $-$18:01:59.0&14.31$^{\rm a}$ &3.83 & 38219 & 5.64 & $-$0.55 & 1.78 &N& 2,5\
HE 1258$+$0113 &13:00:59.23 +00:57:11.7&16.23$^{\rm b}$ &3.80 & 39359 & 5.64 & $-$0.53 & 1.83 &N& 1,5,7\
HE 2218$-$2026 &22:21:13.02 $-$20:11:17.6&16.28&4.40 & 38330 & 5.87 & $-$0.35 & 1.56 &CN&\
HE 1135$-$1134 &11:38:10.66 $-$11:51:03.8&15.45&3.88 & 40079 & 5.68 & $-$0.26 & 1.82 &N& 1\
HE 1136$-$2504 &11:39:10.21 $-$25:20:55.5&13.83$^{\rm a}$ &4.25 & 41381 & 5.84 & $-$0.16 & 1.72 &N& 2,5\
HE 1310$-$2733 &13:12:50.65 $-$27:49:03.1&14.38&3.76 & 40000 & 5.63 & $-$0.15 & 1.87 &N& 2,5\
HE 2359$-$2844 &00:01:38.48 $-$28:27:42.8&15.7 &3.85 & 38325 & 5.65 & $-$0.15 & 1.77 &CN& 3\
PG 2204$+$070 &22:07:16.20 +07:18:36.0&15.74$^{\rm 4}$ &3.72 & 40606 & 5.62 &0.07 & 1.90 &N& 1,4,8\
HE 1256$-$2738 &12:59:01.48 $-$27:54:19.3&16.29$^{\rm a}$ &4.04 & 40029 &5.68 & 0.09 & 1.82 & CN & 2,5\
HE 2203$-$2210 &22:06:29.38 $-$21:56:00.0&15.04&4.95 & 47343 & 6.20 & 0.45 & 1.59 &CN&\
HE 1142$-$2311 &11:44:50.15 $-$23:28:18.0&15.37$^{\rm a}$ &3.86 & 54718 & 5.80 & 0.68 & 2.21 &C$^d$ & 2,5,6\
HE 0111$-$1526 &01:13:38.24 $-$15:11:02.6&14.59$^{\rm a}$ &3.86 & 39152 & 6.31 & 0.83 & 1.16 & CN & 3,8\
HE 1251$+$0159 &12:54:08.35 +01:43:24.0&15.24$^{\rm a}$ &4.47 & 48208 & 5.98 & 1.03 & 1.80 &C& 1\
HE 1511$-$1103 &15:14:17.04 $-$11:14:13.6&14.78$^{\rm a}$ &3.88 & 42298 & 5.68 & 1.10 & 1.91 &CN& 1,5\
HE 1430$-$0815 &14:33:36.93 $-$08:28:24.8&15.69$^{\rm a}$ &2.32 & 61011 & 5.26 & 1.17 & 2.97 &C&1,2,7,8\
HE 1203$-$1048 &12:05:56.59 $-$11:05:29.4&15.69&3.98 & 45439 & 5.91$^{\ast}$ & 1.36$^{\ast}$ & 1.91 &C& 1\
HE 1446$-$1058 &14:49:24.49 $-$11:11:19.0&15.00$^{\rm a}$ &3.91 & 45000 & 5.76 & 1.37 & 1.94 &CN & 2\
HE 0342$-$1702 &03:44:58.82 $-$16:52:42.2&14.75&4.18 & 41914 & 5.78 & 1.40 & 1.79 &N&\
HE 0952$+$0227 &09:55:34.57 +02:12:47.9&14.72&3.84 & 46524$^{\ast}$ & 5.75 & 1.41 & 2.00 &C$^d$ & 1,5\
HE 2347$-$4130 &23:50:19.70 $-$41:14:01.1&15.16&4.07 & 44875 & 5.83 & 1.44 & 1.88 &C&\
HE 0414$-$5429 &04:15:30.23 $-$54:21:58.7&14.60&3.86 & 44678 & 5.76 & 1.57$^{\ast}$ & 1.96 &C$^d$ &\
HE 0914$-$0314 &09:17:15.62 $-$03:53:57.3&14.93&3.97 & 45136 & 5.79 & 1.65 & 1.94 & C$^d$ &1\
BPS CS 22955$-$0024 &20:23:50.26 $-$25:08:28.8&15.50$^{\rm c}$ &4.04 & 44622 & 5.80 & 1.72 & 1.89 &CN& 4\
HE 0155$-$3710 &01:58:01.44 $-$36:56:21.9&15.08&4.24 & 41405 & 5.77 &1.76 & 1.76 &N &3,7\
HE 1136$-$1641 &11:38:54.62 $-$16:58:13.4&14.82$^{\rm a}$ &4.03 & 44646 & 5.80 & 1.81 & 1.88 &C& 2\
HE 0958$-$1151 &10:00:42.64 $-$12:06:00.0&13.81$^{\rm a}$ &3.29 & 44125 & 5.51$^{\ast}$ & 1.85 & 2.18 &C$^d$ & 1,2,4,5\
HE 0016$-$3212 &00:18:53.22 $-$31:56:01.7&14.27&3.73 & 41674 & 5.70$^{\ast}$ & 2.14 & 1.97 & CN & 3,6\
HE 0031$-$5607 &00:34:07.75 $-$55:51:05.9&15.45&5.37 & 41423 & 6.25 & 2.26$^{\ast}$ & 1.31 &N& 8\
HE 1316$-$1834 &13:19:16.93 $-$18:49:52.0&16.24&3.55 & 42811 & 5.56 & 2.32 & 2.05 &N& 2,8\
PG 2258$+$155 &23:00:57.75 +15:48:39.8&15.14$^{\rm c}$ &4.77 & 42084 & 6.08$^{\ast}$ & 2.84$^{\ast}$ & 1.56 &N& 1,4\
PG 1632$+$222 &16:34:16.09 +22:11:40.9&15.13$^{\rm b}$ &4.77 & 39384 &6.16 & 2.84 & 1.31 &C& 1,4,8\
HE 0001$-$2443 &00:04:31.01 $-$24:26:21.1&13.66&4.62 & 40975 & 5.94 & 2.97 & 1.60 &N& 3\
HZ 1 & 04:50:13.52 +17:42:06.2 &12.60$^{\rm a}$ &3.95 & 41344 & 5.68 &3.00 & 1.87 &N& 3,4\
&\
&\
&\
&\
&\
&\
Error determination \[subsec:errors\]
-------------------------------------
The statistical 1-$\sigma$-errors from the fit procedure are typically lower than 100 K, 0.04 dex and 0.04 dex for ${T_\mathrm{eff}}$, ${\log\left(g\right) }$ and ${\log\left(y\right) }$, respectively, as a consequence of the high resolution and the low noise level of our data. Nevertheless we decided to derive a more reliable estimate of the true errors on the basis of the existing two or more exposures of each star, where available. We determined the relative error for ${T_\mathrm{eff}}$, $g$ and $y$ by taking the difference between the fit results of each exposure and dividing by the weighted mean of these fit results (see Section \[subsec:fit\]). Figure \[error\] shows the histogram distribution of the resulting relative errors. We find the 1-$\sigma$-value (sample standard deviation) of these distributions to yield global relative errors of $\Delta {T_\mathrm{eff}}/ {T_\mathrm{eff}}= 0.026$, $\Delta g
/ g = 0.25$ and $\Delta y / y = 0.30$. This corresponds to $\Delta
\log{T_\mathrm{eff}}= 0.011$, $\Delta {\log\left(g\right) }= 0.097$ and $\Delta {\log\left(y\right) }= 0.11$. We note, that some relative errors are particularly large, i.e. more than 2-$\sigma$ (sample standard deviation) of these distributions. These so-called ”outliers” are flagged in Tab. \[tab\_newgrid\].
Spectral analysis: Results, trends and biases {#sec:result2}
=============================================
As outlined in @ede03 and paper I we shall search for trends in parameter space. We shall discuss the distributions of gravity, temperature, and helium abundance, taking into account the CN-classification scheme (see section \[sec:class\]). As systematic differences in atmospheric parameters lead to selection biases, we discuss this issue at the end of this section.
Atmospheric parameters and C & N line strengths
-----------------------------------------------
The following conclusions can be drawn from the distribution of stars in the [${T_\mathrm{eff}}$-${\log\left(y\right) }$]{}-diagram (Fig. \[teffloghe\]):
\(i) SdO stars with a helium abundance below solar, ranging from $0.0003
\le He/H \le 0.08$, are scattered in a wide range of effective temperatures from $35\,\rm{kK} \la {T_\mathrm{eff}}\la
80\,\rm{kK}$, and no correlation can be found.
\(ii) SdO stars with a helium abundance exceeding the solar value, ranging from $0.1\la He/H \la 1$, tightly cluster around ${T_\mathrm{eff}}\approx 40\,\rm{kK}$ and He/H$\approx$0.5.
\(iii) Above a helium abundance of He/H=10 the sdO stars show a clear dependence of the helium abundance on temperature: the helium abundance *decreases* with increasing ${T_\mathrm{eff}}$, opposite to the general trend for sdB stars.
Most strikingly a clear correlation between helium abundance and CN class becomes apparent from Fig. \[teffloghe\]. None of the sdO stars with subsolar helium content shows carbon and/or nitrogen lines and therefore all of them are of CN class 0. The opposite is true for sdO stars with supersolar helium content – all of them show carbon and/or nitrogen lines and therefore are either of class C, N, or CN.
This suggests that the sdO stars should be grouped according to helium content into two classes. Those with supersolar helium abundances will further on be referred to as *helium-enriched* sdO stars while those with subsolar helium abundances will be termed *helium-deficient* sdO stars.
Figs. \[tefflogg\_sdO\] and \[tefflogg\_hesdO\] show the distribution of the full SPY sample of sdO stars in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-plane, in comparison with the sdB stars from paper I. The location of the EHB band [@dor93] is shown, along with the helium zero-age main sequence (He-ZAMS) which indicates the boundary beyond which no stable helium core burning can be established [@pac71].
Both *helium-deficient* (Fig. \[tefflogg\_sdO\]) and *helium-enriched* (Fig. \[tefflogg\_hesdO\]) sdO stars are found at higher temperatures than the sdB stars, most of them between 37 kK and 47 kK. They clearly lie outside the EHB band, which is defined as the region between the zero-age EHB (ZAEHB) and the terminal age EHB (TAEHB) given by the evolutionary calculations of @dor93. A significant fraction (7 out of 33) of *helium-enriched* sdOs appear to lie below the helium main sequence. In view of the non-Gaussian distribution of errors (see Sect. \[error\]), their gravities might have been overestimated. Indeed, two stars are amongst the ”outliers” and the results of another three are based on single spectra. It may therefore be premature to draw reliable conclusions. Additional observations are necessary to derive more precise gravities.
While the *helium-deficient* sdO stars are scattered in a wide [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-range, most *helium-enriched* sdOs populate a relatively narrow region $({T_\mathrm{eff}}$ from $\sim$40 to $\sim$46kK and ${\log\left(g\right) }$ from $\sim$5.5 to $\sim$5.9).
Selection bias \[sec:select\]
-----------------------------
The target objects of SPY were selected from a magnitude limited list of candidate white dwarfs. Therefore all subdwarfs present in the SPY dataset have only been included “by accident”, because they were classified erroneously as white dwarfs mostly from low resolution objective prism spectra. For the sdB stars, we found in paper I that any selection effects cannot be significantly different from typical selection effects present in earlier studies of sdB stars. Besides the usual biases of magnitude limited surveys there may be another selection effect introduced by the SPY observing strategy. As the SPY project aimed at including a significant fraction of helium-rich white dwarfs, preference was given to DB candidates, many of which turned out to be sdO stars. There may thus be a bias in favour of selecting sdO stars *relative to sdB stars* in the SPY project. This must be taken into account when the combined sdB and sdO sample from SPY is compared to predictions from evolutionary calculations.
Evolutionary status \[sec:obstheo\]
===================================
As outlined above, the distribution of the programme stars in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram (see Figs. \[tefflogg\_sdO\] and \[tefflogg\_hesdO\]) as well as with respect to CN class suggests that *helium-enriched* sdO stars form a population different from the *helium-deficient* sdO stars. Published evolutionary scenarios try to explain the origin of sdB and sdO stars either by canonical or non-canonical evolution of single stars or by close binary evolution with mass exchange and common envelope episodes. We now test these scenarios by comparing our observational results to their predictions.
Canonical evolution of single stars
-----------------------------------
Canonical EHB models [e.g. @dor93] are characterized by a core mass of nearly half a solar mass and a tiny inert hydrogen-rich envelope. The core mass is fixed by the onset of the core helium flash at the tip of the red giant branch and depends only slightly on metalicity and helium abundance. Hence the canonical core mass is restricted to a very narrow range of 0.46 to 0.5 ${\mbox{M}_{\odot}}$. This configuration prevents an EHB star from ascending the AGB. The post-EHB evolution proceeds towards higher temperatures until the white dwarf cooling track is reached and gravity increases.
The problem for the formation of EHB stars in this case is, how almost the entire envelope of the RGB progenitor is lost at precisely the same time as the core reaches the mass required for the helium flash. Enhanced mass-loss during or after the red giant branch has been postulated [@dcr96] but no physical mechanism has yet been identified.
In Fig. \[cI\] we compare the distribution of sdO stars to the position of the EHB-band and to post-EHB evolutionary tracks and find that none of the programme stars lies on the EHB-band. However, it is premature to conclude that they have evolved from the EHB. Most of the *helium-enriched* sdO stars cluster in a narrow region of the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram. The calculations of @dor93 indicate that the pace of evolution does not change very much through post-EHB evolution. Hence post-EHB evolution can not explain the clustering of *helium-enriched* sdO stars. Moreover, the post-EHB scenario fails to explain the surface enrichment of helium and it also can not explain why the C and/or N lines in *helium-enriched* sdO stars are stronger than *helium-deficient* ones, since no dredge-up process is predicted to occur.
The *helium-deficient* sdO stars, however, can well be explained by the post-EHB hypothesis as they have similarly low helium abundances as sdB stars. In addition they are scattered in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram, hence no slow-down in their evolution has to be invoked. Therefore it is reasonable that *helium-deficient* sdO stars are post-EHB stars and have evolved from sdB stars, while *helium-enriched* sdO stars have not.
Some of the more luminous sdO stars may not be related to the EHB at all but may have evolved off the AGB. Therefore we included post-AGB evolutionary tracks from @schoe79 [@schoe83] in Fig. \[cI\]. The position of the *helium-deficient* sdO HE 1047$-$0637 (${T_\mathrm{eff}}\approx 60\,000K, {\log\left(g\right) }\approx 5.0$) is matched by these tracks, indicating a possible post-AGB evolutionary stage while the *helium-enriched* sdO HE 1430$-$0815 (${T_\mathrm{eff}}\approx 61\,000K, {\log\left(g\right) }\approx 5.3$) may be either a post-AGB or a post-EHB star. The short evolutionary timescales [about 30000 years from the AGB towards a pre-white dwarf, @schoe79; @schoe83] drastically reduce the probability of finding true post-AGB stars.
Non-canonical evolution of single stars
---------------------------------------
Since canonical single star evolution can not explain the *helium-enriched* sdO stars, in particular their clustering in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-plane, the helium enrichment and the C and/or N line strengths, other scenarios have to be investigated, such as the late hot flasher scenario, in which the core helium flash occurs when the star has already left the RGB and is approaching the white dwarf cooling sequence (delayed helium core flash). During the flash, He and C is dredged-up to the surface [@swe97].
Fig. \[latehe\] compares an evolutionary track for the late hot flasher scenario with the distribution of our [*h*elium-enriched]{} sdO stars. The star encountered a late flash on its way from the RGB towards the white dwarf regime. This late hot flash forces the star to land on or near the helium main sequence, i.e. at the extreme end of the EHB. The final composition of the envelope is helium dominated with Y=0.814, complemented by hydrogen (X=0.154) and enriched with carbon at Z=0.032 [or nitrogen if the hydrogen burning during the helium flash phase burns $^{12}$C into $^{14}$N; @swe97]. Indeed, most of our observed *helium-enriched* sdO stars lie near the model track, suggesting that they indeed may originate from this scenario [see also @lemk97]. However, the evolutionary time scales (1.95$\times 10^6$ yrs for the evolution shown in Fig. \[latehe\]; @swe97) are much shorter than for the core helium burning phase. Accordingly the stars should accumulate near the end of the track, i.e. near the helium main sequence, which is not the case for our programme stars.
Although the late hot flasher scenario can explain the helium enrichment and the line strengths of C and/or N lines as due to dredge up, it fails to reproduce the distribution of the stars in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram in detail.
Binary evolution \[sec:evol\_hesdo\]
------------------------------------
As the evolutionary scenarios for single stars discussed above partially fail to explain the observed properties of *helium-enriched* sdO stars, we now focus on binary evolution. Two flavours of close binary evolution have been envisaged to explain sdB and sdO stars. The formation of core helium burning EHB stars through close binary evolution has extensively been investigated by @han02 [@han03] using the binary population synthesis approach.
However, it may not be taken for granted that hot subluminous stars are indeed core helium burning. They may form from red giants that left the RGB before igniting helium in the core and evolve through the EHB region as helium stars towards the white dwarf cooling sequence, see @heb03bb. Such progenitors of helium core white dwarfs, indeed, have been discovered. @heb03a found that the sdB star HD 188112 has a mass of 0.23${\mbox{M}_{\odot}}$, too small to sustain helium burning.
### Binary evolution scenarios and the role of white dwarf mergers
@han03 [hereafter HPMM] showed in their binary population synthesis study that three channels are relevant for the formation of hot subluminous stars involving either common-envelope ejection, stable Roche lobe overflow or a merger of two helium white dwarfs. Stable Roche lobe overflow is predicted to lead to composite spectrum systems consisting of a hot subdwarf and a main sequence star. There are eight such binaries (sdO + main sequence star) present amongst our programme stars.
In paper I, we compared the atmospheric parameters of the SPY sdB stars to the HPMM models by using two diagnostic tools, namely the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram and the cumulative luminosity function. Our analysis of sdO stars extends the subdwarf sample to higher temperatures which may allow to study the link between sdB and sdO stars. One would thus expect a comparison of our full sdB/sdO sample with the HPMM models to yield more robust results than in paper I. However, we must observe selection biases as discussed in section \[sec:select\].
We present this comparison for the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram in Fig. \[hpmm1\]. The simulation set No. 10 of HPMM was chosen because it came closest to the SPY-sdB distribution (see paper I). The grey shading of the rectangular areas corresponds to the respective number of simulated stars they contain (cf. paper I). Higher number densities of simulated subdwarfs correspond to darker grey shading. We refer the reader to paper I for more details about the general specifications of the HPMM simulations.
From the direct comparison of our derived [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-values to the HPMM simulations, we see two effects. First, sdO stars significantly exceed even the hottest temperatures that result from any HPMM simulation set, with stars reaching up to 80 kK. Even for the sdO stars with lower temperatures, no set is able to reproduce their rather wide range in surface gravity. HPMM therefore covers only $\le 38$% of all sdO stars. Second, by restricting our analysis to stars which come close to the HPMM predictions, i.e. those that are apparently connected with the sdB sample, a strong disagreement of the observational data with the simulation set becomes obvious: the relative amount of hot (sdO) and cool (sdB) stars differs significantly. Apart from possible limitations of the HPMM models (as discussed below in section \[sec:discuss\]), there may be an observational bias in favour of sdO stars, rendering a *quantitative* comparison with HPMM, as performed for the sdB stars in paper I, very difficult or even impossible.
Let us now focus on the white dwarf merger channel, which is supposed to result in stars with very low hydrogen envelope mass. Hence stars resulting from mergers of two helium-core white dwarfs are expected to be found near the hot end of the HPMM distribution. Mergers can also produce stars in a wider range of masses than the other channels predict. The merging process probably induces lots of mixing of nuclear processed material to the stellar surface of the remnant, potentially leading to a helium- and nitrogen-rich surface composition. Depending on the efficiency of nuclear burning and mixing, carbon may also be enriched at the surface. As these predictions match the observed properties of the *helium-enriched* sdO stars at least qualitatively, we regard the merger of helium core white dwarfs as a viable scenario. There are two predictions we can test: (i) Stars formed from a merger should not be radial velocity variable (unless they stem from triple systems) and (ii) the helium enrichment is accompanied by enrichment of nitrogen and/or carbon. While our CN-classification scheme (section \[sec:class\]) is based on the *presence* of C and/or N-lines in the spectrum, the latter issue requires a *quantitative* abundance analysis. We shall address these tests in forthcoming papers.
### Non-core helium burning stars in close binary systems
The evolution of RGB stars whose envelopes get almost completely stripped by Roche lobe overflow in a close binary system *before* helium burning starts has been investigated e.g. by @dri98. These tracks were calculated from a 1${\mbox{M}_{\odot}}$ model sequence starting from the pre-main sequence stage up through the RGB. Large mass loss rates were then adopted and the evolution of the resulting helium star was followed. The remnant finally evolves into a helium core white dwarf.
In Fig. \[driebe\] we compare the programme stars to the predictions of the evolutionary models of @dri98 for different masses. While the position in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram of any of our programme stars can be matched by a post-RGB track of appropriate mass, it is striking that most of the *helium-enriched* sdO stars agree reasonably well with the theoretical predictions for a rather narrow mass range, i.e. between 0.3M$_{\odot}$ and 0.33M$_{\odot}$. Hence the distribution of *helium-enriched* sdO stars could be a sequence of low mass stars evolving into helium white dwarfs. It is also surprising that the predicted masses are close to the minimum mass for the helium main sequence, which might be purely coincidental.
However, the models of @dri98 predict rather thick hydrogen layers. Therefore, this scenario has difficulties to explain the high helium abundances at the stellar surface as observed in *helium-enriched* sdO stars. During the post-RGB evolution hydrogen-shell flashes occur which possibly may lead to a dredge up of helium and nitrogen [@dri99]. This may explain nitrogen strong-lined objects. It is, however, not evident how carbon-strong objects could be formed. Moreover, the models of @dri99 indicate that the occurrence of hydrogen flashes is constrained to a lower mass range of 0.21 to 0.3${\mbox{M}_{\odot}}$.
If the hydrogen were removed from the envelope by some unknown process, the evolution would be speeded up considerably, because most of the luminosity is provided by the hydrogen burning shell, which would be extinguished if the hydrogen mass is too low. Rapid evolution reduces the detectability of such stars.
Therefore we regard the post-RGB scenario as unlikely. However it is a viable scenario to explain some of the helium-*deficient* sdO stars since helium core white dwarf progenitors are known albeit rare amongst the sdB stars [e.g. HD 188112, @heb03a].
Summary and Discussion \[sec:discuss\]
======================================
We analyzed high resolution optical spectra of 58 subluminous O stars. We found spectroscopic and/or photometric evidence for cool companions to eight subluminous O stars. While HE 1502$-$1019 and HE 1513$-$0432 display only weak helium lines, the others have strong helium lines. HE 0301$-$3039 is an unique binary consisting of two sdO stars with spectra dominated by helium lines [@lis04]. Evidence for cool companions to the other helium-enriched stars rests on photometry only. Near-infrared spectroscopy is required for classifying them. As the photometric measurements for our sample are only available for 52 out of 58 stars, the fraction of sdO stars with cool companions may be slightly larger.
A grid of synthetic H-/He-line spectra calculated from NLTE model atmospheres was used to derive the stars’ atmospheric parameters. Reliable atmospheric parameters were obtained for 13 *helium-deficient* sdO and 33 *helium-enriched* sdO stars. A clear-cut correlation between CN class and helium abundance was found. *Helium-deficient* sdO stars did not show any C and/or N lines, while all *helium-enriched* sdO stars do show either C or N lines or both, suggesting that they form two different populations of stars. The *helium-deficient* sdO stars are scattered in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{} diagram, whereas most *helium-enriched* sdO stars cluster in a narrow range.
Comparing the observed distribution to the predictions of evolutionary calculations for single as well as for close binary stars, we conclude that many *helium-deficient* sdO stars can be explained as evolved sdB stars. Both classes of star are helium deficient and have weak metal lines which is caused by atmospheric diffusion processes.
Most of the *helium-enriched* sdO stars cluster in a narrow region of the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{}-diagram at temperatures between 40kK and 50kK. While diffusion is probably causing helium deficiency, it is unlikely to account for the helium enrichment. Non-standard evolutionary scenarios were therefore considered as well. The predictions from the late hot flasher scenario as well as the helium white dwarf merger scenario are roughly consistent with the observed distribution of *helium-enriched* sdO stars but do not match them in detail. The occurrence of a delayed helium core flash as well as the merger of two helium white dwarfs may explain the helium enrichment. In both cases carbon and/or nitrogen can be dredged up to the stellar surface, which would explain the strength of the C and/or N lines in *helium-enriched* sdO stars.
Some high gravity *helium-enriched* sdO stars may lie below the helium main sequence, which is at variance with any core helium burning model. Therefore, we considered models for post-RGB stars with inert helium cores which evolve through the sdB/sdO regime in the [${T_\mathrm{eff}}$-${\log\left(g\right) }$]{} diagram into helium core white dwarfs. This is the only scenario that can explain stars to lie below the helium main sequence. However, it is not obvious how the helium enrichment is brought about as the hydrogen envelopes of post-RGB models has to be relatively thick or, otherwise, the stars would evolve too fast to be observable in large quantities. If the existence of a population of high gravity sdO stars below the helium main sequence could be confirmed by surveys with larger sample size, a possible explanation could be provided by the post-RBG scenario.
Our conclusions can be tested by measuring the binary frequency and the C and N abundances. If *helium-deficient* sdO stars are evolved sdB stars, their binary fractions should be the same. As about 40% of the sdB stars are in close binaries with periods below 10 days [@napi04], we expect 5 *helium-deficient* sdO stars in our sample to be radial velocity variable on time scales of 10 days or less. An investigation of the binary frequency is underway in our group. If *helium-enriched* sdO stars result from mergers, no radial velocity variations would be expected (except for objects arising from triple stars). If material processed by nuclear burning is dredged up, the helium enrichment is expected to be accompanied by enrichment of nitrogen and/or carbon. While our CN-classification scheme (section \[sec:class\]) is based on the *presence* of C and/or N-lines in the spectrum, the latter issue requires a *quantitative* abundance analysis to test the late hot flasher and the merger scenario.
In addition, improvements to our analyses of the SPY-sample are required for a more detailed comparison with predictions of evolutionary calculations. The role of metal-line blanketing in the NLTE model atmospheres needs to be investigated. Due to the complexity of the problem, only few studies of NLTE metal line blanketing in sdO star atmospheres are available [@haas96; @lanz97; @deet00] which indicate that the effective temperatures may have to be reduced. According to these investigations, the most important contribution to line blanketing stems from the iron group elements. However, their abundances can be derived from ultraviolet spectra only. The observed diversity of carbon and nitrogen line strengths indicates that the treatment of metals has to be done carefully on a star-by-star basis.
Apparently, the SPY sample suffers from observational selection biases. We found it likely that sdO stars were selected preferentially for the SPY target list. Hence sdO stars may be overrepresented with respect to sdB stars. To this end the Sloan Digital Sky Survey (SDSS) is a promising source of hot subdwarf stars, as its selection criteria are very different from that of SPY. We have already begun a spectroscopic analysis of SDSS sdO stars [@heb06] which will contribute significantly towards discriminating between the various evolutionary scenarios and hypotheses outlined above.
T.L. gratefully acknowledges support by the Swiss National Science Foundation. R.N. is supported by a PPARC Advanced Fellowship. We thank Iris Traulsen and Thomas Rauch for their help in running the model atmosphere codes, Heiko Hirsch for double checking the CN-classification, and Roy [Ø]{}stensen for creating the subdwarf data base [@oest06], which we used extensively. 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 [@2MASS]. This publication makes use of the VizieR database of astronomical catalogs [@Vizier].
[^1]: Based on observations collected at the Paranal Observatory of the European Southern Observatory for program No. 165.H-0588(A) and 167.D-0407(A).
[^2]: http://cdsweb.u-strasbg.fr/denis.html
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The inclusion of a macroscopic adaptive threshold is studied for the retrieval dynamics of layered feedforward neural network models with synaptic noise. It is shown that if the threshold is chosen appropriately as a function of the cross-talk noise and of the activity of the stored patterns, adapting itself automatically in the course of the recall process, an autonomous functioning of the network is guaranteed. This self-control mechanism considerably improves the quality of retrieval, in particular the storage capacity, the basins of attraction and the mutual information content.'
author:
- 'D. Bollé and R. Heylen'
title: Adaptive Thresholds for Layered Neural Networks with Synaptic Noise
---
Introduction
============
As is common knowledge by now, layered feedforward neural network models are the workhorses in many practical applications in several areas of research and, therefore, any new insight in their capabilities and limitations should thus be welcome. In view of the fact that in many of these applications, e.g., pattern recognition in general, information is mostly encoded by a small fraction of bits and that also in neurophysiological studies the activity level of real neurons is found to be low, any reasonable network model has to allow variable activity of the neurons. The limit of low activity, i.e., sparse coding is then especially interesting. Indeed, sparsely coded models have a very large storage capacity behaving as $1/(a\ln a)$ for small $a$, where $a$ is the activity (see, e.g., [@W; @P; @Ga; @Ok] and references therein). However, for low activity the basins of attraction might become very small and the information content in a single pattern is reduced [@Ok]. Therefore, the necessity of a control of the activity of the neurons has been emphasized such that the latter stays the same as the activity of the stored patterns during the recall process. This has led to several discussions imposing external constraints on the dynamics of the network. However, the enforcement of such a constraint at every time step destroys part of the autonomous functioning of the network, i.e., a functioning that has to be independent precisely from such external constraints or control mechanisms. To solve this problem, quite recently a self-control mechanism has been introduced in the dynamics of networks for so-called diluted architectures [@DB98]. This self-control mechanism introduces a time-dependent threshold in the transfer function [@DB98; @BDA00]. It is determined as a function of both the cross-talk noise and the activity of the stored patterns in the network, and adapts itself in the course of the recall process. It furthermore allows to reach optimal retrieval performance both in the absence and in the presence of synaptic noise [@DB98; @BDA00; @BH04; @DKTE02]. These diluted architectures contain no common ancestors nodes, in contrast with feedforward architectures. It has then been shown that a similar mechanism can be introduced succesfully for layered feedforward architectures but, without synaptic noise [@BM00].
The purpose of the present contribution is to generalise this self-control mechanism for layered architectures when synaptic noise is allowed, and to show that it leads to a substantial improvement of the quality of retrieval, in particular the storage capacity, the basins of attraction and the mutual information content.
The model
=========
Consider a neural network composed of binary neurons arranged in layers, each layer containing $N$ neurons. A neuron can take values $\sigma_{i}(t) \in
\{0,1\}$ where $t=1,\ldots,L$ is the layer index and $i=1, \ldots ,N$ labels the neurons. Each neuron on layer $t$ is unidirectionally connected to all neurons on layer $t+1$. We want to memorize $p$ patterns $\{\xi_i^\mu(t)\},
{{i=1,\ldots,N}, ~{\mu=1,\ldots,p}}$ on each layer $t$, taking the values $\{0,1\}$. They are assumed to be independent identically distributed random variables (i.i.d.r.v.) with respect to $i$, $\mu$ and $t$, determined by the probability distribution: $p(\xi_i^\mu (t))=a\delta(\xi_i^\mu (t)-1)+(1-a)\delta(\xi_i^\mu (t))$. From this form we find that the expectation value and the variance of the patterns are given by $
E[\xi_i^\mu (t)]=E[\xi_i^\mu (t)^2]=a~.
$ Moreover, no statistical correlations occur, in fact for $\mu\neq\nu$ the covariance vanishes.
The state $\sigma_{i}(t+1)$ of neuron $i$ on layer $t+1$ is determined by the state of the neurons on the previous layer $t$ according to the stochastic rule $$\label{eq:stoc}
P(\sigma_{i}(t+1)\mid \sigma_{1}(t), \ldots ,\sigma_{N}(t))
= \{1+\exp[2(2\sigma_i(t+1)-1) \beta{h_i(t)}]\}^{-1}.$$ The right hand side is the logistic function. The “temperature" $T=1/\beta$ controls the stochasticity of the network dynamics, it measures the synaptic noise level [@HKP91]. Given the network state $\{\sigma_i(t)\};{i=1,\ldots,N}$ on layer $t$, the so-called “local field" ${h_i(t)}$ of neuron $i$ on the next layer $t+1$ is given by $$\label{eq:h}
h_i(t)= \sum_{j=1}^{N}
J_{ij}(t)(\sigma_j(t) -a)-\theta(t)$$ with $\theta(t)$ the threshold to be specified later. The couplings $J_{ij}(t)$ are the synaptic strengths of the interaction between neuron $j$ on layer $t$ and neuron $i$ on layer $t+1$. They depend on the stored patterns at different layers according to the covariance rule $$\label{eq:j}
J_{ij}(t)=\frac{1}{N {a(1-a)}} \sum_{\mu=1}^{N}
(\xi_i^\mu (t+1)-a)(\xi_j^\mu (t)-a)~.$$ These couplings then permit to store sets of patterns to be retrieved by the layered network.
The dynamics of this network is defined as follows (see [@DKM]). Initially the first layer (the input) is externally set in some fixed state. In response to that, all neurons of the second layer update synchronously at the next time step, according to the stochastic rule (\[eq:stoc\]), and so on.
At this point we remark that the couplings (\[eq:j\]) are of infinite range (each neuron interacts with infinitely many others) such that our model allows a so-called mean-field theory approximation. This essentially means that we focus on the dynamics of a single neuron while replacing all the other neurons by an average background local field. In other words, no fluctuations of the other neurons are taken into account. In our case this approximation becomes exact because, crudely speaking, $h_{i}(t)$ is the sum of very many terms and a central limit theorem can be applied [@HKP91].
It is standard knowledge by now that mean-field theory dynamics can be solved exactly for these layered architectures (e.g., [@DKM; @B04]). By exact analytic treatment we mean that, given the state of the first layer as initial state, the state on layer $t$ that results from the dynamics is predicted by recursion formulas. This is essentially due to the fact that the representations of the patterns on different layers are chosen independently. Hence, the big advantage is that this will allow us to determine the effects from self-control in an exact way.
The relevant parameters describing the solution of this dynamics are the [*main overlap*]{} of the state of the network and the $\mu$-th pattern, and the [*neural activity*]{} of the neurons $$\label{M(t)}
M^\mu(t) =
\frac{1}{N{a(1-a)}}\sum_{i=1}^N{(\xi_i^\mu(t)-a)}
(\sigma_i(t) -a),
\qquad
q(t) =
\frac{1}{N}\sum_{i=1}^N \sigma_i(t)~.$$
In order to measure the retrieval quality of the recall process, we use the mutual information function [@DB98; @BDA00; @NBP98; @ST98]. In general, it measures the average amount of information that can be received by the user by observing the signal at the output of a channel [@B90; @S48]. For the recall process of stored patterns that we are discussing here, at each layer the process can be regarded as a channel with input $\xi_i^\mu(t)$ and output $\sigma_{i}(t)$ such that this mutual information function can be defined as [@DB98; @B90] $$\label{eq:inf}
I(\sigma_i(t);\xi_i^\mu (t))=
S(\sigma_i(t))-\langle S(\sigma_i(t)|\xi_i^\mu (t))\rangle
_{\xi^{\mu}(t)}$$ where $S(\sigma_i(t))$ and $S(\sigma_i(t)|\xi_i^\mu (t))$ are the entropy and the conditional entropy of the output, respectively $$\begin{aligned}
\label{eq:en}
S(\sigma_i(t))&=& -\sum_{\sigma_i} p(\sigma_i(t))\ln[p(\sigma_i(t))]\\
\label{eq:enc}
S(\sigma_i(t)|\xi_i^\mu (t))&=&
-\sum_{\sigma_i} p(\sigma_i(t)|\xi_i^\mu (t))
\ln[p(\sigma_i(t)|\xi_i^\mu (t))]~.\end{aligned}$$ These information entropies are peculiar to the probability distributions of the output. The quantity $p(\sigma_i(t))$ denotes the probability distribution for the neurons at layer $t$ and $p(\sigma_i(t)|\xi_i^\mu (t))$ indicates the conditional probability that the $i$-th neuron is in a state $\sigma_i(t)$ at layer $t$ given that the $i$-th site of the pattern to be retrieved is $\xi_i^\mu (t)$. Hereby, we have assumed that the conditional probability of all the neurons factorizes, i.e., $p(\{\sigma_i(t)\}|\{\xi_i(t)\})=\prod_j p(\sigma_j(t)|\xi_j(t))$, which is a consequence of the mean-field theory character of our model explained above. We remark that a similar factorization has also been used in Schwenker et al. [@SSP96].
The calculation of the different terms in the expression (\[eq:inf\]) proceeds as follows. Because of the mean-field character of our model the following formula hold for every neuron $i$ on each layer $t$. Formally writing (forgetting about the pattern index $\mu$) $\langle O \rangle
\equiv \langle \langle O \rangle_{\sigma|\xi} \rangle_{\xi}=
\sum_{\xi} p(\xi) \sum_{\sigma} p(\sigma|\xi) O $ for an arbitrary quantity $O$ the conditional probability can be obtained in a rather straightforward way by using the complete knowledge about the system: $\langle \xi \rangle=a, \, \langle \sigma \rangle=q, \,
\langle (\sigma-a)( \xi-a) \rangle=M, \, \langle 1 \rangle=1$.
The result reads $$p(\sigma|\xi)=[\gamma_0\xi+(\gamma_1-\gamma_0)\xi]\delta(\sigma-1)+
[1-\gamma_0-(\gamma_1-\gamma_0)\xi]\delta(\sigma)$$ where $\gamma_0=q-aM$ and $\gamma_1=(1-a)M+q$, and where the $M$ and $q$ are precisely the relevant parameters (\[M(t)\]) for large $N$. Using the probability distribution of the patterns we obtain $$p(\sigma)=q\delta(\sigma-1)+(1-q)\delta(\sigma)~.$$ Hence the entropy (\[eq:en\]) and the conditional entropy (\[eq:enc\]) become $$\begin{aligned}
S(\sigma)=&-&q\ln q -(1-q)\ln(1-q) \\
S(\sigma|\xi)=&-&[\gamma_0+(\gamma_1-\gamma_0)\xi]
\ln[\gamma_0+(\gamma_1-\gamma_0)\xi]
\nonumber \\
&-&[1-\gamma_0-(\gamma_1-\gamma_0)\xi]
\ln[1-\gamma_0-(\gamma_1-\gamma_0)\xi]~.\end{aligned}$$ By averaging the conditional entropy over the pattern $\xi$ we finally get for the mutual information function (\[eq:inf\]) for the layered model $$\begin{aligned}
\label{eq:Ifin}
I(\sigma;\xi) = -q\ln q -(1-q)\ln(1-q)
+a[\gamma_1\ln\gamma_1+(1-\gamma_1)\ln(1-\gamma_1)]
\nonumber\\
+(1-a)[\gamma_0\ln\gamma_0+(1-\gamma_0)\ln(1-\gamma_0)]~. \end{aligned}$$
Adaptive thresholds
===================
It is standard knowledge (e.g., [@DKM]) that the synchronous dynamics for layered architectures can be solved exactly following the method based upon a signal-to-noise analysis of the local field (\[eq:h\]) (e.g., [@Ok; @B04; @A77; @AM88] and references therein). Without loss of generality we focus on the recall of one pattern, say $\mu=1$, meaning that only $M^1(t)$ is macroscopic, i.e., of order $1$ and the rest of the patterns causes a cross-talk noise at each step of the dynamics.
We suppose that the initial state of the network model $\{\sigma_i(1)\}$ is a collection of i.i.d.r.v. with average and variance given by $
E[\sigma_i(1)]=E[(\sigma_i(1))^2]=q_0~.
$ We furthermore assume that this state is correlated with only one stored pattern, say pattern $\mu=1$, such that $ \mbox{\rm Cov}(\xi_i^\mu (1),\sigma_i(1))=\delta_{\mu,1}~M_0^1~{a(1-a)}~.$
Then the full recall proces is described by [@DKM; @B04] $$\begin{aligned}
&& M^1(t+1) = \frac{1}{2}\left\{\int{\cal D} x
\tanh\left[\beta((1-a)M^1(t)-\theta(t)+\sqrt{\alpha D(t)}\,x)\right]
\right. \nonumber \\
&& \hspace*{1cm} + \left. \int{\cal D} x
\tanh\left[\beta(-aM^1(t)-\theta(t)+ \sqrt{\alpha D(t)}\,x)\right]
\right\}
\label{eq:a} \\
&& q(t+1)= aM^1(t+1) \nonumber \\
&& \hspace*{1cm} +\frac{1}{2}
\left\{ 1+\int{\cal D} x
\tanh\left[\beta(-aM^1(t)-\theta(t)+\sqrt{\alpha D(t)}\,x)\right]
\right\}
\label{eq:b} \\
&& D(t+1)= Q(t+1) \nonumber\\
&& \hspace*{1cm} +
\frac{\beta}{2}\left\{1-a\int{\cal D} x\tanh^2\beta\left[(1-a)M^1(t)
-\theta(t)+\sqrt{\alpha D(t)}\,x\right] \right.
\nonumber \\
&& \hspace*{1cm} - \left. (1-a)\int{\cal D} x\tanh^2\beta
\left[-aM^1(t)-\theta(t)+\sqrt{\alpha D(t)}\,x\right]\right\}^2 D(t)
\label{eq:c}\end{aligned}$$ where $\alpha= p/N$, ${\cal D} x$ is the Gaussian measure ${\cal D} x= dx
(2\pi)^{-1/2}\exp(-x^2/2)$, where $Q(t)=[(1-2a)q(t)+a^2]$ and where $D(t)$ contains the influence of the cross-talk noise caused by the patterns $\mu>1$. As mentioned before, $\theta(t)$ is an adaptive threshold that has to be chosen.
In the sequel we discuss two different choices and both will be compared for networks with synaptic noise and various activities. Of course, it is known that the quality of the recall process is influenced by the cross-talk noise. An idea is then to introduce a threshold that adapts itself autonomously in the course of the recall process and that counters, at each layer, the cross-talk noise. This is the self-control method proposed in [@DB98]. This has been studied for layered neural network models without synaptic noise, i.e., at $T=0$, where the rule (\[eq:stoc\]) reduces to the deterministic form $ \sigma_i(t+1)=\Theta({h_i(t)}) $ with $\Theta(x)$ the Heaviside function taking the value $\{0,1\}$. For sparsely coded models, meaning that the pattern activity $a$ is very small and tends to zero for $N$ large, it has been found [@BM00] that $$\label{eq:thr}
\theta(t)_{sc}= c(a)\sqrt{\alpha D(t)}, \quad c(a)=\sqrt{-2\ln a}$$ makes the second term on the r.h.s of Eq.(\[eq:b\]) at $T=0$, asymptotically vanish faster than $a$ such that $q \sim a$. It turns out that the inclusion of this self-control threshold considerably improves the quality of retrieval, in particular the storage capacity, the basins of attraction and the information content.
The second approach chooses a threshold by maximizing the information content, $i=\alpha I$ of the network (recall Eq. (\[eq:Ifin\])). This function depends on $M^1(t)$, $q(t)$, $a$, $\alpha$ and $\beta$. The evolution of $M^1(t)$ and of $q(t)$ (\[eq:a\]), (\[eq:b\]) depends on the specific choice of the threshold through the local field (\[eq:h\]). We consider a layer independent threshold $\theta(t)=\theta$ and calculate the value of (\[eq:Ifin\]) for fixed $a$, $\alpha$, $M_0^1$, $q_0$ and $\beta$. The optimal threshold, $\theta=\theta_{opt}$, is then the one for which the mutual information function is maximal. The latter is non-trivial because it is even rather difficult, especially in the limit of sparse coding, to choose a threshold interval by hand such that $i$ is non-zero. The computational cost will thus be larger compared to the one of the self-control approach. To illustrate this we plot in Figure \[fig:infoth\] the information content $i$ as a function of $\theta$ without self-control or a priori optimization, for $a=0.005$ and different values of $\alpha$.
![ The information $i=\alpha I$ as a function of $\theta$ for $a=0.005$, $T=0.1$ and several values of the load parameter $\alpha=0.1,1,2,4,6$ (bottom to top)[]{data-label="fig:infoth"}](info_theta.eps){width="6.5cm"}
For every value of $\alpha$, below its critical value, there is a range for the threshold where the information content is different from zero and hence, retrieval is possible. This retrieval range becomes very small when the storage capacity approaches its critical value $\alpha_c=6.4$.
Concerning then the self-control approach, the next problem to be posed in analogy with the case without synaptic noise is the following one. Can one determine a form for the threshold $\theta(t)$ such that the integral in the second term on the r.h.s of Eq.(\[eq:b\]) at $T \neq 0$ vanishes asymptotically faster than $a$?
In contrast with the case at zero temperature where due to the simple form of the transfer function, this threshold could be determined analytically (recall Eq. (\[eq:thr\])), a detailed study of the asymptotics of the integral in Eq. (\[eq:b\]) gives no satisfactory analytic solution. Therefore, we have designed a systematic numerical procedure through the following steps:
- Choose a small value for the activity $a'$.
- Determine through numerical integration the threshold $\theta'$ such that $$\int_{-\infty}^{\infty}
\frac{dx \,\,e^{-x^2/ 2 \sigma^2}}{\sigma \sqrt{2\pi }}
\Theta (x- \theta) \leq a' \quad \mbox{for} \quad \theta > \theta'$$ for different values of the variance $\sigma^2={\alpha D(t)}$.
- Determine as a function of $T=1/\beta$, the value for $\theta'_T$ such that $$\int_{-\infty}^{\infty}
\frac{dx \,\,e^{-y^2/ \sigma^2}}{2 \sigma \sqrt{2\pi }}
[1+ \tanh[\beta (x- \theta)]] \leq a' \quad
\mbox{for} \quad \theta > \theta' +\theta'_T.$$
The second step leads precisely to a threshold having the form of Eq. (\[eq:thr\]). The third step determining the temperature-dependent part $\theta'_T$ leads to the final proposal $$\theta_{t}(a,T)=\sqrt{-2 \ln (a)\alpha D(t)} - \frac12 \ln(a) T^2.
\label{threstemp}$$ This dynamical threshold is again a macroscopic parameter, thus no average must be taken over the microscopic random variables at each step $t$ of the recall process.
We have solved these self-controlled dynamics, Eqs.(\[eq:a\])-(\[eq:c\]) and (\[threstemp\]), for our model with synaptic noise, in the limit of sparse coding, numerically. In particular, we have studied in detail the influence of the $T$-dependent part of the threshold. Of course, we are only interested in the retrieval solutions with $M>0$ (we forget about the index $1$) and carrying a non-zero information $i=\alpha I$. The important features of the solution are illustrated, for a typical value of $a$ in Figures \[fig:basinsT\]-\[fig:infoT\]. In Figure \[fig:basinsT\] we show
![The basin of attraction as a function of $\alpha$ for $a=0.005$ and $T=0.2, 0.15, 0.1, 0.05$ (from left to right) with (full lines) and without (dashed lines) the $T$-dependent part in the threshold (\[threstemp\]).[]{data-label="fig:basinsT"}](basins.eps){width="6cm"}
the basin of attraction for the whole retrieval phase for the model with threshold (\[eq:thr\]) (dashed curves) compared to the model with the noise-dependent threshold (\[threstemp\]) (full curves). We see that there is no clear improvement for low $T$ but there is a substantial one for higher $T$. Even near the border of critical storage the results are still improved such that also the storage capacity itself is larger.
This is further illustrated in Figure \[fig:evolM\] where we compare
![The evolution of the main overlap $M(t)$ for several initial values $M_0$ with $T=0.2,~q_0=a=0.005,~\alpha=1$ for the self-control model (\[threstemp\]) without (a) and with $T$-dependent part (b) and for the optimal threshold model (c).[]{data-label="fig:evolM"}](evolution.eps){height="4.2cm" width="10cm"}
the evolution of the retrieval overlap $M(t)$ starting from several initial values, $M_0$, for the model with (Figure \[fig:evolM\] (a)) and without (Figure \[fig:evolM\] (b)) the $T$-correction in the threshold and for the optimal threshold model (Figure \[fig:evolM\] (c)). Here this temperature correction is absolutely crucial to guarantee retrieval, i.e., $M \approx 1$. It really makes the difference between retrieval and non-retrieval in the model. Furthermore, the model with the self-control threshold with noise-correction has even a wider basin of attraction than the model with optimal threshold.
In Figure \[fig:infoT\] we plot the information content $i$ as a function of the temperature for the self-control dynamics with the threshold (\[threstemp\]) (full curves), respectively (\[eq:thr\]) (dashed curves). We see that a substantial improvement of the information content is obtained.
![The information content $i=\alpha I$ as a function of $T$ for several values of the loading $\alpha$ and $a=0.005$ with (full lines) and without (dashed lines) the $T$-correction in the threshold.[]{data-label="fig:infoT"}](infoT.eps){width="6.5cm"}
Finally we show in Figure \[fig:phases\] a $T-\alpha$ plot for $a=0.005$ (a) and $a=0.02$ (b) with (full line) and without (dashed line) noise-correction in the self-control threshold and with optimal threshold (dotted line).
These lines indicate two phases of the layered model: below the lines our model allows recall, above the lines it does not. For $a=0.005$ we see that the $T$-dependent term in the self-control threshold leads to a big improvement in the region for large noise and small loading and in the region of critical loading. For $a=0.02$ the results for the self-control threshold with and without noise-correction and those for the optimal thresholds almost coincide, but we recall that the calculation with self-control is autonomously done by the network and less demanding computationally.
Conclusions
===========
In this work we have studied the inclusion of an adaptive threshold in sparsely coded layered neural networks with synaptic noise. We have presented an analytic form for a self-control threshold, allowing an autonomous functioning of the network, and compared it with an optimal threshold obtained by maximizing the mutual information which has to be calculated externally each time one of the network parameters (activity, loading, temperature) is changed. The consequences of this self-control mechanism on the quality of the recall process have been studied.
We find that the basins of attraction of the retrieval solutions as well as the storage capacity are enlarged. For some activities the self-control threshold even sets the border between retrieval and non-retrieval. This confirms the considerable improvement of the quality of recall by self-control, also for layered network models with synaptic noise.
This allows us to conjecture that self-control might be relevant for other architectures in the presence of synaptic noise, and even for dynamical systems in general, when trying to improve, e.g., basins of attraction .\
[**Acknowledgment**]{}\
This work has been supported by the Fund for Scientific Research- Flanders (Belgium).
[1]{}
Willshaw D J, Buneman O P, and Longuet-Higgins H C, Nonholographic associative memory, [*Nature*]{} [**222**]{} (1969) 960. Palm G, On the storage capacity of an associative memory with random distributed storage elements, [*Biol. Cyber.*]{} [**39**]{} (1981) 125. Gardner E, The space of interactions in neural network models, [*J. Phys. A: Math. Gen.*]{} [**21**]{} (1988) 257. Okada M, Notions of associative memory and sparse coding, [*Neural Networks*]{} [**9**]{} (1996) 1429. Dominguez D R C and Bollé D, Self-control in sparsely coded networks, [*Phys. Rev. Lett.*]{} [**80**]{} (1998) 2961. Bollé D, Dominguez D R C and Amari S, Mutual information of sparsely coded associative memory with self-control and ternary neurons, [*Neural Networks*]{} [**13**]{}(2000) 455. Bollé D and Heylen R, Self-control dynamics for sparsely coded networks with synaptic noise, in [*2004 Proceedings of the IEEE International Joint Conference on Neural Networks*]{}, p.3195 Dominguez D R C, Korutcheva E, Theumann W K and Erichsen Jr. R, Flow diagrams of the quadratic neural network, [*Lecture Notes in Computer Science*]{}, [**2415**]{}, (2002) 129. Bollé D and Massolo G, Thresholds in layered neural networks with variable activity, [*J. Phys. A: Math. Gen.*]{} [**33**]{} (2000) 2597. Hertz J, Krogh A and Palmer R G, [*Introduction to the Theory of Neural Computation*]{}, Addison-Wesley, Redwood City (1991). Domany E, Kinzel W and Meir R, Layered Neural Networks, [*J.Phys. A: Math. Gen.*]{} [**22**]{} (1989) 2081. Bollé D, Multi-state neural networks based upon spin-glasses: a biased overview, in [*Advances in Condensed Matter and Statistical Mechanics*]{} eds. Korutcheva E and Cuerno R., Nova Science Publishers, New-York,(2004 p. 321-349. Nadal J-P, Brunel N and Parga N, Nonlinear feedforward networks with stochastic outputs: infomax implies redundancy reduction, [*Network: Computation in Neural Systems*]{} [**9**]{} (1998) 207. Schultz S and Treves A, Stability of the replica-symmetric solution for the information conveyed by a neural network. [*Phys. Rev. E* ]{} [**57**]{} (1998) 3302. Blahut R E, [*Principles and Practice of Information Theory*]{}, Reading, MA: Addison-Wesley (1990). Shannon C E, A mathematical theory for communication, [*Bell Systems Technical Journal*]{} [**27**]{} (1948) 379. Schwenker F, Sommer F T and Palm G, Iterative retrieval of sparsely coded associative memory patterns, [*Neural Networks*]{} [**9**]{} (1996) 445. Amari S, Neural theory and association of concept information, [*Biol. Cyber.*]{} [**26**]{} (1977) 175. Amari S and Maginu K, Statistical neurodynamics of associative memory, [*Neural Networks*]{} [**1**]{} (1988) 63.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The effect of heart geometry and anisotropy on cardiac re-entry dynamics and self-termination is studied here in anatomically realistic computer simulations of human foetal heart. 20 weeks of gestational age human foetal heart isotropic and anisotropic anatomy models from diffusion tensor MRI data sets are used in the computer simulations. The fiber orientation angles of the heart were obtained from the DT-MRI primary eigenvalues. In a spatially homogeneous electrophysiological mono domain model with the DT-MRI based heart geometries, we initiate simplified Fitz-Hugh-Nagumo kinetics cardiac re-entry at a prescribed location in a 2D slice, and in the full 3D anatomy model. In a slice of the heart, the MRI based fiber anisotropy changes the re-entry dynamics from pinned to anatomical re-entry. In the full 3D MRI based model, the foetal heart fiber anisotropy changes the re-entry dynamics from a persistent re-entry to the re-entry self-termination. Time of re-entry self-termination depends on the re-entry initial position. Anisotropy of the heart speeds up re-entry self-termination.'
author:
- 'I.V.Biktasheva'
- 'R.A.Anderson'
- 'A.V.Holden'
- 'E.Pervolaraki'
- 'F.C.Wen'
title: 'Cardiac re-entry dynamics & self-termination in DT-MRI based model of Human Foetal Heart'
---
[^1]
> The effect of the heart anisotropy and anatomy on cardiac re-entry dynamics, although difficult to demonstrate in experiment, is well appreciated [@Bishop-etal-2010; @Bishop-etal-2011; @Bishop-Plank-2012; @Fukumoto-etal-2016], and has been studied in simplified mathematical and computer models [@Fenton-Karma-1998; @Pertsov-etal-PRL2000; @Wellner-etal-PNAS2002; @RodriguezEasonTrayanova-2006; @Dierckx-etal-PRE2013]. The [BeatBox]{} [@bbx-2017-PONE] High Performance Computing (HPC) cardiac electrophysiology computer simulation environment allows direct incorporation of the high resolution DT-MRI heart anatomy data sets into the biophysically and anatomically realistic computer simulations. In the [BeatBox]{} *in-silico* model, the anisotropy of the tissue is switched “on” and “off” to allow for comparison between the anatomically realistic isotropic and anisotropic conduction, in order to see the specific pure anatomy effects, as well as the interplay between the anisotropy and anatomy of an individual heart. In this paper, we present the DT-MRI based anatomy and myofiber structure realistic computer simulation study of cardiac re-entry dynamics in the *in-silico* model of the human foetal heart [@Pervolaraki-etal-2013]. We demonstrate that, in a 2D slice of the heart, the realistic fiber anisotropy of the tissue changes cardiac re-entry dynamics from pinned into fast anatomical re-entry. In the full 3D DT-MRI based model, depending on the initial location of the re-entry, the isotropic geometry of the heart might sustain a perpetual re-entry even with a positive filament tension; while the same positive filament tension re-entry initiated at the same location of the foetal heart with the realistic fiber anisotropy self-terminates within seconds. Generally, time of re-entry self-termination depends on the re-entry initial position, while the role of the heart anisotropy is to speed up the re-entry self-termination.
\[Intro\]Introduction
=====================
Since the hypothesis over a century ago that cardiac re-entry underlies cardiac arrhythmias [@Mines-1913; @Garey-1914] , and the much later confirmation of the hypothesis in cardiac tissue experiment [@Allessie-etal-1973; @Pertsov-etal-1993], the re-entry (*aka* spiral wave in 2D, cardiac excitation vortex in 3D), its origin and its role in sustained arrhythmias and fibrillation, as well as a possibility of its effective control and defibrillation, have been an object of extensive theoretical study and modelling [@Wiener-Rosenblueth-1946; @Balakhovsky-1965; @Krinsky-1968; @Panfilov-etal-1984; @Davydov-etal-1988; @Keener-1988; @Ermakova-etal-1989; @Biktashev-Holden-1994; @awt; @Fenton-Karma-1998; @Pertsov-etal-PRL2000; @Wellner-etal-PNAS2002; @swd; @orbit; @Biktashev-etal-2011-PONE; @Biktasheva-etal-2015-PRL]. From experiment, it is an established point of view that cardiac arrhythmias are due to a complex combination of electrophysiological [@BoschNattel-CardiovascularResearch2002; @Workman-etal-HeartRythm2008; @Kushiyama-etal-2016], structural [@Pellman-etal-2010; @Eckstein-etal-2011; @Takemoto-etal-2012; @Eckstein-etal-2013], and anatomical [@MacEdo-etal-2010; @Anselmino-etal-2011] factors which sustain cardiac re-entry [@GrayPertsovJalife-1996-Circ; @Wu-etal-1998-CR; @Nattel-Nature2002; @Yamazaki-etal-2012-CVR].
The specific effect of the heart anisotropy and anatomy on cardiac re-entry dynamics is well appreciated [@Bishop-etal-2010; @Bishop-etal-2011; @Bishop-Plank-2012; @Fukumoto-etal-2016], and has been studied in simplified mathematical and computer models [@Fenton-Karma-1998; @Pertsov-etal-PRL2000; @Wellner-etal-PNAS2002; @RodriguezEasonTrayanova-2006; @Dierckx-etal-PRE2013]. [ The anisotropic discontinuities in the heart muscle have been commonly seen as a substrate for rise of cardiac re-entry due to the abrupt change in conduction velocity and wavefront curvature. On the other hand, [extensive mapping of cardiac myocyte orientation in mammalian hearts has shown that the transmural fiber arrangement, including the range of transmural change in fiber angle in ventricular wall, was consistent within a species, and varied between species[@Hunter-etal-CompBiolOfHeart p. 173]. ]{} So that changes in anisotropy seen in healthy hearts can facilitate initiation of arryhthmias. ]{}
The recent advance in DT-MRI technology and High Performance Computing (HPC) allows the obtained DT-MRI data sets with the detailed heart anatomy and myofiber structure to be directly incorporated into the anatomically realistic computer simulations [@bbx-2017-PONE], so that the anisotropy of the tissue in the *in-silico* model can be switched on and off to allow for comparison between the anatomically realistic isotropic and anisotropic conduction in order to see specific anatomy effects as well as the interplay between the anisotropy and anatomy of an individual heart.
In this paper, we present the raw DT-MRI based anatomically and myofiber structure realistic computer simulation study of cardiac re-entry dynamics in the *in-silico* model of human foetal heart. [[[ The raw DT-MRI image data [@Pervolaraki-etal-2013] was segmented into the tissue/non-tissue pixels based on the MRI luminosity threshold, followed by the calculation of the fiber angles at each voxel from the diffusion-weighted DT-MRI images.]{} This very basic segmentation]{}]{} might be seen as a limitation of the study from the cardiac physiology point of view. However, the purpose of our study is not to provide results of immediate physiological or clinical relevance: for these we currently simply have not enough data. Rather, from the non-linear science point of view our rationale is to use the raw DT-MRI data “as is” as an example of an unaltered nature provided medium to study a re-entry dynamics. Although the DT-MRI yields three eigenvalues, the second and the third are often harder to distinguish, so we used only the primary eigenvector to define the local fibre orientations in the simulation study. The focus of the paper is to demonstrate the effect of a real mammalian heart anatomy and anisotropy on a re-entry dynamics. The available MRI data of a foetal heart provide an excellent oportunity for such study. The objectives for the use of foetal heart MRI data are: whether the anatomical settings of the although foetal but a real heart might support a positive filament tension re-entry, and what would it be the role of a real heart anisotropy in that case. So, here we demonstrate that the real heart anisotropy enhances re-entry self-termination.
We demonstrate that, in a 2D slice of the heart, the realistic fiber anisotropy might change the re-entry dynamics from pinned to anatomical re-entry.
In the full 3D DT-MRI based model, depending on the location of the re-entry initiation, the isotropic geometry of the heart might sustain perpetual re-entry even with a positive filament tension kinetics. While the same positive filament tension re-entry initiated at the same location of the foetal heart with the realistic fiber anisotropy self-terminates within seconds. Time of re-entry self-termination depends on the re-entry initial position. Anisotropy of the real heart speeds up re-entry self-termination, and in this sense has a rather anti-arrhythmogenic effect. [The geometry and anisotropy of the heart together ensure the fastest self-termination of cardiac re-entry]{}.
[The novel significance of our findings is that we demonstarte that the real life heart anisotropy might have a rather anti-arrhythmic function as it facilitates fast self-termination of cardiac re-entry. ]{}
\[Methods\]Methods
==================
\[DT-MRI\]DT-MRI based anatomy model
------------------------------------
![\[HF\_slice\_x\_63\] [[**The 143 DGA human foetal heart [@Pervolaraki-etal-2013]**]{}. [BeatBox]{} [@bbx-2017-PONE] geometry [`.bbg`]{} format visualisation[: shown here are projections of the unit vectors of the local fibre orientation onto the cross-section plane. Laminar fibres are well formed except near the surfaces of the outer walls, where chaotic structure from earlier developmental stages is still present. This is seen even better in the colour-coded Figure 4 in Pervolaraki et al [@Pervolaraki-etal-2013 p. 5]. ]{}]{}](Fig1-eps-converted-to.pdf)
The DT-MRI data sets of the $128\times128\times128$ voxels size, with voxel resolution of $\sim100\mu m$, of ethically obtained 143 days of gestational age (DGA) human foetal heart [@Pervolaraki-etal-2013], were converted into the [BeatBox]{} [@bbx-2017-PONE] regular Cartesian mesh [`.bbg`]{} geometry format, containing the DT-MRI cartesian coordinates of the heart tissue points together with the corresponding components of the diffusion tensor primary eigenvectors [@bbx-2017-PONE]. The [[`.bbg`]{}]{} file is an ASCII text file, each line in which describes a point in a regular mesh in the following format: $${\texttt{ x,y,z,status,fibre\_x,fibre\_y,fibre\_z}}$$ Here [`x, y, z`]{} are integer Cartesian coordinates of a DT-MRI voxel, [`status`]{} is a flag with a nonzero-value for a tissue point, and [`fibre_x, fibre_y, fibre_z`]{} are $x$-, $y$- and $z$-components of the fibre orientation vector at that point. To reduce the size of the [[`.bbg`]{}]{} files, only the tissue points, that is points with nonzero [`status`]{} need to be specified, because the [BeatBox]{} solver will ignore the void points with zero status in any case. Although the original DT-MRI images data sets had $128\times128\times128$ voxels size, the actual dimensions of the foetal heart minimum bounding box were $67\times91\times128$, with $181070$ tissue points.
[[[ The raw DT-MRI anatomy data [@Pervolaraki-etal-2013] were segmented into the “tissue”/“not tissue” pixels discretion based on the MRI luminosity threshold, with the cartesian fiber angles at each voxel obtained from the diffusion-weighted DT-MRI images. Only this basic segmentaion of the raw DT-MRI anatomy data [@Pervolaraki-etal-2013] was taken into account in the computer simulation of cardiac re-entry dynamics, so we shall refer to it as the raw DT-MRI based anatomy model.]{}]{}]{}
[In the 2D model, the fibres vectors were projected into the plane, in order to construct the 2D diffusivity tensor. ]{}
[[[ Fig. \[HF\_slice\_x\_63\] shows the cross section of the 143 days of gestational age (DGA) foetal heart with already formed intramural laminar structure and more irregular epicardial, endocardial, and septal fibers, see Figure 4 of Pervolaraki et al [@Pervolaraki-etal-2013 p. 5] for the color-encoded fractional anisotropy (FA) and all the three components of the fiber angles in human foetal hearts. The DT-MRI based foetal heart model]{}]{} offered a unique opportunity to see if the 20 weeks of gestation age intramural heart structure was capable to support cardiac re-entry, as it would not be possible for the re-entry to pin to the endocardial fine features which were yet to be developed later, such as *e.g.* the pinning to pectinate muscles junction with crystae terminalis reported in adult human atria [@Wu-etal-1998-CR; @Yamazaki-etal-2012-CVR; @Kharche-etal-2015-BMRI]. ]{}
\[RD\] Cardiac Tissue Model
---------------------------
To investigate the effects of anatomy on cardiac re-entry dynamics we used *monodomain* tissue model with non-flux boundary conditions $$\begin{aligned}
&
{\frac{{\partial}{{{\mathbf{u}}}}}{{\partial}{\t}}} = \f ({{\mathbf{u}}})
+ \nabla\cdot\hat{{\mathbf{D}}}\nabla {{\mathbf{u}}},
\label{bc} \\ &
\qquad \qquad {\normv \cdot\hat{{\mathbf{D}}}\nabla {{\mathbf{u}}}}\bigg|_G = 0,
\nonumber \end{aligned}$$ where ${{\mathbf{u}}}(\r,t)={(u, v)}^T$, $\r$ is the position vector, $\f(\r,t)={(f, g)}^T$ is the Fitz-Hugh-Nagumo [@Winfree-1991] kinetics column-vector $$\begin{aligned}
f(u,v) &= {\alpha}^{-1}(u-u^3/3-v), \nonumber\\
g(u,v) &= {\alpha}\, ( u + {\beta}- {\gamma}v ), \label{FHN}\end{aligned}$$ with the parameter values ${\alpha}=0.3$, ${\beta}={0.71}$, ${\gamma}=0.5$, which in an infinite excitable medium support a rigidly rotating vortex with positive filament tension [@ft]. The simplified FHN model was intentionally chosen for this study in order to fully eliminate the possible effects of a realistic cell excitation kinetics, such as *e.g.* meander [@Winfree-1991], alternans[@Karma_Chaos1994], negative filament tension[@ft], etc., and in order to enhance and highlight the pure effects of the heart anatomy and anisotropy on the cardiac re-entry outcome. $\hat{{\mathbf{D}}}={{\mathbf{Q}}}\hat P$, where ${{\mathbf{Q}}}=\mathrm{diag}(1,0)=\begin{bmatrix}1&0\\0&0\end{bmatrix}$ is the matrix of the relative diffusion coefficients for $u$ and $v$ components, and $\hat
P=[P_{\j\k}]\in{\mathbb{R}}^{3\times 3}$ is the $u$ component diffusion tensor, which has only two different eigenvalues: the bigger, simple eigenvalue $\Ppar$ corresponding to the direction along the tissue fibers, and the smaller, double eigenvalue $\Port$, corresponding to the directions across the fibres, so that $$P_{\j\k} = \Port{\delta}_{\j\k} +
\left(\Ppar-\Port\right)\fib_\j\fib_\k,
\label{sigma}$$ where $\fibvec=\left(\fib_\k\right)$ is the unit vector of the fiber direction; $\normv$ is the vector normal to the tissue boundary $G$. In the isotropic simulation, $\Ppar$ and $\Port$ values were fixed at $\Ppar=\Port=1$ (corresponding 1D conduction velocity [1.89]{}). In the anisotropic simulations, $\Ppar$ and $\Port$ values were fixed at $\Ppar=2$, $\Port=0.5$ (corresponding conduction velocities [2.68]{} and [1.34]{} respectively). All the conduction velocities have been computed for the period waves with the frequency of the free spiral wave in the model, i.e. [11.36]{}. [[With the isotropic diffusivity ($\Ppar=\Port=1$) equal to the geometric mean between the faster and the slower anisotropic diffusivities ($\Ppar=2, \Port=1/2$), the isotropic conduction velocity 1.89 was almost exactly the same as the geometric mean 1.89 of the faster and slower (2.68 and 1.34 respectively) anisotropic conduction velocities, chosen in order to minimize the maximal relative difference between the isotropic and anisotropic propagation speeds.]{}]{}
All the computer simulations presented here were done using the [BeatBox]{} [@bbx-2017-PONE] software package with the explicit time-step Euler scheme, on the Cartesian regular grid with space step discretization $\Delta x=0.1$, time step discretisation $\Delta t=0.001$; 5-point stencil for isotropic, and 9-point stencil for anisotropic Laplacian approximation in 2D simulations; 7-point stencil for isotropic, and 27-point stencil for anisotropic Laplacian approximation in 3D simulations. The re-entry was initiated by the phase distribution method [@chaos]: in the 2D simulations, at a prescribed location of the cross section of the DT-MRI based anatomical model; in the 3D simulations, at a prescribed location of the full DT-MRI based whole heart anatomical model.
\[RESULTS\]Results
==================
\[2D\] 2D MRI-based “slice” simulations
---------------------------------------
In the 2D simulations, Fig. \[slice\_fig\], a counter-clockwise re-entry was initiated by the phase distribution method [@chaos], with the initial center of rotation placed at the prescribed location $x_0=40, y_0=60$ in the 2D cross section of the DT-MRI based anatomical model shown in Fig. \[HF\_slice\_x\_63\].
In the Fig. \[slice\_fig\](a-b), it can be seen that in both isotropic and anisotropic 2D simulations, at $t=0$, there was identical location of the initial re-entry rotation center: roughly in the middle of the slice, in the vicinity of the septum cuneiform opening.
Fig. \[slice\_fig\](a) shows the isotropic dynamics of the re-entry, that is with the fiber orientation data “turned OFF”, so that only the geometry of the isotropic homogeneous slice affects the dynamics of the re-entry. While it is known that in an infinite medium the chosen FHN kinetics parameter values ${\alpha}=0.3$, ${\beta}={0.71}$, ${\gamma}=0.5$ produce rigidly rotating spiral [@Winfree-1991], the anatomically realistic boundaries of the foetal heart cause the drift of the re-entry. The re-entry does not terminate because of the resonant reflection from the inexcitable boundaries [@Biktashev-Holden-1994], but after the transient first rotation around the septum cuneiform opening, the tip of the re-entry firmly pins to the sharp lower end of the cuneiform opening, see Fig. \[slice\_fig\](a).
Fig. \[slice\_fig\](b) shows the anisotropic dynamics of the re-entry, that is with the fiber orientation data “turned ON”, so that both the anatomically realistic geometry and the anisotropy of the otherwise homogeneous slice of the heart affect the dynamics of the re-entry, causing its drift. In the anisotropic slice, the re-entry also does not terminate at the inexcitable boundaries, but after the faster than in the isotropic case, see the $a.u.$ time labels in the Fig. \[slice\_fig\](a-b), transient first rotation around the septum cuneiform opening, the anatomically realistic anisotropy of the medium turns the initial spiral wave into the fast anatomical re-entry around the septum cuneiform opening, see Fig. \[slice\_fig\](b).
\[3D\] 3D Whole heart MRI-based simulations
-------------------------------------------
In the 3D whole heart MRI-based simulations shown in the Fig. \[3D\_iso\]and Fig. \[3D\_aniso\], a counter-clockwise excitation vortex was initiated by the phase distribution method [@chaos], with the initial position of the transmural vortex filament (yellow line) at the prescribed location [*along the $x$ axis*]{} at $y_0=40, z_0=60$. [It can be seen in Fig. \[3D\_iso\] isotropic, and Fig. \[3D\_aniso\] anisotropic 3D simulations that, at $t=0$, there was identical initial location of the filament of the excitation vortex: that is transmurally, roughly in the middle through the ventricles of the heart. ]{}
Fig. \[3D\_iso\] shows the *isotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned OFF”, so that only the geometry of the otherwise isotropic homogeneous foetal heart affects the dynamics of the vortex. It is known that the chosen FHN kinetics parameter values ${\alpha}=0.3$, ${\beta}={0.71}$, ${\gamma}=0.5$ produce rigidly rotating vortex with the positive filament tension [@ft], which, depending on the topology, either collapses or straightens up between the opposite boundaries of the excitable medium. [[In the 3D anatomically realistic isotropic simulations of the foetal heart, the anatomically realistic boundaries of the heart cause drift of the excitation vortex, and, depending on the initial position of the vortex filament, vortices with the positive filament tension tend to collapse. However, there exist initial locations of the excitation vortex, which although result in the drift of the vortex, still do not lead to the expected collapse of the vortex with positive filament tension. One of such outcomes is shown in the Fig. \[3D\_iso\]. Here, following the geometry of the heart, after a very short transient, the initial vortex filament breaks into the two short pieces, each of which finds its own synchronous perpetual pathway in the “isotropic” foetal heart, resulting in the seemingly perpetual cardiac re-entry, which failed to self-terminate within the extended simulation time, see Fig. \[3D\_iso\].]{}]{}
Fig. \[3D\_aniso\] shows the *anisotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned ON”, so that both the anatomically realistic geometry and the anisotropy of the otherwise homogeneous foetal heart affect the dynamics of the initial vortex.
In the 3D whole heart MRI-based simulations shown in the Fig. \[3D\_X\_iso\]
and Fig. \[3D\_X\_aniso\],
a counter-clockwise excitation vortex was initiated by the phase distribution method [@chaos], with the initial position of the transmural vortex filament (yellow line) at the prescribed location [*along the $y$ axis*]{} at $x_0=40, z_0=60$[, that is *perpendicular* to the initial orientation of the vortex filament shown in Fig. \[3D\_iso\] and Fig. \[3D\_aniso\]. It can be seen in Fig. \[3D\_X\_iso\] isotropic, and in Fig. \[3D\_X\_aniso\] anisotropic 3D simulations, that at $t=0$, there was identical intial location of the filament of the excitation vortex: that is transmurally, roughly in the middle through the ventricles of the foetal heart, and *perpendicular* to the initial orientation of the vortex filament shown in Fig. \[3D\_iso\] and Fig. \[3D\_aniso\]. ]{}
Fig. \[3D\_X\_iso\] shows the *isotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned OFF”, so that only the geometry of the otherwise isotropic homogeneous foetal heart affects the dynamics of the vortex. Here, contrary to the expectation for the positive filament tension vortex to always contract, the organising filament first transiently extends intramurally along the tissue walls, before finally breaking up to the two ring-like pieces, each of which quickly contracts and terminates at the opposite base and apex regions of the heart.
Fig. \[3D\_X\_aniso\] shows the *anisotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned ON”, so that both the anatomically realistic geometry and the anisotropy of the otherwise homogeneous foetal heart affect the dynamics of the initial vortex leading to its really fast termination at the apex of the heart.
[[In the raw DT-MRI model simulations shown in Fig. \[3D\_iso\], Fig. \[3D\_aniso\], Fig. \[3D\_X\_iso\], and Fig. \[3D\_X\_aniso\], it can be seen that although the organising filament of the vortex could not get through into the accidental “leftover” piece of tissue adjacent to the apical region, the piece got activated and might have served as an artificial “capacitor” affecting dynamics of the re-entry. In order to check whether this might be the case, we edited the original raw DT-MRI model by removing in the MRI the foreign piece, and repeated the whole heart isotropic and anisotropic simulations from the same two orthogonal initial locations of the re-entry, similar to the shown in Fig. \[3D\_iso\], Fig. \[3D\_aniso\], Fig. \[3D\_X\_iso\], and Fig. \[3D\_X\_aniso\].]{}]{}
In the 3D whole heart “edited” MRI model simulations shown in the Fig. \[3D\_shaven\_iso\]
and Fig. \[3D\_shaven\_aniso\],
a counter-clockwise excitation vortex was initiated by the phase distribution method [@chaos], with the initial position of the transmural vortex filament (yellow line) at the prescribed location [*along the $x$ axis*]{} at $y_0=40, z_0=60$. [It can be seen in Fig. \[3D\_shaven\_iso\] isotropic, and in Fig. \[3D\_shaven\_aniso\] anisotropic 3D simulations, that, at $t=0$, there was identical initial location of the filament of the excitation vortex: that is transmurally, roughly in the middle through the ventricles of the foetal heart, similar to the initial location of the vortex filament in the raw DT-MRI simulations shown in Fig. \[3D\_iso\] and Fig. \[3D\_aniso\] . ]{}
Fig. \[3D\_shaven\_iso\] shows the *isotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned OFF”, so that only the geometry of the otherwise isotropic homogeneous foetal heart affects the dynamics of the vortex. Here, following the geometry of the heart, the organising filament of the initial vortex [ also breaks into the two short pieces, each of which also finds its own synchronous pathway similar to the beginning of the raw DT-MRI simulation shown in Fig. \[3D\_iso\]. However, this time, after a few rotations, the two re-entries find their end in their also almost synchronous termination of the filaments in the base region of the foetal heart, see Fig. \[3D\_shaven\_iso\].]{}
Fig. \[3D\_shaven\_aniso\] shows the *anisotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned ON”, so that both the anatomically realistic geometry and the anisotropy of the otherwise homogeneous foetal heart affect the dynamics of the initial vortex. Here, the anisotropy of the heart [ also causes the significant transient distortion of the organising filament of the initial vortex, followed by its fast drift towards the apex, and the ultimate termination at the AV border before a completion of a single rotation, very similar to the raw DT-MRI simulation shown in Fig. \[3D\_aniso\]. However, this time, without the “leftover” piece “incidental capcitor” effect, there is just a bit faster repolarisation of the whole heart than it was in the presence of the “incidental capcitor” in the raw DT-MRI simulation shown in Fig. \[3D\_aniso\]. [For the comparison of the re-entry termination times, and the whole heart recovery times, see FIG. \[Table1\]]{}. ]{}
In the 3D whole heart “edited” MRI model simulations shown in the Fig. \[3D\_X\_shaven\_iso\]
and Fig. \[3D\_X\_shaven\_aniso\],
a counter-clockwise excitation vortex was initiated by the phase distribution method [@chaos], with the initial position of the transmural vortex filament (yellow line) at the prescribed location [*along the $y$ axis*]{} at $x_0=40, z_0=60$. [It can be seen in Fig. \[3D\_X\_shaven\_iso\] isotropic, and in Fig. \[3D\_X\_shaven\_aniso\] anisotropic 3D simulations, that at $t=0$, there was the identical initial location of the filament of the excitation vortex: that is transmurally, roughly in the middle through the ventricles of the foetal heart, perpendicular to the initial location of the vortex filament in the “edited” MRI simulations shown in Fig. \[3D\_shaven\_iso\] and Fig. \[3D\_shaven\_aniso\], and similar to the initial location of the vortex filament in the raw DT-MRI simulations shown in Fig. \[3D\_X\_iso\] and Fig. \[3D\_X\_aniso\] . ]{}
[Fig. \[3D\_X\_shaven\_iso\] shows the *isotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned OFF”, so that only the geometry of the otherwise isotropic homogeneous foetal heart affects the dynamics of the vortex. Here, again contrary to the expectation for a positive filament tension vortex to always contract, the organising filament first transiently extends intramurally before [breaking up]{} into the two ring-like pieces, each of which quickly contracts and terminates at the opposite base and apex regions of the heart, identical to what can be seen in the raw DT-MRI simulation shown in Fig. \[3D\_X\_iso\]. So that, this time, for this particular orientation of the initial re-entry, the “leftover” tissue “incidental capacitor” effect does not seem to play any role in the outcomes of the isotropic “heart geometry only” raw DT-MRI simulations shown in Fig. \[3D\_X\_iso\] as opoosed to the outcome of the identical initial re-entry location in the “edited” MRI simulations shown in Fig. \[3D\_X\_shaven\_iso\]. ]{}
Fig. \[3D\_X\_shaven\_aniso\] shows the *anisotropic dynamics* of the excitation vortex, that is with the fiber orientation data “turned ON”, so that both the anatomically realistic geometry and the anisotropy of the otherwise homogeneous foetal heart affect the dynamics of the initial vortex[, which, in the absence of the “incidental capacitor” effect, results in the fastest possible termination of the re-entry at the apex of the heart, before the vortex first rotation ever started. The re-entry termination time here is more than twice shorter than in the raw and “edited” MRI isotropic simulations shown in Fig. \[3D\_X\_iso\] and Fig. \[3D\_X\_shaven\_iso\], shorter than in the analogous simulation with the “incidental capacitor” effect shown in the Fig. \[3D\_X\_aniso\], and times shorter than in any of the simulations of the re-entry with the perpendicular initial location shown in the Fig. \[3D\_iso\], Fig. \[3D\_aniso\], Fig. \[3D\_shaven\_iso\], and Fig. \[3D\_shaven\_aniso\] .]{}
In Fig \[Table1\], we have summarized the results of the simulations shown in Fig. \[3D\_iso\], Fig. \[3D\_aniso\], Fig. \[3D\_X\_iso\], Fig. \[3D\_X\_aniso\], Fig. \[3D\_shaven\_iso\], Fig. \[3D\_shaven\_aniso\], Fig. \[3D\_X\_shaven\_iso\], and Fig. \[3D\_X\_shaven\_aniso\], with the re-entry termination time shown in arbitrary time units under each respected whole heart model and initiation cite panel. It can be seen that the realistic anisotropy of the heart causes at least twice faster termination of re-entry. It also can be seen that indeed the present in the raw DT-MRI model leftover piece of tissue connected to the apical region of the heart has served as an artificial “capacitor” affecting the dynamics of the re-entry, and significantly prolongated life time of the re-entry initiated at particular locations/orientation respective to the “capacitor”.
Finally, the 3D anatomically realistic simulations of the foetal heart show that the realistic anisotropy of the heart causes the fast transient distortion of the vortex filament, and the typical fast drift towards the apex area of the inexcitable boundary of the heart, which ultimately results in the fast self-termination of the excitation vortex, see Figs. \[3D\_iso\]-\[3D\_X\_shaven\_aniso\] and the corresponding movies in the Supplementary Material section \[Suppl\].
\[Discussion\] DISCUSSION AND FUTURE DIRECTIONS
===============================================
Although the role of heart anatomy and anisotropy in the origin and sustainability of cardiac arrhythmias has been appreciated for a long time, the experimental evidence capable to clarify the detail of the effects of the heart anatomy on the persistent cardiac arrhythmias and fibrillation are limited. [[ In particular, the theoretically plausible hypothesis that the anisotropic discontinuities in the heart might be a source of rise for cardiac re-entry due to the abrupt change in conduction velocity and wavefront curvature was in controversy with the observation that the transmural fiber arrangement, including the range of transmural change in fiber angle in ventricular wall, although varied between species[@Hunter-etal-CompBiolOfHeart p. 173], was consistent within a species. So that the question was that, if the pro-arrhythmic mechanism of cardiac re-entry initiation by the anisotropic discontinuities in a heart[@Fenton-Karma-1998; @Spach-CircRes-2001; @Smaill-etal-2004] was correct, what would then have been a reason for the consistent structure [@Hunter-etal-CompBiolOfHeart p. 173] of the anisotropic discontinuities in healthy mammalian hearts.]{} The]{} combination of the High Performance Computing with the high-resolution DT-MRI based anatomy models of the heart allows anatomically realistic *in-silico* testing of the effects of individual heart anatomy and anisotropy on the cardiac re-entry dynamics [@Kharche-etal-2015-BMRI; @Kharche-etal-2015-LNCS; @bbx-2017-PONE]. In this paper, for the first time, we present the anatomy and myofiber structure realistic computer simulation study of the cardiac re-entry dynamics in the DT-MRI based model of the human foetal heart [@Pervolaraki-etal-2013].
The comparative isotropic *vs* anisotropic simulation of the otherwise homogeneous foetal heart shows that, in the 2D slice of the heart, the realistic fiber anisotropy might change the re-entry dynamics from pinning to the sharp end of the septum cuneiform opening, Fig. \[slice\_fig\](a), into a fast anatomical re-entry around the opening, Fig. \[slice\_fig\](b). [Because of the 2D re-entry pinning to either the sharp end of the septum opening in the isotropic simulation in Fig. \[slice\_fig\](a), or to the whole septum opening as an anatomical re-entry in the anisotropic simulation in Fig. \[slice\_fig\](b), despite of the only basic segmentation of the MRI model into the tissue/not tissue points, and the ventricles not being isolated from the atria, the tip of the re-entry never got from the ventricles into the atria, Fig. \[slice\_fig\]. ]{} [Although, from the cardiac physiology point of view, the [[ [ only basic segmentation of the raw DT-MRI data [@Pervolaraki-etal-2013] into the tissue/non-tissue pixels]{} might be seen as]{}]{} a major limitation of the study, from the non-linear science point of view, the use of the raw MRI data as an example of a nature provided medium to study a re-entry dynamics gives an important insight into the pure anatomy induced drift in an otherwise homogeneous 2D medium, and into the possibility of pinning of the re-entry not to a major blood vessel but to a sharp end of an anatomical openning [@Biktasheva-etal-2015-PRL]; and into that a real fiber anisotropy is capable to turn the pinned re-entry into an anatomical one.]{} Importantly though, the 2D simulations in Fig. \[slice\_fig\] are an important step to highlight the role and the necessity of the whole heart structure in the re-entry dynamics and self-termination.
[In the 3D DT-MRI based *isotropic* model of the foetal heart, depending on the initial location/orientation of the organising filament of the excitation vortex, the geometry of the foetal heart might sustain perpetual cardiac re-entry even with a positive filament tension, Fig. \[3D\_iso\]. However, if the same positive filament tension vortex is initiated at the exactly same location/orientation and in the same anatomical environment in the full *anisotropic* 3D DT-MRI based model of the heart, the realistic fiber structure of the foetal heart facilitates fast self-termination of cardiac re-entry, Fig. \[3D\_aniso\]. ]{}
[ From the respective comparison of the “isotropic vs anisotropic” simulations in FIG. \[3D\_iso\] vs FIG. \[3D\_aniso\], and FIG. \[3D\_shaven\_iso\] vs FIG. \[3D\_shaven\_aniso\], it can be seen that, whereas the re-entry filaments were capable to penetrate from the ventricles to atria in the isotropic simulations shown in FIG. \[3D\_iso\] and FIG. \[3D\_shaven\_iso\], the abrupt change in the fiber angles between the atria and the ventricles, which can be seen in FIG. \[HF\_slice\_x\_63\], did not allow the re-entry filaments to get from the ventricles to atria in the anisotropc simulations shown in FIG. \[3D\_aniso\] and FIG. \[3D\_shaven\_aniso\], so that the ventricles’ anisotropy could complete the speedy elimination of the re-entry within its single rotation. ]{}
[The comparison of the re-entry termination times in the raw DT-MRI data model whole heart simulations shown in Fig. \[3D\_iso\], Fig. \[3D\_aniso\], Fig. \[3D\_X\_iso\], and Fig. \[3D\_X\_aniso\], with the corresponding series of the “edited” MRI model whole heart simulations shown in Fig. \[3D\_shaven\_iso\], Fig. \[3D\_shaven\_aniso\], Fig. \[3D\_X\_shaven\_iso\], and Fig. \[3D\_X\_shaven\_aniso\], showed that, although the filament of the re-entry never got through into the small piece of excitable tissue accidentally adjacent to the apical region of the heart, the adjacent tissue served as a “capacitor” significantly prolongating the life time of the re-entry initiated at a particular location/orientation respective to the “capacitor’s” own location/orientation. See for the quantitative comparison of the re-entry termination times Fig. \[Table1\] and Fig. \[quant\], where the bigger number and the total length of the filaments tend to correlate with the faster termination of re-entry, though these fail to identify the persistent re-entry in FIG. \[3D\_iso\] simulation. ]{}
The “isotropic vs anisotropic” comparison of re-entry self-termination time in both the original raw DT-MRI simulations series, and in the “edited” MRI whole heart simulations, confirmed that the real anisotropy of the heart speeds up cardiac re-entry self-termination. [The re-entry self-termination times provided for the summarised comparison in Fig. \[Table1\] and Fig. \[quant\], show that, regardless of with or without the “leftover” piece adjacent to the apex, the anisotropy of the heart speeds up cardiac re-entry self-termination. Fig. \[quant\] shows that anisotropy increases the transient number and the transient total length of the filaments. The bigger transient number and the total length of the filaments tend to correlate with the faster termination of re-entry. The biggest transient total length of the filaments was in case of the re-entry initiated *along the $y$ axis*, see panel [**(d)**]{} in Fig. \[quant\], which ensured the re-entry fastest termination. It can be seen from FIG. \[3D\_X\_iso\], FIG. \[3D\_X\_aniso\], FIG. \[3D\_X\_shaven\_iso\], and FIG. \[3D\_X\_shaven\_aniso\], that the initial position of the filament *along the $y$ axis* allowed the filament to grow intramurally, thus maximally increasing the transient total length of the filaments, and speeding up their termination. ]{}
The simulations with the “edited” MRI image of thus completely isolated heart, Fig. \[3D\_shaven\_iso\], Fig. \[3D\_shaven\_aniso\], Fig. \[3D\_X\_shaven\_iso\], and Fig. \[3D\_X\_shaven\_aniso\], in comparison with the original DT-MRI model simulations, Fig. \[3D\_iso\], Fig. \[3D\_aniso\], Fig. \[3D\_X\_iso\], Fig. \[3D\_X\_aniso\], provide an important new biological insight into the problem of cardiac re-entry dynamics. Namely, that an excitable tissue accidentally adjacent to the heart might serve as a capacitor capable to prolongate time of cardiac re-entry self-termination, see for the respective comparison the simulations in Fig. \[3D\_iso\] against Fig. \[3D\_shaven\_iso\], Fig. \[3D\_aniso\] against Fig. \[3D\_shaven\_aniso\], Fig. \[3D\_X\_iso\] against Fig. \[3D\_X\_shaven\_iso\], and Fig. \[3D\_X\_aniso\] against Fig. \[3D\_X\_shaven\_aniso\], all also summarised in Fig. \[Table1\]. The latter suggests a possible new mechanism for persistent cardiac re-entry. So that if, apart from the major blood vessels normally adjacent to the heart in vivo and affecting re-entry dynamics, there were also an accidental “touching” of the heart by an adjacent excitable tissue, for example, due to the change of posture in the night sleep, the “incidental capacitor” effect could prolongate the time of cardiac re-entry self-termination, or indeed failure to self-terminate, which could be an explanation to the elusive and difficult to reporduce but statistically salient data for longer episodes of arrhythmias reported in the night ECGs as opposed to the on average shorter arrhythmias in the day time ECGs. Although our simulations using the original raw DT-MRI data with the small piece of the foreign leftover tissue, could have been seen a limitation of the study, the real heart in vivo does not exist in complete isolation from the main blood vessels and other neighboring tissues. So, we believe that our “incidental” leftover tissue results only once more confirm the importance and the necessity of taking into account the real anatomical settings and surrounding of the heart for the full appreciation of cardiac re-entry dynamics.
The [BeatBox]{} DT-MRI based *in-silico* model comparative study confirms the cardiac anatomy and anisotropy functional effect on cardiac re-entry sustainability as opposed to its self termination, the pinning of the re-entry to anatomical features, its transformation from pinned to anatomical re-entry, and the re-entry self-termination caused by the anisotropy of the tissue.
One of the limitations of the present study is the use of the simplified Fitz-Hugh-Nagumo [@Winfree-1991] excitation model Eq. (\[FHN\]). The simplified FHN model with the excitation kinetics parameters ${\alpha}=0.3$, ${\beta}={0.71}$, ${\gamma}=0.5$, which, in an infinite homogeneous isotropic excitable medium, supports a rigidly rotating vortex with positive filament tension [@ft], was chosen for this study in order to eliminate the effects of realistic cell excitation kinetics, such as *e.g.* meander [@Winfree-1991], alternans[@Karma_Chaos1994], negative filament tension[@ft], etc., in order to enhance and highlight the pure effects of the heart anatomy and anisotropy on the cardiac re-entry outcome. The realistic cell excitation models should be used in the future studies, in order to clarify the particular effects and interplay of the cell excitation kinetics with the heart anatomy and anisotropy.
As it can be seen from Fig. \[HF\_slice\_x\_63\] ([for the color-encoded fractional anisotropy (FA) and for the color-encoded all the three components of the fiber angles see Figure 4 in Pervolaraki et al [@Pervolaraki-etal-2013 p. 5]]{}), formation of the fiber structure at the epicardium and endocardium is not completed yet in the foetal heart, so that only the already formed intramural laminar structure of the fibers can affect the dynamics of cardiac re-entry. Although the use of the not fully formed foetal heart can be seen as a limitation of the study, on the other hand, it may be said that the chaotic epicardium and endocardium fiber orientation prevents the foetal heart re-entry from pinning to the fine anatomical features which were yet to be developed at the fully formed [@Pervolaraki-etal-2013] endocardium later on. [The possible differences in the anatomy and fiber structure between the foetal heart used here and fully formed/adult hearts in general, could have seriously affected the simulations, such as in the case of *e.g*. reported pinning of cardiac re-entry to the junction of pectinate muscles with crystae terminalis in adult human atrium [@Wu-etal-1998-CR; @Yamazaki-etal-2012-CVR; @Kharche-etal-2015-BMRI]. That is, although it is possible to initiate a cardiac re-entry in the tiny $1.4g$ (at $143$ DGA) foetal heart [@Pervolaraki-etal-2013], the already formed intramural laminar fiber anisotropy of the foetal heart facilitates the re-entry self-termination, Fig. \[Table1\]. With the hindsight of the present study, in a fully formed adult heart, because of the presence of the pinning opportunities provided by the endocardium anatomical features [@Wu-etal-1998-CR; @Yamazaki-etal-2012-CVR; @Kharche-etal-2015-BMRI], there must exist additional mechanisms to facilitate cardiac re-entry self-termination [@Clayton-etal-1993].]{}
The most serious limitation of the study is that [[[ only the basic segmentation of the raw DT-MRI data [@Pervolaraki-etal-2013] into the tissue/non-tissue pixels based on the MRI luminosity threshold, and only the primary eigenvalues of fibres orientation, were taken into account in the [BeatBox]{} [@bbx-2017-PONE] computer simulation of the cardiac re-entry dynamics]{}. Further levels of the model segmentation, in order to take into account e.g. the heart collagen skeleton, isolation of ventricles from atria, etc., will inevitably change the outcome of the re-entry, by adding the electrically impermeable barriers to cardiac re-entry. Currently, this further]{}]{} segmentation is added into DT-MRI based models via complex rule based image post-processing [@Lombaert-etal-2012; @Gahm-etal-2013], which not only limits the available segmented DT-MRI cardiac anatomy data sets, but also inevitably brings in an artificial assumption/limitation element into these models. From the non-linear science point of view[, which]{} we have persued in this initial study, the rationale was to use the raw DT-MRI data as an example of a nature provided medium to study a re-entry dynamics. In the future, the multichannel computer tomography might offer an automatic tissue segmentation, so that the [[multi-level]{}]{} segmented DT-MRI based heart anatomy models might become more available, and be used in the [BeatBox]{} [@bbx-2017-PONE] anatomically and biophysically realistic simulations of cardiac re-entry dynamics.
Finally, we believe [that]{} a simple “mechanistic” explanation, although often craved for, might be rather inadequate/premature [here, and will]{} require better theoretical understanding [of]{} the demonstrated potential effect of [the heart]{} anisotropy on cardiac re-entry dynamics, for it is not a particular feature, or a sequence of anisotropy features, but [rather the whole complex of the shape, anisotropy, and the exact heart position within the body surrounding, which affects the re-entry dynamics in a particular way, and which seems to have had evolved in order to ensure the fastest self-termination of cardiac re-entry. If our hypothesis is correct, it might explain the difficulties with reproducibility of the arrhythmia in vivo and in an isolated heart.]{} [The most important novel finding of the paper is that, contrary to what currently seems to be a commonly accepted view of the pro-arrhythmic nature of cardiac anisotropy, the point of view based on the mainly theoretical and simplified anatomy models studies, for the first time ever, and for the first time in a real whole heart DT-MRI based model, we have demonstrated that the real life heart anisotropy might have rather an anti-arrhythmic effect, as it facilitates the fastest self-termination of cardiac re-entry. ]{}
We acknowledge the support of the UK Medical Research Council grant G1100357 for the human foetal heart DT-MRI data sets. We also wish to acknowledge the support of the BeatBox software development project by EPSRC (UK) grants EP/I029664 and EP/P008690/1. We thank all the developers of the BeatBox HPC Simulation Environment for Biophysically and Anatomically Realistic Cardiac Electrophysiology. We are grateful to Professor V.N.Biktashev for much appreciated advice and discussion.
[55]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{}, @noop [****, ()]{}, @noop [****, ()]{}, @noop [****, ()]{}, @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} in @noop [**]{}, (, , ) pp. @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} in @noop [**]{}, () pp.
[^1]: As submitted to Chaos: An Interdisciplinary Journal of Nonlinear Science, Focus Issue on the topic of Complex Cardiac Dynamics.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recently, it has been argued that in the supersymmetric extension of the seesaw-extended Standard Model, heavy right-handed neutrinos and sneutrinos may give corrections as large as a few GeV to the mass of the lightest neutral CP-even Higgs boson, even if the soft supersymmetry-breaking parameters are of order the electroweak scale. The presence of such large corrections would render precise Higgs masses incalculable from measurable low-energy parameters. We show that this is not the case: decoupling is preserved in the appropriate sense and right-handed (s)neutrinos, if they exist, have negligible impact on the physical Higgs masses.'
author:
- |
Patrick Draper and Howard E. Haber\
*Santa Cruz Institute for Particle Physics, University of California, Santa Cruz CA 95064*
title: |
**Decoupling of the Right-handed Neutrino Contribution to the Higgs Mass in Supersymmetric Models\
**
---
Introduction
============
The discovery of a new boson near 126 GeV [@:2012gu] that resembles the Higgs boson of the Standard Model has stimulated considerable theoretical interpretation. In supersymmetric models, the observed mass is particularly interesting. Whereas 126 GeV is compatible with expectations for the mass ($m_h$) of the lightest neutral CP-even Higgs boson of the Minimal Supersymmetric Standard Model (MSSM), large quantum corrections are indicated in order to raise $m_h$ to a value 40% above $m_Z$ [@Heinemeyer:2011aa]. In the next-to-minimal model (NMSSM), an additional tree-level contribution may also boost the value of $m_h$, but radiative effects are still necessary unless the tri-linear coupling of the singlet and doublet Higgs fields in the superpotential is large [@Hall:2011aa]. Thus the measured Higgs mass provides an important clue to the parameters of the supersymmetric model.
The program of precision calculations of the lightest Higgs boson mass in the MSSM began with one-loop results, given in [@Haber:1990aw], followed by two-loop contributions given in [@Carena:1995bx]. Partial three-loop results are now available [@Martin:2007pg]. In cases with a large hierarchy between the weak scale and the scale of the stop squarks, resummation has been used to obtain precise results, now at the level of three-loop $\beta$-functions in some cases [@Degrassi:2012ry]. Residual theoretical uncertainty estimates vary depending on the type of calculation performed, but in the fixed-order case are perhaps of the order 1 GeV for light spectra below a TeV, and 2-3 GeV for heavier spectra [@Martin:2007pg].
The utility of the computations described above rely on decoupling – very heavy states that do not receive their masses from electroweak symmetry breaking are expected to give negligible contribution to the lightest Higgs mass. Only a limited set of model parameters, which are in principle accessible at future collider experiments, are thought to be required for an accurate calculation of $m_h$. On the other hand, if an inaccessible heavy sector could provide a significant contribution to $m_h$, then only the size of this contribution could be constrained by comparing the measured $m_h$ to the calculation in terms of observable parameters. This would clearly be a much weaker position.
Recently, it has been suggested in Ref. [@Heinemeyer:2010eg] that in the seesaw-extended MSSM [@Grossman:1997is; @mssm-seesaw], a right-handed neutrino and sneutrino provides an example of such a non-decoupling heavy sector, potentially shifting the MSSM prediction for $m_h$ by as much as a few GeV at one-loop order, even if the soft supersymmetry (SUSY) breaking parameters remain at the TeV scale. It was further argued that the large terms appear at order $p^2$ in the relevant two-point functions, which are invisible to effective potential estimates that are based on calculations performed at zero external momenta.[^1]
In light of its importance for the interpretation of the observed Higgs boson with $m_h\simeq 126$ GeV, we have performed a reanalysis of the right-handed neutrino and sneutrino contributions to $m_h$ in the seesaw-extended MSSM. We find that the corrections to $m_h$ due to physics at the seesaw scale are always minuscule, of the order of a billionth of an eV. This decoupling behavior is manifest in renormalization schemes in which the $\tan\beta$ counterterm is completely insensitive to phenomena at scales well above the SUSY-breaking scale. One class of decoupling schemes includes physical schemes, where the $\tan\beta$ counterterm is controlled, for example, by the radiative corrections to the mass of the heavy Higgs boson, or by corrections to the decays of the heavy Higgs bosons to down-type fermions. Another class of decoupling schemes subtracts non-decoupling terms by hand, mocking up the behavior of minimal subtraction schemes where heavy particles are fully integrated out at their thresholds and are absent from the low-energy theory. If a non-decoupling renormalization scheme is employed in the definition of $\tan\beta$, then the decoupling of high-scale physics phenomena in the radiatively-corrected Higgs mass is recovered once $\tan\beta$ is directly related to a low-energy observable. That is, $\tan\beta$ should be regarded as an intermediary quantity, which one is free to define in any scheme. Independently of how one defines $\tan\beta$, the MSSM Higgs mass ultimately depends solely on parameters that can be fixed by experimental measurements at energy scales of order the SUSY-breaking scale and below. Contributions to the Higgs mass from energy scales significantly above the SUSY-breaking scale must be negligible.
This paper is organized as follows. In Section 2 we review the computation of the physical masses of the neutral CP-even Higgs bosons in the MSSM at one-loop order. We provide compact formulae and discuss the role of $\tan\beta$ renormalization in the results. In Section 3 we calculate the leading contributions to the lightest Higgs boson mass from the left and right handed neutrino/sneutrino sectors. We reduce the full diagrammatic result of Ref. [@Heinemeyer:2010eg] to simple, approximate analytic formulae in two different renormalization schemes, and find that in both cases the right-handed neutrino sector exhibits appropriate decoupling. We provide an interpretation of our approximate formulae in the more natural setting of effective field theory. Finally, we study the full one-loop results numerically, finding again that contributions from the right-handed neutrino sector are negligible. Our conclusions are presented in Section 4. Explicit expressions for self-energy functions, tadpoles and the $\tan\beta$ counterterm, which can provide potential non-decoupling contributions in the computation of the Higgs mass, are exhibited in Appendix A. Using these approximate forms, one can check that the non-decoupling terms cancel exactly in the expressions for the one-loop radiatively-corrected Higgs mass when a suitable definition of the $\tan\beta$ counterterm is employed.
Physical Higgs Masses at One Loop in the MSSM
=============================================
We begin our discussion with a review of the one-loop physical Higgs masses in the MSSM with the minimally required two-Higgs doublet Higgs sector. The neutral field content is $$\begin{aligned}
H_{u,d}^0\equiv\frac{\phi_{u,d}^r+i\phi_{u,d}^i}{\sqrt{2}}+v_{u,d}\,\end{aligned}$$ where $v^2\equiv v_u^2+v_d^2=(174~{\rm GeV})^2$. Here $v_u$ \[$v_d$\] are the vacuum expectation values (vevs) of the neutral Higgs fields that couple exclusively to the up-type \[down-type\] quark and lepton fields.
The MSSM Higgs scalar potential is given by $$\begin{aligned}
V=m_1^2H_d^\dagger H_d+m_2^2H_u^\dagger H_u-b(H_d H_u+{\rm h.c.})+\tfrac{1}{8}G^2(H_u^\dagger H_u-H_d^\dagger H_d)^2+\tfrac{1}{2}g^2|H_d^\dagger H_u|^2\; ,\end{aligned}$$ where $G^2\equiv g_1^2+g_2^2$, $m_1^2\equiv m_{d}^2+|\mu|^2$, $m_2^2\equiv m_{u}^2+|\mu|^2$, $\mu$ is the supersymmetric Higgs mass parameter, and $m_u^2$, $m_d^2$, and $b$ are soft SUSY-breaking squared-mass parameters. The linear terms in the potential are given by: $$\begin{aligned}
T_u&\equiv\frac{\partial V}{\partial \phi_u^r}\bigg|_{\phi=0}=\frac{v_u}{\sqrt{2}}\left(2m_2^2+\tfrac{1}{2}G^2(v_u^2-v_d^2)-2b\frac{v_d}{v_u}\right)\;,
\label{baretad1}\end{aligned}$$ $$\begin{aligned}
T_d&\equiv\frac{\partial V}{\partial \phi_d^r}\bigg|_{\phi=0}=\frac{v_d}{\sqrt{2}}\left(2m_1^2+\tfrac{1}{2}G^2(v_d^2-v_u^2)-2b\frac{v_u}{v_d}\right)\;.
\label{baretad2}\end{aligned}$$ The quadratic terms yield $2\times2$ scalar and pseudoscalar squared-mass matrices \[in the $(\phi_d,\phi_u)$ basis\], $$\begin{aligned}
\frac{\partial^2V}{\partial\phi_a^r\partial\phi_b^r}\equiv\mathcal{M}^2_e=\left(
\begin{array}{cc}
m_1^2+\tfrac{1}{4}G^2(3v_d^2-v_u^2) & -\tfrac{1}{2}G^2 v_uv_d-b \\
-\tfrac{1}{2}G^2 v_uv_d-b & m_2^2+\tfrac{1}{4}G^2(3v_u^2-v_d^2) \\
\end{array}
\right)\;,\end{aligned}$$
$$\begin{aligned}
\frac{\partial^2V}{\partial\phi_a^i\partial\phi_b^i}\equiv\mathcal{M}^2_o=\left(
\begin{array}{cc}
m_1^2+\tfrac{1}{4}G^2(v_d^2-v_u^2) & b \\
b & m_2^2+\tfrac{1}{4}G^2(v_u^2-v_d^2) \\
\end{array}
\right).\end{aligned}$$
All parameters appearing in the above formulae should be interpreted as bare parameters.
It is convenient to require that $v_{u,d}$ are stationary points of the full one-loop effective potential, which is achieved via the tadpole cancellation conditions, $$\begin{aligned}
T_{u,d}+A_{u,d}=0\;.
\label{eq:tadzero}\end{aligned}$$ The functions $A_{u,d}$ are the one-loop tadpole diagrams at zero external momentum, and the $T_{u,d}$ are functions of the bare parameters given in Eqs. (\[baretad1\]) and (\[baretad2\]). Using Eq. (\[eq:tadzero\]), the pseudoscalar mass matrix simplifies to $$\begin{aligned}
\mathcal{M}^2_o=\left(
\begin{array}{cc}
b\displaystyle\frac{v_u}{v_d}-\displaystyle\frac{A_d}{\sqrt{2}v_d} & b \\[10pt]
b & b\displaystyle\frac{v_d}{v_u}-\displaystyle\frac{A_u}{\sqrt{2}v_u} \\
\end{array}
\right).\end{aligned}$$ Diagonalizing this matrix and expanding to leading order in $A_{u,d}$, the bare masses for the pseudoscalar $A$ and the Goldstone boson $G$ are found: $$\begin{aligned}
\label{eq:mabare}
m_{A}^2&=\frac{v^2}{v_uv_d}b-\frac{v_u^2}{v^2}\frac{A_d}{\sqrt{2}v_d}-\frac{v_d^2}{v^2}\frac{A_u}{\sqrt{2}v_u}\;,\\
m_{G}^2&=-\frac{1}{\sqrt{2}v^2}\left(A_dv_d+A_uv_u\right)\;.\end{aligned}$$ Solving Eqs. (\[eq:tadzero\]) and (\[eq:mabare\]) for $b$, $m_1^2$ and $m_2^2$ yields $$\begin{aligned}
b&=\left(\frac{v_uv_d}{v^2}\right)m_A^2+\left(\frac{v_u}{v}\right)^4\frac{A_d}{\sqrt{2}v_u}+\left(\frac{v_d}{v}\right)^4\frac{A_u}{\sqrt{2}v_d}\;,\nonumber\\
m_1^2&=\left(\frac{v_u}{v}\right)^2m_A^2+\left(\frac{v_u}{v}\right)^4\frac{A_d}{\sqrt{2}v_d}+\left(\frac{v_dv_u}{v^2}\right)^2\frac{A_u}{\sqrt{2}v_u}+ \frac{G^2}{4}(v_u^2-v_d^2)\;,\nonumber\\
m_2^2&=\left(\frac{v_d}{v}\right)^2m_A^2+\left(\frac{v_uv_d}{v^2}\right)^2\frac{A_d}{\sqrt{2}v_d}+\left(\frac{v_d}{v}\right)^4\frac{A_u}{\sqrt{2}v_u}- \frac{G^2}{4}(v_u^2-v_d^2) \;.\end{aligned}$$ Inserting these results into $\mathcal{M}^2_e$, we obtain $$\begin{aligned}
\mathcal{M}^2_e=\left(
\begin{array}{cc}
m_A^2s^2_\beta+m_Z^2c^2_\beta+\displaystyle\frac{A_d}{\sqrt{2}v_d }s^4_\beta+\displaystyle\frac{A_u}{\sqrt{2}v_u}s^2_\beta c^2_\beta & -(m_A^2+m_Z^2)s_\beta c_\beta - \displaystyle\frac{A_u}{\sqrt{2}v_u}c^3_\beta s_\beta -\displaystyle\frac{A_d}{\sqrt{2}v_d}s^3_\beta c_\beta \\[10pt]
-(m_A^2+m_Z^2)s_\beta c_\beta - \displaystyle\frac{A_u}{\sqrt{2}v_u}c^3_\beta s_\beta -\displaystyle\frac{A_d}{\sqrt{2}v_d}s^3_\beta c_\beta & m_A^2c^2_\beta+m_Z^2s^2_\beta+\displaystyle\frac{A_d}{\sqrt{2}v_d }s^2_\beta c^2_\beta+\displaystyle\frac{A_u}{\sqrt{2}v_u}s^4_\beta
\end{array}
\right)\;,\end{aligned}$$ where $m_Z^2\equiv \tfrac{1}{2}G^2v^2$, $s_\beta\equiv\sin\beta$, and $c_\beta\equiv\cos\beta$. The squared-mass matrix $\mathcal{M}^2_e$ can be diagonalized to obtain the bare masses $m_{h,H}^2$ for the light neutral CP-even Higgs boson $h$ and the heavy neutral CP-even Higgs boson $H$.
At this stage, it is convenient to replace the bare masses by physical masses: $$\begin{aligned}
m_{h,Z,A,H}^2= m_{hP,ZP,AP,HP}^2-\Sigma_{hh,ZZ,AA,HH}(m_{hP,ZP,AP,HP}^2)\;,
\label{eq:pmass}\end{aligned}$$ where the subscript $P$ indicates the corresponding physical parameter. The $\Sigma$ functions are the real parts of the corresponding self-energy functions[^2] through which parameters from other sectors of the theory affect the Higgs masses. At one-loop order, the arguments of $\Sigma_{hh,HH}$ can be consistently replaced with the corresponding tree-level expressions for the physical masses, $$\begin{aligned}
m_{ht,Ht}^2=\frac{1}{2}\left(m_Z^2+m_A^2\mp\sqrt{(m_A^2-m_Z^2)^2+4m_A^2m_Z^2\sin^22\beta}\right).
\label{mhtree}\end{aligned}$$
The replacements of Eq. (\[eq:pmass\]) largely sidestep the need to introduce renormalized mass parameters and counterterms in the calculation of $m_h$ and $m_H$. The only explicit counterterms required are associated with the parameters $v_u$ and $v_d$, which are divergent because they are fixed to the vevs of the bare fields $H_{u,d}$. Rescaling the fields by wave function renormalizations renders the vevs finite, $$v_u\rightarrow \mathcal{Z}_{H_u}^{-1/2} v_u = v_u(1+\tfrac{1}{2}\delta\mathcal{Z}_{H_u})\;,\;\;\;\;\;\;\;v_d\rightarrow \mathcal{Z}_{H_d}^{-1/2} v_d = v_d(1+\tfrac{1}{2}\delta\mathcal{Z}_{H_d})\;.
\label{eq:vevrenorm}$$ At one-loop order the renormalization of the vevs affects the Higgs masses only through the parameter $\tan\beta\equiv v_u/v_d$, which can be replaced by a renormalized parameter and a counterterm that is fixed by Eq. (\[eq:vevrenorm\]): $$\begin{aligned}
&\phantom{line} \nonumber \\[-40pt]
\tan\beta & \rightarrow \tan\beta-\delta\tan\beta\;,\label{eq:tbr} \\[-40pt]
& \phantom{line}\nonumber \end{aligned}$$ where $$\begin{aligned}
&\phantom{line} \nonumber \\[-40pt]
\delta\tan\beta & \equiv\tfrac{1}{2}(\delta\mathcal{Z}_{H_d}-\delta\mathcal{Z}_{H_u})\tan\beta.
\label{eq:dtbZ} \\[-30pt]
& \phantom{line}\nonumber \end{aligned}$$
Making the substitutions of Eqs. (\[eq:pmass\]) and (\[eq:tbr\]) and expanding to leading order in the one-loop functions, we obtain $$\begin{aligned}
\label{mhfull}
m_h^2&=m_{ht}^2-\sin(\beta-\alpha)\frac{\sqrt{2}A_h}{v}-\sin^2(\beta+\alpha)\Sigma_{ZZ}(m_Z^2)+\Sigma_{hh}(m_{ht}^2)\nonumber\\
&\qquad-\cos^2(\beta-\alpha)\Sigma_{AA}(m_A^2)+\sin^2(\beta-\alpha)\Sigma_{GG}(0)-2m_Z^2\cos^2\beta\sin(2(\beta+\alpha))\delta\tan\beta \; ,\end{aligned}$$ and $$\begin{aligned}
\label{mhfull2}
m_H^2&=m_{Ht}^2-\cos(\beta-\alpha)\frac{\sqrt{2}A_H}{v}-\cos^2(\beta+\alpha)\Sigma_{ZZ}(m_Z^2)+\Sigma_{HH}(m_{Ht}^2)\nonumber\\
&\qquad-\sin^2(\beta-\alpha)\Sigma_{AA}(m_A^2)+\cos^2(\beta-\alpha)\Sigma_{GG}(0)+2m_Z^2\cos^2\beta\sin(2(\beta+\alpha))\delta\tan\beta \;,\end{aligned}$$ where $m_{ht,Ht}$ are the tree-level masses of the neutral CP-even Higgs bosons \[cf. Eq. (\[mhtree\])\], $m_A$ and $m_Z$ are the physical masses[^3] (i.e., input parameters taken from experimental measurements), the angle $\alpha$ is the tree-level mixing angle obtained in the diagonalization of $\mathcal{M}_e^2$, and $$A_h\equiv A_u\cos\alpha-A_d\sin\alpha\,,\qquad\quad A_H\equiv A_u\sin\alpha+A_d\cos\alpha\,,$$ are the tadpoles with respect to the neutral CP-even Higgs mass basis.
In obtaining the formulae in Eqs. (\[mhfull\]) and (\[mhfull2\]) we have used the tree level relation that relates $\alpha$ to the parameters $\beta$ and $m_A$ (cf. Eq. (A.20) of Ref. [@Gunion:1986nh]), $$\begin{aligned}
\label{AZ}
m_A^2=-m_Z^2\frac{\sin\bigl(2(\beta+\alpha)\bigr)}{\sin\bigl(2(\beta-\alpha)\bigr)},\end{aligned}$$ as well as the relation between the tadpoles and the Goldstone self-energy imposed by the requirement that the one-loop Goldstone boson mass vanishes, $$\begin{aligned}
\sqrt{2}\,v\Sigma_{GG}(0)=\cos(\beta-\alpha)A_H+\sin(\beta-\alpha)A_h\;.\end{aligned}$$ As a check of our calculation, we note that in the limit of $\beta=\half\pi$ and $m_A>m_Z$ we have $m_{ht}=m_Z$, $m_{Ht}=m_A$, and $\sin(\beta-\alpha)=1$ at tree-level. In this case, Eqs. (\[mhfull\]) and (\[mhfull2\]) reduce to $$\begin{aligned}
\label{eq:largetb}
m_h^2&= m_Z^2+\Sigma_{hh}(m_{Z}^2)-\Sigma_{ZZ}(m_Z^2)-\frac{A_h}{\sqrt{2}\,v}\;,\nonumber\\
m_H^2&=m_A^2+\Sigma_{HH}(m_{A}^2)-\Sigma_{AA}(m_A^2)\;,\end{aligned}$$ which reproduces the result for $m_h$ obtained in Ref. [@Haber:1990aw].
From Eqs. (\[mhfull\]) and (\[mhfull2\]), we see that the only counterterm appearing explicitly in the Higgs masses is $\delta\tan\beta$. If only a prediction for $m_{h}$ is desired, then $\delta\tan\beta$ can be eliminated in favor of $m_{H}$, and all instances of the renormalized $\tan\beta$ parameter appearing in the self-energies may be consistently replaced at one-loop order by solving the tree level formula Eq. (\[mhtree\]) for $\tan\beta$ as a function of $m_H$. The end result, $$\begin{aligned}
m_h^2=m_A^2+m_Z^2-m_H^2+\Sigma_{hh}(m_{ht}^2)+\Sigma_{HH}(m_{Ht}^2)-\Sigma_{ZZ}(m_Z^2)-\Sigma_{AA}(m_A^2)-\Sigma_{GG}(0)\;,
\label{eq:mhsumrule}\end{aligned}$$ coincides with a sum rule derived first in Ref. [@Berger:1989hg].
In the MSSM, the prediction for $m_h$ and $m_H$ depends on $\tan\beta$ and other MSSM mass parameters (such as $m_A$ and the top squark mass and mixing parameters). In particular, since $\tan\beta$ appears in the expressions for $m_{ht}$ and $m_{Ht}$ \[cf. Eq. (\[mhtree\])\], one must define $\delta\tan\beta$ by specifying a subtraction scheme. In principle any scheme to define the parameter $\tan\beta$ is allowed. In practice, it is preferable to employ a scheme that satisfies decoupling, in which case $\tan\beta$ can be determined solely from physical measurements that can be carried out in collider experiments. In contrast, if a non-decoupling scheme is used, then the definition of $\tan\beta$ depends on unknown contributions from inaccessible heavy sectors, in which case the value of $\tan\beta$ (which is needed to predict $m_h$ and $m_H$) cannot be determined from low-energy experimental measurements.
Of course, in the context of a specific model of high scale physics, one can employ a non-decoupling scheme to define $\tan\beta$ and then compute the relation of $\tan\beta$ so defined to some specific low-energy observable. In this case, one can formally eliminate $\tan\beta$ and re-express the MSSM prediction for $m_h$ and $m_H$ in terms of the corresponding low-energy observable. This would then provide a prediction for $m_h$ and $m_H$ in terms of parameters that can be determined solely from low-energy measurements. Following such a procedure, one finds that the predicted values for $m_h$ and $m_H$ are completely insensitive to high-scale physics, as expected from the decoupling properties of quantum field theory (e.g., see Ref. [@Dobado:1997up]). By employing a definition of $\tan\beta$ that respects decoupling, the insensitivity of the predicted values for $m_h$ and $m_H$ to high scale physics is manifest.
Suppose that there are no schemes in which $\tan\beta$ can be determined from a low-scale measurement. As a simple example, consider the case of high-scale SUSY in the decoupling limit, where all the superpartner masses and $m_A$ are taken very large, of order $m_{\rm SUSY}\gg m_Z$. In this case decoupling schemes for $\tan\beta$ are not particularly favored over non-decoupling schemes. On the other hand, the observed Higgs mass is no longer a testable prediction, but rather a scheme-dependent constraint on the two unmeasurable parameters $m_{\rm SUSY}$ and $\tan\beta$. Scheme-dependence is not very important in this case as it can simply be absorbed in an unobservable shift of $\tan\beta$. Furthermore, it does not affect the upper bound on $m_h$ for fixed $m_{\rm SUSY}$, which is obtained in the large $\tan\beta$ limit where scheme-dependent terms vanish. For the rest of this work, we will focus on the case in which the MSSM Higgs mass prediction is testable at colliders.
The standard $\overline{\rm DR}$ scheme [@DR] will not automatically yield decoupling. However, it can be modified slightly (m$\overline{\rm DR}$, in the notation of Ref. [@Heinemeyer:2010eg]) to remove large logarithms by hand. This subtraction reproduces the result one would obtain at leading-log order with effective field theory, in which heavy sectors are integrated out by hand at their thresholds. Hence, at leading-log order the m$\overline{\rm DR}$ scheme respects decoupling. However, beyond leading-log, one should also remove non-decoupling non-logarithmic finite terms that are still present in the m$\overline{\rm DR}$ scheme. This can be achieved in an extension of the m$\overline{\rm DR}$ scheme in which *all* contributions from the heavy sector are subtracted.
A scheme that possesses similar properties, denoted by “DEC” (for decoupling) below, fixes the Higgs wave function counterterms as follows,[^4] $$\begin{aligned}
\left(\delta\mathcal{Z}_{H_d}\right)_{\rm DEC}&=\frac{d\Sigma_{HH}(p^2)}{dp^2}\bigg |_{\alpha=0\,,\,p^2=0}\;,\nonumber\\[10pt]
\left(\delta\mathcal{Z}_{H_u}\right)_{\rm DEC}&=\frac{d\Sigma_{hh}(p^2)}{dp^2}\bigg |_{\alpha=0\,,\,p^2=0}\;.\label{deltaZs}\end{aligned}$$ In this scheme, the $\tan\beta$ counterterm is given by Eq. (\[eq:dtbZ\]), $$\label{dtanbdec}
(\delta\tan\beta)_{\rm DEC}=\tfrac{1}{2}\tan\beta\left\{\frac{d[\Sigma_{HH}(p^2)-\Sigma_{hh}(p^2)]}{dp^2}\right\}_{\alpha=0\,,\,p^2=0}\,.$$ Indeed, the DEC scheme manifestly removes large logarithms and finite terms from heavy sectors (as we exhibit explicitly in Section 3.1). This subtraction scheme also removes additional contributions that depend on the low-energy sectors (without affecting the decoupling behavior of the scheme). In fact, this is reminiscent of the on-shell scheme (the definition of which does not involve the limit $\alpha\rightarrow 0$) which was observed in Ref. [@Heinemeyer:2010eg] to respect decoupling, but was discarded in favor of the m$\overline{\rm DR}$ scheme, as the latter was deemed to be more numerically stable. We emphasize that even with a scheme (such as the DEC scheme) that is not directly related to any particular physical measurement, decoupling is preserved if the effects of the heavy sector that do not vanish in the large mass limit are fully removed by hand. In particular this is how effective field theory analysis should be performed in mass-independent schemes [@Weinberg:1980wa].
Another possibility is to demand that some physical (measurable) quantity is given at one-loop order by its tree-level formula. Two such quantities are the mass $m_H$ and the decay rate $\Gamma(A\rightarrow\tau\tau)$. In the former case \[denoted as the “HiggsMass" (HM) scheme\], the $\tan\beta$ counterterm is obtained by setting $m_H^2=m_{Ht}^2$ in Eq. (\[mhfull2\]), which *defines* $\tan\beta$ in terms of the *low-energy* physical parameters $m_Z$, $m_H$ and $m_A$, so that all one-loop pieces cancel: $$\begin{aligned}
(\delta\tan\beta)_{\rm HM}=\frac{1}{2m_Z^2\cos^2\beta\sin\bigl(2(\beta+\alpha)\bigr)}\bigg(&\cos(\beta-\alpha)\frac{\sqrt{2}A_H}{v}+\cos^2(\beta+\alpha)\Sigma_{ZZ}(m_Z^2)-\Sigma_{HH}(m_{Ht}^2)\nonumber\\
&+\sin^2(\beta-\alpha)\Sigma_{AA}(m_A^2)-\cos^2(\beta-\alpha)\Sigma_{GG}(0)\bigg)\;.\label{deltatbHM}\end{aligned}$$
A detailed and complementary discussion of $\tan\beta$ renormalization appears in Ref. [@Freitas:2002um]. In this reference, the authors do not emphasize decoupling properties, but exhibit other flaws among all available schemes. For example, $\overline{\rm DR}$ is gauge-dependent at one-loop, the HM scheme can lead to large perturbative corrections and numerical instability, and using $\Gamma(A\rightarrow\tau\tau)$ is both technically complicated and introduces flavor dependence into $\tan\beta$. For our purposes of exhibiting decoupling in the next section, we will use the DEC and HM schemes as examples.
Regardless of the scheme used to define $\delta\tan\beta$, measuring $\Gamma(A\rightarrow\tau\tau)$ is a good way to experimentally determine the numerical value of renormalized parameter $\tan\beta$ in the given scheme. Once $\tan\beta$, $m_A$, and the soft parameters are fixed (either by hand or from experimental determinations), $m_h$ and $m_H$ become predictions of the theory.
Right-Handed Sneutrino Contributions to $\boldsymbol{m_h}$
==========================================================
Right-handed neutrinos and sneutrinos obtain supersymmetric masses and couple to the Higgs sector through the following superpotential interactions [@Grossman:1997is; @mssm-seesaw]: $$\begin{aligned}
\label{superpot}
W=\mu H_dH_u + y_\nu LH_uN-y_lLH_dR+\frac{1}{2}m_MNN\;,\end{aligned}$$ where $N$ and $R$ represent the right-handed neutrino and lepton multiplets, respectively, and $m_M$ is the Majorana mass. There are also new soft SUSY-breaking couplings and masses given by the potential $$\begin{aligned}
V_{\rm soft}=m_{\tilde{R}}^2\tilde{N}^*\tilde{N}+(y_\nu A_\nu H^0_U\tilde{\nu}_L\tilde{N}^*+m_M B_\nu\tilde{N}\tilde{N}+\rm{h.c.})\;.\end{aligned}$$ In general all masses and couplings are $3\times 3$ matrices in flavor space, but for simplicity we consider only a single flavor. The resulting neutrino mass matrix is given by $$\begin{aligned}
\label{eq:mnu}
\mathcal{M}_\nu=
\left(
\begin{array}{cc}
0 & m_D \\
m_D & m_M
\end{array}
\right)\;,\end{aligned}$$ where $m_D\equiv y_\nu v_u$. The CP-even/odd $(+/-)$ sneutrino mass matrices are given by [@Grossman:1997is]: $$\begin{aligned}
\mathcal{M}^2_{\tilde{\nu}{\pm}}=
\left(
\begin{array}{cc}
m^2_{\tilde{L}} +m_D^2+\frac{1}{2}m_Z^2\cos 2\beta & \quad m_D (A_\nu-\mu\cot\beta\pm m_M) \\
m_D (A_\nu-\mu\cot\beta\pm m_M) &\quad m^2_{\tilde{R}} +m_D^2+m_M^2\pm 2B_\nu m_M
\end{array}
\right)\;,\end{aligned}$$ where $m^2_{\tilde{L}}$ is the usual soft-breaking mass for the left-handed sneutrinos present in the MSSM.
In the analysis presented in this paper, we consider only a single flavor of right-handed neutrinos and sneutrinos as described above. Nevertheless, our conclusions should not be affected by the presence of additional generations of right-handed neutrinos and sneutrinos.
Approximate Diagrammatic Result
-------------------------------
We expect that the right-handed neutrino and sneutrino contributions to the physical Higgs masses should decouple as the Majorana mass scale becomes much larger than the soft supersymmetry breaking scales, if all other parameters are held fixed. This expectation is based on the fact that the Majorana mass term $m_M$ that appears in the superpotential \[cf. eq. (\[superpot\])\] is a supersymmetry-preserving parameter. Indeed it is well known that the corrections to the tree-level Higgs mass relations in the MSSM are due entirely to SUSY-breaking effects. In contrast, we do not expect decoupling if the SUSY-breaking parameters associated with the right-handed sneutrino sector are taken very large. In the calculations presented in this section, we shall initially assume that all SUSY-breaking masses are no larger than $\mathcal{O}(1~{\rm TeV})$. The consequences of large SUSY-breaking in the right-handed sector will be briefly considered in Section 3.4.
The relevant one-loop tadpoles and self-energy functions are given in the appendix of Ref. [@Heinemeyer:2010eg]. We have independently computed $\Sigma_{hh,ZZ}$ and $A_{h}$ in the $\cot\beta\rightarrow0$ limit and found agreement except for the minus signs in front of the $m_Z^2$ terms in the last and third-to-last lines of Eq. (81) of Ref. [@Heinemeyer:2010eg]. Inserting the formulae for the one-loop tadpoles and self-energy functions into Eqs. (\[mhfull\]) and (\[mhfull2\]), we obtain the full results for $m_{h,H}^2$. To avoid a proliferation of scales tangential to the question of decoupling, we turn off $A_\nu-\mu\cot\beta$ and the $B_\nu$ parameter, and fix a common scale $m_S$, where $m_{\tilde{L}}=m_{\tilde{R}}\equiv
m_S$.[^5] We expand to first order in $m^2/m_M^2$, where $m\in\{m_Z,m_S,m_D\}$, and to leading order in powers of $m_Z$, which is the smallest mass scale when the superpartner masses, the CP-odd Higgs mass $m_A$, and the Dirac mass are large. Note that keeping only the leading order in $m_Z$ is equivalent to taking $\alpha\simeq\beta-\pi/2$ (since the vev $v$ aligns with the light state $h$ in this limit). At leading-logarithmic order, we find that the lightest Higgs mass squared is shifted relative to its tree level value in the two renormalization schemes by an amount $$\begin{aligned}
\bigg(\Delta m_h^2\bigg)_{\rm DEC}&\simeq\frac{g^2 m_Z^2}{48\pi^2 c_W^2}\cos^2 2\beta\log\frac{m_S}{m_Z}-\frac{g^2m_D^4m_S^2}{4\pi^2c_W^2m_M^2m_Z^2} \log\frac{m_M}{m_S}\;,\nonumber\\[8pt]
\bigg(\Delta m_h^2\bigg)_{{\rm HM}}&\simeq\frac{g^2m_Z^2}{48\pi^2 c_W^2}\log\frac{m_S}{m_Z}-\frac{g^2m_D^4m_S^2}{4\pi^2c_W^2m_M^2m_Z^2\sin^2\beta} \log\frac{m_M}{m_S}\;,
\label{eq:mhresult}\end{aligned}$$ where $c_W\equiv\cos\theta_W=m_W/m_Z$.
The first terms on the right-hand side of Eq. (\[eq:mhresult\]) are contributions from left-handed sneutrino loops and are insensitive to the heavy right-handed neutrino scale. These terms also appear in the ordinary MSSM without neutrino masses. For TeV-scale superpartners, these terms shift the Higgs mass by 100–200 MeV. The second terms are leading corrections from the Majorana sector and decouple rapidly as $\log {m_M}/m_M^2$, giving shifts that are generically less than a billionth of an eV. Including corrections of $\mathcal{O}(m_Z^2/m_A^2)$ is equivalent to keeping the tree-level mixing parameter $\alpha$ as a free parameter. In this case, the expressions given in Eq. (\[eq:mhresult\]) are somewhat more complicated (with non-trivial $\alpha$-dependence), but the structure of these results are maintained. Contributions that would be sensitive to the physics of the right-handed neutrino sector would yield additional terms in Eq. (\[eq:mhresult\]) of $\mathcal{O}(m_D^2)$. However, using the explicit expressions given in Appendix A, it is straightforward to verify that such terms *exactly cancel* in both the HM and DEC schemes, independently of the value of $\alpha$.
The decoupling behavior exhibited in Eq. (\[eq:mhresult\]) depends on how the light neutrino masses are allowed to change as $m_M$ is taken large. Since the overall scale of the light neutrino masses is not known, $m_D$ can be held fixed while $m_M$ is increased, in which case both the light neutrino masses and the second terms in Eq. (\[eq:mhresult\]) strictly decrease. On the other hand, one could also hold the light neutrino mass scale fixed. In this case, because of the seesaw mechanism present in Eq. (\[eq:mnu\]), the second terms in Eq. (\[eq:mhresult\]) are proportional to $m_\nu^2$ and lose their $m_M^{-2}$ decoupling behavior. Of course, this loss of decoupling is illusory, as the $m_M^{-2}$ behavior is hidden inside $m_\nu^2$ via the seesaw relation $m_\nu\sim m_D^2/m_M$. Under the assumption that $y_\nu\lsim\mathcal{O}(1)$, it follows that $m_D$ cannot be larger than the electroweak scale, in which case $m_\nu$ is at most of order $1$ eV for a suitably chosen right-handed neutrino mass scale. Hence, the magnitude of the corrections to $m_h$ due to the right-handed neutrino sector are always minuscule.
For the calculation of $\Delta m_h^2$ in the HM scheme, we avoided the direct computation of $\delta\tan\beta$ by taking advantage of the sum rule, substituting everywhere the tree level expression for $m_H^2$. Therefore, as a check of Eq. (\[eq:mhresult\]), we can compute the relation between $\tan\beta$ in the two schemes and see if it is consistent with the difference in the two computations of $\Delta m_h^2$.
The relation between the renormalized $\tan\beta$ parameters is determined by the counterterms, $$\begin{aligned}
\tan\beta_{\rm HM}=\tan\beta_{\rm DEC}+\delta\tan\beta_{\rm HM}-\delta\tan\beta_{\rm DEC}\;,
\label{eq:tbschemes}\end{aligned}$$ where $\delta\tan\beta_{\rm DEC}$ is given by Eq. (\[dtanbdec\]) and $\delta\tan\beta_{\rm HM}$ is given by Eq. (\[deltatbHM\]). Hence, the shift in the one-loop prediction for $m_h^2$ incurred by changing schemes is given by inserting Eq. (\[eq:tbschemes\]) into the tree level formula for $m_h^2$: $$\begin{aligned}
\bigg(\Delta m_h^2\bigg)_{\rm DEC}-\bigg(\Delta m_h^2\bigg)_{{\rm HM}}\simeq -2m_Z^2\cos^2\beta\sin 4\beta\Bigl[\delta\tan\beta_{\rm HM}-\delta\tan\beta_{\rm DEC}\Bigr]\;.
\label{eq:deltamhschemes}\end{aligned}$$ We find, in the approximations used above for $\Delta m_h^2$, $$\begin{aligned}
\delta\tan\beta_{\rm HM}-\delta\tan\beta_{\rm DEC}\simeq\frac{\tan\beta}{\cos 2\beta}\left(\frac{g^2}{96\pi^2c_W^2}\log\frac{m_S}{m_Z}-\frac{g^2m_D^4m_S^2}{32\pi^2c_W^2m_M^2m_Z^4\sin^4\beta}\log\frac{m_M}{m_S}\right)\;.
\label{eq:deltatb}\end{aligned}$$ It is straightforward to check that inserting Eq. (\[eq:deltatb\]) into Eq. (\[eq:deltamhschemes\]), the scheme difference obtained in Eq. (\[eq:mhresult\]) is recovered.
In non-decoupling subtraction schemes such as $\overline{\rm DR}$, the non-decoupling contributions to the one-loop corrected Higgs mass given in Eq. (\[mhfull\]) enter via the $\tan\beta$ counterterm. Using the results of Eqs. (\[dtbhm\]) and (\[dtbdr\]) given in Appendix A, $$\begin{aligned}
\label{eq:DRDEC}
\delta\tan\beta_{\rm DEC}-\delta\tan\beta_{\rm \overline{DR}}\simeq
\frac{g^2m_D^2}{32\pi^2c_W^2m_Z^2\sin 2\beta}\left(1-\log\frac{m_M^2}{Q^2}\right)\;,\end{aligned}$$ where $Q$ is the renormalization scale. As noted in Ref. [@Heinemeyer:2010eg], the partial decoupling-by-hand of the ${{\rm m}\overline{\rm DR}}$ scheme can be achieved in the $\overline{\rm DR}$ scheme by taking $Q^2=m_M^2$. However, a finite non-logarithmic term remains that also must be subtracted by hand if $\tan\beta$ is to be a genuine low-energy parameter that can be determined from experimental measurements far below the seesaw scale. Indeed, one could simply extend the ${{\rm m}\overline{\rm
DR}}$ scheme by performing this extra subtraction. The end result is equivalent to the DEC scheme at leading order in our expansions.
To make further contact with the results of Ref. [@Heinemeyer:2010eg], we first note that Eq. (\[mhfull\]) can be rewritten as $$\begin{aligned}
\label{mhfullalt1}
m_h^2=m_{ht}^2 - \widehat{\Sigma}_{hh}(m_{ht}^2)\;,\end{aligned}$$ where $\widehat{\Sigma}_{hh}(p^2)$ is defined in Eq. (3.7a) of Ref. [@Heinemeyer:2010eg].[^6] If the two-loop contributions generated by products of self-energy functions are neglected in Eq. (3.2) of Ref. [@Heinemeyer:2010eg], then the pole in the matrix propagator corresponding to the light CP-even Higgs mass is given by $$\begin{aligned}
\label{mhfullalt2}
m_h^2=m_{ht}^2 - \widehat{\Sigma}_{hh}(m_h^2)\;,\end{aligned}$$ where $m_h^2$ appearing on the right-hand side above is the one-loop corrected Higgs mass. Note that the fact that the argument of $\widehat\Sigma_{hh}$ is $m_h^2$ rather than $m_{ht}^2$ means that partial two-loop information is being included in the expression for the one-loop corrected Higgs mass. In this case, Eq. (3.7a) of Ref. [@Heinemeyer:2010eg] implies that the loop-corrected Higgs mass given by Eq. (\[mhfullalt2\]) is equivalent to Eq. (\[mhfull\]) with the following replacement, $$\begin{aligned}
\Sigma_{hh}(m_{ht}^2)\rightarrow \bigl[\Sigma_{hh}(p^2)-\delta
\mathcal{Z}_{hh} (p^2-m_{ht}^2)\bigr]\biggl|_{p^2=m_h^2},
\label{mhfullmod}\end{aligned}$$ where \[cf. Eq. (3.10a) of Ref. [@Heinemeyer:2010eg]\], $$\begin{aligned}
\label{zhh}
\delta\mathcal{Z}_{hh}=\sin^2\alpha~\delta\mathcal{Z}_{H_d} + \cos^2\alpha~\delta\mathcal{Z}_{H_u}\;.\end{aligned}$$
We now examine in more detail how decoupling occurs in the expression for the loop-corrected Higgs mass. It is convenient to define a momentum-dependent Higgs squared-mass, $$\begin{aligned}
m_h^2(p^2) &\equiv m_{ht}^2-\widehat\Sigma_{hh}(p^2)\nonumber \\
&= m_h^2(m_{ht}^2)+\Sigma_{hh}(p^2)-\Sigma_{hh}(m_{ht}^2)-\delta\mathcal{Z}_{hh}(p^2-m_{ht}^2)\nonumber \\
&\equiv \overline{m}_h^2(p^2)-\delta\mathcal{Z}_{hh}(p^2-m_{ht}^2)\,,\label{mhfullp}\end{aligned}$$ where $\overline{m}_h^2(p^2)$ corresponds to the result of Eq. (\[mhfull\]) after replacing $\Sigma_{hh}(m_{ht}^2)$ with $\Sigma_{hh}(p^2)$. By choosing either $p^2=m_{ht}^2$ or $p^2=m_{h}^2$, we recover either Eq. (\[mhfullalt1\]) or Eq. (\[mhfullalt2\]), respectively. The potential non-decoupling behavior lies in the $\mathcal{O}(m_D^2)$ contributions to the loop-corrected Higgs mass. In Appendix A, we give the leading terms contributing at $\mathcal{O}(m_D^2)$ in the individual self-energy functions, tadpoles, and the $\tan\beta$ counterterm. None of the individual terms that appear in the expression for the loop-corrected Higgs mass vanish in the large $m_M$ limit. However, given a decoupling scheme for $\delta\tan\beta$ \[and $\delta\mathcal{Z}_{hh}$, if Eq. (\[mhfullmod\]) is used\], then the non-decoupling terms cancel exactly in the Higgs mass prediction, leaving only $m_M^2$-suppressed terms at $\mathcal{O}(m_D^4)$.
It is instructive to evaluate the $\mathcal{O}(m_D^2)$ contributions to $m_h^2(p^2)$ in the DEC scheme. Using the results of Appendix A, we readily find that $$\begin{aligned}
\overline{m}_h^2(p^2)\biggl|_{\mathcal{O}(m_D^2)} & =
-\frac{g^2m_D^2}{64\pi^2c_W^2m_Z^2\sin^2\beta}\bigg(\frac{1}{\epsilon}-\gamma+\log4\pi+1-\log\frac{m_M^2}{Q^2}\bigg)\nonumber\\[6pt]
&\qquad\quad\times\biggl[p^2-m_A^2+m_{Ht}^2-m_Z^2+\cos2\beta(m_Z^2-m_A^2)+\cos2\alpha(p^2-m_{Ht}^2)\biggr]\;,
\label{eq:nondecHM}\end{aligned}$$ where the pole at $\epsilon=0$ indicates that the ultraviolet divergences have not yet canceled \[cf. Eq. (\[Qtilde\]) of Appendix A\]. We can simplify Eq. (\[eq:nondecHM\]) by using the tree-level sum rule $m_{ht}^2=m_A^2-m_{Ht}^2+m_Z^2$ and the tree-level mixing angle relation $$\begin{aligned}
\label{mixang}
\cos2\alpha(m_{Ht}^2-m_{ht}^2)=\cos2\beta(m_Z^2-m_A^2)\;.\end{aligned}$$ The end result is $$\begin{aligned}
\overline{m}_h^2(p^2)\biggl|_{\mathcal{O}(m_D^2)}=-\frac{g^2m_D^2\cos^2\alpha}{32\pi^2c_W^2m_Z^2\sin^2\beta}(p^2-m_{ht}^2)\left(\frac{1}{\epsilon}-\gamma+\log4\pi+1-\log\frac{m_M^2}{Q^2}\right)\;.
\label{eq:nondecmDEC}\end{aligned}$$
To complete the computation of $m_h^2(p^2)$, we make use of Eqs. (\[deltaZs\]) and (\[zhh\]) and the $\mathcal{O}(m_D^2)$ expressions given in Eqs. (\[sighhp\]) and (\[sigHHp\]), $$\begin{aligned}
\label{zhhdec}
\delta\mathcal{Z}_{hh}\biggl|_{\mathcal{O}(m_D^2)}=-\frac{g^2m_D^2\cos^2\alpha}{32\pi^2c_W^2m_Z^2\sin^2\beta}\left(\frac{1}{\epsilon}-\gamma+\log4\pi+1-\log\frac{m_M^2}{Q^2}\right)\;.\end{aligned}$$ Using Eq. (\[mhfullp\]), it follows that the $\mathcal{O}(m_D^2)$ contributions to $m_h^2(p^2)$ exactly cancel in the DEC scheme. This decoupling has already been demonstrated for the one-loop corrected Higgs mass defined by Eq. (\[mhfull\]) in the DEC scheme \[cf. Eq. (\[eq:mhresult\])\].
One can repeat the above calculation in the HM scheme, where $\overline{m}_h^2(p^2)$ is most easily obtained using Eq. (\[eq:mhsumrule\]), which yields $$\overline{m}_h^2(p^2)=m_{ht}^2+\Sigma_{hh}(p^2)+\Sigma_{HH}(m_H^2)-\Sigma_{ZZ}(m_Z^2)-\Sigma_{AA}(m_A^2)-\Sigma_{GG}(0)\,.$$ Evaluating the self-energy functions using the results of Appendix A, we again recover the result of Eq. (\[eq:nondecHM\]). For $p^2=m_{ht}^2$, the $\mathcal{O}(m_D^2)$ terms vanish exactly and the decoupling behavior is established, as previously demonstrated. In the case of $p^2\neq m_{ht}^2$, we need a separate definition of the Higgs wave function counterterms. Here, the natural choice is an on-shell scheme, which fixes the residues of the corresponding pole masses to unity. In this scheme, the $\mathcal{O}(m_D^2)$ contributions to $\delta\mathcal{Z}_{hh}\bigl|_{\mathcal{O}(m_D^2)}$ are the same as those of the DEC scheme, since the $\mathcal{O}(m_D^2)$ contributions to $d\Sigma_{hh}(p^2)/dp^2$ and $d\Sigma_{HH}(p^2)/dp^2$ are independent of $p^2$. Thus, it again follows that the $\mathcal{O}(m_D^2)$ contributions to $m_h^2(p^2)$ exactly cancel in the HM scheme.
In contrast, consider the computation of $m_h^2(p^2)$ in the $\overline{\rm DR}$ scheme. Due to the modification of the $\tan\beta$ counterterm \[cf. Eq. (\[eq:DRDEC\])\], an extra term is obtained in the evaluation of $m_h^2(m_{ht}^2)$ \[cf. Eq. (\[mhfull\])\]. It follows that in the $\overline{\rm DR}$ scheme, $$\begin{aligned}
\overline{m}_h^2(p^2)\biggl|_{\mathcal{O}(m_D^2)}&=\frac{g^2m_D^2}{32\pi^2c_W^2}\biggl\{\cot\beta\sin\bigl(2(\beta+\alpha)\bigr)\left(1-\log\frac{m_M^2}{Q^2}\right) \nonumber \\
&\qquad\qquad\qquad -\frac{\cos^2\alpha}{\sin^2\beta}\left(\frac{p^2-m_{ht}^2}{m_Z^2}\right)\left(\frac{1}{\epsilon}-\gamma+\log4\pi+1-\log\frac{m_M^2}{Q^2}\right)\biggr\}\;.
\label{eq:nondecmDR}\end{aligned}$$ To obtain the corresponding $\overline{\rm DR}$ expression for $\delta\mathcal{Z}_{hh}\bigl|_{\mathcal{O}(m_D^2)}$, we retain $\epsilon^{-1}-\gamma+\log 4\pi$ in Eq. (\[zhhdec\]) and discard the remaining terms. Thus in the $\overline{\rm DR}$ scheme, Eq. (\[mhfullp\]) yields $$\begin{aligned}
{m}_h^2(p^2)\biggl|_{\mathcal{O}(m_D^2)}=\frac{g^2m_D^2}{32\pi^2c_W^2}\left[\cot\beta\sin\bigl(2(\beta+\alpha)\bigr)
-\frac{\cos^2\alpha}{\sin^2\beta}\left(\frac{p^2-m_{ht}^2}{m_Z^2}\right)\right]\left(1-\log\frac{m_M^2}{Q^2}\right)\,.
\label{eq:nondecoupDR}\end{aligned}$$ In the $\rm{m}\overline{\rm DR}$ scheme of Ref. [@Heinemeyer:2010eg], one sets $Q^2=m_M^2$ to remove the logarithm, but the constant term remains and decoupling is not satisfied. The loop-corrected Higgs mass advocated in Ref. [@Heinemeyer:2010eg] corresponds to setting $p^2=m_h^2$ in $m^2_h(p^2)$ \[cf. Eq. (\[mhfullalt2\])\]. In this case, there are two separate contributions to the non-decoupling behavior, corresponding to the two terms obtained in Eq. (\[eq:nondecoupDR\]). In the $\rm{m}\overline{\rm DR}$ scheme, the second term of Eq. (\[eq:nondecoupDR\]) is negative and provides the dominant source of the Higgs mass shift at large $\tan\beta$. Indeed, it is of the correct order of magnitude to explain the decrement in $m_h$ obtained in the numerical analysis of Ref. [@Heinemeyer:2010eg].
Thus, we have located the sources of the non-decoupling behavior found in Ref. [@Heinemeyer:2010eg]. However, we note that even in a consistent one-loop truncation where $p^2=m_{ht}^2$ is taken to evaluate the loop-corrected Higgs mass, there is still a residual non-decoupling behavior in the $\rm{m}\overline{\rm DR}$ scheme, which enters via the $\tan\beta$ counterterm (which fixes the definition of $\tan\beta$). In contrast, by employing a decoupling scheme to fix the $\tan\beta$ counterterm (and the Higgs wave function counterterms if separately needed), one is guaranteed a loop-corrected Higgs mass that is completely insensitive to the physics at the right-handed neutrino scale (assuming this scale lies significantly above the SUSY-breaking scale).
Effective Field Theory Estimates of the Higgs Mass Shift
--------------------------------------------------------
In Ref. [@Heinemeyer:2010eg] it was argued that large corrections to $m_h$ could be traced to terms proportional to the external momenta in the self-energy functions. Such terms would not appear in the usual effective potential calculation. However, we have found that in a consistent one-loop truncation, such large corrections do not appear in the full expression for the physical Higgs mass when expressed in terms of parameters that can be measured directly in the low-energy effective theory. Therefore, it should be possible to derive the parametric properties of the leading terms presented in Section 3.1 directly from corrections to the Higgs quartic coupling in the effective potential, as computed in effective field theory (EFT)—the natural framework for dealing with large mass hierarchies. For simplicity, we will work primarily in the small-$m_Z$ limit, where the vev $v$ aligns with the light state $h$ such that $\alpha\rightarrow\beta-\pi/2$.
The $m_Z^2$ term we found in $\Delta m_h^2$ is just the usual contribution at low scales from the $D$-term coupling $|H_u|^2|\tilde{L}^2$, and is insensitive to the $m_M$ threshold. What about the subleading term? Imagine that we integrate out the right-handed neutrino and sneutrino at the right-handed neutrino mass threshold. Above this scale, the running of $\lambda$ (the coefficient of the quartic self-coupling $\frac{1}{8}h^4$ in the effective Lagrangian) is supersymmetric, but the TeV-scale soft mass splits the scalar and fermion states, leading to a logarithmic correction to $\lambda$ from the right-handed sneutrino bubble diagram: $$\begin{aligned}
\Delta m_h^2 =2(\Delta \lambda) v^2 \sim \frac{m_D^4}{v^2}\log\frac{m_{\tilde{N}}^2}{m_N^2}\sim \frac{m_D^4m_S^2}{v^2m_M^2}\;.\end{aligned}$$ This term is certainly present in the corrections, but it is $m_M$-suppressed and has no log enhancement, so it is not the source of the second terms in Eq. (\[eq:mhresult\]). In addition to direct contributions to $\lambda$, we also generate an approximately supersymmetric higher-dimensional coupling, $$\begin{aligned}
\Delta W= \frac{y_\nu^2}{m_M}LH_uLH_u\;.\end{aligned}$$ This coupling affects the running of $\lambda$ when supersymmetry is broken via the diagrams in Fig. 1. The dominant contribution comes from the sneutrino diagram, $$\begin{aligned}
\frac{\partial\lambda}{\partial\log Q^2}\approx \frac{y_\nu^4 m_S^2 \sin^4\beta}{8\pi^2 m_M^2}\;.\end{aligned}$$ Running the quartic coupling down from $m_M$ to $m_S$ and recalling that $v=\sqrt{2} m_W/g$, we obtain at leading logarithmic order, $$\begin{aligned}
\Delta m_h^2 = -\frac{m_D^4m_S^2}{2\pi^2v^2m_M^2}\log\frac{m_M}{m_S}\;,
\label{eq:DECEFT}\end{aligned}$$ matching the terms in Eq. (\[eq:mhresult\]) in the DEC scheme.
To understand why we obtained the DEC scheme result instead of the HM scheme result, and how the latter can be reproduced, we have to consider the definition of $\tan\beta$ in the effective theory. Up to threshold corrections that are subleading (not log-enhanced), $\tan\beta\lsub{\rm EFT}=\tan\beta\lsub{\rm full}$ at the matching scale $Q=m_M$. Therefore, the tree-level boundary condition for the Higgs self-coupling $\lambda$ takes the usual form, $$\begin{aligned}
\lambda=\tfrac{1}{4}(g_1^2+g_2^2)\cos^2 2\beta\lsub{\rm EFT}\,,\end{aligned}$$ at the matching scale. To obtain the $m_M$-dependent corrections to a low-energy prediction for $m_h$, we should include not only the shift of Eq. (\[eq:DECEFT\]), but also contributions obtained by rewriting $\cos^2 2\beta$ in the EFT at $Q=m_M$ in terms of $\cos^2 2\beta$ in the EFT at $Q=m_S$.
Below $m_M$, the dimension-5 operator contributes to the running of $\tan\beta$ in a scheme-dependent way. It is straightforward to check that the beta-function for $\tan\beta$ does not contain terms proportional to $m_S^2/m_M^2$ in the DEC scheme or any minimal subtraction scheme, where the field-strength renormalization counterterms are set by derivatives of self-energies with respect to $p^2$. The relevant diagrams are obtained by setting two external legs to $v_u$ in Fig. 1, which makes it clear that the sneutrino loop is independent of $p^2$. Therefore, in the DEC scheme, the corrections to $m_h^2$ from the running of $\tan\beta$ are higher-order in the $m_Z$ expansion, and are not required to reproduce Eq. (\[eq:mhresult\]).
In contrast, the $\tan\beta$ counterterm in the HM scheme is controlled by the self-energies themselves instead of their $p^2$ derivatives. At leading order in the $m_Z$ expansion, Eq. (\[deltatbHM\]) with $\alpha=\beta-\pi/2$ yields: $$\begin{aligned}
(\delta\tan\beta)_{\rm HM}=-\frac{1}{2m_Z^2\cos^2\beta\sin4\beta}\bigl[\Sigma_{AA}(m_A^2)-\Sigma_{HH}(m_H^2)\bigr]\;.\end{aligned}$$ Therefore, the sneutrino contributions to $\Sigma_{HH}(m_H^2)$ and $\Sigma_{AA}(m_A^2)$ can provide $m_S^2/m_M^2$ terms in the running of $\tan\beta$. Explicitly, $$\begin{aligned}
\frac{\partial (\tan\beta)_{\rm HM}}{\partial\log Q^2}= \frac{1}{2m_Z^2\cos^2\beta\sin4\beta}\frac{y_\nu^4m_S^2v_u^2\cos^2\beta}{4\pi^2m_M^2}\;,\end{aligned}$$ which at leading-log yields, $$\begin{aligned}
\Delta m_h^2 &= 2m_Z^2\cos^2\beta\sin4\beta\frac{\partial (\tan\beta)_{\rm HM}}{\partial\log Q^2}\log\frac{m_S^2}{m_M^2}\nonumber\\[6pt]
&=-\frac{m_D^4m_S^2}{2\pi^2v^2m_M^2\tan^2\beta}\log\frac{m_M}{m_S}\;.
\label{eq:tbHMrun}\end{aligned}$$ Adding Eq. (\[eq:tbHMrun\]) to Eq. (\[eq:DECEFT\]), we recover the leading HM scheme expression given by the full theory in Eq. (\[eq:mhresult\]).
A more complete effective field theory analysis of the threshold corrections from the right-handed neutrino/sneutrino sector is beyond the scope of this paper. However, our full-theory calculation makes clear how decoupling will manifest at the thresholds. Loop diagrams involving right-handed neutrinos or sneutrinos will indeed provide non-decoupling finite contributions to the low-energy effective Higgs self-coupling $\lambda$ during matching, but these contributions will be absorbed by finite and unobservable shifts in $\tan\beta$.
Numerical Results
-----------------
![Left panel: The shift in the Higgs mass due to right-handed (s)neutrinos in the decoupling (DEC) scheme at different points on the neutrino mass plane, for values of the parameters given in the text. Right panel: The same in the HiggsMass (HM) scheme. $\Delta m_{h,RH}$ is defined in Eq. (\[eq:deltamhdef\]).[]{data-label="fig:fullnumerics"}](DECnumerics.pdf "fig:"){width="48.00000%"} ![Left panel: The shift in the Higgs mass due to right-handed (s)neutrinos in the decoupling (DEC) scheme at different points on the neutrino mass plane, for values of the parameters given in the text. Right panel: The same in the HiggsMass (HM) scheme. $\Delta m_{h,RH}$ is defined in Eq. (\[eq:deltamhdef\]).[]{data-label="fig:fullnumerics"}](HiggsMassnumerics.pdf "fig:"){width="48.00000%"}
The full one-loop analytic formulae for the Higgs mass shifts in the decoupling schemes are too complicated to reproduce here. On the other hand, the approximations used above do not rule out the possibility of large corrections proportional to $m_M^2$ or $\log m_M^2$ appearing at higher order in the $m_Z$ expansion or in non-logarithmic terms. To demonstrate that such terms are not present, we have numerically evaluated the full one-loop (s)neutrino contribution to $m_h$ as a function of $|m_\nu|$ and $m_N$, with the pure left-handed sneutrino contribution subtracted out. For definiteness, we define $$\begin{aligned}
\Delta m_{h,RH}\equiv \sqrt{m_{ht}^2+\Delta m^2_{h,RH}}-m_{ht}\;,
\label{eq:deltamhdef}\end{aligned}$$ which can be thought of as an upper bound on the contribution to $m_h$ from the right-handed (RH) sector. The results are exhibited in Fig. \[fig:fullnumerics\].
If additional sectors are included to raise $m_h$ from $m_{ht}\sim m_Z$ to 126 GeV, $\Delta m_{h,RH}$ will be further suppressed by about 40%, although this is clearly unimportant in light of the overall scale of the corrections in Fig. \[fig:fullnumerics\]. Other parameters in the figure are fixed to the values $A_\nu=B_\nu=m_S=1$ TeV, $\mu=200$ GeV, and $\tan\beta=5$. As $m_M$ is increased for fixed $m_D$, we move towards the upper-left corner of the plot, where the mass shift is minimal: this trend establishes decoupling in the fixed $m_D$-sense. If we increase $m_M$ and $m_D$ so that the light physical neutrino mass $m_\nu$ is fixed, we see that the corrections are roughly constant, also as expected. In either case the overall magnitude of the corrections is never larger than about $10^{-10}$ eV, which is consistent with our estimate from Eq. (\[eq:mhresult\]).
Large SUSY-Breaking in the Right-Handed Sector
----------------------------------------------
Consider the impact of choosing values for the SUSY-breaking parameters $m_{\tilde{R}}^2$ and $B_\nu$ that are large compared to the other SUSY-breaking parameters. If soft squared-mass parameter $m_{\tilde{R}}^2$ becomes of order $m_M^2$, then the contribution to the Higgs quartic coupling from the running between $m_{\tilde{N}}^2$ and $m_N^2$ no longer decouples with large $m_M$. The Higgs mass receives a correction of order $$\begin{aligned}
\Delta m_h^2\sim \frac{m_D^4}{v^2}\log\left(\frac{m_M^2+m_{\tilde{R}}^2}{m_M^2}\right)\;,\end{aligned}$$ in complete analogy to the contribution from the top squarks. However, $m_{\tilde{R}}^2$ also enters into the one-loop RGE for the Higgs mass parameter $m_{H_u}^2$, and therefore exacerbates the little hierarchy problem when $m_{\tilde{R}}^2\gg m_Z^2$. For this reason it is preferable to keep $m_{\tilde{R}}^2$ of the same order as other squark and slepton squared-mass parameters.
When the soft mass parameter $B_\nu$ dominates the SUSY-breaking parameters, it splits the $CP$-even and $CP$-odd right-handed sneutrinos according to $m_{\tilde{N}_{\pm}}\approx m_M\pm B_\nu$. It also alters the running of the Higgs quartic coupling at high energy scales and inhibits decoupling. Running between $m_{\tilde{N}_+}^2$ and $m_{\tilde{N}_-}^2$ yields a correction to the Higgs mass of order $$\begin{aligned}
\Delta m_h^2&\sim \frac{m_D^4}{v^2}\log\left(\frac{m_{\tilde{N}_+}m_{\tilde{N}_-}}{m_N^2}\right)
\simeq\frac{m_D^4}{v^2}\log\left(\frac{m_M^2-B_\nu^2}{m_M^2}\right)\;.\end{aligned}$$ The primary distinction from the case of large $m_{\tilde{R}}^2$ is that $B_\nu$ lowers the geometric mean of the right-handed sneutrino masses, making the logarithm negative and decreasing the Higgs mass. However, as in the case of $m_{\tilde{R}}^2$, there is a good reason to keep $B_\nu\ll m_M$. In particular, a large value of $B_\nu$ generates a large contribution to $\tilde\nu_L$–$\tilde\nu_R$ mixing, which in turn generates a one-loop correction to the light neutrino masses that swamps the tree-level seesaw contribution if $B_\nu\gtrsim 10^3 m_{\tilde{\nu}_L}$ [@Grossman:1997is].
In both the large $m_{\tilde{R}}^2$ and large $B_\nu$ scenarios, the contribution to $m_h$ from the left-handed sector diagrams of Fig. 1 are subdominant. The large right-handed neutrino-sneutrino mass splittings change the argument of the logarithm, but the contribution remains suppressed by the left-handed neutrino-sneutrino mass splitting controlled by $m_{\tilde{L}}^2$.
Conclusions
===========
A recent analysis [@Heinemeyer:2010eg] has argued that adding a right-handed neutrino and sneutrino to the MSSM could generate a sizable radiative contribution to the lightest Higgs boson mass in the case of a large right-handed neutrino mass scale, even if all soft SUSY-breaking parameters remain at the TeV scale. Such a non-decoupling effect would cast doubt on the notion that the Higgs mass can be reliably calculated in a weak-scale supersymmetric theory in terms of measurable TeV-scale parameters. In this paper we have reanalyzed the radiative corrections to the Higgs mass from the right-handed neutrino sector.
In the analysis presented in this work, we began with a review of the computation of one-loop corrections to the physical masses of the neutral Higgs bosons of the MSSM, streamlining the derivation, providing compact general formulae for the spectrum, and reviewing the decoupling properties of various $\tan\beta$ renormalization schemes. In our consideration of the relevance of decoupling, we distinguished two cases. First, we commented briefly on the possibility that $\tan\beta$ cannot be independently measured in any scheme. For example, this could occur simply because all MSSM degrees of freedom are too heavy, in which case the decoupling properties of the scheme used to define $\tan\beta$ are irrelevant. However, the corresponding MSSM Higgs mass prediction cannot be tested, and the most that can be achieved is a scheme-dependent constraint on the superpartner mass scale and $\tan\beta$. Much more relevant for phenomenology is the alternative case, where some MSSM particles with $\tan\beta$-sensitive couplings can be accessed in collider experiments. In this latter case, one can predict the masses of the MSSM Higgs bosons in terms of quantities that are directly accessible to experimental measurements. These predicted masses are completely insensitive to physics at mass scales significantly larger than the scale of SUSY-breaking (such as the high-scale seesaw sector employed in a theory of neutrino masses). Consequently, it is especially convenient to define the parameter $\tan\beta$ using a renormalization scheme that respects decoupling, since the expressions for the MSSM Higgs masses (which depend explicitly on $\tan\beta$) will then manifestly exhibit the expected decoupling behavior.
Applying the general mass formulae to the right-handed neutrino sector, we derived expressions for the leading contributions in two decoupling schemes, and found that the magnitude of the corrections to the Higgs mass are utterly negligible. The expected decoupling behavior is observed if the right-handed neutrino mass scale is taken large while other input parameters are held fixed. The structure of the leading correction terms is easily recovered from effective field theory arguments. Finally, to go beyond the approximate formulae, we performed a numerical analysis including all contributing one-loop terms. We find that the corrections remain negligible and are well-reproduced by the leading terms. Since all the relevant couplings are weak, it is sufficient to work to one-loop order. In particular, the effective field theory analysis gives us confidence that our results will not change with the inclusion of two-loop and higher-order effects. Thus, we conclude that the right-handed neutrino mass scale plays no significant role in the determination of the Higgs spectrum in weak-scale supersymmetric models.
Acknowledgments {#acknowledgments .unnumbered}
---------------
We would like to acknowledge fruitful discussions with Sven Heinemeyer and Maria Herrero concerning the work of Ref. [@Heinemeyer:2010eg]. We are also grateful to the anonymous referee whose critique of earlier versions of this manuscript resulted in significant improvements to the presentation. PD and HEH are supported in part by U.S. Department of Energy grant number DE-FG02-04ER41286.
**APPENDIX**
Approximate Renormalized Self-Energies and Tadpoles
===================================================
It is convenient to have analytic approximations for the self-energy functions and tadpoles in order to see how the terms sensitive to the seesaw scale explicitly cancel in the expressions for the Higgs masses \[Eqs. (\[mhfull\]) and (\[mhfull2\])\]. Following Ref. [@Heinemeyer:2010eg], we perform a series expansion in powers of $m_D^2$. At $\mathcal{O}(m_D^0)$, the contributions are insensitive to the seesaw scale. At $\mathcal{O}(m_D^4)$, each self-energy scales as $m_M^{-2}$, exhibiting decoupling independently, in agreement with Ref. [@Heinemeyer:2010eg]. In contrast, decoupling occurs in the $\mathcal{O}(m_D^2)$ terms due to nontrivial cancellations among the various terms in Eqs. (\[mhfull\]) and (\[mhfull2\]).
Below we give the $\mathcal{O}(m_D^2)$ contributions to the real parts of the self-energy functions and tadpoles[^7] in $d$-spacetime dimensions using dimensional regularization, expanded to leading order with respect to the mass hierarchy $$\label{hierarchy}
\{m_Z^2,p^2,m_A^2,m_H^2\}\ll m_S^2\ll m_M^2\,.$$ It is convenient to adopt the shorthand notation $$\log \widetilde{Q}^2\equiv\frac{1}{\epsilon}-\gamma+\log(4\pi Q^2)\,,
\label{Qtilde}$$ where $Q$ is the renormalization scale, $\epsilon\equiv 2-\half d$ and $\gamma$ is Euler’s constant. The $\mathcal{O}(m_D^2)$ contribution to $\Sigma_{hh}(p^2)$ at leading order in the mass hierarchy \[cf. Eq. (\[hierarchy\])\] is given by $$\begin{aligned}
\label{sighhp}
\Sigma_{hh}(p^2)=& \frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\Biggl\{
\frac{2\cos^2\alpha}{\sin^2\beta}
\bigg[2m_S^2\log \frac{m_M^2}{\widetilde{Q}^2}+(m_Z^2-p^2)\bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)
\bigg]\nonumber\\[6pt]
&+m_Z^2 \bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)
\biggl[\cos^2\alpha (4-3
\cot^2\beta )+2 \sin 2 \alpha \cot \beta -\sin^2\alpha\biggr]\Biggr\}\,,\end{aligned}$$ where $p$ is the incoming four momentum. Likewise, $\Sigma_{HH}(p^2)$ is obtained by making the replacement $\alpha\to\alpha-\half\pi$ in Eq. (\[sighhp\]), $$\begin{aligned}
\label{sigHHp}
\Sigma_{HH}(p^2)=& \frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\,\Biggl\{
\frac{2\sin^2\alpha}{\sin^2\beta}
\bigg[2m_S^2\log \frac{m_M^2}{\widetilde{Q}^2}+(m_Z^2-p^2)\bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)
\bigg]\nonumber\\[6pt]
&+m_Z^2 \bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)
\biggl[\sin^2\alpha (4-3
\cot^2\beta )-2 \sin 2 \alpha \cot \beta -\cos^2\alpha\biggr]\Biggr\}\,.\end{aligned}$$
For completeness, we provide the $\mathcal{O}(m_D^2)$ contribution to the real parts of all the other relevant self-energy functions \[at leading order in the mass hierarchy, Eq. (\[hierarchy\])\], $$\begin{aligned}
\Sigma_{ZZ}(m_Z^2)= &\frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\,2m_Z^2\bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg),\nonumber\\[8pt]
\Sigma_{AA}(m_A^2)=&\frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\Biggl\{\frac{\cos 2 \beta}{\sin^2\beta}
\bigg[2 m_S^2 \log \frac{m_M^2}{\widetilde{Q}^2}-(m_A^2 +m_Z^2)\bigg(1-\log\frac{m_M^2}{\widetilde{Q}^2}\bigg)\bigg]\nonumber\\[6pt]
&\qquad\qquad\qquad +\frac{1}{\sin^2\beta}\left[2 m_S^2\log \frac{m_M^2}{\widetilde{Q}^2}-m_A^2 \bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)\right]+2 \cos2 \beta\, m_Z^2 \bigg(1-\log\frac{m_M^2}{\widetilde{Q}^2}\bigg)\Biggr\}\nonumber,\\[8pt]
\Sigma_{GG}(0)=&~\frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\left\{ 4 m_S^2 \log \frac{m_M^2}{\widetilde{Q}^2}-2 \cos 2 \beta\, m_Z^2 \bigg(1-\log\frac{m_M^2}{\widetilde{Q}^2}\bigg)\right\},\nonumber\\[8pt]
\frac{A_h}{\sqrt{2}\,v}=&\frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\left\{ \frac{\cos \alpha}{\sin\beta} \Bigg[4 m_S^2 \log \frac{m_M^2}{\widetilde{Q}^2}-m_Z^2
\bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)\Bigg]\right.\nonumber \\[6pt]
& \qquad\qquad\qquad \left. +m_Z^2 (\sin \alpha \cos \beta +3 \cos \alpha
\sin \beta ) \bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)\right\},\nonumber\\[8pt]
\frac{A_H}{\sqrt{2}\,v}=&~\frac{g^2 m_D^2}{64 \pi^2 c_W^2 m_Z^2}\left\{\frac{\sin \alpha}{\sin\beta} \Bigg[4 m_S^2 \log \frac{m_M^2}{\widetilde{Q}^2}-m_Z^2
\bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)\Bigg]\right.\nonumber \\[6pt]
& \qquad\qquad\qquad \left. +m_Z^2 (3 \sin \alpha \sin \beta -\cos \alpha
\cos \beta ) \bigg(1-\log \frac{m_M^2}{\widetilde{Q}^2}\bigg)\right\}.\label{expansions}\end{aligned}$$
Next, we compute the $\mathcal{O}(m_D^2)$ contributions \[at leading order in the mass hierarchy, Eq. (\[hierarchy\])\] to the counterterm $\delta\tan\beta$ in the various renormalization schemes. In the HM scheme, $\delta\tan\beta$ is given by Eq. (\[deltatbHM\]). Using the above expressions for the self-energy functions, along with Eq. (\[AZ\]) and the following tree-level relations (cf. Eq. (A.20) of Ref. [@Gunion:1986nh]), $$m_{ht}^2=-\frac{m_Z^2\cos 2\beta\sin(\beta+\alpha)}{\sin(\beta-\alpha)}\,,\qquad\quad
m_{Ht}^2=\frac{m_Z^2\cos 2\beta\cos(\beta+\alpha)}{\cos(\beta-\alpha)}\,,$$ we obtain after considerable simplification, $$\delta\tan\beta_{{\rm HM}}=\delta\tan\beta_{\overline{\rm DEC}}=\frac{g^2m_D^2}{32\pi^2c_W^2m_Z^2\sin2\beta}\bigg({\frac{1}{\epsilon}-\gamma+\log{4\pi}-\log{\frac{m_M^2}{Q^2}}+1}\bigg)\;.\label{dtbhm}$$ Note that the $\mathcal{O}(m_D^2)$ contributions to the counterterm $\delta\tan\beta$ in the HM and DEC schemes are equivalent, in light of the absence of non-decoupling terms in Eq. (\[eq:deltatb\]). Indeed, the $\mathcal{O}(m_D^2)$ contribution to $\delta\tan\beta$ is independent of the tree-level Higgs mixing angle $\alpha$. Although this result is obvious in the DEC scheme (which is defined via Higgs wave function counterterms that are evaluated at $\alpha=0$), the cancellation of the $\alpha$-dependence in the $\mathcal{O}(m_D^2)$ contribution to $\delta\tan\beta_{\rm{HM}}$ \[defined in Eq.(\[deltatbHM\])\] is highly non-trivial.
In contrast, in the $\overline{\rm DR}$ scheme only the $\epsilon^{-1}-\gamma+\log 4\pi$ is retained, so that the corresponding $\mathcal{O}(m_D^2)$ contribution is simply $$\begin{aligned}
\delta\tan\beta_{\overline{\rm DR}}&=\frac{g^2m_D^2}{32\pi^2c_W^2m_Z^2\sin2\beta}\bigg(\frac{1}{\epsilon}-\gamma+\log{4\pi}\bigg)\,.\label{dtbdr}\end{aligned}$$
[999]{}
S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], Phys. Lett. B [**716**]{}, 30 (2012) \[arXiv:1207.7235 \[hep-ex\]\], G. Aad [*et al.*]{} \[ATLAS Collaboration\], Phys. Lett. B [**716**]{}, 1 (2012) \[arXiv:1207.7214 \[hep-ex\]\]. S. Heinemeyer, O. Stal and G. Weiglein, Phys. Lett. B [**710**]{}, 201 (2012) \[arXiv:1112.3026 \[hep-ph\]\], P. Draper, P. Meade, M. Reece and D. Shih, Phys. Rev. D [**85**]{}, 095007 (2012) \[arXiv:1112.3068 \[hep-ph\]\], A. Arbey, M. Battaglia, A. Djouadi, F. Mahmoudi and J. Quevillon, Phys. Lett. B [**708**]{}, 162 (2012) \[arXiv:1112.3028 \[hep-ph\]\]. L. J. Hall, D. Pinner and J. T. Ruderman, JHEP [**1204**]{}, 131 (2012) \[arXiv:1112.2703 \[hep-ph\]\], U. Ellwanger, JHEP [**1203**]{}, 044 (2012) \[arXiv:1112.3548 \[hep-ph\]\]. H. E. Haber and R. Hempfling, Phys. Rev. Lett. [**66**]{}, 1815 (1991); Y. Okada, M. Yamaguchi and T. Yanagida, Prog. Theor. Phys. [**85**]{}, 1 (1991); J. R. Ellis, G. Ridolfi and F. Zwirner, Phys. Lett. B [**262**]{}, 477 (1991). M. S. Carena, J. R. Espinosa, M. Quiros and C. E. M. Wagner, Phys. Lett. B [**355**]{}, 209 (1995) \[arXiv:hep-ph/9504316\]; M. S. Carena, M. Quiros and C. E. M. Wagner, Nucl. Phys. B [**461**]{}, 407 (1996) \[arXiv:hep-ph/9508343\]; H. E. Haber, R. Hempfling and A. H. Hoang, Z. Phys. C [**75**]{}, 539 (1997) \[arXiv:hep-ph/9609331\], S. Heinemeyer, W. Hollik, and G. Weiglein, Phys. Rev. D [**58**]{}, 091701 (1998) \[arXiv:hep-ph/9803277\]; S. Heinemeyer, W. Hollik, and G. Weiglein, Phys. Lett. B [**440**]{}, 296 (1998) \[arXiv:hep-ph/9807423\]; S. Heinemeyer, W. Hollik, and G. Weiglein, Eur. Phys. J. C [**9**]{}, 343 (1999) \[arXiv:hep-ph/9812472\], J. R. Espinosa and R. J. Zhang, J. High Energy Phys. [**0003**]{}, 026 (2000) \[arXiv:hep-ph/9912236\]; J. R. Espinosa and R. J. Zhang, Nucl. Phys. B [**586**]{}, 3 (2000) \[arXiv:hep-ph/0003246\] M. Carena, H. E. Haber, S. Heinemeyer, W. Hollik, C. E. M. Wagner, and G. Weiglein, Nucl. Phys. B [**580**]{}, 29 (2000) \[arXiv:hep-ph/0001002\], G. Degrassi, P. Slavich, and F. Zwirner, Nucl. Phys. B [**611**]{}, 403(2001) \[arXiv:hep-ph/0105096\]; A. Brignole, G. Degrassi, P. Slavich and F. Zwirner, Nucl. Phys. B [**631**]{}, 195 (2002) \[arXiv:hep-ph/0112177\], S. P. Martin, Phys. Rev. D [**67**]{}, 095012 (2003) \[arXiv:hep-ph/0211366\]. S. P. Martin, Phys. Rev. D [**75**]{}, 055005 (2007) \[hep-ph/0701051\] R. V. Harlander, P. Kant, L. Mihaila and M. Steinhauser, Phys. Rev. Lett. [**100**]{}, 191602 (2008) \[Phys. Rev. Lett. [**101**]{}, 039901 (2008)\] \[arXiv:0803.0672 \[hep-ph\]\], P. Kant, R. V. Harlander, L. Mihaila and M. Steinhauser, JHEP [**1008**]{}, 104 (2010) \[arXiv:1005.5709 \[hep-ph\]\]. G. Degrassi, S. Di Vita, J. Elias-Miro, J. R. Espinosa, G. F. Giudice, G. Isidori and A. Strumia, JHEP [**1208**]{}, 098 (2012) \[arXiv:1205.6497 \[hep-ph\]\]. S. Heinemeyer, M. J. Herrero, S. Penaranda and A. M. Rodriguez-Sanchez, JHEP [**1105**]{}, 063 (2011) \[arXiv:1007.5512 \[hep-ph\]\]. Y. Grossman and H. E. Haber, Phys. Rev. Lett. [**78**]{}, 3438 (1997) \[hep-ph/9702421\]. J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi and T. Yanagida, Phys. Lett. B [**357**]{}, 579 (1995) \[hep-ph/9501407\]; Phys. Rev. D [**53**]{}, 2442 (1996) \[hep-ph/9510309\]; J. R. Ellis, J. Hisano, M. Raidal and Y. Shimizu, Phys. Rev. D [**66**]{}, 115013 (2002) \[hep-ph/0206110\]; A. Masiero, S. K. Vempati and O. Vives, New J. Phys. [**6**]{}, 202 (2004) \[hep-ph/0407325\]; E. Arganda, A. M. Curiel, M. J. Herrero and D. Temes, Phys. Rev. D [**71**]{}, 035011 (2005) \[hep-ph/0407302\]; J. R. Ellis and O. Lebedev, Phys. Lett. B [**653**]{}, 411 (2007) \[arXiv:0707.3419 \[hep-ph\]\]; A. Dedes, H. E. Haber and J. Rosiek, JHEP [**0711**]{}, 059 (2007) \[arXiv:0707.3718 \[hep-ph\]\]; M. Asano, T. Kubo, S. Matsumoto and M. Senami, Phys. Rev. D [**80**]{}, 095017 (2009) \[arXiv:0807.4922 \[hep-ph\]\]; M. Hirsch, F. R. Joaquim and A. Vicente, JHEP [**1211**]{}, 105 (2012) \[arXiv:1207.6635 \[hep-ph\]\]. W. Wang, J. M. Yang and L. L. You, arXiv:1303.6465 \[hep-ph\]. J. F. Gunion and H. E. Haber, Nucl. Phys. B [**278**]{}, 449 (1986). A. Dobado, M. J. Herrero and S. Penaranda, Eur. Phys. J. C [**7**]{}, 313 (1999) \[hep-ph/9710313\]; Eur. Phys. J. C [**12**]{}, 673 (2000) \[hep-ph/9903211\]. M. S. Berger, Phys. Rev. D [**41**]{}, 225 (1990). W. Siegel, Phys. Lett. B [**84**]{}, 193 (1979); D. M. Capper, D. R. T. Jones and P. van Nieuwenhuizen, Nucl. Phys. B [**167**]{}, 479 (1980). For a recent review, see T. Jones, PoS LL [**2012**]{}, 011 (2012).
S. Weinberg, Phys. Lett. B [**91**]{}, 51 (1980). A. Freitas and D. Stockinger, Phys. Rev. D [**66**]{}, 095014 (2002) \[hep-ph/0205281\].
[^1]: These contributions to $m_h$ are therefore quite distinct from corrections that have been found in certain parameter ranges of the NMSSM with TeV-scale right-handed neutrinos [@Wang:2013jya].
[^2]: In our notation, the sum of all one-loop Feynman graphs contributing to the $\phi\phi$ ($\phi=h,A,H$) and $ZZ$ self-energy functions are denoted by $-iC_{\phi\phi}(p^2)$ and $iA_{ZZ}(p^2)g_{\mu\nu}+iB_{ZZ}(p^2)p_{\mu\nu}$, respectively, where $p$ is the four-momentum of the incoming boson. Only $\Sigma_{ZZ}(p^2)\equiv{\rm
Re~} A_{ZZ}(p^2)$ and $\Sigma_{\phi\phi}(p^2)\equiv{\rm Re~}
C_{\phi\phi}(p^2)$ are needed to define the physical on-shell boson masses. Note that the opposite sign choice in the definition of $\Sigma(p^2)$ is sometimes employed in the literature.
[^3]: To simplify the typography, we remove all $P$ subscripts. However, all masses in the subsequent formulae should now be interpreted as (finite) physical masses.
[^4]: The choice of evaluating the $p^2$–derivatives of the self-energies at $p^2=0$ is one of many possible choices. Employing a different value of $p^2$ would simply yield a $\tan\beta$ definition that differs at the one-loop level. In the approximations used in this paper, the difference in the two definitions of $\tan\beta$ is subdominant and can thus be neglected.
[^5]: The case where $m_{\tilde R}$ and/or $B_\nu$ are parametrically larger than the electroweak scale will be briefly considered in Section 3.4.
[^6]: Note that the self-energy and tadpole functions in the conventions of Ref. [@Heinemeyer:2010eg] differ by an overall sign from those defined in this paper. This is the origin of the minus sign in Eq. (\[mhfullalt1\]).
[^7]: Note that the self-energy and tadpole functions in the conventions of Ref. [@Heinemeyer:2010eg] differ by an overall sign from those defined in this paper.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The dissipation properties of a fine sand system are investigated by a low-frequency mechanical spectroscopy. The experiments show many interesting profiles of the relative energy dissipation, which imply that some structural transition of force chains in dense granular media has occurred. The following data and discussion indicate that the transition of force chains will lead to the small deformation of arrangement in the granular system, which is responsible for the historical effects. We hope this research can improve our knowledge of the microstructure of the granular materials.'
author:
- 'Wan-Jing Wang, Kai-Wei Yang, Xue-Bang Wu, Yu-Bing Wang, and Zhen-Gang Zhu'
title: Structural transition of force chains observed by mechanical spectroscopy
---
\[sec:sec1\]introduction
========================
Granular materials [@Gennes] are ubiquitous in everyday¡¯s life, but a satisfying comprehension of their complex dynamics has not yet been achieved. Force chains [@Liu1; @Behringer] play a key important role in understanding the static structure and dynamics properties of the densely packed granular materials, such as jamming [@Negel], sound transmission [@Liu2], force propagation [@Goldenberg] and memory effects [@Geng]. Many experiments and theoretical simulations have been carried out and focused mainly on the characters of force chains [@Coppersmith; @Peters]. The most common method used to examine force chains and theirs key characteristics is to visualize these contact forces by stress-induced birefringence within assemblies of photoelastic grains [@Liu1; @Behringer]. Experiments and simulations [@Negel; @Radjai] all show that the transmission of force chains appears as a complex force network that is highly ramified and distributes inhomogeneous throughout the whole system. In fact, the force chains are quite sensitive to small perturbations in the packing geometry of the grains [@Liu1; @Liu2; @Goldenberg; @Brujic]. For example, if a granular materials are driven by a small shear, the force chains will change dramatically and evolve in a complex random way [@Mueth; @Anna; @Goldenberg]. So recently an alternative method has been developed to probe the force chains and its change [@Anna1; @Wang]. This method is based on the relationship between the energy dissipation and the structural changes of force chains in the granular system, which is related to the stress relaxation process [@Brujic].
In Ref. [@Anna1; @Wang], the dissipation properties of a sheared granular medium have been studied and a simple rheological model is presented, which suggests that small slides in the inhomogeneous granular materials are responsible for the energy dissipation. However, the microscopic picture about the slides is unclear. In this work, our investigation will concentrate on the details of the small slides in the microscopic picture, which we call the structural transition of force chains. The mechanical spectroscopy exhibits a fact that a small changes of the force network structure in granular system can lead to a corresponding energy dissipation. In addition, our experimental data show that in the granular system the arrangement and the volume fraction of the grains will change with the structural transition of force chains. The following discussion indicates that it is the small change of the arrangement that leads to the well-known historical effects, which mean that the state and properties of granular materials are quite dependent on their past history.
\[sec:sec2\]experiment
======================
In the experiment, the mechanical spectroscopy measurements were conducted on a developed low-frequency inverted torsion pendulum using the forced-vibration method. Fig. \[fig:Fig1\] shows the sketch of the pendulum. The inverted torsion pendulum consists of a cylinder being able to rotate around its axis, but prevented from moving sideways by two suspension wires fixed to two ends of the cylinder. The cylinder is forced into torsional vibration by a time-dependent force $F(t)=F_0 \sin(2\pi \nu t)$, exerted by applying a pair of permanent magnets fixed to the pendulum and external coils (circulating an ac current), where $\nu$ is the forced frequency. The angular displacement function of the cylinder, $A(t)$, is measured optically. In the case here, the response of the argument $A(t)=A_0 \sin(2\pi \nu t+
\delta\alpha)$, where $\delta\alpha$ is the phasic difference between $A(t)$ and $F(t)$. According to this measurement technique, the relative energy dissipation (RED) is calculated by measuring the loss angle between applied stress and resulting strain. Meanwhile, the relative modulus (RM) is calculated from the ratio between the stress and strain.
Before each experiment, the granular system is flattened and vibrated by external vibrations to ensure the accuracy of the measurements. And the whole system is placed on an antivibrational table to prevent undesired vibration-induced effects. In our experiments the granular materials are composed of fine sand sieved diameter $d=0.054\sim0.11mm$ and the maximum angular displacement is below $0.4^{o}$ ($0.4\pi R/180<0.1d$, which indicates that the real displacement of the cylinder is less than $0.1d$, relative to the averaged particle size. So the system can be considered as quasistatic.). The forced frequency $\nu=0.6 Hz$ or $\nu=1.0 Hz$ is chosen, which is well below the inherent frequency of the pendulum (about $36 Hz$).
\[sec:sec3\]results
===================
In our experiments the RED and the RM are given by $\eta=\tan(\delta\alpha)$ and $G_0=F_0/A_0$, respectively. The mechanical spectroscopy (RED and the RM) as a function of the amplitude $A_0$, are shown as follow.
![\[fig:Fig2\] The RED ($\triangle$) and RM ($\square$) virus the amplitude $A_{0}$(degree) in a fine sand system with the oscillating cylinder covered by a layer of grains, for the frequency $\nu=0.6 Hz$ and the immersed depth $h=60 d$.](Fig2){width="8.0cm"}
![\[fig:Fig3\] The historical effects of granular materials. (a) The RED virus the amplitude with different path. (b) RM virus the amplitude with different path. The immersed depth $h=60 d$ and the forced frequency $\nu=0.6 Hz$.](Fig3){width="8.0cm"}
Figure \[fig:Fig2\] shows clearly the relative energy dissipation increases with the amplitude. Meanwhile, we observe that the relative modulus decreases monotonously with increasing amplitude. These data are obtained when we swept the amplitude down repeatedly. Whether there is hysteresis in the system, i.e. are the peaks still there if the forced amplitude is swept up and down? This is an important question and relates to the historical effects have been found in weakly vibrated granular systems [@Umbanhowar]. As shown in Fig. \[fig:Fig3\](a), when we increase the amplitude step by step, we obtain a slowly decreasing RED. In succession, when the amplitude is swept down slowly, two different profiles (RED and RM) are obtained, as the profile (2) in Fig. \[fig:Fig3\] shows. We find that the differences between them are obvious: a rather small RED is obtained and there is a loss peak, while RM increases monotonously with decreasing amplitude. This is the historical effects, which indicate that the state and properties of granular materials are quite dependent on their past history.
\[sec:sec4\]structural transition of force chains
=================================================
In Ref. [@Anna1; @Wang], a simple rheological model is presented to reproduce the RED measured in granular materials. In the model, the granular medium is characterized by slide unit and spring unit. The model suggests that small slides in the inhomogeneous granular materials are responsible for the energy dissipation. However, the microscopic picture about the slides is unclear. Below we will try to give some details of the small slides in the microscopic picture and then give a explanation of the historical effects.
![\[fig:Fig4\] The structural transition of force chains. (a) Sketch of the forced torsion pendulum immersed into the granular medium. The cylinder is covered by a layer sand glued on with epoxy. (b) Schematic (horizontal direction) and (c) the microscopic picture (vertical direction) of the distribution of force chains in a steady state. (e) Schematic (horizontal direction) and (f) the microscopic picture (vertical direction) of the distribution of force chains opposing the rotation of the pendulum. The dark grains are glued on the cylinder by an epoxy. The force chains are represented as bonds connecting the light grey grains. Picture (d) shows the force chains in a horizontal layer near the cylinder. In this picture, the grains besides those on the force chains have not been shown.](Fig4){width="9.0cm"}
Granular materials are relatively discrete medium and force transmission through a granular system can only occur via the interparticle contacts. Under the gravity force, the force distribution in a static packing of grains in a cylinder should be inhomogeneous and could form many force chains [@Geng; @Radjai]. Near the cylinder, the spatial force distribution will be organized along directions almost normal to the cylindrical probe [@Anna1], where maximum strain be built up between chains of grains, due to the gravity. These grain chains of forces have two characters: first, the contact force between them is rather stronger and carries almost all the weight load of the above grains, and second, a chain is a quasilinear arrangement of three or more grains [@Peters]. The chain lengths are defined as the grain number in the chain, e.g., the shortest chain is two-grain chain. Meanwhile the magnitude of the contact force denotes the correlation strength of the chain. Here we present a sketch of the grain chains in Fig. \[fig:Fig4\](b) and (c). Fig.\[fig:Fig4\](c) shows a part of the microscopic picture (vertical direction) of the distribution of force chains in static state. As this picture shows, these force chains are oriented almost in the vertical direction because of the weight of the grains. In other words, the forces originate from the weight of the grains and the weight of an above particle is transmitted to a neighbor underlying particle, as shown in q-model [@Liu1; @Coppersmith]. So we often find the force chains arrange as roots of trees. In addition, we know that the qualities and strengths of chains will increase with the depth (“Janssen effects").
In the granular systems, the grains can not move independently. More often some degree of freedom of a grain are partially frozen, so that the motion of the grains is a correlated motion. When a shear stress is applied to the granular material, rather than deforming uniformly, the system such as dry sand develop shear bands [@Drake] — narrow zones of strongly correlated particles, with essentially rigid adjacent regions. Similarly, when a cylindrical probe is rotating in the granular medium, the spatial force distribution around the probe will be organized along directions almost tangent to the cylinder, where the maximum stress also build up many chains of grains. These correlations of the grain chains have been shown by the radial profiles of azimuthal velocity in Ref. [@Mueth; @Cates]. Here we present a sketch of the grain chains under shear stress in Fig. \[fig:Fig4\] (e) and (f). Compared the conditions before the shear, many differences will appear as follow. First, under shear stress more neighbor grains of the cylindrical probe will join in the force chains, although, the configuration of grains could not change, as Fig. \[fig:Fig4\] (c) and (f) show. Second, the directions of force chains will change a lot, i.e., the force chains oriented almost in the horizontal direction in contrast to almost in the vertical direction. The reason is that the shear stress changes the origin of the force chains. Then the shear stress leads to the structural changes of the force chains. Except for changing the arrangement of the force chains, the applied shear will also influence the characters of the force chains, e.g., the chain lengths will get longer and the correlation strength of the chains will become stronger. In order to clearly show the schematic of force chains, we give a microscopic picture of the force chains in a horizontal layer near the cylinder as shown in Fig. \[fig:Fig4\] (d), where the other grains besides the grains on the force chains have not been shown.
Here we call this change of the force chains without configurational change as the structural transition of force chains. Now let us discuss the relationship between the structural transition of force chains and the energy dissipation. The definition of force chains indicates that the drag force resisting a solid object moving slowly through a granular medium originates not only in the grains immediately in front of the object but also in the successive layers of grains supporting them. When a cylindrical probe begin to rotate in the granular medium, the spatial force distribution around the probe will change dramatically, as shown above. During this structural transition of force chains, some change of configuration in microscopic length scale, such as the formation and break of adhesive junctions between the surface asperities, and other forms of localized dissipative processes, must have occurred [@Anna1]. Figure \[fig:Fig2\] shows clearly the relative energy dissipation in these processes. When we increase the vibration amplitude, the structural transition, such as the break of adhesive junctions between the surface asperities, will increase, i.e., the transition quantity will increase. When the transition quantity increases the energy dissipation must will increase. As expected, the experimental results also show that the energy dissipation increases with the amplitude. Meanwhile, we observe that the relative modulus decreases monotonously with increasing amplitude, which indicates the soften of the dense granular system under shear strain [@Wang]. These analysis indicate that the structural transition of force chains leads to the energy dissipation and the dissipation will increase with the transition quantity.
In our experiments, the granular materials are composed of fine sand sieved diameter $d=0.054\sim0.11mm$. In this scale, granular matter is a well-known example of athermal system, that is a system where classical thermodynamics does not apply since thermal energy ($k_{B}T$) is insignificant compared to the gravitational energy of a macroscopic grain. A static packing of grains is therefore in a metastable state, indefinitely trapped in a local minimum of the total potential energy. However, the forced vibration will break the jamming of granular packing [@Anna]. From Ref. [@Philippe] we know that after a succession of high amplitude vibration a rather looser packing of grains (the volume fraction is small) is obtained. However, we do not know what about the mechanical spectroscopy in such a looser packing of grains under low-amplitude shear. And we do not know how the volume fraction of grains influences the energy dissipation of granular system. In the following we will discuss the energy dissipation in the condition of historical effects. As we know, the data in Fig. \[fig:Fig2\] are obtained in the experiments performed in the direction that the forced amplitude is swept down repeatedly. In the process we decrease the amplitude step by step and we obtain a slowly decreasing RED. The beginning of the profile (1) in Fig. \[fig:Fig3\] shows the dissipation at low amplitude obtained immediately after a succession of high amplitude vibration. In other words, at this moment the granular packing is rather looser. But at the beginning a rather larger RED and a rather lower RM are obtained. Then Fig. \[fig:Fig3\](a) shows that the RED decreases monotonously with increasing amplitude. In succession, when the amplitude is swept down slowly, a denser packing is obtained [@Philippe] and two different profiles (RED and RM) are obtained, as the profile (2) in Fig. \[fig:Fig3\] shows that a rather small RED is obtained, while RM increases monotonously with decreasing amplitude (see Fig. \[fig:Fig3\](b)). The above discussion implies that the energy dissipation is related to the volume fraction of granular system, i.e., the looser packing is more dissipative. Considering that the energy dissipation of granular system increases with the transition quantity of force chains, we can say that the structural transition will occur more easily in the looser packing.
![\[fig:Fig5\] The RED virus the time with different amplitude $A_0$, as noted. Firstly the amplitude is swept down (a), and then the amplitude is swept up (b). The immersed depth $h=70 d$ and the forced frequency $\nu=1.0 Hz$.](Fig5){width="8.0cm"}
In order to understand the historical effects better, below we will discuss the aging effects in granular materials. Fig. \[fig:Fig5\] shows the RED as a function of the time with different amplitude $A_0$, as noted. At the beginning of all these experiments, the packing is rather looser. Then under a series of vibrations with the amplitude $A_0=0.284^{o}$, the granular system will be compacted [@Philippe]. According to the relationship between the volume fraction and the energy dissipation of granular system, we know that when the granular system is compacted the RED will decrease with the vibration time. As expected, Fig. \[fig:Fig5\](a) shows the results: firstly the RED decreases quickly and then slowly approaches a saturation value. In other words, at the beginning the structural transition of force chains will occur continuously with the energy dissipation of system. Then the transition quantity will decrease with the compaction of the granular packing, which is the reason of the decrease of energy dissipation. These analysis indicate that the structural transition of force chains has slowly changed the arrangement of the granular materials, i.e., the compaction of granular materials. And it is this arrangement change that leads to the change of energy dissipation with vibration time. The profiles in Fig. \[fig:Fig5\](a) can be well fitted with an exponential decay law, which is a fundamental character of the relaxation of granular systems [@Philippe]. This is the aging effects. Fig. \[fig:Fig5\](a) presents many similar profiles with different amplitude. The difference between them is that the quantity of RED decreases with decreasing forced amplitude. However, when we increase the amplitude step by step, we obtain some different profiles. As Fig. \[fig:Fig5\](b) shows, at the beginning the RED increases quickly and then slowly approaches a saturation quantity, what indicates that the relaxation of granular system has occurred and the arrangement of granular system must have changed. As mentioned above, it also is the structural transition of the force chains that leads to this change. However, the profiles in Fig. \[fig:Fig5\](b) follow the exponential grow law, which shows that the trend of the profiles is opposite with the profiles in Fig. \[fig:Fig5\](a). These differences between the profiles indicate that the arrangement of the grain packing changes differently when we increase the amplitude step by step. This is a process that the granular packing changes looser, in contrast to the compaction. In addition, the results confirms an earlier experimental observation: as the intensity of vibration decreases both the volume fraction of stationary and the compaction time increases [@Philippe]. The reason is that the transition quantity will be very small when the amplitude of vibration is lower. In according to the above analysis, we know that the forced vibration actually changes the microstructure (the volume fraction and the arrangement) of the granular system because of the slowly structural transition of force chains. This change of the microstructure is associated with the distribution of the various-size empty apace between grains, i.e., the defects motion and annihilation, which is the fundament of the historical effects.
The above discussion indicates that the transition of force chains is a relaxation process companied with the energy dissipation, as shown in the explanation of experimental data. The mechanical spectroscopy also show that a minute change of the arrangement of the granular system is sufficient to significantly change the amplitude response. It is these changes of the arrangement that are responsible for the well-known historical effects in the granular system preparation.
\[sec:sec5\]conclusion
======================
No body will be surprised that when he moves a solid intruder in a granular medium some deformations and some shear bands near interface occur [@Gennes; @Drake]. And we all know that the shear bands are the key to explain the “flowability" of granular materials. However, the shear banding behavior is very complex and is difficult to study. Many experiments and theoretical simulations have carried out and shown that the shear band thickness and shape are dependent on the shear strain and the boundary roughness conditions [@Goldenberg; @Siavoshi]. Here we investigated the structural transition of force chains in the shear bands from a view of energy dissipation. We found that the shape of the shear bands will change with the shear stress, as the microstructure of force chains shows. Of course, there are also many questions in understanding what happens on the grain scale [@Knight].
Different from the most common method to study force chains using birefringent materials [@Liu1; @Behringer], our investigation focus on the relationship between the energy dissipation and the force chains in granular systems, which gives a microscopic picture of the force chains in dense granular materials. In the experiments, we observed many interesting physical effects, such as the historical effects and aging effects. The results indicate that some structural transition of force chains, even changes of arrangement, have occurred. While the energy dissipation in this regime is a dynamic quality, we find that it is determined by the static structure of the medium. So we can say this experiment offers a new bridge between recent developments in understanding static configuration in granular media and the dynamic properties.
This work is financially supported by the National Natural Foundation of China under Grant No. 10674135 and the Knowledge Innovation Program of Chinese Academy of Sciences under Grant No. KJCX2-SW-W17.
[1]{} P. G. de Gennes, Rev. Mod. Phys. [**71**]{}, S374 (1999).
C.-h. Liu, S. R. Nagel, D. A. Schecter, S. N. Coppersmith, S. Majumdar, O. Narayan, and T. A. Witten, Science [**269**]{}, 513 (1995).
R. P. Behringer, D. Howell, L. Kondic, S. Tennakoon, and C. Veje, Physica D [**133**]{}, 1(1999).
A. J. Liu and S. R. Negel, *Jamming and Rheology: Constrained Dynamics on Microscopic and Macroscopic Scales* (Taylor & Franics, London, 2001).
C.-h. Liu and S. R. Nagel, Phys. Rev. Lett. [**68**]{}, 2301(1992).
C. Goldenberg and I. Goldhirsch, Nature (London) [**435**]{}, 188 (2005); S. Luding, Nature (London) [**435**]{}, 159 (2005). J. Geng *et al.*, Phys. Rev. Lett. [**87**]{}, 035506 (2001).
S. N. Coppersmith, C.-h. Liu, S. Majumdar, O. Narayan, and T. A. Witten, Phys. Rev. E [**53**]{}, 4673(1996).
J. F. Peters, M. Muthuswamy, J. Wibowo, and A. Tordesillas, Phys. Rev. E [**72**]{} 041307(2005); G. Lois and J. M. Carlson, Europhys. Lett. [**80**]{}, 58001(2007).
F. Radjai, M. Jean, J.-J. Moreau, and S. Roux, Phys. Rev. Lett. [**77**]{}, 274(1996).
J. Brujic, P. Wang, C. Song, D. L. Johnson, O. Sindt, and H. A. Makse, Phys. Rev. Lett. [**95**]{}, 128001 (2005).
D. M. Mueth, *et al.* Nature (London) [**406**]{}, 385 (2000). G. D’Anna, and G. Gremaud, Nature (London) [**413**]{}, 407 (2001).
G. D’Anna, Phys. Rev. E [**62**]{} 982 (2000).
W.-J. Wang, X.-Z. Kong, and Z.-G. Zhu, Phys. Rev. E [**75**]{}, 041302 (2007).
T. G. Drake, J. Grophys. Res. [**95**]{}, 8681(1990).
M. E. Cates, J. P. Wittmer, J.-P. Bouchaud, and P. Claudin, Phys. Rev. Lett. [**81**]{}, 1841 (1998); A. J. Liu, and S. R. Nagel, Nature (London) [**396**]{}, 21 (1998).
P. Umbanhowar and M. van Hecke, Phys. Rev. E [**72**]{}, 030301(R) (2005).
P. Philippe and D. Bideau, Europhys. Lett. [**60**]{}, 677 (2002).
S. Siavoshi, A. V. Orpe, and A. Kudrolli, Phys. Rev. E [**73**]{} 010301(R) (2006).
J. B. Knight, *et al.* Phys. Rev. E [**51**]{} 3957 [1995]{}; P. Richard, *et al.* Nature Materials [**4**]{} 121 [2005]{}.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recent experiments (Decelle et al., Phys. Rev. Lett. **102**, 027203 (2009), Ref. [@Decelle-09]) discovered an ultra-fast regime of spin avalanches in crystals of magnetic magnets, which was three orders of magnitude faster than the traditionally studied magnetic deflagration. The new regime has been hypothetically identified as magnetic detonation. Here we demonstrate the possibility of magnetic detonation in the crystals, as a front consisting of a leading shock and a zone of Zeeman energy release. We study the dependence of the magnetic detonation parameters on the applied magnetic field. We find that the magnetic detonation speed only slightly exceeds the sound speed in agreement with the experimental observations.'
author:
- 'M. Modestov'
- 'V. Bychkov'
- 'M. Marklund'
title: 'Ultra-fast spin avalanches in crystals of molecular magnets in terms of magnetic detonation'
---
Molecular magnetism is a rapidly developing interdisciplinary research area within material science [@Gatteschi-review-03; @Barco-review-05]. One of the widely investigated materials in the subject is $\rm{Mn}_{12} $-acetate, with a high spin number ($S = 10$) and strong magnetic anisotropy [@Gatteschi-review-03; @Barco-review-05; @Sessoli-Nature-93; @Friedman-PRL-96; @Thomas-Nature-96]. At sufficiently low temperature all the spins of the molecules become oriented along an applied external magnetic field, thus occupying the ground state (e.g. $S_{z} = 10$); in this state the magnetization reaches its saturation value. When the magnetic field direction is switched to the opposite one, the former ground state becomes metastable with an increased potential energy (the Zeeman energy) and a barrier separating it from the new ground state. Active research on the subject demonstrated that spin-relaxation from the metastable to the ground state often happens in the form of a narrow front spreading in a sample with velocity of a few meters per second [@Suzuki-05; @Hernandez-PRL-05; @McHugh-07; @Garanin-Chudnovsky-2007; @Hernandez-08; @Villuendas-08; @Modestov-2011]. Still, all these works focused on magnetic deflagration, i.e. a front of energy release propagating due to thermal conduction at velocities much smaller than the sound speed.
However, in contrast to other studies, recent experiments by Decelle et al., Ref. [@Decelle-09], discovered a new fast regime of the magnetic avalanches in $\rm{Mn}_{12}$-acetate with a front velocity estimated to be (2000-3000) m/s, which exceeds the typical magnetic deflagration speed by three orders of magnitude. Though a limited number of sensors led to rather large uncertainty in measuring the front velocity, the experiments still indicated clearly that it was comparable to the sound speed in the crystals. Furthermore, Decelle et al. [@Decelle-09] hypothetically interpreted the new regime as *magnetic detonation*. Although this hypothesis looked reasonable, it still required much theoretical work to be justified. In particular, detonations in combustion problems demonstrate a propagation velocity larger than the sound speed by order of magnitude and destructively high pressure [@LL-Fluidmechanics; @Law-book]. Besides, even in combustion science, the phenomenon of deflagration-to-detonation transition has remained one of the least understood processes for more than seventy years, despite its extreme importance [@Law-book; @Dorofeev-2011]. It is only recently that a quantitative theoretical understanding of this process has been achieved [@Bychkov-et-al-2005; @Akkerman-et-al-2006; @Bychkov-et-al-2008; @Valiev-et-al-2010].
In this Letter we demonstrate the possibility of magnetic detonation, in the form of a front with a leading shock and a zone of Zeeman energy release. We study the dependence of the magnetic detonation parameters on the applied magnetic field. We find that the magnetic detonation speed is only slightly greater than the sound speed, in agreement with experimental observations.
In line with the experiments [@Decelle-09], we consider magnetic avalanches in $\rm{Mn}_{12}$-acetate with the Hamiltonian $$\label{eq1}
\mathcal{H} = - \beta S_{z}^{2} - g\mu_{B} H_{z} S_{z},$$ suggested in [@Garanin-Chudnovsky-2007]. Here $S_{z}$ is the spin projection, $\beta \approx 0.65K$ is the magnetic anisotropy constant, $g \approx 1.94$ is the gyromagnetic factor, $\mu_{B}$ is the Bohr magneton, and $H_{z} $ is the external magnetic field. The Hamiltonian (1) determines the Zeeman energy release and the energy barrier of the spin transition (in temperature units), which depend on the magnetic field as $$\begin{aligned}
\label{eq2}
& E_{a}&= \beta S_{z}^{2} - g\mu_{B}H_{z}S_{z} + \frac{g^2}{4\beta}\mu_{B}^{2}H_{z}^{2},\\
\label{eq3}
& Q&= 2g\mu _{B} H_{z} S_{z},\end{aligned}$$ respectively, with $S_{z} =10$. The energy barrier decreases with the field while the Zeeman energy increases linearly. Next, we consider a stationary magnetic detonation in a crystal of molecular magnets. Similar to the theory of shocks in solids [@Zeldovich-Raizer], we adopt the reference frame of the detonation front and find the conservation laws of mass, momentum and energy according to $$\begin{aligned}
\label{eq4}
\rho_{0}D &=& \rho u,\\
\label{eq5}
P_{0} + \rho_{0} D^{2} &=& P + \rho u^{2},\\
\label{eq6}
\varepsilon_{0} + \frac{P_{0}}{\rho_{0}} + \frac{1}{2}D^{2} + Q &=&
\varepsilon + \frac{P}{\rho} + \frac{1}{2}u^{2} + Q a,\end{aligned}$$ where the label 0 designates the initial state, $D$ is the detonation speed, $u$ is velocity produced by the detonation, $\varepsilon$ is thermal energy per molecule, and $a$ is the fraction of molecules in the metastable state. Here we neglect the thermal conduction, since this is a comparatively slow process. In the theory of shock waves one introduces the volume per unit mass $V \equiv 1 / \rho$ instead of density. The conservation laws Eqs. (\[eq4\])–(\[eq6\]) have to be complemented by an equation of state. Following Ref. [@Zeldovich-Raizer], we represent the pressure and energy of condensed matter at low temperature as a combination of elastic and thermal components according to $$\label{eq7}
P = \frac{c_{0}^{2}}{V_{0} n}\! \left[ \left( \frac{V_{0}}{V}
\right)^{n}\!- 1 \right] \!+ \frac{\Gamma }{V} \frac{Ak_{B} T^{\alpha+1}} {(\alpha+1)\Theta_{D}^{\alpha}},$$ $$\label{eq8}
\varepsilon = \frac{c_{0}^{2}}{n} \!\left\{ {\frac{1}{n -
1}{\left[ {\left( \frac{V_{0}} {V} \right)^{n - 1} \!\!\!- 1} \right]}\! +
\!\frac{V}{V_{0}} - 1} \right\}\! + \frac{Ak_{B} T^{\alpha+1}} {(\alpha+1)\Theta_{D}^{\alpha}},$$ where $c_0$ is the sound speed (we take $c_0\approx 2000\ $m/s in accordance to [@Decelle-09]), the power exponent $n\approx 4$ as suggested in [@Zeldovich-Raizer], $\Gamma \approx 2$ is the Gruneisen coefficient, $\Theta_{D}$ is the Debye temperature with $\Theta_{D} = 38K$ for $\rm{Mn}_{12}$, $k_{B}$ is the Boltzmann constant, $A = 12\pi^{4} / 5$ corresponds to the simple crystal model, $\alpha=3$ is the problem dimension. Thus, in Eqs. (\[eq4\])–(\[eq8\]) we have a complete system for describing magnetic detonation in molecular magnets.
The properties of shocks and detonations are represented by the Hugoniot/detonation curve $P=P(V)$, see Ref. [@LL-Fluidmechanics; @Law-book; @Zeldovich-Raizer]. We also introduce the scaled density ratio $r = \rho / \rho_{0} = V_{0}/V$ which characterizes the matter compression. Using Eqs. (\[eq4\])-(\[eq8\]), we derive the following implicit form for the detonation relation $$\begin{aligned}
\label{eq9}
&& \left( \frac{1}{\Gamma} - \frac{r - 1}{2}
\right)\frac{P}{\rho _{0}} = \,rQ\left( {1 - a} \right) + \left( r + \frac{r - 1}{2}\Gamma \right)\varepsilon_{0} \nonumber\\
&& +\frac{c_{0}^{2}}{n - 1} \left[ {r - 1 -
\left( {1 - \frac{n-1}{\Gamma}}\right)\frac{r^{n} - 1}{n}} \right].\end{aligned}$$ In the case of zero energy release ($a=1$), Eq. (\[eq9\]) reduces to the Hugoniot equation for a shock wave (which we denote by the subscript $s$). In the detonation, the leading shock compresses the sample, increases temperature and hence facilitates the spin reversal with the Zeeman energy release. The released Zeeman energy provides expansion of the medium, which acts like a piston and supports the leading shock. In the case of the completed spin reversal ($a=0$), Eq. (\[eq9\]) describes the final state behind the detonation front (which we denote by the subscript $d$).
{width="3.4in" height="2.5in"}
The insert of Fig. 1 shows the Hugoniot and detonation curves found using Eq. (\[eq9\]) for $H=4$ T. We assume that there is no external atmospheric pressure and the initial temperature is negligible, which corresponds to the initial point ($V=V_0;\, r=1;\ P=0$). Because of the energy release, the detonation curve is always above the Hugoniot one. In the case of $\rm{Mn}_{12}$ we find that the elastic contribution to the pressure and energy dominates over the thermal one, which leads to a rather weak detonation with the shock and detonation curves almost coinciding as shown at the inset of Fig. 1. A self-supporting detonation corresponds to the Chapman-Jouguet (CJ) regime, for which velocity of the products in the reference frame of the front is equal to the local sound speed [@LL-Fluidmechanics]. The CJ point at the detonation curve is determined by the tangent line connecting the initial state and the detonation curve. Since the detonation and Hugoniot curves are extremely close at the insert of Fig. 1, the intersection of the tangent line cannot be seen in the traditional representation of the curves. In order to make the figure illustrative, we subtract this tangent line from the Hugoniot and detonation curves in Fig. 1. In the new representation, the tangent line corresponds to the zero line, while the Hugoniot and detonation curves may be distinguished quite well. The CJ point in Fig. 1 corresponds to the final state behind the detonation front. The shock point indicates the strength of the leading shock as determined by the Zeeman energy release for the CJ regime; the density and the pressure acquire maximum values at the shock front. The Zeeman energy release behind the shock produces expansion of matter with an ensuing pressure reduction.
We notice from Fig. 1 that the $\textrm{Mn}_{12}$ crystal is compressed by few percents in the detonation wave, which makes an analytical theory for the detonation front parameters possible using expansion $r=1+\delta$ with $\delta\ll1$. Then, to leading order in $\delta$, Eqs. (\[eq7\]), (\[eq9\]) reduce to $$\begin{aligned}
\label{eq10}
P &=& \rho_{0} \left[ {\Gamma Q (1-a) + c_{0}^{2} \delta} \right],\\
\label{eq11}
T^{\alpha + 1} &=& \left( {\alpha + 1} \right) \frac{\Theta_{D}^{\alpha
}}{Ak_{B}} \left[ {Q (1-a) + \frac{n + 1}{12}c_{0}^{2} \delta ^{3}}\right].\end{aligned}$$ We find the final compression behind the detonation front as $$\label{eq12}
\delta_{d} = \frac{1}{c_{0}} \sqrt {\frac{2\Gamma Q}{n + 1}}.$$ The compression behind the leading shock is larger by a factor of 2, $\delta_s \approx 2 \delta_d$, as may be seen from the parabolic shape of the Hugoniot and detonation curves in Fig. 1. The detonation speed may be found from Eq. (\[eq4\]) as $$\label{eq13}
D = c_{0} \left( {1 + \frac{n + 1}{4}\delta_{s} }\right),$$ which means that the magnetic detonation speed slightly exceeds the sound speed in agreement with the experimental observations [@Decelle-09]. Substituting $\delta_s$, $\delta_d$, and $a=1; \, 0$ into Eqs. (\[eq10\]), (\[eq11\]), we find the analytical formulas for pressure and temperature at the shock and behind the detonation front, respectively. Taking into account Eq. (\[eq3\]), these formulas specify the dependence of the detonation parameters on the external magnetic field. In Fig. 2 we compare the analytical theory to the numerical solution to Eq. (\[eq9\]). The curves of Fig. 2a show the density at the shock wave and behind the detonation front, when all the spins have been aligned along the magnetic field. We see that the density at the shock increases by less than 3 percents if the magnetic field is below 10 T. Because of this small compression, the analytical theory (\[eq12\]) is in a very good agreement with the numerical solution. The maximum value of shock pressure is about 1.2 atm for 10 T. Thus, due to the small compression and the moderate pressure increase, the magnetic detonation does not destroy the magnetic properties of the crystals.
{width="3.4in" height="2.45in"} {width="3.4in" height="2.45in"}
At the same time, the crystal temperature increases considerably because of the shock, and this stimulates a fast spin reversal and a further temperature increase. Fig. 2b illustrates the temperature increase at the shock wave and behind the detonation front. Again, we observe very good agreement between the analytical theory and the numerical solution. The temperature at the leading shock is comparable to that expected for the magnetic deflagration [@Suzuki-05; @Hernandez-PRL-05; @McHugh-07; @Hernandez-08], which also makes the reaction time comparable in both processes. In classical combustion, the temperature at the leading shock in the detonation wave is still quite small in comparison with the activation energy of the chemical reactions, so that the active reaction zone lags considerably behind the shock [@Law-book]. The situation may be quite different in magnetic detonation. When the magnetic field is stronger than $2-3$ T, the shock temperature is relatively high ($E_a/T_s<5$) so that active spin reversal starts right at the shock wave. Figure 2b presents also the energy barrier as a function of the external magnetic field. The energy barrier decreases with the growth of the magnetic field, Eq. (\[eq2\]), as shown in Fig. 2b. When the magnetic field exceeds 10 T, the energy barrier vanishes, the metastable state turns unstable, and the molecules may settle down freely to the ground state. Hence, one may interpret magnetic avalanches as detonation or deflagration only for the fields below 10 T.
Finally, we describe the internal structure of the magnetic detonation front. In the reference frame of the moving front, the molecule fraction with the spin opposite to the field direction is determined by [@Garanin-Chudnovsky-2007] $$\label{eq14}
u{\frac{\partial a}{\partial x}} = \frac{a}{\tau _{R}}\exp \left(
{ - \frac{E_{a}}{T}} \right),$$ where $\tau_R$ is a constant of time dimension characterizing the spin reversal. We integrate Eq. (\[eq14\]) numerically together with Eqs. (\[eq4\]) and (\[eq7\]) along the tangent line in Fig. 1, from the shock to the CJ point; the obtained profiles are depicted in Fig. 3 for $H=3$ T.
{width="0.95\columnwidth"}
The background shading represents the energy release due to the spin reversal; the temperature and the pressure are scaled to their maximal values. The coordinate is scaled by the characteristic length $L_0=c_0\tau_R \approx 2 \cdot 10^{-4} \textrm{m}$, where we take $\tau_R \approx 10^{-7}\textrm{s}$ as obtained in several experiments [@Suzuki-05; @Hernandez-PRL-05; @McHugh-07; @Hernandez-08]. Using this value we can estimate the characteristic width of the stationary detonation wave to a few millimeters. The applied magnetic field influences strongly the reaction rate and thus the front width. For magnetic fields higher than 5 T, the detonation width is $< 1$ mm, while for a weaker field the width may increase considerably. For this reason, the detonation mechanism in molecular magnets may only be observed in experiments utilizing high enough magnetic fields, since the typical sample size is of order of several millimeters. The typical scales in the experiments of Ref. [@Decelle-09] were also about a few millimeters. Thus the experimentally observed fast avalanche regime was, presumably, a non-stationary detonation in the process of developing.
To summarize, in this Letter we have developed a theory of magnetic detonation in molecular magnets, which explains a new regime of ultra-fast spin avalanches discovered recently in the experiments of Ref. [@Decelle-09]. The detonation regime is two to three orders of magnitude faster than the magnetic deflagration observed before [@Suzuki-05; @Hernandez-PRL-05; @McHugh-07; @Hernandez-08]. We have shown that the leading shock triggers the spin reversal in these magnetic systems, and that the magnetic detonation propagates with velocities slightly larger than the sound speed. In contrast to traditional detonations in combustion, which are characterized by strongly supersonic velocities and ultra-high pressure, magnetic detonations involve rather moderate pressure increase, which is about 1 atm even for considerable magnetic fields. For this reason, magnetic detonation does not destroy magnetic properties of the crystals, a very important conclusion in view of possible applications of molecular magnets to, e.g., quantum computing.
This work was supported by the Swedish Research Council and by the Kempe Foundation. The authors thank Petter Minnhagen, Bertil Sundqvist, Thomas W[å]{}gberg, Sune Pettersson, Tatiana Makarova and Valeria Zagainova for useful discussions.
W. Decelle, J. Vanacken, V. V. Moshchalkov, J. Tejada, J. M. Hernandez, and F. Macia, Phys. Rev. Lett. **102**, 027203 (2009).
D. Gatteschi and R. Sessoli, Angew. Chem., Int. Ed. **42**, 268 (2003).
E. del Barco, A. D. Kent, S. Hill, J. M. North, N. S. Dalal, E. Rumberger, D. N. Hendrikson, N. Chakov, and G Christou, J. Low Temp. Phys. **140**, 119 (2005).
R. Sessoli, D. Gatteschi, A. Caneschi, and M. A. Novak, Nature (London) **365**, 141 (1993).
J. R. Friedman, M. P. Sarachik, J. Tejada, and R. Ziolo, Phys Rev. Lett. **76**, 3830 (1996).
L. Thomas, F. Lionti, R. Ballou, D. Gatteschi, R. Sessoli, and B. Barbara, Nature (London) **383**, 145 (1996).
Y. Suzuki, M. P. Sarachik, E. M. Chudnovsky, S. McHugh, R. Gonzalez-Rubio, N. Avraham, Y. Myasoedov, E. Zeldov, H. Shtrikman, N. E. Chakov, and G. Christou, Phys. Rev. Lett. **95**, 147201 (2005).
A. Hernandez-Minguez, J. M. Hernandez, F. Macia, A. Garcia-Santiago, J. Tejada, and P. V. Santos, Phys. Rev. Lett. **95**, 217205 (2005).
S. McHugh, R. Jaafar, M. P. Sarachik, Y. Myasoedov, A. Finkler, H. Shtrikman, E. Zeldov, R. Bagai, and G. Christou, Phys. Rev. B **76**, 172410 (2007).
A. Hernandez-Minguez, F. Macia, J. M. Hernandez, J. Tejada, and P. V. Santos, J. Magn. Magn. Mater. **320**, 1457 (2008).
D. A. Garanin and E. M. Chudnovsky, Phys. Rev. B. **76**, 054410 (2007).
D. Villuendas, D. Gheorghe, A. Hernandez-Minguez, F. Macia, J. M. Hernandez, J. Tejada, R. J. Wijngaarden, EPL (Europhysics Letters) **84**, 67010 (2008).
M. Modestov, V. Bychkov, and M. Marklund, accepted in Phys. Rev. B.
L. Landau and E. Lifshitz, *Fluid Mechanics*, Pergamon Press, Oxford, 1989.
C.K. Law, *Combustion Physics*, Cambridge University Press, NY, 2006.
S. Dorofeev, Proc. Combust. Inst. **33** 2161 (2011).
V. Bychkov, A. Petchenko, V. Akkerman, L.-E. Eriksson, Phys. Rev. E **72**, 046307 (2005).
V. Akkerman, V. Bychkov, A. Petchenko, L.-E. Eriksson, Combust. Flame **145**, 206 (2006).
V. Bychkov, D. Valiev, L.-E. Eriksson, Phys. Rev. Lett. **101**, 164501 (2008).
D. Valiev, V. Bychkov, V. Akkerman, C. K. Law, L.-E. Eriksson, Combust. Flame **157**, 1012 (2010)
Ya. B. Zeldovich and Yu. P. Raizer, *Physics of Shock Wave and High-Temperature Hydrodynamic Phenomena*, Dover Publications, Inc. Mineola, New York (2002).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We consider a microscopic model (a system of self-propelled particles) to study the behaviour of a large group of pedestrians walking in a corridor. Our point of interest is the effect of anisotropic interactions on the global behaviour of the crowd. The anisotropy we have in mind reflects the fact that people do not perceive (i.e. see, hear, feel or smell) their environment equally well in all directions. The dynamics of the individuals in our model follow from a system of Newton-like equations in the overdamped limit. The instantaneous velocity is modelled in such a way that it accounts for the angle under which an individual perceives another individual.\
We investigate the effects of this perception anisotropy by means of simulations, very much in the spirit of molecular dynamics. We define a number of characteristic quantifiers (including the polarization index and Morisita index) that serve as measures for e.g. organization and clustering, and we use these indices to investigate the influence of anisotropy on the global behaviour of the crowd. The goal of the paper is to investigate the potentiality of this model; extensive statistical analysis of simulation data, or reproducing any specific real-life situation are beyond its scope.
address:
- '$^1$ Department of Mathematics and Computer Science, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands'
- '$^2$ Centre for Analysis, Scientific computing and Applications, Department of Mathematics and Computer Science & Institute for Complex Molecular Systems, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands'
- '$^3$ Theory of Polymers and Soft Matter, Department of Applied Physics, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands'
author:
- 'Lennart Gulikers$^1$, Joep Evers$^2$, Adrian Muntean$^2$ and Alexey Lyulin$^3$'
title: The effect of perception anisotropy on particle systems describing pedestrian flows in corridors
---
[*Keywords*]{}: Traffic and crowd dynamics, interacting agent models, self-propelled particles, pattern formation (Theory)
\[sec:level1\]Introduction
==========================
During the last two, three decades, the field of crowd dynamics has emerged as the *natural* sciences’ reaction to questions arising from *social* sciences, population biology and urban planning. See e.g. [@Stott] for an example of a problem addressed in psychology, or [@Carroll] for an illustration of the civil engineering aspects. The roots and philosophy of crowd dynamics are very much in the spirit of statistical mechanics, molecular dynamics, interacting particle systems methods and the theory of granular matter, as such treating individual humans nearly as non-living material (cf. e.g. the nice overview [@Schadschneider2011] and references cited therein). A justification for this approach lies in the fact that the individuals’ personal will is more or less averaged out if one looks at the crowd as a whole. From this perspective, it can be considered as (stochastic) noise, superimposed on some ‘clean’ (deterministic) dynamics.\
To illustrate the thin borderlines between several fields of study, the reader is referred to e.g. [@Lutz2; @Lutz1] for the dynamics of non-living particles, [@Lega; @Peruani] for studies of tumbling or self-propelled living particles (like bacteria), or [@Molnar; @CRAS; @CristianiTosin] for crowd dynamics. Although these fields all focus their own specific real-world scenario, their way of thinking and posed questions are very much alike.\
\
However, an evident and important difference between people and molecules or grains (apart from people’s own opinions, irritations etc.) is the fact that people clearly have front and back sides. Our degree of perceiving our surroundings highly depends on the direction of looking. We mainly base our walking behaviour on what we see, and clearly what happens in front of us thus has more influence than what happens behind us (this statement is also supported e.g. by [@Guo12]). A modification or extension of physics-inspired models is needed to incorporate this kind of anisotropy in the interactions between individuals. This paper investigates the effect of anisotropy on the global behaviour of a group of pedestrians.\
\
Our focus is on the simulation of a scenario where pedestrians move in a long corridor. We might relate this situation to evacuation of people from a building (cf. [@Armin; @HelbingNature]). It is sane to assume that these evacuees have an intrinsic *drive* to move towards the exit (i.e. one side of the corridor), and moreover that there view is focused in the same direction. Investigating the effect of anisotropy on the large-scale behaviour of the crowd therefore relates to assessing the escape process.\
\
In Section \[sec:level1\] of this paper the model is presented and explained. Section \[sec:sim setup and results\] is the main part of the paper. It describes the exact scenario of our simulations and the definitions of the quantities we use for assessing the results (polarization index, projected density, Morisita index). Moreover, in this section the simulation results are presented and discussed. Conclusions and an outlook on possible future work are given in Section \[sec:conclusion\].
\[sec:level1\]A model for anisotropic interactions between pedestrians
======================================================================
We represent pedestrians by point particles[^1] having masses $m_i$. They are located in a long corridor of length $\tilde{L}$ and width $B$. Here, the word ‘long’ refers to the fact that at the time scales we focus on, the pedestrians are not able to reach the end of the corridor. Interactions between pedestrians are short-ranged. We therefore suppose that the correlation length in the system is less than or equal to a certain $L\ll\tilde{L}$ and we can subdivide the corridor in an array of rectangles (width $B$ and length $L$), which are all duplicates of each other. We thus have a scenario with periodic boundary conditions. Our domain of interest is therefore a rectangular box $$\Omega:= [ -\frac{L}{2} , \frac{L}{2} ] \times [ -\frac{B}{2} , \frac{B}{2} ],$$ with periodic boundary conditions in one direction and impermeable walls in the other direction. The corridor contains $N < \infty$ pedestrians. For all $i$ $\in \{1, ... , N\}$ and $t\geq 0$, the vector $\vec{r}_i (t) = (x_i(t) , y_i(t)) \in \Omega$ represents the position of the $i$-th pedestrian at time $t$. We denote its velocity by $\vec{v}_i(t)$.\
\
We assume that the governing equation of motion is $$\frac{m_i}{\tau_{drive}} (\vec{v}_i(t) - \vec{v}_{des}) = \vec{F}_i^{soc} + \vec{F}_i^{phys}.
\label{newton_overdamped}$$ The equation describes the motion of the $i$-th individual, which has mass $m_i$ and which moves with velocity $\vec{v}_i(t)$. However, he/she *tries* to move according to its desired velocity $\vec{v}_{des}$. Here, $\tau_{drive}$ is the characteristic relaxation time related to attaining the desired velocity. Its actual velocity is moreover perturbed by two ‘forces’. The word ‘force’ is used since (\[newton\_overdamped\]) can be regarded as an overdamped limit of a Newton-like equation (cf. [@Molnar] for this Newton-like way of modelling).\
One could argue whether the social force is the right concept to use to drive the pedestrians, or maybe ideas like social pressure (as in a Darcy-like law) or cognitive-based heuristics (see e.g. [@Moussaid]) are more appropriate. Here we avoid any polemic by deciding to choose a framework based on social forces and leave for later any further developments of other possible approaches.\
\
There is a physical force $\vec{F}_i^{phys}$ that acts on the individual to describe the effect of the non-living environment (geometry). In this paper we only take into account the influence of walls on pedestrians, that is: $\vec{F}_i^{phys}$ = $\vec{F}_i^{wall}$. Furthermore, pedestrian $i$ experiences a so-called social force $\vec{F}_i^{soc}$ due to the presence of other individuals, which influences the motion of this particular pedestrian $i$.\
\
Individuals are influenced by the walls as soon as they come too close, i.e. within a distance $R_{wall}$. We model these impermeable walls by means of a strong repulsive force $\vec{F}_i^{wall}$ acting on pedestrian $i$: $$\vec{F}_i^{wall} = \left\{
\begin{array}{ll}
F_{Wall}\, (1 - \frac{R_{wall}}{d}) \vec{n}, & \hbox{if $d<R_{wall}$;} \\
\vec{0}, & \hbox{otherwise.}
\end{array}
\right.
\label{FRight}$$ Here, $\vec{n}$ is the unit normal pointing from the corresponding wall into the corridor, $F_{Wall}$ is the strength of the repulsive force and $d$ is the distance to the wall for pedestrian $i$. The word ‘strong’ here implies that this force is not just a contact force, but has a longer range. Typically, this makes individuals avoid walls before touching them.[^2]\
\
Furthermore, very much in the spirit of [@Molnar], we specify the social force by $$\vec{F}_i^{soc} = \sum_{\vec{r}_j \in \Omega_i} - \nabla W(\vec{r}_i - \vec{r}_j),
\label{Fsoc}$$ where:
- $\Omega_i$ is the collection of the position vectors of all individuals which are within a distance $R_{cut}$ to pedestrian $i$. In other words, pedestrians interact only when they are close enough to each other;
- we assume that the interaction potential $W$ depends only on the relative position of the two pedestrians $i$ and $j$ and not on their relative velocity.
(0,0)–(0,1)node\[anchor=north east\][$\vec{e}_y$]{}; (0,0)–(1,0)node\[anchor=north east\][$\vec{e}_x$]{};
(0,0)–(1,2)node\[anchor=north east\][$\vec{r}_i$ ]{}; (0,0)–(3,4)node\[anchor=north east\][ ]{};
(1,2)–(3,4) node\[midway, sloped, above\][$\vec{r}_j-\vec{r}_i$]{} node\[anchor=north west\][$\vec{r}_j$]{};
(1,2)–(3.5,2) node\[midway, sloped, below\][$\vec{v}_{des}$]{};
(2.0, 2) arc (0:45:1.0); at (2.25,2.5) [$\theta_{r_i r_j}$]{};
Specifically, $W$ takes the form, $$\label{W}
W(\vec{r}_i - \vec{r}_j) = U(|\vec{r}_i - \vec{r}_j|)\, \left( 1-\frac{\alpha}{2}(1-\cos\theta_{r_i r_j})\right ).$$ Here, $\alpha \in [0,1]$ is an *anisotropy parameter*, for which $\alpha=0$ means that the potential is isotropic, and $\alpha=1$ means that the anisotropy effects are maximal. The angle of perception $\theta_{r_i r_j}$ is the angle under which an individual positioned in $\vec{r}_i$, while moving in the direction of $\vec{v}_{des}$, perceives location $\vec{r}_j$, see Figure \[figure geometry\]. The precise type of interaction is hidden in the structure of the function $U$. In particular, we distinguish between two types of interactions, namely
- *only repulsive interaction* (for simplicity denoted by ). Pedestrians repel each other when their separation distance is smaller than $R_r^R$ (called radius of repulsion in ) and do not interact at larger distances;
- *both repulsive and attractive interaction* (denoted by ). For this type of interaction, individuals repel each other when their separation distance is smaller than $R_r^{AR}$ (called repulsive radius in ). However, when they are separated by a distance between $R_r^{AR}$ and $R_a^{AR}$ (called attractive radius in AR case), they are attracted to one another. They do not interact outside these regions.
For these two cases $R_{cut}=R_r^R$, respectively $R_{cut}=R_a^{AR}$. The second case is an extension of the first one. The we might regard as a population of individualistic people that simply try to avoid each other. In the there is also some *social cohesion*, as they try to keep the group together. In the following, we describe the precise structure of our potentials.
R case
------
In the repulsive case we take $$\label{U repulsive}
U(s) = \left\{
\begin{array}{ll}
F^R(s-R^R_r-R^R_r\ln\frac{s}{R^R_r}) & \hbox{if $s<R^R_r$;} \\
0 & \hbox{if $s>R^R_r$.}
\end{array}
\right.$$ See Figure \[prim\_r\] for an example of an interaction potential of the above form.
![Typical example of the interaction potential as given in (\[U repulsive\]).[]{data-label="prim_r"}](figure2.pdf)
AR case
-------
If we want to involve a both attractive and repulsive way of interactions, then we take $$U(s) = \left\{
\begin{array}{ll}
F^{AR}(s-R^{AR}_r\ln\frac{s}{R^{AR}_r})+C_2 & \hbox{if $s<R^{AR}_r$;} \\
\tilde{U}(s) & \hbox{if $R^{AR}_r<s<R^{AR}_a$;} \\
0 & \hbox{if $s>R^{AR}_a$;}
\end{array}
\right.
\label{U_attr_rep}$$ where $$\label{U attr-rep parabola}
\tilde{U}(s)=\frac{F^{AR}(\frac{s^3}{3}-(R^{AR}_r+R^{AR}_a)s^2+R^{AR}_rR^{AR}_as)}{R^{AR}_r(R^{AR}_a-R^{AR}_r)}+C_1.$$ In the above, the constants $C_1$ and $C_2$ are such that $U$ is a continuous function. Figure \[prim\_ar\] shows an example of this kind of interaction potentials.
![Typical example of the interaction potential as given in (\[U\_attr\_rep\])–(\[U attr-rep parabola\]).[]{data-label="prim_ar"}](figure3.pdf)
\[sec:sim setup and results\]Simulation: set-up and results
===========================================================
For all simulations, the number of pedestrians $N$ is an integer multiple of 10. Initially, we place these individuals on a lattice of $N_x = \frac{N}{10}$ rows of 10 pedestrians each. The distance between two succeeding rows is always $\Delta x = \frac{L}{N}$. Note that there is an $N$ here, not $N_x$: the pedestrians are thus distributed initially in a domain about one tenth the length of the total corridor (see Figure \[pos\_0\]). The distance between two pedestrians in the same row is $\Delta y = \frac{B}{10}$. Let $(l_i,y_j)$ denote the position of pedestrian $k:= i + 10(j-1)$, where $i=1,\ldots,10$ and $j=1,\ldots,N_x$. Then the coordinates are $l_i = \frac{-L}{2} + \Delta x (i - \frac{1}{2})$ and $y_j = \frac{-B}{2} + \Delta y (j - \frac{1}{2})$. We use these initial conditions for *all* simulations.\
\
![A snapshot of the positions of the 100 pedestrians at time $t=0$ s. This initial configuration is used in all simulations. The markers that indicate the individuals’ positions are smaller than in Figures \[pos\_r\_00\]–\[pos\_ar\_10\], just to avoid overlap here.[]{data-label="pos_0"}](figure4.pdf)
The simulation time corresponds to a real life time $\tau_{obs} = 100$ s. This time is large enough to witness a stable profile for all simulations. We give in Table \[modpar\] a summary of the parameters and their values used in both the ‘repulsive’ and ‘attractive and repulsive’ interaction potentials.[^3]\
----------------- ------- ---------------------
$N$ - 20, 40, 60, 80, 100
$\alpha$ - 0.0, 0.5, 1.0
$\Delta t$ s $0.0001$
$\tau_{drive}$ s $1.0$
$\tau_{obs}$ s $100$
$m$ kg $50$
$B$ m 10
$L$ m 40
$\mathcal{M}$ - 64
$S_x \cdot S_y$ m$^2$ 2.5$\cdot$2.5
$F_r^R$ N 15
$F^{AR}$ N 15
$R_r^R$ m 4.0
$R_r^{AR}$ m 1.5
$R_a^{AR}$ m 3.0
----------------- ------- ---------------------
: Model parameters, symbols, units and reference values. We do simulations in two cases. The : pedestrians interact in a both attractive and repulsive way and the : pedestrians only interact in a repulsive way. NB: $\mathcal{M}$ and $S_x \cdot S_y$ relate to the Morisita index and are specified in Section \[sect:def Morisita\].\
[]{data-label="modpar"}
For more technical details about this kind of simulations, the reader is referred e.g. to [@Ascher; @Bornemann].\
\
To illustrate simulation results, in Figures \[pos\_r\_00\]–\[pos\_ar\_10\] snapshots of the positions of $N=100$ pedestrians at time $t = 100$ s are shown, in both the and the for $\alpha = 1.0 , \alpha = 0.5$ and $\alpha = 0.0$ (fully isotropic), respectively. Some individuals’ positions might seem to coincide with the walls. Note however that our individuals are point particles and the walls have no width. The overlap is therefore only a result of the fact that we need to give size to individuals and walls in order to visualize them.
![A snapshot of the positions of the 100 pedestrians at time $t = 100$ s, plot for the R potential with $\alpha = 1.0$ (most anisotropic interactions).[]{data-label="pos_r_00"}](figure5.pdf)
![A snapshot of the positions of the 100 pedestrians at time $t = 100$ s, plot for the R potential with $\alpha = 0.5$.[]{data-label="pos_r_05"}](figure6.pdf)
![A snapshot of the positions of the 100 pedestrians at time $t = 100$ s, plot for the R potential with $\alpha = 0.0$ (isotropic interactions).[]{data-label="pos_r_10"}](figure7.pdf)
![A snapshot of the positions of the 100 pedestrians at time $t = 100$ s, plot for the AR potential with $\alpha = 1.0$ (most anisotropic interactions). Note that the symmetry is broken due to discretization errors.[]{data-label="pos_ar_00"}](figure8.pdf)
![A snapshot of the positions of the 100 pedestrians at time $t = 100$ s, plot for the AR potential with $\alpha = 0.5$.[]{data-label="pos_ar_05"}](figure9.pdf)
![A snapshot of the positions of the 100 pedestrians at time $t = 100$ s, plot for the AR potential with $\alpha = 0.0$ (isotropic interactions). See main text for the used initial conditions.[]{data-label="pos_ar_10"}](figure10.pdf)
Except for Figure \[pos\_ar\_00\], all figures depict a completely symmetric profile with respect to the line $y = 0$. This is a natural result of the symmetry in the initial data and symmetry w.r.t. the direction of $\vec{v}_{des}$ in the interactions; it is just a property of the hyperbolic conservation laws to govern the mean-field behaviour, cf. [@Dafermos]. Note also the strong tendency of the system to maintain (or produce) organized patterns. These issues will be addressed in more detail in Section \[sect: measured quant results\].\
Figure \[pos\_ar\_00\] is an exception in the sense that the distribution of the crowd is not symmetric around the line $y=0$. This cannot be explained from the model equations, as these imply that the distribution *should* be symmetric. This effect must be due to round-off errors and their propagation with respect to time.\
In the for $\alpha = 0.0$ (cf. Figure \[pos\_ar\_10\],) we see that the people move in a crystal-like formation parallel to the $\vec{e}_x$ - axis, which remotely resembles the 2D crystallization patterns at low temperature pointed out in [@Theil]. The group as a whole is very compact. In the sequel we will use the words clusters/clustering for a situation in which pedestrians move close together.\
However, if we compare Figure \[pos\_ar\_10\] to Figures \[pos\_ar\_00\] and \[pos\_ar\_05\], we observe that introducing anisotropy by setting $\alpha>0$ decreases the amount of structure and clustering. In the , all three values of $\alpha$ allow for a well-structured way of moving, but there is no clear clustering. It is evident from the figures that the people are spread over the whole corridor.\
\
Now, we introduce a number of measurable quantities to help quantify the above statements.
\[sec:level2\]Definitions of the measured quantities
----------------------------------------------------
### Polarization index
Inspired by [@Lega], we define the (time-dependent) polarization index $p$ of a group of people as the average angular deviation from the mean propagation direction. Here, the average is taken over all individuals. Note that zero polarization means that all the people move in parallel.\
We first introduce the following auxiliary definitions:
- $\langle q \rangle_t := \frac{ \sum_{j = 1}^{M} q(t_j) } {M}$ denotes the time average of an arbitrary quantity $q$, based on the values $q(t_j)$ at time $t_j$, $j \in \{ 1,2, ... , M \}$;
- $\theta_i(t) \in (-\pi , \pi]$ is the direction of motion of the pedestrian $i$ at time $t$. It is defined to be the angle $\theta_i(t)$ such that $$\mbox{tan}(\theta_i(t)) = \frac{ \vec{v}_i(t) \cdot \vec{e}_y }{ \vec{v}_i(t) \cdot \vec{e}_x };$$
- $\theta(t) \in (-\pi , \pi]$ is the mean direction of motion of the pedestrians group. It is formally defined as the angle such that $$\mbox{tan}(\theta(t)) = \frac{ \langle \vec{v}(t) \cdot \vec{e}_2 \rangle_N}{\langle \vec{v}(t) \cdot \vec{e}_1 \rangle_N},$$ where $\langle \vec{v}(t) \cdot \vec{e}_\xi \rangle_N$ denotes averaging over the total number of $N$ individuals: $$\langle \vec{v}(t) \cdot \vec{e}_\xi \rangle_N := \frac{ \sum_{j = 1}^{N} \vec{v}_j(t) \cdot \vec{e}_\xi } {N},\, \xi \in \{ x,y \};$$
- $d\left(\theta_i(t),\theta(t)\right) := \min_{k\in\mathbb{Z}} |\theta_i(t)-\theta(t) + 2k\pi|$ denotes the angle in $[0 , \pi]$ between $\vec{v}_i(t)$ and the average direction of motion.
Now, the time-dependent polarization is defined as $$p(t) := \frac{1}{N} \sum_{i = 1}^{N} d\left(\theta_i(t),\theta(t)\right).$$ We are also interested in the time average of $p$, which is defined as $$P := \left\langle p(t) \right\rangle_t = \left\langle \frac{1}{N} \sum_{i = 1}^{N} d\left(\theta_i(t),\theta(t)\right) \right\rangle_t.$$
### Projection of the pedestrian density on the $\vec{e}_x$ axis {#sect:Sim, projection}
To examine the distribution of our crowd in the direction that corresponds to the desired velocity, we can consider the number of pedestrians that are located in $$S_{\eta}^{\epsilon} := \{ (x,y) \in \Omega : -\frac{B}{2} \leq y \leq \frac{B}{2}, |\eta - x|\leq \epsilon\}.$$ Here, $0 < \epsilon \ll 1$, and $S_{\eta}^{\epsilon}$ denotes a narrow strip parallel to the $\vec{e}_y$ axis, centered at $x$-position $\eta$ and of width $\eta$. We could simply plot the number of people in $S_{\eta}^{\epsilon}$ as a function of $\eta$. Since pedestrians are point masses in our model, this procedure would produce a discontinuous, histogram-like graph. Moreover, the number of discontinuities would highly depend on the value of $\epsilon$. In order to smoothen the results, the individuals’ coordinates have been projected on their corresponding $x$-coordinates. Next, we assign an ‘induced density’ $\rho_{\vec{e}_{x,i}}$ to each individual $i$, which can be considered as a mollified Dirac delta distribution, see Figure \[indDensity\]. It was constructed in such a way that it has support of width $\frac{L}{4}$.\
![The induced density on the $\vec{e}_x$ axis $\rho_{\vec{e}_{x,i}}$ prescribed by an individual as a function of the spatial coordinate $\eta$, relative to its horizontal coordinate $x_{i}$.[]{data-label="indDensity"}](figure11.pdf)
The total density in a point $\eta \in [ - \frac{L}{2} , \frac{L}{2} ]$ is obtained by adding all individual contributions: $$\rho_{\vec{e}_x}^N (\eta) := \sum_{i = 1}^{N} \rho_{\vec{e}_x,i} (\eta),$$ where the induced density $\rho_{\vec{e}_x,i}$ is given by,
$$\rho_{\vec{e}_x,i} (\eta) := \left\{
\begin{array}{ll}
\frac{10}{BL} & \hbox{if $| x_i - \eta | \leq \frac{L}{32}$;} \\
\frac{L}{96 B | x_i - \eta |^2} - \frac{2}{3 BL} & \hbox{if $\frac{L}{32} \leq | x_i - \eta | \leq \frac{L}{8}$;} \\
0 & \hbox{otherwise.}
\end{array}
\right.$$
Note that this density distribution is normalized:
$$\int_{\Omega} \rho_{\vec{e}_x}^N d\Omega = \sum_{i = 1}^{N} B \int_{-\frac{L}{2}}^{\frac{L}{2}} \rho_{\vec{e}_x,i} (x) dx = \sum_{i = 1}^{N} 1 = N.$$
### Morisita index {#sect:def Morisita}
We subdivide the domain $\Omega$ in $\mathcal{M}$ equally sized rectangular boxes of dimensions $S_x \cdot S_y$. Inspired by [@Lega], we define the Morisita index $I_{\mathcal{M}}$ for our system, which is a measure for the degree of dispersion (or of clustering) in our crowd. The index is the product of $\mathcal{M}$ and $\delta_\mathcal{M}$, where $\delta_\mathcal{M}$ is an estimator of Simpson’s measure of diversity; cf. [@Simpson]. This estimator is defined as $$\delta_{\mathcal{M}} := \frac{ \sum_{i = 1}^{\mathcal{M}} n_i (n_i - 1) } { N (N-1) } ,$$ where $n_i$ equals the number of pedestrians in box $i$. This is $\frac12\sum_{i = 1}^{\mathcal{M}} n_i (n_i - 1)$, the number of pairs consisting of two individuals within the same box, divided by $N (N-1)/2$: the total number of pairs in the system. Thus, $\delta_\mathcal{M}$ is an (unbiased) estimator of the probability that two randomly and independently chosen individuals are in the same box.\
Now, assume that we draw two independent samples from a *uniform* distribution on our corridor. The probability that both turn out to be in box $i$ is $(1/\mathcal{M})^2$, since they are independent and all boxes have the same size. Summation over all boxes leads to the probability that both sampled individuals are in the same, yet arbitrary, box: $\sum_{i = 1}^{\mathcal{M}} (1/\mathcal{M})^2 = 1/\mathcal{M}$.\
\
As mentioned above, the Morisita index, introduced in [@Morisita59; @Morisita62], is defined as $$\label{Morisita def}
I_{\mathcal{M}} := \mathcal{M}\, \delta_{\mathcal{M}}.$$ The Morisita index can thus be interpreted as (an estimator of) the probability that two arbitrary pedestrians in our system are in the same box, divided by the probability that two uniformly distributed individuals are in the same box.\
See Figure \[Mor\] for an example that illustrates the use of the Morisita index. The figure also shows that the value of the Morisita index depends on the number of boxes $\mathcal{M}$ and their distribution.
(6.5,0)–(6.5,1)node\[anchor=north east\][$\vec{e}_y$]{}; (6.5,0)–(7.5,0)node\[anchor=north east\][$\vec{e}_x$]{};
(4,2) rectangle (-4,2.5); (4,2)–(-4,2); (4,-2) rectangle (-4,-2.5); (4,-2)–(-4,-2);
(4,2)–(4,-2); (3,2)–(3,-2); (2,2)–(2,-2); (1,2)–(1,-2); (-1,2)–(-1,-2); (-2,2)–(-2,-2); (-3,2)–(-3,-2); (-4,2)–(-4,-2);
(4,-1.04)–(-4,-1.05); (4,1.05)–(-4,1.05);
(0.2,0.3) circle (0.1cm); (0.5,0.5) circle (0.1cm); (0.8,0.2) circle (0.1cm);
(1.2,1.6) circle (0.1cm); (1.5,1.2) circle (0.1cm);
(0,2)node\[anchor=south\][$\frac{B}{2}$]{} –(0,-2)node\[anchor=north\][$-\frac{B}{2}$]{}; (4,0)node\[anchor=west\][$\frac{L}{2}$]{}–(-4,0)node\[anchor=east\][$-\frac{L}{2}$]{};
(-2,0)–(-2,-1)node\[midway, right\][$S_y$]{}; (-2,0)–(-3,0)node\[midway, anchor=south\][$S_x$]{};
### Connection between the Morisita index and entropy {#sect:Morisita Entropy}
We already mentioned the word ‘clustering’. In information theory, this concept refers to the process of ordering items: each *document* should be assigned to a *cluster*. The reader is referred to [@Manning], Chapter 16, for more details. A measure to evaluate the quality of a clustering is the *entropy*. Our aim here is to point out that this entropy is related to the Morisita index, be it that there is no one-to-one correspondence.\
Let $K$ be a fixed integer number (the number of clusters) and define $\Omega=\{\omega_1,\ldots,\omega_K\}$ to be the set of clusters. One should see $\omega_k$ as the set of all documents assigned to cluster $k$. In total there are $N$ documents. We follow [@Manning], pp. 328–329, in defining the entropy $H(\Omega)$ as $$\label{entropy def}
H(\Omega) = -\sum_{k=1}^{K}\frac{|\omega_k|}{N}\log\frac{|\omega_k|}{N}.$$ Note that this functional form is often used for entropy in many other contexts different than information theory.\
To find the relation between entropy and the Morisita index, we order our *individuals* according to the box in which they are situated. We take $K$ to be the number of non-empty boxes. If we give these $K$ boxes indices $k$, then $|\omega_k|$ denotes the number of individuals in (non-empty) box $k$. Note that there is a slight subtlety here. Our boxes are what are called *clusters* in the jargon of information theory (i.e. to which individuals are assigned). These should not be confused with our intuitive interpretation of ‘clusters’: aggregations of individuals. Only in specific cases (dense aggregations or large boxes) such aggregation will be contained in a box.\
![Scatter plot of the entropy against the Morisita index for 10000 random configurations. A linear fit is added to the plot to emphasize the clear correlation between the two concepts.[]{data-label="Morisita Entropy fit"}](figure13.pdf)
We investigate numerically the relation between entropy and the Morisita index. Each individual in a group of size $N=1000$ is assigned randomly to one of in total $\mathcal{M}=64$ boxes. Note that this value of $\mathcal{M}$ corresponds to the reference value in Table \[modpar\], which will also be used in the sequel. Using the obtained configuration we calculate the Morisita index (\[Morisita def\]) and the entropy (\[entropy def\]). This procedure is repeated 10000 times. For both quantities the exact position of an individual does not matter; it is only important to know in which box he is. Therefore we only need to assign a box (randomly). In Figure \[Morisita Entropy fit\] we plot the entropy against the Morisita index for each realization. There is a clear correlation between the two, which we indicate by adding a linear fit of the data.\
\
Although entropy and Morisita are not the same concepts (and do not have the same interpretation), our numerical investigation supports the idea that they are similar in some sense. There is a clear (negative) correlation, which means that *mutatis mutandis* we can deduce the same information from either of the two measures.
Measured quantities: results {#sect: measured quant results}
----------------------------
### Polarization {#sect:results polarization}
To show the kind of information we can deduce from the polarization index, we start by examining the evolution of $p(t)$ for $N=100$. The combined results are shown in Figures \[pol t r\] and \[pol t ar\].\
![The instantaneous polarization $p$ as a function of time. Results in the for $N=100$, for several different values of $\alpha$.[]{data-label="pol t ar"}](figure14.pdf)
![The instantaneous polarization $p$ as a function of time. Results in the for $N=100$, for several different values of $\alpha$.[]{data-label="pol t ar"}](figure15.pdf)
In each of the six cases we identify a relatively short period of time just after $t=0$, during which the polarization decreases rapidly. Afterwards, there is a state in which $p$ fluctuates around a certain average level; this state might be an equilibrium. For each $\alpha$ in the , and for $\alpha=0.0$ in the , this average level is zero (we needed to zoom in in Figure \[pol t r\] to verify that these curves really decay below the level as in Figure \[pol t ar\]). The scenarios $\alpha=1.0$ and $\alpha=0.5$ in the are different in the sense that they do not decay to zero. Moreover, if one would zoom in, one would see that the oscillations in the are less rapid and of smaller amplitude than in the . This holds especially for $\alpha=0.5$ and $\alpha=0.0$. Comparing R to AR requires some extra care, however. It is difficult to compare them in a fair way, because of their intrinsically different nature, and because of the (in)compatibility of the tested values for the interaction radii (cf. Table \[modpar\]).\
\
Let us focus on the initial rapid decay of $p$: this suggests that the initial configuration is not a favourable state for the system to be in. An immediate relaxation takes place, implying spreading of the individuals in all possible directions until a more preferable situation is reached. Figures \[pos\_r\_00\]–\[pos\_ar\_05\] support this statement. Most clearly in Figures \[pos\_ar\_00\] and \[pos\_ar\_05\] the system evolved away from the initial configuration. The same statement is true for Figures \[pos\_r\_00\], \[pos\_r\_05\] and \[pos\_r\_10\], although this is less evident. Once we realize however that the particles move in what seem to be *six* horizontal rows, we must indeed conclude that the particles have deviated from the initial situation in which there were *ten* rows. Figure \[pos\_ar\_10\] is somewhat different, even though we recognize relaxation, as the initial occupation was only about one tenth of the corridor length. Spreading in a direction parallel to the mean direction of motion does not explain the peak in $p$ just after $t=0$ however. This is because fluctuations in the magnitude of the velocity (that is: the speed) do not affect the polarization if all individuals move in the same direction. The peak shows that there must have been some vertical displacement too.\
\
We remark that in the $p$ does not necessarily decay to zero, while it does (or at least: *seems to do* up to fluctuations and noise) in the . Regarding Figures \[pos\_r\_00\]–\[pos\_ar\_10\], we can distinguish the graphs for $\alpha=1.0$ and $\alpha=0.5$ in the from the four others, since they do not seem to possess the degree of order and structure that the other graphs do have. Strikingly, these are exactly the cases in Figure \[pol t ar\] where $p$ does not tend to zero. From this we conclude that there is a strong relation between the polarization tending to zero, and the preservation or favouring of patterns and organization in the system. The cases in which $p$ oscillates around a non-zero average in the long run, are exactly those in which the initial ordered configuration has disappeared after some time. Moreover, the fact that the polarization remains positive is an indicator that the configurations in Figures \[pos\_ar\_00\] and \[pos\_ar\_05\] are not stable.\
![The time average of the polarization $P$ as a function of the number of pedestrians $N$. Results in the , for several different values of $\alpha$.[]{data-label="npar"}](figure16.pdf)
![The time average of the polarization $P$ as a function of the number of pedestrians $N$. Results in the , for several different values of $\alpha$.[]{data-label="npar"}](figure17.pdf)
In Figures \[npr\] and \[npar\] the time average polarization index $P$ is shown. For each $N$, we recognize in $P$ the same ordering with respect to $\alpha$ as the ordering we have seen before in $p$ (just for $N=100$, cf. Figures \[pol t r\] and \[pol t ar\]). The ordering between $\alpha=0.5$ and $\alpha=0.0$ () is even much clearer when considering the time average.\
Note that, comparing R and AR, the absolute differences in $P$ are much smaller than those in $p$. This is mainly because the $p$-curves in the have smaller range but larger domain where they are non-negligible.\
\
Up to now we only used initial conditions in which the individuals are positioned on a grid in only a part of the corridor. We wish to explore the influence of these initial conditions, by allowing randomness. More specifically, we sample the initial position of each individual from a random distribution on the whole corridor. The corresponding time averages of the polarization are given in Table \[pol random IC\], represented by the mean and standard deviation of 10 random runs. Both the and the are considered.\
---------------------------- -------- --------
Mean 0.0387 0.0147
Standard deviation 0.0019 0.0018
Regular initial conditions 0.023 0.037
Ratio 0.5941 2.5170
---------------------------- -------- --------
: Time average polarization after time $t = 100$s, for $N=60$ individuals and $\alpha=1.0$. The mean and standard deviation of 10 independent simulation runs are given. In each run the initial conditions were drawn from a random distribution, such that the initial positions are random over the whole corridor. We compare the outcome to the previously used initial conditions (i.e. on a lattice in a section of the corridor). The ratio: polarization for regular initial conditions divided by the average over 10 random runs is also given. Calculations were performed both for the and the .\
[]{data-label="pol random IC"}
Unfortunately, these results do not provide an unambiguous conclusion about the effect of random initial conditions. In the the mean polarization is bigger for random initial conditions than for regular ones. In the this is the other way around. An issue in interpreting the ratios in the bottom line of the table is that we are dividing relatively small numbers. However, this is something that needs to be stressed: the polarization remains relatively small, also for random initial conditions. We will see later (see e.g. Table \[Morisita random IC\]) that for the Morisita index the presented ratio does make sense, and that we can use them to draw conclusions about the effect of randomness.
### Projection of the pedestrian density on the $\vec{e}_x$ axis {#sec: result projection}
Figures \[vdr\_1\_0\] () and \[vdar\] () show the projected mollified density that was introduced in Section \[sect:Sim, projection\]. The graphs show that in the isotropic case ($\alpha = 0.0$) the density profile is completely symmetric around the center of mass. Note that this is only the case if the initial conditions are symmetric. For $\alpha=1.0$ and $\alpha=0.5$ in the symmetry is no longer present. This is perfectly sane, since anisotropy (i.e. $\alpha>0$) was introduced to incorporate asymmetric interactions in our model: a pedestrian is more influenced by an other individual in front of him than by one behind him. The graphs corresponding to $\alpha=1.0$ and $\alpha=0.5$ in the are not given. They possess the same oscillatory behaviour as Figure \[vdr\_1\_0\], but without being symmetric around the center of mass. They do not provide any further information or insight, and thus are omitted.\
![The projection of the pedestrian density on the $\vec{e}_x$ axis $\rho_{e_x}$ as a function of the spatial coordinate $\eta$, relative to the center of mass $\eta_{cm}$. Results in the for several different values of $\alpha$ at time $t = 100$s.[]{data-label="vdar"}](figure18.pdf)
![The projection of the pedestrian density on the $\vec{e}_x$ axis $\rho_{e_x}$ as a function of the spatial coordinate $\eta$, relative to the center of mass $\eta_{cm}$. Results in the for several different values of $\alpha$ at time $t = 100$s.[]{data-label="vdar"}](figure19.pdf)
The highly oscillatory behaviour in Figure \[vdr\_1\_0\] and the omitted graphs reflects the ordered, lattice-like, structures we already observed in Figures \[pos\_r\_00\]–\[pos\_r\_10\]. Following this reasoning, one would expect the same kind of oscillations also in the for $\alpha=0.0$ (cf. the ordered pattern in Figure \[pos\_ar\_10\]). The reason for not seeing this in Figure \[vdar\] is simple: The average distance in $\vec{e}_x$ direction is smaller than the width of the support of our induced density. The total density therefore smoothens out the periodic structure of the individuals’ positions. What we do see is the fact that in this situation the individuals do not occupy the whole corridor, but are confined to a certain section of it.\
\
What we observe is that including attraction in the interaction has some regularizing effect. For each of the three choices for $\alpha$, there is a core of high density around the center of mass, which forms the heart of our crowd. This core is present due to the initial condition that was concentrated on a part of the corridor. Without attraction the repulsive interactions would drive it apart. However, attraction is not able to completely diminish the effect of repulsion. Especially when $\alpha$ increases, the projected density of the core decreases, while there is a tail of mass just behind it (that is, in the graph on the left of the center). This effect can be explained by the fact that individuals in the anisotropic case are driven backwards if they are too close together (like in the high density core). There is no (sufficient) compensation driving the individual forward as this effect is decreased/switched off by increasing $\alpha$.\
Apparently the mechanism that drives individuals to the back has more effect as $\alpha$ increases. In Figure \[vdar\] we namely see that due the periodic boundaries the tail already reaches again the front of the core group. We expect that the Morisita index (see Section \[sect:results Morisita\]) will provide extra support for our observations here; this index should be high when pedestrians concentrate more and more in a particular region.
### Morisita index {#sect:results Morisita}
Figures \[nir\] and \[niar\] show the Morisita index as a function of the number of individuals $N$. The corridor is subdivided in $\mathcal{M}=64$ boxes.\
![The Morisita index $I$ as a function of the number of pedestrians $N$. Results in the , for several different values of $\alpha$ at time $t = 100$s.[]{data-label="niar"}](figure20.pdf)
![The Morisita index $I$ as a function of the number of pedestrians $N$. Results in the , for several different values of $\alpha$ at time $t = 100$s.[]{data-label="niar"}](figure21.pdf)
The increase in the Morisita index as $N$ increases (in the ) can be explained. For repulsive interactions, the individuals have a tendency to move as far apart as possible (if possible until they are a distance $R^R_r$ apart). This was illustrated by Figures \[pos\_r\_00\]–\[pos\_r\_10\]. As $N$ increases, they are however packed together more and more, thus leading to an increase in Morisita index. If we assume (to obtain an approximate result) that the individuals are distributed uniformly, then $n_i=N/\mathcal{M}$ for all $i$. Taking into consideration the ordered distribution in Figures \[pos\_r\_00\]–\[pos\_r\_10\], this assumption is justifiable (see also below). It follows that $I_{\mathcal{M}}=(N-\mathcal{M})/(N-1)=1-(\mathcal{M}-1)/(N-1)\rightarrow 1$ as $N\rightarrow\infty$ for fixed $\mathcal{M}$. Moreover, this implies that $I_{\mathcal{M}}$ tends to its limit value from below. This matches with the increase of the curves in the (Figure \[nir\]), and we conjecture that the Morisita index will tend to 1 for $N$ increasing beyond $N=100$.\
\
There is actually more support for the assumption that the individuals in the are distributed uniformly. This support is provided by Figure \[size i r\], in which (for $N=100$) we show the Morisita index as a function of the box size. In [@Morisita59] an overview is given of the information that can be derived from such graph. Note that the Morisita index equals 1 if the number of boxes $\mathcal{M}=1$. This corresponds to the largest possible box size. The value 1 is also indicated in the graph by the dashed grey line. In Figure \[size i ar\] we see that for each of the three values of $\alpha$ the Morisita index approaches the value 1 from below as the box size increases. This particular trend in the graph corresponds to a uniform distribution of particles (cf. Figure 1 in [@Morisita59]).
![The Morisita index $I$ as a function of the box size $S_x\cdot S_y$. Results in the , for $N=100$, and for several different values of $\alpha$ at time $t = 100$s. The grey dashed line indicates the value 1. Note that in this figure we thus deviate from the (fixed) values for $\mathcal{M}$ and $S_x\cdot S_y$ given in Table \[modpar\].[]{data-label="size i r"}](figure22.pdf)
In the (Figure \[niar\]), we observe that the Morisita index is *not* monotonic in $\alpha$ for $N$ smaller than $N\approx 35$ (this is where the graphs for $\alpha=1.0$ and $\alpha=0.5$ intersect). This point of intersection is hard to explain and requires further investigation. Moreover, it is hard to draw any conclusion about the precise dependence of the Morisita index on $N$ or $\alpha$ in this case. One could argue that increasing $\alpha$ corresponds to lower Morisita index. At least this is the case if one compares $\alpha=0.0$ to $\alpha>0$.\
\
We now consider the influence of varying the box size in the . The trend, as shown in Figure \[size i ar\], is different from the (Figure \[size i r\]). The most importance difference is that the value 1 is approached *from above* for increasing box size. This case is described in Figure 1 of [@Morisita59] as ‘contagious distribution’ with ‘clump(s)’. In our terminology this would correspond to clustering. We already observed that in the the individuals do not occupy the whole corridor (cf. Section \[sec: result projection\]).\
![The Morisita index $I$ as a function of the box size $S_x\cdot S_y$. Results in the , for $N=100$, and for several different values of $\alpha$ at time $t = 100$s. The grey dashed line indicates the value 1. In this figure we also deviate from the values for $\mathcal{M}$ and $S_x\cdot S_y$ given in Table \[modpar\].[]{data-label="size i ar"}](figure23.pdf)
The shape of the graph and the way of approximating 1 from above, is linked in [@Morisita59] to the size of clumps (clusters) and the distribution of particles therein. We will not go into detail here, mostly since one can debate about the question to which class in Figure 1 of [@Morisita59] the graphs in Figure \[size i ar\] should be assigned.\
---------------------------- -------- --------
Mean 0.1555 0.6979
Standard deviation 0.0639 0.1079
Regular initial conditions 0.3616 1.0847
Ratio 2.3256 1.5544
---------------------------- -------- --------
: Morisita index after time $t = 100$s, for $N=60$ individuals and $\alpha=1.0$. The mean and standard deviation of 10 independent simulation runs are given. In each run the initial conditions were drawn from a random distribution, such that the initial positions are random over the whole corridor. We compare the outcome to the previously used initial conditions (i.e. on a lattice in a section of the corridor). The ratio: Morisita index for regular initial conditions divided by the average over 10 random runs is also given. Calculations were performed both for the and the .\
[]{data-label="Morisita random IC"}
To investigate the influence of the initial conditions, we also calculate the Morisita index corresponding to random initial conditions. More details are given in Table \[Morisita random IC\]. Both in the and in the the mean Morisita index over 10 random runs is significantly lower than the Morisita index for regular initial conditions. The difference is even much larger than the standard deviation.\
However, there is consistency, in the sense that in both cases (R and AR) the values are lower.
Conclusions and outlook {#sec:conclusion}
=======================
The behaviour of a crowd of pedestrians inside a corridor, in which the individuals interact via an anisotropic way, can be distinguished clearly from the case in which pedestrians interact in a completely isotropic way. In particular, we observe the following differences compared to the isotropic case:
1. the polarization index increases with increasing anisotropy (i.e. increasing $\alpha$);
2. the projected density along the $\vec{e}_x$ axis shows a symmetric profile around the center of mass for the isotropic case. However, increasing anisotropy implies loss of symmetry;
3. the Morisita index, as a measure of clustering, depends clearly on the anisotropy. It increases (with increasing anisotropy) in the and, roughly speaking, decreases in the .
4. In case of repulsive interactions, the crowd tends to fill the whole corridor. If attraction is included, the group stays compact. Increasing $\alpha$ however seems to diminish this kind of *social cohesion* as individuals do not look behind.
As a result of our study, many new questions arose. Future research should be concentrated on the following three directions:
- Most obviously: what is the effect of a further increase of the number of pedestrians? Do the observed relations between the measured quantities and the number of pedestrians still hold? How do the observed limiting values of the polarization for large $t$ depend on $N$? What about the limit $N\rightarrow\infty$ in this case?\
Can we gain more insight in the strange issues of the (cf. Figure \[niar\]: intersecting curves for $\alpha=1.0$ and $\alpha=0.5$) by further measurements of the Morisita index for increasing $N$? E.g. can we extrapolate information for $N>100$ back into the interval $[0,100]$? Some preliminary comments in this direction are given in \[Appendix\].\
If $N$ increases, the natural thing to do is to consider the discrete-to-continuum limit (i.e. construct educated procedures to derive mean-field limit equations). Does such limit exist, can we derive it, and can we compare the effect of anisotropy in the limit to the observations of the current work?
- How much does the large time behaviour of the crowd depend on the initial conditions? The initial distribution of pedestrians in this paper is not a realistic situation. People starting to enter a corridor are in real life never distributed in a crystalline structure. However, for an escape situation (for example in the case of fire) it seems reasonable to assume that a group of people starts, being clustered, at one side of a corridor. Therefore, as an extension of this research, we propose to use as initial distribution a more realistic configuration in which people are placed at one side of the corridor, with their positions (slightly) perturbed from the grid points. Averaging over a large collection of such perturbed initial distributions, will lead to effective results. Are these averaged results comparable to the ones presented in this paper? In other words: is averaging the results basically the same as removing the fluctuations from the initial conditions?\
In the paper we have included some preliminary results (in the ends of Sections \[sect:results polarization\] and \[sect:results Morisita\]) in this direction. There we took the other extreme: random initial conditions over the whole corridor.
- What happens if we try to make our model more realistic: e.g. change the shape of the domain $\Omega$, or allow variation in the direction and magnitude of individuals’ desired velocity? Including more sophisticated active parts in the boundary (doors) or impermeable objects within the domain, automatically leads to questions about the efficiency of the flow (such issues are also addressed e.g. in [@Guo11; @Guy]). Which geometry leads to the fastest evacuation? First steps in this direction have been made in [@EversMuntean].
The issues addressed in this paper show that anisotropy related to perception has nontrivial effects on the global dynamics of a crowd. Certainly, these effects cannot be neglected. More work, both numerically and analytically, is needed to extend and formalize our results.
The authors thank C. Storm, F. Toschi, F. van de Ven, H. Wyss (all with TU Eindhoven) and R. Fetecău (Simon Fraser Univ. Canada) for a series of fruitful discussions on the dynamics of self-propelled particles with anisotropic motion potentials. They are also grateful to J. Lega (Univ. of Arizona USA) for sharing her thoughts. Moreover, they thank the anonymous referees for their comments and suggestions for improvements. JE kindly acknowledges the financial support of the Netherlands Organisation for Scientific Research (NWO), Graduate Programme 2010.
Towards a bigger number of particles {#Appendix}
====================================
Let us start by saying that simulation of large numbers of individuals is beyond the scope of this paper. Our current implementation is inadequate for simulating system sizes one is used to in molecular dynamics. In our perspective this paper aims primarily at getting insight about what features to expect. A second stage (and follow-up paper) is to optimize the implementation and increase the system size.\
\
Looking ahead, we provide here some preliminary results for $N=1000$ in the . In Figures \[p1000\] and \[i 1000\] we show the time-averaged polarization and the Morisita index, respectively, as a function of $N$. Compared to Figures \[npar\] and \[niar\], the graphs have been continued by incorporation of the values at $N=1000$. Note the logarithmic scale of the horizontal axes.\
![The Morisita index $I$ as a function of the number of pedestrians $N$. Results in the , for several different values of $\alpha$ at time $t = 100$s. This is an extension of Figure \[niar\] including the value for $N=1000$ with logarithmic scaling on the horizontal axis.[]{data-label="i 1000"}](figure24.pdf)
![The Morisita index $I$ as a function of the number of pedestrians $N$. Results in the , for several different values of $\alpha$ at time $t = 100$s. This is an extension of Figure \[niar\] including the value for $N=1000$ with logarithmic scaling on the horizontal axis.[]{data-label="i 1000"}](figure25.pdf)
Note moreover that we have extended Figures \[npar\] and \[niar\] by only one data point each. The linear interpolation between the values at $N=100$ and $N=1000$ is therefore probably not very meaningful. What we are interested in, is the general trend.\
\
For the polarization, the ordering as a function of $\alpha$ remains as we observed it. What requires more investigation is the increasing trend: that is, increasing $P$ for increasing $N$. In Figure \[npar\] the graphs seem to stop growing as $N$ goes towards $100$. An issue here might be that the time interval of 100 s is simply too short for larger $N$.\
\
In the Morisita plots we see a downward trend. We see that the curve for $\alpha=1.0$ remains beneath the other two, also for $N=1000$. We can thus regard the absence of ordering of the curves for $N$ smaller than $N\approx 35$ as an exception.\
A remark needs to be made about the fact that, at $N=1000$, the Morisita index for $\alpha=0.0$ is smaller than for $\alpha=0.5$. The values are $I= 0.9515$ for $\alpha=1.0$, $I=0.9583$ for $\alpha=0.5$ and $I=0.9566$ for $\alpha=0.0$. The ordering of the curves therefore seems to be lost here also, be it that the difference is small compared to the magnitude of $I$.
References {#references .unnumbered}
==========
[99]{} U. Ascher and L. R. Petzold, Computer Methods for Ordinary Differential Equations and Differential-Algebraic Equations, Philadephia, SIAM, 1988.
N. Bellomo, B. Piccoli, A. Tosin, Modeling crowd dynamics from a complex system viewpoint, , 22(supp 02), 1230004, 2012.
F. Bornemann, Scientific Computing with Ordinary Differential Equations, Springer Verlag, 2002.
S. P. Carroll, J. S. Owen, M. F. M. Hussein, Modelling crowd-bridge dynamic interaction with a discretely defined crowd, , 331: 2685–2709, 2012.
E. N. M. Cirillo and A. Muntean, Can cooperation slow down emergency evacuations?, , 340(9): 625–628, 2012.
E. Cristiani, B. Piccoli, A. Tosin, Multiscale modeling of granular flows with application to crowd dynamics, , 9(1): 155–182, 2011.
C. M. Dafermos, Hyperbolic [C]{}onservation [L]{}aws in [C]{}ontinuum [P]{}hysics, Springer Verlag, Berlin, 2010.
J. H. M. Evers and A. Muntean, Modeling micro-macro pedestrian counterflow in heterogeneous domains, , 14(1):27–37, 2011.
R.-Y. Guo, H.-J. Huang, Route choice in pedestrian evacuation: formulated using a potential field, , P04018, 2011.
R.-Y. Guo, S. C. Wong, Y.-H. Xia, H.-J. Huang, W. H. K. Lam, K. Choi, Empirical evidence for the look-ahead behavior of pedestrians in bi-directional flows, , 29(6), 068901, 2012.
S. J. Guy, S. Curtis, M. C. Lin, D. Manocha, Least-effort trajectories lead to emergent crowd behaviors, , 85, 016110, 2012.
D. Helbing and P. Molnar, Social force model for pedestrian dynamics, , 51(5):4282–4286, 1995.
D. Helbing, I. Farkas, T. Vicsek, Simulating dynamical features of escape panic, , 407: 487–490, 2000.
J. Lega, Collective behaviors in two-dimensional systems of interacting particles, , 10: 1231–1231, 2011.
C. D. Manning, P. Raghavan, H. Schütze, Introduction to [I]{}nformation [R]{}etrieval, Cambridge University Press, 2008.
S. Martens, I. M. Sokolov, L. Schimansky-Geier, Impact of inertia on biased Brownian transport in confined geometries, , 136(11): 111102, 2012.
M. Morisita, Measuring of the dispersion of individuals and analysis of the distributional patterns, , 2(4): 215–235, 1959.
M. Morisita, $I_{\delta}$-index, a measure of dispersion of individuals, , 4: 1–7, 1962.
M. Moussaïd, D. Helbing, G. Theraulaz, How simple rules determine pedestrian behavior and crowd disasters, , 108(17): 6884–6888, 2011.
F. Peruani and L. G. Morel, Self-propelled particles with fluctuating speed and direction of motion in two dimensions, , 99(1): 010602–010606, 2007.
P. Romanczuk, M. B[ä]{}r, W. Ebeling, B. Lindner, L. Schimansky-Geier, Active [B]{}rownian particles: [F]{}rom individual to collective stochastic dynamics, , 202(1): 1–162, 2012.
A. Schadschneider, W. Klingsch, H. Kluepfel, T. Kretz, C. Rogsch, A. Seyfried, Evacuation Dynamics: Empirical Results, Modeling and Applications, , Springer Verlag, Berlin, 2009.
A. Schadschneider, D. Chowdhury, K. Nishinari, Stochastic Transport in Complex Systems, Elsevier, 2011.
E. H. Simpson, Measurement of diversity, , 163: 688, 1949.
C. Stott and S. Reicher, Crowd action as intergroup process: introducing the police perspective, , 28: 509–529, 1998.
F. Theil, A proof of crystallization in two dimensions, , 262(1): 209–236, 2006.
[^1]: Note that this is a modelling choice, and that one can view crowds from different perspectives, e.g. micro-, meso- and macroscopic, or a suitable combination of these. See [@Bellomo] for a critical review, or the broader overview in [@Schadschneider2011].
[^2]: We will see later that the type of repulsion in $\vec{F}_i^{wall}$ for small $d$ is the same as in the interactions between individuals (cf. (\[Fsoc\])–(\[U\_attr\_rep\])), be it with different parameters.
[^3]: Regarding the values in the table, we remark that one might be hesitant actually to call a domain of width $B=10$ m a *corridor*. In our setting, the word *corridor* refers to the ratio between length and width, rather than to the absolute values of $B$ and $L$. Note that this ratio of the ‘real’ corridor is in fact even bigger than $L/B$, namely $\tilde{L}/B$, cf. the explanation in the beginning of Section \[sec:level1\].
| {
"pile_set_name": "ArXiv"
} |
---
address:
- '$^1$Walther Meissner Institut, Bayerische Akademie der Wissenschaften, 85748 Garching, Germany'
- |
$^2$Theoretische Physik III, Elektronische Korrelationen und Magnetismus,\
Institut für Physik, Universität Augsburg, 86135 Augsburg, Germany
- '$^3$Department of Physics, University of Waterloo, Waterloo, ON, Canada, N2L 3G1'
-
author:
- 'F. Venturini$^1$, U. Michelucci$^2$, T.P. Devereaux$^3$, and A.P. Kampf$^2$'
title: Collective Spin Fluctuation Mode and Raman Scattering in Superconducting Cuprates
---
Electronic Raman scattering has proven to be a useful tool in exploring the superconducting state in the cuprate materials. The possibility of probing selectively electronic excitations in different regions of the Brillouin zone (BZ) by the choice of polarization geometries has allowed to explore the superconducting gap anisotropy. The successful explanation of the Raman data in $B_{1g}$ and $B_{2g}$ scattering geometries [@pol; @tpd3] has provided one piece of evidence for the by now widely accepted $d_{x^2-y^2}$ pairing symmetry in hole-doped cuprate superconductors [@Annett]. In the context of impurity effects as a testing ground for unconventional superconductivity, the observed $\omega^3$ to $\omega$ crossover in the low frequency $B_{1g}$ Raman response fits consistently with the power law crossovers at low temperatures in the NMR relaxation rate and in the magnetic penetration depth. An even quantitatively consistent picture of electronic Raman scattering and infrared conductivity was achieved when the T-matrix approximation in the “dirty” d-wave scenario is extended to include a spatial extension of the impurity potential [@tpd1].
However up to now the discrepancy between Raman data in $A_{1g}$ and $B_{1g}, B_{2g}$ geometries has remained unresolved [@multi; @Chub]. Previous results for the A$_{1g}$ scattering geometry were found to be very sensitive to changes in the Raman vertex function $\gamma({\bf k})$ [@multi] making a comprehensive explanation difficult for the experimental data in different cuprate materials.
In this paper we calculate the Raman response of a $d_{x^2-y^2}$ superconductor including the contribution from a collective spin fluctuation (SF) mode which is identified with the $(\pi,\pi,\pi)$-resonance (in short $\pi$-resonance) near $\omega_R
\approx 41$ meV observed by inelastic neutron scattering (INS) on bilayer cuprates [@fong; @mook]. Our results suggest that the $A_{1g}$ peak position is largely controlled by the strength and frequency of the $\pi$-resonance mode which on the other hand does not affect the Raman response in the $B_{1g}$ and $B_{2g}$ channels. The inclusion of the collective SF mode allows for a simultaneous fit of the Raman data in all channels in optimally doped materials. Furthermore we find that the inclusion of the SF term significantly reduces the sensitivity to the special choice of the underlying tight-binding band structure, i.e. the sensitivity to the choice of the Raman vertex in $A_{1g}$ symmetry, resolving the previously encountered problems in the symmetry analysis of the light scattering amplitude [@multi; @tpd3].
On the basis of the observation of the collective SF mode in Y-123 and Bi-2212, we consider a bilayer model represented by a tight-binding band structure $$\epsilon_{\bf k} = -2 t (\cos(k_x)+\cos(k_y))+4t'\cos(k_x) \cos(k_y)
- t_\perp({\bf k}),$$ with an inter-plane hopping given by [@bulut] $$t_\perp({\bf k}) = 2 t_\perp \cos (k_z) (\cos(k_x)-\cos(k_y))^2
\label{tperp}$$ where $k_z$ is $0$ or $\pi$ for the bonding or anti-bonding bands of the bilayer, respectively.
The spin susceptibility ($\chi_s$) is modeled by extending the weak coupling form of a BCS superconductor in a $d_{x^2-y^2}$ pairing state to include antiferromagnetic spin fluctuations by a RPA form with an effective interaction $\bar U$; i.e. $\chi_s({\bf q},i\omega)=\chi^0({\bf q},i\omega)/(1-\bar U\chi^0
({\bf q},i\omega))$ where [@bulut] $$\chi^0({\bf q},i\omega)={1\over{\beta}}\hbox{Tr} \sum_{{\bf k},i\omega'}
\hat G({\bf k},i\omega') \hat G({\bf k+q},i\omega'+i\omega).$$ Tr denotes the trace and $\beta=T^{-1}$. $\hat G({\bf k},i\omega)$ is the BCS Green’s function in Nambu space $$\hat G({\bf k},i\omega)=
\frac{i\omega \hat\tau_0 + \xi_{\bf k} \hat\tau_3 + \Delta_{\bf k}
\hat\tau_1} {(i\omega)^2 -\xi_{\bf k}^2 - \Delta_{\bf k}^2}
\label{bcsg}$$ with $\hat\tau_i$ (i=1,2,3) the Pauli matrices, $\hat\tau_0$ the $2\times 2$ unit matrix, $\xi_{\bf k}=\epsilon_{\bf k}-\mu$ and $\Delta_{\bf k}=\Delta_0[\cos(k_x)-\cos(k_y)]/2$. This form of the spin susceptibility contains a strong magnetic resonance peak at ${\bf q}={\bf Q}\equiv (\pi,\pi,\pi)$ which was proposed [@bulut] to explain the INS resonance at energies near 41 meV in Y-123 [@mook] and Bi-2212 [@fong]. Other forms for the spin susceptibility can be straightforwardly used within our model. However, the results are mainly determined by the collective mode at ${\bf Q}$. Therefore we take the bilayer susceptibility for a representative calculation.
The intensity of scattered light $I(\omega)$ in Raman experiments is proportional to the imaginary part of the response function for the effective density operator $$\tilde \rho_{\bf q}=\sum_{{\bf k},\sigma} \hat\gamma ({\bf k})
c^\dag_{\sigma,{\bf k+q}} c_{\sigma,{\bf k}}
\label{rho}$$ in the long wavelength limit ${\bf q\rightarrow 0}$. Specifically $$\begin{aligned}
I(\Omega)&\propto &\left(1+n(\Omega)\right)\hbox{Im} \chi(\Omega+i0^+)
\nonumber \\
\chi(i\Omega)&=&\int_0^{1/T}{\rm d}\tau\, e^{-i\Omega\tau}\,\langle T_\tau [
{\tilde\rho}(\tau),{\tilde\rho}(0)]\rangle ,\end{aligned}$$ with the Bose function $n(\omega)$ and the time ordering operator $T_\tau$.
The bare Raman vertex $\hat\gamma({\bf k})=\tau_{3}\gamma({\bf k})$ in different scattering geometries are classified according to the elements of the $D^{4h}$ point group. For the limiting case of vanishingly small scattered ($\omega_S$) and incident ($\omega_I$) photon energies, it can be represented in the effective-mass approximation (EMA) $$\gamma({\bf k}) = \sum_{\alpha, \beta} e_\alpha^I \frac{\partial^2
\epsilon_{\bf k}}{\partial k_\alpha \partial k_\beta} e_\beta^S
\label{effmass}$$ where ${\bf e}^I$ and ${\bf e}^S$ are the unit vectors for in-plane polarizations (i.e. $\alpha,\beta\in \{x,y\}$) of the incoming and the scattered light, respectively. Using Eq. (\[effmass\]) and the bilayer tight-binding dispersion Eq. \[tperp\] we obtain $$\begin{aligned}
\gamma_{\bf k}^{B_{1g}}&=&2t\gamma_{\bf k}^d\left(1+{4t_\perp\over t}\cos(k_z)
[\cos(k_x)+\cos(k_y)]\right) \\
\gamma^{A_{1g}}_{\bf k}&=&2 t\gamma_{\bf k}^s-2 t_\perp \cos(k_z)
(\cos(2 k_x)+ \cos(2 k_y))\nonumber \\
&&-4\cos(k_x) \cos(k_y)[t'+2 t_\perp \cos(k_z)] \end{aligned}$$ where $\gamma_{\bf k}^{d,s}=(\cos(k_x) \mp \cos(k_y))/2$. However, the EMA has a questionable region of validity for all Raman measurements[@multi] on the cuprates since the incoming photons have energy $\sim 2$eV, which is on the order of the bandwidth and the inter-band excitations according to local density calculations[@lda]. EMA based arguments in previous works about relative Raman intensities for different channels are therefore questionable [@Chub]. We hence consider other forms for the vertices as well which obey the proper symmetry transformations. For the $A_{1g}$ geometry some symmetry compatible choices are $\gamma({\bf k})=\cos(k_x)+
\cos(k_y)$ and $\gamma({\bf k})=\cos(k_x)\cos(k_y)$. These basis functions assign weight to different regions of the BZ and this is the reason why previous results for the $A_{1g}$ response were particularly sensitive to the specific choice of the bare Raman coupling vertex.
The spin fluctuations lead to an additional contribution for the Raman response via a 2-magnon like process as shown diagrammatically in Fig. \[diagr\] [@kampf]. Here, the SF propagator is incorporated in its RPA form for the bilayer (as described above) by the ladder diagram series with an effective on-site Hubbard interaction ${\bar U}$.
We therefore write the Raman response function at finite temperature as the sum of a pair-breaking (PB) and a SF contribution $$\chi_{\gamma\gamma}({\bf q},i\omega)=\chi^{PB}_{\gamma\gamma}({\bf q},i\omega)+
\chi^{SF}_{\gamma\gamma}({\bf q},i\omega)$$ with the Raman vertex specifying the scattering geometry. In the limit ${\bf q}
\rightarrow{\bf 0}$ the diagram for the SF contribution translates into $$\begin{aligned}
\chi^{SF}_{\gamma\gamma}(i\Omega)&=&\displaystyle\frac{1}{\beta}\sum_{
{\bf q}',i\omega}\medskip V^\gamma({\bf q}',i\Omega,i\omega)\chi_s(-{\bf
q'},-i\omega) \nonumber \\
&&\times \chi_s({\bf q}',i\omega+i\Omega)V^\gamma({\bf q}',-i\Omega,-i\omega).
\label{chisf}\end{aligned}$$ The vertex function $V^\gamma({\bf q}',i\Omega,i
\omega)$ includes the bare Raman vertex and is evaluated as $$\begin{aligned}
V^\gamma({\bf q}',i\Omega,i\omega)=\hbox{Tr}\bigg\{\displaystyle\frac{1}{\beta
}\sum_{{\bf k},i\omega'}\hat \gamma({\bf k})
\hat G({\bf k},i\omega'+i\Omega) \hat \tau_0 \bar U
\nonumber\\
\hat G({\bf k+q}',i\omega'+i\Omega+i\omega) \tau_0 \bar U
\hat G({\bf k},i\omega')\bigg\} .
\label{vertex}\end{aligned}$$ $i\Omega$, $i\omega$ denote bosonic and $i\omega'$ fermionic Matsubara frequencies. Similarly, $\chi^{PB}_{\gamma\gamma}$ is evaluated as $$\chi^{PB}_{\gamma\gamma}(i\Omega)={\rm{Tr}\over{\beta}}
\sum_{{\bf k},i\omega'}{\hat
\gamma}({\bf k})\hat G({\bf k},i\omega'){\hat\gamma}({\bf k})
\hat G({\bf k},i\omega'+i\Omega) .$$
The total Raman response is calculated in the gauge invariant form which results from taking into account the long wavelength fluctuations of the order parameter to guarantee local charge conservation [@tpd3; @note]. The total Raman susceptibility thus follows as $$\chi(i\Omega)=\chi_{\gamma \gamma}(i\Omega)-
\frac{\chi^2_{\gamma 1}(i\Omega)}{\chi_{11}(i\Omega)}
\label{screening}$$ where $\chi_{1\gamma}$ and $\chi_{11}$ are obtained by the replacement $\gamma({\bf k})\rightarrow 1$ in one or both bare Raman vertices in the vertex function Eq. (\[vertex\]). The analytical continuation to the real axis is performed using Padé approximants[@pade].
The band structure parameters are chosen for all the numerical calculations to be applicable to optimally doped systems: $\langle n\rangle =0.85$, $t'/t=0.45$, $t_\perp/t=0.1$ [@multi] while the gap has been chosen as $\Delta_0/t=0.25$. We have evaluated the Raman response at a temperature $T/t=0.08$.
Let’s first consider $\chi_{SF}$ alone. In Fig. \[before\] we plot $\chi''_{SF}$ for the $A_{1g}$ channel versus frequency. $\chi_{SF}$ is a convolution of two spin susceptibilities (Eq. (\[chisf\])) and is thus peaked at an energy near twice the magnetic resonance energy. An important point is that in the $B_{1g}$ and $B_{2g}$ geometries the SF term introduces vanishingly small corrections to the total response, rendering the presence of the SF term important only in the A$_{1g}$ geometry. This is due to the sharpness in momentum space of the resonance peak at ${\bf Q}$ in the SF propagator. In fact, if the transfer is taken only at ${\bf Q}$, both the $B_{1g}$ and $B_{2g}$ contributions to $\chi_{SF}$ vanish identically, as can been seen from Eq. \[vertex\]. Therefore for these channels the response is given by the PB term alone.
In Fig. \[after3\] we illustrate the frequency dependence of $\chi''(\Omega)$ in the A$_{1g}$ geometry for three different values of the effective interaction with both the PB and SF contributions included. The shape of the Raman response is modified varying $\bar U/t$ and in particular the position of the resonance is shifted towards higher energies for increasing $\bar U$. With the inclusion of the SF term we now obtain a peak slightly above $\Delta_0$ as observed experimentally in Y-123 and Bi-2212.
A comment is in order on the relative magnitude of the SF and the PB term. Comparing Figs. \[before\] and \[after3\] it is clear that the SF term is much smaller than the PB term, but the effect of this new term is nevertheless visible since the backflow (the second term in Eq. (\[screening\])) mixes in a non trivial way the two contributions. Since the SF term varies as $\bar U^{4}$ in our model it starts to dominate for larger $\bar U$, leading to a shift in spectral weight out towards $2\omega_{R}$, which also changes with $\bar U$.
The experimental position of the peak is not sample dependent in the $A_{1g}$ geometry as already mentioned, and is almost the same in different cuprates. On the other hand the theoretical description with the PB term alone is very sensitive to Raman vertex changes, which can produce variation of its position between $\Delta_0$ and $2\Delta_0$ [@multi], not allowing for a comprehensive modeling of different cuprates.
In Fig. \[sens\] we address the problem of the sensitivity of the result to changes in the bare Raman vertex $\gamma({\bf k})$. To investigate the effect of changes of the vertex function, we have calculated the final response using the three different forms for the vertex $\gamma({\bf k})=\cos(k_x)+
\cos(k_y)$, $\gamma({\bf k})=\cos(k_x)\cos(k_y)$ and in the EMA which posses the correct transformation properties required by symmetry. In the first panel of Fig. \[sens\], we have plotted the Raman response for $\bar U=0$, i.e. the PB contribution alone, and in the second panel for $\bar U/t=1.3$. All curves are renormalized to their peak height to allow for an easier comparison. Clearly the strong sensitivity to changes of the bare Raman vertex (first panel) is much reduced when the SF term is added (second panel).
In Fig. \[fit\] we compare the theoretical results with experimental data on optimally doped Bi-2212 [@multi]. Adding the SF contribution leads to a shift of the peak position from near $\sim
\Delta_0$ for $\bar U=0$ to higher frequencies, and thus to a better agreement with the experimental relative peak positions in $A_{1g}$ and $B_{1g}$ geometries. For the fit we have adjusted $t$ to achieve a good agreement with the $B_{1g}$ channel, and then adjusted $\bar U$ to match the $A_{1g}$ peak position.
The value of $t$ obtained from the fit is $t=130$ meV. This value has to be compared with $t \approx 105$ meV, which results from the condition $\omega_R \approx 40$ meV. This slight discrepancy is most probably related to our simple modeling of the propagators which neglects strong renormalizations from interactions as well as impurities.
From this work we conclude that including the SF contribution in the Raman response solves the previously unexplained sensitivity of the $A_{1g}$ response to small changes in the Raman vertex. Also, within our model it is now possible to obtain the correct relative peak positions of the $A_{1g}$ and the $B_{1g}$ scattering geometry. Whereas the SF (two-magnon) contribution controls the $A_{1g}$ peak, the $B_{1g}$ and $B_{2g}$ scattering geometries are essentially unaffected and determined by pair breaking processes alone.
We would like to thank R. Hackl for numerous discussions. One of the authors (F.V.) would like to thank the Gottlieb Daimler and Karl Benz Foundation for financial support. This work was partially supported by the Deutsche Forschungsgemeinschaft through SFB 484.
[99]{}
T.P. Devereaux, D. Einzel, B. Stadlober, R. Hackl, D.H. Leach, and J.J. Neumeier, Phys. Rev. Lett. [**72**]{}, 396 (1994). T.P. Devereaux and D. Einzel, Phys. Rev. B [**51**]{}, 16336 (1995), [*ibid.*]{} [**54**]{}, 15547 (1996). J. Annett, N. Goldenfeld, and A.J. Leggett, in [*Physical Properties of High Temperature Superconductors*]{}, Ed. D. M. Ginsberg (World Scientific, Singapore, 1996), Vol. 5. T.P. Devereaux and A.P. Kampf , Phys. Rev. B [**59**]{}, 6411 (1999). T.P. Devereaux, A. Virosztek, and A. Zawadowski, Phys. Rev. B, [**54**]{} 12523 (1996). F. Wenger and M. Käll, Phys. Rev. B [**55**]{}, 97 (1997); T. Strohm and M. Cardona, Phys. Rev. B [**55**]{}, 12725 (1997); ibid. [**58**]{}, 8839 (1998); D. Manske, C. Rieck, R. Das Sharma, A. Bock, and D. Fay, Phys. Rev. B [**56**]{}, R2940 (1997); ibid. [**58**]{} 8841 (1998); A.V. Chubukov, D.K. Moor, and G. Blumberg, Sol. State Comm. [**112**]{}, 183 (1999). H.F. Fong, B. Keimer, P.W. Anderson, D. Reznik, F. Dogan, and I.A. Aksay, Phys. Rev. Lett. [**75**]{}, 316 (1995); H.F. Fong, P. Bourges, Y. Siddis, L.P. Regnault, A. Ivanov, G.D. Gu, N. Koshizuka, and B. Keimer, Nature [**398**]{}, 588 (1999). H.A. Mook, M. Yethiraj, G. Aeppli, T.E. Mason, and T. Armstrong, Phys. Rev. Lett. [**70**]{}, 3490 (1993). N. Bulut and D.J. Scalapino, Phys. Rev. B [**53**]{}, 5149 (1996). O.K. Andersen, O. Jepsen, A.I. Liechtenstein, and I.I. Mazin, Phys. Rev. B [**49**]{}, 4145 (1994). A.P. Kampf and W. Brenig, Z. Phys. B- Cond. Mat. [**89**]{}, 313 (1992). Note that the term “screening” has been used to describe the backflow yielding the second term in Eq. (\[screening\]). However, the backflow is independent of the charge of the carriers and comes into play solely from the pair interactions responsible for superconductivity. It yields the well known Anderson-Bogoliubov mode which restores gauge invariance. Details can be found in Ref. [@tpd3]. H.J. Vidberg and J.W. Serene, J. Low Temp. Phys. [**29**]{}, 179 (1977).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The near maximal value for the atmospheric neutrino mixing angle together with the fact that the solar mixing angle satisfies the relation $\sin^2\theta_\odot\simeq \frac{1}{3}$ is the basis for the so called tri-bimaximal mixing when $\theta_{13}=0$. In this note, we explore the possibility that tri-bimaximal mixing is an indication of a softly broken higher leptonic symmetry $S_3$, the permutation of three lepton families that embeds the $\mu-\tau$ exchange symmetry of leptons.'
author:
- '**R.N. Mohapatra, S. Nasri and Hai-Bo Yu**'
date: 'May, 2006'
title: '**$S_3$ Symmetry and Tri-bimaximal Mixing**'
---
Introduction
============
Observation of nonzero neutrino masses and determination of two of their three mixing parameters by experiments have raised the hope that neutrinos may hold the key to unraveling the flavor puzzle for quarks and leptons [@rev]. In order to make progress towards realizing this goal, one must first decipher the underlying reason for the observed leptonic mixing pattern and then search for a unified description of quarks and leptons to understand the quark flavor puzzle.
An interesting possibility to explore is that lepton mixings are indications of underlying flavor symmetries. Two tantalizing hints in favor of this are the near maximal atmospheric mixing angle and vanishing mixing angle $\theta_{13}$. It has been speculated that they are consequences of an approximate discrete $\mu-\tau$ symmetry[@mutau; @moh] for leptons. Its presence can be tested under certain circumstances. Even though there is no such apparent “$\mu-\tau$” symmetry among quarks and charged leptons, it has been shown that unified description of quarks and leptons with this symmetry is possible[@nishiura]. A question raised by this is whether there are higher underlying symmetries of leptons.
A hint for a higher symmetry may be coming from the observation that the solar angle in the PMNS mixing matrix satisfies the relation $\sin^2\theta_\odot\simeq \frac{1}{3}$. The resulting PMNS matrix has the simple form[@tbm]: $$\begin{aligned}
U_{PMNS}~=~\pmatrix{\sqrt{\frac{2}{3}} & \frac{1}{\sqrt{3}} & 0\cr
-\frac{1}{\sqrt{6}} & \frac{1}{\sqrt{3}} & \frac{1}{\sqrt{2}}\cr
-\frac{1}{\sqrt{6}}& \frac{1}{\sqrt{3}} & -\frac{1}{\sqrt{2}}}\end{aligned}$$ and is called a tri-bimaximal mixing. The true nature of the symmetry responsible for this pattern is not clear, although there are many interesting suggestions[@a4; @o3; @s3].
In this brief note, we explore the possibility that the relevant symmetry may be the permutation symmetry $S_3$ of three lepton generations. Such connections have been considered in literature from phenomenological points of view[@s3] and it will be interesting to see to what extent a full gauge model for tri-bimaximal pattern based on $S_3$ group can be developed. In this note we take some steps in this direction. We show that a softly broken $S_3$ symmetry for leptons can lead to tri-bimaximal mixing pattern if we use a combination of type I and type II seesaw mechanism to understand the smallness of neutrino mass. This approach appears to be different from previous attempts at building models for tri-bimaximal mixing[@a4; @o3; @s3].
We proceed in two steps: we first show how in a basis where charged leptons are diagonal, one can derive the mixing pattern in Eq. (1) using softly broken $S_3$ symmetry under certain assumptions. We then show how this the $S_3$ symmetry combined with $ Z_{2e}\times Z_{2\mu}\times Z_{2\tau}$ symmetry can lead to a diagonal charged lepton mass matrix. We then extrapolate the neutrino mass matrix from the seesaw scale to the weak scale and obtain constraints on the mass ratios $m_1/m_3$ and $m_2/m_3$ so that the mixing angles match the observations. We obtain a prediction for $\theta_{13}$, which turns out to be extremely small ($\sim 0.004$). We further show that if the neutrino masses are quasi-degenerate and have the same CP property (i.e. are all positive), then the radiative corrections in the extrapolation to the weak scale are so large that the solar mixing angle is in disagreement with observations. This implies that in supersymmetric theories with large tan$\beta$, seesaw scale tri-bimaximal mixing and degenerate neutrinos are mutually exclusive.
An $S_3$ model
==============
We start with the Majorana neutrino mass matrix whose diagonalization at the seesaw scale leads to the tri-bimaximal mixing matrix: $$\begin{aligned}
{\cal M}_\nu~=~\pmatrix{a & b & b\cr b & a-c & b+c\cr b & b+c &
a-c}\end{aligned}$$ Diagonalizing this matrix leads to the $U_{PMNS}$ of Eq. (1) and the neutrino masses: $m_1=a-b; m_2=a+2b$ and $m_3=a-b-2c$. Clearly if $|a|\simeq |b| \ll |c|$, we get a normal hierarchy for masses.
We now show that the mass matrix in Eq.(2) can be obtained from a softly broken $S_3$ symmetry in the neutrino sector. For this purpose, we assign the three lepton doublets of the standard model $(L_e,L_\mu,L_\tau)$ to transform into each other under permutation. The three right handed neutrinos $(\nu_{R,i=1,2,3})$ transform under three permutation and two cyclic operations of $S_3$ as: $$\begin{aligned}
\nonumber e\leftrightarrow \mu:~ \nu_{R,1}\leftrightarrow -\nu_{R,1}
;~ \nu_{R,2}\leftrightarrow -\nu_{R,3}\\ \nonumber\mu\leftrightarrow
\tau:~ \nu_{R,2}\leftrightarrow -\nu_{R,2};~\nu_{R,1}\leftrightarrow
-\nu_{R,3} \\\nonumber \tau\leftrightarrow e:~
\nu_{R,3}\leftrightarrow -\nu_{R,3};~\nu_{R,1}\leftrightarrow
-\nu_{R,2}\\
\nonumber e\rightarrow\mu\rightarrow\tau:~
\nu_{R,1}\rightarrow\nu_{R,2};~\nu_{R,2}\rightarrow\nu_{R,3};~
\nu_{R,3}\rightarrow\nu_{R,1}\\
e\rightarrow\tau\rightarrow\mu:~\nu_{R,1}\rightarrow\nu_{R,3};~
\nu_{R,2}\rightarrow\nu_{R,1};~\nu_{R,3}\rightarrow\nu_{R,2}\end{aligned}$$ In order to obtain the neutrino mass matrix, we assume that there is a standard model triplet Higgs field $\Delta$ with $Y=2$ which is $S_3$ singlet that couples to the two lepton doublets and an $S_3$ singlet Higgs doublet field $H$ that gives the Dirac mass for the neutrinos. The triplet vev can be made small and of the desired order if the mass of the triplet Higgs field is around $10^{14}$ GeV or so[@rev].
The first point to note is that the most general $S_3$ invariant coupling of the triplet i.e. $f_{ab}L_aL_b \Delta$ is given by the coupling matrix: $$\begin{aligned}
f~=~\pmatrix{a & b & b\cr b & a & b\cr b & b & a}\end{aligned}$$ For the Dirac neutrino coupling we choose to keep the following $S_3$ invariant term: $$\begin{aligned}
{\cal
L}_D~=~h_\nu[\overline{\nu_{R,1}}H(L_e-L_\mu)+
\overline{\nu_{R,2}}H(L_\mu-L_\tau)+\overline{\nu_{R,3}}H(L_\tau-L_e)]+h.c.\end{aligned}$$ One other $S_3$ invariant coupling is set to zero. This is natural in a supersymmetric theory due to the nonrenormalization theorem. We then get for the Dirac mass matrix for neutrinos $$\begin{aligned}
M_D=\left[\matrix{d&-d&0\cr 0&d&-d\cr -d&0&d}\right].\end{aligned}$$ where $d=h_\nu<H>$. If we now assume the following hierarchy among the right handed neutrinos, i.e. $M_{\nu_{R, 1,3}}\gg M_{\nu_{R,2}}$ so that a single right handed neutrino dominates the type I contribution to the seesaw formula[@king], then in the strict decoupling limit, using the mixed type I+II seesaw formula: $$\begin{aligned}
{\cal M}_\nu~=~M_0-M^T_DM^{-1}_{\nu_R}M_D,\end{aligned}$$ we get the desired form for the neutrino Majorana mass matrix (Eq.(2)) which leads to tri-bimaximal mixing. Note that the right handed neutrino masses being dimension three operators break the $S_3$ softly.
In this discussion we have assumed that the charged lepton mass matrix is diagonal. A major challenge for any model for neutrino mixings is to have a consistent picture for both the charged lepton and neutrino sectors simultaneously so that the combination $U^{\dagger}_\ell U_\nu$ equals the observed PMNS matrix. Since in our case, the neutrino sector by itself gives the tri-bimaximal form for the PMNS matrix, the charged lepton sector should be diagonal or nearly so. We will now show that we can obtain a diagonal charged lepton mass matrix in a simple way using the $S_3$ symmetry, provided we choose only one of two allowed $S_3$ invariant Yukawa coupling terms.
In order to achieve this, we assume that there are three standard model Higgs doublets $(H_e,H_\mu,H_\tau)$ transforming like the lepton doublets above under $S_3$. We also assume that the right handed charged leptons $(e_R,\mu_R, \tau_R)$ transform under $S_3$ same way. We then assume a product of discrete symmetries $Z_{2e}\times Z_{2\mu}\times Z_{2\tau}$ under which all fields except the following are even: $(e_R,H_e)$ odd under only $Z_{2e}$ and similarly $(\mu_R, H_\mu)$ are odd only under $Z_{2\mu}$ and $(\tau_R, H_\tau)$ odd under $Z_{2\tau}$. The Yukawa couplings invariant under this are: $$\begin{aligned}
\nonumber {\cal L}'_Y~=~h_e (\bar{L}_eH_ee_R +\bar{L}_\mu
H_\mu\mu_R+\bar{L}_\tau
H_\tau\tau_R)~+~h'_e(\bar{L}_eH_\mu\mu_R+\bar{L}_\mu H_e
e_R\\+\bar{L}_\mu H_\tau\tau_R+\bar{L}_\tau
H_\mu\mu_R+\bar{L}_eH_\tau\tau_R+\bar{L}_\tau H_e e_R)+h.c.\end{aligned}$$ By softly breaking the global $S_3$ symmetry in the Higgs potential for the $H_{e,\mu,\tau}$, we can get $<H_e>\ll <H_\mu>\ll <H_\tau>$ which allows us to obtain a realistic diagonal charged lepton mass matrix if we assume $h'_e=0$. This model then gives us a tri-bimaximal neutrino mixing at the seesaw scale.
Some implications
=================
In order to compare this model with observations, we need to extrapolate the seesaw scale neutrino mass matrix in Eq.(2) down to the weak scale[@rod] and then calculate the masses and mixing angles. This extrapolation depends on the mass hierarchy of the neutrinos. So comparing with observations, we can put limits on the mass hierarchy at low scale. From the expressions for the neutrino masses derived after Eq.(2), one might think that degenerate masses are compatible with tri-bimaximal pattern since there are three parameters and three masses to be fitted. However, in supersymmetric models, mixing angles can receive substantial contributions from RGE effects (specially for large tan$\beta$) and will in general lead to distortion of the mixing angles away from the tri-bimaximal values. For the specific case of tan$\beta=50$ we calculate the radiative corrections to the solar mixing angle $\theta_{12}$ in Fig. 1. We plot sin$^2\theta_{12}$ against $m_2/m_3$ with the input constraint being that $\Delta m^2_\odot/\Delta m^2_{ATM}$ is within 3 $\sigma$ of its present value i.e. $0.024\leq\Delta m^2_{\odot}/\Delta m^2_{ATM}\leq
0.060$[@valle]. We see that for $m_2/m_3 > 0.3$ or so, the solar mixing angle goes outside the observed range and the agreement gets worse for larger values of this mass ratio which corresponds to quasi-degenerate neutrino spectrum. This leads us to conclude that tri-bimaximal mixing at the seesaw scale is incompatible with quasi-degenerate neutrinos for large values of $ \tan \beta$.
For the same value of $\tan \beta$, we show in Fig.2 the allowed ranges for the neutrino mass ratios for the case of normal hierarchy and in Fig.3, the prediction for $\theta_{13}$ for this model. In these figures, we have used the above $3\sigma$ experimental bounds for $\Delta m^2_{\odot}/\Delta m^2_{ATM}$ and also 3 $\sigma$ bounds for $0.23\leq\sin\theta^2_{12}\leq 0.38$ and $0.34\leq\sin\theta^2_{23}\leq
0.68$[@valle]. We find that the prediction for $\theta_{13}\sim 0.004$ which is much too small to be observable in near future. This is because the low energy theory in the absence of radiative corrections is $\mu-\tau$ symmetric. Clearly observation of $\theta_{13}$ higher than this value will rule out this model and indeed any simple model for tri-bimaximal mixing at the seesaw scale for the case of normal mass hierarchy.
Conclusion
==========
In conclusion, in this brief note, we have presented a new way to obtain the tri-bimaximal mixing pattern for neutrinos by embedding $\mu-\tau$ symmetry of the neutrino mass matrix into a softly broken $S_3$ permutation symmetry for leptons and using a simple combination of the type I and type II seesaw formulae along with the dominance of a single right handed neutrino[@king]. We also find that tri-bimaximal mixing at the seesaw scale is incompatible with degenerate neutrino spectrum due to large radiative correction effects for large $ \tan \beta$.
This work is supported by the National Science Foundation grant no. Phy-0354401
[99]{}
For a recent review of the theoretical issues in neutrino physics, see R. N. Mohapatra and A. Y. Smirnov, to appear in Ann. Rev. Nucl. Science, [**56**]{} (2006).
T. Fukuyama and H. Nishiura, hep-ph/9702253; R. N. Mohapatra and S. Nussinov, Phys. Rev. [**D 60**]{}, 013002 (1999); E. Ma and M. Raidal, Phys. Rev. Lett. [**87**]{}, 011802 (2001); C. S. Lam, hep-ph/0104116; T. Kitabayashi and M. Yasue, Phys.Rev. [**D67**]{} 015006 (2003); W. Grimus and L. Lavoura, hep-ph/0305046; 0309050; Y. Koide, Phys.Rev. [**D69**]{}, 093001 (2004);Y. H. Ahn, Sin Kyu Kang, C. S. Kim, Jake Lee, hep-ph/0602160; A. Ghosal, hep-ph/0304090; for examples of such theories, see W. Grimus and L. Lavoura, Phys. Lett. B [**572**]{}, 189 (2003); W. Grimus and L. Lavoura, J. Phys. G [**30**]{}, 73 (2004).
W. Grimus, A. S.Joshipura, S. Kaneko, L. Lavoura, H. Sawanaka, M. Tanimoto, hep-ph/0408123; R. N. Mohapatra, JHEP, [**0410**]{}, 027 (2004); A. de Gouvea, Phys.Rev. [**D69**]{}, 093007 (2004); R. N. Mohapatra and W. Rodejohann, Phys. Rev. [**D 72**]{}, 053001 (2005); T. Kitabayashi and M. Yasue, Phys. Lett,. [**B 621**]{}, 133 (2005); R. N. Mohapatra and S. Nasri, Phys. Rev. D [**71**]{}, 033001 (2005);R. N. Mohapatra, S. Nasri and H. B. Yu, Phys. Lett. B [**615**]{}, 231 (2005).
K. Matsuda and H. Nishiura, Phys. Rev. D [**73**]{}, 013008 (2006); A. Joshipura, hep-ph/0512252; R. N. Mohapatra, S. Nasri and H. B. Yu, Phys. Lett. [**B 636**]{}, 114 (2006).
L. Wolfenstein, Phys. Rev. [**D 18** ]{}, 958 (1978); P. F. Harrison, D. Perkins and W. G. Scott, Phys. Lett. [**B 530**]{}, 167 (2002); P. F. Harrison and W. G. Scott, Phys. lett. [**B 535**]{}, 163 (2002); Z. z. Xing, Phys. Lett. [**B 533**]{}, 85 (2002).
K. S. Babu and X. G. He, hep-ph/0507217; G. Altarelli and F. Feruglio, hep-ph/0512103; B. Adhikary, B. Brahmachari, A. Ghosal, E. Ma and M. K. Parida,hep-ph/0603059.
C. I. Low and R. R. Volkas, Phys. Rev. D [**68**]{}, 033007 (2003);I. de Medeiros Varzielas, S. F. King and G. G. Ross, hep-ph/0512313; W. Grimus and L. Lavoura, JHEP [**0601**]{}, 018 (2006); N. Haba, A. Watanabe and K. Yoshioka, hep-ph/0603116; P. Kovtun and A. Zee, hep-ph/0604169.
P. F. Harrison and W. G. Scott, Phys. Lett. B [**557**]{}, 76 (2003); F. Caravaglios and S. Morisi, hep-ph/0503234; R. Jora, S. Nasri and J. Schechter, hep-ph/0605069.
F. Plentinger and W. Rodejohann, Phys. Lett. B [**625**]{}, 264 (2005)
S. F. King, Nucl. Phys. B [**576**]{}, 85 (2000); S. F. King and N. N. Singh, Nucl. Phys. B [**591**]{}, 3 (2000).
M. Maltoni, T. Schwetz, M. A. Tortola and J. W. F. Valle, New J. Phys. [**6**]{}, 122 (2004)
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'F. M. Brunbauer,'
- 'C. Chatterjee,'
- 'G. Cicala,'
- 'A. Cicuttin,'
- 'P. Ciliberti,'
- 'M. L. Crespo,'
- 'D. D‘Ago,'
- 'S. Dalla Torre,'
- 'S. Dasgupta,'
- 'M. Gregori,'
- 'T. Ligonzo,'
- 'S. Levorato,'
- 'M. Lisowska,'
- 'G. Menon,'
- 'F. Tessarotto,'
- 'L. Ropelewski,'
- 'Triloki,'
- 'A. Valentini,'
- 'L. Velardi,'
- 'Y. X. Zhao'
title: 'Nanodiamond photocathodes for MPGD-based single photon detectors at future EIC'
---
Introduction {#sec:intro}
============
The future Electron Ion Collider (EIC) [@EIC] is the facility dedicated to understanding quantum chromodynamics (QCD), including the elusive non-perturbative effects and the answer to key questions, pending since long. Among them: the origin of nucleon mass and spin and the properties of dense gluon systems. The experimental activity at EIC requires efficient hadron Particle IDentification (PID) in a wide momentum range, including the challenging scope of hadron PID at high momenta, namely larger than $6$-$8~GeV/c$. A gaseous Ring Imaging CHerenkov (RICH) is the only possible choice for this specific task. The number of Cherenkov photons generated in a light radiator is limited. In spectrometer setups, these number of photons is recovered by using long radiators. The compact design of the experimental setup at the EIC collider imposes limitations on the radiator length, requiring a dedicated strategy. In the far UltraViolet (UV) spectral region ($\sim120$ nm), the number of generated Cherenkov photon increases, according to the Frank-Tamm distribution [@Frank:1937fk]. This suggests the detection of photons in the very far UV range. The standard fused-silica windows are opaque for wavelengths below 165 nm. Therefore, a windowless RICH is a potential option. The approach also points to the use of gaseous photon detectors operated with the radiator gas itself [@windowless-RICH].
The MicroPattern Gaseous Detector (MPGD)-based Photon Detectors (PD) have recently been demonstrated as effective devices [@PM18] for the detection of single photon in Cherenkov imaging counters. These PDs are composed of a hybrid structure, where two layers of THick GEM (THGEM) multipliers [@thgem] are followed by a MICRO-MEsh GAseous Structure (MICROMEGAS) [@mm] stage; the top layer of the first THGEM is coated with a reflective CsI PhotoCathode (PC).
CsI PC is, so far, the only feasible option for gaseous detectors thanks to its relatively high work function that makes it more robust than other ones commonly used in vacuum-based detectors. CsI has high Quantum Efficiency (QE) in the far UV spectral region. In spite of its successful application [@RD26], it presents problematic aspects. It is hygroscopic: the absorbed water vapour splits the CsI molecule, causing a degradation in QE [@NIMA_695_2012_279]. Therefore, the handling of CsI PC is a very delicate operation. QE degradation also appears after intense ion bombardment, when the integrated charge is $1~mC/cm^{2}$ [@NIMA_574_2007_28] or larger. In gaseous detectors, ion bombardment of the cathode is by the ion avalanche produced in the multiplication process. The fraction of ions reaching the cathode depends on the detector architecture. In recent years, MPGD schemes with enhanced ion blocking capability have been developed [@PM18; @ibf-blocking].
The search for an alternative UV sensitive photocathode overcoming these limitations is therefore an important goal for the research and development (R&D) program for the experiments at the EIC. In the present article, we present the preliminary results on NanoDiamond (ND) and Hydrogenated-NanoDiamond (H-ND) coated THGEM detectors. They also include preliminary results about QE robustness with respect to ion bombardment.
Nanodiamond based PC as an alternative of CsI PC
================================================
The high QE value of CsI photocathode makes it the mostly used photoconverter for the UV sensitive devices. This high QE value is related to its low electron affinity ($0.1~eV$) and wide band gap ($6.2~eV$) [@JAP_77_1995_2138]. The ND particles have a comparable band gap of $5.5~eV$ and low electron affinity of $0.35$-$0.50~eV$. H-NDs exhibit chemical inertness and radiation hardness. ND hydrogenation lowers the electron affinity to -$1.27~eV$. The Negative Electron Affinity (NEA) allows an efficient escape into vacuum of the generated photoelectrons without an energy barrier at the surface [@NDRep-1]. A novel ND hydrogenation procedure, developed in Bari [@NDRep-1; @NDreport], provides high and stable QE. A comparison of CsI and ND QE can be extracted from literature [@NDRep-1; @NIMA_502_2003_76].
The R&D activity
================
THGEM characterization
----------------------
The initial phase of our R&D studies consisted in coating five THGEMs with ND and H-ND powder.
THGEMs are robust gaseous electron multipliers based on GEM principle scaling the geometrical parameters. It is obtained via standard PCB drilling and etching processes. The 35 $\mu$m copper layer is coated with $\approx$5 $\mu$m of Ni, followed by 200 nm Au. The THGEMs used for our studies have an active area of $30\times30~mm^{2}$ with a hole diameter of 0.4 mm, a pitch of 0.8 mm and a thickness of 470 $\mu$m. THGEMs with different rim i.e. the clearance ring around the hole edge have been used: $\le5\mu$m (no rim), $\sim10~\mu$m and $\sim20~\mu$m.
Each THGEM is characterized in the setup schematized in figure \[fig:Schematic\_of\_Detector\_setup\]. A plane of drift wires above it and a segmented readout anode plane, both properly biased, provide the drift and induction field respectively. The detector is operated with various gas mixtures, all including Ar. The electrons from $^{55}Fe$ converted by Ar are collected and multiplied in the hole region of the THGEM. The electron avalanche generated in the multiplication process, while drifting towards the anode, induces the detected signal.
![Top-left panel: The schematic of our detector assembly. Bottom-left panel: The detector is illuminated with an ${}^{55}Fe$ X-ray source . Top-right panel: A typical ${}^{55}Fe$ X-ray spectrum obtained in $Ar-CO_{2}~(70\%-30\%)$ gas mixture when the applied voltages at drift, top and bottom of THGEM are -2520 V, -1720 V and -500 V respectively, while the anode is at ground. Bottom-right panel: shows the gain dependence of the THGEM versus the applied voltage.[]{data-label="fig:Schematic_of_Detector_setup"}](Schematic_of_Detector_setup.jpg){width="\textwidth"}
All THGEMs used for our studies have been characterized using different gas mixtures at INFN Trieste before applying coating procedures: the goal is to perform comparative studies after coating them with UV sensitive films.
A typical ${}^{55}Fe$ X-ray spectrum obtained in $Ar-CO_{2}~(70\%-30\%)$ gas mixture is shown in figure \[fig:Schematic\_of\_Detector\_setup\], top-right panel. The bottom right panel of figure \[fig:Schematic\_of\_Detector\_setup\] shows the gain dependence of THGEM versus the voltages applied between the two faces.
Coating procedure
-----------------
![ (A) Au\_PCB of 1 inch diameter substrate used for the QE measurement. (B) Uncoated THGEM of active area 30 mm$\times$30 mm. (C) Half uncoated and half coated THGEM, mounted into the test chamber and zoomed view of the both coated (D) and uncoated (E) part. (F) test chamber with readout pad where the THGEMs are tested. (G) The test chamber after installation of a THGEM, illuminated by an ${}^{55}Fe$ X-ray source.[]{data-label="fig:Detector_Images"}](Detector_Images.png){width="\textwidth"}
THGEMs have been coated with raw ND, namely as-received powder or with H-ND. ND powder with an average grain size of 250 nm produced by Diamonds & Tools srl has been employed. The coating is covering either the whole surface of one of the THGEM faces, or half of it.
The standard procedure of hydrogenation of ND powder photocathodes is performed by using the MicroWave Plasma Enhanced Chemical Vapor Deposition (MWPECVD) technique. For the treatment in microwave (mw) H2 plasma, 30 mg of ND powder was placed in a tungsten boat (overall length 32 mm, trough 12 mm long × 5 mm wide × 1 mm deep, Agar Scientific Ltd) positioned on a heatable substrate holder of an ASTeX-type reactor evacuated to a base pressure better than $7\times{10}^{-7}$ mbar. The powder was heated to $650^{0}$C using an external radiative heater (via a Proportional-Integral-Derivative feedback control system), then H2 gas was flowed in the chamber at 200 sccm, the pressure and the mw power were maintained at 50 mbar and 1250 W, respectively. The heating due to the mw power increases further the temperature of the powders up to $1138^{0}$C as determined by a dual wavelength ($\lambda$1 = 2.1 $\mu m$ and $\lambda$2 =2.4 $\mu m$) infrared pyrometer (Williamson Pro 9240). After 1 h of H2 plasma exposure, the hydrogenated powder were cooled to room temperature under high vacuum.This procedure can not be used for THGEMs which are made by fiberglass, which does not tolerate temperatures above $180^{0}$C.
This limitation is overcome by the novel and low-cost technique developed at INFN Bari [@coating-I; @coating-II]. The H-ND is obtained by treating the as-received powder in $H_{2}$ microwave plasma for one hour before deposition.
The ND and H-ND powder were separately dispersed in deionized water and sonicated for 30 minutes by a Bandelin Sonoplus HD2070 system. Then, the emulsion was sprayed on the THGEM at $120 ^{0}$C or slightly higher temperature.
Four THGEMs with different geometrical characteristics have been coated as listed below:
- 0 $\mu m$ rim - ND half coated
- 0 $\mu m$ rim - H-ND half coated
- 10 $\mu m$ rim - H-ND full coated
- 20 $\mu m$ rim - ND half coated
A fifth THGEM with 10 $\mu m$ rim was coated with a reflective CsI film by thermal evaporation technique at INFN Bari.
Images of the coated substrates and the setup for the characterization are provided in Fig. [\[fig:Detector\_Images\]]{}.
![The gain of the THGEM with 10 $\mu$m rim measured before and after CsI coating are compared. Left panel: Gain versus applied voltage across the THGEM electrodes. Right panel: gain evolution versus time.[]{data-label="fig:CsI_Comp"}](EG_CsI_Comp.png "fig:"){width=".56\linewidth"}![The gain of the THGEM with 10 $\mu$m rim measured before and after CsI coating are compared. Left panel: Gain versus applied voltage across the THGEM electrodes. Right panel: gain evolution versus time.[]{data-label="fig:CsI_Comp"}](10um_id2_CsI.png "fig:"){width=".44\linewidth"}
Post-coating THGEM characterization
-----------------------------------
The characterization of THGEMs coated with CsI, ND and H-ND provides interesting indications. The gain in the coated part tends to be larger than the gain for the uncoated part in all the three cases. However, the increase for the coated part depends on the coating materials as well as on rim size.
The THGEM with $\sim10~\mu$m rim size, coated with a reflective CsI showed a 20% gain increment in comparison to the uncoated ones as shown in Fig. \[fig:CsI\_Comp\]. A tentative explanation of the observed gain increase is the lower rate of charging-up of the free dielectric surface. The surface resistivity is decreased when the coating is present due to the resistivity of the coating film.
The left panel of Fig. \[fig:X\_Ray\_Spectra\], shows the amplitude distribution for both uncoated and ND coated parts of the THGEM with $\sim20~\mu$m rim size. The voltages applied to drift, top and bottom of the THGEM electrodes are 3510 V, 2110 V and 750 V respectively. The gain of the ND coated part is $\approx $ 2 times higher compared to the one of the uncoated part.
![ Left panel: ${}^{55}Fe$ X-ray spectra obtained with a 20 $\mu$m rim THGEM half-coated with ND powder in $Ar-CO_{2} (70\%-30\%)$ gas mixture. The voltages applied to drift, top and bottom of the THGEM electrodes are 3510 V, 2110 V and 750 V respectively, while the anode is kept at ground. Right panel: ${}^{55}Fe$ X-ray spectra obtained with a 0 $\mu$m rim uncoated THGEM. The same measurement after coating the same THGEM with a H-ND emulsion prepared 17 months earlier in a $Ar-CH_{4}~(50\%-50\%)$ gas mixture[]{data-label="fig:X_Ray_Spectra"}](THGEM_ND_Comp_new.pdf "fig:"){width=".485\linewidth"} ![ Left panel: ${}^{55}Fe$ X-ray spectra obtained with a 20 $\mu$m rim THGEM half-coated with ND powder in $Ar-CO_{2} (70\%-30\%)$ gas mixture. The voltages applied to drift, top and bottom of the THGEM electrodes are 3510 V, 2110 V and 750 V respectively, while the anode is kept at ground. Right panel: ${}^{55}Fe$ X-ray spectra obtained with a 0 $\mu$m rim uncoated THGEM. The same measurement after coating the same THGEM with a H-ND emulsion prepared 17 months earlier in a $Ar-CH_{4}~(50\%-50\%)$ gas mixture[]{data-label="fig:X_Ray_Spectra"}](BeforeAfterCoating_new.pdf "fig:"){width=".48\linewidth"}
In case of a ND coated THGEM with no rim the gain of the coated part is larger by a factor of $\sim$1.4 as shown in Fig.\[fig:Nor\_Gain\_LR\] left panel. The gain is maximum when the X-ray source starts illuminating and it decreases gradually by ${\sim}$ 20% in about 500 minutes. This effect is observed both for the ND coated and uncoated THGEM parts as illustrated in Fig.\[fig:Nor\_Gain\_LR\], right panel. The tentative explanation of the gain increase is the same one already proposed for the gain increase observed with CsI coating. The increase is higher when the open dielectric surface is larger, namely for the $\sim20~\mu$m rim THGEM.
![Evolution versus time of the effective gain behavior of a THGEM with 0 $\mu$m rim, half-coated with ND. Gain versus time (left panel); the source is moved to the coated region at 700 minutes. The same data normalized to the maximum gain measured in the coating region versus time, where t=0 is when the illumination of a region starts (right panel).[]{data-label="fig:Nor_Gain_LR"}](Gain_Evaluation_Uncoated_ND_coated_THGEM_Corrected.pdf "fig:"){width=".48\linewidth"}![Evolution versus time of the effective gain behavior of a THGEM with 0 $\mu$m rim, half-coated with ND. Gain versus time (left panel); the source is moved to the coated region at 700 minutes. The same data normalized to the maximum gain measured in the coating region versus time, where t=0 is when the illumination of a region starts (right panel).[]{data-label="fig:Nor_Gain_LR"}](Nor_Gain_Evaluation_Corrected.pdf "fig:"){width=".48\linewidth"}
The H-ND coated THGEMs with 0 $\mu m$ and 10 $\mu m$ rim show a lower electrical stability as compared to the uncoated THGEMs and cannot be operated at nominal voltage. In order to study this unexpected behaviour a second exercise was performed using a new THGEM with no rim fully coated with H-ND and $Ar-CH_{4}~(50\%-50\%)$ gas mixture. Immediately after the coating the THGEM could not be operated at the nominal voltage. A heat treatment in an electric oven at ${120}~^{0}C$ for 24 hours allowed to perform the characterization. The right panel of figure \[fig:X\_Ray\_Spectra\] shows the signal amplitude distributions measured before and after the H-ND coating followed by the heat treatment. No evidence for increase of gain is observed. This observation suggests that the electrical instability present before the heat treatment can be related to water molecules present at the H-ND surface.
A consistent picture is emerging in spite of the initially unexpected results obtained characterising the coated THGEMs. The consolidation of this picture requires further investigation.
Ageing test of H-ND PC
----------------------
QE measurements can be performed with the available setup only in case of small-size samples. Therefore, the same coating procedure used for the THGEM samples has been applied to disk substrates, one inch diameter. The disk material and surface preparation are the same of the THGEMs, even if no hole structure is present (Fig. \[fig:Detector\_Images\] A). The H-ND coating is with an emulsion of ND powder hydrogenated 17 months earlier.
![Quantum efficiency as a function of wavelength for fresh and various charge accumulations ($0.263mC{/}cm^{2}, ~2.895mC{/}cm^{2}, ~5.527mC{/}cm^{2} and ~8.159mC{/}cm^{2}$) due to ion bombardment on H-ND coated Au\_PCB substrate.[]{data-label="fig:Ageing_New"}](X_ray_Ageing_Corrected.pdf){width="\textwidth"}
The H-ND coated PCB disc was irradiated with a mini X-ray source in $Ar-CO_{2} (70\%-30\%)$ gas mixture. For the QE measurement a McPherson VUV monochromator (model 234/302) was employed. The RD-51 ASSET system [@asset] at CERN has been used. In this setup both the QE measurement and ion bombardment can be performed in the same system without exposure to air. Ions are generated by a gaseous multiplier metallic grid set at 5 mm from the disc surface and they impinge on the sample surface. The QE of an H-ND coated disc was measured before irradiation in a wavelength range from 130 nm to 180 nm with a scan step of 5 nm. The sample was then moved to the X-ray irradiation chamber using an automated manipulator under vacuum ($\approx {1\times{10}^{-7} mbar}$). The QE has been measured again after controlled doses of accumulated charge and the measurements are reported in Fig. \[fig:Ageing\_New\]. The QE before irradiation and after an accumulated charge of $0.263~ mC{/}cm^{2}$ are similar in whole wavelength range. This strongly supports the hypothesis that H-ND photocathode are more robust than CsI once respect to ion bombardment. In fact, in case of a CsI photocathode, a 25% drop in QE is observed for an irradiation of $1.0~ mC{/}cm^{2}$ [@NIMA_574_2007_28]. For the H-ND coated sample, a decrease in the QE of 42$\%$ and 74$\%$ was observed as charge accumulation reached the values of $2.895~ mC{/}cm^{2}$ and $5.527 ~mC{/}cm^{2}$, respectively. Interestingly, we did not observe any further degradation in the QE for an accumulated charge of $8.159~ mC{/}cm^{2}$. This is the first preliminary irradiation ageing study of H-ND photocathodes ever performed.
Conclusion
==========
THGEM samples coated with different types of photosensitive layers (CsI, ND and H-ND) have been studied. An increase in the gain response for the CsI and ND coated THGEMs was observed compared to the uncoated ones. The electrical instability of the H-ND coated THGEM, initially observed, is overcome by a heat treatment. No gain enhancement is observed.
The X-ray irradiation study on H-ND photocathodes performed by us for the first time indicates that H-ND is more robust than CsI. We can conclude that H-ND photocathode material is a promising alternative to CsI based photocathodes for all applications in the far VUV domain requiring high robustness.
Acknowledgment
==============
This R&D activity is partially supported by
- EU Horizon 2020 research and innovation programme, STRONG-2020 project, under grant agreement No 824093;
- the Program Detector Generic R&D for an Electron Ion Collider by Brookhaven National Laboratory, in association with Jefferson Lab and the DOE Office of Nuclear Physics.
[99]{}
A. Accardi et al., *Electron-Ion Collider: The next QCD frontier*, Eur. Phys. J. A **52** (2016) 268.
I. Frank and I. Tamm, *“Coherent visible radiation of fast electrons passing through matter,”* Compt. Rend. Acad. Sci. URSS **14**, **no.3**, (1937) 109-114. M. Blatnik et al., *Performance of a Quintuple-GEM Based RICH Detector Prototype*, IEEE NS **62** (2015) 3256.
J. Agarwala et al., *The MPGD-based photon detectors for the upgrade of COMPASS RICH-1 and beyond*, Nucl. Instr. and Meth. A **936** (2019) 416.
L. Periale et al., *Detection of the primary scintillation light from dense Ar, and Xe with novel photosensitive gaseous detectors*, Nucl. Instr. and Meth. A **478** (2002) 377; P. Jeanneret, *Time Projection Chambers and detection of neutrinos, PhD thesis*, Neuchatel University, 2001; P.S. Barbeau et al, *Toward coherent neutrino detection using low-background micropattern gas detectors* IEEE **NS-50** (2003) 1285; R. Chechik et al, *Thick GEM-like hole multipliers: properties and possible applications*, Nucl. Instr. and Meth. A **535** (2004) 303.
Y. Giomataris et al., *MICROMEGAS: a high-granularity positionsensitive gaseous detector for high particle-flux environments*, Nucl. Instr. and Meth. A **376** (1996) 29.
The RD26 Collaboration, *RD26 status reports: CERN/DRDC* **93-36** (1993), CERN/DRDC **94-49** (194), CERN/DRDC (1996).
Triloki n, B.Dutta,B.K.Singh, *Influence of humidity on the photoemission properties and surface morphology of cesium iodide photocathode*, Nucl. Instr. and Meth. A, **695** (2012) 279.
H.Hoedlmoser et al., *Long term performance and ageing of CsI photocathodes for the ALICE/HMPID detector*, Nucl. Instr. and Meth. A, **574** (2007) 28.
A. Bondar et al., *Study of ion feedback in multi-GEM structures*, Nucl. Instr. and Meth. A **496** (2003) 325; A. Breskin et al., *Sealed GEM photomultiplier with a CsI photocathode: ion feedback and ageing*, Nucl. Instr. and Meth. A **478** (2002) 225; J.F.C.A. Veloso et al., *A proposed new microstructure for gas radiation detectors: The microhole and strip plate*, Rev. Sc. Instr. **71** (2000) 2371; A.V. Lyashenko et al., *Further progress in ion back-flow reduction with patterned gaseous hole - multipliers*,JINST **2** (2007) P08004;A.V. Lyashenko et al., *Efficient ion blocking in gaseous detectors and its application to gas-avalanche photomultipliers sensitive in the visible-light range*, Nucl. Instr. and Meth. A **598** (2009) 116;M. Alexeev et al., *Ion backflow in thick GEM-based detectors of single photons*, JINST **8** (2013) P01021. M. Bari et al., *RHIP, a Radio-controlled High-Voltage Insulated Picoammeter and its usage in studying ion backflow in MPGD-based photon detectors*; PoS **MPGD2017** 068; M.M. Aggarwal et al, (ALICE TPC Collaboration), *Particle identification studies with a full-size 4-GEM prototype for the ALICE TPC upgrade*, Nucl. Instr. and Meth. A **903** (2018) 215.
A.Buzulutskov, A.Breskin, R.Chechik, *Field enhancement of the photoelectric and secondary electron emission from CsI* ,Journal of Applied Physics **77** (1995) 2138.
L. Velardi, A. Valentini, G. Cicala, *Highly efficient and stable UV photocathode based on nanodiamond particles*, Appl. Phys. Lett. **108** (2016) 083503-1-5.
A. Valentini, D. Melisi, G. De Pascali, G. Cicala, L. Velardi, A. Massaro, *High-efficiency nanodiamond-based ultraviolet photocathodes*, 30-03-2017 Patent n. **WO 2017/051318 A9**; International Patent n. **PCT/IB2016/055616** of September 21, 2016; National Patent Italia - n. 102015000053374 del 21 Settembre 2015, Istituto Nazionale di Fisica Nucleare e Consiglio Nazionale delle Ricerche.
F. Piuz, *Ring imaging Cherenkov systems based on gaseous photo-detectors: trends and limits around particle accelerators*, Nucl. Instr. and Meth. A , **502** (2003) 76.
G. Cicala, A. Massaro, L. Velardi, G. S. Senesi, A. Valentini, *Self-Assembled Pillar-Like Structures in Nanodiamond Layers by Pulsed Spray Technique*, ACS Appl. Mater. Interfaces **6** (2014) 21101.
L. Velardi, A. Valentini, G. Cicala, *UV photocathodes based on nanodiamond particles: effect of carbon hybridization on the efficiency*, Diam. Relat. Mater. **76** (2017) 1-8.
$https://indico.cern.ch/event/872501/contributions/3726017/attachments/1985809/3308869/Marta\_Lisowska\_-\_RD51\_Mini\_Week\_-\_Asset\_photocathode\_characterisation\_device.pdf$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We demonstrate a new spectroscopic method for studying electronic transitions in molecules with extremely broad range of angular momentum. We employ an optical centrifuge to create narrow rotational wave packets in the ground electronic state of $^{16}$O$_2$. Using the technique of resonance-enhanced multi-photon ionization, we record the spectrum of multiple ro-vibrational transitions between [[$X^3\Sigma_{g}^{-}$ ]{}]{}and [[$C^3\Pi_{g}$ ]{}]{}electronic manifolds of oxygen. Direct control of rotational excitation, extending to rotational quantum numbers as high as $N\gtrsim 120$, enables us to interpret the complex structure of rotational spectra of [[$C^3\Pi_{g}$ ]{}]{}beyond thermally accessible levels.'
author:
- Aleksey Korobenko
- 'Alexander A. Milner'
- 'John W. Hepburn'
- Valery Milner
title: Rotational spectroscopy with an optical centrifuge
---
Ro-vibrational spectroscopy of molecules often employs the technique of resonance-enhanced multi-photon ionization (REMPI) due to the high sensitivity, spectral resolution and versatility of the latter [@Letokhov1987]. In a typical setup, molecules are ionized through a resonant multi-photon electronic transition with a tunable laser, and the induced ion current is measured as a function of the laser wavelength. The frequency of the observed REMPI resonances reveals the ro-vibrational spectrum of the molecule, while their strength is proportional to the population of a given initial level and the transition probability to the excited state.
One traditional approach to rotational spectroscopy involves interpreting REMPI spectra by fitting them to the theoretical model of a molecular potential, while using appropriate spectroscopic constants (e.g. rotational constant $B_{v}$) as fitting parameters. This method is hardly applicable to the study of high-energy rotational states because of the two main reasons. Firstly, populating high rotational states by means of ro-vibrational energy transfer [@Mullin1995] or with laser light [@Karczmarek1999; @Li2000; @Cryan2011] is rather challenging, whereas thermal excitation may require unreasonably high temperatures, e.g. $\approx$50,000 K for reaching $N>100$ in oxygen.
The second problem stems from the high number of populated rotational levels leading to high complexity of the observed spectra. For example, spin-rotation and spin-orbit level splitting both in the ground and excited states of [[O$_{2}$ ]{}]{}, together with the two-photon selection rules $\Delta J=0,\pm 1,\pm 2$ in a typically used two-photon transition $C^3\Pi_g(v'=2)\leftarrow\leftarrow X^3\Sigma_g^-(v''=0)$, would result in as many as 21 peaks for each $N$[@Sur1986]. For a broadly populated thermal ensemble this would lead to 21 largely overlapping branches. Finally, theoretical treatment of extremely broad range of angular momenta may require using intermediate Hund’s cases or considering perturbations by other states[@Morrill1999], which often result in a high level of uncertainty.
Optical centrifuge is an alternative tool for exciting molecules to extremely high rotational states by means of non-resonant laser fields [@Karczmarek1999; @Villeneuve2000; @Yuan2011; @Yuan2011a]. In a recent study, we have shown that the centrifuge can be used to produce and control the so-called “super rotor” states - coherent rotational wave packets with ultra-high angular momentum $N$ and narrow distribution width $\Delta N \ll N$[@korobenko2013]. Here we utilize this unique capability of the centrifuge for the purpose of obtaining and interpreting complex REMPI spectra of oxygen super rotors ($0 < N \lesssim 120$). We excite oxygen to a narrow rotational wave packet whose center is accurately tuned across the broad range of well defined $N$ values. The centrifuge excitation is then followed by a REMPI measurement. Owing to the narrow $N$ distribution, the detected spectrum becomes significantly less congested, and identifying rotational resonances is greatly simplified.
{width="1.95\columnwidth"}
Following the original recipe by Karczmarek et al. [@Karczmarek1999], we utilize the output of an 800 nm, 35 fs (full width at half-maximum, FWHM), Ti:Sapphire regenerative amplifier. We split its spectrum in half at around the central wavelength (Fig. \[fig\_centrifuge\][**a**]{}), in a Fourier plane of a pulse shaper. Frequency chirps of equal magnitude (0.26 ps$^{-2}$) and opposite signs are applied to the “red” and “blue” arms of the centrifuge, as demonstrated by the cross-correlation frequency-resolved optical gating (XFROG) measurement (Fig. \[fig\_centrifuge\][**c**]{}). The two arms are combined with a polarizing beam splitter cube, and polarized with an opposite sense of circular polarization. Optical interference of the two circularly polarized frequency-chirped laser fields results in a pulse with rotating linear polarization (inset to Fig. \[fig\_centrifuge\]**c**). Because of the anisotropic polarizability, molecular axes line up along the axis of laser polarization, and then follow it adiabatically as the plane of polarization rotates with increasing angular frequency. Given the available spectral bandwidth, the accelerating centrifuge can reach angular frequencies on the order of 10 THz, which in the case of oxygen corresponds to the rotational quantum number $N \approx 119$.
As we have demonstrated in [@korobenko2013], truncating the spectrum of the centrifuge in a Fourier plane of the pulse shaper by a movable shutter (inset to Fig. \[fig\_centrifuge\][**a**]{}) enables accurate control of the rotational state of the centrifuged molecules. Characterizing the centrifuge field with XFROG allowed us to calibrate the final rotation speed of the centrifuge, and hence the corresponding molecular angular momentum, as a function of the shutter position.
![ Detection setup. Centrifuge beam is combined with a tunable UV laser pulse and focused inside a vacuum chamber on a supersonically expanded oxygen jet between the charged plates of a time-of-flight (TOF) mass spectrometer. The ionization rate is measured with a multi-channel plate (MCP).[]{data-label="fig_setup"}](setup){width="1\columnwidth"}
REMPI detection was carried out using narrowband nanosecond probe pulses tunable from 279 nm to 288 nm (0.1 cm$^{-1}$ linewidth, 500 $\mu $J, 50 Hz repetition rate). Probe beam was combined with a centrifuge beam on a dichroic mirror (Fig. \[fig\_setup\]), and focused with a 35 mm focal length spherical aluminium mirror on a supersonically expanded molecular jet passing between the charged plates of the time-of-flight (TOF) spectrometer. The jet was generated by an Even-Lavie pulsed valve ($25~\mu\mathrm{s}$ opening time, $150~\mu\mathrm{m}$ nozzle diameter) located $20~\mathrm{cm}$ away from the detection region. Ion current was detected with a microchannel plate (MCP) detector. Initial rotational temperature of the sample, extracted from the REMPI spectrum taken without the centrifuge field (Fig. \[fig\_overall\][**a**]{}), was about 10 K.
{width="1.95\columnwidth"}
The main result of this work is shown in Figure \[fig\_overall\], where the detected ion count is plotted against the probe wavelength (horizontal axis) and the final rotation speed $\Omega $ of the truncated centrifuge (vertical axis). The latter is expressed in terms of the angular momentum $N$ of an oxygen molecule rotating with the angular frequency $\Omega $, according to: $$\Omega=2\pi c \left[ E(N)-E(N-1) \right],$$ $$E(N)=BN(N+1)-DN^2(N+1)^2,$$ where $E$ is the energy of state $|N\rangle$, $c$ is the speed of light in vacuum, $B=1.438$ cm$^{-1}$ and $D=4.839\times 10^{-6}$ cm$^{-1}$ [@NIST]. The validity of Dunham expansion of rotational energy to second order in $N(N+1)$ at extremely high values of $N$ has been demonstrated in our previous study[@korobenko2013].
Each peak in the two-dimensional REMPI spectrogram of Fig. \[fig\_overall\][**b**]{} corresponds to a two-photon transition between a rotational level in the electronic ground state, [[$X^3\Sigma_{g}^{-}$ ]{}]{}, and a rotational level of [[$C^3\Pi_{g}$ ]{}]{}. Finite horizontal width of the observed peaks stems from the finite transition linewidth (as in conventional “1D REMPI” detection), whereas finite vertical spread reflects the narrow width of the excited rotational wave packet created by the centrifuge.
The complexity of the two-photon absorption line structure in rotationally hot oxygen gas is illustrated by red and green lines in Fig.\[fig\_overall\][**a**]{} which correspond to the hot thermal ensemble (simulated numerically) and the ensemble of centrifuged molecules (experimentally observed 2D spectrogram integrated along its vertical dimension), respectively. In sharp contrast to conventional 1D REMPI spectroscopy, controlled centrifuge spinning offers direct assignment of rotational quantum numbers to the observed REMPI peaks, as well as significantly better peak separation due to their distribution along the added second dimension.
Vertical traces originating from bright resonance peaks in Figure \[fig\_overall\][**b**]{} (examples are marked with white arrows) correspond to molecules which “leaked out” of the weakened centrifuge potential before reaching the terminal angular frequency of the centrifuge. After escaping the centrifuge, these molecules continue their free rotation while the trap is accelerating further. The three bright vertical stripes reproduce the initial cold beam spectrum (blue line in Fig. \[fig\_overall\][**a**]{}) and correspond to the molecules which were not trapped by the centrifuge. The width of the final rotational wave packets can be readily extracted as $\delta N\approx 7$ (FWHM), from the vertical cross sections, shown in Fig. \[fig\_overall\][**c**]{}. Here, we detected rotational states with $N$ as high as $\sim 80$. Rotational line broadening above $N\approx 60$ can be attributed to the increasing Rydberg-valence interaction (governed by the Frank-Condon overlap with the continuum wavefunctions) similarly to the previously observed rotational broadening in the lower vibrational states ($v'=0,1$) of the excited potential[@Sur1986].
{width="1.95\columnwidth"}
One can see that the peaks in Fig.\[fig\_overall\] are grouped in regular patterns, resembling Fortrat parabolas corresponding to different rotational branches. Within a single branch, the center of each consecutive resonant peak is shifted by $\Delta N =2$ (Fig. \[fig\_overall\][**c**]{}), reflecting the smallest step in the rotational ladder climbing executed by the centrifuge. Circularly polarized light can be used to further simplify the spectrum. As shown in Fig. \[fig\_fits\], the signal strength of different rotational branches depends on the handedness of probe polarization. This is due to the highly non-uniform population distribution among the magnetic sub-levels in the centrifuged wave packet, with most of the population concentrated at $M_J\simeq J$ (or $M_J\simeq-J$) [@Karczmarek1999].
![Comparison of the observed REMPI data for the perturbed $F_1$ spin-orbit component with the calculations based on spectroscopic constants from this work (red circles), White *et al.*[@white2004] (blue triangles) and Lewis *et al.*[@Lewis1999] (purple squares).[]{data-label="fig_comparison"}](comparison.pdf){width="0.95\columnwidth"}
To identify different rotational branches, we use three sets of spectroscopic constants (for $F_1,F_2$ and $F_3$ spin-orbit components of the excited state, respectively) from the previous studies on thermally excited ensembles[@Lewis1999; @white2004]. These constants are listed in Table \[tab\_constants\]. For $F_2$ and $F_3$ components, our results are well described by the constants provided by Lewis *et al.*[@Lewis1999]. On the other hand, the observed $F_1$ peaks do not agree well with the suggested numerical values ($\nu_0=69366$ cm$^{-1}$ and $B_0=1.6$ cm$^{-1}$), as shown in Figure \[fig\_comparison\]. This can be attributed to the complexity of the broadened and highly overlapping structure of $F_1$ lines, which makes it hard to interpret and fit the data from a thermally populated ensemble. Centrifuge spectroscopy enabled us to correct the values of $F_1$ spectroscopic parameters (Table \[tab\_constants\]) by performing the fit of the most pronounced $\Delta N=-2$ branch (Fig. \[fig\_comparison\]).
[**Spin-orbit branch**]{} $\mathbf{\nu_0}$[**, cm**]{}$^\mathbf{-1}$ $\mathbf{B_0}$[**, cm**]{}$^\mathbf{-1}$ [**D, cm**]{}$^\mathbf{-1}$
--------------------------- -------------------------------------------- ------------------------------------------ -----------------------------
$^3\Pi_0(F_1)$ 69375 1.585 $2.5\times 10^{-7}$
$^3\Pi_1(F_2)$ 69445 1.648 $1.0\times 10^{-5}$
$^3\Pi_2(F_3)$ 69550 1.685 $1.3\times 10^{-5}$
: Spectroscopic constants used to fit the data in Fig.\[fig\_fits\][]{data-label="tab_constants"}
![Ultra high-$N$ resonances. Probe polarization in [**a**]{} and [**b**]{} is the same as in Fig. \[fig\_fits\]](high.pdf){width="0.95\columnwidth"}
. \[fig\_high\]
In addition to the identified branches of $C^3\Pi_g(v'=2)\leftarrow\leftarrow X^3\Sigma_g^-(v''=0)$, washing out at $N\gtrsim 80$, we observed well resolved ultra-high rotational lines with probe polarizations both counter-rotating (Fig. \[fig\_high\][**a**]{}, branch “a”) and co-rotating (Fig. \[fig\_high\][**b**]{}, branch “b”) with respect to molecular motion. The corresponding rotational numbers exceeded 100 and reached 123 for the highest detected transition. Line separation in branch “a” was $84~\mathrm{cm}^{-1}$ at $N=117$, and $102~\mathrm{cm}^{-1}$ in branch “b” at $N=113$. The lines appear in pairs, with a splitting of about $12~\mathrm{cm}^{-1}$ for both branches, which can be attributed to $\Lambda$-doubling. Strong linewidth dependence on $N$ was detected in branch “b”: the doublet structure could not be resolved at $N=103$, whereas it was easily identified at $N=115$, where the linewidth of each individual resonance was found to be equal $7~\mathrm{cm}^{-1}$. While “b” branch resonances can still be attributed to $v'=2\leftarrow\leftarrow v''=0$ transition, branch “a” lies $\sim 1000~\mathrm{cm}^{-1}$ lower, pointing at its possible origin from a different vibrational state.
In conclusion, we have demonstrated a new spectroscopic method for studying the rotational structure of electronic transitions in molecules. The method is based on controlled molecular spinning with an optical centrifuge. We have applied this technique to $C^3\Pi_g(v'=2)\leftarrow\leftarrow X^3\Sigma_g^-(v''=0)$ in [[O$_{2}$ ]{}]{}, and showed an agreement with previously known spectroscopic constants for $F_2$ and $F_3$ spin-orbit components. In the case of strongly perturbed $F_1$ components, higher resolution of the implemented method enabled us to improve the previously known spectroscopic constants. We also observed narrow ($\sim 7~\mathrm{cm}^{-1}$) resonances from ultra-high rotational states with $N>100$ with highly non-monotonic linewidth dependence. Some of these resonances were attributed to a different excited vibrational state, which could not be resolved in previous studies at low $N$.
This work has been supported by the CFI, BCKDF and NSERC, and carried out under the auspices of the Center for Research on Ultra-Cold Systems (CRUCS). We gratefully acknowledge stimulating discussions with R. Krems, V. Petrovic, M. Shapiro and E. Grant.
[15]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [**]{} (, ) @noop [****, ()]{} [****, ()](http://link.aps.org/doi/10.1103/PhysRevLett.82.3420) [****, ()](http://link.aip.org/link/jcpsa6/v112/i14/p6255/s1) @noop [****, ()]{} [****, ()](http://link.aip.org/link/?JCPSA6/84/69/1) [****, ()](http://link.aip.org/link/?JCPSA6/111/173/1) [****, ()](http://link.aps.org/doi/10.1103/PhysRevLett.85.542) @noop [****, ()]{} [****, ()](http://pubs.rsc.org/en/content/articlehtml/2011/fd/c0fd00021c) [ ()](http://arxiv.org/abs/1304.0438) @noop [“,” ]{} () @noop [**]{}, , Vol. (, , ) [****, ()](http://link.aip.org/link/?jcp/120/2445/1) [****, ()](http://link.aip.org/link/?JCPSA6/111/186/1)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We obtain a sharp local well-posedness result for the Gradient Nonlinear Wave Equation on a nonsmooth curved background. In the process we introduce variable coefficient versions of Bourgain’s $X^{s,b}$ spaces, and use a trilinear multiscale wave packet decomposition in order to prove a key trilinear estimate.'
address:
- |
Department of Mathematics, Hylan Building\
University of Rochester, Rochester, NY 14627
- |
Department of Mathematics, Evans Hall\
University of California at Berkeley, Berkeley, CA 94720-3840
author:
- 'Dan-Andrei Geba and Daniel Tataru'
bibliography:
- 'tril.bib'
title: Gradient NLW on curved background in $4+1$ dimensions
---
Introduction
============
In this article we are investigating the issue of local well-posedness for a variable coefficient semilinear wave equation in $4+1$ dimensions. To describe the context and motivate the interest in our problem we introduce three related equations. We begin with a generic gradient NLW equation in ${\mathbb R}^{n+1}$, $$\Box u\,=\,\Gamma(u)(\nabla u)^2
\label{sem}$$ with the nonlinearity $$\Gamma(u) (\nabla u)^2\,=\,q^{ij}(u) {\partial}_i u\, {\partial}_j u$$ where $q^{ij}$ are smooth functions and the standard summation convention is used.
Then we move on to a similar equation but on a curved background, $$\Box_g u\,=\,\Gamma(u)(\nabla u)^2
\label{eq}$$ with $\Box_{g}\,=\,g^{ij}\,\partial_i\partial_j$, where the summation occurs from $0$ to $n$ and the index $0$ stands for the time variable. To insure hyperbolicity we assume that the matrix $g^{ij}$ has signature $(1,n)$ and the time level sets $x_0=const$ are space-like, i.e. $g^{00}>0$. In effect to simplify some of the computations we make the harmless assumption $g^{00}=1$.
Finally, we consider a corresponding quasilinear equation $$\Box_{g(u)} u\,=\,\Gamma(u)(\nabla u)^2
\label{qua}$$ with similar assumptions on the matrix $g$.
In all three cases we are interested in the local well-posedness of the Cauchy problem in Sobolev spaces $H^s({\mathbb R}^n)\times H^{s-1}({\mathbb R}^n)$ with initial data $$u(0,x)\,=\,u_0(x),\qquad \partial_t u(0,x)\,=\,u_1(x)
\label{id}$$ The first equation is the best understood so far, and is known to be locally well-posed for $s$ in the range $$s>\max\{\frac{n}{2},\frac{n+5}{4}\}$$ This range is sharp. The $\frac{n}{2}$ obstruction comes from scaling, while the $\frac{n+5}{4}$ is related to concentration along light rays, see Lindblad [@MR1375301]. The proof of the positive result is fairly straightforward in dimension $2+1$ and $3+1$, where it suffices to rely on the Strichartz estimates. In $4+1$ dimensions this no longer works and one needs to use instead the $X^{s,\theta}$ spaces, see Foschi-Klainerman [@MR1755116]. These are multiplier weighted $L^2$ spaces associated to the wave operator as the Sobolev spaces $H^s$ are connected to the Laplace operator $\Delta$, see Klainerman-Machedon [@MR1420552]: $$\| u \|_{X^{s,\theta}} = \| (1+|\xi|^2)^{\frac s2}\cdot
(1+||\tau|-|\xi||^2)^{\frac \theta2}\cdot |\hat{u}(\tau,\xi)|
\|_{L^2}
\label{standardxs}$$ where $\hat u=\hat{u}(\tau,\xi)$ is the space-time Fourier transform of function $u=u(t,x)$. Finally, in the most difficult case, $n\geq 5$, this was proved by Tataru [@MR1739207], using a suitable modification of the $X^{s,\theta}$ spaces, needed in order to control the interaction of high and low frequencies in the multiplicative estimates.
For the quasilinear problem (\[qua\]) the sharp result is only known to hold in dimensions $n=2,3$. This was proved by Smith-Tataru [@MR2178963] (see also Lindblad’s counterexample [@MR1666844]). The argument there still requires the use of Strichartz estimates. These are derived from a wave packet parametrix construction for a wave equation with very rough coefficients, which in turn is obtained via a very delicate analysis of the Hamilton flow. A different proof of this result in the special case of the Einstein vacuum equation was independently obtained by Klainerman-Rodnianski [@MR1885093], [@MR2180401], [@MR2052472]. In dimensions $n\geq 4$ it is still unclear which is the optimal threshold, the best results so far being contained in the above mentioned paper of Smith-Tataru [@MR2178963] and in an earlier one, Tataru [@MR1887639]: $$\aligned &n=4,5\quad \quad s>\frac{n}{2}+\frac{1}{2}\\
&n\geq 6\qquad \quad s>\frac{n}{2}+\frac{2}{3}
\endaligned$$ In the same direction but somewhat closer in spirit to the present paper is Bahouri and Chemin’s work [@MR2082388; @MR2003418]. The equation considered there is still quasilinear, but the main estimates are frequency localized versions of the Strichartz estimates for the wave equation on a rough background.
As an intermediate step toward understanding the higher dimensional quasilinear problem, we consider here the semilinear problem on a curved background and we prove the sharp result:
Let $n=4$ and assume that the coefficients $g^{ij}$ satisfy $\partial^2 g \in L^2 L^\infty$. Then the Cauchy problem , is locally well-posed in $H^s\times
H^{s-1}$ for $s>\frac{9}{4}$. \[main\]
Here well-posedness is understood in the strongest sense, i.e. the solutions have Lipschitz dependence on the initial data and they exist on a time interval which only depends on the size of the initial data.
One contribution of the present paper is to introduce variable coefficient versions of the $X^{s,b}$ spaces, study their properties and obtain the corresponding Strichartz type embeddings. However, the main novelty, contained in the last two sections, is a new method, based on a trilinear wave packet decomposition, to prove a key trilinear bound which cannot be obtained directly from the Strichartz estimates.
The first step in the proof is to reduce the problem to the case when the initial data is small, using scaling and the finite speed of propagation. This is a routine argument for which we refer the reader to [@MR2178963]. Once we know that the initial data is small, we can fix the time interval and set it to $[-1,1]$. This will be the case throughout the rest of the paper.
To solve the problem for small data we use a fixed point argument. Let $S(u_0,u_1)$ and $\Box_{g}^{-1}$ be respectively the homogeneous and inhomogeneous solution operators $$\begin{aligned}
&\Box_{g}S(u_0,u_1)\,=\,0,\quad S(u_0,u_1)(0)\,=\,u_0,\quad
\partial_t S(u_0,u_1)(0)\,=\,u_1
\label{hom}\\
&\Box_{g}(\Box_{g}^{-1}H)\,=\,H,\qquad
(\Box_{g}^{-1}H)(0)\,=\,0,\qquad
\partial_t(\Box_{g}^{-1}H)(0)\,=\,0
\label{inhom}\end{aligned}$$ Then a solution $u$ for (\[eq\]) in $[-1,1]$ is also a fixed point for the functional $$F(u)\,= S(u_0,u_1) + \Box_{g}^{-1}( \Gamma(u)(\nabla
u)^2)
\label{funct}$$
In order to apply a fixed point argument for $F$ we need to find two Banach spaces $X$ and $Y$ for which the following mapping properties hold: $$\begin{aligned}
&\,\| S(u_0,u_1) \|_X \lesssim \|(u_0,u_1)\|_{H^s\times H^{s-1}}
\label{h}\\
&\,\|\Box_g^{-1}H\|_X \lesssim \|H\|_Y
\label{i}\\
&\,\|u \cdot w\|_X \lesssim \|u\|_X \|w\|_X
\label{xx}\\
&\|\Gamma(u)\|_{X} \lesssim C(\|u\|_{L^\infty}) (1 + \|u\|_{X}^5)
\label{moser} \\
&\,\|u \cdot w\|_Y \lesssim \|u\|_X \|w\|_Y
\label{xy} \\
&\,\|\nabla v\cdot \nabla w\|_Y\lesssim \|v\|_X\cdot \|w\|_X
\label{n}\end{aligned}$$ where $C=C(\|u\|_{L^\infty})$ is a constant that depends solely on $\|u\|_{L^\infty}$. In the flat case (\[sem\]), for dimension $n=4$, one can make this argument work by choosing $$X\,=\,X^{s,\theta}\qquad Y\,=\,X^{s-1,\theta-1}$$ with $$s\,=\,\theta\,+\frac 32\qquad \theta\,>\,\frac 34 \label{st}$$
For our problem the challenge is twofold: first we need to find suitable variable coefficient versions for the $X^{s,\theta}$ spaces and then, in this new context, prove the corresponding estimates (\[h\])-(\[n\]).
Such spaces were previously introduced by Tataru [@MR1391526], where they are used in the context of a unique continuation problem. There, for a hyperbolic operator $P$ one defines $$X^{s,0}\,=\,H^s,\quad X^{s,1}\,=\,\{u\in H^s| Pu\in H^{s-1}\}$$ Then all the other spaces are defined through interpolation and duality.
In this article we choose to follow a different path based on dyadic decompositions with respect to the spatial frequency and the distance to the characteristic cone. Likely one should be able to prove that the two approaches are equivalent, but we choose not to pursue this here.
Our article is structured as follows. In the next section we define the $X^{s,\theta}$ spaces and prove that they satisfy the linear estimates , . Our definition of the $X^{s,\theta}$ is slightly different from the standard one in the constant coefficient case. Precisely, in the constant coefficient case our definition gives $$\| u \|_{X^{s,\theta}} \approx \| (1+|\xi|^2)^{\frac s2}
(1+||\tau|-|\xi||^2)^{\frac \theta2}\cdot \hat{u}(\tau,\xi)
\|_{L^2} + \|\Box u\|_{L^2_t H^{s+\theta-2}_x}, \quad 0 < \theta <
1 \label{newxs}$$ and one can see that the second term above alters the behavior at high modulations $|\tau| \gg |\xi|$. Correspondingly, for negative $\theta$ we have $$\| u \|_{X^{s,\theta}} \approx \| (1+|\xi|^2)^{\frac s2}
(1+||\tau|-|\xi||^2)^{\frac \theta2}\cdot \hat{u}(\tau,\xi)
\|_{L^2} + \| u\|_{L^2_t H^{s+\theta}_x}, \quad -1 < \theta < 0
\label{newxs1}$$ This change is consistent with scaling and simplifies somewhat the study of high modulation interactions.
In Section \[secse\] we discuss the Strichartz estimates for $\Box_g$, which translate into embeddings for the $X^{s,\theta}$ spaces. These turn out to suffice for the proof of the algebra properties - and for the high-high frequency interactions in .
The difficult part is to study the high-low frequency interactions in . For this we first take advantage of the duality relation $$(X^{s,\theta}+ L^2 H^{s+\theta})'\,=\,X^{-s,-\theta} \label{dual}
\qquad s \in {\mathbb R}, \quad 0 < \theta < \frac 12$$ This is consistent with and . Using this duality, after factoring out high modulation interactions, the bound (\[n\]) is transformed into the trilinear estimate: $$\left|\int u\cdot v\cdot w\,\,dx \,dt\right|\lesssim \|u\|_{X^{1-s,1-\theta}}
\|v\|_{X^{s-1,\theta}} \|w\|_{X^{s-1,\theta}}
\label{trilin}$$ with $(s,\theta)$ verifying (\[st\]). The last section of the paper is devoted to proving this bound. The argument is based on a multiscale trilinear wave packet decomposition for linear waves.
The $X^{s,\theta}$ spaces
=========================
We first introduce Littlewood-Paley decompositions. As a general rule, all frequency localizations in the sequel are only with respect to the spatial variables. There is a single exception to this. Precisely, the coefficients $g^{ij}$ are truncated using space-time multipliers. In order for these truncations to work, we need for these coefficients to be defined globally in time. Hence we assume they have been extended to functions with similar properties in all of ${\mathbb R}^{n+1}$.
Let $\phi$ be a smooth function supported in $\{\frac12 \leq |\xi|
\leq 2 \}$ with the property that $$1 = \sum_{j=-\infty}^\infty \phi(2^{-j} \xi)$$ We consider a spatial Littlewood-Paley decomposition, $$1 = \sum_{{\lambda}=1}^\infty S_{{\lambda}}(D_x)$$ where for dyadic ${\lambda}> 1$ we have $$\qquad S_{{\lambda}}(\xi) = \phi\left(\frac{\xi}{{\lambda}}\right)$$ while $S_1$ incorporates the low frequency contribution in $\{|\xi|\leq 1\}$. Set $$S_{<{\lambda}} = \sum_{\mu=1}^{\frac{\lambda}2} S_\mu$$ We will also use spatial multipliers ${{\tilde S}}_{\lambda}$ with slightly larger support, with $S_{\lambda}{{\tilde S}}_{\lambda}= S_{\lambda}$. We say that a function $u$ is localized at frequency ${\lambda}$ if its Fourier transform is supported in the annulus $\{ \frac{\lambda}8 \leq |\xi| \leq 8{\lambda}\}$.
For the paradifferential type calculus we also need to truncate the coefficients of $\Box_g$ in frequency. Given $\Box_g$ in (\[eq\]) we define the modified operators $$\Box_{g_{<{\lambda}}} =(S_{<{\lambda}}(D_x,D_t) g^{\alpha\beta}) \partial_\alpha
\partial_\beta$$ In the sequel we omit the space and time variables in our function space notations, i.e. $L^p:= L^p_{x,t}$, $L^2 H^s:= L^2_t H^s_x$, $L^p
L^q:= L^p_t L^q_x$, etc. We are ready now to define our spaces:
Let $\theta\in (0,1)$ and $s \in {\mathbb R}$. Then $X^{s,\theta}$ is the space of functions $u \in L^2(-1,1; H^s({\mathbb R}^n))$ for which the following norm is finite: $$\|u\|^2_{X^{s,\theta}}\,=\,\inf \left \{ \sum_{{\lambda}=1}^\infty
\sum_{d=1}^\lambda \|u_{\lambda,d}\|^2_{X_{{\lambda},d}^{s,\theta}} ;\ u
=\sum_{{\lambda}=1}^\infty\sum_{d=1}^{{\lambda}}S_{{\lambda}}
u_{{\lambda},d}\right\}
\label{tp}$$ where ${\lambda}$, $d$ take dyadic values and $$\|u_{{\lambda},d}\|_{X_{{\lambda},d}^{s,\theta}}^2 =
{\lambda}^{2s}d^{2\theta}\|u_{{\lambda},d}\|^2_{L^2} +
{\lambda}^{2s-2}d^{2\theta-2}\|\Box_{g_{<{\sqrt{\lambda}}}}u_{{\lambda},d}\|^2_{L^2}
\label{cl}$$
We also define the space $X^{s-1,\theta-1}$ of functions for which the following norm is finite: $$\begin{split}
\| f\|_{X^{s-1,\theta-1}}^2 = \inf \left\{ \|f_0\|_{L^2 H^{s-1}}^2
+ \sum_{{\lambda}=1}^\infty\right.& \sum_{d=1}^\lambda
\|f_{\lambda,d}\|^2_{X_{{\lambda},d}^{s,\theta}} ;\ \\ & \left. f =f_0+
\sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}}\Box_{g_{<{\sqrt{\lambda}}}} S_{{\lambda}}
f_{{\lambda},d} \right\}
\end{split}
\label{tn}$$ \[xst\]
Intuitively $d$ stands for the modulation of the $u_{{\lambda},d}$ piece. Indeed, in the constant coefficient case one can easily see that $u_{\lambda,d}$ mainly contributes to $u$ in the region where $||\tau|-|\xi|| \approx d$. The condition $1 \leq d$ is related to the spatial localization on the unit scale in our problem. The condition $d \leq \lambda$ reflects the fact that at high modulation we use a simpler structure, see e.g. , .
The cutoff at frequency less than $\sqrt{\lambda}$ for the coefficients $\Box_g$ is related to the regularity of the coefficients, $\partial^2 g \in L^2 L^\infty$. This implies that $\Box_{g_{\geq{\sqrt{\lambda}}}}u_{{\lambda},d}$ is an allowable error term.\
We begin our analysis of the $X^{s,\theta}$ spaces with a simple observation, namely that without any restriction in generality one can assume that the functions $u_{{\lambda},d}$ and $f_{{\lambda},d}$ in Definition \[xst\] are localized at frequency ${\lambda}$. Precisely, we have the stronger result:
The following estimate holds: $${\lambda}^{s-1}d^{\theta}\|\nabla S_{\lambda}v\|_{L^2} +
{\lambda}^{s-1}d^{\theta-1}\|\Box_{g_{<{\sqrt{\lambda}}}}S_{\lambda}v \|_{L^2}
\lesssim \|v\|_{X_{{\lambda},d}^{s,\theta}} \label{strongst}$$
We first bound the time derivatives of $v$ in negative Sobolev spaces, $${\lambda}^{s} d^{\theta} (\| {\partial}_t^2 v\|_{L^2 (H^{-2}+{\lambda}^2 L^2)} + \|{\partial}_t
v\|_{L^2(H^{-1}+{\lambda}L^2)} ) \lesssim \|v\|_{X_{{\lambda},d}^{s,\theta}}
\label{lowv}$$ This follows by Cauchy-Schwartz from the interpolation inequality $$\|{\partial}_t v\|_{L^2 ( H^{-1}+{\lambda}L^2)}^2 \lesssim (\| {\partial}_t^2 v\|_{L^2 (H^{-2}+{\lambda}^2L^2)}
+\|v\|_{L^2})\|v\|_{L^2}$$ combined with the bound $$\| {\partial}_t^2 v\|_{L^2 (H^{-2}+{\lambda}^2 L^2)} \lesssim {\lambda}^{-2} \|
\Box_{g_{<{\sqrt{\lambda}}}} v\|_{L^2} + \|{\partial}_t v\|_{L^2 ( H^{-1}+{\lambda}L^2)} + \|v\|_{L^2}$$ To prove this last estimate we only use the $L^\infty$ regularity of $g$ together with the condition $g^{00} = 1$. Then we need the fixed time bounds $$\| g_{<{\sqrt{{\lambda}}}} {\partial}_x {\partial}_t v\|_{H^{-2}+{\lambda}^2 L^2} \lesssim
\|{\partial}_t v\|_{H^{-1} + {\lambda}L^2}$$ $$\| g_{<{\sqrt{{\lambda}}}} {\partial}_x^2 v\|_{ H^{-2}+{\lambda}^2 L^2} \lesssim
\|v\|_{L^2}$$ They are similar, so we only discuss the second one. We write $$g_{<{\sqrt{{\lambda}}}} {\partial}_x^2 v = {\partial}_x^2 (g_{<{\sqrt{{\lambda}}}} v) - 2{\partial}_x
({\partial}_x g_{<{\sqrt{{\lambda}}}} v) + {\partial}_x^2 g_{<{\sqrt{{\lambda}}}} v$$ and use the uniform bounds $$|g_{<{\sqrt{{\lambda}}}}| \lesssim 1, \qquad |{\partial}_x g_{<{\sqrt{{\lambda}}}}|
\lesssim {\lambda},
\qquad |{\partial}_x^2 g_{<{\sqrt{{\lambda}}}}| \lesssim {\lambda}^2$$ This concludes the proof of .
The first term in is directly bounded using . For the second it suffices to prove the commutator estimate $$\|[\Box_{g_{<{\sqrt{\lambda}}}},S_{\lambda}] v\|_{L^2} \lesssim {\lambda}\|v\|_{L^2}+
\|{\partial}_t v\|_{ L^2} \label{smallgcom}$$ We have $$[\Box_{g_{<{\sqrt{\lambda}}}},S_{\lambda}] = [g_{<{\sqrt{{\lambda}}}},S_{\lambda}] {\partial}_t {\partial}_x +
[g_{<{\sqrt{{\lambda}}}},S_{\lambda}] {\partial}_x^2$$ and the commutators are localized at frequency ${\lambda}$ so the spatial derivatives only contribute factors of ${\lambda}$. Hence follows from the standard commutator estimate $$\|[g_{<{\sqrt{{\lambda}}}},S_{\lambda}]\|_{L^2 \to L^2} \lesssim {\lambda}^{-1}
\|\nabla g\|_{L^\infty}$$
Applying the above Lemma with $S_{\lambda}$ replaced by ${{\tilde S}}_{\lambda}$ we obtain
One can replace the $X^{s,\theta}_{{\lambda},d}$ norm in the definition of $X^{s,\theta}$ and $X^{s-1,\theta-1}$ by the norm $$\| v\|_{\tilde{X}^{s,\theta}_{{\lambda},d}} = {\lambda}^{s-1}d^{\theta}\|\nabla
v\|_{L^2} + {\lambda}^{s-1}d^{\theta-1}\|\Box_{g_{<{\sqrt{\lambda}}}} v\|_{L^2}$$ \[tx\]
For the proof of the duality relation it is convenient to work with a selfadjoint operator. Thus we consider the selfadjoint counterpart $\tilde{\Box}_g$ of $\Box_g$ $$\tilde{\Box}_g = \partial_i g^{ij} \partial_j$$ Then for $v$ localized at frequency $\lambda$ we commute and estimate the frequency localized difference $$\| \tilde \Box_{g_{<{\sqrt{\lambda}}}} v - \Box_{g_{<{\sqrt{\lambda}}}} v\|_{L^2}
\lesssim \|\nabla v\|_{L^2}$$ This leads directly to
One can replace the $\Box_{g_{<{\sqrt{\lambda}}}} $ operator in the definition of $X^{s,\theta}$ and $X^{s-1,\theta-1}$ by the similar operator in divergence form $\tilde\Box_{g_{<{\sqrt{\lambda}}}} $. \[self\]
As a consequence of the second part of we have
The following embedding holds for $-1 < \theta < 0$: $$X^{s,\theta} \subset L^2 H^{s+\theta}$$ \[sw\]
Another use of this is to establish energy estimates. A direct application of energy estimates for the wave equation yields the bound $$\| \nabla v\|_{L^\infty L^2}^2 \lesssim \|\nabla v\|_{L^2}^2 +
\|\nabla v\|_{L^2} \| \Box_{g} v\|_{L^2}$$ This leads to $${\lambda}^{s-1} d^{\theta-\frac12} \| \nabla S_{\lambda}v\|_{L^\infty L^2}
\lesssim \| v\|_{\tilde{X}^{s,\theta}_{{\lambda},d}} \label{litwo}$$ Going back to Definition \[xst\], this implies
Assume that $\theta > \frac12$. Then $$\|u\|_{L^\infty H^s} + \|u_t\|_{L^\infty H^{s-1}} \lesssim \|u \|_{X^{s,\theta}}$$ \[trace\]
To prove the estimates and in the context of the $X^{s,\theta}$ spaces we need to switch from the frequency truncated coefficients to the full coefficients $g^{ij}$. The tool needed to do that is contained in the following:
Assume that $0 \leq s \leq 3$. Then the following fixed time estimate holds: $$\sum_{{\lambda}=1}^\infty \lambda^{2(s-1)}\|{{\tilde S}}_{\lambda}({g_{> \sqrt{{\lambda}}}} u)
\|_{L^2}^2 \lesssim (M (\|{\partial}^2 g\|_{L^\infty}))^2 \| u\|_{
H^{s-2}}^2 \label{largeg}$$ where $M$ stands for the maximal function with respect to time. We also have the dual estimate $$\| \sum_{{\lambda}=1}^\infty {g_{> \sqrt{{\lambda}}}} {{\tilde S}}_{\lambda}f_{\lambda}\|_{H^{2-s}}^2 \lesssim \sum_{{\lambda}=1}^\infty \lambda^{2(1-s)}\| f_{\lambda}\|_{L^2}^2 \label{largegdual}$$
We take a Littlewood-Paley decomposition of both factors, $${{\tilde S}}_{\lambda}(g_{> \sqrt{{\lambda}}} u) = \sum_{\mu=1}^\infty \sum_{\nu =
\sqrt{{\lambda}}}^\infty {{\tilde S}}_{\lambda}( g_\nu u_\mu)$$ The $(\mu,\nu)$ term is nonzero only in the following situations:
\(i) $\nu \ll {\lambda}$, $\mu \approx {\lambda}$. Then we estimate $$\| {{\tilde S}}_{\lambda}(g_\nu u_\mu)\|_{L^2} \lesssim \|g_\nu\|_{L^\infty}
\|u_\mu\|_{ L^2} \lesssim \nu^{-2} M (\|{\partial}^2 g\|_{L^\infty})
\|u_\mu\|_{ L^2}$$ and use the square summability with respect to ${\lambda}$ together with the relation $\nu^{-2} \lesssim {\lambda}^{-1}$.
\(ii) $\nu \approx {\lambda}$, $\mu \ll {\lambda}$. Then $$\| {{\tilde S}}_{\lambda}(g_\nu u_\mu)\|_{L^2} \lesssim \|g_\nu\|_{L^\infty}
\|u_\mu\|_{ L^2} \lesssim {\lambda}^{-2} M (\|{\partial}^2 g\|_{L^\infty})
\|u_\mu\|_{ L^2}$$ This is tight only when $s=3$ and $\mu=1$, otherwise there is a gain which insures the summability in ${\lambda}$, $\mu$.
\(iii) $\nu \approx \mu \gtrsim {\lambda}$. Then $$\| {{\tilde S}}_{\lambda}(g_\nu u_\mu)\|_{L^2} \lesssim \|g_\nu\|_{L^\infty}
\|u_\mu\|_{ L^2} \lesssim \mu^{-2} M (\|{\partial}^2 g\|_{L^\infty}) \|
u_\mu\|_{ L^2}$$ This is always stronger than we need. The proof of the lemma is concluded.
We now establish some simple properties of the linear equation $$\Box_g u = f, \qquad u(0) = u_0, \qquad u_t(0) = u_1.
\label{leq}$$ Then
The linear equation is well-posed in $H^s \times
H^{s-1}$ for $0 \leq s \leq 3$. \[lwp\]
The proof follows easily from energy estimates, see [@MR2153517].
We use this to prove , namely
Assume that $0 \leq s \leq 3$ and $\theta > 0$. Then the solution $u$ to verifies $$\| u \|_{X^{s,\theta}} \lesssim \|u_0\|_{H^s} +
\|u_1\|_{H^{s-1}} + \|f\|_{L^2 H^{s-1}}$$ \[nhom\]
We decompose the solution $u$ as $$u = \sum_{{\lambda}=1}^\infty S_\lambda {{\tilde S}}_{\lambda}u$$ and think of this as a part of the sum in which corresponds to $d=1$. Then $$\begin{split}
\|u\|_{X^{s,\theta}}^2 &\lesssim \sum_{{\lambda}=1}^\infty \|{{\tilde S}}_{\lambda}u\|_{X^{s,\theta}_{\lambda,1}}^2 \\
& \approx \sum_{{\lambda}=1}^\infty {\lambda}^{2s} \| {{\tilde S}}_{\lambda}u\|^2_{L^2} +
\lambda^{2(s-1)} \|\Box_{g_{<{\sqrt{\lambda}}}} {{\tilde S}}_{\lambda}u\|^2_{L^2}
\\ &\lesssim \| u\|^2_{L^2 H^s} + \sum_{{\lambda}=1}^\infty
\lambda^{2(s-1)} \|\Box_{g_{<{\sqrt{\lambda}}}} {{\tilde S}}_{\lambda}u - {{\tilde S}}_{\lambda}\Box_g
u \|^2_{L^2} + \|f\|^2_{L^2 H^{s-1}}
\end{split}$$ The first term is easily controlled by energy estimates. The second is decomposed as follows: $$\Box_{g_{<{\sqrt{\lambda}}}} {{\tilde S}}_{\lambda}u - {{\tilde S}}_{\lambda}\Box_g u =
[\Box_{g_{<{\sqrt{\lambda}}}},{{\tilde S}}_{\lambda}] u - {{\tilde S}}_{\lambda}\Box_{g_{>{\sqrt{\lambda}}}} u$$ For the commutator we use the fixed time bound along with square summability in $\lambda$. The second part is controlled by .
The result in the next Lemma implies the estimate for the spaces $X,Y$:
Assume that $0 \leq s \leq 3$ and $\frac12 < \theta < 1$. Then the operator $\Box_g^{-1}$ has the mapping property $$\Box_g^{-1}: X^{s-1,\theta-1} \to X^{s,\theta}$$
Let $f \in X^{s-1,\theta-1}$. We use the representation in , $$f = f_0 + \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}}\Box_{g_{<{\sqrt{\lambda}}}}
S_{{\lambda}} f_{{\lambda},d}$$ By Definition the function $$u = \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} S_{{\lambda}} f_{{\lambda},d}$$ belongs to $X^{s,\theta}$. The difference $v = u - \Box_g^{-1} f$ solves $$\Box_g v = \Box_g u - f, \qquad v(0) = u(0), \qquad v_t(0) = u_t(0)$$ To estimate it we use Lemma \[nhom\]. The initial data is controlled due to Corollary \[trace\], so it remains to bound the inhomogeneous term in $L^2 H^{s-1}$. Thus we need to show that $$\| \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} \Box_{g_{>{\sqrt{\lambda}}}}
S_{{\lambda}} f_{{\lambda},d} \|_{L^2H^{s-1}}^2 \lesssim \sum_{{\lambda},d}
\|f_{{\lambda},d}\|_{X^{s,\theta}_{{\lambda},d}}^2$$ Considering the trace regularity result in Corollary \[trace\] this would follow from $$\| \sum_{{\lambda}=1}^\infty \Box_{g_{>{\sqrt{\lambda}}}} S_{{\lambda}} f_{{\lambda}}
\|_{L^2H^{s-1}}^2 \lesssim \sum_{{\lambda}} \|\nabla f_{{\lambda}}\|_{L^\infty
H^{s-1}}^2, \qquad f_{\lambda}= \sum_{d=1}^{\lambda}f_{{\lambda},d}$$ which in turn is a consequence of the fixed time bound .
We finish this section by proving a key duality relation between $X^{s,\theta}$ spaces with positive, respectively negative $\theta$.
For $ 0 < \theta < \frac12$ we have the duality relation $$X^{-s,-\theta}= (X^{s,\theta} + L^2 H^{s+\theta})'
\label{dual1}$$
a\) We first show that $$X^{-s,-\theta} \subset (X^{s,\theta} + L^2 H^{s+\theta})'$$ From Corollary \[sw\] we obtain $
X^{-s,-\theta} \subset (L^2 H^{s+\theta})' $. It remains to prove the bound $$\left|\int u\cdot f\,\, dx\,dt\right|\lesssim
\|u\|_{X^{s,\theta}}\,\|f\|_{X^{-s,-\theta}}$$ We consider Littlewood-Paley decompositions of $u$ and $v$ as in Definition \[xst\], $$u = \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} S_\lambda u_{\lambda,d},
\qquad f = f_0 + \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} \tilde \Box_{g_{<\sqrt{\lambda}}}
S_\lambda f_{\lambda,d}$$ with $\tilde \Box_{g_{<\sqrt{\lambda}}}$ in divergence form, see Corollary \[self\]. The summation with respect to ${\lambda}$ is essentially diagonal therefore it follows by orthogonality. To handle the $d$ summation it suffices to obtain the off-diagonal decay $$\begin{split}
\left | \int S_{\lambda}u_{\lambda,d_1} \cdot\tilde\Box_{g_{<\sqrt{\lambda}}}
S_{\lambda}f_{\lambda,d_2} dx dt\right| \lesssim &
\min\left\{\left(\frac{d_2}{d_1}\right)^\theta,
\left (\frac{d_1}{d_2}\right)^{\frac12-\theta}\right\}
\\ & \|u_{\lambda,d_1}\|_{X^{s,\theta}_{\lambda,d_1}
}\,\|f_{\lambda,d_2}\|_{X^{1-s,1-\theta}_{\lambda,d_2}}
\end{split}$$ If $d_2 < d_1$ then this follows directly from and . Otherwise we integrate by parts $$\begin{split}
\int S_{\lambda}u_{\lambda,d_1}\cdot \, &\tilde\Box_{g_{<\sqrt{\lambda}}} S_{\lambda}f_{\lambda,d_2} dx
dt = \int\tilde \Box_{g_{<\sqrt{\lambda}}} S_{\lambda}u_{\lambda,d_1} \cdot
S_{\lambda}f_{\lambda,d_2} dx dt \\ + & \left. \int (S_{\lambda}u_{\lambda,d_1}\cdot g^{0\alpha}_{<\sqrt{{\lambda}}}\partial_\alpha S_{\lambda}f_{\lambda,d_2} -
g^{0\alpha}_{<\sqrt{{\lambda}}}\partial_\alpha S_{\lambda}u_{\lambda,d_1} \cdot S_{\lambda}f_{\lambda,d_2}) dx \right|_{-1}^1
\end{split}$$ For the first term we use and . For the second we use the trace regularity result in .
b\) We now show that $$(X^{s,\theta}+ L^2 H^{s+\theta})' \subset X^{-s,-\theta}$$ Let $T$ be a bounded linear functional on $X^{s,\theta} + L^2
H^{s+\theta}$. Due to the second term we can identify $T$ with a function $u \in L^2 H^{-s-\theta}$.
On the other hand, we can apply it to functions $v \in X^{s,\theta}$ of the form $$v = \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} S_\lambda v_{\lambda,d}$$ Then we must have the bound $$|Tv|^2 \lesssim \|v\|_{X^{s,\theta}}^2 \lesssim \sum_{{\lambda},d}
\|v_{\lambda,d}\|^2_{X^{s,\theta}_{\lambda,d}} \lesssim
\sum_{{\lambda},d} \left( {\lambda}^{2s}d^{2\theta}\|v_{{\lambda},d}\|^2_{L^2} +
{\lambda}^{2s-2}d^{2\theta-2}\|\tilde
\Box_{g_{<{\sqrt{\lambda}}}}v_{{\lambda},d}\|^2_{L^2}\right)$$ Given the definition of the $X^{s,\theta}_{\lambda,d}$ norms, using succesively the Hahn-Banach theorem and Riesz’s theorem it follows that we can find functions $f_{\lambda,d}$ and $h_{\lambda,d}$ with $$\sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} \lambda^{-2s} d^{-2\theta}
\|f_{\lambda,d}\|_{L^2}^2
+ \lambda^{2(1-s)} d^{2(1-\theta)} \|h_{\lambda,d}\|_{L^2}^2 = M < \infty
\label{M}$$ so that $$Tv = \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} \int f_{\lambda,d}\,
v_{\lambda,d} + h_{\lambda,d}\cdot \tilde \Box_{g_{<
\sqrt\lambda}}v_{\lambda,d}\, dx dt$$ In particular this must hold for $v$ of the form $v = S_\lambda
v_{\lambda,d}$, $$\int u \, S_\lambda v_{\lambda,d} dx dt = \int f_{\lambda,d}\,
v_{\lambda,d} + h_{\lambda,d}\cdot \tilde \Box_{g_{<
\sqrt\lambda}}v_{\lambda,d} \, dx dt$$ For each $\lambda,d$ this yields $$S_\lambda u = f_{\lambda,d} + \tilde \Box_{g_{< \sqrt\lambda}} h_{\lambda,d}$$ Then we can represent $S_\lambda u$ in the form $$S_\lambda u = f_{\lambda,1} + \sum_{d=1}^{\frac\lambda 2}
\tilde \Box_{g_{< \sqrt\lambda}} u_{\lambda,d} + \tilde \Box_{g_{<
\sqrt\lambda}} h_{\lambda,\lambda} \qquad u_{\lambda,d} =
h_{\lambda,d} -h_{\lambda,2d}
\label{py}$$ This yields for $u$ the representation $$u = \sum_{\lambda = 1}^\infty \tilde S_\lambda \left( f_{\lambda,1} + \sum_{d=1}^{\frac\lambda 2}
\tilde \Box_{g_{< \sqrt\lambda}} u_{\lambda,d} + \tilde \Box_{g_{<
\sqrt\lambda}} h_{\lambda,\lambda}\right)
\label{pya}$$ This is very close to but not exactly the form in . However the multipliers $\tilde S_\lambda$ can be easily replaced by $S_\lambda$ by reapplying the Paley-Littlewood decomposition on the right, and then $S_\lambda$ can be commuted to the right of $\tilde \Box_{g_{<
\sqrt\lambda}}$ due to the Corollary \[tx\] and the commutator bound . Hence we have $$\| u\|_{X^{-s,-\theta} }^2 \lesssim \sum_{\lambda = 1}^\infty
\left(\lambda^{-2s} \|f_{\lambda,1} \|_{L^2}^2 +
\sum_{d=1}^{\lambda/2}
\| u_{\lambda,d}\|_{X_{\lambda,d}^{1-s,1-\theta}}^2 +
\| h_{\lambda,\lambda}\|_{X_{\lambda,\lambda}^{1-s,1-\theta}}^2\right)$$ and due to it remains to bound the right hand side by $$M + \| u\|_{L^2 H^{-s-\theta}}^2$$ There is nothing to do for the $f_{\lambda,1}$ term. On the other hand we can bound $$\begin{split}
\|u_{\lambda,d}\|_{X^{1-s,1-\theta}_{\lambda,d}}^2 \lesssim &\
\lambda^{2(1-s)} d^{2(1-\theta)} \| u_{\lambda,d}\|_{L^2}^2
+ \lambda^{-2s} d^{-2\theta} \| \tilde \Box_{g_{<
\sqrt\lambda}} u_{\lambda,d}\|_{L^2}^2 \\
= &\ \lambda^{2(1-s)} d^{2(1-\theta)} \| h_{\lambda,d} - h_{\lambda,2d}\|_{L^2}^2
+ \lambda^{-2s} d^{-2\theta} \| f_{\lambda,d} - f_{\lambda,2d}\|_{L^2}^2
\\
\lesssim &\ \lambda^{-2s} d^{-2\theta}
(\|f_{\lambda,d}\|_{L^2}^2+ \|f_{\lambda,2d}\|_{L^2}^2)
\\ &\ + \lambda^{2(1-s)} d^{2(1-\theta)} (\|h_{\lambda,2d}\|_{L^2}^2
+\|h_{\lambda,d}\|_{L^2}^2)
\end{split}$$ Finally, for the last term we have $$\begin{split}
\| h_{\lambda,\lambda}\|_{X_{\lambda,\lambda}^{1-s,1-\theta}}^2
\lesssim &\ \lambda^{2(1-s)} \lambda^{2(1-\theta)} \| h_{\lambda,\lambda}\|_{L^2}^2
+ \lambda^{-2s} \lambda^{-2\theta} \| \tilde \Box_{g_{<
\sqrt\lambda}} h_{\lambda,\lambda}\|_{L^2}^2
\\ = &\ \lambda^{2(1-s)} \lambda^{2(1-\theta)} \| h_{\lambda,\lambda}\|_{L^2}^2
+ \lambda^{-2s} \lambda^{-2\theta} \| S_\lambda u - f_{\lambda,\lambda}\|_{L^2}^2
\\ \lesssim&\ \lambda^{2(1-s)} \lambda^{2(1-\theta)} \|
h_{\lambda,\lambda}\|_{L^2}^2 + \lambda^{-2s} \lambda^{-2\theta} \| f_{\lambda,\lambda}\|_{L^2}^2
+ \| S_\lambda u\|_{L^2 H^{-s-\theta}}^2
\end{split}$$ The proof is concluded.
Strichartz estimates and applications. {#secse}
======================================
The Strichartz estimates for the variable coefficient wave equation, as proved in [@MR1887639], have the form:
(Tataru [@MR1887639]) Assume that the coefficients $g^{ij}$ of $\Box_g$ satisfy ${\partial}^2 g^{ij} \in L^1L^\infty$. Then the solutions to the wave equation in $n+1$ dimensions satisfy the bounds $$\| D^{\sigma} \nabla u\|_{L^p L^q} \lesssim \|u(0)\|_{H^1} + \|u_t(0)\|_{L^2} +
\|\Box_g u\|_{L^1 L^2}
\label{se}$$ where $$\sigma = -\frac{n}2 + \frac{1}p +\frac{n}q, \qquad \frac{2}p +\frac{n-1}q
\leq \frac{n-1}2, \qquad 2 \leq p \leq \infty, \ \ 2 \leq q < \infty
\label{pq}$$
Applying this bound on an interval $I$ of size $\epsilon^2$ we obtain by Cauchy-Schwartz $$\| D^{\sigma} \nabla u\|_{L^p(I; L^q)} \lesssim \frac{1}\epsilon \|u\|_{H^1(I
\times {\mathbb R}^n)} + \epsilon \|\Box_g u\|_{L^2(I \times {\mathbb R}^n)}, \qquad
\epsilon \leq 1$$ Summing up over small intervals this extends to intervals of arbitrary lengths. Optimizing over $\epsilon$ yields $$\| D^{\sigma} \nabla u\|_{L^p L^q}^2 \lesssim \| u\|_{H^1}^2 +
\| u\|_{H^1} \|\Box_g u\|_{L^2}$$ We want to apply this result to the functions $S_{\lambda}u_{{\lambda},d}$ in Definition \[xst\]. By we obtain
a\) Let $(\sigma,p,q)$ verifying $$\sigma = -\frac{n}2 + \frac{1}p +\frac{n}q, \qquad 2 \leq p \leq
\infty, \ \ 2 \leq q < \infty$$ Then for $(\sigma,p,q)$ as in we have $$\| S_{\lambda}\nabla u\|_{L^p L^q} \lesssim \lambda^{1-s-\sigma}
d^{\frac12-\theta} \|u\|_{X^{s,\theta}_{{\lambda},d}}$$ If additionally $\theta > \frac12$ then $$\| S_{\lambda}\nabla u\|_{L^p L^q} \lesssim \lambda^{1-s-\sigma}
\|u\|_{X^{s,\theta}}$$ b) If instead $$\frac{2}p +\frac{n-1}q \geq \frac{n-1}2$$ then $$\| S_{\lambda}\nabla u\|_{L^p L^q} \lesssim \lambda^{1-
s-\sigma+\frac12(\frac{2}p +\frac{n-1}q - \frac{n-1}2)}
d^{\frac12-\theta-\frac12(\frac{2}p +\frac{n-1}q - \frac{n-1}2)}
\|u\|_{X^{s,\theta}_{{\lambda},d}}$$
\[sob\]
The interesting triplets of indices for $(\sigma,p,q)$ in $4+1$ dimensions are $$(0,\infty,2) \text{(energy)} \qquad (-\frac12,\frac{10}3,\frac{10}3)
\text{(Strichartz)} \qquad (-\frac56,2,6) \text{(Pecher)}$$ In addition, we can also use the index $q=\infty$. Thus we obtain the triplets $$(-2,\infty,\infty), \qquad (-\frac32,2,\infty)$$ For the case when $\theta<\frac12$, we rely on the additional triplets $$(-\frac16,2,3),\qquad (\frac14,4,2)$$
For convenience we summarize the bounds we need for $\tilde
X^{s,\theta}_{{\lambda},d}$:
For $0 < \theta < 1$ we have $${\lambda}^{s-1} \|S_{\lambda}\nabla u\|_{L^\infty L^2} + {\lambda}^{s-\frac52} \|S_{\lambda}\nabla u\|_{L^2
L^\infty} + {\lambda}^{s-3} \| S_{\lambda}\nabla u\|_{L^\infty} \lesssim
d^{\frac12-\theta} \|u\|_{\tilde X^{s,\theta}_{{\lambda},d}}$$ $${\lambda}^{s-\frac{17}{12}} \|S_{\lambda}\nabla u\|_{L^2
L^3} + {\lambda}^{s-1} \| S_{\lambda}\nabla u\|_{L^4 L^2} \lesssim
d^{\frac14-\theta} \|u\|_{\tilde X^{s,\theta}_{{\lambda},d}}$$ \[d\]
The reason we include the gradient is to have also bounds for $u_t$. Because of the frequency localization, if we drop the gradient the same bounds hold with one less power of ${\lambda}$.
In our estimates later on we also need to work with $X^{s,b}$ functions which are concentrated into a smaller modulation range. For this we introduce the additional norm $$\| u\|_{\tilde X^{s,\theta}_{{\lambda},<d}}^2 = \inf \left\{
\sum_{h=1}^d \|u_h\|_{\tilde X^{s,\theta}_{{\lambda},h}}^2;\ u = \sum_{h=1}^d u_h\right\}$$ If $d = {\lambda}$ we simply write $\tilde X^{s,\theta}_{{\lambda}}$. A simple argument leads to $$\|u\|^2_{X^{s,\theta}}\,=\,\inf \left \{ \sum_{{\lambda}=1}^\infty
\|S_{{\lambda}} u_{{\lambda}}\|_{\tilde
X^{s,\theta}_{{\lambda}}}^2 ;\ u =\sum_{{\lambda}=1}^\infty S_{{\lambda}}
u_{{\lambda}}\right\}$$ We also have
a\) Assume that $\theta > \frac12$. Then $${\lambda}^{s-1} \|S_{\lambda}\nabla u\|_{L^\infty L^2} + {\lambda}^{s-\frac52} \|S_{\lambda}\nabla u\|_{L^2
L^\infty} + {\lambda}^{s-3} \| S_{\lambda}\nabla u\|_{L^\infty} \lesssim \|u\|_{{{\tilde{X}}}^{s,\theta}_{{\lambda},<d}}$$
b\) Assume that $\theta < \frac12$. Then $${\lambda}^{s-1} \|S_{\lambda}\nabla u\|_{L^\infty L^2} + {\lambda}^{s-\frac52} \|S_{\lambda}\nabla u\|_{L^2
L^\infty} + {\lambda}^{s-3} \| S_{\lambda}\nabla u\|_{L^\infty} \lesssim
d^{\frac12-\theta} \|u\|_{\tilde X^{s,\theta}_{{\lambda},<d}}$$ \[lessd\]
In preparation for proving bilinear estimates for the $X^{s,\theta}$ spaces we first investigate which multiplications leave the ${{\tilde{X}}}^{s,\theta}_{{\lambda},d}$ space unchanged. For this we define the algebras $M_d$, $M_{<d}$ with the norms $$\|f\|_{M_d} = \| f\|_{L^\infty} + d^{-1} \|f_t\|_{L^\infty}
+ d^{-\frac12} \|f_t\|_{L^2 L^\infty} + d^{-\frac32} \|f_{tt}\|_{L^2 L^\infty}$$ $$\|f\|_{M_{<d}} = \|f\|_{M_d} + d^{\frac12} \| f\|_{L^2 L^\infty}$$ Then we have the multiplicative properties
Assume that $f$ is localized at frequency $d \leq {\lambda}$. Then we have $$\|f S_{\lambda}u\|_{\tilde X^{s,\theta}_{{\lambda},d}} \lesssim \|f\|_{M_d} \|u\|_{\tilde
X^{s,\theta}_{{\lambda},d}}$$ respectively $$\begin{split}
\|f S_{\lambda}u\|_{\tilde X^{s,\theta}_{{\lambda},d}} & \lesssim \|f\|_{M_{<d}} \|u\|_{\tilde
X^{s,\theta}_{{\lambda},<d}}, \qquad \theta < \frac12 \\
\|f S_{\lambda}u\|_{\tilde X^{s,\theta}_{{\lambda},d}} & \lesssim
d^{\theta-\frac12} \|f\|_{M_{<d}} \|u\|_{\tilde
X^{s,\theta}_{{\lambda},<d}}, \qquad \theta > \frac12
\end{split}$$ \[md\]
The proof is straightforward, using Leibnitz’s rule and the energy estimate . To bound functions in the $M_d$, respectively $M_{<d}$ norms we use Corollary \[lessd\] with $d = {\lambda}$ to obtain:
a\) Assume that $\theta > \frac12$. Then $$\| S_{\lambda}u\|_{M_{<{\lambda}}} \leq {\lambda}^{2-s} \|u\|_{\tilde
X^{s,\theta}_{{\lambda}}}\qquad \| S_{<{\lambda}} u\|_{M_{{\lambda}}} \leq
\max\{1,{\lambda}^{2-s}\} \|u\|_{X^{s,\theta}}$$ $$\| S_{<{\lambda}} u\|_{M_{<{\lambda}}} \leq
\max\{{\lambda}^ \frac12,{\lambda}^{2-s}\} \|u\|_{X^{s,\theta}}$$
b\) Assume that $\theta < \frac12$. Then $$\| S_{\lambda}u\|_{M_{<{\lambda}}} \leq {\lambda}^{\frac52-\theta-s} \|u\|_{\tilde
X^{s,\theta}_{{\lambda}}}\qquad \| S_{<{\lambda}} u\|_{M_{{\lambda}}} \leq
\max\{1,{\lambda}^{\frac52-\theta-s}\} \|u\|_{X^{s,\theta}}$$ $$\| S_{<{\lambda}} u\|_{M_{<{\lambda}}} \leq
\max\{{\lambda}^ \frac12,{\lambda}^{\frac52-\theta-s}\} \|u\|_{X^{s,\theta}}$$ \[xmd\]
Using the above property we prove the algebra property for the space $X$.
Assume that $s > 2$ and $\frac12 < \theta < s-\frac32$ . Then $X^{s,\theta}$ is an algebra. \[xxx\]
Let $u,v \in X^{s,\theta}$. For both we consider the decomposition in Definition \[xst\], $$u = \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} S_{\lambda}u_{{\lambda},d}, \qquad
v = \sum_{{\lambda}=1}^\infty \sum_{d=1}^{{\lambda}} S_{\lambda}v_{{\lambda},d},$$ For the terms in the decomposition we use the $\tilde{X}^{s,\theta}_{{\lambda},d}$ norms, as allowed by Corollary \[tx\]. We denote $$u_{\lambda}= \sum_{d=1}^{{\lambda}} u_{{\lambda},d}, \qquad u_{{\lambda},<d} = \sum_{h=1}^{d} u_{{\lambda},h}$$ Then we write $$uv = \sum_{\mu =1}^\infty S_{\mu} (uv) = \sum_{\mu =1}^\infty
\sum_{{\lambda}_1 =1}^\infty \sum_{{\lambda}_2 =1}^\infty S_{\mu} (S_{{\lambda}_1}u_{{\lambda}_1} S_{{\lambda}_2}v_{{\lambda}_2})$$ There are two cases when the above summand is nonzero, namely if ${\lambda}_1 \approx {\lambda}_2 \gtrsim \mu$ and if $\max\{{\lambda}_1,{\lambda}_2\} \approx \mu$. We consider them separately.
[**Case 1**]{}, $ {\lambda}_1,{\lambda}_2 \approx {\lambda}\gtrsim \mu$. In this case the summability with respect to ${\lambda}$ is trivial, so it suffices to look at the product $S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda}$ for fixed ${\lambda}$. This is localized at frequency $\leq {\lambda}$. Combining the $L^\infty L^2$ and the $L^2 L^\infty$ bounds in Corollary \[lessd\] we obtain $$\| S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda}\|_{L^2} +{\lambda}^{-1} \| {\partial}_t (S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda})\|_{L^2} \lesssim {\lambda}^{-2s+\frac32}
\|u_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}} \|v_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}}
\label{lla}$$ Using the equation we can also bound the second time derivative, $${\lambda}^{-2} \| {\partial}_t^2 (S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda})\|_{L^2} \lesssim
{\lambda}^{-2s+\frac32}
\|u_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}} \|v_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}}
\label{llb}$$ The three bounds above allow us to estimate for $\mu \leq {\lambda}$ $$\|S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda}\|_{X^{s,\theta}_{\mu,\mu}} \lesssim
\mu^{s+\theta-2} {\lambda}^{-2s+\frac72}
\|u_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}} \|v_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}}$$ This suffices provided that $\theta < s-\frac32$, which is insured by our hypothesis.
[**Case 2**]{}. Here we consider products of the form $S_\mu v_\mu S_{\lambda}u_{\lambda}$ where $\mu \ll {\lambda}$. Then the product is localized at frequency ${\lambda}$. The summation with respect to ${\lambda}$ is trivial, but not the one with respect to $\mu$. We write $$S_\mu v_\mu S_{\lambda}u_{\lambda}= S_\mu v_{\mu} S_{\lambda}u_{{\lambda},<\mu} +
\sum_{d = \mu}^{\lambda}S_\mu v_{\mu} S_{{\lambda}} u_{{\lambda},d}$$ Using Lemma \[md\] and Lemma \[xmd\] we obtain $$\begin{split}
\|S_{\lambda}u_{\lambda}S_\mu v_\mu&\|_{\tilde X^{s,\theta}_{{\lambda}}}^2 \lesssim
\|S_\mu v_{\mu} S_{\lambda}u_{{\lambda},<\mu}\|_{\tilde X^{s,\theta}_{{\lambda},\mu}}^2
+ \sum_{d = \mu}^{\lambda}\|S_\mu v_{\mu} S_{{\lambda}} u_{{\lambda},d}\|_{\tilde X^{s,\theta}_{{\lambda},d}}^2
\\ &\lesssim \mu^{2\theta-1}
\|S_\mu v_\mu\|^2_{M_{<\mu}} \|u_{{\lambda},<\mu}\|_{\tilde X^{s,\theta}_{{\lambda},<\mu}}^2
+ \|S_\mu v_\mu\|^2_{M_{\mu}} \sum_{d=\mu}^{\lambda}\|u_{{\lambda},d}\|_{\tilde
X^{s,\theta}_{{\lambda},d}}^2
\\ &\lesssim \mu^{2\theta-1}\|S_\mu
v_\mu\|^2_{M_{<\mu}}\sum_{d=1}^{\lambda}\|u_{{\lambda},d}\|_{\tilde X^{s,\theta}_{{\lambda},d}}^2
\\ &\lesssim \mu^{3+2\theta-2s} \|v_\mu\|^2_{ {{\tilde{X}}}^{s,\theta}_{\mu}}\sum_{d=1}^{\lambda}\|u_{{\lambda},d}\|_{\tilde X^{s,\theta}_{{\lambda},d}}^2
\end{split}$$ The summation with respect to $\mu$ is trivial since $\theta < s-\frac32$.
We next prove .
Assume that $s > 2$ and $\frac12 < \theta < s-\frac32$ . Then we have the multiplicative estimate $$X^{s,\theta} \cdot X^{s-1,\theta-1} \subset X^{s-1,\theta-1}$$
By duality this reduces to the multiplicative estimate $$X^{s,\theta} \cdot (X^{1-s,1-\theta}+L^2 H^{2-s-\theta}) \subset
X^{1-s,1-\theta} + L^2 H^{2-s-\theta}$$ Since $s > 2$ we have the fixed time multiplication $$H^s \cdot H^{2-s-\theta} \subset H^{2-s-\theta}$$ which implies the space-time bound $$L^\infty H^s \cdot L^2 H^{2-s-\theta} \subset L^2 H^{2-s-\theta}$$ Due to the energy estimate for $X^{s,\theta}$ it remains to show that $$X^{s,\theta} \cdot X^{1-s,1-\theta} \subset
X^{1-s,1-\theta} + L^2 H^{2-s-\theta}$$ We consider a product $S_{\lambda}u_{\lambda}S_\mu v_\mu$ which we decompose as in the previous proof. Because of the lack of symmetry we now need to consider three cases.
[**Case 1**]{}. Here we estimate $ S_\mu (S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda})$ where $\mu
\lesssim {\lambda}$. By Corollary \[lessd\] we obtain $$\|S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda}\|_{L^2 L^\frac{3}{2}} \lesssim \|S_{\lambda}u_{\lambda}\|_{L^4 L^3} \|S_{\lambda}v_{\lambda}\|_{L^4 L^3} \lesssim
{\lambda}^{\theta-\frac{2}{3}} \|u_{\lambda}\|_{\tilde X^{s,\theta}_{{\lambda}}}
\|v_{\lambda}\|_{\tilde
X^{1-s,1-\theta}_{{\lambda}}}$$
Using then Sobolev embeddings we obtain
$$\|S_\mu(S_{\lambda}u_{\lambda}S_{\lambda}v_{\lambda})\|_{L^2 H^{2-s-\theta}} \lesssim
\mu^{\frac 83 -s-\theta}{\lambda}^{\theta-\frac{2}{3}} \|u_{\lambda}\|_{\tilde
X^{s,\theta}_{{\lambda}}} \|v_{\lambda}\|_{\tilde
X^{1-s,1-\theta}_{{\lambda}}}$$
[**Case 2**]{}. Here we bound $S_\mu u_\mu S_{\lambda}v_{\lambda}$, $\mu \ll {\lambda}$. The product is localized at frequency ${\lambda}$, and the analysis is almost identical to Case 2 in Proposition \[xxx\].
[**Case 3**]{}. Here we bound $S_{\lambda}u_{\lambda}S_\mu v_\mu$, $\mu \ll {\lambda}$. The same argument applies, the only difference is that we gain some extra $\mu/{\lambda}$ factors.
We continue with the Moser estimates in , which follow from
Assume that $s > 2$ and $\frac12 < \theta < s-\frac32$ . Let $\Gamma$ be a smooth function. Then $$\|\Gamma(u)\|_{X^{s,\theta}} \lesssim C(\|u\|_{L^\infty}) (1 +
\|u\|_{X^{s,\theta}}^5)$$
We write $$\Gamma(u)-\Gamma(v) = (u-v) f(u,v)$$ and $$f(u,v) - f(x,y) = (u-x) g_1(u,v,x,y) + (v-y)g_2(u,v,x,y)$$ where $f$, $g_1$ and $g_2$ are smooth functions. Then we have $$\begin{split}
\Gamma(u) =& \Gamma(u_{1}) + \sum_{{\lambda}=1}^\infty \Gamma(u_{\leq 2{\lambda}}) - \Gamma(u_{\leq {\lambda}})
\\ =& \Gamma(u_{1}) + \sum_{{\lambda}=1}^\infty u_{2{\lambda}} f(u_{\leq 2{\lambda}},u_{\leq {\lambda}})
\\ = & \Gamma(u_{1}) + \sum_{{\lambda}=1}^\infty u_{2{\lambda}} [ f(u_{\leq 2},u_1) +
\sum_{\mu=2}^{\lambda}( f(u_{\leq 2\mu},u_{\leq \mu}) -f(u_{\leq
\mu},u_{\leq \mu/2}))]
\\ =& \Gamma(u_{1}) + \sum_{{\lambda}=1}^\infty u_{2{\lambda}} [ f(u_{\leq 2},u_1) +
\sum_{\mu=2}^{\lambda}(u_{2\mu}\, g_1(u_{\leq 2\mu},u_{\leq
\mu},u_{\leq
\mu/2}) \\ &+ u_\mu \, g_2(u_{\leq 2\mu},u_{\leq \mu},u_{\leq
\mu/2}))]
\end{split}$$ Hence we need to bound expressions of the form $$S_{\lambda}u_{\lambda}\, S_\mu v_\mu \, h(S_{<\mu} w), \qquad \mu \leq {\lambda}$$ There are two different cases to consider:
[**Case 1**]{}. $\mu \approx {\lambda}$. Then the product has the form $$S_{\lambda}u_{\lambda}\, S_{\lambda}v_{\lambda}\, h(S_{<{\lambda}} w)$$ The first product is localized at frequency ${\lambda}$ and can be estimated as in , . For the nonlinear expression we use Lemma \[xmd\] to obtain $$\| S_{<{\lambda}} w\|_{M_{\lambda}} \lesssim \|w\|_{X^{s,\theta}}$$ On one hand by the chain rule we obtain $$\| h(S_{<{\lambda}} w)\|_{M_{\lambda}} \lesssim C(\|w\|_{L^\infty}) (1 +
\|w\|_{X^{s,\theta}}^3) \label{mlx}$$ On the other hand because of the frequency localization we also have the improved high frequency bound $$\| {{\tilde S}}_\mu h(S_{<{\lambda}} w)\|_{M_\mu} \lesssim C(\|w\|_{L^\infty})
\left(\frac{\lambda}\mu\right)^N (1+ \|w\|_{X^{s,\theta}}^3), \qquad \mu
\gg {\lambda}\label{mlxmu}$$ Taking this into account and repeatedly using Leibnitz’s rule we get $$\begin{split}
& \| S_{\lambda}u_{\lambda}\, S_{\lambda}v_{\lambda}\, h(S_{<{\lambda}} w) \|_{X^{s,\theta}}^2 \\
&\lesssim \sum_{\mu=1}^\infty \| S_\mu ( S_{\lambda}u_{\lambda}\,S_{\lambda}v_{\lambda}\,
h(S_{<{\lambda}} w)) \|_{\tilde X^{s,\theta}_\mu}^2 \\ & \lesssim \sum_{\mu
\lesssim {\lambda}} \| S_{\lambda}u_{\lambda}\, S_{\lambda}v_{\lambda}\, h(S_{<{\lambda}} w) \|_{\tilde
X^{s,\theta}_\mu}^2 + \sum_{\mu \gg {\lambda}} \| S_{\lambda}u_{\lambda}\, S_{\lambda}v_{\lambda}\,
{{\tilde S}}_\mu h(S_{<{\lambda}} w)) \|_{\tilde X^{s,\theta}_\mu}^2 \\ &
\lesssim C(\|w\|_{L^\infty})
\left (\sum_{\mu \lesssim {\lambda}} \mu^{2s+2\theta-4} {\lambda}^{-4s + 7} +
\sum_{\mu \gg {\lambda}} {\lambda}^{2\theta + 3 - 2s} \left(\frac{\lambda}\mu\right)^N
\right) \|u_{\lambda}\|^2_{\tilde X^{s,\theta}_{\lambda}} \|v_{\lambda}\|^2_{\tilde
X^{s,\theta}_{\lambda}} (1+\|w\|_{X^{s,\theta}}^6) \\ & \lesssim
C(\|w\|_{L^\infty})\, {\lambda}^{2\theta + 3 - 2s} \, \|u_{\lambda}\|^2_{\tilde X^{s,\theta}_{\lambda}}
\|v_{\lambda}\|^2_{\tilde X^{s,\theta}_{\lambda}} (1+\|w\|_{X^{s,\theta}}^6)
\end{split}$$ This is trivially summable with respect to ${\lambda}$.
[**Case 2**]{}. $\mu \ll {\lambda}$. Then the product has the form $$S_{\lambda}u_{\lambda}\,S_\mu v_\mu\, h(S_{<\mu} w) =$$ $$\begin{split}
&S_{\lambda}u_{{\lambda},<\mu}\, S_\mu v_\mu\, S_{<\mu} h(S_{<\mu} w) +
\sum_{\mu \leq
d \ll {\lambda}} S_{{\lambda}} u_{{\lambda},<d}\, S_\mu v_\mu \, S_d h(S_{<\mu} w)
\\ & + \sum_{\mu \leq d \ll {\lambda}} S_{{\lambda}}
u_{{\lambda},d} \,S_\mu v_\mu \, S_{<d} h(S_{<\mu} w) + S_{\lambda}u_{\lambda}\, S_\mu
v_\mu\, S_{{\lambda}} h(S_{<\mu} w)
\\ & + \sum_{\nu \gg {\lambda}} S_{\lambda}u_{\lambda}\, S_\mu v_\mu\, S_{\nu} h(S_{<\mu} w)
= f_1+f_2 +f_3 + f_4 + f_5
\end{split}$$ For $f_1$ we use Lemma \[md\], Lemma \[xmd\] and to obtain $$\begin{split}
\|f_1\|_{\tilde X^{s,\theta}_{{\lambda},\mu}} &\lesssim
\|S_{\lambda}u_{{\lambda},<\mu}\, S_\mu v_\mu\|_{\tilde X^{s,\theta}_{{\lambda},\mu}}
\| h(S_{<\mu} w)\|_{M_\mu} \\ &\lesssim \mu^{\theta-\frac12}
\|u_{{\lambda},<\mu}\|_{\tilde X^{s,\theta}_{{\lambda},<\mu}} \| S_\mu
v_\mu\|_{M_{<\mu}} \| h(S_{<\mu} w)\|_{M_\mu}
\\ &\lesssim C(\|w\|_{L^\infty}) \mu^{\theta+\frac32 -s}
\|u_{{\lambda},<\mu}\|_{\tilde X^{s,\theta}_{{\lambda},<\mu}} \|v_\mu\|_{\tilde
X^{s,\theta}_{\mu}} (1+\|w\|_{X^{s,\theta}}^3)
\end{split}$$ The summation with respect to $\mu$ is trivial and the square summability with respect to ${\lambda}$ is inherited from the first factor.
For $f_2$ we apply the same argument. There is a loss of a small power of $(d/\mu)^\theta$ from the first product, but this is compensated by the gain of arbitrary powers of $\mu/d$ due to . The same works for $f_3$ but there is no $(d/\mu)^\theta$ loss. In the case of $f_4$ we need to worry about the ${\lambda}$ summability, but the $(\mu/{\lambda})^N$ gain in settles this. Finally, for $f_5$ there is a $(\mu/\nu)^N$ gain which cancels again all the losses.
Summing up the pieces we obtain $$\begin{split}
&\| S_\nu (S_{\lambda}u_{\lambda}\,S_\mu v_\mu\, h(S_{<\mu} w))
\|_{X^{s,\theta}}
\\ &\lesssim C(\|w\|_{L^\infty}) \nu^{s+\theta-2} {\lambda}^{-2s+\frac72} (\frac{\mu}{{\lambda}})^N \|u_{\lambda}\|_{\tilde X^{s,\theta}_{\lambda}} \|v_\mu\|_{\tilde X^{s,\theta}_\mu }
(1+\|w\|_{X^{s,\theta}}^3)
\end{split}$$ for $\nu \ll {\lambda}$, $$\| S_\nu (S_{\lambda}u_{\lambda}\, S_\mu v_\mu \,h(S_{<\mu} w))
\|_{X^{s,\theta}} \lesssim C(\|w\|_{L^\infty}) \mu^{\theta+\frac32
-s} \|u_{\lambda}\|_{\tilde X^{s,\theta}_{\lambda}} \|v_\mu\|_{\tilde
X^{s,\theta}_\mu } (1+\|w\|_{X^{s,\theta}}^3)$$ for $\nu \approx {\lambda}$, respectively $$\| S_\nu (S_{\lambda}u_{\lambda}\, S_\mu v_\mu\, h(S_{<\mu} w))
\|_{X^{s,\theta}} \lesssim C(\|w\|_{L^\infty}) \mu^{\theta+\frac32
-s} (\frac{\mu}{\nu})^N \|u_{\lambda}\|_{\tilde X^{s,\theta}_{\lambda}}
\|v_\mu\|_{\tilde X^{s,\theta}_\mu } (1+\|w\|_{X^{s,\theta}}^3)$$ for $\nu \gg {\lambda}$.
This concludes the proof of the proposition.
Finally, we consider the bilinear estimate in , which follows from the next Proposition. Its proof cannot be completed using the type of arguments we have employed so far. Instead, we contend ourselves with reducing it to the trilinear estimate in , to the proof of which we devote the rest of the paper.
Assume that $s > \frac94$ and $\frac34 < \theta < s-\frac32$ . Then we have the multiplicative estimate $$\|\nabla u \nabla v\|_{X^{s-1,\theta-1}} \lesssim \|u\|_{X^{s,\theta}} \|v\|_{X^{s,\theta}}
\label{mainp}$$
We begin our analysis with a simple observation, namely that
If $u \in X^{s,\theta}$ then $\nabla u \in \tilde X^{s-1,\theta}$ where $$\tilde X^{s-1,\theta} = X^{s-1,\theta} + (L^2H^{s+\theta-1} \cap
H^1 H^{s+\theta-2}).$$
We first consider spatial derivatives, for which we prove the better bound $$\| \nabla_x u\|_{X^{s-1,\theta}} \lesssim \|u\|_{ X^{s,\theta}}$$ By Definition \[xst\] and Corollary \[tx\] it suffices to show that for functions $v$ localized at frequency $\lambda$ we have $$\| \nabla_x v\|_{X^{s-1,\theta}_{\lambda,d}} \lesssim
\|v\|_{\tilde X^{s,\theta}_{\lambda,d}}$$ But this follows from the straightforward commutator bound $$\| [ \Box_{g_{< \sqrt\lambda}},\nabla] v\|_{L^2} \lesssim
\lambda \|\nabla v\|_{L^2}
\label{ccv}$$ Here we recall that $g^{00}=1$, therefore every term in the commutator has at least one spatial derivative.
Next we consider time derivatives, where it suffices to show that for functions $v$ localized at frequency $\lambda$ we can write $v = v_1+
v_2$ where $v_1$, $v_2$ have the same frequency localization and $$\| \partial_t v_1\|_{X^{s-1,\theta}_{\lambda,d}} +
\left(\frac{\lambda}{d}\right)^{1-\theta}
\|\partial_t v_2\|_{ (L^2H^{s+\theta-1} \cap
H^1 H^{s+\theta-2})} \lesssim
\|v\|_{\tilde X^{s,\theta}_{\lambda,d}}
\label{vud}$$ Roughly speaking $v_1$ accounts for the low modulation ($\lesssim
\lambda$) part of $v$ while $v_2$ accounts for the high modulation part. We define $v_2$ as $$v_2 = (\Delta_{x,t})^{-1} \Box_{g_{< \sqrt\lambda}} v$$ This satisfies the bound $$\|\nabla^2 v_2\|_{L^2} \lesssim
\|\Box_{g_{< \sqrt\lambda}} v\|_{L^2}$$ which implies both the $v_2$ bound in and an $H^2$ bound for $ v_1$ which gives the correct $L^2$ bound for ${\partial}_t
v_1$, $$\lambda^{s-1} d^\theta \| \partial_t v_1\|_{L^2} +
\left(\frac{\lambda}{d}\right)^{1-\theta} \|\partial_t v_2\|_{
(L^2H^{s+\theta-1} \cap H^1 H^{s+\theta-2})} \lesssim \|v\|_{\tilde
X^{s,\theta}_{\lambda,d}}$$
It remains to estimate $ \Box_{g_{< \sqrt\lambda}} \partial_t v_1$. We have $$\| \Box_{g_{< \sqrt\lambda}} \partial_t v_1 \|_{L^2} \leq \| [
\Box_{g_{< \sqrt\lambda}},\partial_t] v_1\|_{L^2} + \| \partial_t
\Box_{g_{< \sqrt\lambda}} v_1 \|_{L^2}$$ For the first term we use again . For the second we compute $$\Box_{g_{< \sqrt\lambda}} v_1 = (-\Box_{g_{< \sqrt\lambda}}
+\Delta_{x,t}) (\Delta_{x,t})^{-1} \Box_{g_{< \sqrt\lambda}} v$$ Since the difference $\Box_{g_{< \sqrt\lambda}} -\Delta_{x,t}$ contains no second order time derivatives this yields the bound $$\| \partial_t \Box_{g_{< \sqrt\lambda}} v_1\|_{L^2} \lesssim \lambda
\|\Box_{g_{< \sqrt\lambda}} v\|_{L^2}$$ This allows us to conclude the proof of and therefore the proof of the lemma.
We now return to the estimate . Using the duality in , reduces to $$\begin{split}
\left|\int uvw dx dt\right| \lesssim & \|u\|_{\tilde X^{s-1,\theta} }
\|v\|_{\tilde X^{s-1,\theta}}
\|w\|_{X^{1-s,1-\theta}+L^2 H^{2-s-\theta}}
\end{split}
\label{tri}$$ We do a trilinear Littlewood-Paley decomposition. Due to symmetry, we need to consider two cases.
[**Case 1**]{}. Here we consider high-high-low interactions and bound $$I = \int S_{\lambda}u\, S_{\lambda}v\, S_\mu w \,dx dt, \qquad \mu \lesssim
{\lambda}$$ We have $$|I| \lesssim \|S_{\lambda}u\|_{L^\infty L^2} \|S_{\lambda}v\|_{L^2 L^6}
\|S_\mu w\|_{L^2 L^3}$$ which by the embeddings in Corollary \[sob\] give $$|I| \lesssim {\lambda}^{\frac56-2s+2} \mu^{s+\theta - \frac43}
\|u\|_{\tilde X^{s-1,\theta} }
\|v\|_{\tilde X^{s-1,\theta}}
\|w\|_{X^{1-s,1-\theta}+L^2 H^{2-s-\theta}}$$ This suffices since both the exponent of ${\lambda}$ and the sum of the two exponents are negative.
[**Case 2**]{}. Here we consider high-low-high interactions and seek to bound $$I = \int S_{\lambda}u \,S_\mu v\, S_{\lambda}w \,dx dt, \qquad \mu \ll {\lambda}$$ As a first simplification we dispense with the auxiliary $L^2$ norms. Begin with $$\begin{split}
|I| &\lesssim \|S_{\lambda}u\|_{L^2} \|S_\mu v\|_{L^\infty} \|S_{\lambda}w\|_{L^2}
\\ &\lesssim
{\lambda}^{s-1}\mu^{\frac34} \|S_{\lambda}u\|_{ L^2}\ \mu^{\frac94-s}
\|v\|_{\tilde X^{s-1,\theta}}\ {\lambda}^{1-s}\|S_{\lambda}w\|_{L^2}
\end{split}$$ This allows us to dispense not only with the $L^2H^{s+\theta-1}$ part of $u$, but also with its $X^{s-1,\theta}_{{\lambda},>\mu}$ component.
If $v \in L^2H^{s+\theta-1} \cap H^1 H^{s+\theta-2}$ then we bound $$\begin{split}
|I| & \lesssim \|S_{\lambda}u\|_{L^\infty L^2} \|S_\mu v\|_{L^2
L^\infty} \|S_{\lambda}w\|_{L^2}
\\ & \lesssim \mu^{3-s-\theta} \|u\|_{\tilde X^{s-1,\theta} }
\ \|v\|_{L^2H^{s+\theta-1}}\ {\lambda}^{1-s}\|S_{\lambda}w\|_{L^2}
\end{split}$$
Finally, if $w \in L^2 H^{2-s-\theta}$ then we can also estimate $$\begin{split}
|I| &\leq \|S_{\lambda}u\|_{L^\infty L^2} \|S_\mu v\|_{L^2 L^\infty}
\|S_{\lambda}w\|_{L^2}
\\ & \lesssim \mu^{\frac32-s+\theta} \|u\|_{\tilde X^{s-1,\theta} }
\ \|v\|_{\tilde X^{s-1,\theta}}\ {\lambda}^{1-s} \mu^{1-\theta} \|S_{\lambda}w\|_{L^2 }
\end{split}$$ which suffices for both the $L^2 H^{2-s-\theta}$ and the $X^{1-s,1-\theta}_{{\lambda},> \mu}$ components of $w$. Hence we have reduced to the bound $$|I| \lesssim \|S_{\lambda}u\|_{ X^{s-1,\theta}_{{\lambda},<\mu }} \|S_\mu
v\|_{X^{s-1,\theta}} \|S_{\lambda}w\|_{X^{1-s,1-\theta}_{{\lambda},<\mu}}
\qquad \mu \ll {\lambda}\label{tri1}$$
Unfortunately we cannot fully prove this using Strichartz type estimates. However, we can use scaling to simplify this further and reduce it to $$\left|\int S_{\lambda}u \,S_\mu v \,S_{\lambda}w dx dt\right| \lesssim \ln \mu
\ \|u\|_{X^{0,1}_{{\lambda},1}}
\|v\|_{X^{\frac54,1}_{\mu,1}} \|w\|_{X^{0,\frac14}_{{\lambda},d}} \qquad \mu
\ll {\lambda}\label{finaltri}$$ For now we show that implies . The remaining sections of the paper are devoted to the proof of .
After cancelling the powers of the high frequency the estimate follows after summation with respect to $1 \leq d_1, d_2,
d_3 \leq \mu$ from the bounds $$\left|\int S_{\lambda}u \,S_\mu v \,S_{\lambda}w dx dt\right| \lesssim \ln \mu \
d_{min}^{\frac12} d_{mid}^\frac12 d_{max}^\frac14 \| u\|_{
X^{0,0}_{{\lambda},d_1}} \|v\|_{ X^{\frac54,0}_{\mu,d_2}} \|
w\|_{X^{0,0}_{{\lambda},d_3}}
\label{casea}$$ if $d_2 < d_{max}$, respectively $$\left|\int S_{\lambda}u \,S_\mu v \,S_{\lambda}w dx dt\right| \lesssim \ln{\mu}\
d_{min}^{\frac12} d_{max}^\frac34 \|u\|_{ X^{0,0}_{{\lambda},d_1}} \|
v\|_{ X^{\frac54,0}_{\mu,d_2}} \| w\|_{X^{0,0}_{{\lambda},d_3}}
\label{caseb}$$ if $d_2 = d_{max}$.
To reduce all these cases to we use scaling combined with a time decomposition argument. Precisely, for $1 < d < \lambda$ we consider a smooth partition of unity in time with respect to time intervals of length $d^{-1}$, $$1 = \sum \chi_d^j(t)$$ Then a simple commutation argument shows that we can localize the $\tilde X^{s,\theta}_{\lambda,d}$ norm to the $d^{-1}$ time intervals while retaining square summability, $$\| u\|_{ \tilde X^{s,\theta}_{\lambda,d}}^2 \approx
\sum_j \| \chi_d^j u\|_{ \tilde X^{s,\theta}_{\lambda,d}}^2
\label{timedec}$$
We use such time decompositions in order to carry out the following three reduction steps:
\(i) Reduction to $d_{min}=1$. By all three norms are square summable with respect to time intervals of length $d_{min}^{-1}$. Hence it suffices to prove the bounds on $d_{min}^{-1}$ time intervals. Rescaling such time intervals back to time $1$ we arrive at the case $d_{min}=1$. The regularity of the coefficients improves after the rescaling, here and below. Also we note that by Duhamel’s formula we can replace the factor corresponding to $d_{min}$ by a solution to the homogeneous equation.
\(ii) Reduction to $d_{mid} =1$. By the norms corresponding to $d_{max}$ and $d_{mid}$ are square summable with respect to time intervals of length $d_{mid}^{-1}$. Hence it suffices to prove the bounds on $d_{mid}^{-1}$ time intervals. Rescaling such time intervals back to time $1$ we arrive at the case $d_{mid}=1$. Again by Duhamel’s formula we also replace the factor corresponding to $d_{mid}$ by a solution to the homogeneous equation.
\(iii) Here we are in the case where two of the factors are solutions for the homogeneous equation. In the case of the remaining factor is at high frequency $\lambda$; then we use directly .
In the case of the remaining factor is at low frequency $\mu$, so we need to prove that $$\left|\int S_{\lambda}u \,S_\mu v \,S_{\lambda}w dx dt\right| \lesssim \ln{\mu}\
d^\frac34 \|u\|_{ X^{0,0}_{{\lambda},1}} \|
v\|_{ X^{\frac54,0}_{\mu,d}} \| w\|_{X^{0,0}_{{\lambda},1}}$$ Partitioning the unit time into about $d$ time intervals of length $d^{-1}$ this would follow from $$\left|\int \chi_d^i S_{\lambda}u \,S_\mu v \,S_{\lambda}w dx dt\right| \lesssim \ln{\mu}\
d^\frac14 \|u\|_{ X^{0,0}_{{\lambda},1}} \|
v\|_{ X^{\frac54,0}_{\mu,d}} \| w\|_{X^{0,0}_{{\lambda},1}}$$ Rescaling the small time intervals to unit size this becomes exactly .
Half-waves and angular localization operators
=============================================
We write the symbol for $\Box_g$, $$p(t,x,\tau,\xi) = \tau^2 - 2 g^{0j} \tau \xi_j - g^{ij} \xi_i
\xi_j$$ in the form $$p(t,x,\tau,\xi) = (\tau + a^+(t,x,\xi)) (\tau + a^-(t,x,\xi))$$ This leads to a decomposition of solutions to the wave equation into two half-waves:
(Geba-Tataru [@MR2153517]) Let $u$ be a solution to the inhomogeneous equation for $\Box_g$. Then there is a representation $$\nabla u = u^+ + u^-$$ where $$\begin{split}
\|u^+\|_{L^2} + \|(D_t+A^+(t,x,D)) u^+\|_{L^2} &+ \|u^-\|_{L^2} +
\|(D_t+A^-(t,x,D)) u^-\|_{L^2} \\ & \lesssim \|u\|_{H^1} +\|\Box_g
u\|_{L^2}
\end{split}$$
As a consequence, in we are allowed to replace solutions to the $\Box_g$ equation by solutions to the $D_t+A^+$, respectively $D_t+A^-$ equation. We also denote $$\aligned \|u\|_{X_\pm}\,=\,\|u\|_{L^2} &+ \|(D_t+A^\pm(t,x,D))
u\|_{L^2}\\
\|u\|_{X_{\pm,d}}\,=\,d^\frac14\|u\|_{L^2} &+
d^{-\frac34}\|(D_t+A^\pm(t,x,D)) u\|_{L^2}
\endaligned$$
In order to facilitate the use of microlocal analysis tools it is convenient to replace the symbols $a^\pm$ with mollified versions $a^{\pm}_{<\mu}$ defined by $$a^{\pm}_{<\mu}(t,x,\xi) = S_{<\mu}(D_x) a(t,x,\xi)$$
Given an angular scale $\alpha$ we consider the $\pm$ Hamilton flows for $D_t+A^{\pm}_{<\alpha^{-1}}$. $$\left\{ \begin{array}{c} \frac{d}{dt} x_t^\pm = {\partial}_\xi
a^\pm_{<\alpha^{-1}}(t,x_t^\pm,\xi_t^\pm) \cr \cr
\frac{d}{dt} \xi_t^\pm = -{\partial}_x a^\pm_{<\alpha^{-1}}(t,x_t^\pm,\xi_t^\pm)
\end{array}\right . \qquad
\left\{ \begin{array}{c} x_0^\pm=x \cr \cr \xi_0^\pm = \xi\end{array}
\right .
\label{hf}$$ These are bilipschitz flows, homogeneous with respect to the $\xi$ variable. The angular scale is relevant in that the Hamilton flow for $D_t+A^{\pm}_{<\alpha^{-1}}$ serves as a good approximation to the Hamilton flow for $D_t+A^{\pm}$ up to an $O(\alpha)$ angular difference.
To characterize the higher regularity properties of these flows is convenient to introduce (see [@TG]) a metric $g_\alpha$ in the phase space, defined by $$ds^2 = |\xi|^{-4} (\xi d\xi)^2 + |\xi|^{-4} \alpha^{-2} (\xi \wedge
d\xi)^2 + \alpha^{-4} |\xi|^{-2} (\xi dx)^2 + |\xi|^{-2} \alpha^{-2}
(\xi \wedge dx)^2$$ Then as in [@TG] we obtain
The Hamilton flow maps $(x_t^\pm, \xi_t^\pm)$ are $g_\alpha$-smooth canonical transformations. \[flowreg\]
Given a direction $\theta \in S^{n-1}$ at time $t=0$ we introduce the size $\alpha$ sectors $$S_\alpha(\theta) = \{ \xi; \ \angle(\xi,\theta) < \alpha\}$$ $$\tilde S_\alpha(\theta) = \{ \xi; \ C \alpha < \angle(\xi,\theta) < 2
C\alpha\}$$ where $C$ is a fixed large constant. The images of ${\mathbb R}^n \times S_\alpha(\theta)$, respectively ${\mathbb R}^n \times \tilde S_\alpha(\theta)$ along the Hamilton flow for $D_t+A^{\pm}_{<\alpha^{-1}}$ are denoted by $H_\alpha^\pm
S_\alpha(\theta)$, respectively $H_\alpha^\pm \tilde S_\alpha(\theta)$.
Let $\xi_\theta^\alpha =\xi_\theta^\alpha(x,t)$ be the Fourier variable which is defined by the $D_t +A^{+}_{< \alpha^{-1}}$ Hamilton flow with initial data $\xi_\theta^\alpha(x,0)=\theta$ (i.e. $\xi_\theta^\alpha(x,t)=\xi^+_t(t)$ is the solution of the flow with initial data $\xi^+_0=\xi$, for which $x^+_t(t)=x$). This is well defined at least for a short time, precisely for as long as caustics do not occur. From Lemma \[flowreg\] one also sees that $\xi_\theta^\alpha$ is a $g_\alpha$-smooth function of $x$.
We consider a maximal set $O_\alpha$ of $\alpha$-separated directions and a partition of unity at time $0$ $$1 =\sum_{\theta \in O_\alpha} \chi^{\pm,\alpha}_\theta(0,x,\xi)$$ consisting of $0$-homogeneous symbols supported in $S_\alpha(\theta)$ which are smooth on the corresponding scale. Transporting these symbols along the $\pm$ Hamilton flows by $$\chi^{\pm,\alpha}_\theta(0,x,\xi)\,=\,\chi^{\pm,\alpha}_\theta(t,x_t^\pm,\xi_t^\pm)$$ produces a time dependent partition of unity $$1 =\sum_{\theta \in O_\alpha} \chi^{\pm,\alpha}_\theta(t,x,\xi)
\label{linpart}$$ so that the support of $\chi^{\pm,\alpha}_\theta(t,x,\xi)$ is contained in $H_\alpha^\pm S_\alpha(\theta)$.
The regularity of these symbols is easily obtained from the transport equations (see again [@TG]):
The symbols $\chi^{\pm,\alpha}_\theta(t,x,\xi)$ belong to the class $S(1,g_{\alpha})$[^1]. \[wpsymbols\]
We use the above partition of unity in the phase space to produce a corresponding pseudodifferential partition of unity. Given a frequency $\lambda > \alpha^{-2}$ we define the symbols $$\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,\xi) = S_{<
\lambda /8}(D_x) \chi^{\pm,\alpha}_{\theta}(t,x,\xi) \tilde s_\lambda(\xi)$$ These are used in order to split general frequency localized waves into square summable superpositions of directionally localized waves, $$S_\lambda u = \sum_{\theta \in
O_\alpha}\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D) S_\lambda u$$ This decomposition is closely related to a wave packet decomposition, see [@MR1644105], [@MR2178963], [@MR2153517], and [@TG]. The difference is that here we skip the spatial localization part since it brings no additional benefit. The above localization at spatial frequencies less than $\lambda/8$ insures that the output of the operators $\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
S_\lambda$ is still localized at frequency $\lambda$. This localization is otherwise harmless:
The symbols $\chi^{\pm,\alpha}_{\theta,\lambda} (t,x,\xi)$ belong to the class $S(1,g_{\alpha})$. In addition, we have similar bounds for the Poisson bracket $$\{ \tau+a^\pm_{<\alpha^{-1}} (t,x,\xi),
\chi^{\pm,\alpha}_{\theta,\lambda} (t,x,\xi)\}
\in S(1,g_{\alpha})
\label{poib}$$ \[wpsymbols\]
The fact that $\chi^{\pm,\alpha}_{\theta,\lambda} (t,x,\xi) \in S(1,g_{\alpha})$ is straightforward since the multiplier $S_{<\lambda/8}$ is a mollifier on the $\lambda^{-1}$ spatial scale, which is less that the spatial scale of the $g_\alpha$ balls.
Since $\chi^{\pm,\alpha}_{\theta}$ is transported along the $a^\pm_{<\alpha^{-1}} (t,x,\xi)$ flow, the Poisson bracket is expressed in the form $$\{ a^\pm_{<\alpha^{-1}} (t,x,\xi), \tilde s_\lambda(\xi)\}
\chi^{\pm,\alpha}_{\theta,\lambda} (t,x,\xi) + \tilde s_\lambda(\xi) [
H_{a^\pm_{<\alpha^{-1}}},S_{<\lambda/8}(D_x)]
\chi^{\pm,\alpha}_{\theta}(t,x,\xi)$$ Here $H_{a^\pm_{<\alpha^{-1}}}$ is the Hamiltonian operator associated to the $a^\pm_{<\alpha^{-1}} (t,x,\xi)$ flow. It is easy to see that the first term belongs to $S(1,g_\alpha)$, therefore it remains to consider the commutator term. We have $$[ H_{a^\pm_{<\alpha^{-1}}},S_{<\lambda/8}(D_x)] = [\partial_\xi
a^\pm_{<\alpha^{-1}}, S_{<\lambda/8}(D_x)] \partial_x - [\partial_x
a^\pm_{<\alpha^{-1}}, S_{<\lambda/8}(D_x)] \partial_\xi$$ The commutator of a scalar function $g$ with $S_{< \lambda/8}$ can be expressed as a rapidly convergent series of the form $$[g,S_{< \lambda/8}] = \lambda^{-1} \sum_j S_{< \lambda/8}^{1,j} \nabla g S_{<
\lambda/8}^{2,j}$$ where the multipliers $S_{< \lambda/8}^{1,j}$ and $S_{<
\lambda/8}^{2,j}$ have the same properties as $S_{<\lambda/8}$ and decay rapidly with respect to $j$. Then the above commutator term is expressed as $$[ H_{a^\pm_{<\alpha^{-1}}},S_{<\lambda/8}(D_x)] =
\lambda^{-1} \sum_j S_{< \lambda/8}^{1,j} \left( \partial_x \partial_\xi
a^\pm_{<\alpha^{-1}} \partial_x - \partial_x^2
a^\pm_{<\alpha^{-1}}\partial_\xi \right) S_{<\lambda/8}^{j,2}$$ At this stage the effect of the mollifiers is negligible and we can use the regularity properties of $a^\pm$ and $\chi^{\pm,\alpha}_{\theta}$ to directly compute $$\tilde s_\lambda(\xi) [
H_{a^\pm_{<\alpha^{-1}}},S_{<\lambda/8}(D_x)]
\chi^{\pm,\alpha}_{\theta}(t,x,\xi) \in S(\frac{1}{\alpha^2
\lambda},g_\alpha)$$
To better understand the phase space localization provided by $\chi^{\pm,\alpha}_{\theta,\lambda} $ consider some point $(x_0,t_0)$ and the corresponding center direction $\xi_\theta^\alpha(x_0,t_0)$. A spatial unit $g_\alpha$ ball $B_\theta^\alpha(x_0,t_0)$ centered at $(x_0,t_0)$ has dimensions[^2] $\alpha^2 \times \alpha^{n-1}$ with the long sides normal to $\xi_\theta^\alpha(x_0,t_0)$. Within the ball $B_\theta^\alpha(x_0,t_0)$, $\chi^{\pm,\alpha}_{\theta,\lambda}
\tilde S_\lambda$ localizes frequencies to a sector of angle $\alpha$ centered at $\xi_\theta^\alpha(x_0,t_0)$. Thus the frequencies are localized to a radial rectangle centered at $\lambda \xi_\theta^\alpha(x_0,t_0)$ of size $\lambda \times
(\alpha \lambda)^{n-1}$. In this picture, angle $\alpha$ wave packets correspond to a spatial localization on the scale of the above ball $B_\theta^\alpha(x_0,t_0)$, constructed along a fixed ray of the Hamilton flow.
The $g_\alpha$ metric restricted to frequency $\lambda$ is slowly varying and temperate at frequencies [^3] $\lambda \geq\alpha^{-2}$, and in our analysis we will always be above this threshold. Hence there is a good pseudodifferential calculus for operators with $S(1,g_\alpha)$ symbols. The semiclassical parameter $h=h(\alpha,\lambda)$ in the $S(1,g_\alpha)$ calculus at frequency $\lambda$ is given by $$h(\alpha,\lambda) = (\alpha^2 \lambda)^{-1}$$ The $S(1,g_\alpha)$ symbols at frequency $\lambda$ satisfy the bounds $$\left|(\xi_\theta^{\alpha} \partial_x)^\sigma
(\xi_\theta^{\alpha} \wedge \partial_x)^\beta
\partial_\xi^\nu (\xi \partial_\xi)^\gamma q(t,x,\xi)\right|
\lesssim \alpha^{-2\sigma -|\beta|} (\alpha \lambda)^{-\nu}
\label{simbolga}$$ Due to the $L^2$ in time regularity of the second order derivatives of the coefficients we also introduce the space of symbols $L^2
S(1,g_\alpha)$ which at frequency $\lambda$ satisfy $$\left|(\xi_\theta^{\alpha} \partial_x)^\sigma
(\xi_\theta^{\alpha} \wedge \partial_x)^\beta
\partial_\xi^\nu (\xi \partial_\xi)^\gamma q(t,x,\xi)\right|
\lesssim \alpha^{-2\sigma -|\beta|} (\alpha \lambda)^{-\nu} f(t)
\label{l2simbolga}$$ for some $f \in L^2$. In all the operators we consider here, the function $f$ is the same: $$f(t) = M(\|\nabla^2 g(t)\|_{L^\infty})
\label{fM}$$ In some of our estimates we need to deal with two distinct scales at a given frequency $\lambda$, namely the angular scale $\alpha$ and the $\lambda^{\frac12}$ scale at which the coefficients are truncated. Correspondingly we introduce additional symbol classes $C^k_\lambda
S(1,g_\alpha)$ of symbols $q$ localized at frequency $\lambda$ which satisfy the $S(1,g_\alpha)$ bounds for $\sigma+|\beta| \leq k$, respectively the weaker estimate $$\left|(\xi_\theta^{\alpha} \partial_x)^\sigma
(\xi_\theta^{\alpha} \wedge \partial_x)^\beta
\partial_\xi^\nu (\xi \partial_\xi)^\gamma q(t,x,\xi)\right|
\lesssim (\alpha^{-2\sigma -|\beta|}+\alpha^{-k}
\lambda^{\frac{\sigma+|\beta|}2}) (\alpha \lambda)^{-\nu}
\label{simbolgal}$$ for $\sigma+|\beta| > k$. There is still a calculus for such symbols, since the above bounds are stronger than the $S(1,g_{\lambda^{\frac12}})$ bounds. The related classes of symbols $L^2 C^k_\lambda S(1,g_\alpha)$ are defined in a manner which is similar to .
Using the calculus for the above symbol classes one can prove that the partition of unity in yields an almost orthogonal decomposition of functions, namely
Fix a frequency ${\lambda}$ and let $\alpha > {\lambda}^{-\frac12}$. Then for each function $u$ which is localized at frequency $\lambda$ we have $$\sum_{\theta \in O_\alpha} \| \chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{X_\pm}^2 \approx \| u\|_{X_\pm}^2
\label{linsumeq}$$ \[linsum\]
We only outline the proof, since this result is essentially contained in [@MR2153517]. There are two bounds to prove. The first $$\sum_{\theta \in O_\alpha} \| \chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{L^2}^2 \approx \| u\|_{L^2} ^2
\label{aort}$$ follows from the almost orthogonality of the operators $\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)$. This in turn is due to the almost disjoint supports[^4] of $\chi^{\pm,\alpha}_{\theta,\lambda}$ and to the $S(1,g_{\alpha})$ calculus.
Consider now the second bound $$\sum_{\theta \in O_\alpha} \| (D_t + A^\pm) \chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{L^2}^2 \approx \|(D_t + A^\pm) u\|_{L^2}^2 +O(
\| u\|_{L^2}^2)
\label{aortc}$$ We first establish it with $A^\pm$ replaced by $A^\pm_{< \lambda^{\frac12}}$, $$\sum_{\theta \in O_\alpha} \| (D_t + A^\pm_{< \lambda^{\frac12}})
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{L^2}^2 \approx \|(D_t + A^\pm_{< \lambda^{\frac12}}) u\|_{L^2}^2
+O( \| u\|_{L^2}^2)
\label{aortd}$$ Due to and the energy bound $$\| u\|_{L^\infty L^2}^2 \lesssim \| u\|_{L^2}^2
+ \| u\|_{L^2}\|(D_t + A^\pm_{< \lambda^{\frac12}}) u\|_{L^2}$$ it suffices to prove the commutator estimate $$\sum_{\theta \in O_\alpha}\| [ D_t+A^\pm_{<\lambda^{\frac12}},
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] u\|_{L^2}^2 \lesssim \|
u\|_{L^\infty L^2}^2 \label{aorte}$$ which we split into two components.
For the low frequency part of the coefficients we use a second order commutator $$\sum_{\theta \in O_\alpha}\| [ D_t+A^\pm_{<\alpha^{-1}},
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] u\|_{L^2}^2 \lesssim \|
u\|_{L^\infty L^2}^2
\label{lowcom}$$ For this it suffices to prove that $$[ D_t+A^\pm_{<\alpha^{-1}}, \chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)]
\in OP L^2 S(1,g_\alpha)
\label{lowcoma}$$ The summation with respect to $\theta \in O_\alpha$ follows by orthogonality since the symbols for the above commutators will retain the rapid decay away from the support of $\chi^{\pm,\alpha}_{\theta,\lambda}$. Here it is important that applies uniformly.
Due to the Poisson bracket bound in it suffices to show that $$[A^\pm_{<\alpha^{-1}},
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] +
i \{a^\pm_{< \alpha^{-1}}, \chi^{\pm,\alpha}_{\theta,\lambda}\} (t,x,D) \in OP L^2 S(1,g_\alpha)$$ Due to the frequency localization of $\chi^{\pm,\alpha}_{\theta,\lambda}$, only the values of $a^\pm(t,x,\xi)$ in the region $|\xi| \approx \lambda$ can affect the above operator. At this point it is no longer important that $a^\pm_{<
\alpha^{-1}}$ and $\chi^{\pm,\alpha}_{\theta,\lambda}$ are related. We consider a rapidly convergent spherical harmonics expansion of $a^{\pm}$, $$a^{\pm}(t,x,\xi) = \sum_j b_j (t,x) \phi_j(\xi)$$ where $b_j$ have the same regularity as the coefficients $g^{ij}$ while $ \phi_j(\xi) $ are homogeneous of order $1$. It suffices to consider a single term $b(t,x) \phi(\xi)$ in this expansion and show that $$[b_{<\alpha^{-1}} \phi(D),
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] + i \{b_{<\alpha^{-1}} \phi
, \chi^{\pm,\alpha}_{\theta,\lambda}\} (t,x,D) \in OPL^2S(1,g_\alpha)
\label{scom}$$ To see this we consider the commutators with $b$ and with $\phi$. The commutator term with $b$ has the form $$C_b = ([b_{<\alpha^{-1}} ,
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] +
i \{b_{<\alpha^{-1}},\chi^{\pm,\alpha}_{\theta,\lambda}\} (t,x,D))\phi(D)$$ Since $\partial_x^2 b_{<\alpha^{-1}} \in L^2S(1,g_\alpha)$, $\partial_\xi^2
\chi^{\pm,\alpha}_{\theta,\lambda} \in S(\alpha^{-2}
\lambda^{-2},g_\alpha)$ and $\phi \in S(\lambda,g_\alpha)$, the $S(g_\alpha)$ calculus at frequency $\lambda$ yields the better result $C_b \in OPL^2S(\alpha^{-2} \lambda^{-1},g_\alpha)$, which is tight only when $\alpha= \lambda^{-\frac12}$.
The commutator term with $\phi$ has the form $$C_\phi = b_{<\alpha^{-1}} ([\phi(D),
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] + i \{\phi
, \chi^{\pm,\alpha}_{\theta,\lambda}\} (t,x,D))$$ The $b_{<\alpha^{-1}}$ factor belongs to $S(1,g_\alpha)$ and can be neglected. The argument for the remaining part is somewhat more delicate since it hinges on the homogeneity of $\phi$. With $b = 1$ denote by $\xi$ the input frequency for $C_\phi$ and by $\eta$ the output frequency. Due to the homogeneity of $\phi$ we have the representation $$\phi(\eta) - \phi(\xi) = (\eta-\xi) \nabla \phi(\xi) +
\psi(\xi,\eta)(\xi \wedge (\xi-\eta))^2
\label{phixieta}$$ where $\psi$ is a smooth and homogeneous of order $-3$ matrix valued function. For $|\xi|,|\eta| \approx \lambda$ we can separate variables in $\psi$ and express it as a rapidly convergent series $$\psi(\xi,\eta) = \lambda^{-3}\sum_{j} \psi^1_j(\eta) \psi^2_j(\xi)$$ This gives a representation for $C_\phi$ of the form $$C_\phi = \lambda^{-3}\sum_{j} \psi^1_j(D) ((\xi \wedge \partial_x)^2
\chi^{\pm,\alpha}_{\theta,\lambda})(t,x,D) \psi^2_j(D)$$ Since $\chi^{\pm,\alpha}_{\theta,\lambda}(x,D) \in S(1,g_\alpha)$ we obtain $(\xi \wedge \partial_x)^2 \chi^{\pm,\alpha}_{\theta,\lambda}
\in S(\lambda^2 \alpha^{-2},g_\alpha)$ which shows that $C_\phi \in
OPS(\alpha^{-2} \lambda^{-1},g_\alpha)$. This concludes the proof of and thus the proof of .
For the intermediate frequency part of the coefficients we have a first order commutator estimate $$\sum_{\theta \in O_\alpha}\| [ A^\pm_{\alpha^{-1} < \cdot <
\lambda^{\frac12} }, \chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] u\|_{L^2}^2
\lesssim \| u\|_{L^\infty L^2}^2
\label{incom}$$ Together with this implies .
This follows from first order commutator estimate $$[ A^\pm_{\alpha^{-1} < \cdot < \lambda^{\frac12} },
\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)] \in OPL^2 C^1_\lambda S(1,g_\alpha)
\label{incoma}$$ Indeed, for a scalar function $b$ we can estimate $$\alpha^{-2} \| b_{\alpha^{-1} < \cdot < \lambda^{\frac12} } \|_{L^2
L^\infty}
+ \alpha^{-1} \| \partial_x b_{\alpha^{-1} < \cdot < \lambda^{\frac12} } \|_{L^2
L^\infty}
\lesssim \|\partial^2 b\|_{L^2 L^\infty}$$ Applied to the the symbol $a^\pm$ as a function of $x$ this shows that $$a^\pm_{\alpha^{-1} < \cdot <
\lambda^{\frac12} }\in L^2C^2_\lambda S(\alpha^2 \lambda,g_\alpha)$$ Since $\chi^{\pm,\alpha}_{\theta,\lambda} \in S(1,g_\alpha)$, the estimate follows by pdo calculus. The square summability with respect to $\theta$ is again due to the almost disjoint supports of the symbols $\chi^{\pm,\alpha}_\theta$.
It remains to pass from to . Due to the energy bound $$\| u\|_{L^\infty L^2}^2 \lesssim \| u\|_{L^2}^2
+ \| u\|_{L^2}\|(D_t + A^\pm_{< \lambda^{\frac12}}) u\|_{L^2}$$ this is a consequence of the estimate $$\|A^\pm_{> \lambda^{\frac12}} u\|_{L^2} \lesssim \| u\|_{L^\infty L^2}$$ applied to both $u$ and $\chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u$. Using the spherical harmonics decomposition of the symbols $a^{\pm}$ as above this reduces to the straightforward bound $$\| b_{> \lambda^{\frac12}} u\|_{L^2} \lesssim \lambda^{-1} \|\partial^2
b\|_{L^2 L^\infty}
\|u\|_{L^\infty L^2}$$
The frequency localization in $\chi^{\pm ,\alpha} _{\theta,\lambda}$ contributes to improved Strichartz type estimates above the critical range of exponents. Begin for instance with the endpoint $L^2 L^6$ Strichartz estimate $$\|\chi^{\pm ,\alpha} _{\theta,\lambda}(t,x,D) u\|_{L^2 L^6}
\lesssim {\lambda}^\frac56 \| u\|_{X_\pm} \label{specher}$$ Here the angular frequency localization plays no role. However, suppose we want to use Bernstein’s inequality to replace this by an $L^2 L^\infty$ estimate. Modulo rapidly decaying tails, within each spatial $g_\alpha$ ball $B^\alpha_\theta(x_0,t_0)$ the function $ \chi^{\pm ,\alpha} _{\theta,\lambda} (t,x,D) u$ is frequency localized in a dyadic sector section of size $\lambda
\times (\alpha \lambda)^3$. Then the constant in Bernstein’s inequality is $$[\lambda \times (\alpha \lambda)^3]^\frac16 = \lambda^\frac23 \alpha^\frac12$$ Hence we obtain the better $L^2 L^\infty$ bound $$\|\chi^{\pm ,\alpha} _{\theta,\lambda}(t,x,D) u\|_{L^2 L^\infty} \lesssim
\alpha^\frac12 \lambda^\frac32 \| u\|_{X_\pm},
\qquad \alpha > \lambda^{-\frac12}
\label{pecher}$$ A simpler related uniform bound is derived directly from the energy estimates, $$\|\chi^{\pm ,\alpha} _{\theta,\lambda}(t,x,D) u\|_{L^\infty} \lesssim
\alpha^\frac32 \lambda^2 \| u\|_{X_\pm},
\qquad \alpha > \lambda^{-\frac12}
\label{pechera}$$ A similar bound holds for the right hand side of the $\chi^{\pm
,\alpha} _{\theta,\lambda}(t,x,D) u$ equation. Indeed, for $u \in X_\pm$ we can write $$(D_t + A^\pm) \chi^{\pm ,\alpha}_{\theta,\lambda}(t,x,D) u =
(D_t + A^\pm_{<{\lambda}^\frac12}) \chi^{\pm ,\alpha}_{\theta,\lambda}(t,x,D)
u + A^\pm_{>{\lambda}^\frac12} \chi^{\pm ,\alpha}_{\theta,\lambda}(t,x,D)
u$$ The first term belongs to $L^2$ and has a similar frequency localization as $\chi^{\pm ,\alpha}_{\theta,\lambda}(t,x,D)
u$. The second is estimated directly using . This yields $$\|(D_t + A^\pm) \chi^{\pm ,\alpha} _{\theta,\lambda}(t,x,D) u\|_{L^2 L^\infty} \lesssim
\alpha^\frac32 \lambda^2 \| u\|_{X_\pm},
\qquad \alpha > \lambda^{-\frac12}
\label{pecherb}$$
Another way of taking advantage of the angular localization is in corresponding bounds for derivatives. Consider the differentiation operators $\xi_\theta^\alpha \wedge D$ whose symbol vanishes in the $\xi_\theta^\alpha$ direction. Then in the support of $\chi^{\pm ,\alpha} _{\theta,\lambda}$ these symbols have size $\alpha \lambda$. Hence from we also obtain $$\|(\xi_\theta^\alpha \wedge D)\chi^{\pm ,\alpha} _{\theta,\lambda}(t,x,D) u\|_{L^2
L^6} \lesssim (\alpha {\lambda}) {\lambda}^\frac56 \| u\|_{X_\pm}
\label{spechera}$$ We can argue in the same way for the energy estimates or for the $L^2
L^\infty$ bound in . For convenience we collect several such bounds in a single norm, $$\begin{split}
\| v\|_{X_{\pm}^{\lambda, \alpha, \theta}} =&\ \|v\|_{X_{\pm}} +\|
v\|_{L^\infty L^2} + \lambda^{-\frac56} \| v\|_{L^2 L^6}+
\alpha^{-\frac12} \lambda^{-\frac32} \| v\|_{L^2 L^\infty}
+ \alpha^{-\frac32} \lambda^{-2} \|v\|_{L^\infty}
\\ &\ + \alpha^{-\frac32} \lambda^{-2} \| (D_t + A^\pm) v\|_{L^2 L^\infty}
+ (\alpha
\lambda)^{-1}\| (\xi_\theta^\alpha \wedge D) v\|_{L^\infty
L^2} \\ &\ + (\alpha
\lambda)^{-1} (\lambda^{-\frac56} \| (\xi_\theta^\alpha \wedge D) v\|_{L^2 L^6} +
\alpha^{-\frac12} \lambda^{-\frac32} \| (\xi_\theta^\alpha \wedge D)
v\|_{L^2 L^\infty})
\end{split}$$ and use it to state a corresponding version of , $$\sum_{\theta \in O_\alpha} \| \chi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{X_\pm^{\lambda,\alpha,\theta}}^2 \approx \|\tilde S_{\lambda}u\|_{X_\pm}^2
\label{limsumeqa}$$
We want to replace the partition of unity in first with a bilinear one and next with a trilinear one. Given two frequencies $\mu < \lambda$, we denote $\alpha_\mu= \mu^{-\frac12}$ and introduce a corresponding bilinear partition of unity which is useful when estimating the frequency $\mu$ output of the product of two frequency $\lambda$ waves. The main contribution corresponds to opposite frequencies $\xi$ and $\eta$, therefore we organize the following decomposition based on the dyadic angle $\alpha_\mu \leq \alpha \leq 1$ between $\xi$ and $-\eta$. Precisely, by superimposing the $\alpha$ angular decompositions for $\alpha$ in the above range we obtain $$\begin{split}
& \tilde s_\lambda(\xi) \tilde s_\lambda(\eta)=
\\ & \sum_{\theta_1,\theta_2\in
O_{\alpha_\mu}}^{|\theta_1 + \theta_2| \leq 2C
\alpha_\mu} \ \sum_{\theta_3,\theta_4 \in
O_{2\alpha_\mu}}^{|\theta_3 +\theta_4| \leq 4C
\alpha_\mu}
\chi^{\pm, \alpha_\mu}_{\theta_1,\lambda}(t,x,\xi)
\chi^{\mp, \alpha_\mu}_{\theta_2,\lambda}(t,x,\eta)
\chi^{\pm,2 \alpha_\mu}_{\theta_3,\lambda}(t,x,\xi)
\chi^{\mp,2 \alpha_\mu}_{\theta_4,\lambda}(t,x,\eta)
\\ &+ \!\!
\sum_{\alpha=\alpha_\mu}^1 \!\!
\sum_{\theta_1,\theta_2 \in O_\alpha}^{C \alpha
\leq |\theta_1+\theta_2| \leq 2 C
\alpha}
\ \sum_{\theta_3,\theta_4 \in
O_{2\alpha}}^{|\theta_3+\theta_4| \leq 4C
\alpha} \!\!
\chi^{\pm,\alpha}_{\theta_1,\lambda}(t,x,\xi)
\chi^{\mp,\alpha}_{\theta_2,\lambda}(t,x,\eta)
\chi^{\pm,2\alpha}_{\theta_3,\lambda}(t,x,\xi)
\chi^{\mp,2\alpha}_{\theta_4,\lambda}(t,x,\eta)
\end{split}$$ To shorten this expression we redenote factors and harmlessly simplify the summation notations to $$1 = \sum_{\theta \in O_{\alpha_\mu}}
\phi^{\pm,\alpha_\mu}_{\theta,\lambda}(t,x,\xi) \phi^{\mp,\alpha_\mu}_{-\theta,\lambda}(t,x,\eta)
+ \sum_{\alpha=\alpha_\mu}^1 \ \sum_{\theta \in O_\alpha}
\phi^{\pm,\alpha}_{\theta,\lambda}(t,x,\xi)
{{\tilde{\phi}}}^{\mp,\alpha}_{-\theta,\lambda}(t,x,\eta)
\label{bilinsum}$$ where the tilde in $ {{\tilde{\phi}}}^{\pm,\alpha}_{\theta,\lambda}$ indicates an $O(C\alpha)$ angular separation from $\theta$. The symbols $\phi^{\pm,\alpha}_{\theta,\lambda}$, respectively ${{\tilde{\phi}}}^{\pm,\alpha}_{\theta,\lambda}$ retain the same properties as $\chi^{\pm,\alpha}_{\theta,\lambda}$, namely $$\phi^{\pm,\alpha}_{\theta,\lambda} \in
S(1,g_\alpha),
\qquad \{ \tau+a^\pm_{<\alpha^{-1}} (t,x,\xi),
\phi^{\pm,\alpha}_{\theta,\lambda} (t,x,\xi)\}
\in S(1,g_{\alpha})$$ and the same for ${{\tilde{\phi}}}^{\pm,\alpha}_{\theta,\lambda}$. In particular the counterpart of is still valid, $$\sum_{\theta \in O_\alpha} \| \phi^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{X_\pm^{\lambda,\alpha,\theta}}^2 +
\| {{\tilde{\phi}}}^{\pm,\alpha}_{\theta,\lambda}(t,x,D)
u\|_{X_\pm^{\lambda,\alpha,\theta}}^2\approx \|\tilde S_{\lambda}u\|_{X_\pm}^2
\label{limsumeqb}$$
Finally, we arrive at the main trilinear symbol decomposition. Its aim is to achieve a simultaneous angular decomposition in trilinear expressions of the form $$\int u v w dx dt$$ We denote the three corresponding frequencies by $\xi, \eta$ and $\zeta$. We assume that each of the factors has a dyadic frequency localization, $$|\xi| \approx |\eta| \approx \lambda, \qquad |\zeta| \approx \mu,
\qquad 1 \ll \mu \leq \lambda$$ If the trilinear decomposition were translation invariant then only its structure on the diagonal $\xi+\eta+\zeta=0$ is relevant. However, in our case we are working with variable coefficient operators therefore a neighborhood of the diagonal is relevant. The size of this neighborhood is determined by the spatial regularity of the symbols via the uncertainty principle.
Corresponding to the first term in we consider a decomposition in $\zeta$ with respect to the dyadic angle between $\zeta$ and $\theta$, $$\tilde s_\mu(\zeta) = \phi^{\pm,\alpha_\mu}_{\theta,\mu} (t,x,\zeta)
+ \sum_{\alpha > \alpha_\mu} {{\tilde{\phi}}}^{\pm,\alpha}_{\theta,\mu}
(t,x,\zeta)$$ To understand the $\zeta$ decomposition corresponding to the second term in we first identify the location of the diagonal $\xi+\eta+\zeta = 0$. Given the above dyadic localization of $\xi,\eta$ and $\zeta$, if the angle between $\xi$ and $-\eta$ is of order $\alpha$, then the angle between $\xi$ and $\pm \zeta$ must be of order $\alpha \lambda \mu^{-1}$ which is larger than $\alpha$. Thus the interesting angular separation threshold for $\zeta$ is $\alpha \lambda \mu^{-1}$. It would appear that there are two cases to consider, namely when the angle between $\xi$ and $\zeta$ is small, and when the angle between $-\xi$ and $\zeta$ is small. However, due to our choice of the $\pm$ signs corresponding to $\xi$, $\eta$ and $\zeta$, the latter case leads to nonresonant wave interactions and loses its relevance. Hence, the significant dyadic parameter here is the angle between $\xi$ and $\zeta$, and the $\zeta$ decomposition has the form $$\tilde s_\mu(\zeta) = \phi^{\pm,\alpha \mu^{-1}
\lambda}_{\theta,\mu} (t,x,\zeta) + {{\tilde{\phi}}}^{\pm,\alpha \mu^{-1}
\lambda}_{\theta,\mu} (t,x,\zeta) + \sum_{\beta
>\alpha \mu^{-1}
\lambda } {{\tilde{\phi}}}^{\pm,\beta}_{\theta,\mu} (t,x,\zeta)$$ Then the full trilinear decomposition has the form $$\begin{split}
\tilde s_{\lambda}(\xi) \tilde s_\lambda(\eta) \tilde s_\mu(\zeta)
= & \sum_{\theta \in O_{\alpha_\mu}}
\phi^{\pm,\alpha_\mu}_{\theta,\lambda}(t,x,\xi)
\phi^{\mp,\alpha_\mu}_{-\theta,\lambda}(t,x,\eta)
\phi^{\pm,\alpha_\mu}_{\theta,\mu}(t,x,\zeta)
\\ +& \sum_{\theta \in O_{\alpha_\mu}} \phi^{\pm,\alpha_\mu}_{\theta,\lambda}(t,x,\xi)
\phi^{\mp,\alpha_\mu}_{-\theta,\lambda}(t,x,\eta)
\sum_{\alpha > \alpha_\mu}
\tilde \phi^{\pm,\alpha}_{\theta,\mu}(t,x,\zeta)
\\ +& \sum_{ \alpha > \alpha_\mu }\sum_{\theta \in
O_{\alpha}}
\phi^{\pm,\alpha}_{\theta,\lambda}(t,x,\xi) \tilde
\phi^{\mp,\alpha}_{-\theta,\lambda}(t,x,\eta) \tilde \phi^{\pm,\alpha
\mu^{-1} \lambda}_{\theta,\mu}(t,x,\zeta)
\\ +& \sum_{ \alpha > \alpha_\mu }\sum_{\theta \in
O_{\alpha}}
\phi^{\pm,\alpha}_{\theta,\lambda}(t,x,\xi) \tilde
\phi^{\mp,\alpha}_{-\theta,\lambda}(t,x,\eta) \phi^{\pm,\alpha \mu^{-1}
\lambda}_{\theta,\mu}(t,x,\zeta)
\\ +& \sum_{ \alpha > \alpha_\mu }\sum_{\theta \in
O_{\alpha}}
\phi^{\pm,\alpha}_{\theta,\lambda}(t,x,\xi) \tilde
\phi^{\mp,\alpha}_{-\theta,\lambda}(t,x,\eta) \sum_{\beta > \alpha \mu^{-1}
\lambda} \tilde\phi^{\pm,\beta}_{\theta,\mu}(t,x,\zeta)
\end{split}
\label{trilindec}$$ In the above sum the first three terms are the main ones, as they account for the behavior near the diagonal. The remaining terms have off diagonal support, and their contribution to trilinear forms as above is negligible.
Proof of the trilinear estimate
================================
As noted in the previous section, we can replace the spaces $X^{s,\theta}_{{\lambda},d}$ in with the $X_{\pm}$ spaces. Hence we restate in the form
For any choice of the $\pm$ signs and $1 < d < \mu \ll \lambda$ we have $$\left|\int S_{\lambda}u\, S_{\lambda}v\, S_\mu w dx dt \right | \lesssim \ln \mu \cdot
\mu^\frac54\|S_\lambda u\|_{X_{\pm}} \|S_\lambda v\|_{X_{\pm,d}}
\|S_\mu w\|_{X_\pm}
\label{lastone}$$
We begin with several simple observations. First, by localizing to a fixed smaller space-time scale and rescaling back to unit scale we can insure that the coefficients $g^{ij}$ vary slowly inside a unit cube, $$|\nabla_{x,t} g^{ij}| \ll 1$$ This in turn insures that the Fourier variable does not vary much along the Hamilton flow, $$|\xi_\theta^\alpha - \theta| \ll 1$$
We can also localize all factors in frequency to angular regions of small size, say $< \frac{1}{20}$. The corresponding localization multipliers are easily seen to be bounded in $X_{\pm}$ and $X_{\pm,d}$.
If the first two $\pm$ signs are identical then the product $S_{\lambda}u\, S_{\lambda}v$ is concentrated at a time frequency of the order of ${\lambda}$ which makes it almost orthogonal to $S_\mu w$, hence the estimate above is much easier. Therefore without any restriction in generality we fix the first sign to $+$ and the second one to $-$. Even though the problem is not symmetric with respect to the first two factors, the sign in the third factor plays no role whatsoever, so we fix it to $+$. We denote $$a(t,x,\xi)\,=\,a^+(t,x,\xi)$$ Then $$a^-(t,x,\xi)\,=\,-a(t,x,-\xi)$$ We note that, for the purpose of the above estimates, in the definition of $X_{\pm}$ at frequency $\lambda$ we can replace the symbols $a(x,\xi)$ with their regularized versions, namely $a_{<{\lambda}^\frac12}(x,\xi)$.
To keep the number of parameters small we first present the argument in the case when $d=1$. Once this is done, we show what changes are necessary for $d > 1$.
[**Case 1**]{}: $d = 1$. Corresponding to the trilinear symbol decomposition of the identity we consider the corresponding pseudodifferential decomposition of the trilinear expression in . The we estimate each of the five terms. We remark that, since $S_{\lambda}u$, $S_{\lambda}v$ and $S_\mu w$ are frequency localized in a small angle, so are all the factors in .
[**Case 1, term I:**]{} $$I = \sum_{\theta \in O_{\alpha_\mu}} \int
\phi^{+,\alpha_\mu}_{\theta,\lambda}(t,x,D) S_{\lambda}u\
\phi^{-,\alpha_\mu}_{-\theta,\lambda}(t,x,D) S_{\lambda}v\
\phi^{+,\alpha_\mu}_{\theta,\mu}(t,x,D) S_\mu w \,dx dt$$ We use the energy estimate for the first two factors and the $L^2 L^\infty$ bound for the third to obtain $$|I| \lesssim \mu^{\frac54} \|
\phi^{+,\alpha_\mu}_{\theta,\lambda}(x,D) S_{\lambda}u\|_{X^{\lambda,\alpha_\mu,\theta}_{+}} \|
\phi^{-,\alpha_\mu}_{-\theta,\lambda}(x,D) S_{\lambda}v\|_{X^{\lambda,\alpha_\mu,\theta}_{-}} \|
\phi^{+,\alpha_\mu}_{\theta,\mu}(t,x,D) S_\mu w
\|_{X^{\mu,\alpha_\mu,\theta}_{+}}$$ The summation with respect to $\theta$ is straightforward due to .
[**Case 1, term II:**]{} This is the most difficult term, $$II = \sum_{\theta \in O_{\alpha_\mu}} \int
\phi^{+,\alpha_\mu}_{\theta,\lambda}(t,x,D) S_{\lambda}u\
\phi^{-,\alpha_\mu}_{-\theta,\lambda}(t,x,D)S_{\lambda}v
\sum_{\alpha > \alpha_\mu} \tilde
\phi^{+,\alpha}_{\theta,\mu}(t,x,D) S_\mu w \,dx dt$$
The summation with respect to $\theta$ is easily done using . Hence, in what follows, we fix $\theta$ and redenote $$u_\theta = \phi^{+,\alpha_\mu}_{\theta,\lambda}(t,x,D) S_{\lambda}u, \qquad
v_\theta = \phi^{-,\alpha_\mu}_{-\theta,\lambda}(t,x,D) S_{\lambda}v, \qquad
w_{\theta}^\alpha = {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D)S_\mu w$$ The factors $u_\theta$ and $v_\theta$ are frequency localized in small angles around $\theta$, respectively $-\theta$; $w_{\theta}^\alpha$ has a similar localization around $\pm \theta$ provided that $\alpha
\ll 1$.
We denote by ${{\tilde{a}}}_{<\mu^\frac12} (t,x,\xi)$ the linearization of $a_{<\mu^\frac12}(t,x,\xi)$ with respect to $\xi$ around $\xi = \xi_\theta^{\alpha_\mu}(t,x)$. Since $a_{<\mu^\frac12}(t,x,\xi)$ is a homogeneous symbol of order $1$, we have $${{\tilde{a}}}_{<\mu^\frac12} (t,x,\xi) = \xi \partial_\xi
a_{<\mu^\frac12}(t,x,\xi_\theta^{\alpha_\mu})$$ Consider now the difference $$e = a_{<\mu^\frac12}- {{\tilde{a}}}_{<\mu^\frac12}$$ It vanishes of second order on the half line ${\mathbb R}^+ \xi_\theta$. Due to the uniform (nonradial) convexity of the characteristic cone $\{
\tau + a_{<\mu^\frac12}(t,x,\xi)=0\}$, it follows that $e$ is nonzero when $\xi$ is not collinear with $\xi_\theta^{\alpha_\mu}$. Precisely, we can estimate it in terms of the angle $\angle(\xi,\xi_\theta^{\alpha_\mu})$ as $$e(t,x,\xi) \approx |\xi|
|\angle(\xi,\xi_\theta^{\alpha_\mu})|^2$$ In particular in the support of the symbol ${{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$ the above angle has size $\alpha$ and the frequency has size $\mu$. Hence[^5] $$e(t,x,\zeta)
\approx \alpha^2 \mu, \qquad (t,x,\zeta) \in \text{supp }
{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$$ Here it may help to think of the constant coefficient case where $\xi_\theta^{\alpha_\mu}=\theta$, while $a-{{\tilde{a}}}= |\xi| - \xi \theta$. We introduce a local inverse for $ e(t,x,\zeta)$ in the support of ${{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$, namely $$l(t,x,\zeta) = \tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,\zeta)
e^{-1}(t,x,\zeta)$$ The cutoff symbol $ \tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$ is similar to ${{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$ but has a slightly larger support and equals $1$ in a neighbourhood of the support of ${{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$.
As defined, the operator $L(t,x,D)$ is not localized at frequency $\mu$. To remedy this we truncate its output in frequency and set $$\tilde L = \tilde S_\mu(D) L(t,x,D)$$ The properties of the operator $\tilde L$ are summarized in the following
The operator $\tilde L$ satisfies the following estimates:
a\) fixed time $L^p$ mapping properties: $$\| \tilde L \|_{L^p \to L^p} \lesssim \alpha^{-2} \mu^{-1}, \qquad 1 \leq p \leq \infty$$ b) fixed time approximate inverse of $A(t,x,D) - {{\tilde{A}}}(t,x,D)$: $$\|(A(t,x,D)- {{\tilde{A}}}(t,x,D)) \tilde L - \tilde{{\tilde{\phi}}}(t,x,D)\|_{L^p
\to L^p} \lesssim \mu^{-\frac12} +\alpha^{-2} \mu^{-1}, \quad 1 \leq
p \leq \infty$$ c) space-time $X_+$ mapping properties: $$\| \tilde L \|_{X_{+} \to X_+} \lesssim \alpha^{-2} \mu^{-1}$$
\[L\]
We first compute the regularity of the symbol $e(t,x,\zeta)$ within the support of $l$. With respect to $\xi$ this is smooth and homogeneous, therefore we only have to keep track of the order of vanishing when $\xi$ is in the $\xi_\theta^{\alpha_\mu}$ direction. With respect to $x$ there is the dependence coming from the symbol $a$, as well as the dependence due to the $\xi_\theta^{\alpha_\mu}$ direction occuring in the linearization. Since $a$ is Lipschitz in $x$ and $\xi_\theta$ is Lipschitz in $x$ and smooth on the $\alpha_\mu$ scale, within the support of $l$ we obtain $$e \in C^1_\mu S(\alpha^2 \mu,g_\alpha)
\label{amureg}$$
Combining this with the regularity of the symbol $\tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu} \in S(1,g_\alpha) $ we obtain the symbol regularity for $l$, $$l \in C^1_\mu S((\alpha^2 \mu)^{-1},g_\alpha)
\label{amurega}$$
To prove part (a) of the Lemma we observe that for fixed $(t,x)$ the symbol $l(t,x,\xi)$ is a smooth bump function of size $(\alpha^2 \mu)^{-1}$ in a rectangle of size $ \mu \times (\alpha
\mu)^{n-1}$ oriented in the $ \xi_\theta^{\alpha_\mu}$ direction. This implies that its kernel $K(t,x,y)$ is bounded by $(\alpha^2
\mu)^{-1}$ times an integrable bump function on the dual scale, $$|K(t,x,y)| \lesssim (\alpha^2 \mu)^{-1} \mu (\alpha \mu)^{n-1}
(1 + \mu |\xi_\theta^{\alpha_\mu}(t,x)(x-y)| +
\alpha \mu |\xi_\theta^{\alpha_\mu}(t,x)\wedge (x-y)|)^{-N}$$ This bound is symmetric; indeed, since $\xi_\theta^{\alpha_\mu}(t,x)$ is Lipschitz in $x$ we can replace it by $\xi_\theta^{\alpha_\mu}(t,y)$ in the above bound. Thus integrating we have $$\sup_x \int |K(t,x,y)| dy \lesssim (\alpha^2 \mu)^{-1}, \qquad
\sup_y \int |K(t,x,y)| dx \lesssim (\alpha^2 \mu)^{-1}$$ The $L^p$ bounds for $L(t,x,D)$ and also for $\tilde L$ immediately follow.
For later use in the proof we observe that within the support of $l$ we have $$|\xi_\theta^{\alpha_\mu}(t,x) \wedge \xi | \lesssim \alpha \mu$$ Then the same argument as above yields the additional bounds $$\| (\xi_\theta^{\alpha_\mu}(t,x) \wedge D)^\beta \tilde L u\|_{L^p}
\lesssim (\alpha \mu)^{|\beta|} (\alpha^2 \mu)^{-1} \| u\|_{L^p}
\label{wedga}$$
For part (b) we write $$(A(t,x,D)- {{\tilde{A}}}(t,x,D)) \tilde L - \tilde{{\tilde{\phi}}}(t,x,D) = R_1(t,x,D) +
R_2(t,x,D)$$ where $$R_1(t,x,D)=E(t,x,D)
\tilde S_\mu(D) L(t,x,D) - \tilde {{\tilde{\phi}}}(t,x,D),$$ respectively $$R_2(t,x,D)=(A_{>\mu^\frac12}(t,x,D)- {{\tilde{A}}}_{>\mu^\frac12}(t,x,D))
\tilde S_\mu(D) L(t,x,D),$$
The operator $R_1$ is localized at frequency $\mu$. The principal part cancels, and since $e \in C^1_\mu S(\alpha^2 \mu,g_\alpha)$ and $l \in C^1_\mu S((\alpha^2 \mu)^{-1},g_\alpha)$ by the pseudodifferential calculus it follows that $$R_1(t,x,D) \in C^0_\mu S((\alpha^2 \mu)^{-1},g_\alpha)$$ In addition, the symbol of $R_1$ decays rapidly away from the support of $\tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$. Hence we obtain the same kernel and $L^p$ bounds as in the case of $L(t,x,D)$.
Consider now the operator $R_2$. Since $a(t,x,\zeta)$ is Lipschitz in $x$ it follows that $|a_{>\mu^\frac12}(t,x,\zeta)| \lesssim
\mu^{-\frac12} |\zeta|$. Expanding $a_{>\mu^\frac12}(t,x,\zeta)$ in a rapidly decreasing series of spherical harmonics with respect to $\zeta$, we can separate variables and reduce the problem to the simpler case when $a_{>\mu^\frac12}(t,x,\zeta) = b(t,x)
c(\zeta)$ with $|b| < \mu^{-\frac12}$ and $c$ is smooth and homogeneous of order $1$. For the symbol $c-{{\tilde{c}}}$ we use the representation $$c(\zeta) - {{\tilde{c}}}(t,x,\zeta) = \psi(\xi_\theta^{\alpha_\mu}, \zeta)
(\xi_\theta^{\alpha_\mu}(t,x) \wedge \zeta)^2$$ where $\psi$ is smooth in both arguments and homogeneous of order $-1$ in $\zeta$. Separating variables in $\psi$ we can assume without any restriction in generality that $\psi$ depends only on $\zeta$. Then after some simple commutations we obtain $$c(D) - {{\tilde{c}}}(t,x,D) = (\xi_\theta^{\alpha_\mu}(t,x) \wedge D)^2
\psi(D) + O(1)_{L^p \to L^p}$$ To estimate this we use . The factor $\psi(D)\tilde S_\mu(D)$ yields an extra $\mu^{-1}$ factor in the $L^p$ bounds, therefore we obtain $$\| R_2(t,x,D)\|_{L^p \to
L^p} \lesssim \mu^{-\frac12}$$
Finally we prove part (c). By (a), $\tilde L$ is $L^2$ bounded with norm $O(\alpha^{-2} \mu^{-1})$, therefore it remains to prove the commutator estimate $$\| [ D_t+ A_{<\mu^\frac12}(t,x,D),\tilde S_\mu L(t,x,D)]\|_{L^\infty L^2 \to L^2}
\lesssim \alpha^{-2} \mu^{-1}
\label{lcom}$$ This is a consequence of the operator bound $$[ D_t+ A_{<\mu^\frac12}(t,x,D),\tilde S_\mu L(t,x,D)] \in L^2 C^0_\mu S(\alpha^{-2} \mu^{-1},g_\alpha)$$ To prove it we use the pdo calculus to represent the commutator as a principal term plus a second order error, $$[ D_t+ A_{<\mu^\frac12}(t,x,D),\tilde S_\mu L(t,x,D)] = \tilde S_\mu Q(t,x,D) + R(t,x,D)$$ where the principal part $q$ has symbol $$q(t,x,\xi) =-i \{\tau + a_{<\mu^\frac12}(t,x,\xi), l(t,x,\xi)\}$$ The remainder $R$ is localized at frequency $\mu$. A direct computation, using , shows that its symbol satisfies $$r \in L^2 C^0_\mu S(\alpha^{-2} \mu^{-1},g_\alpha)$$ It remains to consider the above Poisson bracket and prove that $$q \in L^2 C^0_\mu S(\alpha^{-2} \mu^{-1},g_\alpha)
\label{qsal}$$ For this we write $q$ in the form $$iq = - \tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu} q_1 e^{-2} + q_2 e^{-1}
+ q_3 e^{-1}$$ where $$q_1 (t,x,\xi)= \left\{\tau +
a_{<\mu^\frac12} , e \right\}, \qquad
q_2 (t,x,\xi)= \left\{\tau + a_{<\alpha^{-1}}, \tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}\right\}$$ respectively $$q_3 (t,x,\xi)=
\left\{ a_{\alpha^{-1}<\cdot < \mu^{\frac12}}, \tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}\right\}$$
Within the support of $\tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$ we know that $e \in C^1_\mu S(\alpha^2 \mu, g_\alpha)$ is an elliptic symbol. Hence for the first term it suffices to show that $q_1 \in
C^0_\mu S(\alpha^2 \mu, g_\alpha)$. Indeed, by definition $q_1$ is a homogeneous symbol of order $1$ which is continuous in $x$ and homogeneous in $\zeta$. In addition, we know that $e(t,x,\zeta)$ vanishes of second order in $\zeta$ at $
(t,x,\xi_\theta^{\alpha_\mu}(x,t))$ which is also invariant with respect to the $\tau+ a_{<\mu^\frac12}$ Hamilton flow. Then $q$ must vanish of second order in $\zeta$ at $(t,x,\xi_\theta^{\alpha_\mu}(x,t))$. Arguing as in the case of $e$, this implies that within the support of $\tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}$ we have $q_1 \in C^0_\mu
S(\alpha^2 \mu, g_\alpha)$.
As in we know that $q_2 \in S(1,g_\alpha)$. Also we have $a_{\alpha^{-1}<\cdot < \mu^{\frac12}} \in L^2 C^2_\mu
S(\alpha^2 \mu, g_\alpha)$ and $\tilde{{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}
\in S(1,g_\alpha)$ therefore $q_3 \in L^2 C^1_\mu S(1,g_\alpha)$.
This concludes the proof of and therefore the proof of the lemma.
To continue the estimate of term II in Case 1 we define the auxiliary trilinear form $$\begin{aligned}
E(u,v,{{\tilde{w}}}) &=& \int (D_t + A(t,x,D)) u\, v {{\tilde{w}}}dx dt + \int u (D_t -
A(t,x,-D)) v\, {{\tilde{w}}}dx dt
\\ &+& \int u v\, (D_t + {{\tilde{A}}}(t,x,D)) {{\tilde{w}}}dx dt\end{aligned}$$ With $\tilde w= \tilde L w_{\theta}^\alpha$ we write $$\begin{split}
\int u_\theta v_\theta w_\theta^\alpha dx dt =& -\int u_\theta
v_\theta\, ((A(t,x,D)- {{\tilde{A}}}(t,x,D))\tilde L-1)w_\theta^\alpha dx dt
\\ +& \int (D_t + A(t,x,D)) u_\theta\, v_\theta {{\tilde{w}}}dx dt \\ +& \int u_\theta (D_t - A(t,x,-D)) v_\theta\,
{{\tilde{w}}}dx dt
\\ +& \int u_\theta v_\theta (D_t + A(t,x,D)) {{\tilde{w}}}dx dt
\\ -& E(u_\theta,v_\theta,{{\tilde{w}}})
\end{split}
\label{longsum}$$ We bound each term separately. For the first one we write $$\begin{split}
(A(t,x,D)- {{\tilde{A}}}(t,x,D)) \tilde{ L}-1 =& (A(t,x,D)-{{\tilde{A}}}(t,x,D))\tilde
L-\tilde {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D)
\\ & + (\tilde {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D)-1)
\end{split}$$ The contribution of the first line is estimated using Lemma \[L\] (b) and for $w_\theta^\alpha$, $$\begin{split}
\bigg| \int u_\theta v_\theta [&(A(t,x,D)-{{\tilde{A}}}(t,x,D)) \tilde L
- \tilde {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D)]w_\theta^\alpha dx
dt \bigg|
\\& \lesssim (\alpha^{-2}\mu^{-1}+\mu^{-\frac12}) \|u_\theta\|_{L^\infty L^2} \|v_\theta\|_{L^\infty L^2} \|w_\theta^\alpha \|_{L^{2}
L^{\infty}}
\\& \lesssim (\alpha^{-2}\mu^{-1} +\mu^{-\frac12}) \|u_\theta\|_{X_+} \|v_\theta\|_{X_-} \|w_\theta^\alpha \|_{L^{2}
L^{\infty}}
\\& \lesssim (\alpha^{-2}\mu^{-1}+\mu^{-\frac12}) \alpha^\frac12 \mu^\frac32
\|u_\theta\|_{X_+} \|v_\theta\|_{X_-}
\|w_\theta^\alpha\|_{X_+^{\mu,\alpha,\theta}}
\end{split}$$ For the contribution of the second line we observe that $$(\tilde {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D)-1) w_\theta^\alpha = (\tilde {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D)-1) \tilde
\phi_\theta^{+,\alpha} S_\mu w$$ where the symbols $\tilde {{\tilde{\phi}}}_{\theta,\mu}^{+,\alpha} -1$ and $\tilde \phi_{\theta,\mu}^{+,\alpha} s_\mu $ have disjoint supports. Since they both belong to $S(1,g_\alpha)$, this yields a gain of a factor $(\alpha^2 \mu)^{-N}$ in , with $N$ arbitrarily large: $$\sum_\theta \| (\tilde {{\tilde{\phi}}}^{+,\alpha}_{\theta,\mu}(t,x,D) -1) w_\theta^\alpha\|_{L^2 L^\infty}^2 \lesssim
\mu^{\frac52} (\alpha^2 \mu)^{-N} \|w_\theta^\alpha\|_{X_+^{\mu,\alpha,\theta}}^2$$ This is more than we need.
For the second term in we use the $L^2$ bound for $(D_t+A)u_\theta$, the energy bound for $v_\theta$ and for ${{\tilde{w}}}$. This yields $$\left| \int (D_t +A(t,x,D)) u_\theta\, v_\theta {{\tilde{w}}}dx dt \right|
\lesssim
\alpha^{-\frac32} \mu^{\frac12} \|u_\theta\|_{X_+} \|v_\theta\|_{X_-}
\|w_\theta^\alpha\|_{X_+^{\mu,\alpha,\theta}}$$ The third term is similar.
For the fourth term in we use the energy for the first two factors combined with Bernstein derived $L^2 L^\infty$ bound for the third, $$\begin{split}
\left| \int u_\theta v_\theta\, (D_t + A(t,x,D)) {{\tilde{w}}}dx dt\right| &
\lesssim \|u\|_{X_+} \|v\|_{X_-}
\|(D_t + A(t,x,D)) {{\tilde{w}}}\|_{L^2 L^\infty} \\ &\lesssim (\alpha^2
\mu)^{-1} (\mu (\alpha
\mu)^3)^\frac12 \|u_\theta\|_{X_+} \|v_\theta\|_{X_-}
\|w_\theta^\alpha\|_{X_+^{\mu,\alpha,\theta}}
\end{split}$$
It remains to prove the estimate for $E$. Observe that the time derivatives in $E$ can be integrated out, producing contributions of the form $$\int u_\theta\, v_\theta \, {{\tilde{w}}}dx
\label{puvtw}$$ at the initial and the final time. These are estimated using energy bounds for the first two factors and the pointwise bound arising from Bernstein’s inequality for the last factor, $$\| {{\tilde{w}}}\|_{L^\infty} \lesssim (\alpha^2 \mu)^{-1} \|
w^\alpha_\theta\|_{L^\infty} \lesssim (\alpha^2 \mu)^{-1} (\mu
(\alpha \mu)^3)^\frac12
\|w^\alpha_\theta\|_{X_+^{\mu,\alpha,\theta}} = (\alpha^2
\mu)^{-\frac14}
\mu^\frac54\|w^\alpha_\theta\|_{X_+^{\mu,\alpha,\theta}}$$
This leaves us with a purely spatial trilinear form, $$\int E_0(u_\theta,v_\theta,{{\tilde{w}}}) dt$$ where $$E_0(u,v,{{\tilde{w}}}) = \int A(t,x,D) u\, v {{\tilde{w}}}- u A(t,x,-D) v \,{{\tilde{w}}}+ u
v \, {{\tilde{A}}}(t,x,D) {{\tilde{w}}}\ dx$$ The main bound for $E_0$ is provided in the next lemma.
\[leob\] Let $1 \leq \mu \lesssim \lambda$. Assume that $\xi_\theta$ is a Lipschitz function of $x$ with $|\xi_\theta -\theta| \ll 1$ and that $a \in C^1 S^1_{hom}$. Then the trilinear form $E_0$ satisfies the fixed time estimate: $$\begin{split}
|E_0(u,v,{{\tilde{w}}})| \lesssim& \ \|u\|_{L^{p_1}}
\|v\|_{L^{q_1}} \|{{\tilde{w}}}\|_{L^{r_1}} \\ & + \lambda^{-1} \| (\xi_\theta \wedge D) u\|_{L^{p_2}}
\|v\|_{L^{q_2}} \| (\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{r_2}}
\\ & + \lambda^{-1} \| u\|_{L^{p_2}}
\| (\xi_\theta \wedge D) v\|_{L^{q_2}} \| (\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{r_2}}
\\ & + \mu \lambda^{-2} \| (\xi_\theta \wedge D) u\|_{L^{p_3}}
\| (\xi_\theta \wedge D) v\|_{L^{q_3}} \|{{\tilde{w}}}\|_{L^{r_3}}
\end{split}
\label{eob}$$ for all indices $$\frac{1}{p_i} + \frac{1}{q_i} + \frac{1}{r_i} = 1, \qquad 1 \leq
p_i,q_i,r_i \leq \infty$$ and for all functions $u$, $v$ localized at frequency $\lambda$ in a small angular neighbourhood of $\theta$, respectively $-\theta$ and all $w$ localized at frequency $\mu$. \[ee\]
While any choice of $L^p$ norms is allowed in the lemma, in order to conclude the proof of the estimate for $E$ it suffices to use the set of indices $(2,2,\infty)$. We apply the lemma with $u = u_\theta$, $v = v_\theta$ and ${{\tilde{w}}}= \tilde L w^\alpha_\theta$ as above. This yields $$\begin{split}
\left|\int E_0(u_\theta,v_\theta,{{\tilde{w}}}) dt\right| \lesssim & \
\|u_\theta\|_{L^{\infty} L^2}
\|v_\theta\|_{L^{\infty} L^2} \|{{\tilde{w}}}\|_{L^{2} L^\infty} \\ & + \lambda^{-1} \| (\xi_\theta \wedge D) u_\theta\|_{L^{\infty} L^2}
\|v_\theta\|_{L^{\infty} L^2} \| (\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{2} L^\infty}
\\ & + \lambda^{-1} \| u_\theta\|_{L^{\infty} L^2}
\| (\xi_\theta \wedge D) v_\theta\|_{L^{\infty} L^2} \| (\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{2} L^\infty}
\\ & + \mu \lambda^{-2} \| (\xi_\theta \wedge D) u_\theta\|_{L^{\infty} L^2}
\| (\xi_\theta \wedge D) v_\theta\|_{L^{\infty} L^2} \|{{\tilde{w}}}\|_{L^{2} L^\infty}
\end{split}
\label{secondee}$$ Due to the angular localization, the operator $ (\xi_\theta \wedge D) $ yields a factor of $\mu^{-\frac12} \lambda$ when applied to $u_\theta$ or $v_\theta$, respectively a factor of $\alpha \mu$ when applied to $\tilde w$. Hence we obtain $$\left|\int E_0(u_\theta,v_\theta,{{\tilde{w}}}) dt\right| \lesssim
\frac{\mu^{\frac32} \alpha^\frac12}{\alpha^2 \mu} (1+ \alpha
\mu^\frac12 +\alpha \mu^\frac12+ 1)
\|u_\theta\|_{X_+^{{\lambda},\alpha,\theta}}
\|v_\theta\|_{X_-^{{\lambda},\alpha,\theta}}
\|w^\alpha_\theta\|_{X_+^{\mu,\alpha,\theta}}$$ which is acceptable since $\alpha^2 \mu \geq 1$.
Since the symbol $a$ is smooth and homogeneous of order $1$ with respect to $\xi$, we can use its representation in terms of the spherical harmonics and reduce the problem to the case when $a$ has the form $$a(x,\xi) = b(x) c(\xi)$$ where $b$ is Lipschitz continuous.
We denote by $\xi$, respectively $\eta$ the frequencies for the $u_\theta$, respectively $v_\theta$ factors in $E_0$. Then $\xi$ and $\eta$ have size $\lambda$ and are in a small angular neighbourhood of $\theta$. We expand $c$ around the line generated by $\xi_\theta$ into a linear term and a quadratic error, $$c(\xi) = \xi (\nabla c)(\xi_\theta) + \xi B(\xi,\xi_\theta) \xi$$ where $B$ is homogeneous of order $-1$ with respect to $\xi$ and can be chosen so that $$\xi_\theta B(\xi,\xi_\theta) = 0, \qquad B(\xi,\xi_\theta) \xi_\theta = 0$$ To see that this is possible we observe that after a rigid rotation we can assume that $\xi_\theta = e_1$. For $\xi = (1,\xi')$ with $|\xi'|
\ll 1$ we write the first order Taylor polynomial with integral remainder $$\begin{split}
c(1,\xi') = &\ c(1,0) + \xi' c_{\xi'}(1,0) + \xi' B(1,\xi') \xi'
\\ = &\ c_{\xi_1}(1,0) + \xi' c_{\xi'}(1,0) + \xi' B(1,\xi') \xi'
\end{split}$$ where $B$ is given by $$B(1,\xi') = \int_0^1 (1-h) \nabla^2_{\xi'} a(1,h\xi') dh$$ This extends by homogeneity to all $\xi$ in a small angle around $\theta$.
We represent $B$ as a rapidly convergent sum of terms of the form $$\lambda^{-1} F(\xi_\theta) g(\xi)$$ where $g$ is a scalar function which is bounded and smooth on the $\lambda$ scale and $F$ is a matrix inheriting the above property of $B$, $$\xi_\theta F(\xi_\theta) = 0, \qquad F(\xi_\theta) \xi_\theta = 0
\label{null}$$ So we have $$c(\xi) = \xi (\nabla c)(\xi_\theta) + \lambda^{-1} \sum
\xi F(\xi_\theta) \xi g(\xi)$$ Then we obtain the rapidly convergent series representation $$\begin{aligned}
c(\xi) - c(\eta) &=& (\xi-\eta) (\nabla c)(\xi_\theta) +\lambda^{-1} \sum (\xi - \eta)
F(\xi_\theta) \xi g(\xi) \\ &+& \lambda^{-1} \sum \eta F(\xi_\theta) (\xi -\eta) g(\eta)
\\ &+& \lambda^{-2} \sum \eta F(\xi_\theta) \xi (\xi - \eta) h(\xi) k(\eta)\end{aligned}$$ where $h$ and $k$ are smooth and bounded on the $\lambda$ dyadic scale.
We use this representation for the first two components in $E_0$. The contribution of the first term above cancels the principal part of the third component in $E_0$. We retain the other three terms though, therefore this yields the following rapidly convergent series representation for $E_0$: $$E_0(u,v,{{\tilde{w}}}) = \int u v {{\tilde{w}}}D(b (\nabla c)(\xi_\theta)) dx
+ \sum E_0^1 + \sum E_0^2 + \sum E_0^3$$ The first term is easily estimated since $b (\nabla
c)(\xi_\theta)$ is Lipschitz continuous. The first summand has the form $$\begin{split}
E_0^1 &= {\lambda}^{-1} \int F(\xi_\theta) D (Dg(D) u v)\, {{\tilde{w}}}dx
\\ & = - {\lambda}^{-1} \int D F(\xi_\theta) D g(D) u v {{\tilde{w}}}+ F(\xi_\theta) D g(D) u v D {{\tilde{w}}}\ dx
\end{split}$$ In the first term $F(\xi_\theta)$ is Lipschitz in $x$ and the $u$ derivative yields a factor of $\lambda$. For the second term on the other hand we use to estimate $$| D g(D) u F(\xi_\theta) D {{\tilde{w}}}| \lesssim |(\xi_\theta \wedge D) g(D) u| |(\xi_\theta \wedge D) {{\tilde{w}}}|$$ Commuting $g(D)$ with $(\xi_\theta \wedge D)$ we get $$|(\xi_\theta \wedge D) g(D) u| \leq |g(D) (\xi_\theta \wedge D) u| + | [
g(D), \xi_\theta \wedge D] u|$$ with the commutator $[ g(D), \xi_\theta \wedge D]$ bounded in all $L^p$ spaces. Hence $$|E_0^1| \lesssim \| u\|_{L^{p_1}} \|v\|_{L^{q_1}} \|{{\tilde{w}}}\|_{L^{r_1}} +
\lambda^{-1} \| (\xi_\theta \wedge D) u\|_{L^{p_2}} \|v\|_{L^{q_2}}
\|(\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{r_2}}$$ The second summand of $E_0$ is similar but with the roles of $u$ and $v$ reversed.
Finally, $$\begin{split}
E_0^3 &= \lambda^{-2} \int F(\xi_\theta)\, D (D h(D) u \,D k(D) v)\, {{\tilde{w}}}dx
\\ &= -\lambda^{-2} \int D F(\xi_\theta) \,D h(D) u\, D k(D) v \, {{\tilde{w}}}+ F(\xi_\theta)\, D h(D) u\, D k(D) v \, D {{\tilde{w}}}\ dx
\end{split}$$ where the matrix $F(\xi_\theta)$ is paired with the $u$ and $v$ derivatives. In the first term the two derivatives on $u$ and $v$ yield a $\lambda^2$ factor. In the second term we use as before and commute out the $h(D)$ and $k(D)$ multipliers. We obtain $$|E_0^3| \lesssim\| u\|_{L^{p_1}} \|v\|_{L^{q_1}} \|{{\tilde{w}}}\|_{L^{r_1}} + \mu
\lambda^{-2}\| (\xi_\theta \wedge D) u\|_{L^{p_3}} \| (\xi_\theta \wedge D) v\|_{L^{q_3}} \|
{{\tilde{w}}}\|_{L^{r_3}}$$ Summing up the results we get the conclusion of the Lemma.
[**Case 1, term III**]{}: This has the form $$III = \int \sum_{ \alpha > \mu^{-\frac12} }\sum_{\theta \in
O_{\alpha}}
\phi^{+,\alpha}_{\theta,\lambda}(t,x,D) S_{\lambda}u \ \tilde
\phi^{-,\alpha}_{-\theta,\lambda}(t,x,D) S_{\lambda}v \ \tilde
\phi^{+,\alpha \mu^{-1} \lambda}_{\theta,\mu}(t,x,D) S_\mu w \,dx
dt$$ In this case the summation with respect to $\theta$ is accomplished by , while for the $\alpha$ summation we simply accept a $\ln \mu$ loss. Fixing $\alpha$ and $\theta$ we set $$u_\theta^\alpha = \phi^{+,\alpha}_{\theta,\lambda}(t,x,D) S_{\lambda}u, \quad v_\theta^\alpha =
\phi^{-,\alpha}_{-\theta,\lambda}(t,x,D) S_{\lambda}v, \quad w_\theta^\alpha=\tilde \phi^{+,\alpha
\mu^{-1} \lambda}_{\theta,\mu}(t,x,D) S_\mu w.$$ and repeat the analysis for Case 1, term II. The angular localization of $u_\theta^\alpha $ and $v_\theta^\alpha$ is not used in the bounds for the first four terms in , therefore that part of the argument rests unchanged. The same applies to the bound for the fixed time integral in .
It remains to consider the bound for $E(u_\theta^\alpha,v_\theta^\alpha,{{\tilde{w}}})$. The $\alpha$ localization angle for $w_\theta^\alpha$ is now $\alpha
\mu^{-1} \lambda$, therefore part (b) of Lemma \[L\] gives $$\| {{\tilde{w}}}\|_{X_+} \lesssim \frac{\mu}{\alpha^2 \lambda^2}
\|w_\theta^\alpha\|_{X_+}$$ This is stronger than in the previous case because it gives a high frequency gain. Now we are able to use Lemma \[ee\] with exponents $(3,2,6)$ to obtain $$\begin{split}
|\int E_0(u_\theta^\alpha,v_\theta^\alpha,{{\tilde{w}}}) dt| \lesssim &\
\|u_\theta^\alpha\|_{L^2 L^3} \|v_\theta^\alpha\|_{L^{\infty} L^2}
\|{{\tilde{w}}}\|_{L^{2} L^6}
\\ & + \lambda^{-1} \| (\xi_\theta \wedge D) u_\theta^\alpha\|_{L^2 L^3} \|v_\theta^\alpha\|_{L^{\infty} L^2}
\| (\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{2} L^6} \\ & + \lambda^{-1} \|
u_\theta^\alpha\|_{L^2 L^3} \| (\xi_\theta \wedge D)
v_\theta^\alpha\|_{L^{\infty} L^2} \| (\xi_\theta \wedge D) {{\tilde{w}}}\|_{L^{2}
L^6} \\ & + \mu \lambda^{-2} \| (\xi_\theta \wedge D)
u_\theta^\alpha\|_{L^2 L^3} \| (\xi_\theta \wedge D)
v_\theta^\alpha\|_{L^{\infty} L^2} \|{{\tilde{w}}}\|_{L^{2} L^6}
\end{split}$$ Due to the angular localization on the $\alpha$ scale for $u_\theta^\alpha$ and $v_\theta^\alpha$, respectively on the $\alpha \mu^{-1} \lambda$ scale for $w_\theta^\alpha$, all $(\xi_\theta \wedge D)$ operators above yield $\alpha \lambda$ factors. Hence, taking advantage of the Strichartz estimates, we obtain $$\begin{split}
|\int E_0(u_\theta^\alpha,v_\theta^\alpha,{{\tilde{w}}}) dt| \lesssim&\
\frac{\mu}{\alpha^2 \lambda^2} \alpha^2 \lambda\ \lambda^{\frac5{12}} \mu^\frac56
\|u_\theta^\alpha\|_{X_+^{{\lambda},\alpha,\theta}} \|v_\theta^\alpha\|_{X_-^{{\lambda},\alpha,\theta}}
\|w_\theta^\alpha\|_{X_+^{\mu,\frac{\alpha{\lambda}}{\mu},\theta}}
\\
=&\ \lambda^{-\frac7{12}} \mu^\frac{11}6
\|u_\theta^\alpha\|_{X_+^{{\lambda},\alpha,\theta}} \|v_\theta^\alpha\|_{X_-^{{\lambda},\alpha,\theta}}
\|w_\theta^\alpha\|_{X_+^{\mu,\frac{\alpha{\lambda}}{\mu},\theta}}
\end{split}$$ which is satisfactory since $\lambda \gtrsim \mu$.
We conclude this case with two remarks. First, in this context the proof of Lemma \[ee\] is somewhat of an overkill. In fact, it would suffice to linearize separately $a(t,x,\xi)$ and $a(t,x,\eta)$ around $\xi_\theta$ and use the fact that the symbol $a(t,x,\xi) - {{\tilde{a}}}(t,x,\xi)$ has size $\alpha^2 \lambda$ at frequency ${\lambda}$ in $H_\alpha S_\alpha(\theta)$. Secondly, the endpoint Strichartz estimate is only used here for convenience; there is some flexibility in choosing the indices.
[**Case 1, term IV.**]{} This has the form $$IV = \int \sum_{ \alpha > \mu^{-\frac12} }\sum_{\theta \in
O_{\alpha}}
\phi^{+,\alpha}_{\theta,\lambda}(t,x,D) S_{\lambda}u \ \tilde
\phi^{-,\alpha}_{-\theta,\lambda}(t,x,D) S_{\lambda}v \ \phi^{+,\alpha
\mu^{-1} \lambda}_{\theta,\mu}(t,x,D) S_\mu w \,dx dt$$ Again the summation with respect to $\theta$ is accomplished by , while for the $\alpha$ summation we simply accept a $\ln \mu$ loss. This term is better behaved because the symbol $$\phi^{+,\alpha}_{\theta,\lambda}(x,\xi)\, \tilde
\phi^{-,\alpha}_{-\theta,\lambda}(x,\eta) \,\phi^{+,\alpha \mu^{-1}
\lambda}_{\theta,\mu}(x,\zeta)$$ vanishes on $H=\{\xi+\eta + \zeta = 0\}$. Precisely, in the support of the above symbol we have $$|\xi| \approx \lambda,\ |\xi \wedge \xi_\theta^\alpha| \lesssim \alpha
\lambda, \qquad |\eta| \approx \lambda,\ |\eta \wedge
\xi_\theta^\alpha| \approx C \alpha
\lambda, \qquad |\zeta| \approx \lambda,\ |\zeta \wedge
\xi_\theta^\alpha| \lesssim \alpha
\lambda.$$ This leads to $$|(\xi + \eta +\zeta) \wedge
\xi_\theta^\alpha| \approx C \alpha
\lambda
\label{offdiag}$$ This can be taken advantage of in a direct computation in the above formula. Including the dyadic frequency localizations into the $\phi$’s, each term in $IV$ has the integral representation $$\int \phi^{+,\alpha}_{\theta,\lambda}(t,x,\xi) \hat u(\xi) \
\tilde \phi^{-,\alpha}_{-\theta,\lambda}(t,x,\eta) \hat v(\eta) \
\phi^{+,\alpha \mu^{-1} \lambda}_{\theta,\mu}(t,x,\zeta) \hat
w(\zeta)\, e^{i x(\xi+\eta+\zeta)} \, d\xi d\eta d\zeta dx dt$$ Defining the spatial elliptic operator $F$ with symbol $$f(t,x,\xi) = (\xi \wedge \xi_\theta^\alpha)^{2N}$$ we have $$F(t,x,D_x) e^{i x(\xi+\eta+\zeta)} = |(\xi + \eta +\zeta) \wedge
\xi_\theta^\alpha|^{2N} e^{i x(\xi+\eta+\zeta)}$$ Hence integration by parts in the above formula leads to $$\int \psi(t,x,\xi,\eta,\zeta) \hat u(\xi) \, \hat v(\eta) \, w(\zeta) e^{i
x(\xi+\eta+\zeta)} \, d\xi d\eta d\zeta dx dt$$ where the new symbol $\psi$ is $$\psi(t,x,\xi,\eta,\zeta) = F^*(t,x,D_x) \left( \frac{\phi^{+,\alpha}_{\theta,\lambda}(t,x,\xi) \tilde
\phi^{-,\alpha}_{-\theta,\lambda}(t,x,\eta) \phi^{+,\alpha
\mu^{-1} \lambda}_{\theta,\mu}(t,x,\zeta) }{ |(\xi + \eta +\zeta) \wedge
\xi_\theta^\alpha|^{2N}} \right)$$ In the support of the numerator the bound holds. Hence separating the variables we can represent the denominator as a rapidly convergent series with terms $$(\alpha \lambda)^{-2N} \chi_{< \alpha {\lambda}}( \xi \wedge \xi_\theta^\alpha) \chi_{C \alpha {\lambda}}
( \eta \wedge \xi_\theta^\alpha)
\chi_{< \alpha {\lambda}}( \zeta \wedge \xi_\theta^\alpha)$$ where each of the $\chi$’s above is a unit bump function on the $\alpha \lambda$ scale. Thus they can be included in the corresponding $\phi$ factors. Due to the $S(g_\alpha)$ regularity of the $\phi$ factors, each derivative $ \xi_\theta^\alpha \wedge D$ applied to them yields an $\alpha^{-1}$ factor. Thus $\psi$ is represented as a rapidly convergent series of products of the form $$(\alpha^2 \lambda)^{-2N} \psi^{+,\alpha}_{\theta,\lambda}(t,x,\xi) \tilde
\psi^{-,\alpha}_{-\theta,\lambda}(t,x,\eta) \psi^{+,\alpha
\mu^{-1} \lambda}_{\theta,\mu}(t,x,\zeta)$$ where the $\psi$ factors have the same support and regularity as the corresponding $\phi$’s. The integral above is similarly represented as a rapidly convergent series with terms of the form $$(\alpha^2 \lambda)^{-2N} \int \psi^{+,\alpha}_{\theta,\lambda}(t,x,D) S_\lambda u \,\tilde
\psi^{-,\alpha}_{-\theta,\lambda}(t,x,D) S_{\lambda}v \, \psi^{+,\alpha
\mu^{-1} \lambda}_{\theta,\mu}(t,x,D) S_\mu w \,dx dt$$ Since $\alpha > \mu^{-\frac12}$, the factor in front of the above integral allows us to exchange low frequencies for high frequencies. This suffices in order to bound the last integral using Strichartz estimates.
[**Case 1, term V**]{}
This is similar to Case 1, term $IV$. This time in the support of the symbol $$\phi^{+,\alpha}_{\theta,{\lambda}}(t,x,\xi)\, \tilde
\phi^{-,\alpha}_{-\theta,{\lambda}}(t,x,\eta)\,
\tilde\phi^{+,\beta}_{\theta,\mu}(t,x,\zeta)$$ we have $$|\xi| \approx \lambda,\ |\xi \wedge \xi_\theta^\alpha| \lesssim \alpha
\lambda, \qquad |\eta| \approx \lambda,\ |\eta \wedge
\xi_\theta^\alpha| \approx C \alpha
\lambda, \qquad |\zeta| \approx \lambda,\ |\zeta \wedge
\xi_\theta^\alpha| \approx C\beta
\lambda.$$ Hence $$|(\xi + \eta +\zeta) \wedge
\xi_\theta^\alpha| \approx C \alpha
\lambda$$ therefore the symbol above is supported at distance $\beta {\lambda}$ from the diagonal $H$. Hence integrating by parts as in the previous case we gain arbitrary powers of $(\alpha \beta {\lambda})^{-1}$. Then we can close the argument using Strichartz type estimates.
[**Case 2**]{}, $1 < d < \mu$. This requires only minor changes, which we describe in what follows. We still consider the five terms in the trilinear decomposition , but we replace the smallest localization angle $\mu^{-\frac12}$ by $d^{\frac12} \mu^{-\frac12}$.
[**Case 2, term I**]{}. Here we need to sum expressions of the form $$I = \int \phi^{+,d^{\frac12}\mu^{-\frac12}}_{\theta,\lambda}(t,x,D) S_{\lambda}u \
\phi^{-,d^{\frac12}\mu^{-\frac12}}_{-\theta,\lambda}(t,x,D) S_{\lambda}v\
\phi^{+,d^{\frac12}\mu^{-\frac12}}_{\theta,\mu}(t,x,D) S_\mu w \,dx$$ over $\theta \in O_{d^{\frac12}\mu^{-\frac12}}$. Each term is bounded by combining the energy estimate for the first factor, the $L^4 L^2$ bound for the second and for the third.
[**Case 2, term II.**]{} Here we use for the summation of expressions of the form $$II = \int \phi^{+,d^{\frac12}\mu^{-\frac12}}_{\theta,\lambda}(t,x,D) S_{\lambda}u\
\phi^{-,d^{\frac12}\mu^{-\frac12}}_{-\theta,\lambda}(t,x,D) S_{\lambda}v\,
\sum_{\alpha > d^{\frac12}\mu^{-\frac12}} \tilde
\phi^{+,\alpha}_{\theta,\lambda}(t,x,D) S_\mu w dx$$ over $\theta \in O_{d^{\frac12}\mu^{-\frac12}}$. We use the same operator $L$, the same function $\tilde w$ and the same trilinear form $E$. In the first, second and fourth terms are estimated in the same way, but using the $L^4 L^2$ bound for the second factor. In the third term we lose a power of $d$, $$\begin{aligned}
\left|\int u_\theta (D_t - A(t,x,-D)) v_\theta {{\tilde{w}}}dx\right| &\lesssim&
\|u_\theta\|_{L^\infty L^2} \|(D_t - A(t,x,-D)) v_\theta\|_{L^2}
\|{{\tilde{w}}}\|_{L^2 L^\infty}\\ & \lesssim&
\|u_\theta\|_{X_+} d^{\frac34} \|v_\theta\|_{X_{-,d}} \frac{1}{\alpha^2 \mu}
\alpha^{\frac12} \mu^{\frac32} \|w_\theta^\alpha\|_{X_+^{\mu,\alpha,\theta}}
\\ &\lesssim & \left(\frac{d}{\alpha^2 \mu}\right)^\frac34 \mu^{\frac54}
\|u_\theta\|_{X_+}\|v_\theta\|_{X_{-,d}} \|w_\theta\|_{X_+^{\mu,\alpha,\theta}}\end{aligned}$$ But this is still acceptable due to the reduced range for $\alpha$, namely $\alpha^2 \mu \geq d$.
In the expression there is a $d^{\frac14}$ loss in the $L^2$ bound for $v_\theta$, but this is compensated for by the previously unused $(\alpha^2 \mu)^{-\frac14}$ factor in the pointwise bound for ${{\tilde{w}}}$.
Finally, for the $E_0$ bounds we reuse but with all the $v_\theta$ factors estimated in $L^2$. This produces an extra $d^{-\frac14}$ gain. On the other hand, the angular localization for $u_\theta$ and $v_\theta$ is worse. Precisely, the operator $
(\xi_\theta \wedge D) $ yields a factor of $d^\frac12 \mu^{-\frac12}
\lambda$ when applied to $u_\theta$ or $v_\theta$, respectively a factor of $\alpha \mu$ when applied to $\tilde w$. Hence we obtain $$\left|\int E_0(u_\theta,v_\theta,{{\tilde{w}}}) dt\right| \lesssim
\frac{\mu^{\frac32} \alpha^\frac12}{d^{\frac14} \alpha^2 \mu} (1+
d^\frac12 \alpha
\mu^\frac12 +d^\frac12 \alpha
\mu^\frac12+ d) \|u_\theta^\alpha\|_{X_+^{{\lambda},\theta,\alpha}}
\|v_\theta^\alpha\|_{X_-^{\lambda,\theta,\alpha}}
\|w_\theta^\alpha\|_{X_+^{\mu,\theta,\alpha}}$$ This is still acceptable since $\alpha^2 \mu \geq d$.
[**Case 2, term III.**]{} Compared to the similar argument in Case 1, the following modifications are required:
\(i) The third term in is treated as in Case 2, term II.
\(ii) In the bound for $E_0$, the $L^\infty L^2$ norms are replaced by $L^4 L^2$ in all the $v_\theta$ factors.
[**Case 2, terms IV,V.**]{} These are identical to Case 1.
**Acknowledgement**
Both authors would like to thank MSRI for the hospitality in the Fall 2005 semester, where part of this article was written. Both authors were supported in part by the NSF grant DMS-0301122.
[^1]: Throughout this paper we will use the standard notation $S(m,g)$, while in [@TG] we used for $S(1,g)$ the shorter one: $S(g)$.
[^2]: Here $n$ stands for the space dimension
[^3]: This corresponds to the classical wave packets which are localized on the scale of the uncertainty principle. Above this threshold we are dealing with generalized wave packets, which may have a more complex structure, see [@MR2153517] and [@TG]
[^4]: modulo tails which are rapidly decreasing on the $g_\alpha$ scale
[^5]: here we switch to the letter $\zeta$ for the frequency, as the following analysis refers to the region at low frequency $\mu$ corresponding to the last factor $w$ in the trilinear form.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present an experimental and theoretical study of cascaded high order Stimulated Brillouin Scatterings (SBS) in single mode fibers. It is shown that because of the back-scattering nature of the process, feedback in the input port is needed for obtaining a significant cascaded effect in nonresonant systems. We also discuss similarities to nonlinear photorefractive processes.'
author:
- 'Assaf Ben-Bassat, Ariel Gordon and Baruch Fischer'
title: 'High order stimulated Brillouin scattering in single-mode fibers with strong feedback'
---
Introduction
============
Stimulated Brillouin Scattering (SBS) has a long research history as a basic phenomenon and as a tool in many contexts and materials. We mention here two important connections to SBS that were intensively studied in the last three decades. The first one is the link to phase conjugation [@zeldovich]. It was found that the reflection of focused light beams in various media gave in some cases a phase conjugate replica of the input beam. This method gave maybe the first demonstration of phase conjugation, and later generated many activities. Another wave of research was related to SBS in fibers [@agrawalbook], first in multi-mode fibers, and then more intensively to single mode fibers. Here the conventional meaning of phase conjugation in the spatial or pictorial aspect is meaningless, especially when we consider single mode fibers. Nevertheless, the nonlinear coupling efficiency becomes very high even for low light powers at the mW regime, due to the light confinement along large distances in the fiber, compared to limited focused lengths that can be obtained in free space propagation. Therefore, SBS became a crucial factor that has to be considered in fiber-optic communications. It is usually an effect that must be eliminated to allow the light propagation without losing a big fraction of it to reflections. It is similar to another effect that was used for phase conjugation, the stimulated Raman scattering (SRS). In the SRS case however, there were found important uses in fiber-optic communications. The main one is the use as broadband amplifiers, especially at the important $1.5 \mu m$ wavelength regime. A difference between SRS and SBS is the magnitude of the frequency shift of the reflected light (Stokes wave) compared to the input, originating from the vibration frequency of the relevant medium entity involved in the nonlinear process. This frequency shift is in fibers on the order of $~10 GHz$ for SBS and $~13THz$ in fibers for SRS at the $1.5\mu
m$ wavelength regime.
In this paper we focus our attention on a cascaded SBS process in single-mode fibers. Therefore the present work doesn’t offer any direct use for phase conjugation. Nevertheless, it can be meaningful for other SBS schemes, in free space and multi-mode fibers, where “spatial” phase conjugation is applicable. Additionally, one might find possible applications by using the self frequency shifts that are in the order of future dense WDM (wavelenght division multiplexing) technologies, believed to be heavily used in future fiber-optic communications.
For this paper, presented in the context of works on dynamic holography and photorefractive optics, it is worthwhile to mention some similarities between SBS, SRS and photorefractive four-wave mixing. Pioneering work was done at the early stages when photorefractive materials have been started to be a part of the field of nonlinear-optics, and was used for wave mixing, phase conjugation and oscillators, at a few places around the world: in Kiev [@Kiev] (by Kukhtarev, Markov, Odulov, Soskin, and Vinetskii), at Thomson CSF [@Huignard] (by Huignard, Spitz, Aubourg, and Herriau, at the University of Southern California [@Feinberg] (by Feinberg and Hellwarth), and at Caltech [@PR0; @PR1] (by Cronin-Golomb, Fischer, White and Yariv. Later, a huge stream of research was done around the world in many aspects of photorefractivity. We mention a few works on photorefractive wave-mixing, done in our group at Technion [@PR2; @PR3; @PR4; @PR5; @PR6; @PR7; @PR8; @PR9] (by Fischer, Sternklar, Weiss and Segev), that can be associated to the present work on SBS. The first link is to a class of self oscillation processes in photorefractive media [@Feinberg; @PR1; @PR2]. Like in SBS, passive or self-pumped phase conjugate mirrors can be obtained. Here four-wave mixing [@PR3; @PR4] gives spontaneous phase conjugate reflection via pump beams which are self generated and can be regarded as the “internal” crystal “waves” (albeit light waves), like the self generated “sound waves” in the SBS case [@zeldovich]. The phase conjugation property can be also explained by similar arguments, that among all possible scattering, the phase conjugate pattern which is an oppositely propagating replica of the input light wave (and therefore coincide in space), experiences the highest gain, and thus wins out and prevails over all other scatterings [@zeldovich; @PR5; @PR6]. Another similarity is the self frequency shift of the reflection with respect to the input light beam [@PR7; @PR8; @PR9]. In the photorefractive case, the shift is typically in the $1-10^{-3}Hz$ region, depending on the photorefractive buildup time constant, which is much slower that the relevant nonlinear effect in the SBS and SRS cases. Additionally, for photorefractive wave mixing one can also think of cascaded self reflections in an open or closed cavity. Specific examples can be two-beam coupling, via reflection gratings where the beams are almost counter-propagating, or resonators that give high order oscillations.
It is also worthwhile to mention connections of fibers to phase conjugation. In fact, one of the first suggestions for methods of phase conjugation and its uses dealt with the restoration of images transmitted through multi-mode fibers [@PC1]. It was proposed there to use nonlinear three-wave mixing to phase conjugate the distorted image transmitted through a fiber, and retransmit it through an identical fiber section, such that the second propagation exactly cancels the phase distortion of the first section. The idea was later demonstrated [@PC2] in a single section fiber with a round-trip propagation in the same fiber, because of the difficulty to get two identical multi mode fibers. Another idea in the early stages of research that gained a lot of recent attention in the fiber-optic communication community was to compensate for dispersion in single mode fibers by using the phase conjugation property of flipping the spectral band of a time dependent signal [@PC3]. Again a two section scheme with phase conjugation between them, can provide a perfect compensation.
SBS in fibers has been studied intensively throughout the years. Input light at power levels on the order of 10 mW is strongly backscattered, producing a frequency down shifted Stokes wave, due to nonlinear interaction of light and sound waves. The associated threshold depends on the light losses. The simplest configuration for studying SBS in optical fibers is just a long enough optical fiber, typically of a few km, with a good termination at its far end, to avoid feedback. Much work has been done analyzing this system, the SBS threshold [@feedback] and the reflection strength, which is the ratio between the final power of the Stokes wave and the initial power of the pump.
When feedback is added to the fiber from the far end termination, or from other reflectors or simply by forming a ring cavity, the SBS threshold can be lowered significantly and can even result in oscillation and a Brillouin laser.
In a fiber with no feedback, SBS is well described by the common “three wave model”: the pump wave, the Stokes wave and the mediating sound wave. Second order SBS [@2order], which is the generation of yet another, secondary Stokes wave, by SBS from the first SBS wave, is known but considered to be weak for systems without gain, and is usually neglected for such systems. However, for systems with strong feedback, higher SBS orders can be significant, and taking them into account is crucial for understanding the physics of such systems.
In this work we investigate a system with strong feedback, where several orders, a cascade of SBS, are generated, in a non-resonant system (open cavity, with only one side feedback ). We realize that it is necessary to put the feedback at the input port of the fiber to allow the each SBS backscattering order to generate its own SBS in an optimized intensity profile along the fiber interaction path. We compare the experimental results to theoretical analysis, and trace clearly the vast effects of second and third order SBS. We find good agreement between the experimental data and the multiple-order SBS theory.
SBS without feedback
====================
In the simple Brillouin scattering scheme in long fibers most of the energy of the input laser can be transferred to the Stokes wave. Therefore one would expect that the Stokes wave would pass its energy to a counter propagating Stokes wave moving again in the direction of the input laser beam, and with a frequency of $\omega_0-2\Omega$. Then one can ask if and what order can be reached in such a chain of cascaded SBS system? (the n-th order having an additional frequency shift, such that $\omega_0-n\Omega$.) We will mathematically solve the coupled wave equations for these three waves and then show experimentally that for a regular system, even the generation of the second order (third wave) SBS is negligible. Later we show that a chain process that builds many strong high order SBS is possible by adding feedback via a simple reflector at the input port.
The common three-wave SBS model (one acoustic and two light waves) in steady state, is described by two coupled differential equations, for the intensities of the pump and the Stokes wave [@boydbook]: $$\frac{dI_{1}}{dz}=-gI_{1}I_{2}$$ $$\label{I2}
\frac{dI_{2}}{dz}=-gI_{1}I_{2},$$ where $I_{1}$ and $I_{2}$ are the intensities of the incident and the Stokes waves respectively. $g$ is the Brillouin gain parameter, that depends on the fiber. These equations neglect losses in the fibers and can be integrated analytically (as well as the ones with nonzero losses [@Chen]) to yield the following intensity profiles: $$\begin{aligned}
\label{IntI2}
I_{2}(z)=\frac{I_{2}(0)(I_{1}(0)-I_{2}(0))}{I_{1}(0)e^{gz(I_{1}(0)-I_{2}(0))}-I_{2}(0)}
\cr \cr I_{1}(z)= I_{1}(0)-I_{2}(0)+I_{2}(z)\end{aligned}$$ We choose our origin $z=0$ at the pump input port of the fiber. Then $I_{1}(0)$ is the intensity of the incident wave, and $I_{2}(0)$ is the output intensity of the Stokes wave.
It is seen that if $I_2(L)$, the intensity of the Stokes wave at the far side of the fiber, is zero, then $I_2(z) \equiv 0$, which means that there is no Stokes wave at all. This reflects the fact that the Stokes wave must start from a seed, whose source is noise in the system. According to several approaches [@boyd1; @boyd2], it is important to think of the noise as distributed all over the fiber. We shall follow here the simple case of a noise seed at the far side of the fiber, on the order of 1nW.
Requiring $I_{2}(L)=\varepsilon$ in (\[IntI2\]) yields a relation between the incident power $I_1(0)$ and the power of the Stokes wave $I_2(0)$ in terms of a transcendental equation. We shall denote this relation by $$\label{B}
I_2(0)=B(I_1(0)).$$ The function $B$ can be approximated, for low input intensities, by $$\label{BB}
I_2(0)=\varepsilon e^{g L I_1(0)}$$ This approximation is good as long as $\varepsilon e^{g L
I_1(0)}<<I_1(0)$ i. e. for small enough incident intensities $I_1(0)$, and can be obtained as well by the non-depleted pump approximation.
Equations (\[I2\]) neglect losses in the fiber. One of the outcomes of losses in the fiber is the existence of a threshold for SBS. It starts only if the incident beam is intense enough. Otherwise, losses suppress the Stokes wave. Moreover, the transfer of energy from the incident beam to the Stokes wave lasts only while the intensity of the incident wave remains above the threshold. Losses are also known to shorten the effective length of the fiber, so the physical length in \[BB\] is replaced by an effective length [@agrawalbook] $$\label{leff}
L_{eff}= \frac 1 \alpha (1-e^{-\alpha L}),$$ where $\alpha$ is the fiber loss coefficient.
Second order SBS requires a three optical wave model. The coupled wave equations for the intensities are given by [@2order; @boydbook]:
$$\begin{aligned}
\label{3waves}
\frac{dI_1}{dz}&=&-gI_1I_2\cr\cr\frac{dI_2}{dz}&=&-gI_1I_2+gI_2I_3\cr\cr\frac{dI_3}{dz}&=&gI_2I_3\end{aligned}$$
$I_{1}$ the incident wave, $I_{2}$ the back scattered Stokes wave, $I_{3}$ the Stokes wave generated by $I_{2}$, which propagates in the same direction as $I_1$, etc. For three optical waves the equations can be integrated analytically [@2order], but unfortunately the three integration constants appearing in the solution are again transcendental functions of the boundary conditions. One can easily verify that $$\label{C}
C_1=I_{1}- I_{2}+ I_{3} \quad{\rm and}\quad C_2=I_{1} I_{3}$$ are constants of motion. Then we can find that $$\label{I1}
I_{1}(z)=\frac{C_1}{2}+(\frac q 2) \frac{1+C_3 e^{-qz}}{1-C_3
e^{-qz}}; \quad q=\sqrt{C_1^2-4C_2}$$ $C_3$ is the third integral, and $I_{2}(z), I_{3}(z)$ can be obtained from \[I1\] through \[C\]
In spite of the similarity of the terms $gI_{1} I_{2}$ and $
gI_{2} I_{3}$ in \[3waves\], power exchange between $I_{1}$ and $I_{2}$ is much more efficient than between $I_{2}$ and $I_{3}$. The difference stems form the boundary conditions. For a system without feedback $I_{3}$ has the initial value of $\varepsilon$ at $z=0$ and grows as $z$ increases, whereas $I_{2}$ decays as $z$ increases, keeping their product small all the way to $z=L$.
For a system without feedback or gain, the second order SBS is weak compared to the pump and to the first order SBS. Indeed, from the second relation of Eq. \[C\] one obtains $$\label{3waves1}
I_{3}(z)=I_{1}(0) \frac{\varepsilon}{I_{1}(z)}$$ Since $I_{1}(0)$ and $I_{1}(z)$ are roughly of the same order of magnitude, one concludes that the second order SBS $I_{3}(z)$ is not high above the noise level $\varepsilon$.
To verify Eq.(\[3waves1\]), we measured $I_3$ using a very simple experimental setup shown in Fig. \[3setup\]. The fiber we used was SMF-28, of 25 kilometers long. The output spectrum at $Z=L$, shown in Fig. \[osa3\], is composed of three wavelengths, $I_3(L),I_1(L)$ and the reflections from the input isolator of the first Stokes wave $I_2$. In Fig. \[i3g\] we have plotted $I_{3}(L)$ vs. $I_{1}(0)$ . In the range of input intensities we have applied, $I_{1}(L)$ was weakly dependent on $I_{1}(0)$ and was about 1.5mW. We observe that $I_{3}(L)$ changes linearly with $I_{1}(0)$ as expected from Eq. (\[3waves1\]). The slope can be related to $\varepsilon$ to yield $\varepsilon\approx 1nW$.
![Experimental setup (regular - without feedback elements) for examining SBS .[]{data-label="3setup"}](3setup.eps){width="7.7cm"}
![Optical spectrum of a simple SBS process.[]{data-label="osa3"}](osa3.eps){width="7.7cm"}
![Output power of second stokes $(I_3)$ vs. the input power $(I_1)$ for a double SBS process without feedback.[]{data-label="i3g"}](i3g.eps){width="7.7cm"}
We thus summarize that a cascaded SBS beyond the first order SBS without feedback elements is very weak. Nevertheless we show below that with proper boundary conditions with one reflector, strong higher order SBS can be generated.
System with feedback and high order SBS
========================================
We have seen that the Stokes waves generated by SBS don’t generate their own Stokes waves because of the opposite growth direction along the fiber of the “pump” and its SBS product. One way to cascade many Brillouin scattered waves is by intervening in the setup, causing every set of waves to resemble a two wave system. Fig. \[multi\] represents a suggested setup we check experimentally. The input laser beam enters the system through a fiber Bragg grating. The initial wave that start the cascading process is obtained at the output of the grating. We will denote this wave as $I_1$. It propagates to the right, generates a Stokes wave $I_2$ that propagates to the left. We know that $I_2$ doesn’t generates its own Stokes, however when $I_2$ is reflected back from the grating it will create a Stokes wave travelling again to the left, since after the reflection the waves $I_3$ and $I_4$ behave according to the two wave system equations. $I_3$ begins with a large power at the Bragg grating and is depleted only by its Stokes wave $I_4$, which means that the coupled equations for two waves can be used. This behavior is repeated for $I_3$ and $I_4$, $I_5$ and $I_6$ and so on. From understanding how this system works we can easily conclude that the coupling between every pair of waves is only through the boundary conditions of each pair. For the first pair the known boundary conditions are given by $I_1(0)$ and $I_2(l_{eff})$, and for the second pair by $I_2(0)$, which is the solution of the first pair, and by $I_3(l_{eff})$ and so on. Each pair of waves gets it boundary condition from the solution of the previous pair.
We first present the experimental result showing the generation of strong high orders. The experimental setup is given in Fig. \[multisetup\]. We used a 25km long single-mode fiber. Since every Stokes wave is down shifted by 10.3GHz from its “pump”, we used a broadband Bragg grating which can reflect all the Stokes waves with approximately the same reflectivity. Knowing the input power to our system and the reflection function of the grating is sufficient to calculate the output spectra of both output1 and output2. We expect to obtain at output port 2 a multiple-peaks spectrum, with a 10.3GHz spacing, each of power $I_{th}$, since we took a long enough fiber for the incident wave to be exhausted down to the SBS threshold power. At output port 1 we should also see a multiple-peaks spectrum. The first peak is due to the direct reflection from the Bragg grating of the input laser with power of $I_{in}\cdot r$, and all the rest are the back scattered Stokes waves. In Figs. \[out1\],\[out2\] we show the output spectra from ports 1 and 2. The central four strong lines belong to the input and the first three SBS orders. We can see additional two weaker lines at the right side giving the 4th and 5th SBS orders, and at outpot port 2 additional three four-wave mixing products, resulting mainly from the mixing of the input wave with the strong first SBS orders.
![Cascaded SBS setup.[]{data-label="multi"}](multi1.eps){width="7.7cm"}
For the theoretical part we write the coupled wave equations. We solved numerically the equations for the first eight waves in the system and compared them to the experiment.
The coupled equations for the eight waves are: $$\begin{aligned}
\label{8waves}
\frac{dI_1}{dz}&=&-gI_1I_2\cr\cr
\frac{dI_2}{dz}&=&-gI_1I_2+gI_2I_5\cr\cr
\frac{dI_3}{dz}&=&-gI_3I_4\cr\cr
\frac{dI_4}{dz}&=&-gI_3I_4+gI_4I_7\cr\cr
\frac{dI_5}{dz}&=&-gI_5I_6+gI_2I_5\cr\cr
\frac{dI_6}{dz}&=&-gI_5I_6\cr\cr
\frac{dI_7}{dz}&=&-gI_7I_8+gI_4I_7\cr\cr
\frac{dI_8}{dz}&=&-gI_7I_8\end{aligned}$$
![Experimental arrangement for multi stage SBS.[]{data-label="multisetup"}](multisetup.eps){width="7.7cm"}
![Output spectra of multi stage SBS measured at output 1.[]{data-label="out1"}](out1.eps){width="7.7cm"}
![The experimental output spectrum of multi order SBS measured at the output port 2.[]{data-label="out2"}](out2.eps){width="7.9cm"}
We note that the all SBS orders (even $n$ waves) are generated as they propagate to the left direction, but then they are reflected by the mirror at the left (input) side. This reflection enables the cascaded process by generating the SBS. Thus the definition of the eight wave used in the equations is as follows:$I_1$: frequency $\omega$ propagating to the right.$I_2$: frequency $\omega-\Delta\omega$ propagating to the left. $I_3$: frequency $\omega-\Delta\omega$ propagating to the right. $I_4$: frequency $\omega-2\Delta\omega$ propagating to the left. $I_5$: frequency $\omega-2\Delta\omega$ propagating to the right. $I_6$: frequency $\omega-3\Delta\omega$ propagating to the left. $I_7$: frequency $\omega-3\Delta\omega$ propagating to the right. $I_8$: frequency $\omega-4\Delta\omega$ propagating to the left.
For the boundary conditions we have at the left side ($z=0$) the reflectivity ratio $r$ between waves $I_3 \,\&\, I_2$, $I_5 \,\&\,
I_4$ and $I_7 \,\&\, I_6$, and at the right side ($z=L$) the thermal noise ($I_2,I_4,I_6\& I_8$) needed for the SBS. Thus: $$\begin{aligned}
\label{bc}
&& I_{1}(z=0)=I_p \cr && I_{n}(z=L)=\varepsilon \phantom{I_{n}}
\quad \, for \, n=2,4... \cr && I_{n}(z=0) = r I_{n-1} \quad \,
for \, n=3,5...\end{aligned}$$
We don’t elaborate here on the simulation results, that will be given elsewhere, but note that they show a plausible match, although not complete, to the experiments. There remains questions and ingredients that have to be considered. An important point is the way that the seeding noise is incorporated into the system. In a realistic model it should be taken as a stochastic source distributed along the fiber. Additionally, other elements, such as four-wave mixing and losses, should be included in some cases in the calculations.
Brillouin Laser
===============
Understanding the simple cascading process for the multiple Stokes waves can lead to a much more efficient setup for creating the multi wavelength comb of Brillouin Stokes waves. The setup, shown in Fig. \[blaser\] is in the form of a long laser with feedback in both sides of the cavity, and an external injected seed to start the scattering process.
![Experimental setup for a Brillouin laser.[]{data-label="blaser"}](blaser.eps){width="7.7cm"}
When the laser operates without the externally injected signal its spectrum is governed only by the reflectivity spectrum of both gratings, but when we start injecting an external laser source through one of the gratings a process similar to the process in the multi Stokes system happens and multi SBS Stokes appear. In the laser configuration the Stokes waves have feedbacks on both sides and the Stokes are travelling in an amplifying media, therefore they are amplified inside the cavity which gives the potential ability for many more Stokes waves. The In this configuration, of course, light is generated only in longitudinal modes which meet the cavities longitudinal mode restriction, but in a long cavity with relatively narrow spaced modes we see all the Brillouin Stokes waves develop. In addition to the high order SBS it is also possible to have products of four-wave mixing (4WM). Every two SBS waves, can generate a new wave by 4WM. The result are waves with the same frequency spacing, but here also with a possible positive frequency shift; thus also getting new lines with higher frequencies (or lower wavelengths).
In Fig. \[Blaserout\] we show the output of the Brillouin laser described above. Comparing the spectrum of the Brillouin laser to that of the multi Stokes open system (Fig. (\[out2\]) we see the numerous number of Brillouin lines and also the lines generated by four-wave mixing (4WM), especially those above the input frequency.
![The experimental Brillouin laser output spectrum.[]{data-label="Blaserout"}](blaserout.eps){width="7.7cm"}
In this experiment we used a chirped grating for one of the ’mirrors’ and deliberately chose a grating that compensates for the dispersion of one round trip in the laser, this selection increased the amplitude of the 4WM terms compared to the same laser with a non-chirped grating, but had no effect on the terms created by SBS. We tested this view to show that SBS terms are phased matched and that the peaks we see are mostly SBS and not other non-linear phenomena.
Discussion and Summary
======================
We have demonstrated that the SBS process does not cascade by itself in open system configurations due to the power profile of the waves in optical fibers. In order to cascade the SBS process we must intervene in the system to change the basic configuration of the interacting waves. One way of achieving this is by the use of a Bragg reflector which changes the power profile to be favorable for the generation of higher stokes reflection. In this simple setup each pair of waves, signal and its Stokes, can be treated as a simple SBS reflection and all pairs are related through the boundary conditions of the setup. The boundary condition relations make it simple to design the output power of each Stokes wave by changing the input power and the Bragg reflector’s reflectivity. We have also demonstrated a closed system of a laser configuration which is much more efficient than the non-feedback setup and can generate many more SBS reflections, but is not as simple to analyze and design.
[10]{}
B.Ya. Zel’Dovich, N.F. Pilipetsky, V.V. Shkunov, *Principles of Phase Conjugation*, Springer Series in Optical Sciences, [**42** ]{}, 1985.
G. P. Agrawal, *Nonlinear Fiber Optics*, Academic Press, Second Edition, 1995.
N. V. Kukhtarev, V. B. Markov, S. G. Odulov, M. S. Soskin, V, L. Vinetskii, " Holograpghic Storage in Electrooptic Crystals 1@2. Steady State and Beam Coupling - Light Amplification; Ferroelectrics [**22** ]{}, 961, 1979
J. P. Huignard, E. Spitz, P. Aubourg, J.P Herriau, “Phase-Conjugate Wavefront Generation via Real-Time Holography in $B_12SiO_{20}$ rystals”, Opt. Lett. [**4** ]{}, 21, 1979 J. Feinberg, R. W. Hellwarth, “Phase Conjugating Mirror with Continuous-Wave Gain”, Opt. Lett. [**5** ]{}, 519, 1980
B. Fischer, M. Cronin-Golomb, J.O. White, and A. Yariv,“amplified reflection, Transmission, and Self-Oscillation in Real-Time Holography”, Opt. Lett. [**6** ]{}, 519, 1981
J.O. White, B. Fischer, M. Cronin-Golomb and A. Yariv, “Coherent Oscillation by Self-Induced Gratings in the Photorefractive Crystal BaTiO3”, Appl. Phys. Lett., [**40**]{}, 450, 1982.
B. Fischer, S. Sternklar and S. Weiss, “Photorefractive Oscillators”, IEEE, J. Quantum Electronics, [**25**]{}, 550, 1989.
M. Cronin-Golomb, B. Fischer, J.O. White and A. Yariv, “Theory and Application of Four Wave Mixing in Photorefractive Materials”, (Invited Paper), IEEE J. Quantum Electr.,[**QE-20** ]{}, 12, 1984.
S. Weiss, S. Sternklar and B. Fischer, “Double Phase Conjugate Mirrors: Analysis, Operation and Applications”, Optics Lett., [**12**]{}, 114, 1987.
37\. B. Fischer, S. Weiss and S. Sternklar, “Spatial Light Modulation and Filtering Effects in Photorefractive Wave Mixing”, Appl. Phys. Lett., [**50**]{}, 483, 1987.
B. Fischer, S. Sternklar and S. Weiss, “Photorefractive Laser Oscillation with Intracavity Image and Multimode Fibers”, Appl. Phys. Lett., [**48**]{}, 1567, 1986.
B. Fischer, “Theory of Self Frequency Detuning of Oscillators by Wave Mixing in Photorefractive Crystals”, Optics Lett., [**11**]{}, 236, 1986.
S. Sternklar, S. Weiss and B. Fischer, “Tunable Frequency Shift of Photorefractive Oscillators”, Optics Lett., [**11**]{}, 165, 1986.
S. Weiss, M. Segev and B. Fischer, “Line Narrowing and Self Frequency Scanning of Laser Diode Arrays Coupled to a Photorefractive Oscillator”, IEEE, J. Quantum Electronics, JQE, [**24**]{}, 706, 1988.
A. Yariv, “3-Dimensional Pictorial Transmission in Optical Fibers”, Appl. Phys. Lett. [**28**]{}, 88 1976.
B. Fischer and S. Sternklar, “Image Transmission and Interferometry Through Multimode Fibers using Self-Pumped Phase Conjugation”, Appl. Phys. Lett., [**46**]{}, 113, 1985.
A. Yariv, D. Fekete and D. Pepper, “Compensation for channel dispersion by non-linear optical phase conjugation”, Opt. Lett. Vol. [**4**]{}, 52, 1979.
M. F. Ferreira, “Effect of stimulated Brillouin scattering on distributed fibre amplifiers”, Electon. Lett. [**30**]{}, 40, 1994. Alexander L. Gaeta and Robert W. Boyd, “Stochastic dynamics of stimulated Brillouin scattering in an optical fiber”, Phys. Rev. A [**44**]{}, 3205, 1991. R. W. Boyd and K. Rzazewski, “Noise initiation of stimulated Brillouin scattering” , Phys. Rev. A [**42**]{}, 5514, 1990. M. Dämmig, G. Zinner, F. Mitschke and H. Welling “Stimulated Brillouin scattering in fibers with and without external feedback”, Phys. Rev. A [**48** ]{}, 3301, 1993
N. Shibata, R. G. Waarts and R. P. Braun, “Brillouin-gain spectra for single-mode fibers having pure-silica, GeO$_2$-doped, and P$_2$O$_5$-doped cores”, Optics Letters, [**12**]{}, 269, 1987. R. W. Tkach, A. R. Charplyvy and R. M. Derosier, “Spontaneous Brillouin Scattering for Single-Mode Optical-Fi.bre Characerisation”, Electronics Letters, [**22**]{}, 1011, 1986.
R. W. Boyd, *Nonlinear Optics*, Academic Press, 1992.
D. Cotter, “Transient stimulated Brillouin scattering in long single-mode fibres ”, Electronics Letters, [**18**]{}, 504, 1982; Journal of LightWave Technology, [**6**]{}, 710, 1988.
D. S. Lim. H. K. Lee, K. H. Kim, S. B. Kang ,J. T . Ahn and Min-Yong Jeon, “Generation of multiorder Stokes and anti-stokes lines in a Brillouin erbium-fiber laser with sagnac loop mirror”, Optics Letters, [**23**]{}, 1671, 1998. D. Park, J. Park, N. Park, J. Lee and J. Chang, “53-line multi-wavelength generation of Brillouin/erbium fiber laser with enhanced Stokes feedback coupling”, OFC (Optical fiber communication) conference, Technical Digest Postconference Edition, [**3**]{}, 11, 2000 A. Yariv, *Optical Electronics in Modern Communicartions*, Oxford University Press, Fifth Edition, 1997.
L. Chen and X. Bao, “Analytical numerical solutions for steady state stimulated Brillouin scattering in a single-mode fiber”, Optics Communications, [**152**]{}, 65, 1998.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Sebastian Szyller
- Buse Gul Atli
- Samuel Marchal
- 'N. Asokan'
bibliography:
- 'bibliography.bib'
title: ': Dynamic Adversarial Watermarking of Neural Networks'
---
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a continuation of our theoretical research into the influence of co-solvent polarizability on a differential capacitance of the electric double layer. We formulate a modified Poisson-Boltzmann theory, using the formalism of density functional approach on the level of local density approximation taking into account the electrostatic interactions of ions and co-solvent molecules as well as their excluded volume. We derive the modified Poisson-Boltzmann equation, considering the three-component symmetric lattice gas model as a reference system and minimizing the grand thermodynamic potential with respect to the electrostatic potential. We apply present modified Poisson-Boltzmann equation to the electric double layer theory, showing that accounting for the excluded volume of co-solvent molecules and ions slightly changes the main result of our previous simplified theory. Namely, in the case of small co-solvent polarizability with its increase under the enough small surface potentials of electrode the differential capacitance undergoes the significant growth. Oppositely, when the surface potential exceeds some threshold value (which is slightly smaller than the saturation potential), the increase in the co-solvent polarizability results in a differential capacitance decrease. However, when the co-solvent polarizability exceeds some threshold value, its increase generates a considerable enhancement of the differential capacitance in a wide range of surface potentials. We demonstrate that two qualitatively different behaviors of the differential capacitance are related to the depletion and adsorption of co-solvent molecules at the charged electrode. We show that an additive of the strongly polarizable co-solvent to an electrolyte solution can shift significantly the saturation potential in two qualitatively different manners. Namely, a small additive of strongly polarizable co-solvent results in a shift of saturation potential to higher surface potentials. On the contrary, a sufficiently large additive of co-solvent shifts the saturation potential to lower surface potentials. We obtain that an increase in the co-solvent polarizability makes the electrostatic potential profile longer-ranged. However, increase in the co-solvent concentration in the bulk leads to non-monotonic behavior of the electrostatic potential profile. An increase in the co-solvent concentration in the bulk at its sufficiently small values makes the electrostatic potential profile longer-ranged. Oppositely, when the co-solvent concentration in the bulk exceeds some threshold value, its further increase leads to decrease in electrostatic potential at all distances from the electrode.'
author:
-
-
-
title: 'On the theory of electric double layer with explicit account of a polarizable co-solvent'
---
Introduction
============
The Poisson-Boltzmann (PB) equation is the simplest and very efficient tool for describing distribution of charged particles near the macroscopic charged objects in many areas, such as biophysics, electrochemistry, chemical engineering, etc [@Israelachvili]. As is well known, the PB equation is based on the mean-field theory that makes its application to real systems quite problematic. Firstly, the mean-field theory itself does not allow us to take into account the effects of the ionic correlations that is crucial for medium and high concentrated electrolyte solutions. Secondly, considering the solvent as a continuous dielectric medium makes it impossible to study the effects of the solvent molecular structure. These two factors have motivated the researchers to improve the PB equation in the last two decades [@Podgornik_Review; @Ben-Yaakov_Review]. At present, great efforts have been made to modify the PB equation with respect to ionic correlations [@Podgornik_1989; @Netz; @Moreira_Netz; @Netz_Orland; @Forsman; @Bazant2011], the dipole structure of the solvent [@Coalson_1996; @Andelman_2007; @Andelman_2012; @Buyukdagli_2013; @Buyukdagli_2014], polarizability and permanent dipole of ions [@Frydel; @Buyukdagli_2013] as well as their excluded volume [@Andelman_1997; @Antypov_2005; @Kornyshev; @Maggs2015; @Buyukdagli; @Buyukdagli_2], the dielectric decrements of ions [@Ben-Yaakov_2011; @Andelman_2015; @Hatlo], and finally solvent quadrupolarizability [@Slavchov].
Most of these researches are devoted to the influence of the different microscopic ionic parameters on the macroscopic quantities of the electric double layer, such as local concentration of ions on the electrode, disjoining pressure, and double layer differential capacitance. The latter is one of the most important quantities for the electrochemical applications. In a recent work [@Budkov] we showed in the framework of the field-theoretical approach that if an electrolyte solution is mixed with some strongly polarizable dielectric co-solvent, then the variation of the differential capacitance becomes the greater the stronger polarizability grows. We also demonstrated that in contrast to the co-solvent polarizability the permanent dipole of the co-solvent molecules only slightly affects the differential capacitance. Moreover, due to the fact that the above mentioned theory described the ions and co-solvent molecules as point particles, the effects of the excluded volume were fully ignored. However, as was clearly showed by Kornyshev [@Kornyshev] in the framework of lattice gas model, the excluded volume of ions must strongly affects the value of differential capacitance in the region of high surface potentials.
In this work we continue our theoretical research into the co-solvent polarizability influence on the double layer differential capacitance. We obtain the expression for the grand thermodynamic potential as a functional of electrostatic potential profile within the density functional approach on the level of local density approximation, taking into account the electrostatic interactions of ions and co-solvent molecules as well as their excluded volume. We derive the modified PB equation, considering the three-component symmetric lattice gas model as a reference system and minimizing the grand thermodynamic potential with respect to the electrostatic potential. We apply this equation to the theory of electric double layer, studying the behavior of differential capacitance and local co-solvent concentration on the electrode as the functions of surface potential as well as a behavior of electrostatic potential profile with varying the polarizability and concentration of co-solvent in the bulk solution.
Theory
======
General formalism
-----------------
We consider an electrolyte solution containing $N_{+}$ ions carrying a charge $q_{+}>0$, $N_{-}$ ions carrying a charge $q_{-}<0$, and a solvent which we shall model as a continuous dielectric medium with dielectric permittivity $\varepsilon_{s}$. Moreover, we consider $N$ molecules of a co-solvent which have a polarizability $\alpha$. To describe the thermodynamic properties of such system, we shall use the variant of density functional theory at the level of local density approximation developed recently in the work [@Maggs2015].
The grand thermodynamic potential of the electrolyte solution mixed with the polarizable co-solvent can be written as $$\nonumber
\Omega=-\int \frac{\varepsilon(\bold{r})\left(\nabla{\psi}(\bold{r})\right)^2}{8\pi} d\bold{r}+\int \rho_{c}(\bold{r})\psi(\bold{r})d\bold{r}$$ $$\label{eq:Omega1}
+\int\left(f(c_{+}(\bold{r}),c_{-}(\bold{r}),n(\bold{r}))-\mu_{+}c_{+}(\bold{r})-\mu_{-}c_{-}(\bold{r})-\mu n(\bold{r})\right)d\bold{r} ,$$ where $\varepsilon(\bold{r})=\varepsilon_{s}+4\pi \alpha n(\bold{r})$ is the local dielectric permittivity, $c_{\pm}(\bold{r})$ is the local concentrations of ions, $n(\bold{r})$ is the local concentration of co-solvent, $\rho_{c}(\bold{r})=q_{+}c_{+}(\bold{r})+q_{-}c_{-}(\bold{r})$ is the charge density, $f$ is the density of free energy of the reference system (see below).
Rewriting the grand thermodynamic potential (\[eq:Omega1\]) as $$\label{eq:Omega2}
\Omega=\int \left(-\frac{\varepsilon_{s}\left(\nabla{\psi}\right)^2}{8\pi}+f(c_{+},c_{-},n)-(\mu_{+}-q_{+}\psi)c_{+}-(\mu_{-}-q_{-}\psi)c_{-}-\left(\mu+\frac{\alpha}{2}\left(\nabla{\psi}\right)^2\right) n\right)d\bold{r},$$ and using the thermodynamic relation for the pressure $$P=c_{+}\mu_{+}+c_{-}\mu_{-}+n\mu- f,$$ we eventually obtain $$\label{eq:Omega3}
\Omega[\psi]=-\int \left(\frac{\varepsilon_{s}\left(\nabla{\psi}\right)^2}{8\pi}+ P(\mu_{+}-q_{+}\psi,\mu_{-}-q_{-}\psi,\mu+\frac{\alpha}{2}\left(\nabla{\psi}\right)^2)\right) d\bold{r}.$$ Thus, if the explicit function $P=P(\mu_{+},\mu_{-},\mu)$ is known, one can obtain the explicit equation for the electrostatic potential $\psi(\bold{r})$ by minimizing the functional (\[eq:Omega3\]). To take into account the excluded volume of co-solvent and ions, we consider the lattice gas model (without the attractive Van-der-Waals interactions between the particles) as a reference system for which the explicit dependence $P=P(\mu_{+},\mu_{-},\mu)$ is well known: $$\label{eq:P}
P=\frac{k_{B}T}{v}\ln\left(1+e^{\beta\mu_{+}}+e^{\beta\mu_{-}}+e^{\beta\mu}\right),$$ where $v$ is the volume occupied by a particle of lattice gas, $T$ is the temperature, $k_{B}$ is the Boltzmann constant, $\beta=1/k_{B}T$.
Therefore, we obtain the following functional: $$\label{eq:Omega4}
\Omega[\psi]=-\int \left(\frac{\varepsilon_{s}\left(\nabla{\psi}\right)^2}{8\pi}+\frac{k_{B}T}{v}\ln\left(1+e^{\beta\left(\mu_{+}-q_{+}\psi\right)}+e^{\beta\left(\mu_{-}-q_{-}\psi\right)}+e^{\beta\left(\mu+\frac{\alpha}{2}\left(\nabla{\psi}\right)^2\right)}\right)\right)d\bold{r}.$$ Further, minimizing the functional (\[eq:Omega4\]) and using the expressions for the chemical potentials of species $$\label{eq:mu}
\mu_{\pm}=k_{B}T\ln{\frac{c_{\pm,b}v}{1-v(c_{+,b}+c_{-,b}+n_{b})}}, ~~\mu=k_{B}T\ln{\frac{n_{b}v}{1-v(c_{+,b}+c_{-,b}+n_{b})}},$$ we arrive at the modified Poisson-Boltzmann equation with accounting for the polarizability of co-solvent molecules, their excluded volume, and the excluded volume of electrolyte ions $$\label{eq:PBeq1}
\nabla(\varepsilon(\bold{r})\nabla{\psi}(\bold{r}))=-\frac{4\pi \left(q_{+}c_{+,b}e^{-\beta q_{+}\psi(\bold{r})}+q_{-}c_{-,b}e^{-\beta q_{-}\psi(\bold{r})}\right)}{1+v\left(c_{+,b}\left(e^{-\beta q_{+}\psi(\bold{r})}-1\right)+c_{-,b}\left(e^{-\beta q_{-}\psi(\bold{r})}-1\right)+n_{b}\left(e^{\frac{\beta\alpha}{2}\left(\nabla{\psi}(\bold{r})\right)^2}-1\right)\right)},$$ where $c_{\pm,b}$ is the bulk concentrations of ions, $n_{b}$ is the bulk co-solvent concentration; $$\label{eq:perm1}
\varepsilon(\bold{r})=\varepsilon_{s}+\frac{4\pi\alpha n_{b} e^{\frac{\beta\alpha}{2}\left(\nabla{\psi}(\bold{r})\right)^2}}{1+v\left(c_{+,b}\left(e^{-\beta q_{+}\psi(\bold{r})}-1\right)+c_{-,b}\left(e^{-\beta q_{-}\psi(\bold{r})}-1\right)+n_{b}\left(e^{\frac{\beta\alpha}{2}\left(\nabla{\psi}(\bold{r})\right)^2}-1\right)\right)}$$ is the local dielectric permittivity of the electrolyte solution. When there are no co-solvent molecules in the electrolyte solution ($n_{b}=0$), we arrive at the equation obtained firstly by Borukhov et al [@Andelman_1997] and Kornyshev [@Kornyshev] $$\varepsilon_{s}\nabla^2\psi(\bold{r})=-\frac{4\pi\left(q_{+}c_{+,b}e^{-\beta q_{+}\psi(\bold{r})}+q_{-}c_{-,b}e^{-\beta q_{-}\psi(\bold{r})}\right)}{1+v\left(c_{+,b}\left(e^{-\beta q_{+}\psi(\bold{r})}-1\right)+c_{-,b}\left(e^{-\beta q_{-}\psi(\bold{r})}-1\right)\right)}.$$
In limit of the point particles (when $v\rightarrow 0$) equation (\[eq:PBeq1\]) looks as follows $$\label{eq:PBeq2}
\nabla(\varepsilon(\bold{r})\nabla{\psi}(\bold{r}))=-4\pi \left(q_{+}c_{+,b}e^{-\beta q_{+}\psi(\bold{r})}+q_{-}c_{-,b}e^{-\beta q_{-}\psi(\bold{r})}\right),$$ where $\varepsilon(\bold{r})=\varepsilon_{s}+4\pi\alpha n_{b}e^{\frac{\beta\alpha}{2}\left(\nabla{\psi}(\bold{r})\right)^2}$ is the local dielectric permittivity in the approximation of point particles. It should be noted that equation (\[eq:PBeq2\]) was obtained in the recent work [@Budkov] within the field-theoretical approach.
Theory of electric double layer
-------------------------------
As an application of the modified PB equation (\[eq:PBeq1\]-\[eq:perm1\]), we formulate the generalized Kornyshev’s theory [@Kornyshev; @Barrat_Hansen]. We consider a system containing a charged electrode, which we shall model as a charged flat surface with a surface charge density $\sigma$, the ions of 1:1 electrolyte (i.e. when $q_{+}=-q_{-}=e$; $e$ is the elementary charge), and the molecules of the polarizable co-solvent with a polarizability $\alpha$. In this case the average concentrations of ions in the bulk are equal, i.e., $c_{+,b}=c_{-,b}=c$. Choosing $z$ axis perpendicular to the electrode and placing the origin on it, one can write the grand thermodynamic potential per unit area of the electrode as follows: $$\label{eq:Omega5}
\Omega[\psi]=-\int\limits_{0}^{\infty}dz \left(\frac{\varepsilon_{s}\left(\psi^{\prime}(z)\right)^2}{8\pi}+\frac{k_{B}T}{v}\ln\left(1+e^{\beta\left(\mu_{+}-e\psi(z)\right)}+e^{\beta\left(\mu_{-}+e\psi(z)\right)}+e^{\beta\left(\mu+\frac{\alpha}{2}\left(\psi^{\prime}(z)\right)^2\right)}\right)\right),$$
Since the integrand in (\[eq:Omega5\]) does not depend on coordinate $z$ explicitly, the Euler-Lagrange equation has a first integral which determines the condition of the solution mechanical equilibrium $$\label{eq:press}
P\left(\mu_{+}-e\psi,\mu_{-}+e\psi,\mu+\frac{\alpha\mathcal{E}^2}{2}\right)-\frac{\varepsilon_{s}\mathcal{E}^2}{8\pi}-\alpha\mathcal{E}^2 n\left(\mu_{+}-e\psi,\mu_{-}+e\psi,\mu+\frac{\alpha\mathcal{E}^2}{2}\right)=P\left(\mu_{+},\mu_{-},\mu\right),$$ where the local electric field $\mathcal{E}(z)=-\psi^{\prime}(z)$ and the local co-solvent concentration $n=\partial{P}/\partial{\mu}$ are introduced. The first term in the left-hand side of eq. (\[eq:press\]) determines the pressure which is related to the excluded volume of particles, whereas the second and third terms determine the so-called disjoining pressure contribution which is due to the electrostatic interactions [@Barrat_Hansen].
Further, substituting the expressions for the bulk chemical potentials of species (\[eq:mu\]) and for the pressure (\[eq:P\]) into the equation (\[eq:press\]), we eventually obtain $$\label{eq:E1}
1+\left(2c\left(\cosh(\beta e\psi(z))-1\right)+n_{b}\left(e^{\frac{\beta\alpha \mathcal{E}^2(z)}{2}}-1\right)\right)v=e^{\frac{v}{k_{B}T}\left(\frac{\varepsilon_{s}\mathcal{E}^2(z)}{8\pi}+\frac{\alpha n_{b}\mathcal{E}^2(z)e^{\frac{\beta\alpha\mathcal{E}^2(z)}{2}}}{1+\left(2c\left(\cosh{\beta e\psi(z)}-1\right)+n_{b}\left(e^{\frac{\beta\alpha\mathcal{E}^2(z)}{2}}-1\right)\right)v}\right)}.$$
In the limit $v\rightarrow 0$ we obtain the following equation $$\label{eq:E2}
\frac{\varepsilon_{s}\mathcal{E}^2(z)}{8\pi}+n_{b} k_{B}T\left(1-e^{\frac{\beta\alpha \mathcal{E}^2(z)}{2}}\right)+n_{b}\alpha \mathcal{E}^2(z)e^{\frac{\beta\alpha \mathcal{E}^2(z)}{2}}=2c k_{B}T\left(\cosh{\beta e\psi(z)}-1\right)$$ which was first obtained in the work [@Budkov].
To obtain the potential profile $\psi(z)$, we should first solve the eq. (\[eq:E1\]) as a transcendental equation numerically (for instance, by Newton’s method) with respect to $\mathcal{E}=-\psi^{\prime}(z)$ at different values of $\psi$. Thus, we obtain the function $\mathcal{E}=\mathcal{E}(\psi)$. In order to obtain the potential profile $\psi(z)$, we solve numerically the equation $\psi^{\prime}=-\mathcal{E}(\psi)$ with use of the standard boundary condition $$\begin{aligned}
\label{eq:boundary_cond1}
-\varepsilon(0)\psi^{\prime}(0)=4\pi\sigma,\end{aligned}$$ where the local dielectric permittivity of the electrolyte solution $$\varepsilon(z)=\varepsilon_{s}+4\pi\alpha n(z)$$ is introduced. The local co-solvent concentration can be expressed as follows $$n(z)=\frac{n_{b}e^{\frac{\beta\alpha \mathcal{E}^2(z)}{2}}}{1+v\left(2c\left(\cosh{\beta e\psi(z)}-1\right)+n_{b}\left(e^{\frac{\beta\alpha \mathcal{E}^2(z)}{2}}-1\right)\right)}.$$
To calculate the differential capacitance $C=\partial{\sigma}/\partial{\psi_{0}}$ as a function of the surface electrostatic potential $\psi_{0}=\psi(0)$ which is usually an experimentally controllable parameter, we should calculate the surface charge density $\sigma$. For this purpose we use the first integral (\[eq:E1\]) written for $z=0$ $$\label{eq:E3}
1+\left(2c\left(\cosh(\beta e\psi_{0})-1\right)+n_{b}\left(e^{\frac{\beta\alpha \mathcal{E}_{0}^2}{2}}-1\right)\right)v=e^{\frac{v}{k_{B}T}\left(\frac{\varepsilon_{s}\mathcal{E}_{0}^2}{8\pi}+\frac{\alpha n_{b}\mathcal{E}_{0}^2e^{\frac{\beta\alpha\mathcal{E}_{0}^2}{2}}}{1+\left(2c\left(\cosh{\beta e\psi_{0}}-1\right)+n_{b}\left(e^{\frac{\beta\alpha\mathcal{E}_{0}^2}{2}}-1\right)\right)v}\right)},$$ where $\mathcal{E}_{0}=\mathcal{E}(0)$, and the boundary condition (\[eq:boundary\_cond1\]) may be rewritten in the form $$\label{eq:boundary_cond2}
\sigma=\frac{1}{4\pi}\left(\varepsilon_{s}+\frac{4\pi\alpha n_{b} e^{\frac{\beta\alpha\mathcal{E}_{0}^2}{2}}}{1+v\left(2c\left(\cosh{\beta e\psi_{0}}-1\right)+n_{b}\left(e^{\frac{\beta\alpha \mathcal{E}_{0}^2}{2}}-1\right)\right)}\right)\mathcal{E}_{0}.$$
Solving the system of coupled nonlinear equations (\[eq:E3\]-\[eq:boundary\_cond2\]) numerically with respect to $\mathcal{E}_{0}$ and $\sigma$ at different values of $\psi_{0}$, we obtain the dependence $\sigma=\sigma(\psi_{0})$ that allows us to obtain the differential capacitance profile $C=C(\psi_{0})$ (see the next section).
Numerical results and discussion
================================
Turning to the numerical calculations, we determine the following reduced parameters: $\tilde{n}_{b}=n_{b}v$, $\tilde{\alpha}=\alpha/v\varepsilon_{s}$, $\tilde{\mathcal{E}}=\beta ev^{1/3}\mathcal{E}$, $u=\beta e\psi$, $\tilde{z}=z/v^{1/3}$, and $\tilde{\sigma}=\sigma\beta ev^{1/3}/\varepsilon_{s}$. We first discuss the behavior of the differential capacitance as the function of surface potential. The reduced differential capacitance $\tilde{C}=C/v^{1/3}\varepsilon_{s}$ can be calculated as $$\tilde{C}=\frac{\partial{\tilde{\sigma}}}{\partial{u}_{0}},$$ where $u_{0}=u(0)$. The system of coupled equations (\[eq:E3\]-\[eq:boundary\_cond2\]) can be rewritten in the dimensionless form as $$\label{eq:E4}
1+2\tilde{c}\left(\cosh{u_{0}}-1\right)+\tilde{n}_{b}\left(e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}_{0}^2}{2\xi}}-1\right)=e^{\frac{1}{\xi}\left(\frac{\tilde{\mathcal{E}}_{0}^2}{8\pi}+\frac{\tilde{\alpha} \tilde{n}_{b}\tilde{\mathcal{E}}_{0}^2e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}_{0}^2}{2\xi}}}{1+2\tilde{c}\left(\cosh{u_{0}}-1\right)+\tilde{n}_{b}\left(e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}_{0}^2}{2\xi}}-1\right)}\right)},$$ and $$\label{eq:boundary_cond3}
\tilde{\sigma}=\frac{1}{4\pi}\left(1+\frac{4\pi\tilde{\alpha} \tilde{n}_{b} e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}_{0}^2}{2\xi}}}{1+2\tilde{c}\left(\cosh{u_{0}}-1\right)+\tilde{n}_{b}\left(e^{\frac{\tilde{\alpha} \tilde{\mathcal{E}}_{0}^2}{2\xi}}-1\right)}\right)\tilde{\mathcal{E}}_{0},$$ where $\xi=l_{B}/v^{1/3}$, $l_B=e^2/{\varepsilon_{s}k_{B}T}$ is the Bjerrum length.
The first integral (\[eq:E1\]) of the modified Poisson-Boltzmann equation can be also rewritten in the dimensionless form as follows $$\label{eq:E5}
1+2\tilde{c}\left(\cosh{u(\tilde{z})}-1\right)+\tilde{n}_{b}\left(e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}^2(\tilde{z})}{2\xi}}-1\right)=e^{\frac{1}{\xi}\left(\frac{\tilde{\mathcal{E}}^2(\tilde{z})}{8\pi}+\frac{\tilde{\alpha} \tilde{n}_{b}\tilde{\mathcal{E}}^2(\tilde{z})e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}^2(\tilde{z})}{2\xi}}}{1+2\tilde{c}\left(\cosh{u(\tilde{z})}-1\right)+\tilde{n}_{b}\left(e^{\frac{\tilde{\alpha}\tilde{\mathcal{E}}^2(\tilde{z})}{2\xi}}-1\right)}\right)}.$$
We use the following values of the physical parameters $\varepsilon_{s}=80$, $T=300~K$, $c=0.1~mol/L$ $v^{1/3}=0.3~nm$ which yield a set of the reduced parameters: $\tilde{c}=1.63\times 10^{-3}$, $\xi=2.32$. Fig. 1a demonstrates the differential capacitance profiles $\tilde{C}=\tilde{C}(u_{0})$ for the small co-solvent polarizabilities and the fixed bulk co-solvent concentration $\tilde{n}_{b}=0.5$. As is seen, increasing the co-solvent polarizability may generate a differential capacitance enhancement in the region of surface potentials less than the ’saturation’ potential $u_{sat}$ (a surface potential at which the maximum of the differential capacitance is achieved). However, if the surface potential is in the region of electric double layer saturation, increase in the co-solvent polarizability provokes a decrease in the differential capacitance (see Fig. 1a). Oppositely, when the co-solvent polarizability exceeds some critical value, its increase leads to a different behavior of the differential capacitance. Namely, increasing the co-solvent polarizability in this case generates a significant growth of the differential capacitance in the wide range of surface potentials (see Fig. 1b). In order to understand these two qualitatively different regimes, let us consider the behavior of co-solvent concentration on the electrode $\tilde{n}_{s}=\tilde{n}(0)$ as the function of surface potential $u_{0}$ at different co-solvent polarizabilities $\tilde{\alpha}$. Fig. 2 demonstrates the values of $\tilde{n}_{s}$ as the functions of surface potential at different co-solvent polarizabilities. As one can see, at sufficiently small co-solvent polarizability the cosolvent molecules are depleted at the electrode. On the contrary, when the co-solvent polarizability exceeds some threshold value, the co-solvent molecules create an adsorption layer on the charged electrode. These two regimes are clearly demonstrated by fig. 3, where the co-solvent concentration profiles $\tilde{n}(\tilde{z})$ are depicted. Thus, two different regimes of the differential capacitance behavior are related to the depletion and adsorption of co-solvent molecules at the charged electrode.
Figures 4a,b show the differential capacitance profiles at different values of the dimensionless co-solvent concentration $\tilde{n}_{b}$ at the fixed co-solvent polarizability $\tilde{\alpha}=0.3$. As one can see, an increase in the co-solvent concentration in the bulk solution can shift significantly the maximum of differential capacitance by two qualitatively different manners. Namely, at the sufficiently small co-solvent concentration its increase leads to a shift of the differential capacitance maximum to the region of higher surface potentials (see fig. 4a). It means that an additive of the small quantity of the polarizable co-solvent to the electrolyte solution prevents the saturation of the electric double layer. In the case, when the co-solvent concentration exceeds the threshold value, the maximum of differential capacitance shifts to the region of lower surface potentials (see fig.4b). Figure 5 shows the dependencies of the saturation potential $u_{sat}$ on the co-solvent concentration $\tilde{n}_{b}$ at different values of co-solvent polarizability $\tilde{\alpha}$. As it is shown, the non-monotonic behavior of the saturation potential with varying co-solvent concentration occurs at sufficiently large co-solvent polarizability only. However, an additive of the co-solvent with sufficiently small polarizability leads to the shift of the saturation potential to lower potentials for all the considered co-solvent concentrations. It should be noted that non-monotonic behavior of the saturation potential with increasing co-solvent concentration can be of interest to electrochemical applications, where it is necessary to control the differential capacitance.
Fig. 6 demonstrates the comparison between the differential capacitance profiles obtained by the present theory and our previous theory. As one can see, previous theory is valid at small surface potentials only. Indeed, accounting for the excluded volume of both ions and molecules of the co-solvent results in a decrease in the differential capacitance in the region of high surface potential compared to the simplified theory of point particles. The latter means that the dramatic increase in the differential capacitance at high surface potentials predicted in work [@Budkov] is unphysical.
Finally, we discuss the influence of the co-solvent concentration and co-solvent polarizability on the electrostatic potential profile $u(\tilde{z})$. As well as in our previous theory, an increase in the co-solvent polarizability leads to longer-ranged electrostatic potential profiles (Fig. 7). The latter is due to the fact that an increase in this variable results in higher local dielectric permittivity that, in turn, leads to a decrease in the electrode charge screening. However, an increase in the bulk co-solvent concentration leads to more complex behavior of the electrostatic potential profile. Namely, increasing the bulk co-solvent concentration at its sufficiently small values makes the electrostatic potential profile longer-ranged. Nevertheless, when the co-solvent concentration in the bulk exceeds some threshold value, its further increase leads to a decrease in the electrostatic potential at all distances from the electrode (Fig. 8). Such behavior of the electrostatic potential depending on the bulk co-solvent concentration is different on that predicted by our previous simplified theory of point particles, where the potential profile becomes longer-ranged at all co-solvent concentartions.
Conclusion
==========
In this work based on the density functional formalism on the level of local density approximation, we have developed a modified Poisson-Boltzmann equation with an explicit account of the polarizable co-solvent in combination with the excluded volume of ions and co-solvent molecules. We have applied the modified Poisson-Boltzmann equation to electric double layer theory and shown that like in our previous simplified theory [@Budkov] (where all particles of the electrolyte solution were considered as point ones), the present theory predicts the influence of the co-solvent polarizability on the differential capacitance. Namely, in the case of small co-solvent polarizabilities under sufficiently small surface potentials of electrode the differential capacitance grows significantly with increasing of the co-solvent polarizability as well as bulk co-solvent concentration. Oppositely, when the surface potential exceeds some threshold value (which is close to the saturation potential), the growth of the co-solvent polarizability and bulk co-solvent concentration results in decrease in the differential capacitance. However, when the co-solvent polarizability exceeds some threshold value, its increase generates a considerable growth of the differential capacitance in the region of the double layer saturation. We have established that two qualitatively different regimes of the differential capacitance behavior are caused by the depletion and adsorption of co-solvent molecules at the charged electrode. We have also shown that an additive of the sufficiently strong polarizable co-solvent to an electrolyte solution can significantly shift the maximum of differential capacitance by two qualitatively different ways. Namely, a small additive of co-solvent results in the shift of differential capacitance maximum to the higher surface potentials. However, when the bulk co-solvent concentration exceeds the threshold value, the maximum of differential capacitance shifts to the lower surface potentials. We have shown that increase in the co-solvent polarizability results in longer-ranged electrostatic potential profile. Finally, we have obtained that at sufficiently small co-solvent concentration in the bulk its increase makes the electrostatic potential profile longer-ranged. Nevertheless, when the co-solvent concentration in the bulk exceeds some threshold value, its further increase leads to a decrease in electrostatic potential.
Now we would like to discuss the limitations of the present theory. It is well known, the lattice gas model highly underestimates the pressure in the bulk at high number densities of particles for the off-lattice hard spheres system [@Sanchez1976]. Moreover, the lattice gas model highly overestimates the differential capacitance obtained by MD computer simulations in the wide range of surface potential [@Fedorov2008]. That is why the lattice gas model cannot be used for quantitative predictions of both thermodynamic and electrochemical variables, but only for their qualitative evaluations. To get more reliable quantitative results, one can use more precise Percus-Yevick or Carnahan-Starling equations of state. However, the application of these equations of state will involve more difficult numerical calculations [@Maggs2015]. The next limitation is related to the fact that the present theory is based on the local density approximation and fully ignores the nonlocal packing effects which have a short-range nature and must be important for the ions and co-solvent molecules near the electrode [@Howard2010; @Bazant2011]. However, we believe that such short-ranged effects could not drastically affect the double layer differential capacitance which should be determined mostly by the long-range correlations of particles. On the other hand, the effects of co-solvent polarizability related to the long-range correlations of particles [@Schroder_2015; @Cavalcante_2014] should be qualitatively described on the level of mean-field approximation. Unfortunately, we cannot give an [*a priori*]{} estimate within this formalism of the results obtained. The latter requires calculations based on the nonlocal density functional theory or computer simulations. In the present theory, we have considered the solvent as continuous dielectric medium with fixed dielectric permittivity. In other words, we have assumed that the solvent dielectric permittivity near the charged electrode is the same as that in the bulk solution. However, as is well known, such assumption cannot be correct for the sufficiently large surface charge density of the electrode. Indeed, the application of sufficiently large electric field can lead to significant decrease of the water dielectric permittivity [@Booth1951; @Sutman1998; @Yeh1999; @Gongadze2015]. That is why our theory gives highly overestimated polarizabilities of the co-solvent molecules $\alpha\simeq 200~\textup{\AA}^3$ ($\tilde{\alpha}\simeq 0.1$) for which the discussed phenomena might be realized. We believe that accounting for the effect of dielectric permittivity renormalization near the charged electrode might reduce the polarizability to the physically reasonable values ($\alpha\simeq 10~\textup{\AA}^3$). Nevertheless, we hope that our self-consistent field theory may be of use for qualitative evaluations in various electrochemical applications. Finally, it is worth noting that the present theory makes sense only in the case when the co-solvent polarizability significantly greater than the polarizability of solvent. Indeed, only in such case the consideration of the solvent as a continuous dielectric medium at a sufficient distance from the electrode may be justified. Evidently, this condition can be satisfied for the aromatic compounds dissolved in some aqueous electrolyte solution.
In conclusion, we would like to speculate on the possible application of our theory to the experimental systems. In our opinion, it can be applied to the theoretical description of the aromatic compounds solubilization in aqueous micellar solutions of amphiphilic imidazolium ionic liquids [@Luczak2013].
We thank N. Georgi, A.I. Victorov and E.A. Safonova for fruitful discussions. We thank Reviewers for valuable comments that helped us to improve this work. This research was supported by grant from the President of the Russian Federation (No MK-2823.2015.3).
[99]{} (Academic Press, 2011).
J. Chem. Phys. $\bold{139}$, 150901 (2013).
J.Phys.: Condens. Matter. $\bold{21}$, 424106 (2009).
Eur. Phys. J. E $\bold{5}$, 557 (2001).
EPL $\bold{52}$, 705 (2000).
J. Chem. Phys. $\bold{91}$, 5840 (1989).
The European Physical Journal E $\bold{2-3}$, 203 (2000).
J. Phys. Chem. B $\bold{108}$, 9236 (2004).
PRL $\bold{106}$, 046102 (2011).
J. Phys. Chem. B $\bold{100}$, 2612 (1996).
PRL $\bold{99}$, 077801 (2007).
PRL $\bold{108}$, 227801 (2012).
J. Chem. Phys. $\bold{140}$, 234903 (2014).
J. Chem. Phys. $\bold{134}$, 234704 (2011).
Phys. Rev. E $\bold{87}$, 063201 (2013).
PRL $\bold{79}$, 435 (1997).
PRE $\bold{71}$, 061106, (2005).
J. Phys. Chem. B $\bold{111}$, 5545 (2007).
Soft Matter $\bold{12}$, 1219 (2015).
EPL $\bold{98}$, 60003 (2012).
J. Stat. Mech. $\bold{P05033}$, 1 (2011).
J. Chem. Phys. $\bold{134}$, 074705 (2011) .
J. Chem. Phys. $\bold{142}$, 044706 (2015).
EPL $\bold{97}$, 28010 (2012).
J. Chem. Phys. $\bold{140}$, 164510 (2014).
EPL $\bold{111}$, 28002 (2015).
(University Press, Cambridge, 2003).
The Journal of Physical Chemistry $\bold{80}$ (2), 1 (1976).
Electrochimica Acta $\bold{53}$, 6835 (2008).
J. Phys. Chem. B $\bold{114}$, 6074 (2010).
J. Chem. Phys. $\bold{142}$, 064503 (2015).
J. Chem. Phys. $\bold{140}$, 144108 (2014).
J. Chem. Phys. $\bold{142}$, 391 (1951).
J. Electroanal. Chem. $\bold{450}$, 289 (1998).
J. Chem. Phys. $\bold{110}$, 7935 (1999).
Electrochimica Acta $\bold{178}$, 541 (2015).
J. Phys. Chem. B, $\bold{117}$, 5653 (2013).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'High-fidelity speech can be synthesized by end-to-end text-to-speech models in recent years. However, accessing and controlling speech attributes such as speaker identity, prosody, and emotion in a text-to-speech system remains a challenge. This paper presents a system involving feedback constraints for multispeaker speech synthesis. We manage to enhance the knowledge transfer from the speaker verification to the speech synthesis by engaging the speaker verification network. The constraint is taken by an added loss related to the speaker identity, which is centralized to improve the speaker similarity between the synthesized speech and its natural reference audio. The model is trained and evaluated on publicly available datasets. Experimental results, including visualization on speaker embedding space, show significant improvement in terms of speaker identity cloning in the spectrogram level. In addition, synthesized samples are available online for listening. [^1]'
address: |
$^{\star}$Electrial & Computer Engineering, Duke University, Durham, NC, United States\
$^{\dag}$ Data Science Research Center, Duke Kunshan University, Kunshan, China
bibliography:
- 'mybib.bib'
title: |
From Speaker Verification to Multispeaker Speech Synthesis,\
Deep Transfer with Feedback Constraint
---
**Index Terms**: Text-to-speech, multi-speaker speech synthesis, speaker embedding, end-to-end
Introduction
============
Speech synthesis, also known as text-to-speech (TTS), specifies the technique that achieves the transformation from text to audio waveform. It has been widely used in our daily life, e.g., navigation systems, audiobooks, and virtual assistants. The performance of the TTS system has been further improved recently by adopting the end-to-end neural network framework [@mehri-srnn-2017-iclr; @shen2018natural; @ping2018clarinet; @sotelo2017char2wav]. The end-to-end principle is applied in the TTS model by a cohesive and autoregressive chain of neural network structures that are connected by well-defined input-output features. For instance, the state-of-the-art system Tacotron2 [@shen2018natural] consists of an encoder-decoder architecture and a neural vocoder Wavenet[@oord2016wavenet].
Extensions on these models have been developed for allowing the TTS system to control the speech characteristics. These extensions are able to enrich the expressiveness of the synthesized voice and further enhance the robustness of TTS systems. For example, Yuxuan Wang et al. proposed the style tokens to uncover the latent space regarding speech attributes that are hard to define and label [@wang2018style; @wang2017uncovering]. The models are jointly trained with the Tacotron-based TTS architecture in an unsupervised manner. On the other hand, controlling speech attributes that have easily found labels (e.g., language, emotion, and speaker identity) have also been investigated [@zhang2019learning; @yu2019durian; @jia2018transfer]. Typically, the speech attribute is controlled with a TTS model by conditioning the synthesizer with the vector representation called embedding.
The multispeaker TTS system is one of the extensions, which is developed to clone and manage distinct voices either seen or not seen during training. Most systems use the speaker embedding to characterize the expected voice and speaking style in the multispeaker TTS system [@jia2018transfer; @Ji2019Multi; @alvarez2019problem], while speaker adaptation can also be used for speaker transfer TTS modeling [@fan2015multi]. Voice cloning by speaker adaptation acquires more data and computational resource for the target voice and usually is less robust compared with cloning by speaker embedding [@arik2018neural]. The speaker verification system plays an essential role in the multispeaker TTS system for cloning unseen voices. Eliya Nachmani et al. has proposed an approach where the speaker verification system and the synthesizer are jointly trained [@nachmani2018fitting]. However, the knowledge for discriminative speaker representations is limited by the training dataset in this case. Then Ye Jia et al. further investigated the knowledge transfer in terms of speaker characteristics by decoupling these two tasks, where the speaker verification network is trained with a dataset that contains a larger amount of speakers but is not suitable for TTS training [@jia2018transfer]. The discriminative speaker embedding extracted from the speaker verification network is used for conditioning the TTS synthesizer and leads to better performance on open-set voice cloning.
Although the model proposed in [@jia2018transfer] increases the robustness of the synthesizer for open-set multispeaker synthesis, the speaker’s similarity is not close between the synthesized speech and the speaker’s natural speech. Concerning the same speaker, the embeddings extracted from synthesized speech and those extracted from natural speech may have two distinct distributions. To further transfer the knowledge from a speaker verification model to the speech synthesizer, we propose a multispeaker TTS model with the feedback constraint toward the speaker embedding space. Specifically, an added score associated with the speaker similarity is performed by the verification network for forcing the synthesizer to derive the knowledge for speaker identity cloning. The proposed method is evaluated on publicly available datasets. As demonstrated in the visualization of the embedding space, speaker embeddings extracted from our synthesized speech lies in the same cluster as those from natural speech. Therefore, the model may be useful for data augmentation and the white-box spoofing attack toward speaker verification in the future.
This paper is organized as follows: section \[related\_work\] describes the related works in terms of speaker verification and speech synthesis. Our proposed system is presented in section \[system\_describtion\]. Experimental setup and results are shown in section \[result\]. Finally, we conclude the paper in section \[conclusion\].
Related works {#related_work}
=============
Speaker verification
--------------------
Open-set multispeaker TTS highly depends on speaker representations for conditioning the synthesizer to copy the desired voice. To that end, speaker verification systems, especially text-independent systems, are often used for feature extraction regarding their discriminative speaker representations.
In the speaker verification field, deep speaker feature learning systems proposed in recent years have achieved comparable performance or even surpassed the classical i-vector systems [@cai2018analysis; @snyder2018x; @Cai_2018_Odyssey]. The overall training architecture of the deep speaker verification system is shown in figure \[fig:spk\_framework\]. Specifically, the speaker verification model takes variable-length audio signal $x = [x_1, x_2, x_3 ... x_n]$ as input and convert the signal into frequency-domain acoustic features, e.g., filter-bank energy or Mel frequency cepstral coefficients (MFCCs). Acoustic features are then fed into a neural network-based extractor to obtain the fixed dimensional speaker representation $z \in \mathcal{R}^d $, where $d$ is the dimension of the speaker embedding. Note that the back-end discriminator here in the training phase is different from the one in the evaluation phase. The discriminator in the training phase is to classify embeddings according to their target speaker labels in order to train a discriminative speaker embedding space, while the one used in the evaluation phase is to verify if two embeddings come from the same speaker.
Among deep speaker embedding systems that are developed with various DNN architectures [@snyder2018x; @snyder2016deep; @cai2020fly], we follow the ResNet-based verification system [@cai2020fly] in our work for embedding extraction to extract time-invariant speaker embedding.
Multispeaker speech synthesis
-----------------------------
Cloning and controlling speech attributes have been studied for decades in the text-to-speech (TTS) field. For voice synthesis, Yamagishi et al. proposed feature-space adaptive training for speaker-adaptive TTS [@yamagishi2009robust]. The system aims to reduce the size of data and the cost for building different voices when developing statistical parametric speech systems based on Hidden Markov models (HMMs). After Tacotron2 demonstrated its ability to synthesize high-quality speech that can be as natural as human speech, extensions of Tacotron2 were proposed for speech attribute cloning by conditioning the linguistic encoder output with attribute embeddings. For instance, Yuxuan Wang et al. introduced global style tokens (GSTs) as the attribute embeddings to achieve style-control TTS synthesis [@wang2018style]. The proposed model in [@wang2018style] , where GSTs are trained in an unsupervised manner, also helps improve the speech intelligibility when it is used for multispeaker TTS training.
On the other hand, in order to achieve zero-shot voice cloning, the speaker representation is commonly extracted by a separated model and used as the conditioned feature in multispeaker TTS models [@jia2018transfer; @nachmani2018fitting]. In this case, the multispeaker TTS model developed for zero-shot voice cloning consists of two models, one for speaker embedding extraction and the other for TTS conversion. When the two models are trained jointly, the TTS system yields moderate performance in synthesizing voices that are unseen in the training data [@nachmani2018fitting]. The reason might be because the datasets collected for speech synthesis have limited speakers, and the datasets collected for speaker analysis have no transcriptions for TTS training. Jia Ye et al. chose to train the two models individually, where the TTS model learns the speaker representation knowledge by the embedding extracted from the speaker verification model [@jia2018transfer]. This method improves the robustness with the ability to clone unseen voices. However, two distinct clusters, representing synthesized speech and natural speech from the same speaker, are observed in the embedding space as shown in [@jia2018transfer]. To further investigate this problem and enhance the knowledge transfer, we propose a model with a feedback constraint that engages the speaker embedding extractor. We show that by showing that embeddings from different speakers result in distinct distributions in the vector space, while embeddings from the same speaker, whether synthesized or natural, lie in the same cluster.
In our work, we use a speaker embedding extractor that is different from the model described in [@jia2018transfer]. By the time we finalized our work, Erica et al. published a study investigating how different speaker embedding networks affects the multispeaker synthesis system [@cooper2019zero]. In that study, the author claim that LDE-based embedding could improve speaker similarity and naturalness. Our model has a similar speech encoder as the learnable dictionary encoding-based (LDE-based) systems described in [@cooper2019zero].
Methods {#system_describtion}
=======
Our proposed multispeaker TTS framework is shown in figure \[fig:tts\]. We follow the baseline end-to-end speaker verification system presented in [@cai2018analysis] as our embedding extraction network. The Mel-spectrogram is used as the acoustic feature for both the speaker embedding extraction system and the multispeaker TTS system. As for the speaker embedding network, ResNet34 architecture is used as the encoder network, followed by a pooling layer that calculates the mean and the standard deviation of encoder outputs. Then the speaker embedding is obtained by concatenating the mean and the standard deviation. While in the training phase, a back-end classification network consisting of two fully connected layers maps embeddings to target speakers.
We use the tacotron-based model as the Mel-spectrogram prediction model. The input character sequence is converted into a vector sequence by a trainable lookup table. Then the encoder, which consists of 5 convolutional layers and a bi-directional long short-term memory (BLSTM) layer, consumes the embedding sequence and delivers the memory that represents the context and linguistic characteristics of the input text. Speaker embedding, extracted from the target audio signal, is then concatenated with the encoder output memory globally as the final encoding states.
The decoder takes steps to predict Mel-spectrograms with three modules, which are the attention mechanism, the RNN decoder and the PostNet. The attention mechanism provides the context vector for the decoder RNN to generate spectrograms that associate with specific encoder states for each decoding time step. In addition, it provides soft alignment between the input encoder states and the target Mel-spectrogram. For each decoding step, the decoder RNN predicts the spectrogram with respect to the context vector and the predicted result from the previous time step, where the previous predicted frame is taken by the PreNet module. Two linear projection layers are followed by the decoder RNN for predicting Mel-spectrograms and stop tokens, respectively. Stop tokens are the binary sequence that specifies the valid decoding frames, where $0$ denotes a valid frame, and $1$ indicates the end of the decoding process. The PostNet takes the predicted Mel-spectrogram as input to obtain the residual parameters that are related to future context since the decoder RNN is unable to foresee future frames. The predicted Mel-spectrograms are finetuned with the PostNet, which leads to high-quality outputs.
The speaker embedding network is engaged after the PostNet during the training phase. It is added as a feedback constraint to force the TTS model to learn the speaker variety knowledge sufficiently so that the speaker characteristics extracted from synthesized Mel-spectrogram lays in the same distribution as those extracted from the natural speech from the same person. In this case, the parameters of the speaker embedding network are not updated during the training phase.
We use the cosine distance between the ground truth speaker embedding and the one extracted from the predicted Mel-spectrogram by speaker embedding network as one of the loss functions for optimizing the TTS network. Other than that, mean square error (MSE) between predicted Mel-spectrograms and the ground truth spectrogram, classification loss of the stop tokens, and the regularization loss for encoder-decoder parameters are applied to ensure correct predictions.
The Mel-spectrogram is converted back to the audio signal by the neural vocoder WaveRNN [@kalchbrenner2018efficient], which is able to generate high-quality speech at fast speed.
Experiments {#result}
===========
We used four publicly available datasets for training and evaluation. All data from Voxceleb1 [@nagrani2017voxceleb] and Voxceleb2 [@chung2018voxceleb2], with more than 7, 000 speakers, are used for training the speaker verification system. The VCTK English dataset [@veaux2016superseded], which contains 109 speakers with various accents, is used for TTS model training, while data from 8 speakers are randomly excluded as the VCTK test set. For each speaker in the training set, 8 utterances are randomly picked out as the VCTK validation set. 7 speakers from the Librispeech dataset [@panayotov2015librispeech] are randomly selected as the cross-domain test set. All audios are downsampled to 16 kHz in our experiments.
![Subjective preference result[]{data-label="fig:objeva"}](objeva.pdf){width="47.00000%"}
We evaluate the performance by comparing the proposed system, which has added feedback constraint (FC), with the multispeaker TTS baseline system without FC. The two systems are identified by **‘baseline’** and **‘FC’** in the following subjective and objective results. We first trained the baseline model until it can synthesize intelligible speech. Then the FC model is trained from the pre-trained baseline model while engaging the speaker embedding network. Both models are then trained to the same total training steps with the same batch size.
Subjective evaluation
---------------------
We asked 12 people to choose their preferable speech for pairs that contain speech synthesized by both systems. Audios in each pair are synthesized with the same text content and the conditioned embedding from the same reference audio. Each person chose their preference concerning speaker similarity and naturalness from 38 pairs, which are randomly selected from the VCTK test set, the VCTK validation set, and the Librispeech set. Preference results are shown in figure \[fig:objeva\]. For subjective evaluation, the FC system outperforms the baseline system on all three evaluation sets. For seen speakers in the training data, the speech synthesized by both systems is close. Hence people do not have a preferred choice for more than $50\%$ pairs in the VCTK validation set. Given these points, both systems can copy seen voices well, while the FC system obtains better performance on unseen voice cloning.
[|l|c|c|c|]{} & Systems &
-------------
SV-EER (%)
Dep / Indep
-------------
: Objective evaluation results[]{data-label="tbl:objeva"}
&
-------------------
Average
cosine similarity
Dep / Indep
-------------------
: Objective evaluation results[]{data-label="tbl:objeva"}
\
& natural & 1.76 & -\
& baseline & 14.72 / 13.18 & 0.403 / 0.333\
& FC & 8.22 / 7.68 & 0.764 / 0.577\
& natural & 1.61 & -\
& baseline & 9.22 / 8.23 & 0.472 / 0.394\
& FC & 5.02 / 3.42 & 0.842 / 0.67\
& natural & 5.26 & -\
& baseline & 26.84 / 26.46 & 0.222 / 0.139\
& FC & 16.54 / 16.11 & 0.626 / 0.389\
Objective evaluation
--------------------
For each utterance from all evaluation datasets, we synthesized speech according to the given transcript and the embedding extracted from the original speaker’s voice. Two different synthesized results were collected for each utterance. Although both are synthesized with the same reference voice, one is synthesized based on the speaker embedding extracted from the utterance that has the exact same content, while the other result is synthesized with the speaker embedding extracted from a randomly selected utterance with different content. These are identified as text-dependent **(Dep)** result and text-independent **(Indep)** result in table \[tbl:objeva\]. Speaker verification equal error rate **(SV-EER)** is used to evaluate the speaker discrimination performance for a set of embeddings. We randomly generate enrollment-verification pairs for each experiment, where half of the trials are cross-speaker pairs. We also compute the average cosine similarity between embeddings extracted from synthesized speech and the ground truth embeddings to measure the speaker similarity performance objectively. As shown in table \[tbl:objeva\], the FC system obtains significantly lower EERs than the baseline system on all evaluation sets, whether text-dependent or text-independent. The FC system also has higher average cosine similarities than the baseline system. In either case, we can conclude that the voice synthesized by the FC system is more close to the reference voice than the baseline system. The similarity is improved with the feedback constraint network.
Likewise, we can visualize the results from the embedding space by the t-Distributed Stochastic Neighbor Embedding (t-SNE), as shown in figure \[fig:vctk\_textdep\]. The utterances synthesized by the baseline system with reference embeddings from the same speaker is in the same cluster, but do not have the same distribution with reference embeddings, even is closer to another speaker. For example, as shown in figure \[fig:vctk\_textdep\] $(a)$, embeddings from ‘p225-syn’ have a distribution that is close to ‘p376-org’ other than its reference speaker ‘p225-org’. For the FC system, the embeddings extracted from the same voice, either synthesized or natural, lie in the same distribution in the embedding space. Therefore, the synthesized voice is more close to the original speaker for utterances synthesized by the FC system.
Conclusion
==========
In this paper, a multispeaker TTS approach that explores the use of a speaker verification system is presented. A trained speaker verification system is combined into the TTS framework acting as the feedback constraint to facilitate voice cloning. Experimental evaluations, including both subjective and objective evaluations, demonstrate that our proposed system enhances the knowledge transfer from speaker verification to speech synthesis. Accordingly, our proposed method achieves significant improvement regarding voice cloning, which can be used for data augmentation or white-box spoofing attack in the future.
[^1]: https://caizexin.github.io/mlspk-syn-samples/index.html
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We analyze the electric power load in the Czech Republic (CR) which exhibits a seasonality as well as other oscillations typical for European countries. Moreover, we detect 1/f noise property of electrical power load with extra additional peaks that allows to separate it into a deterministic and stochastic part. We then focus on the analysis of the stochastic part using improved Multi-fractal Detrended Fluctuation Analysis method (MFDFA) to investigate power load datasets with a minute resolution. Extracting the noise part of the signal by using Fourier transform allows us to apply this method to obtain the fluctuation function and to estimate the generalized Hurst exponent together with the correlated Hurst exponent, its improvement for the non-Gaussian datasets. The results exhibit a strong presence of persistent and the dataset is characterized by a non-Gaussian skewed distribution. There are also indications for the presence of the probability distribution that has heavier tail than the Gaussian distribution.'
address:
- 'Charles University in Prague, Faculty of Social Sciences, Institute of Economic Studies, Opletalova 26. CZ-11000 Prague 1, Czech Republic'
- 'Czech Technical University in Prague, Faculty of Nuclear Sciences and Physical Engineering, Department of Physics, Břehová 7, CZ-11519 Prague 1, Czech Republic'
- 'Bogolyubov Laboratory of Theoretical Physics, Joint Institute of Nuclear Research, RU-141980 Dubna, Russia'
author:
- Jiří Kracík
- Hynek Lavička
title: Fluctuation analysis of high frequency electric power load in the Czech Republic
---
MFDFA, electric power load, Hurst exponent, persistent process, 1/f noise, non-Gaussian distribution
Introduction
============
The responsibility for the safe and reliable operation is one of the basic duties of the national Transmission System Operator (TSO). The gradual liberalization of the European electricity market led to a necessity of the integration of mutually uncoordinated transmission systems. The enhancements of these transmission systems are very intensive in terms of both the time as well as capital investments and due to this the current energy networks are reaching their technical limits. That is mostly obvious in case of a massive increase of the offshore wind power plant installations located in the distant parts, hundreds of kilometers far from the end consumer. The electricity, which cannot pass through the under-dimensioned transmission lines or so called congestions, flows through the surrounding system which must accommodate these unscheduled flows. Unfortunately, the market with electricity and its mechanisms do not reflect this fact. In our work, we analyze high frequency data of electricity consumption in the Czech Republic and we also determine the degree of the uncertainty of the behavior of the consumers.
The analysis of the electricity prices and loads has been discussed by R. Weron [@key02]. He stated that the electricity loads, which are non-stationary time series, are combinations of both the trends and the periodic cycles with a random component. It is known from literature that electricity loads are correlated with the weather (e.g., the temperature, see [@key02; @Peirson1994235; @Lee2011896]) as well as with socio-economical changes and processes.
The first method (R/S method) for a non-stationary time series analysis was invented by H.E. Hurst [@key29]. Since its introduction the method has been tested on various datasets and also implemented very effectively on computer [@key08; @key16; @key17; @key20]. The method estimates the Hurst exponent of dataset that is related to the exponent of the autocorrelation function from the theory of fractional Brownian motion [@key32; @key24]. A modern alternative of the Hurst exponent estimation for series with local trends is the Detrended Fluctuation Analysis (DFA) which was introduced in [@key14; @key15] and used for economy datasets [@key03], heart rate dynamics [@key23; @key25], DNA sequences [@key26; @key15; @key14], long-time weather records [@key13], electricity prices time series [@key16; @key17] and wind speed records [@key37]. Recently, the DFA was improved to quantify the fluctuation function of datasets using different metrics [@key06; @key07]. The MFDFA is able to estimate the exponent of the autocorrelation function and also the exponent of the probability distribution function. In recent years, there has been a considerable focus on the investigation of multifractal cross-correlation between a pair of synchronized datasets [@key40].
There is a broad literature of modelling and forecasting methods of both price and/or load time series. It usually incorportes the Autoregressive Moving Average processes (ARMA), the Vector Autoregression (VAR), the Vector Error Correction (VECM), machine learning, an adaptive neuro-fuzzy network and a customers segmentation. Fixed mean, restricted variance and normally distributed error term represent basic assumptions for finding the best linear unbiased estimation, for a summary see Ref [@key42; @key45; @key46].
In this paper we study a dataset of electric power load in the Czech Republic since $2008$ till $2011$ with a one-minute time step. We focus on the properties of the fluctuation function where the first periodic part of the signal is filtered from the dataset and then the MFDFA is used. Our main aim is to determine the Hurst exponent which provides information about the autocorrelation function as well as the probability distribution. We also validate the assumptions of the normal (Gaussian) noise distribution and the short-range correlations.
The paper is organized as follows: In section 2 we describe the methodology of data processing. We first describe the Fourier filtering method and then the MFDFA. In section 3 we analyze the dataset using the methodology from section 2. Finally, in section 4, we draw the conclusions of our study.
![The electric power load in Czech Republic between January 1$^{st}$ 2008 and December 31$^{st}$ 2011 (top) and the stochastic part $P^{stoch}$ obtained by the filtration of the signal (bottom).[]{data-label="fig:Load"}](load_2008-2011)
![The power spectrum of the electric power load between 1$^{st}$ of January 2008 and 31$^{st}$ of December 2011 spanning $4$ years in total showing the 1/f property with the extra peaks. The blue points show one-year (Y), one-week (W), one-day (D) and 12 hour periods (12H). In the inset of the plot, there is the dependence of the RMSE on $\beta$ for the different exponents of filtration (parameter $\alpha$). The vertical black line shows the actual location of the parameter $\beta$ used for the filtration procedure.[]{data-label="fig:Power_spectrum"}](power_spectrum_2008-2011)
Methodology
===========
Human behavior datasets typically exhibit the oscillations with the periods related to the units of calendar [@key02; @key13] and the same applies to the electric power load. The one-year and one-week oscillations are clearly visible in Fig. \[fig:Load\] but the presence of other frequencies is not so easily observable. To obtain the information regarding the strength of the oscillations we employ the Power spectrum which is shown in Fig. \[fig:Power\_spectrum\]. It depicts additional periods with the lengths of one day and $12$ hours beside the others. Moreover, since the periods of the power loads do not follow harmonic functions, we can also observe peaks at the positions of the integer multiples of a typical trend. The reconstruction of the original load on the basis of these most significant trend components is influenced by randomness which is represented by less significant components of the Power spectrum.
Motivation
----------
In our study, we focus on the properties of the random part and we use the MFDFA which is popular among scientists [@key16; @key14; @key15; @key23; @key25; @key03; @key06; @key07; @key11] as an effective tool for extracting the properties of a long-range memory within the time series.
Since time series generally might be non-stationary, polynomial trends may still govern them. The basic idea of the DFA is to strip off the trends and use the residues for the further analysis. In the MFDFA, we are looking for typical patterns, which govern the time series manifesting a self-affine property defined by $X\left(c\cdot t\right)=c^{H}\cdot X\left(t\right)$. The generalized Hurst exponent H, determined by the method, is the measure of the long term memory in the time series and it is directly related to the non-integer fractal dimension D.
The disadvantage of this method is that the periodic trends disturb the estimation of the Hurst exponent [@key09] and therefore, before we employ the method, we have to filter out the oscillations from the signal.
We use the Fourier transform to execute the filtration. The MFDFA itself then removes the polynomial trends. The resulting signal is decomposed as
$$\begin{aligned}
P\left(t\right) & = & P^{stoch}\left(t\right)+P^{deter}\left(t\right),\label{eq:Decomposition_of_signal}\end{aligned}$$
where $P^{deter}$ describes the periodic behavior of the system, while $P^{stoch}$ stands for the random part.
We used a regression model with dummy variables indicating holidays and we perform the method described below. We observed negligible differencies for the low orders of the MFDFA but the observable differences for the higher orders of the MFDFA. However, the widths of the multifractal spectrums are negligible in both cases.
Mathematical description
------------------------
We execute our analysis in three steps. First, we perform the Fourier transform to separate the signal into the stochastic and the deterministic part by the Fourier transform. In the next step, we execute the MFDFA. Finally, in the last step, we calculate the correlated Hurst exponent, which requires shuffling of the original-time series. It is an improvement of the typically used generalized Hurst exponent, exploited in cases, where we have an assumption of the non-normally distributed time series.
### Fourier transform filtering
We deal with a signal in the discrete time-steps $P\left(t_{n}\right)$ where $t_{n}=t_{1}+n\cdot\Delta t$ and $n\in M\equiv\{1,\ldots,N\}$. Since the Discrete Fourier Transform of the signal is $\widehat{P}\left(m\right)=\frac{1}{\sqrt{N}}\sum_{n\in M}\exp\left(-\frac{2\pi\boldsymbol{i}\cdot n\cdot m}{M}\right)P\left(t_{n}\right)$, and the related Power spectrum $S\left(m\right)=\widehat{P}\left(m\right)\cdot\widehat{P}^{*}\left(m\right)$, where $x^{*}$ stands for conjugation). The Power spectrum, see Fig. \[fig:Power\_spectrum\], of the signal $P\left(t_{n}\right)$ exhibits a power law-like shape with extra peaks and each coefficient of the Fourier transform is separated into two parts according to the threshold $\beta\cdot m^{-\alpha}$:
- discrete significant coefficients in the Power spectrum for certain frequencies above the threshold forms $\vert\widehat{P^{deter}}\left(m\right)\vert$;
- coefficients below the threshold forms $\widehat{P^{stoch}}\left(m\right)$ ;
where $\alpha$ and $\beta$ are the parameters set with regard to the chosen RMSE level. We also note that if $\widehat{P^{deter}}\left(m\right)\neq0$ then we define $\arg\widehat{P^{deter}}\left(m\right)=\arg\widehat{P^{stoch}}\left(m\right)=\arg\widehat{P}\left(m\right)$. Otherwise $\arg\widehat{P^{deter}}\left(m\right)$ is not defined. The Fourier transform of the sub-signals $\widehat{P^{deter}}\left(m\right)$ and $\widehat{P^{stoch}}\left(m\right)$ then follow $\widehat{P}\left(m\right)=\widehat{P^{deter}}\left(m\right)+\widehat{P^{stoch}}\left(m\right)$, which is the Fourier transform of . \[eq:Decomposition\_of\_signal\]. By executing the the inverse Fourier transform $P\left(t_{n}\right)=\frac{1}{\sqrt{N}}\sum_{n\in M}\exp\left(\frac{2\pi\boldsymbol{i}\cdot n\cdot m}{M}\right)\widehat{P}\left(m\right)$ we obtain a deterministic part $P^{deter}\left(t_{n}\right)$ from $\widehat{P^{deter}}\left(m\right)$ . The later part $\widehat{P^{stoch}}\left(m\right)$ is transformed to $P^{stoch}\left(t\right)$.
To measure the quality of the filter we use a root mean square error, see inset of Fig. \[fig:Power\_spectrum\], defined as follows:
$$RMSE=\frac{\sqrt{N\sum_{i=1}^{N}\left(P^{deter}\left(t_{i}\right)-P\left(t_{i}\right)\right)^{2}}}{\sum_{i=1}^{N}P\left(t_{i}\right)}.\label{eq:RMSE}$$
The level of the error was determined both to decrease the RMSE and to prevent $P^{stoch}$ from incorporating a periodic function that produces the artificial behavior of the fluctuation function.
### Multi-fractal Detrended Fluctuation Analysis
We employ the Multi-fractal Detrended Fluctuation Analysis (MFDFA) for analyzing the filtered signal $P^{stoch}\left(t_{i}\right)$. The method is employed as an effective tool to avoid the artificial (see Ref. [@key22]) in the autocorrelation function or in the Power spectrum due to the oscillation of the electric power loads and the presence of the peaks in the Power spectrum, see Fig. \[fig:Power\_spectrum\].
Each element $\left\{ x_{i}\equiv P^{stoch}\left(t_{i}\right)\right\} $ of the dataset is indexed by $i\in M$. The application of the MFDFA consists of five steps:
Step 1. Integration of the dataset to produce the dataset $X_{j}=\sum_{i=1}^{j}x_{i}$. The “double” integration of the dataset $\tilde{X}_{j}=\sum_{i=1}^{j}X_{i}$ is also performed.
Step 2. Division of the dataset $X_{i}$ into $L_{s}\equiv\left\lfloor \frac{N}{t}\right\rfloor $ overlapping segments $X_{j,k}$ with length $s$ and $j\in\left\{ 1,\ldots,s\right\} $.
Step 3. Use of a standard (least-square) regression method of fixed order $M$ on each segment $X_{j,k}$ to obtain the local trend $T_{k}\left(x\right)$ in the region $x\in\left[1,s\right]$.
Step 4. Calculation of the sample variance for each of the $L_{s}$ segments of the original dataset $$V\left(k\right)\equiv\frac{1}{s}\sum_{j=1}^{s}\left(X_{j,k}-T_{k}\left(j\right)\right)^{2}.$$
Step 5. Averaging over all the segments of the original dataset to obtain the multi-fractal fluctuation function
$$F_{q}\left(s\right)\equiv\begin{cases}
\left(\frac{1}{L_{s}}\sum_{k=1}^{L_{s}}V^{\frac{q}{2}}\left(k\right)\right)^{\frac{1}{q}} & \textrm{if}\ q\neq0\\
\exp\left(\frac{1}{2\cdot L_{s}}\sum_{k=1}^{L_{s}}\ln V\left(k\right)\right) & \textrm{if}\ q=0
\end{cases}.\label{eq:Fluctuation_function}$$
In the analysis we investigate the properties of the fluctuation function $F_{q}\left(s\right)$ on the window of the size $s$ and on the measure $q$. Generally, $F_{q}\left(s\right)$ grows with increasing $s$ for all $q$ (see Fig. \[fig:MF-DFA-2\] or follow original literature [@key06; @key07; @key09; @key15; @key14; @key23; @key27]), following the power law
$$\begin{aligned}
F_{q}\left(s\right) & \sim & s^{H\left(q\right)+1}.\label{eq:Definition of Hurst exponent}\end{aligned}$$
The exponent $H\left(q\right)$ is called the Hurst exponent, see Ref. [@key29]. Generally, it is related to the long-term autocorrelation or the heavy-tailed distribution of the governing process, see Ref. [@key06; @key07]. We also note that $+1$ in Eq. \[eq:Definition of Hurst exponent\] stands due to the application of the double integration instead of the single integration of dataset, for discussion, please, see Ref. [@key06].
We exploit a fractal spectrum to analyze whether the dataset is governed by a single exponent or by a set of exponents. We define a scaling function by formula:
$$\tau\left(q\right)=q\cdot H\left(q\right)-1.\label{eq:scaling_function}$$
We define a fractal spectrum as the Legendre transform of $\tau\left(q\right)$ using the definition of a new variable $\pi=\frac{d\tau}{dq}$:
$$f\left(\pi\right)=q\cdot\pi-\tau.\label{eq:multifractal_spectrum}$$
Generally, the fractal spectrum allows to distinguish mono- and multifractal processes. The width of the fractal spectrum is defined by the formula $\Delta\pi=\max_{q\in\mathbb{{R}}}\pi-\min_{q\in\mathbb{{R}}}\pi$. The value of $\pi$ in peak of $f\left(\pi\right)$ denoted by $\pi^{max}$ represents the most frequent value of the exponent. As the width of the fractal spectrum goes wider, the number of admitted exponents increases and the monofractality shifts to the multifractality.
### Shuffling of the stochastic part of the time series
Generally, if a stochastic process generates the time series following a non-normal (non-Gaussian) distribution, the generalized Hurst exponent $H\left(q\right)$ combines the information about the autocorrelation function influenced by the properties of its probability distribution. We extract the correlation Hurst exponent $H^{cor}\left(q\right)$ that separates the generalized Hurst exponents calculated using the original time series and calculated using the shuffled one[^1].
While executing the shuffling procedure, we destroy the autocorrelations (if present) within the sample. Then we use a standard MFDFA described in previous section to the calculate shuffled fluctuation function:
$$F_{q}^{shuf}(s)=\overline{F_{q}\left(\left\{ x_{i}\right\} ^{shuf}\right)(s)},$$ where $\overline{x}$ stands for the averaging samples of shuffling and $\left\{ x_{i}\right\} ^{shuf}$ means shuffling of the time serie $x_{i}$. Finally, we estimate the generalized Hurst exponent of the shuffled time serie $H^{shuf}\left(q\right)$. As it was noted in the previous paragraph, the correlation Hurst exponent is then defined by following formula:
$$H^{cor}(q)=H(q)-H^{shuf}(q).\label{eq:CorrelatedHurstexponent}$$
Analogically to the generalized Hurst exponent $H\left(q\right)$ we can define the correlation fractal spectrum $f^{cor}\left(\alpha\right)$ related to $H^{cor}\left(q\right)$ by the formulas \[eq:scaling\_function\] and \[eq:multifractal\_spectrum\].
Implementation of the method
----------------------------
We used a multi-threaded implementation of the MFDFA with Zarja library [@key01] [^2] which can effectively run on multi-core cluster computers. We also compared the results with the implementations used in [@key27; @key26; @key23]. The filtration of dataset was executed in the Python using the NumPy and SciPy modules [@key30; @Key31].
Analysis of dataset
===================
![The quantile diagram of the probability density function generated from $P^{stoch}$ and its counterpart generated from it by the regression model with dummy variables $P^{stoch,dummy}$ and their comparison with the normal distribution with the same mean $\mu$ and variance $\sigma^{2}$. In the insets, we show the comparison of the probability density functions of $P^{stoch}$ with the appropriate normal distribution (the normal plot at the bottom, the log-normal scale at the top).[]{data-label="fig:Quantile diagram"}](quantile_diagram_2008-2011)
![The fluctuation function $F_{q}\left(s\right)$ of the signal $P^{stoch}$ obtained using the MFDFA of the order $4$ for various $q$s. We present $q\in\left\{ -10,-5,-2,0,2,5,10\right\} $ from the bottom to the top, respectively. Each plot is multiplied by factor $10$ from its predecessor. []{data-label="fig:MF-DFA-2"}](DFA_comparison_2008-2011)
![The correlation Hurst exponent $H^{cor}\left(q\right)$ estimated using the MFDFA of orders $2-6$. In the inset, we show the generalized Hurst exponent $H\left(q\right)$ for the same MFDFA orders. We used the dataset obtained by the regression model with dummy variables indicating the holidays. The dataset without use of the method follows the analogous pattern.[]{data-label="fig:Hurst exponent"}](hurst_exponent_2008-2011)
![The multifractal spectrum $f\left(\pi\right)$ of $P^{stoch}$ for various orders of the MFDFA method and initial detreding with parameter $\alpha$ is shown on the top. In the middle we present the correlation and shuffled multifractal spectrum, respectively.[]{data-label="fig:multifractal_spectrum"}](multifractal_spectra_2008-2011)
![The generalized Hurst exponents (top) and the multifractal spectrums (middle, bottom) of the surrogate datasets which underwent phase randomization. The middle figure show the multifractal spectrums for the surogate dataset and bottom one illustrates dependence on the subset of the dataset (each color shows the different subset). The top and middle figure plots dependence on the order of the method. The bottom subplot is for order $4$ of the method.[]{data-label="fig:surrogate_multifractal_spectra"}](surrogate_multifractal_spectra_2008-2011)
Data description
----------------
Our dataset describes the electric power load of the Czech Republic which is monitored by national Transmission System Operator (TSO), ČEPS a.s. It was calculated with high frequency from the stored data using the formula:
$$P\left(t\right)=\sum_{i\in1}^{M}T\left(t,i\right)-E\left(t\right)+I\left(t\right)+P_{u}\left(t\right),\label{eq:Electric power load}$$
where $T\left(t,i\right)$ stands for $i$-th turbo-generator output of the total number $M$. The turbo-generators are directly measured from their minimal value of $100\ kW$ of installed capacity. $E\left(t\right)$ and $I\left(t\right)$ are the exports and imports, respectively. Generally, they are a kind of bottlenecks because there are only few direct transmission lines between the Czech Republic and the neighboring countries. Finally, $P_{u}\left(t\right)$ stands for the balance of the pumped-storage hydroelectricity[^3].
The dataset is calculated in real time from various sources and the datalinks are not generally completely error-proof. Each datapoint is thus accompanied with the confidence flag indicating the credibility of the source. Some datapoints are calculated, using Eq. \[eq:Electric power load\], others are interpolated.
Our dataset consists of $N=2,103,840$ datapoints and it spans $4$ years since $2008$ till $2011$ with a one-minute time step. In our analysis, we neglect the confidence flag and we use the electric power load measured in $\mathrm{MW}$ only.
Results of Fourier filtering
----------------------------
The electric power load dataset of the Czech Republic is depicted in Fig. \[fig:Load\], where the Power spectrum $S\left(\omega\right)$ exhibits the power law with extra significant peaks, see Fig. \[fig:Power\_spectrum\], and therefore we first execute the Fourier filtering of the dataset where we assume $\vert\widehat{P^{stoch}}\left(m\right)\vert=\beta\cdot m^{-\alpha}$ with parameters $\alpha$ and $\beta$ yet to be determined. In our study we mainly choose $\text{\ensuremath{\alpha}}=0.7$ as an approximation of the best fit of this exponent and in order to the prove robustness of the method, we also plot the $RMSE$ for the two other values close to the chosen value of the parameter $\alpha$, see inset of Fig. \[fig:Power\_spectrum\]. The extensive test of the dependence of the fractal spectra on the exponent $\alpha$ is shown in Fig. \[fig:multifractal\_spectrum\].
Then we construct the dependence of the $RMSE$ on the parameter $\alpha$ and we choose the break-point of this dependence as an $\alpha$ value. The $RMSE$ is defined by Eq. \[eq:RMSE\] and at $\beta=7\cdot10^{8}\ \mathrm{MW}$ (we note that it is the equivalent of $S\left(m\right)\sim m^{-2\alpha}$). The $P^{stoch}$ does not exhibit large periodic fluctuations (see the bottom part of Fig. \[fig:Load\]) and the quantile diagram as well as the probability density distribution around the mean behave close to the normal (Gaussian) distribution (Fig. \[fig:Quantile diagram\]). We note that the choice of $\beta=2\cdot10^{9}\ \mathrm{MW}$ leads to both the significant deviation from normal distribution in its center part as well as to the increase of the periodicity in the stochastic part. The filtered signal is shown at the bottom of Fig. \[fig:Load\] and it is then more analyzed.
Results of application of MFDFA
-------------------------------
Firstly, we investigate the probability distribution function of the time series $P^{stoch}$ despite of the fact that there can still be temporary trends, see Fig. \[fig:Quantile diagram\]. The comparison of the quantile diagram, the mean and the variance of $P^{stoch}$ with quantiles of the normal (Gaussian) distribution is presented in Fig. \[fig:Quantile diagram\]. It clearly shows the deviations for the small values of the power load. In the lower right inset in Fig. \[fig:Quantile diagram\], the comparison of the histogram of $P^{stoch}$ with the appropriate normal distribution exhibits a good approximation about the average. In the upper left inset in Fig. \[fig:Quantile diagram\], we can observe the deviations of the small values of the power load from the normal distribution in the semi-logarithmic scale.
In the next step, we perform the MFDFA to calculate the fluctuation function $F_{q}\left(s\right)$ and we estimate the generalized Hurst exponent $H\left(q\right)$, see the inset of Fig. \[fig:Hurst exponent\] in range $[2\cdot10^{3},2\cdot10^{5}]$. The generalized Hurst exponent depends on $q$ we expect presence of multifractality. To get valuable information about the autocorrelation function, we shuffle the dataset to calculate the fluctuation function $F_{q}^{shuf}\left(s\right)$. The ratio of the original fluctuation function $F_{q}\left(s\right)$ against the fluctuation function of the shuffled dataset $F_{q}^{shuf}\left(s\right)$ formulated as $F_{q}^{cor}\left(s\right)=\frac{F_{q}\left(s\right)}{F_{q}^{shuf}\left(s\right)}$ follows the power law similarily as $F_{q}\left(s\right)$ see Fig. \[fig:MF-DFA-2\]. Then the calculation of the correlation Hurst exponent $H_{q}^{cor}\left(s\right)$ is performed using the formula \[eq:CorrelatedHurstexponent\]. We show $H_{q}^{cor}\left(s\right)$ in Fig. \[fig:Hurst exponent\] and the exponent stands between the values of $0.55$ till $0.8$ (in contradiction to the calculation of the generalized Hurst exponent based on the normally distributed time series), showing a strong persistence. Additionally we note that the estimation of the Hurst exponents is stable with regard to the orders of the MFDFA.
In Fig. \[fig:multifractal\_spectrum\], the fractal spectrum $f\left(\pi\right)$, the correlation fractal spectrum $f^{cor}\left(\pi\right)$ and also the shuffled fractal spectrum $f^{shuf}\left(\pi\right)$ of the stochastic part $P^{stoch}$ are not concentrated at single $\pi$ but they are broadly spread among the wide range of $\pi$s. conclude that the processes are multifractal in the distribution as well as in the correlation function. However, multifractality of the correlation function is stronger $\Delta\pi^{cor}\cong0.3$ in contrast to the multifractality of the distribution function $\Delta\pi^{shuf}=0.15$ for the same order of the method.
Tests of stability of the results
---------------------------------
The above mentioned results of the analysis may depend on additional factors. To address the factors we execute additional tests to show the invariance of the conclusions.
### Stability of results with respect to the filter
As a test of the stability of the results, we performed multiple calculations of MFDFA for different values of the parameters $\alpha$ and $\beta$. The generalized Hurst exponent as well as the fractal spectrum depend on a particular value of $\alpha$ and it is independent on the order of the method, see Fig. \[fig:multifractal\_spectrum\]. The change of the order does not significantly imply the change of the width of the fractal spectra. On the other hand, the shuffled fractal spectrum is independent on the value of $\alpha$ and it is localized around $\frac{1}{2}$ – the value of the Gaussian distribution. The persistence of the time series is conserved in the proximity of $\alpha=0.7$, see the middle of the Fig. \[fig:multifractal\_spectrum\].
### Surrogate data test
Generally, there are usually two reasons of the multifractality in time series:
- long range correlations of small and large fluctuations within the time serie;
- heavy-tailed probability distribution function (not necessarily the Lévy $\alpha$-stable distribution, see Ref. [@key55]).
The long-range correlation property and the fat-tailed probability distribution are investigated by shuffling and by a phase randomization. Shuffling destroys the correlations within the time series but it preserves the probability distribution. On the other hand, the phase randomization preserves the correlation function but weakens both the non-Gaussian and non-linear properties of the time serie. The procedures were firstly proposed in Ref. [@key50] and a review of its use can be found in Ref. [@key51]. We note that this method was initially used in the context of the MFDFA in Ref. [@key52].
We practically performed the test on $50$ samples of the surrogate datasets and we present the results in Fig. \[fig:surrogate\_multifractal\_spectra\]. In the graph in the top we can see similar results of $H$ as in the inset in the Fig. \[fig:Hurst exponent\]. In the middle graphs there is the result comparable with the top graphs in the Fig. \[fig:multifractal\_spectrum\]. We obtained the width of the fractal spectra $\Delta\pi\cong0.3$ and the location of the maximum is around $\pi^{max}\cong1.2$. We conclude that the multifractality is not caused by non-linearity and beside that there are the indications of the presence of a distribution with the tail heavier than the Gaussian distribution possess. From theory of the stable distributions and the stochastic processes, Refs. [@key55; @key60], the Gaussian distribution possess $H\left(2\right)=\frac{1}{2}$ and the Lévy $\alpha$-stable distribution $H\left(2\right)=\frac{1}{\omega}$ where $\omega$ is the exponent of the tail (for the Gaussian distribution we have $\omega=2$). We obtained for the shuffled multifractal spectra, where shuffling erases the autocorrelations with in the time series, see the bottom of Fig. \[fig:multifractal\_spectrum\], wide peak around $\pi^{shuf}=\frac{1}{2}$. Based on the assumption that the probability distribution is stable we admit presence the Lévy $\alpha$-stable distribution with the exponents $\omega$ close to the values of the Gaussian distribution. We also note that the result is independent of the set up of the initial filtering method. Additionally the Lévy $\alpha$-stable distribution must be skewed due to indications in Fig. . We also tested the influence of using a regression model with dummy variables for the decrease of the effect of holidays. As you can see on Fig. \[fig:Quantile diagram\], the result is not significant.
### Problems of stationarity and deficient random generators
We applied the Augmented Durbin-Watson test on the $P_{stoch}$ and we rejected the null hypothesis of non-stationarity at the $5\%$ significance level.
As a test of the stability of the results we separated the original dataset into $8$ sets with equal size and we executed the proposed method for each segment. The results of the method are at the bottom of Fig. \[fig:surrogate\_multifractal\_spectra\] where the curves representing the surrogate fractal spectra show the overlap with the width of fractal spectra $\Delta\pi^{sur}\cong0.3$ and $\pi^{max}\cong1.3$. These values are approximately equal to the results of the complete dataset. Thus, we conclude, that the results of the method are stable with the respect to the change of the scale.
Conclusions and Outlook
=======================
The main contribution of this paper is an analysis of the high-frequency electric power loads dataset of the Czech Republic using the improved MFDFA methodology. We discovered that the power spectrum of the signal exhibits 1/f noise property with the additional peaks that are caused by a periodic behavior of the electricity consumption. Based on that fact, we first separated the noise from modulating signal and then we applied the MFDFA without dealing with an artificial behavior of the fluctuation function, see Ref. [@key09]. After that we exploited the MFDFA for the analysis of the dataset to obtain information about the autocorrelation function. The major part of the power load is governed by oscillations. Beside that we report a strong persistence of the power loads where the distribution function exhibits non-Gaussian properties. The fractal spectra of both the distribution as well as the autocorrelation function indicate the presence of multifractality. We also performed a test using surrogate datasets as well as a test of stationarity to validate the strength of our conclusions. The analysis suggests the presence of the probability distribution with the tails heavier than the Gaussian distribution.
Some of our results are in contradiction with the previously published work analyzing electricity consumption and also with the assumptions of electricity load prediction models [@NowickaZagrajek20021903; @key02; @Pappas20081353; @Pappas2010256]. First, our analysis indicates that the stochastic part of the signal is not normally distributed, second, the distribution function is skewed and it may even have infinite moments of the probability distribution and third, the autocorrelation function is persistent. We also conclude that the estimations of risks based on traditional forecasting methods using the Gaussian distribution and short-range correlations are not usable due to both the long-range autocorrelation and the probability distribution’s extremes. The main part of the load constituting approximately $95\%$ of the signal was filtered out and it is systematically driven by external factors. Modeling by means of a regime-switching model makes a good sense to us.
The Czech transmission system is sufficiently dimensioned to cope with electricity consumption fluctuations contained in the dataset we had at our disposal. The problem that attracts actual attention of the TSO is dealing with the unexpected flows from north to south of Europe through the Czech Republic, see Ref. [@key28], which are caused by inhomogeneity of sources generating electricity and consumption of electricity in Europe. The presented approach might also be applied to solve a more complex problem, where in addition to the uncertainty of the electricity consumption, we may also consider the uncertainty caused by real power inflows and outflows (imported and exported electricity) or the uncertainty due to differences between cross-border trading and real electricity flows (obeying Kirchhoff’s laws). The level of uncertainty is expressed as a deviation from foreseeable behavior described by polynomial trends and periodic oscillations.
Author contributions {#author-contributions .unnumbered}
====================
J.K. obtained and prepared the dataset. H.L. prepared the tool for analysis and performed the analysis. J.K. and H.L. contributed to the writing of the manuscript. The work described in this paper will be used in J.K.’s Ph.D. thesis.
Acknowledgement {#acknowledgement .unnumbered}
===============
This article was supported by Czech Ministry of Education RVO68407700 and it also was written with the support of SVV project Strengthenning Doctoral Research in Economics and Finance. We thank for the fruitful discussion to P. Jizba, J. Lavička, A.M. Povolotsky, V.B. Priezzhev, E. Lutz, T. Kiss, G. Alber and H.E.Stanley.
Parameters and symbols of the methodology {#parameters-and-symbols-of-the-methodology .unnumbered}
=========================================
Value Symbol Unit
--------------------------------------- --------------------------- ----------------
Electric load $P$ $\mathrm{MW}$
Stochastic part of electric load $P^{stoch}$ $\mathrm{MW}$
Window size $s$ $\mathrm{min}$
Multifractal measure (parameter) $q$ $1$
Multifractal fluctuation function $F_{q}\left(s\right)$ $\mathrm{MW}$
Generalized Hurst exponent $H\left(q\right)$ $1$
Hurst exponent $H\equiv H\left(2\right)$ $1$
Correlation Hurst exponent $H^{cor}\left(q\right)$ $1$
Hurst exponent of shuffled time serie $H^{shuf}\left(q\right)$ $1$
Scaling exponent $\tau\left(q\right)$ $1$
[10]{} url \#1[`#1`]{}urlprefixhref \#1\#2[\#2]{} \#1[\#1]{}
R. Weron, Modeling and Forecasting Electricity Loads and Prices: A Statistical Approach, John Wiley and Sons Ltd, 2006.
J. Peirson, A. Henley, Electricity load and temperature: Issues in dynamic specification, Energy Economics 16 (4) (1994) 235 – 243.
C.-C. Lee, Y.-B. Chiu, Electricity demand elasticities and temperature: Evidence from panel smooth transition regression with instrumental variable approach, Energy Economics 33 (5) (2011) 896 – 902.
H. E. Hurst, Long term storage capacity of reservoirs, Trans. Am. Soc. Civ. Eng. 116 (770).
T. Preis, P. Virnau, W. Paul, J. J. Schneider, Accelerated fluctuation analysis by graphic cards and complex pattern formation in financial markets, New J. Phys. 11 (2009) 093024.
R. Weron, A. Przybylowicz, Hurst analysis of electricity price dynamics, Physica A 283.
R. Weron, Energy price risk management, Physica A 285 (2000) 127.
T.Preis, W. Paul, J. J. Schneider, Fluctuation patterns in high-frequency financial asset returns, Europhys. Lett. 82 (2008) 68005.
B. B. Mandelbrot, J. W. [Van Ness]{}, Fractional brownian motions, fractional noises and applications, SIAM Review 10 (4) (1968) 422–437.
H. A.Makse, S. Havlin, M. Schwartz, H. E. Stanley, Method for generating long-range correlations for large systems, Phys. Rev. E 53 (1996) 5445–5449.
C.-K. Peng, S. V. Buldyrev, S. Havlin, M. Simons, H. E. Stanley, A. L. Goldberger, Mosaic organization of [DNA]{} nucleotides, Phys. Rev. E 49 (1994) 1685–1689.
C.-K. Peng, S. V. Buldyrev, A. L. Goldberger, S. Havlin, F. Sciortino, M. Simons, H. E. Stanley, Long-range correlations in nucleotide sequence, Nature 356 (1992) 168–170.
R. Mantegna, H. Stanley, An introduction to econophysics: correlations and complexity in finance, Cambridge University Press, New York, NY, USA, 2000.
C.-K. Peng, S. Havlin, H. E. Stanley, A. L.Goldberger, Quantification of scaling exponents and crossover phenomena in nonstationary heartbeat time series, Chaos 5 (1995) 82–87.
J. W. Kantelhardt, Y. Askenazy, P. C. Ivanov, A. Bunde, S. Havlin, T. Penzel, J.-H. Peter, H. E. Stanley, Characterization of sleep stages by correlations in the magnitude and sign of heartbeat increments, Phys. Rev. E 051908.
A. L. Goldberger, L. A. N. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov, R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, H. E. Stanley, [PhysioToolkit]{}, and [PhysioNet]{}: Components of a new research resource for complex physiologic signals, Circulation 101 (2000) e215–e220.
P. Talkner, R. O. Weber, Power spectrum and detrended fluctaution analysis: Application to daily temperatures, Phys. Rev. E 64 (2000) 150–160.
K. Koçak, Examination of persistence properties of wind speed records using detrended fluctuation analysis, Energy 34 (2009) 1980–1985.
J. Kantelhardt, S. Zschiegner, E. Koscielny-Bunde, S. Havlin, A. Bunde, H. Stanley, Multifractal detrended fluctuation analysis of nonstationary time series, Physica A 316 (2002) 87.
J. Kantelhardt, E. Koscielny-Bunde, H. Rego, S. Havlin, A. Bunde, Detecting long-range correlations with detrended fluctuation analysis, Physica A 295 (2001) 441.
D. Horvatic, H. Staley, B. Podobnik, Detrended cross-correlation analysis for non-stationary time series with periodic trends, Europhys. Lett. 94 (18007).
S. Chan, K. Tsui, H. Wu, Y. Hou, Y.-C. Wu, F. Wu, Load/price forecasting and managing demand response for smart grids: Methodologies and challenges, IEEE Signal Processing Magazine (2012) 68 – 85.
T. Hong, P. Pinson, S. Fan, Global energy forecasting competition 2012, Int. J. Forecasting 30 (2) (2014) 357–363.
R. Weron, Electricity price forecasting: A review of the state-of-the-art with a look into the future, Int. J. Forecasting 30 (2014) 1030–1081.
A. Bunde, S. Havlin, J. W. Kanterhardt, T. Penzel, J. H. Peter, K. Voigt, Correlated and uncorrelated regions in heart-rate fluctuations during sleep, Phys. Rev. Lett. 85 (2000) 3736–3739.
K. Hu, P. C. Ivanov, Z. Chen, P. Carpena, , H. E. Stanley, Effect of trends on detrended fluctuation analysis, Phys. Rev. E 64 (2001) 011114.
S. Akselrod, D. Gordon, F. A.Ubel, D. C. Shannon, A. C.Barger, R. J.Cohen, Science 213 (1981) 220.
C.-K. Peng, J. Mietus, J. M. Hausdorff, S. Havlin, H. E. Stanley, A. L. Goldberger, Long-range anticorrelations and non-gaussian behavior of the heartbeat, Phys. Rev. Lett 70 (9) (1993) 1343–1346.
H. Lavička, Simulations of Agents on Social Network, LAP Lambert Academic Publishing, 2010.
H. P. Langtangen, A Primer on Scientific Programming with Python, Springer, 2009.
I. Idris, NumPy 1.5 Beginner’s Guid, Packt Publishing, 2011.
K.-I. Sato, Lévy Processes and Infinitely Divisible Distributions, Cambridge University Press, 1999.
J. Theiler, S. Eubank, A. Longtin, B. Galdrikian, J. Farmer, Testing for nonnonlinear in time series: the method of surrogate data, Physica D 58 (1992) 77–94.
T. Schreiber, A. Schmitz, Surrogate time series, Physica D 142 (2000) 346–382.
P. Norouzzadeh, W. Dullaert, B. Rahmani, Anti-correlation and multifractal features of spain electricity spot market, Physica A 380 (2007) 333–342.
G. Samorodnitsky, M. S. Taqqu, Stable Non-Gaussian Random Processes: Stochastic Models with Infinite Variance., Chapman and Hall, New York, 1994.
J. Nowicka-Zagrajek, R. Weron, Modeling electricity loads in [California]{}: [ARMA]{} models with hyperbolic noise, Signal Processing 82 (12) (2002) 1903 – 1915.
S. S. Pappas, L. Ekonomou, D. C. Karamousantas, G. E. Chatzarakis, S. K. Katsikas, P. Liatsis, Electricity demand loads modeling using autoregressive moving average [(ARMA)]{} models, Energy 33 (9) (2008) 1353 – 1360.
S. S. Pappas, L. Ekonomou, P. Karampelas, D. C. Karamousantas, S. K. Katsikas, G. E. Chatzarakis, P. D. Skafidas, Electricity demand load forecasting of the hellenic power system using an [ARMA]{} model, Electric Power Systems Research 80 (3) (2010) 256 – 264.
Z. Boldi[š]{}, Czech electricity grid challenged by [German]{} wind, Europhys. News 44 (4) (2013) 16–18. [](http://dx.doi.org/http://dx.doi.org/10.1051/epn/2013401).
[^1]: To shuffle the dataset we utilized Fisher-Yates algorithm that is effective even in the case of large dataset. In our case, we used the average of 100 samples of shuffling.
[^2]: Source code can be found at <http://zarja.sourceforge.net>.
[^3]: There are three of them - Dlouhé stráně $600\ MW$, Dale¨ice $450\ MW$ and ¦těchovice with $48\ MW$ of installed capacity.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This paper calculates the quantized energy levels of the hydrogen atom, using a metaplectic-c prequantization bundle and a definition of a quantized energy level that was introduced by the author in a previous paper [@v1]. The calculation makes use of a computational technique also demonstrated in that paper. The result is consistent with the standard quantum mechanical prediction. Unlike other treatments of the hydrogen atom, this approach does not require the construction of the symplectic reduction, but takes place over a regular level set of the energy function. The Ligon-Schaaf regularization map is used to transform the problem into one of determining the quantized energy levels of a free particle on ${\ensuremath{TS^3}}$.'
author:
- |
Jennifer Vaughan\
`[email protected]`
title: 'Metaplectic-c Quantized Energy Levels of the Hydrogen Atom'
---
\[section\] \[theorem\][Example]{}\[theorem\][Definition]{}\[theorem\][Lemma]{}
Introduction
============
The quantized energy levels of the hydrogen atom have been calculated for various physical models, using various flavors of geometric quantization. Notable examples include:
- Simms [@sim3], who used the observation that the space of orbits corresponding to a fixed negative energy is isomorphic to $S^2\times S^2$, and determined those energies for which the reduced manifold admits a prequantization circle bundle;
- Sniatycki [@sn2], who looked at the 2-dimensional relativistic Kepler problem and computed a Bohr-Sommerfeld condition for the completely integrable system given by considering the energy and angular momentum functions simultaneously;
- Duval, Elhadad and Tuynman [@det], who took the phase space to include the spins of the electron and proton, then chose a polarization and determined the Kostant-Souriau quantized operator corresponding to the energy function with fine and hyperfine interaction terms.
These examples exist at one of two possible extremes. On one hand, the quantized energy condition can be evaluated over the symplectic reduction of a particular level set of the energy function, as in [@sim3]. This definition only looks at one energy level at a time, but it requires constructing the symplectic reduction and establishing that the result is a smooth manifold. On the other hand, the quantized energy levels can be determined from properties of the quantized system as a whole, as in [@sn2] or [@det]. These approaches are characterized by requiring a polarization, or the equivalent information – recall that the Hamiltonian vector fields corresponding to the commuting functions in a completely integrable system generate a real polarization – and the calculation is not restricted to a single level set of the energy function in question.
In our prior paper [@v1], which builds on work by Robinson [@rob1], we propose an alternative definition of a quantized energy level that acts as a middle ground between these two extremes. While this definition can be applied to Kostant-Souriau quantization, it was developed for use with metaplectic-c quantization, and that is the context in which we will apply it here.
Metaplectic-c quantization, due to Robinson and Rawnsley [@rr1], is a quantization procedure that replaces the prequantization circle bundle and metaplectic structure from the Kostant-Souriau recipe by a single object called a metaplectic-c prequantization. Given a metaplectic-c prequantizable symplectic manifold $(M,\omega)$ and a function $H:M\rightarrow{\ensuremath{\mathbb{R}}}$, our quantized energy condition is evaluated over a regular level set of $H$. It does not require the symplectic reduction, nor does it depend on a polarization. The objective of this paper is to apply the metaplectic-c quantized energy condition to the hydrogen atom, using the physical model that is equivalent to the Kepler problem. We will show that the quantized energy levels are in agreement with the quantum mechanical prediction.
In Section \[sec:mpc\], we review the fundamentals of metaplectic-c prequantization and the quantized energy condition. In Section \[sec:hatom\], we set up our model of the hydrogen atom and construct a metaplectic-c prequantization for its phase space. Section \[sec:ls\] presents the Ligon-Schaaf regularization map, which is a symplectomorphism from the negative-energy domain of the hydrogen atom to an open submanifold of ${\ensuremath{TS^3}}$. We show how to relate the energy function for the hydrogen atom to that of a free particle on ${\ensuremath{TS^3}}$. Finally, in Section \[sec:freeTS\], we determine the quantized energy levels of a free particle on ${\ensuremath{TS^3}}$, and use these to determine the quantized energy levels for the hydrogen atom.
Metaplectic-c Prequantization and the Quantization Condition {#sec:mpc}
============================================================
In this section, we briefly outline the key features of metaplectic-c prequantization and our quantized energy condition. A more detailed overview of similar material appears in [@v1], and proofs of the properties of the quantized energy condition are given there. For the original presentations of metaplectic-c quantization and the constructions of the associated bundles used in the quantized energy condition, see [@rr1] and [@rob1].
Section \[subsec:stdvec\] establishes standard notation and conventions that will be used throughout the rest of the paper. Readers who are familiar with the background should nevertheless refer to this section before proceeding to the calculations.
The Metaplectic-c Group {#subsec:vtspc}
-----------------------
Let $(V,\Omega)$ be a $2n$-dimensional symplectic vector space, and assume that it is equipped with a compatible complex structure $J$. The metaplectic group $\operatorname{Mp}(V)$ is the unique connected double cover of the symplectic group $\operatorname{Sp}(V)$, and the metaplectic-c group $\operatorname{Mp}^c(V)$ is defined to be $$\operatorname{Mp}^c(V)=\operatorname{Mp}(V)\times_{{\ensuremath{\mathbb{Z}}}_2}U(1).$$
By construction, $\operatorname{Mp}^c(V)$ contains $U(1)$ and $\operatorname{Mp}(V)$ as subgroups. The inclusion of each subgroup into $\operatorname{Mp}^c(V)$ yields a short exact sequence and a group homomorphism on $\operatorname{Mp}^c(V)$. One such sequence is $$1\rightarrow U(1)\rightarrow\operatorname{Mp}^c(V){\ensuremath{\stackrel{\sigma}{\longrightarrow}}}\operatorname{Sp}(V)\rightarrow 1,$$ where the group homomorphism $\sigma$ is called the *projection map*, and its restriction to $\operatorname{Mp}(V)\subset\operatorname{Mp}^c(V)$ is the double covering map. The other is $$1\rightarrow\operatorname{Mp}(V)\rightarrow\operatorname{Mp}^c(V){\ensuremath{\stackrel{\eta}{\longrightarrow}}}U(1)\rightarrow 1,$$ where the group homomorphism $\eta$ is called the *determinant map*, and it has the property that for any $\lambda\in U(1)\subset\operatorname{Mp}^c(V)$, $\eta(\lambda)=\lambda^2$. The Lie algebra $\mathfrak{mp}^c(V)$ is identified with $\mathfrak{sp}(V)\oplus\mathfrak{u}(1)$ under the map $\sigma_*\oplus\frac{1}{2}\eta_*$.
For all $g\in\operatorname{Sp}(V)$, define $$C_g=\frac{1}{2}(g-JgJ).$$ Then $C_g$ is a complex automorphism of $V$. Using this definition, we construct an embedding of $\operatorname{Mp}^c(V)$ into $\operatorname{Sp}(V)\times{\ensuremath{\mathbb{C}}}\setminus{\left\{0\right\}}$ in the following way. Given any $a\in\operatorname{Mp}^c(V)$, let $g=\sigma(a)\in\operatorname{Sp}(V)$, and let $\mu\in{\ensuremath{\mathbb{C}}}\setminus{\left\{0\right\}}$ be such that $\eta(a)=\mu^2{\ensuremath{\mbox{Det}}}_{\ensuremath{\mathbb{C}}}C_g$. To remove the ambiguity from the definition of $\mu$, assume that if $a=I$, then $\mu=1$. The desired embedding is given by the map $a\mapsto(g,\mu)$. The properties of this map are established in [@rr1]; following their terminology, we refer to the pair $(g,\mu)$ as the *parameters* of $a$. Note that if $a\in\operatorname{Mp}(V)$, then its parameters $(g,\mu)$ satisfy $\mu^2{\ensuremath{\mbox{Det}}}_{\ensuremath{\mathbb{C}}}C_g=1$.
Fix a subspace $W\subset V$ with codimension $1$. Its symplectic orthogonal $W^\perp$ is a one-dimensional subspace of $W$, and the quotient space $W/W^\perp$ acquires a symplectic structure from the symplectic form $\Omega$. This new symplectic vector space has symplectic group $\operatorname{Sp}(W/W^\perp)$ and metaplectic-c group $\operatorname{Mp}^c(W/W^\perp)$. The complex structure $J$ on $V$ induces a complex structure on $W/W^\perp$ in such a way that $$\label{eq:JW}
W/W^\perp\cong(W^\perp\oplus JW^\perp)^\perp,$$ where the isomorphism respects both the symplectic structures and the complex structures.
Let $\operatorname{Sp}(V;W)\subset\operatorname{Sp}(V)$ be the subgroup whose elements preserve the subspace $W$. Then there is a natural group homomorphism $\operatorname{Sp}(V;W){\ensuremath{\stackrel{\nu}{\longrightarrow}}}\operatorname{Sp}(W/W^\perp)$. Now let $\operatorname{Mp}^c(V;W)=\sigma^{-1}(\operatorname{Sp}(V;W))\subset\operatorname{Mp}^c(V)$. As shown in [@rob1], there is a group homomorphism $\operatorname{Mp}^c(V;W){\ensuremath{\stackrel{\hat{\nu}}{\longrightarrow}}}\operatorname{Mp}^c(W/W^\perp)$ such that the following commutes.
$\xymatrix{
\operatorname{Mp}^c(V)\supset\operatorname{Mp}^c(V;W) \ar[r]^(0.6){\hat{\nu}} \ar[d]^\sigma & \operatorname{Mp}^c(W/W^\perp) \ar[d]^\sigma\\
\operatorname{Sp}(V)\supset\operatorname{Sp}(V;W) \ar[r]^(0.6){\nu} & \operatorname{Sp}(W/W^\perp)
}$
On the level of Lie algebras, $\eta_*\circ\hat{\nu}_*=\eta_*$.
Metaplectic-c Prequantization of Symplectic Manifolds
-----------------------------------------------------
Assume that $(V,\Omega)$ has been fixed as in the previous section. Let $(M,\omega)$ be a $2n$-dimensional symplectic manifold. We view the symplectic frame bundle $\operatorname{Sp}(M,\omega){\ensuremath{\stackrel{\rho}{\longrightarrow}}}M$ as a right principal $\operatorname{Sp}(V)$ bundle whose fibers take the form $$\operatorname{Sp}(M,\omega)_m={\left\{b:V\rightarrow T_mM:b\mbox{ is a symplectic isomorphism}\right\}},\ \ \forall m\in M.$$ The group action is by precomposition.
A *metaplectic-c prequantization* for $(M,\omega)$ is the triple $(P,\Sigma,\gamma)$, where:
(1) $P{\ensuremath{\stackrel{\Pi}{\longrightarrow}}}M$ is a right principle $\operatorname{Mp}^c(V)$ bundle over $M$;
(2) $\Sigma:P\rightarrow\operatorname{Sp}(M,\omega)$ is a map such that $\rho\circ\Sigma=\Pi$ and $\Sigma(q\cdot a)=\Sigma(q)\cdot\sigma(a)$ for all $q\in P$ and $a\in\operatorname{Mp}^c(V)$; and
(3) $\gamma$ is a $\mathfrak{u}(1)$-valued one-form on $P$ such that
(a) $\gamma$ is invariant under the $\operatorname{Mp}^c(V)$ action;
(b) if $\partial_\alpha$ is the vector field on $P$ generated by $\alpha\in\mathfrak{mp}^c(V)$, then $\gamma(\partial_\alpha)=\frac{1}{2}\eta_*\alpha$; and
(c) $d\gamma=\frac{1}{i\hbar}\Pi^*\omega$.
Note that $(P,\gamma)$ is a principal circle bundle with connection one-form over $\operatorname{Sp}(M,\omega)$, with projection map $\Sigma$. We will view $P$ as a bundle over $M$ or a bundle over $\operatorname{Sp}(M,\omega)$ as the circumstance demands.
Two metaplectic-c prequantizations for $(M,\omega)$ are considered equivalent if there is a diffeomorphism between them that preserves the one-forms and commutes with the respective maps to $\operatorname{Sp}(M,\omega)$. If $(M,\omega)$ is metaplectic-c prequantizable, then the set of equivalence classes of metaplectic-c prequantizations for $(M,\omega)$ is in one-to-one correspondence with the locally constant cohomology group $H^1(M,U(1))$. In particular, if $H^1(M,U(1))$ is trivial, then the metaplectic-c prequantization of $(M,\omega)$ is unique up to isomorphism.
Metaplectic-c Quantized Energy Condition {#subsec:mpcquant}
----------------------------------------
Assume that $(M,\omega)$ admits a metaplectic-c prequantization $(P,\Sigma,\gamma)$. Let $H:M\rightarrow{\ensuremath{\mathbb{R}}}$ be a smooth function. We denote its Hamiltonian vector field by $\xi_H$, and adopt the convention that $\xi_H{\ensuremath{\lrcorner\,}}\omega=dH$. Let $\xi_H$ have flow $\phi^t$ on $M$. There is a lift of $\phi^t$ to $\operatorname{Sp}(M,\omega)$ given by $${\ensuremath{\widetilde{\phi}}}^t(b)=\phi^t_*\circ b,\ \ \forall b\in\operatorname{Sp}(M,\omega).$$ Let the corresponding vector field on $\operatorname{Sp}(M,\omega)$ be ${\ensuremath{\widetilde{\xi}}}_H$. Now let ${\ensuremath{\widehat{\xi}}}_H$ be the lift of ${\ensuremath{\widetilde{\xi}}}_H$ to $P$ such that $\gamma({\ensuremath{\widehat{\xi}}}_H)=0$, and let ${\ensuremath{\widehat{\phi}}}^t$ be the flow of ${\ensuremath{\widehat{\xi}}}_H$.
Given a regular value $E$ of $H$, the level set $S=H^{-1}(E)$ is an embedded submanifold of $M$ with codimension $1$. Assume that a model subspace $W\subset V$ of codimension $1$ has been fixed. The groups defined in Section \[subsec:vtspc\] will be the structure groups for the bundles that we now define over $S$.
Let $TS^\perp$ denote the null foliation, which is the bundle over $S$ given by $$T_sS^\perp={\left\{\zeta\in T_sM:\omega(\zeta,T_sS)=0\right\}},\ \ \forall s\in S.$$ At each $s\in S$, $T_sS^\perp$ is a one-dimensional subspace of $T_sS$, spanned by $\xi_H(s)$. The quotient bundle $TS/TS^\perp$ is a symplectic vector bundle over $S$, and we model its symplectic frame bundle $\operatorname{Sp}(TS/TS^\perp)$ on $W/W^\perp$: $$\operatorname{Sp}(TS/TS^\perp)_s={\left\{b':W/W^\perp\rightarrow T_sS/T_sS^\perp:b'\mbox{ is a symplectic isomorphism}\right\}},\ \ \forall s\in S.$$ Let $\operatorname{Sp}(M,\omega;S)\subset\operatorname{Sp}(M,\omega)|_{S}$ be the subbundle defined by $$\operatorname{Sp}(M,\omega;S)_s={\left\{b\in\operatorname{Sp}(M,\omega):bW=T_sS\right\}},\ \ \forall s\in S.$$ This is a principal $\operatorname{Sp}(V;W)$ bundle over $S$. The symplectic frame bundle $\operatorname{Sp}(TS/TS^\perp)$ is naturally identified with the bundle associated to $\operatorname{Sp}(M,\omega;S)$ by the group homomorphism $\operatorname{Sp}(V;W){\ensuremath{\stackrel{\nu}{\longrightarrow}}}\operatorname{Sp}(W/W^\perp)$.
On the level of metaplectic-c bundles, let $P^S$ be the principal $\operatorname{Mp}^c(V;W)$ bundle over $S$ obtained by restricting $P$ to $\operatorname{Sp}(M,\omega;S)$. Let $P_S$ be the bundle associated to $P^S$ by the group homomorphism $\operatorname{Mp}^c(V;W){\ensuremath{\stackrel{\hat{\nu}}{\longrightarrow}}}\operatorname{Mp}^c(W/W^\perp)$, and let the projection map to $\operatorname{Sp}(TS/TS^\perp)$ be $P_S{\ensuremath{\stackrel{\Sigma_S}{\longrightarrow}}}\operatorname{Sp}(TS/TS^\perp)$. The connection one-form $\gamma$ on $P$ pulls back to $\gamma^S$ on $P^S$, and induces a connection one-form $\gamma_S$ on $P_S$. The following diagram shows the various bundles over $S$, together with their structure groups. $$\xymatrix{
(P^S,\gamma^S) \ar[dd]_{\operatorname{Mp}^c(V;W)} \ar[dr]_{U(1)} \ar[drrr]^{\hat{\nu}} \\
& \operatorname{Sp}(M,\omega;S) \ar[dl]^{\operatorname{Sp}(V;W)} \ar[drrr]^(0.4){\nu} & & (P_S,\gamma_S) \ar[dd]_{\operatorname{Mp}^c(W/W^\perp)} \ar[dr]^{U(1)} \\
S \ar[drrr]^{=} & & & & \operatorname{Sp}(TS/TS^\perp) \ar[dl]^{\operatorname{Sp}(W/W^\perp)} \\
& & & S
}$$
It is clear that the flow $\phi^t$ of $\xi_H$ preserves the manifold $S$. Further, the lifted flows ${\ensuremath{\widetilde{\phi}}}^t$ and ${\ensuremath{\widehat{\phi}}}^t$ preserve the manifolds $\operatorname{Sp}(M,\omega;S)$ and $P^S$, respectively, and they induce flows on $\operatorname{Sp}(TS/TS^\perp)$ and $P_S$. We also denote these induced flows by ${\ensuremath{\widetilde{\phi}}}^t$ and ${\ensuremath{\widehat{\phi}}}^t$, and let the corresponding vector fields on $\operatorname{Sp}(TS/TS^\perp)$ and $P_S$ be ${\ensuremath{\widetilde{\xi}}}_H$ and ${\ensuremath{\widehat{\xi}}}_H$.
Having established all of these constructions, we can now state the definition of a quantized energy level for the system $(M,\omega,H)$. This definition first appeared in [@v1], which contains a detailed examination of its properties.
\[def:quantE\] The regular value $E$ of $H$ is a **quantized energy level** of $(M,\omega,H)$ if the connection one-form $\gamma_S$ on $P_S$ has trivial holonomy over all closed orbits of ${\ensuremath{\widetilde{\phi}}}^t$ on $\operatorname{Sp}(TS/TS^\perp)$.
The following property of Definition \[def:quantE\] was proved in [@v1].
Suppose $H_1,H_2:M\rightarrow{\ensuremath{\mathbb{R}}}$ are smooth functions, and suppose there are $E_1,E_2\in{\ensuremath{\mathbb{R}}}$ such that $E_j$ is a regular value of $H_j$ for $j=1,2$ and $H_1^{-1}(E_1)=H_2^{-1}(E_2)$. Then $E_1$ is a quantized energy level for $(M,\omega,H_1)$ if and only if $E_2$ is a quantized energy level for $(M,\omega,H_2)$.
In other words, the quantized energy condition depends on the geometry of the level set $S$ and not on a particular choice of $H$, a property that we have termed dynamical invariance. As a special case of this theorem, suppose $H_1,H_2:M\rightarrow{\ensuremath{\mathbb{R}}}$ are smooth functions such that $H_2=f\circ H_1$ for some diffeomorphism $f:\mbox{range}(H_1)\rightarrow\mbox{range}(H_2)$. Then $E$ is a quantized energy level for $H_1$ if and only if $f(E)$ is a quantized energy level for $H_2$. We will make use of this fact in Section \[subsec:rescale\].
Choices for Subsequent Calculations {#subsec:stdvec}
-----------------------------------
In the sections that follow, we will need model symplectic vector spaces of several different dimensions. Let us fix some standard definitions and notation.
For any $n\in{\ensuremath{\mathbb{N}}}$, let $(V_n,\Omega_n)$ be a $2n$-dimensional symplectic vector space. Let $({\ensuremath{\widehat{v}}}_1,\ldots,{\ensuremath{\widehat{v}}}_n,$ ${\ensuremath{\widehat{w}}}_1,\ldots,{\ensuremath{\widehat{w}}}_n)$ be a symplectic basis for $V_n$, and write all elements of $V_n$ as ordered $2n$-tuples with respect to this basis. The symplectic form can be written in terms of the dual basis as $$\Omega_n=\sum_{j=1}^n{\ensuremath{\widehat{v}^*}}_j\wedge{\ensuremath{\widehat{w}^*}}_j.$$ Assume that each real vector $(a_1,\ldots,a_n,b_1,\ldots,b_n)\in V_n$ is identified with the complex vector $(b_1+ia_1\ldots,b_n+ia_n)\in{\ensuremath{\mathbb{C}}}^n$. The resulting complex structure $J$ on $V_n$ is written in matrix form as $J={\ensuremath{{\left(\begin{array}{cc}{0}&{I}\\{-I}&{0}\end{array}\right)}}}$, where $I$ is the $n\times n$ identity matrix.
When we require a subspace of $V_n$ of codimension $1$, we choose $$W_n=\mbox{span}{\left\{{\ensuremath{\widehat{v}}}_1,\ldots,{\ensuremath{\widehat{v}}}_n,{\ensuremath{\widehat{w}}}_1,\ldots,{\ensuremath{\widehat{w}}}_{n-1}\right\}}.$$ Then $$W_n^\perp=\mbox{span}{\left\{{\ensuremath{\widehat{v}}}_n\right\}}\ \ \mbox{and}\ \ W_n/W_n^\perp=\mbox{span}{\left\{[{\ensuremath{\widehat{v}}}_1],\ldots,[{\ensuremath{\widehat{v}}}_{n-1}],[{\ensuremath{\widehat{w}}}_1],\ldots,[{\ensuremath{\widehat{w}}}_{n-1}]\right\}}.$$ Using equation \[eq:JW\], it is immediate that $W_n/W_n^\perp$ is isomorphic to $V_{n-1}$ as a symplectic vector space and a complex vector space. The commutative diagram from Section \[subsec:vtspc\] containing the group homomorphisms $\nu$ and $\hat{\nu}$ can be rewritten as follows. $$\xymatrix{
\operatorname{Mp}^c(V_n)\supset\operatorname{Mp}^c(V_n;W_n) \ar[r]^(0.65){\hat{\nu}} \ar[d]^\sigma & \operatorname{Mp}^c(V_{n-1}) \ar[d]^{\sigma} \\
\operatorname{Sp}(V_n)\supset\operatorname{Sp}(V_n;W_n) \ar[r]^(0.65){\nu} \ar[r] & \operatorname{Sp}(V_{n-1})
}$$
The Hydrogen Atom {#sec:hatom}
=================
Setup {#subsec:setupH}
-----
Let $\dot{{\ensuremath{\mathbb{R}}}}^3$ represent ${\ensuremath{\mathbb{R}}}^3$ with the origin removed, and let $M=T\dot{{\ensuremath{\mathbb{R}}}}^3=\dot{{\ensuremath{\mathbb{R}}}}^3\times{\ensuremath{\mathbb{R}}}^3$. We use Cartesian coordinates $q=(q_1,q_2,q_3)$ on $\dot{{\ensuremath{\mathbb{R}}}}^3$ and $p=(p_1,p_2,p_3)$ on ${\ensuremath{\mathbb{R}}}^3$. Equip $M$ with the symplectic form $\omega=\sum_{j=1}^3dq_j\wedge dp_j$.
We consider the model of the hydrogen atom that is equivalent to the Kepler problem. Assume that a proton is fixed at the origin in ${\ensuremath{\mathbb{R}}}^3$, and that an electron of mass $m_e$ interacts with it via the electrostatic force, which obeys an inverse-square law with constant of proportionality $k$. Then $M$ is the phase space for the motion of the electron, where $q$ and $p$ represent its position and momentum, respectively. The Hamiltonian energy function is $$H=\frac{1}{2m_e}|p|^2-\frac{k}{|q|},$$ and the corresponding Hamiltonian vector field is $$\xi_H=\sum_{j=1}^3{\left(\frac{1}{m_e}p_j{\frac{\partial{}}{\partial{q_j}}}-\frac{k}{|q|^3}q_j{\frac{\partial{}}{\partial{p_j}}}\right)}.$$ Let $\xi_H$ have flow $\phi^t$ on $M$.
The solutions to the Kepler problem are well known [@a1; @cb]. The angular momentum vector $L=q\times p$ is a constant of the motion, as is the eccentricity vector $e=\frac{1}{m_ek}p\times L-\frac{q}{|q|}$. In position space, the orbits corresponding to a given energy $E\in{\ensuremath{\mathbb{R}}}$ are conic sections with eccentricity $$\label{eq:ecc}
|e|=\sqrt{1+\frac{2E|L|^2}{m_ek^2}},$$ having the origin as a focus.
In particular, suppose $E<0$. Then the value of $|L|$ lies in the interval ${\left[0,\sqrt{-\frac{m_ek^2}{2E}}\right]}$. If $|L|>0$, then the orbit is an ellipse, with $|L|=\sqrt{-\frac{m_ek^2}{2E}}$ being the special case of a circle. All elliptical orbits with energy $E$ have period $\frac{2\pi}{\Lambda}$, where $\Lambda=\sqrt{-\frac{8E^3}{m_ek^2}}$. If $|L|=0$, however, then $e=1$ and the orbit is a line segment. Physically, this represents the case where the electron begins from rest and collapses on a straight-line trajectory into the proton. Such motion is not periodic, which implies that the level set $H^{-1}(E)$ contains orbits of $\xi_H$ that are not closed. Further, the collapse occurs in finite time, meaning that the vector field $\xi_H$ is not complete.
The objective of this paper is to determine the quantized energy levels for $(M,\omega,H)$. In the next section, we construct a metaplectic-c prequantization for $(M,\omega)$, and formulate our approach for performing the quantized energy calculation.
Metaplectic-c Prequantization of $(M,\omega)$ {#subsec:mpcP}
---------------------------------------------
We choose the model symplectic vector space $V_3$, as described in Section \[subsec:stdvec\]. The tangent bundle $TM$ can be identified with $M\times V_3$ with respect to the global trivialization $${\ensuremath{\widehat{v}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{q_j}}}\right|_{m}},\ \ {\ensuremath{\widehat{w}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{p_j}}}\right|_{m}},\ \ \forall m\in M,\ j=1,2,3.$$ This yields an identification of the symplectic frame bundle $\operatorname{Sp}(M,\omega)$ with $M\times\operatorname{Sp}(V_3)$.
Let $P=M\times\operatorname{Mp}^c(V_3)$, with bundle projection map $P{\ensuremath{\stackrel{\Pi}{\longrightarrow}}}M$. Define the map $\Sigma:P\rightarrow\operatorname{Sp}(M,\omega)$ by $$\Sigma(m,a)=(m,\sigma(a)),\ \ \forall m\in M,\forall a\in\operatorname{Mp}^c(V_3),$$ where the right-hand side is written with respect to the global trivialization above. Let $$\label{eq:beta}
\beta=\sum_{j=1}^3{\left(q_jdp_j+d(q_jp_j)\right)}$$ on $M$, so that $d\beta=\omega$. The reason for this choice of $\beta$ will be made clear in Section \[subsec:lsts\]. Let $\gamma$ be the $\mathfrak{u}(1)$-valued one-form on $P$ given by $$\gamma=\frac{1}{i\hbar}\Pi^*\beta+\frac{1}{2}\eta_*\vartheta_0,$$ where $\vartheta_0$ is the trivial connection on the product bundle. Then $(P,\Sigma,\gamma)$ is a metaplectic-c prequantization for $(M,\omega)$, and it is unique up to isomorphism.
The quantized energy levels of $(M,\omega,H)$ are those regular values $E$ of $H$ such that the holonomy of $\gamma_S$ is trivial over all closed orbits of ${\ensuremath{\widetilde{\xi}}}_H$ on $\operatorname{Sp}(TS/TS^\perp)$, where $S=H^{-1}(E)$. If $E\geq 0$, then the quantization condition can be evaluated immediately. From equation \[eq:ecc\] for the eccentricity of the orbit, we see that if $E=0$, then the orbits are parabolas in position space, and if $E>0$, then they are hyperbolas. In these cases, $\xi_H$ has no closed orbits in $S$, which implies that ${\ensuremath{\widetilde{\xi}}}_H$ cannot have any closed orbits in $\operatorname{Sp}(TS/TS^\perp)$. Therefore the holonomy condition is satisfied vacuously, and all nonnegative energy levels are quantized energy levels. This is consistent with the physical prediction from quantum mechanics: a particle that is not spatially confined has a continuous energy spectrum.
It remains to consider the orbits corresponding to negative energy. Let $$N={\left\{m\in M:H(m)<0\right\}}.$$ Then $N$ is an open, simply connected submanifold of $M$. By restriction, the symplectic form on $M$ induces one on $N$, and the metaplectic-c prequantization for $M$ induces one for $N$. We use the same symbols to denote the restricted objects: $(N,\omega)$ is a symplectic manifold, and $(P,\Sigma,\gamma)$ is its metaplectic-c prequantization. Since $N$ is simply connected, $(P,\Sigma,\gamma)$ is unique up to isomorphism.
Let $E<0$ be fixed, and let $S=H^{-1}(E)\subset N$. Through the process described in Section \[subsec:mpcquant\], we obtain the three-level structures $$(P^S,\gamma^S)\rightarrow\operatorname{Sp}(N,\omega;S)\rightarrow S$$ and $$(P_S,\gamma_S)\rightarrow\operatorname{Sp}(TS/TS^\perp)\rightarrow S.$$ Lift $\xi_H$ to ${\ensuremath{\widetilde{\xi}}}_H$ on $\operatorname{Sp}(N,\omega;S)$, and let it induce a vector field on $\operatorname{Sp}(TS/TS^\perp)$.
To evaluate the quantization condition using these bundles, we would have to determine the closed orbits of ${\ensuremath{\widetilde{\xi}}}_H$ on $\operatorname{Sp}(TS/TS^\perp)$, then lift ${\ensuremath{\widetilde{\xi}}}_H$ to ${\ensuremath{\widehat{\xi}}}_H$ on $P_S$, horizontally with respect to $\gamma_S$, and ensure that every lift of a closed orbit in $\operatorname{Sp}(TS/TS^\perp)$ is closed in $P_S$. However, this procedure is computationally prohibitive in all but the special case of the circular orbit. Instead, we will use Ligon-Schaaf regularization to transform the quantized energy calculation on $(N,\omega)$ into one on an open submanifold of ${\ensuremath{TS^3}}$. This is the subject of Section \[sec:ls\].
Ligon-Schaaf Regularization {#sec:ls}
===========================
Let ${\ensuremath{T^+S^3}}$ represent the result of removing the zero section from ${\ensuremath{TS^3}}$. In Section \[subsec:setupH\], we noted that the vector field $\xi_H$ is not complete. The Ligon-Schaaf map is a symplectomorphism from $(N,\omega)$ to an open submanifold of ${\ensuremath{T^+S^3}}$, having the property that $\xi_H$ is mapped to a vector field that is complete on ${\ensuremath{T^+S^3}}$. This map was presented in [@ls], and an in-depth discussion of it can be found in [@cb].
In this section, we will state the Ligon-Schaaf map and list its relevant properties. Then we will construct a metaplectic-c prequantization for ${\ensuremath{TS^3}}$, and show how to lift the Ligon-Schaaf map to the level of symplectic frame bundles, and of metaplectic-c prequantizations. Using the lifted maps, we will be able to relate the quantized energy levels of the hydrogen atom to those of a free particle on ${\ensuremath{TS^3}}$.
Our conventions and notation largely follow those in [@cb]. We state results without proof; much more detail can be found in [@cb] and works cited.
${\ensuremath{TS^3}}$ and the Ligon-Schaaf Map {#subsec:lsts}
----------------------------------------------
Consider $T{\ensuremath{\mathbb{R}}}^4={\ensuremath{\mathbb{R}}}^4\times{\ensuremath{\mathbb{R}}}^4$ with Cartesian coordinates $(x,y)$, where $x=(x_1,x_2,x_3,x_4)$ and $y=(y_1,y_2,y_3,y_4)$. Let $T{\ensuremath{\mathbb{R}}}^4$ have symplectic structure $$\omega_4=\sum_{j=1}^4 dx_j\wedge dy_j,$$ and let $$\label{eq:beta4}
\beta_4=-\sum_{j=1}^4y_jdx_j$$ on $T{\ensuremath{\mathbb{R}}}^4$, so that $d\beta_4=\omega_4$. The submanifold ${\ensuremath{TS^3}}$ is given by $${\ensuremath{TS^3}}={\left\{(x,y)\in T{\ensuremath{\mathbb{R}}}^4:|x|^2=1,x\cdot y=0\right\}}\subset T{\ensuremath{\mathbb{R}}}^4,$$ where we take the usual Euclidean inner product on ${\ensuremath{\mathbb{R}}}^4$. From now on, we abbreviate $T{\ensuremath{\mathbb{R}}}^4$ by $Z$.
By treating $({\ensuremath{TS^3}},\omega_3)$ as a level set of the constraint functions $c_1(x,y)=\frac{1}{2}(|x|^2-1)$ and $c_2(x,y)=x\cdot y$ on $(Z,\omega_4)$, it can be shown that the restriction of $\omega_4$ to ${\ensuremath{TS^3}}$ yields a symplectic form on ${\ensuremath{TS^3}}$. Let $({\ensuremath{TS^3}},\omega_3)$ be the resulting symplectic manifold. Let $\beta_3$ be the restriction of $\beta_4$ to ${\ensuremath{TS^3}}$, so that $d\beta_3=\omega_3$.
Let ${\ensuremath{T^+S^3}}$ represent ${\ensuremath{TS^3}}$ with the zero section removed: $${\ensuremath{T^+S^3}}={\left\{(x,y)\in{\ensuremath{TS^3}}:|y|>0\right\}}.$$ Define the map $D:{\ensuremath{T^+S^3}}\rightarrow{\ensuremath{\mathbb{R}}}$ by $$D(x,y)=-\frac{m_ek^2}{2|y|^2},\ \ \forall (x,y)\in{\ensuremath{T^+S^3}}.$$ As shown in [@cb], the Hamiltonian vector field for $D$ on ${\ensuremath{T^+S^3}}$ is $$\xi_D=\sum_{j=1}^4{\left(\frac{m_ek^2}{|y|^4}y_j{\frac{\partial{}}{\partial{x_j}}}-\frac{m_ek^2}{|y|^2}x_j{\frac{\partial{}}{\partial{y_j}}}\right)},$$ and its flow is $$\psi^t_D(x,y)={\ensuremath{{\left(\begin{array}{cc}{\cos\frac{m_ek^2}{|y|^3}t}&{\frac{1}{|y|}\sin\frac{m_ek^2}{|y|^3}t}\\{-|y|\sin\frac{m_ek^2}{|y|^3}t}&{\cos\frac{m_ek^2}{|y|^3}t}\end{array}\right)}}}{\ensuremath{{\left(\begin{array}{c}{x}\\{y}\end{array}\right)}}}.$$ The map $D$ is called the Delaunay Hamiltonian, and $\xi_D$ is the Delaunay vector field. Since the orbits of $\xi_D$ must preserve $|y|$, it is clear from the form of $\psi^t_D$ that $\xi_D$ is complete on ${\ensuremath{T^+S^3}}$, and every orbit is closed.
Now let $S^3_{n}$ represent $S^3$ with the north pole $(0,0,0,1)$ removed. The Ligon-Schaaf map $LS:N\rightarrow{\ensuremath{T^+S^3_{n}}}$ is given by $$LS(q,p)=(A\sin\varphi+B\cos\varphi,-\nu A\cos\varphi+\nu B\sin\varphi),$$ where $$\begin{array}{ll}
\displaystyle\nu=\sqrt{-\frac{m_ek^2}{2H(q,p)}},&\displaystyle\varphi=\frac{1}{\nu}(q\cdot p),\\
\displaystyle A={\left(\frac{q}{|q|}-\frac{1}{m_ek}(q\cdot p)p,\frac{1}{\nu}(q\cdot p)\right)},&\displaystyle B={\left(\frac{1}{\nu}|q|p,\frac{1}{m_ek}|p|^2|q|-1\right)}.\\
\end{array}$$ This map has the following properties:
(1) $LS$ is a diffeomorphism between $N$ and ${\ensuremath{T^+S^3_{n}}}$;
(2) $LS^*\beta_3=\sum_{j=1}^3{\left(q_jdp_j+d(q_jp_j)\right)}=\beta$ (recall equation \[eq:beta\]);
(3) $H=D\circ LS$.
Assertions (1) and (2) imply that $LS$ is a symplectomorphism; from that and (3), it follows that $LS_*\xi_H=\xi_D$.
Metaplectic-c Prequantization for ${\ensuremath{TS^3}}$ {#subsec:mpcTS}
-------------------------------------------------------
We begin by constructing a metaplectic-c prequantization for $(Z=T{\ensuremath{\mathbb{R}}}^4,\omega_4)$, and let it induce a metaplectic-c prequantization for $({\ensuremath{TS^3}},\omega_3)$. For $(Z,\omega_4)$, we proceed in precisely the same way as we constructed $(P,\Sigma,\gamma)$ for $(M,\omega)$ in Section \[subsec:mpcP\]. Choose the model vector space $V_4$, and identify the symplectic frame bundle $\operatorname{Sp}(Z,\omega_4)$ with $Z\times\operatorname{Sp}(V_4)$ using the global trivialization for the tangent bundle given by $${\ensuremath{\widehat{v}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{x_j}}}\right|_{(x,y)}},\ {\ensuremath{\widehat{w}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{q_j}}}\right|_{(x,y)}},\ \ \forall (x,y)\in Z,\ j=1,\ldots,4.$$ Let $Q_4=Z\times\operatorname{Mp}^c(V_4)$ with bundle projection map $Q_4{\ensuremath{\stackrel{\Pi_4}{\longrightarrow}}} Z$. Define the map $\Gamma_4:Q_4\rightarrow\operatorname{Sp}(T{\ensuremath{\mathbb{R}}}^4,\omega_4)$ by $$\Gamma_4(x,y,a)=(x,y,\sigma(a)),\ \ \forall (x,y)\in Z,\forall a\in\operatorname{Mp}^c(V_4).$$ Lastly, define the $\mathfrak{u}(1)$-valued one-form $\delta_4$ on $Q_4$ by $$\delta_4=\frac{1}{i\hbar}\Pi_4^*\beta_4+\frac{1}{2}\eta_*\vartheta_0,$$ where $\vartheta_0$ is the trivial connection on $Q_4$, and where $\beta_4$ was defined in equation \[eq:beta4\]. Then $(Q_4,\Gamma_4,\delta_4)$ is the unique metaplectic-c prequantization for $(Z,\omega_4)$ up to isomorphism.
In order to construct the metaplectic-c prequantization of $({\ensuremath{TS^3}},\omega_3)$ induced by $(Q_4,\Gamma_4,\delta_4)$, we proceed by the process of symplectic reduction. Let $R:Z\rightarrow{\ensuremath{\mathbb{R}}}$ be given by $$R(x,y)=\frac{1}{2}|x|^2,\ \ \forall(x,y)\in Z.$$ The Hamiltonian vector field for $R$ on $Z$ is $$\xi_R=-\sum_{j=1}^4 x_j{\frac{\partial{}}{\partial{y_j}}},$$ and its flow on $Z$ is $$\psi^t_R(x,y)=(x,y-tx),\ \ \forall (x,y)\in Z.$$ It is straightforward to show that ${\ensuremath{TS^3}}$ can be identified with the space of orbits of $\xi_R$ on the level set $A=R^{-1}(\frac{1}{2})$. The orbit projection map $p:A\rightarrow{\ensuremath{TS^3}}$ is $$p(x,y)=(x,y-(x\cdot y)x),\ \ \forall(x,y)\in A.$$ Let $i:A\rightarrow Z$ be the inclusion map. Recall that $\omega_3$ is the symplectic form on ${\ensuremath{TS^3}}$ obtained by restricting $\omega_4$. A calculation shows that $p^*\omega_3=i^*\omega_4$. Therefore $({\ensuremath{TS^3}},\omega_3)$ is the symplectic reduction of $(Z,\omega_4)$ at the level set corresponding to $R=\frac{1}{2}$.
Choose the model subspace $W_4$ as described in Section \[subsec:stdvec\], and construct the three-level structure $$(Q_{4A},\delta_{4A}){\ensuremath{\stackrel{\Gamma_{4A}}{\longrightarrow}}}\operatorname{Sp}(TA/TA^\perp)\rightarrow A.$$ Then $\operatorname{Sp}(TA/TA^\perp)$ is a principal $\operatorname{Sp}(V_3)$ bundle over $A$, and $Q_{4A}$ is a principal $\operatorname{Mp}^c(V_3)$ bundle over $A$. Let the symplectic frame bundle $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ be modeled on $V_3$, so that $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ is a principal $\operatorname{Sp}(V_3)$ bundle over ${\ensuremath{TS^3}}$.
For each $z\in A$, $p_*|_z$ appears in the short exact sequence $$0\rightarrow T_zA^\perp\rightarrow T_zA{\ensuremath{\stackrel{p_*|_z}{\longrightarrow}}}T_{p(z)}{\ensuremath{TS^3}}\rightarrow 0.$$ Note that the projection map $p$ is the identity on ${\ensuremath{TS^3}}\subset A$. It follows that for all $z\in{\ensuremath{TS^3}}$, $p_*|_z:T_zA/T_zA^\perp\rightarrow T_{z}{\ensuremath{TS^3}}$ is a symplectic isomorphism. Therefore $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ and $\operatorname{Sp}(TA/TA^\perp)|_{{\ensuremath{TS^3}}}$ are isomorphic as principal $\operatorname{Sp}(V_3)$ bundles over ${\ensuremath{TS^3}}$. Using that isomorphism, we view $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ as a subbundle of $\operatorname{Sp}(TA/TA^\perp)$.
Let $(Q,\Gamma,\delta)$ be the result of restricting $(Q_{4A},\Gamma_{4A},\delta_{4A})$ to $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)\subset\operatorname{Sp}(TA/TA^\perp)$. Then $(Q,\Gamma,\delta)$ is a metaplectic-c prequantization for $({\ensuremath{TS^3}},\omega_3)$, and it is unique up to isomorphism. We will also use the notation $(Q,\Gamma,\delta)$ to denote the metaplectic-c prequantizations for ${\ensuremath{T^+S^3}}$ and ${\ensuremath{T^+S^3_{n}}}$ obtained by restriction.
Lifting the Ligon-Schaaf Map
----------------------------
Recall that the Ligon-Schaaf map $LS:N\rightarrow{\ensuremath{T^+S^3_{n}}}$ is a symplectomorphism. Define the map $\widetilde{LS}:\operatorname{Sp}(N,\omega)\rightarrow\operatorname{Sp}({\ensuremath{T^+S^3_{n}}},\omega_3)$ by $$\widetilde{LS}(b)=LS_*\circ b,\ \ \forall b\in\operatorname{Sp}(N,\omega),$$ and observe that this is an isomorphism of principal $\operatorname{Sp}(V_3)$ bundles.
In Section \[subsec:mpcP\], we used the global trivialization $${\ensuremath{\widehat{v}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{q_j}}}\right|_{m}},\ \ {\ensuremath{\widehat{w}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{p_j}}}\right|_{m}},\ \ \forall m\in N,\ j=1,2,3,$$ to identify $\operatorname{Sp}(N,\omega)$ with $N\times\operatorname{Sp}(V_3)$. From the map $\widetilde{LS}$, we see that $\operatorname{Sp}({\ensuremath{T^+S^3_{n}}},\omega_3)$ can be identified with ${\ensuremath{T^+S^3_{n}}}\times\operatorname{Sp}(V_3)$ with respect to the global trivialization $${\ensuremath{\widehat{v}}}_j\mapsto{\left.LS_*{\frac{\partial{}}{\partial{q_j}}}\right|_{LS(m)}},\ \ {\ensuremath{\widehat{w}}}_j\mapsto{\left.LS_*{\frac{\partial{}}{\partial{q_j}}}\right|_{LS(m)}},\ \ \forall LS(m)\in{\ensuremath{T^+S^3_{n}}},\ j=1,2,3.$$ In terms of these trivializations for $\operatorname{Sp}(N,\omega)$ and $\operatorname{Sp}({\ensuremath{T^+S^3_{n}}},\omega_3)$, $\widetilde{LS}$ is given simply by $$\label{eq:lstilde}
\widetilde{LS}(m,g)=(LS(m),g),\ \ \forall m\in N,\forall g\in\operatorname{Sp}(V_3).$$
As we have seen before, once we have a global trivialization for the symplectic frame bundle, it is straightforward to construct a metaplectic-c prequantization. Let $Q'={\ensuremath{T^+S^3_{n}}}\times\operatorname{Mp}^c(V_3)$ with bundle projection map $Q'{\ensuremath{\stackrel{\Pi'}{\longrightarrow}}}{\ensuremath{T^+S^3_{n}}}$. Define the map $\Gamma':{\ensuremath{T^+S^3_{n}}}\rightarrow\operatorname{Sp}({\ensuremath{T^+S^3_{n}}},\omega_3)$ by $$\Gamma'(z,a)=(z,\sigma(a)),\ \ \forall z\in{\ensuremath{T^+S^3_{n}}},\forall a\in\operatorname{Mp}^c(V_3).$$ Let $$\delta'=\frac{1}{i\hbar}\Pi'^*\beta_3+\frac{1}{2}\eta_*\vartheta_0,$$ where $\beta_3$ is the restriction to ${\ensuremath{T^+S^3_{n}}}$ of the one-form defined in equation \[eq:beta4\], and where $\vartheta_0$ is the trivial connection on $Q'$. Then $(Q',\Gamma',\delta')$ is a metaplectic-c prequantization for $({\ensuremath{T^+S^3_{n}}},\omega_3)$. Since ${\ensuremath{T^+S^3_{n}}}$ is simply connected, $(Q',\Gamma',\delta')$ must be isomorphic to $(Q,\Gamma,\delta)$.
Recall that $P=M\times\operatorname{Mp}^c(V_3)$. Define $\widehat{LS}:P\rightarrow Q'$ by $$\widehat{LS}(m,a)=(LS(m),a).$$ This is clearly an isomorphism of principal $\operatorname{Mp}^c(V_3)$ bundles. Since $LS^*\beta_3=\beta$, we have $LS^*\delta'=\gamma$. Lastly, it follows from equation \[eq:lstilde\] that $\Gamma'\circ\widehat{LS}=\widetilde{LS}\circ\Sigma$. Therefore $\widehat{LS}$ is an isomorphism of metaplectic-c prequantizations.
All of the preceding observations combine to yield the following commutative diagram. $$\xymatrix{(P,\gamma) \ar[d]^{\Sigma} \ar[r]^(0.45){\widehat{LS}} & (Q',\delta') \ar[d]^{\Gamma'}\\
\operatorname{Sp}(N,\omega) \ar[d] \ar[r]^(0.45){\widetilde{LS}} & \operatorname{Sp}({\ensuremath{T^+S^3_{n}}},\omega_3) \ar[d]\\
(N,\omega) \ar[r]^(0.45){LS}& ({\ensuremath{T^+S^3_{n}}},\omega_3)
}$$ Each of the maps $LS$, $\widetilde{LS}$, and $\widehat{LS}$ is an isomorphism. From these isomorphisms and the fact that $D\circ LS=H$, it follows that $E<0$ is a quantized energy level of $(N,\omega,H)$ if and only if it is a quantized energy level of $({\ensuremath{T^+S^3_{n}}},\omega_3,D)$.
In fact, we claim that $E<0$ is a quantized energy level of $(N,\omega,H)$ if and only if it is a quantized energy level of $({\ensuremath{T^+S^3}},\omega_3,D)$. When we replace the north pole in $S^3$, we acquire more closed orbits: namely, those with $x$-components that pass through $(1,0,0,0)$. However, due to the rotational symmetry of the system $({\ensuremath{T^+S^3}},\omega_3,D)$, an orbit that passes through $(1,0,0,0)$ can always be transformed into one that does not, without altering the holonomy condition. Therefore the quantized energy levels of $({\ensuremath{T^+S^3_{n}}},\omega_3,D)$ and $({\ensuremath{T^+S^3}},\omega_3,D)$ are identical.
Rescaling the Delaunay Hamiltonian {#subsec:rescale}
----------------------------------
So far, we have shown that the negative quantized energy levels of the hydrogen atom are the same as the quantized energy levels of the Delaunay Hamiltonian on ${\ensuremath{T^+S^3}}$. In this section, we will make one final transformation that relates these energies to the quantized energy levels of a free particle on ${\ensuremath{TS^3}}$.
Let $K:TS^3\rightarrow{\ensuremath{\mathbb{R}}}$ be given by $$K(x,y)=\frac{1}{2}|y|^2,\ \ \forall(x,y)\in{\ensuremath{TS^3}}.$$ In the context of classical mechanics, this energy function describes a free particle on ${\ensuremath{TS^3}}$. It is shown in [@cb] that the corresponding Hamiltonian vector field on ${\ensuremath{TS^3}}$ is $$\xi_K=\sum_{j=1}^4{\left(y_j{\frac{\partial{}}{\partial{x_j}}}-|y|^2x_j{\frac{\partial{}}{\partial{y_j}}}\right)},$$ and the flow of this vector field is $$\label{eq:flowK}
\psi^t_K(x,y)={\ensuremath{{\left(\begin{array}{cc}{\cos|y|t}&{\frac{1}{|y|}\sin|y|t}\\{-|y|\sin|y|t}&{\cos|y|t}\end{array}\right)}}}{\ensuremath{{\left(\begin{array}{c}{x}\\{y}\end{array}\right)}}},\ \ \forall(x,y)\in{\ensuremath{TS^3}}\ \mbox{such that}\ |y|>0.$$ If $|y|=0$, then the particle is stationary, and the flow is simply $\psi^t_K(x,0)=(x,0)$.
On the submanifold ${\ensuremath{T^+S^3}}$, the range of $K$ is ${\ensuremath{\mathbb{R}}}_{>0}$, while the range of the Delaunay Hamiltonian $D$ is ${\ensuremath{\mathbb{R}}}_{<0}$. Let $f:{\ensuremath{\mathbb{R}}}_{>0}\rightarrow{\ensuremath{\mathbb{R}}}_{<0}$ be given by $$\label{eq:diffeo}
f(z)=-\frac{m_ek^2}{4z},\ \ \forall z\in{\ensuremath{\mathbb{R}}}_{>0}.$$ Note that $f$ is a diffeomorphism, and $D=f\circ K$ on ${\ensuremath{T^+S^3}}$. Using the dynamical invariance property of quantized energy levels that was stated in Section \[subsec:mpcquant\], we see that $E$ is a quantized energy level of $({\ensuremath{T^+S^3}},\omega_3,D)$ if and only if $f^{-1}(E)$ is a quantized energy level of $({\ensuremath{T^+S^3}},\omega_3,K)$. Thus the quantized energy levels of $(N,\omega,H)$ are exactly the images of the positive quantized energy levels of $({\ensuremath{TS^3}},\omega_3,K)$ under the diffeomorphism $f$.
Quantization of a Free Particle on ${\ensuremath{TS^3}}$ {#sec:freeTS}
========================================================
Orbits of $\xi_K$ and Local Coordinates for ${\ensuremath{TS^3}}$
-----------------------------------------------------------------
Let ${\ensuremath{\mathcal{E}}}>0$ be arbitrary, and let ${\ensuremath{\mathcal{S}}}=K^{-1}({\ensuremath{\mathcal{E}}})\subset{\ensuremath{TS^3}}$. From the flow of $\xi_K$ in equation \[eq:flowK\], it is apparent that all orbits of $\xi_K$ in ${\ensuremath{\mathcal{S}}}$ are closed, with period $\frac{2\pi}{\sqrt{2{\ensuremath{\mathcal{E}}}}}$. Let $z_0=(x_0,y_0)\in{\ensuremath{\mathcal{S}}}$ be an arbitrary initial point, and let ${\ensuremath{\mathcal{C}}}\subset{\ensuremath{\mathcal{S}}}$ be the the orbit of $\xi_K$ through $z_0$: $${\ensuremath{\mathcal{C}}}={\left\{\psi^t_K(z_0):t\in{\ensuremath{\mathbb{R}}}\right\}}\subset{\ensuremath{TS^3}},$$
We can use the rotational symmetry of the system $({\ensuremath{TS^3}},\omega_3,K)$ to make some simplifying assumptions about ${\ensuremath{\mathcal{C}}}$. If we view $x_0$ and $y_0$ as two perpendicular vectors in ${\ensuremath{\mathbb{R}}}^4$, then there is some rotation about the origin in ${\ensuremath{\mathbb{R}}}^4$ that carries them both to the $x_3x_4$-plane. By performing this rotation in $x$-space and $y$-space, we can assume without loss of generality that $x_0$ and $y_0$ take the form $x_0=(0,0,x_{30},x_{40})$ and $y_0=(0,0,y_{30},y_{40})$, where $|x|^2=x_{30}^2+x_{40}^2=1$, $x\cdot y=x_{30}y_{30}+x_{40}y_{40}=0$, and $|y|^2=y_{30}^2+y_{40}^2=2k$. The orbit ${\ensuremath{\mathcal{C}}}$ then lies in $x_3x_4y_3y_4$-space.
Thus far, we have treated ${\ensuremath{TS^3}}$ as a submanifold of $Z=T{\ensuremath{\mathbb{R}}}^4$, using the coordinates $(x,y)$ on $Z$ to describe points in ${\ensuremath{TS^3}}$. Now we make a local change of coordinates on $Z$ that will yield symplectic coordinates for ${\ensuremath{TS^3}}$ on a neighborhood that contains ${\ensuremath{\mathcal{C}}}$. Specifically, we introduce 4-dimensional spherical coordinates and their conjugate momenta. Let $U\subset Z$ be the open set $$U={\left\{(x,y)\in T{\ensuremath{\mathbb{R}}}^4:x_3^2+x_4^2>0\right\}}.$$ On $U$, let the new spatial coordinates be $(a,b,c,r)$, where $$\begin{aligned}
a=&\arctan\frac{\sqrt{x_2^2+x_3^2+x_4^2}}{x_1},\\
b=&\arctan\frac{\sqrt{x_3^2+x_4^2}}{x_2},\\
c=&\arctan\frac{x_4}{x_3},\\
r=&\sqrt{x_1^2+x_2^2+x_3^2+x_4^2}.\end{aligned}$$ The angles $a$ and $b$ are defined modulo $\pi$, and the angle $c$ is defined modulo $2\pi$. Let $\rho_1=\sqrt{x_2^2+x_3^2+x_4^2}$ and $\rho_2=\sqrt{x_3^2+x_4^2}$. The conjugate momenta corresponding to the spherical coordinates are $$\begin{aligned}
p_a=&\frac{x_1}{\rho_1}(x_2y_2+x_3y_3+x_4y_4)-\rho_1y_1,\\
p_b=&\frac{x_2}{\rho_2}(x_3y_3+x_4y_4)-\rho_2y_2,\\
p_c=&x_3y_4-x_4y_3,\\
p_r=&\frac{1}{r}(x_1y_1+x_2y_2+x_3y_3+x_4y_4).\end{aligned}$$ Later, we will need the inverse transformations, which are $$\label{eq:convert1}
\begin{array}{l}
x_1=r\cos a,\\
x_2=r\sin a\cos b,\\
x_3=r\sin a\sin b\cos c,\\
x_4=r\sin a\sin b\sin c,\\
\end{array}$$
$$\label{eq:convert2}
\begin{array}{l}
\displaystyle y_1=p_r\cos a-\frac{p_a}{r}\sin a,\\
\displaystyle y_2=p_r\sin a\cos b+\frac{p_a}{r}\cos a\cos b-\frac{p_b}{r\sin a}\sin b\\
\displaystyle y_3=p_r\sin a\sin b\cos c+\frac{p_a}{r}\cos a\sin b\cos c+\frac{p_b}{r\sin a}\cos b\cos c-\frac{p_c}{r\sin a\sin b}\sin c,\\
\displaystyle y_4=p_r\sin a\sin b\sin c+\frac{p_a}{r}\cos a\sin b\cos c+\frac{p_b}{r\sin a}\cos b\sin c+\frac{p_c}{r\sin a\sin b}\cos c.
\end{array}$$
For convenience, we let $a_j$, $j=1,\ldots,4$, range over $a,b,c,r$.
On $U$, one can verify that $$\label{eq:beta4s}
\beta_4=-\sum_{j=1}^4y_jdx_j=-\sum_{j=1}^4p_{a_j}da_j,$$ and so $$\omega_4=\sum_{j=1}^4dx_j\wedge dy_j=\sum_{j=1}^4 da_j\wedge dp_j.$$ The submanifold ${\ensuremath{TS^3}}$ is characterized by the constant values $r=1$ and $p_r=0$, which implies that the restrictions to ${\ensuremath{TS^3}}\cap U$ of $\beta_4$ and $\omega_4$ are $$\beta_3=-\sum_{j=1}^3 p_{a_j}da_j,\ \ \omega_3=\sum_{j=1}^3da_j\wedge dp_{a_j}.$$ Thus $(a,b,c,p_a,p_b,p_c)$ are symplectic coordinates for ${\ensuremath{TS^3}}\cap U$. On this neighborhood, the map $K$ takes the form $$K=\frac{1}{2}{\left(p_a^2+\frac{p_b^2}{\sin^2a}+\frac{p_c^2}{\sin^2a\sin^2b}\right)}.$$
Several times now, once we had a set of symplectic coordinates such as $(a,b,c,r,p_a,p_b,p_c,p_r)$, we used the trivialization of the symplectic frame bundle given by the coordinate vector fields to construct a metaplectic-c prequantization. We could apply this same procedure to $U$; however, since $U$ is not simply connected, it is not necessarily the case that the metaplectic-c prequantization so constructed would be isomorphic to the result of restricting $(Q_4,\Gamma_4,\delta_4)$ to $U$. Instead, we must show how the local change of variables from Cartesian to spherical coordinates can be lifted from $Z$ to $Q_4$. This is the subject of the next section.
Change of variables over ${\ensuremath{\mathcal{C}}}$
-----------------------------------------------------
Recall that the integral curve of $\xi_K$ through the initial point $z_0=(x_0,y_0)$ is $$\psi^t_K(x_0,y_0)={\ensuremath{{\left(\begin{array}{cc}{\cos\sqrt{2{\ensuremath{\mathcal{E}}}}t}&{\frac{1}{\sqrt{2{\ensuremath{\mathcal{E}}}}}\sin\sqrt{2E}t}\\{-\sqrt{2{\ensuremath{\mathcal{E}}}}\sin\sqrt{2{\ensuremath{\mathcal{E}}}}t}&{\cos\sqrt{2{\ensuremath{\mathcal{E}}}}t}\end{array}\right)}}}{\ensuremath{{\left(\begin{array}{c}{x_0}\\{y_0}\end{array}\right)}}},$$ and its image ${\ensuremath{\mathcal{C}}}$ is a closed curve lying in $x_3x_4y_3y_4$-space. Since ${\ensuremath{\mathcal{C}}}\subset U$, the points in ${\ensuremath{\mathcal{C}}}$ can be rewritten in spherical coordinates. Upon converting, we find that any point in ${\ensuremath{\mathcal{C}}}$ satisfies $p_a=p_b=0$ and $a=b=\frac{\pi}{2}$. Further, $p_c$ is a constant value over ${\ensuremath{\mathcal{C}}}$ satisfying $p_c^2=2{\ensuremath{\mathcal{E}}}$. Since ${\ensuremath{\mathcal{C}}}\subset{\ensuremath{TS^3}}$, we also have $r=1$ and $p_r=0$. Therefore, in spherical coordinates, the orbit takes the form $${\ensuremath{\mathcal{C}}}={\left\{{\left(\frac{\pi}{2},\frac{\pi}{2},c,1,0,0,p_c,0\right)}:c\in{\ensuremath{\mathbb{R}}}/2\pi{\ensuremath{\mathbb{Z}}}\right\}}.$$ Let $z(c)$ represent the point ${\left(\frac{\pi}{2},\frac{\pi}{2},c,1,0,0,p_c\right)}\in{\ensuremath{\mathcal{C}}}$.
The change of coordinates from Cartesian to spherical must now be lifted to the symplectic frame bundle and the metaplectic-c prequantization for $(Z,\omega_4)$. The change of coordinates on $\operatorname{Sp}(Z,\omega_4)$ will take place over $U$, and that on $Q_4$ will take place over ${\ensuremath{\mathcal{C}}}$. This procedure was first demonstrated in [@v1], and we follow the same steps now.
On the neighborhood $U$, we have two different coordinate maps: the Cartesian map $\Phi_c:U\rightarrow{\ensuremath{\mathbb{R}}}^8$, and the spherical map $\Phi_s:U\rightarrow {\ensuremath{\mathbb{R}}}^4\times({\ensuremath{\mathbb{R}}}/\pi{\ensuremath{\mathbb{Z}}})^2\times{\ensuremath{\mathbb{R}}}/2\pi{\ensuremath{\mathbb{Z}}}\times{\ensuremath{\mathbb{R}}}$. The change of variables on $U$ is simply the transition map $F=\Phi_s\circ\Phi_c^{-1}$. Let $\Phi_c(U)=U_c$ and $\Phi_s(U)=U_s$. Then each of the following maps is a diffeomorphism.
$\xymatrix{
& U \ar[ld]_{\Phi_c} \ar[rd]^{\Phi_s} & \\
U_c \ar[rr]^F & & U_s
}$
These observations are straightforward on $U$, but they will motivate the constructions on the symplectic frame bundle and the metaplectic-c prequantization.
Let $b_c$ be the section of $\operatorname{Sp}(Z,\omega_4)$ over $U$ given by $$b_c(z):V_4\rightarrow T_{z}Z\ \ \mbox{such that}\ \ {\ensuremath{\widehat{v}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{x_j}}}\right|_{z}},\ {\ensuremath{\widehat{w}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{y_j}}}\right|_{z}},\ \ \forall z\in U,\ j=1,\ldots,4.$$ That is, $b_c$ is the section that defines the trivialization of $\operatorname{Sp}(Z,\omega_4)|_U$ with respect to Cartesian coordinates. Let the map $\widetilde{\Phi}_c:\operatorname{Sp}(Z,\omega)|_{U}\rightarrow U_c\times\operatorname{Sp}(V_4)$ be given by $$\widetilde{\Phi}_c(b_c(z)\cdot g)=(\Phi_c(z),g),\ \ \forall z\in U,\ \forall g\in\operatorname{Sp}(V_4).$$ Similarly, let $b_s$ be the section of $\operatorname{Sp}(Z,\omega_4)|_U$ given by $$b_s(z):V_4\rightarrow T_zZ\ \ \mbox{such that}\ \ {\ensuremath{\widehat{v}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{a_j}}}\right|_{z}},\ {\ensuremath{\widehat{w}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{p_{a_j}}}}\right|_{z}},\ \ \forall z\in U,\ j=1,\ldots,4,$$ and define the map $\widetilde{\Phi}_s:\operatorname{Sp}(Z,\omega)|_U\rightarrow U_s\times\operatorname{Sp}(V_4)$ by $$\widetilde{\Phi}_s(b_s(z)\cdot g)=(\Phi_s(z),g),\ \ \forall z\in U,\ \forall g\in\operatorname{Sp}(V_4).$$
To perform the change of coordinates on the level of the symplectic frame bundle, we must lift $F$ to a map $\widetilde{F}:U_c\times\operatorname{Sp}(V_4)\rightarrow U_s\times\operatorname{Sp}(V_4)$ in such a way that the following diagram commutes.
$\xymatrix{
& \operatorname{Sp}(Z,\omega)|_{U} \ar[ld]_{\widetilde{\Phi}_c} \ar[rd]^{\widetilde{\Phi}_s} & \\
U_c\times\operatorname{Sp}(V_4) \ar[rr]^{\widetilde{F}} & & U_s\times\operatorname{Sp}(V_4)
}$
In equations \[eq:convert1\] and \[eq:convert2\], we gave explicit formulas for $x_j$ and $y_j$ in terms of $a_k$ and $p_{a_k}$. At each $z\in U$, let $G(z)$ be the $8\times 8$ matrix consisting of the partial derivatives of the Cartesian coordinates with respect to the spherical ones: $$G(z)={\ensuremath{{\left(\begin{array}{cc}{{\left.{\frac{\partial{x_k}}{\partial{a_j}}}\right|_{z}}}&{{\left.{\frac{\partial{y_k}}{\partial{a_j}}}\right|_{z}}}\\{{\left.{\frac{\partial{x_k}}{\partial{p_{a_j}}}}\right|_{z}}}&{{\left.{\frac{\partial{y_k}}{\partial{p_{a_j}}}}\right|_{z}}}\end{array}\right)}}}_{1\leq j,k\leq 4}.$$ Then $$G(z){\ensuremath{{\left(\begin{array}{c}{{\left.{\frac{\partial{}}{\partial{x_k}}}\right|_{z}}}\\{{\left.{\frac{\partial{}}{\partial{y_k}}}\right|_{z}}}\end{array}\right)}}}_{1\leq k\leq 4}={\ensuremath{{\left(\begin{array}{c}{{\left.{\frac{\partial{}}{\partial{a_j}}}\right|_{z}}}\\{{\left.{\frac{\partial{}}{\partial{p_{a_j}}}}\right|_{z}}}\end{array}\right)}}}_{1\leq j\leq 4},\ \ \forall z\in U.$$ Since the Cartesian and spherical coordinate vectors are both symplectic bases for $T_zZ$, $G(z)$ is a symplectic matrix and can be treated as an element of $\operatorname{Sp}(V_4)$. We claim that the desired map $\widetilde{F}$ is given by $$\widetilde{F}(\Phi_c(z),g)=(\Phi_s(z),G(z)g),\ \ \forall z\in U,\ \forall g\in\operatorname{Sp}(V_4).$$
To prove this claim, it suffices to show that $$\widetilde{\Phi}_s(b_c(z))=\widetilde{F}\circ\widetilde{\Phi}_c(b_c(z));$$ the images under $\widetilde{F}$ of all other elements of the fiber $\operatorname{Sp}(Z,\omega)_z$ are determined by the group action. On the right-hand side, we have $$\widetilde{F}\circ\widetilde{\Phi}_c(b_c(z))=\widetilde{F}(\Phi_c(z),I)=(\Phi_s(z),G(z)).$$ Using the definition of $b_c$, we find that $$b_c(z):G(z){\ensuremath{{\left(\begin{array}{c}{{\ensuremath{\widehat{v}}}_j}\\{{\ensuremath{\widehat{w}}}_j}\end{array}\right)}}}_{1\leq j\leq 4}\mapsto G(z){\ensuremath{{\left(\begin{array}{c}{{\left.{\frac{\partial{}}{\partial{x_j}}}\right|_{z}}}\\{{\left.{\frac{\partial{}}{\partial{y_j}}}\right|_{z}}}\end{array}\right)}}}_{1\leq j\leq 4}={\ensuremath{{\left(\begin{array}{c}{{\left.{\frac{\partial{}}{\partial{a_j}}}\right|_{z}}}\\{{\left.{\frac{\partial{}}{\partial{p_{a_j}}}}\right|_{z}}}\end{array}\right)}}}_{1\leq j\leq 4}.$$ Therefore, with respect to the spherical trivialization, $$\widetilde{\Phi}_s(b_c(z))=(\Phi_s(z),G(z))=\widetilde{F}\circ\widetilde{\Phi}_c(b_c(z)),$$ as needed. Hence the map $\widetilde{F}$ effects the change of coordinates from Cartesian to spherical on $\operatorname{Sp}(Z,\omega_4)|_{U}$.
The final lift to the metaplectic-c prequantization will take place over ${\ensuremath{\mathcal{C}}}$. Let ${\ensuremath{\mathcal{C}}}_c$ and ${\ensuremath{\mathcal{C}}}_s$ be the images of ${\ensuremath{\mathcal{C}}}$ under $\Phi_c$ and $\Phi_s$, respectively. Recall that points in ${\ensuremath{\mathcal{C}}}$ are denoted by $z(c)$ with $c\in{\ensuremath{\mathbb{R}}}/2\pi{\ensuremath{\mathbb{Z}}}$. We write $G(c)$ for $G(z(c))$. The components of $G(c)$ are single-valued with respect to $c$, so $G(c)$ is a closed path through $\operatorname{Sp}(V_4)$.
It is clear from the construction of $Q_4$ how to define a local trivialization with respect to the Cartesian coordinates. Let $\widehat{\Phi}_c:Q_4|_{{\ensuremath{\mathcal{C}}}}\rightarrow{\ensuremath{\mathcal{C}}}_c\times\operatorname{Mp}^c(V_4)$ be given by $\widehat{\Phi}_c(z(c),a)=(\Phi_c(z(c)),a)$ for all $z(c)\in{\ensuremath{\mathcal{C}}}$ and $a\in\operatorname{Mp}^c(V_4)$. Then $\widehat{\Phi}_c$ and $\widetilde{\Phi}_c$ make the following diagram commute.
$\xymatrix{
Q_4|_{{\ensuremath{\mathcal{C}}}} \ar[r]^{\Gamma_4} \ar[d]^{\widehat{\Phi}_c} & \operatorname{Sp}(Z,\omega_4)|_{{\ensuremath{\mathcal{C}}}} \ar[d]^{\widetilde{\Phi}_c} \\
{\ensuremath{\mathcal{C}}}_{c}\times\operatorname{Mp}^c(V_4) \ar[r]^{\sigma} & {\ensuremath{\mathcal{C}}}_c\times\operatorname{Sp}(V_4)
}$
We require a local trivialization of $Q_4$ with respect to the spherical coordinates that has the analogous relationship to $\widetilde{\Phi}_s$. To find the appropriate map $\widehat{\Phi}_s:Q_4|_{{\ensuremath{\mathcal{C}}}}\rightarrow{\ensuremath{\mathcal{C}}}_s\times\operatorname{Mp}^c(V_4)$, we will construct the map $\widehat{F}$ shown below, then define $\widehat{\Phi}_s=\widehat{F}\circ\widehat{\Phi}_c$.
$\xymatrix{
& Q_4|_{{\ensuremath{\mathcal{C}}}} \ar[dl]_{\widehat{\Phi}_c} \ar[rd] \ar[rrrd]^{\Sigma} & \\
{\ensuremath{\mathcal{C}}}_c\times\operatorname{Mp}^c(V_4) \ar[rr]^{\widehat{F}} \ar[rrrd]^{\sigma} & & {\ensuremath{\mathcal{C}}}_s\times\operatorname{Mp}^c(V_4) \ar[rrrd]^(.25){\sigma} & & \operatorname{Sp}(Z,\omega_4)|_{{\ensuremath{\mathcal{C}}}} \ar[ld]_(.4){\widetilde{\Phi}_c} \ar[rd]^(.4){\widetilde{\Phi}_s} & \\
& & & {\ensuremath{\mathcal{C}}}_c\times\operatorname{Sp}(V_4) \ar[rr]^{\widetilde{F}} & & {\ensuremath{\mathcal{C}}}_s\times\operatorname{Sp}(V_4)
}$
An examination of the condition $\sigma\circ\widehat{F}\circ\widehat{\Phi}_c=\widetilde{F}\circ\sigma\circ\widehat{\Phi}_c$ shows that if $\widehat{F}$ exists, then it has the form $$\widehat{F}(\Phi_c(z(c)),a)=(\Phi_s(z(c)),\widehat{G}(c)a),\ \ \forall z(c)\in{\ensuremath{\mathcal{C}}},\forall a\in\operatorname{Mp}^c(V_4),$$ where $\widehat{G}(c)\in\operatorname{Mp}^c(V_4)$ and $\sigma(\widehat{G}(c))=G(c)$ for all $c$. That is, $\widehat{G}(c)$ must be a lift of the path $G(c)$ to $\operatorname{Mp}^c(V_4)$. Further, for $\widehat{F}$ to be a well-defined function, $\widehat{G}(c)$ must be single-valued with respect to $c$.
Since $\operatorname{Mp}^c(V_4)$ is a circle extension of $\operatorname{Sp}(V_4)$, there are many possible lifts of $G(c)$ to $\operatorname{Mp}^c(V_4)$. However, it would be ideal if $\widehat{G}(c)$ were a path not only through $\operatorname{Mp}^c(V_4)$ but through the subgroup $\operatorname{Mp}(V_4)$. Recall that the one-form $\delta_4$ on $Q_4$ is $$\delta_4=\frac{1}{i\hbar}\Pi^*\beta_4+\frac{1}{2}\eta_*\vartheta_0,$$ where $\vartheta_0$ is the trivial connection on $Z\times\operatorname{Mp}^c(V_4)$. Since $\ker\eta=\operatorname{Mp}(V_4)$, a map $\widehat{F}$ that involves only $\widehat{G}(c)\in\operatorname{Mp}(V)$ would not alter the form of the $\frac{1}{2}\eta_*\vartheta_0$ term in $\delta_4$. Therefore we will lift $G(c)$ to $\widehat{G}(c)\in\operatorname{Mp}(V_4)$, and determine whether the lifted path satisfies $\widehat{G}(c+2\pi)=\widehat{G}(c)$.
If $\widehat{G}(c)\in\operatorname{Mp}(V_4)$, then its parameters take the form $(G(c),\mu(c))$, where $\mu(c)^2\mbox{Det}_{\ensuremath{\mathbb{C}}}C_{G(c)}=1$. To determine $\mu(c)$, we must calculate $G(c)$. Using the expressions in equations \[eq:convert1\] and \[eq:convert2\], we compute the partial derivatives that form the matrix $G$, and evaluate at the point $z(c)={\left(\frac{\pi}{2},\frac{\pi}{2},c,1,0,0,p_c,0\right)}\in{\ensuremath{\mathcal{C}}}_s$. Using the abbreviations $S(c)=\sin c$ and $C(c)=\cos c$, the result is $$G(c)={\left(\begin{array}{cccccccc}
-1&0&0&0&0&0&0&0\\
0&-1&0&0&0&0&0&0\\
0&0&-S(c)&C(c)&0&0&-p_cC(c)&-p_cS(c)\\
0&0&C(c)&S(c)&0&0&p_cS(c)&-p_cC(c)\\
0&0&0&0&-1&0&0&0\\
0&0&0&0&0&-1&0&0\\
0&0&0&0&0&0&-S(c)&C(c)\\
0&0&0&0&0&0&C(c)&S(c)
\end{array}\right)}.$$
Next, we evaluate $C_{G(c)}=\frac{1}{2}(G(c)-JG(c)J)$, using the matrix form for $J$ noted in Section \[subsec:stdvec\], and convert it to a $4\times 4$ complex matrix. We find that $$C_{G(c)}={\left(\begin{array}{cccc}
-1&0&0&0\\
0&-1&0&0\\
0&0&-S(c)-\frac{i}{2}p_cC(c)&C(c)-\frac{i}{2}p_cS(c)\\
0&0&C(c)+\frac{i}{2}p_cS(c)&S(c)-\frac{i}{2}p_cC(c)
\end{array}\right)},$$ which has complex determinant $${\ensuremath{\mbox{Det}}}_{\ensuremath{\mathbb{C}}}C_{G(c)}=-1-\frac{1}{4}p_c^2.$$ This value is real and constant over ${\ensuremath{\mathcal{C}}}$. Thus we can define $\widehat{G}(c)$ to be the element of $\operatorname{Mp}(V_4)\subset\operatorname{Mp}^c(V_4)$ with parameters ${\left(G(c),i{\left(1+\frac{1}{4}p_c^2\right)}^{-1/2}\right)}$, for all $c\in{\ensuremath{\mathbb{R}}}/2\pi{\ensuremath{\mathbb{Z}}}$, whereupon $\widehat{G}(c)$ is the desired closed path through $\operatorname{Mp}(V_4)$.
Having determined $\widehat{G}(c)$, we now define the map $\widehat{F}:{\ensuremath{\mathcal{C}}}_c\times\operatorname{Mp}^c(V_4)\rightarrow{\ensuremath{\mathcal{C}}}_s\times\operatorname{Mp}^c(V_4)$ by $$\widehat{F}(\Phi_c(z(c)),a)=(\Phi_s(z(c)),\widehat{G}(c)a)\ \ \forall z(c)\in{\ensuremath{\mathcal{C}}},\forall a\in\operatorname{Mp}^c(V_4).$$ The local trivialization of $Q_4|_{{\ensuremath{\mathcal{C}}}}$ that is compatible with spherical coordinates comes about by setting $\widehat{\Phi}_s=\widehat{F}\circ\widehat{\Phi}_c$. The one-form $\delta_4$ on $Q_4$ induces a one-form $\delta_{4s}$ on ${\ensuremath{\mathcal{C}}}_s\times\operatorname{Mp}^c(V_4)$ that takes the form $$\delta_{4s}=\frac{1}{i\hbar}\Pi^*\beta_4+\frac{1}{2}\eta_*\vartheta_0,$$ where $\beta_4$ is written in spherical coordinates as in equation \[eq:beta4s\], and where $\vartheta_0$ is now the trivial connection on ${\ensuremath{\mathcal{C}}}_s\times\operatorname{Mp}^c(V_4)$.
Restrictions to ${\ensuremath{\mathcal{C}}}\subset{\ensuremath{TS^3}}$
----------------------------------------------------------------------
From now on, we no longer write the maps $\Phi_s$, $\widetilde{\Phi}_s$ and $\widehat{\Phi}_s$ explicitly, but write elements of $Z$, $\operatorname{Sp}(Z,\omega_4)|_U$ and $Q_4|_{{\ensuremath{\mathcal{C}}}}$ with respect to spherical coordinates and the local spherical trivializations. Further, we treat ${\ensuremath{TS^3}}$ as a six-dimensional manifold with local coordinates $(a,b,c,p_a,p_b,p_c)$ on ${\ensuremath{TS^3}}\cap U$, and with symplectic form $\omega_3=\sum_{j=1}^3 da_j\wedge dp_{a_j}$.
Recall from Section \[subsec:mpcTS\] that we defined the map $R(x,y)=\frac{1}{2}|x|^2$ on $Z$, and set $A=R^{-1}(\frac{1}{2})$. We argued that the symplectic frame bundle $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ is isomorphic to $\operatorname{Sp}(TA/TA^\perp)|_{{\ensuremath{TS^3}}}$, and that the metaplectic-c prequantization $(Q,\Gamma,\delta)$ for $({\ensuremath{TS^3}},\omega_3)$ is obtained by restricting $(Q_{4A},\Gamma_{4A},\delta_{4A})$ to $\operatorname{Sp}(TA/TA^\perp)|_{{\ensuremath{TS^3}}}$.
In spherical coordinates, we have $$R(z)=\frac{1}{2}r^2,\ \ \forall z\in Z\cap U,$$ which has Hamiltonian vector field $$\xi_R=r{\frac{\partial{}}{\partial{p_r}}}.$$ It is immediate that at each point $z\in A\cap U$, $$T_zA=\mbox{span}{\left.{\left\{{\frac{\partial{}}{\partial{a}}},{\frac{\partial{}}{\partial{b}}},{\frac{\partial{}}{\partial{c}}},{\frac{\partial{}}{\partial{p_a}}},{\frac{\partial{}}{\partial{p_b}}},{\frac{\partial{}}{\partial{p_c}}},{\frac{\partial{}}{\partial{p_r}}}\right\}}\right|_{z}}\ \ \mbox{and}\ \ T_zA^\perp=\mbox{span}{\left.{\left\{{\frac{\partial{}}{\partial{p_r}}}\right\}}\right|_{z}},$$ which implies that $$T_zA/T_zA^\perp=\mbox{span}{\left.{\left\{{\left[{\frac{\partial{}}{\partial{a}}}\right]},{\left[{\frac{\partial{}}{\partial{b}}}\right]},{\left[{\frac{\partial{}}{\partial{c}}}\right]},{\left[{\frac{\partial{}}{\partial{p_a}}}\right]},{\left[{\frac{\partial{}}{\partial{p_b}}}\right]},{\left[{\frac{\partial{}}{\partial{p_c}}}\right]}\right\}}\right|_{z}}.$$ We see that over $A\cap U$, the local trivialization of $\operatorname{Sp}(Z,\omega_4)|_U$ with respect to spherical coordinates induces a local trivialization $$\operatorname{Sp}(Z,\omega_4;A)|_{A\cap U}=A\cap U\times\operatorname{Sp}(V_4;W_4).$$ The group homomorphism $\nu:\operatorname{Sp}(V_4;W_4)\rightarrow\operatorname{Sp}(V_3)$ then induces a local trivialization $$\operatorname{Sp}(TA/TA^\perp)|_{A\cap U}=A\cap U\times\operatorname{Sp}(V_3).$$ Restricting further to ${\ensuremath{TS^3}}\cap U\subset A\cap U$, we find that $$\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)|_{{\ensuremath{TS^3}}\cap U}={\ensuremath{TS^3}}\cap U\times\operatorname{Sp}(V_3),$$ with the local trivialization given by $${\ensuremath{\widehat{v}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{a_j}}}\right|_{z}},\ \ {\ensuremath{\widehat{w}}}_j\mapsto{\left.{\frac{\partial{}}{\partial{p_{a_j}}}}\right|_{z}},\ \ \forall z\in{\ensuremath{TS^3}}\cap U,\ j=1,2,3.$$
On the level of metaplectic-c bundles, we can write $$Q_4^A|_{{\ensuremath{\mathcal{C}}}}={\ensuremath{\mathcal{C}}}\times\operatorname{Mp}^c(V_4;W_4)$$ by restricting $Q_4|_{{\ensuremath{\mathcal{C}}}}$ to $\operatorname{Sp}(Z,\omega;A)|_{{\ensuremath{\mathcal{C}}}}$. Then the group homomorphism $\hat{\nu}:\operatorname{Mp}^c(V_4;W_4)\rightarrow\operatorname{Mp}^c(V_3)$ induces the local trivialization $$Q_{4A}|_{{\ensuremath{\mathcal{C}}}}=Q|_{{\ensuremath{\mathcal{C}}}}={\ensuremath{\mathcal{C}}}\times\operatorname{Mp}^c(V_3).$$ The prequantization one-form on $Q$ is $$\delta=\frac{1}{i\hbar}\Pi^*\beta_3+\frac{1}{2}\eta_*\vartheta_0,$$ where $\vartheta_0$ is the trivial connection on ${\ensuremath{\mathcal{C}}}\times\operatorname{Mp}^c(V_3)$, and where $$\beta_3=-\sum_{j=1}^3 p_jda_j=-p_cdc$$ over ${\ensuremath{\mathcal{C}}}$.
Quantized Energy Levels for $({\ensuremath{TS^3}},\omega_3,K)$
--------------------------------------------------------------
In the local coordinates $(a,b,c,p_a,p_b,p_c)$, the map $K:{\ensuremath{TS^3}}\rightarrow{\ensuremath{\mathbb{R}}}$ is given by $$K=\frac{1}{2}{\left(p_a^2+\frac{p_b^2}{\sin^2a}+\frac{p_c^2}{\sin^2a\sin^2b}\right)},$$ and its Hamiltonian vector field is $$\xi_K=p_a{\frac{\partial{}}{\partial{a}}}+\frac{p_b}{\sin^2a}{\frac{\partial{}}{\partial{b}}}+\frac{p_c}{\sin^2a\sin^2b}{\frac{\partial{}}{\partial{c}}}+{\left(\frac{p_b^2}{\sin^3 a}\cos a+\frac{p_c^2\cos a}{\sin^3 a\sin^2b}\right)}{\frac{\partial{}}{\partial{p_a}}}+\frac{p_c^2\cos b}{\sin^2a\sin^3b}{\frac{\partial{}}{\partial{p_b}}}.$$ Let $\xi_K$ have flow $\psi_K^t$ on ${\ensuremath{TS^3}}$.
Recall that we fixed ${\ensuremath{\mathcal{E}}}>0$ and defined ${\ensuremath{\mathcal{S}}}=K^{-1}({\ensuremath{\mathcal{E}}})$, and that $z_0$ is a point in ${\ensuremath{\mathcal{S}}}$. The closed curve ${\ensuremath{\mathcal{C}}}$ is the orbit of $\xi_K$ through $z_0$. We will show that the orbit of ${\ensuremath{\widetilde{\xi}}}_K$ through $(z_0,I)\in\operatorname{Sp}(T{\ensuremath{\mathcal{S}}}/T{\ensuremath{\mathcal{S}}}^\perp)$ is closed, and then we will determine the values of ${\ensuremath{\mathcal{E}}}$ for which the orbit of ${\ensuremath{\widehat{\xi}}}_K$ through $(z_0,I)\in Q_{\ensuremath{\mathcal{S}}}$ is also closed.
The lift of $\psi^t_K$ to $\widetilde{\psi}^t_K$ on $\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ is given by $$\widetilde{\psi}^t_K(z,I)=(z,\psi^t_{K*}|_z),\ \ \forall z\in{\ensuremath{TS^3}}\cap U,$$ which implies that the lifted vector field ${\ensuremath{\widetilde{\xi}}}_K$ is $${\ensuremath{\widetilde{\xi}}}_{K}(z,I)={\left(\xi_K(z),{\left.{\frac{d{}}{d{t}}}\right|_{t=0}}\psi^t_{K*}|_z\right)},\ \ \forall z\in{\ensuremath{TS^3}}\cap U.$$ As an $8\times 8$ matrix, ${\left.{\frac{d{}}{d{t}}}\right|_{t=0}}\psi^t_{K*}|_z$ can be interpreted as an element of the Lie algebra $\mathfrak{sp}(V_3)$, and its components are $${\left({\left.{\frac{d{}}{d{t}}}\right|_{t=0}}\psi^t_*|_{z}\right)}_{jk}={\left.{\frac{\partial{{\left(\xi_K\right)}_j}}{\partial{X_k}}}\right|_{X=z}},$$ where $X_k$ ranges over $(a,b,c,p_a,p_b,p_c)$. In particular, when we evaluate this matrix of partial derivatives at $z(c)\in{\ensuremath{\mathcal{C}}}$, we obtain $${\left.{\frac{d{}}{d{t}}}\right|_{t=0}}\psi^t_{K*}|_{z(c)}=
{\left(\begin{array}{cccccc}
0&0&0&1&0&0\\
0&0&0&0&1&0\\
0&0&0&0&0&1\\
-p_c^2&0&0&0&0&0\\
0&-p_c^2&0&0&0&0\\
0&0&0&0&0&0\end{array}\right)}.$$ This value is constant over ${\ensuremath{\mathcal{C}}}$; we denote it by $\kappa$. Thus $${\ensuremath{\widetilde{\xi}}}_K(z(c),I)=(\xi_K(z(c)),\kappa),\ \ \forall z(c)\in{\ensuremath{\mathcal{C}}},$$ which implies that the flow of ${\ensuremath{\widetilde{\xi}}}_K$ through $(z_0,I)\in\operatorname{Sp}({\ensuremath{TS^3}},\omega_3)$ is $$\widetilde{\psi}_K^t(z_0,I)=(\psi_K^t(z_0),\exp(t\kappa)).$$ Let $\lambda=\sqrt{p_c^2}=\sqrt{2{\ensuremath{\mathcal{E}}}}$. A calculation shows that $$\exp(t\kappa)=
{\left(\begin{array}{cccccc}
\cos\lambda t&0&0&\frac{1}{\lambda}\sin\lambda t&0&0\\
0&\cos\lambda t&0&0&\frac{1}{\lambda}\sin\lambda t&0\\
0&0&1&0&0&0\\
-\lambda\sin\lambda t&0&0&\cos\lambda t&0&0\\
0&-\lambda\sin\lambda t&0&0&\cos\lambda t&0\\
0&0&0&0&0&1\end{array}\right)}.$$ Thus the orbit through $(x_0,I)$ is closed, with period $\frac{2\pi}{\lambda}$.
Over ${\ensuremath{\mathcal{C}}}$, $\xi_K$ reduces to $$\xi_K=p_c{\frac{\partial{}}{\partial{c}}}.$$ Therefore, for all $z(c)\in{\ensuremath{\mathcal{C}}}$, we have $$T_{z(c)}{\ensuremath{\mathcal{S}}}^\perp=\mbox{span}{\left.{\left\{{\frac{\partial{}}{\partial{c}}}\right\}}\right|_{z(c)}},\ \ T_{z(c)}{\ensuremath{\mathcal{S}}}=\mbox{span}{\left.{\left\{{\frac{\partial{}}{\partial{a}}},{\frac{\partial{}}{\partial{b}}},{\frac{\partial{}}{\partial{c}}},{\frac{\partial{}}{\partial{p_a}}},{\frac{\partial{}}{\partial{p_b}}}\right\}}\right|_{z(c)}},$$ and $$T_{z(c)}{\ensuremath{\mathcal{S}}}/T_{z(c)}{\ensuremath{\mathcal{S}}}^\perp=\mbox{span}{\left.{\left\{{\left[{\frac{\partial{}}{\partial{a}}}\right]},{\left[{\frac{\partial{}}{\partial{b}}}\right]},{\left[{\frac{\partial{}}{\partial{p_a}}}\right]},{\left[{\frac{\partial{}}{\partial{p_b}}}\right]}\right\}}\right|_{z(c)}}.$$ We identify $\operatorname{Sp}(T{\ensuremath{\mathcal{S}}}/T{\ensuremath{\mathcal{S}}}^\perp)|_{{\ensuremath{\mathcal{C}}}}$ with ${\ensuremath{\mathcal{C}}}\times\operatorname{Sp}(V_2)$ in the obvious way.
Upon descending to $\operatorname{Sp}(T{\ensuremath{\mathcal{S}}}/T{\ensuremath{\mathcal{S}}}^\perp)|_{{\ensuremath{\mathcal{C}}}}$, the induced flow takes the form $$\widetilde{\psi}_K^t(z_0,I)=(\psi^t_K(z_0),\nu(\exp(t\kappa))),$$ where $\nu$ is the group homomorphism $\operatorname{Sp}(V_3;W_3){\ensuremath{\stackrel{\nu}{\longrightarrow}}}\operatorname{Sp}(V_2)$. We calculate that $$\nu(\exp(t\kappa))={\left(\begin{array}{cccc}\cos\lambda t&0&\frac{1}{\lambda}\sin\lambda t&0\\0&\cos\lambda t&0&\frac{1}{\lambda}\sin\lambda t\\-\lambda\sin\lambda t&0&\cos\lambda t&0\\0&-\lambda\sin\lambda t&0&\cos\lambda t\end{array}\right)}.$$ The corresponding vector field on $\operatorname{Sp}(T{\ensuremath{\mathcal{S}}}/T{\ensuremath{\mathcal{S}}}^\perp)|_{{\ensuremath{\mathcal{C}}}}$ is $${\ensuremath{\widetilde{\xi}}}_K(z(c),I)=(\xi_K(z(c)),\overline{\kappa}),\ \ \forall z(c)\in{\ensuremath{\mathcal{C}}},$$ where $$\overline{\kappa}={\left.{\frac{d{}}{d{t}}}\right|_{t=0}}\nu(\exp(t\kappa))={\left(\begin{array}{cccc}0&0&1&0\\0&0&0&1\\-\lambda^2&0&0&0\\0&-\lambda^2&0&0\end{array}\right)}\in\mathfrak{sp}(V_2).$$
The local trivializations lift to the level of metaplectic-c bundles, so that $Q_{\ensuremath{\mathcal{S}}}|_{\ensuremath{\mathcal{C}}}={\ensuremath{\mathcal{C}}}\times\operatorname{Mp}^c(V_2)$. The induced one-form $\delta_{\ensuremath{\mathcal{S}}}$ on $Q_{{\ensuremath{\mathcal{S}}}}$ takes the form $$\delta_{{\ensuremath{\mathcal{S}}}}=-\frac{1}{i\hbar}p_c dc+\frac{1}{2}\eta_*\vartheta_0$$ over ${\ensuremath{\mathcal{C}}}$. Therefore the horizontal lift of ${\ensuremath{\widetilde{\xi}}}_K$ to $Q_{{\ensuremath{\mathcal{S}}}}|_{{\ensuremath{\mathcal{C}}}}$ is $${\ensuremath{\widehat{\xi}}}_K(z(c),I)={\left(\xi_K(z(c)),\overline{\kappa}\oplus\frac{\lambda^2}{i\hbar}\right)},\ \ \forall z(c)\in{\ensuremath{\mathcal{C}}}.$$ Since the $\mathfrak{mp}^c(V_2)$ component is constant, the orbit of ${\ensuremath{\widehat{\xi}}}_K$ through $(z_0,I)\in Q_{\ensuremath{\mathcal{S}}}$ is $$\widehat{\psi}_K^t(z_0,I)={\left(\psi_K^t(z_0),\exp{\left(t\overline{\kappa}\oplus\frac{\lambda^2t}{i\hbar}\right)}\right)}.$$ We can write the $\operatorname{Mp}^c(V_2)$ component as $\exp(t\kappa\oplus 0)e^{\lambda^2t/i\hbar}$, where $\exp(t\overline{\kappa}\oplus 0)\in\operatorname{Mp}(V_2)\subset\operatorname{Mp}^c(V_2)$ and $e^{\lambda^2t/i\hbar}\in U(1)\subset\operatorname{Mp}^c(V_2)$.
The parameters of $\exp(t\overline{\kappa}\oplus 0)$ have the form $(\exp{t\overline{\kappa}},\mu(t))$ where $\mu(t)^2{\ensuremath{\mbox{Det}}}_{{\ensuremath{\mathbb{C}}}}C_{\exp(t\overline{\kappa})}=1$. As a $2\times 2$ complex matrix, $$C_{\exp(t\overline{\kappa})}={\ensuremath{{\left(\begin{array}{cc}{\cos\lambda t+\frac{i}{2}{\left(\lambda+\frac{1}{\lambda}\right)}\sin\lambda t}&{0}\\{0}&{\cos\lambda t+\frac{i}{2}{\left(\lambda+\frac{1}{\lambda}\right)}\sin\lambda t}\end{array}\right)}}},$$ which has determinant $${\ensuremath{\mbox{Det}}}_{\ensuremath{\mathbb{C}}}C_{\exp(t\overline{\kappa})}={\left(\cos\lambda t+\frac{i}{2}{\left(\lambda+\frac{1}{\lambda}\right)}\sin\lambda t\right)}^2.$$ This value circles the origin twice as $t$ ranges from $0$ to $2\pi/\lambda$. Therefore the parameters of $\exp(t\overline{\kappa}\oplus 0)$ are $${\left(\exp(t\overline\kappa),{\left(\cos\lambda t+\frac{i}{2}{\left(\lambda+\frac{1}{\lambda}\right)}\sin\lambda t\right)}^{-1}\right)},$$ which describes a closed orbit in $\operatorname{Mp}(V_2)$ with period $2\pi/\lambda$.
Thus the orbit in $Q_{\ensuremath{\mathcal{S}}}$ is closed if and only if the $U(1)$ term is also a closed orbit with period $2\pi/\lambda$. We require $e^{\lambda^2 t/i\hbar}=1$ when $t=2\pi/\lambda$, or equivalently, $\frac{\lambda^2}{i\hbar}\frac{2\pi}{\lambda}=-2\pi in$ for some $n\in{\ensuremath{\mathbb{Z}}}$. Rearranging and recalling that $\lambda^2=p_c^2=2{\ensuremath{\mathcal{E}}}$ results in ${\ensuremath{\mathcal{E}}}=\frac{1}{2}n^2\hbar^2$ for some $n\in{\ensuremath{\mathbb{Z}}}$.
We now recall the argument made in Section \[subsec:rescale\]. We are only concerned with the strictly positive quantized energies of $K$, because those correspond to quantized energies of the Delaunay Hamiltonian $D$ on ${\ensuremath{T^+S^3}}$. After dismissing $n<0$ as redundant, this leaves us with $${\ensuremath{\mathcal{E}}}_n=\frac{1}{2}n^2\hbar^2,\ \ n\in{\ensuremath{\mathbb{N}}}.$$ The positive value ${\ensuremath{\mathcal{E}}}_n$ is a quantized energy level of $K$ if and only if $f({\ensuremath{\mathcal{E}}}_n)$ is a quantized energy level of $D$, where the diffeomorphism $f$ is given in equation \[eq:diffeo\], and the quantized energy levels of $D$ are precisely the negative quantized energy levels of the hydrogen atom. At last, we find that the quantized energy levels of the hydrogen atom are $$E_n=f({\ensuremath{\mathcal{E}}}_n)=-\frac{m_ek^2}{2n^2\hbar^2},\ \ n\in{\ensuremath{\mathbb{N}}},$$ which is exactly the quantum mechanical prediction for our model of the hydrogen atom.
[99]{}
V.I. Arnold, *Mathematical Methods of Classical Mechanics*, Second Edition, trans. K. Vogtmann and A. Weinstein, Springer-Verlag New York, 1989.
R. Cushman and L. Bates, *Global Aspects of Classical Integrable Systems*, Birkhaüser Verlag, Basel, 1997.
C. Duval, J. Elhadad and G.M. Tuynman, Hyperfine interaction in a classical hydrogen atom and geometric quantization, *J. Geom. Phys.* **3** no. 3, 1986, pp. 401-420.
T. Ligon and M. Schaaf, On the global symmetry of the classical Kepler problem, *Rep. Math. Phys.* **9**, 1976, pp. 281-300.
P.L. Robinson, $\operatorname{Mp}^c$ structures and energy surfaces, *Quart. J. Math.* **41**(3), 1990, pp 325-334.
P.L. Robinson and J.H. Rawnsley, The metaplectic representation, $\operatorname{Mp}^c$-structures and geometric quantization, *Memoirs of the A.M.S.* vol 81, no. 410, AMS, Providence RI, 1989.
D.J. Simms, Bohr-Sommerfeld orbits and quantizable symplectic manifolds, *Math. Proc. Cambridge Philos. Soc.* **73**, 1973, pp 489-491.
J. Śniatycki, Bohr-Sommerfeld Conditions in Geometric Quantization, *Rep. Math. Phys.* **7** no. 2, 1975, pp. 303-311.
J. Vaughan, Dynamical Invariance of a New Metaplectic-c Quantization Condition, 2015, arxiv:1507.06720.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recently multiserver queues with setup times have been extensively studied because they have applications in power-saving data centers. A challenging model is the M/M/$c$/Setup queue where a server is turned off when it is idle and is turned on if there are some waiting jobs. Recently, Gandhi et al. [@Gandhi13; @Gandhi14] obtain the generating function for the number of jobs in the system using the recursive renewal reward approach. In this paper, we derive exact solutions for the joint stationary queue length distribution of the same model using two alternative methodologies: generating function approach and matrix analytic method. The generating function approach yields exact closed form expressions for the joint stationary queue length distribution and the conditional decomposition formula. On the other hand, the matrix analytic approach leads to an exact recursive algorithm to calculate the joint stationary distribution and performance measures so as to provide some application insights.'
author:
- |
Tuan PHUNG-DUC\
Department of Mathematical and Computing Sciences\
Tokyo Institute of Technology\
Email: [email protected]
title: 'Exact Solutions for M/M/c/Setup Queues'
---
Introduction
============
The core part of cloud computing is data center where a large number of servers are available. These servers consume a large amount of energy. Thus, the key issue for the management of these server farms is to minimize the power consumption while keeping acceptable service level for users. It is reported that under the current technology an idle server still consumes about 60% of its peak when processing jobs [@Barroso07]. A natural suggestion to save power is to turn off idle servers. However, off servers need some setup time to be active during which they consume power but cannot process jobs. Thus, there exists a trade-off between power-saving and performance. This motivates the study of multiserver queues with setup times.
Although queues with setup times have been extensively investigated in the literature, most papers deal with single server case [@Takagi90; @Bischof01; @Choudhury98; @Choudhury00]. These papers analyze single server queues with a general service time distribution. Artalejo et al. [@Artalejo05] present an analysis for the multiserver queue with setup times where the authors consider the case in which at most one server can be in the setup mode at a time. This policy is later referred to as staggered setup in the literature [@Gandhi10]. Artalejo et al. [@Artalejo05] derive an analytical solution by solving the set of balance equations for the joint stationary distribution of the number of active servers and that of jobs in the system using a difference equation approach. The solution of the staggered setup model is significantly simplified by Gandhi et al. [@Gandhi10].
Recently, motivated by applications in data centers, multiserver queues with setup times have been extensively investigated in the literature. In particular, Gandhi et al. [@Gandhi10] extensively analyze multiserver queues with setup times. They obtain some closed form approximations for the ON-OFF policy where any number of servers can be in the setup mode at a time. As is pointed out in Gandhi et al. [@Gandhi10], from an analytical point of view the most challenging model is the ON-OFF policy where the number of servers in setup mode is not limited. Recently, Gandhi et al. [@Gandhi13; @Gandhi14] analyze the M/M/$c$/Setup model with the ON-OFF policy using a recursive renewal reward approach. Gandhi et al. [@Gandhi13; @Gandhi14] obtain the generating function of the number of jobs in the system and investigate the response time distribution.
The main aim of our current paper is to derive explicit solutions for the joint queue length distribution for the M/M/$c$/Setup model with ON-OFF policy via two standard methodologies, i.e., generating function approach and matrix analytic method. The advantage of the generating function approach is that it provides detailed results for the joint stationary distribution, i.e., exact expressions for the joint stationary queue length distribution, generating functions and factorial moments of any order. Furthermore, the generating function approach gives a new look to the conditional decomposition for the queue length. On the other hand, the matrix analytic method yields an efficient algorithm where the rate matrix ($R$) and the first passage probability matrix ($G$) are explicitly obtained. In the two methods of this paper, we exploit special structure of the non-homogeneous part of the underlying Markov chain to have significant reductions of the computational complexity in comparison with existing methods in the literature [@Gandhi13; @Gandhi14; @Benny_Johan11].
Some closely related works are as follows. Mitrani [@Mitrani11; @Mitrani13] considers models for server farms with setup costs. The author analyzes the models where a group of reserve servers are shutdown instantaneously if the number of jobs in the system is smaller than some lower threshold and are powered up instantaneously when the queue length exceeds some upper threshold. Because of this instantaneous shutdown and setup, the underlying Markov chain in [@Mitrani13] has a simple birth and death structure which allows closed form solutions. The author investigates the optimal lower and upper thresholds for the system. Mitrani [@Mitrani11] extends [@Mitrani13] to the case where each job has an exponentially distributed random timer exceeding which the job leaves the system. Schwartz et al. [@Schwartz12] consider a similar model to that in [@Mitrani11]. A finite buffer model is presented and analyzed in [@phungduc15] while a model with impatient customers is analyzed in [@phungduc14].
The rest of this paper is organized as follows. Section \[model:sec\] presents the model in detail while Section \[components:sec\] is devoted to the analysis of the model via generating functions. Section \[matrix\_ana:sec\] is devoted to the analysis via matrix analytic methods. Section \[comparison\_approach:sec\] presents a comparison of the several approaches that can be used to analyze our M/M/c/Setup model. Section 6 presents some variant models for which the methodologies in this paper can be easily adapted. Some numerical examples are presented in Section \[numerical:sec\] to show insights into the performance of the system. Concluding remarks are presented in Section \[conclusion:sec\].
Model and Markov Chain {#model:sec}
======================
Model
-----
We consider M/M/$c$/Setup queueing systems with ON-OFF policy. Jobs arrive at the system according to a Poisson process with rate $\lambda$. We assume that the service time of jobs follows an exponential distribution with mean $1/\mu$. In this system, upon service completion, a server is turned off immediately if there are no waiting jobs. Otherwise, it immediately takes a waiting job to process. Upon the arrival of a job, an OFF server (if any) is turned on and the job is placed in the buffer. However, a server needs some setup time to be active so as to serve waiting jobs. We assume that the setup time follows the exponential distribution with mean $1/\alpha$. Assuming that there are two jobs in the system, one job is receiving service and the other job in the buffer is waiting for a server in setup process. Under this situation, if the service completes before the setup, the waiting job is served immediately by the active server and the server in setup process is turned off.
Let $j$ denote the number of customers in the system and $i$ denote the number of active servers. The number of servers in setup process is $\min(j-i, c-i)$. Under these assumptions, the number of active servers is smaller than or equal to the number of jobs in the system. Therefore, in this model a server is in either BUSY or OFF or SETUP. We assume that waiting jobs are served according to a first-come-first-served (FCFS) manner. We call this model an M/M/$c$/Setup queue. The exponential assumptions for the inter-arrival, setup time and service time allow us to construct a Markov chain whose stationary distribution is explicitly obtained. It should be noted that we can easily construct a Markov chain for a more general model with Markovian arrival process (MAP) and phase-type service and setup time distributions. However, the number of states of the resulting Markov chain explodes and thus analytical solutions do not exist.
Markov chain and notations
--------------------------
It is easy to see that the stability condition for the system is $\lambda < c \mu$ because all the servers are eventually active if the number of jobs in the system is large enough. Let $C(t)$ and $N(t)$ denote the number of busy servers and the total number of jobs in the system, respectively. Under the assumptions made in Section \[model:sec\], it is easy to see that $\{X(t) = (C(t), N(t)); t \geq 0\}$ forms a Markov chain in the state space $$\mathcal{S} = \{ (i,j); i = 0,1,\dots,c, j = i,i+1,\dots \}.$$ See Figure \[m:fig\] for the transitions among states. Let $$\pi_{i,j} = \lim_{t \to \infty} \mathbb{P} (C(t) = i, N(t) = j), \qquad (i,j) \in \mathcal{S}.$$ It should be noted that at the state $(i,j)$ the number of waiting jobs is $j-i$. We define the generating functions for the number of waiting jobs as follows. $$\Pi_i (z) = \sum_{j=i}^\infty \pi_{i,j} z^{j-i}, \qquad i = 0,1,\dots,c.$$ We are also interested in finding the factorial moments defined by $\Pi_i^{(n)} (1)$, where $f^{(n)} (x)$ denotes the $n$-th derivative of $f(x)$. We denote the set of non-negative integers and that of positive integers as follows. $${\mathbb{Z}}_+ = \{ 0,1,2,\dots \}, \qquad {\mathbb{N}}= \{ 1,2,3, \dots \}.$$
For $\phi \in \mathbb{R}$, the Pochhammer symbol is defined as follows. $$(\phi)_n = \left \{
\begin{array}{ll}
1 & n = 0, \\
\phi (\phi + 1) \cdots (\phi + n-1), & n \in {\mathbb{N}}.
\end{array}
\right.$$
![State transition diagram.[]{data-label="m:fig"}](model.eps)
Generating Function Approach {#components:sec}
============================
In this section, we derive explicit expressions for the generating functions and the factorial moments. The term “explicit" means that these expressions do not contain [*limits*]{} and they can be exactly calculated using a finite procedure.
Explicit expressions
--------------------
The balance equations for the case $i=0$ read as follows. $$\begin{aligned}
\label{pi_00:eq}
\lambda \pi_{0,0} & = \mu \pi_{1,1}, \qquad j = 0, \\
\label{pi_0jleqc1:eq}
(\lambda + j \alpha) \pi_{0,j} & = \lambda \pi_{0,j-1} , \qquad j = 1,2,\dots,c-1, \\
\label{pi_0jgeqc:eq}
(\lambda + c \alpha) \pi_{0,j} & = \lambda \pi_{0,j-1}, \qquad j \geq c.\end{aligned}$$ Let $\widehat{\Pi}_0 (z) = \sum_{j=c}^\infty \pi_{0,j} z^j$. Multiplying (\[pi\_0jgeqc:eq\]) by $z^j$ and summing over $j \geq c$, we obtain $$\label{widehatp0z:eq}
\widehat{\Pi}_0 (z) = \frac{ \lambda \pi_{ 0, c-1} z^c }{ \lambda + c \alpha - \lambda z} = z^c \frac{A_{0,0}}{\hat{z}_0 - z}, \qquad \Pi_0 (z) = \sum_{j=0}^{c-1} \pi_{0,j} z^j + \widehat{\Pi}_0 (z),$$ where $$A_{0,0} = \lambda \pi_{0,c-1}, \qquad \hat{z}_0 = \frac{\lambda + c\alpha}{\lambda}.$$ Equation (\[pi\_0jleqc1:eq\]) yields $$\pi_{0,j} = \pi_{0,0} \prod_{i=0}^j \frac{ \lambda }{ \lambda + j \alpha }, \qquad j = 1,2,\dots,c-1.$$ Furthermore, from the first equation in (\[widehatp0z:eq\]), we obtain $$\pi_{0,j} = \frac{ \lambda \pi_{ 0, c-1} }{ \lambda + c\mu} \left( \frac{\lambda}{\lambda + c\alpha} \right)^{j-c} = \frac{A_{0,0}}{\hat{z}_0} \left( \frac{1}{\hat{z}_0} \right)^{j-c}, \qquad j \geq c.$$
At this moment, we have the fact that $\pi_{0,j} $ $(j \geq 1)$ and $\pi_{1,1}$ are expressed in terms of $\pi_{0,0}$.
Differentiating (\[widehatp0z:eq\]) $n$ times yields the following recursive formulae for the factorial moments. $$\begin{aligned}
\widehat{\Pi}_0^{(n)} (1) & = \frac{\lambda}{c\mu} \widehat{\Pi}_0^{(n-1)} (1) + \frac{\lambda}{c\mu} \pi_{0,c-1} (c-n)_n, \\
{\Pi}_0^{(n)} (1) & = \sum_{j = 0}^{c-1} \pi_{0,j} (j-n+1)_n + \widehat{\Pi}_0^{(n)} (1),\end{aligned}$$ for $n \in {\mathbb{N}}$.
We shift to the case $i=1$. The balance equations are given as follows. $$\begin{aligned}
\label{pi1j_j1:eq}
(\lambda + \mu) \pi_{1,1} & = \alpha \pi_{0,1} + \mu \pi_{1,2} + 2\mu \pi_{2,2}, \\
\label{pi1j:eq_original}
(\lambda + \mu + (j-1) \alpha) \pi_{1,j} & = j \alpha \pi_{0,j} + \lambda \pi_{1,j-1} + \mu \pi_{1,j+1}, \qquad 2 \leq j \leq c-1, \\
\label{pi1j_jgeqc}
(\lambda + \mu + (c-1)\alpha ) \pi_{1,j} & = c \alpha \pi_{0,j} + \lambda \pi_{1,j-1} + \mu \pi_{1,j+1}, \qquad j \geq c.\end{aligned}$$ Letting $\widehat{\Pi}_1 (z) = \sum_{j=c}^\infty \pi_{1,j} z^{j-1}$, we have $\Pi_1 (z) = \sum_{j=1}^{c-1} \pi_{1,j} z^{j-1} + \widehat{\Pi}_1 (z)$. Multiplying (\[pi1j\_jgeqc\]) by $z^{j-1}$ and summing up over $j \geq c$ yields, $$(\lambda + \mu + (c-1) \alpha) \widehat{\Pi}_1(z) = \frac{c \alpha}{z} \widehat{\Pi}_0 (z) + \lambda z \widehat{\Pi}_1 (z) + \lambda \pi_{1,c-1} z^{c-1} + \frac{\mu}{z} (\widehat{\Pi}_1(z) - \pi_{1,c} z^{c-1}).$$ Rearranging this equation we obtain $$\begin{aligned}
\label{pi1z:eq}
[ (\lambda + \mu + (c-1) \alpha) z - \lambda z^2 -\mu ] \widehat{\Pi}_1 (z) = c \alpha \widehat{\Pi}_0 (z) + \lambda \pi_{1,c-1} z^c - \mu \pi_{1,c} z^{c-1}.
$$ Let $f_1 (z) = (\lambda + \mu + (c-1) \alpha) z - \lambda z^2 -\mu$. Because $f_1(0) = -\mu < 0$, $f_1(1) = (c-1)\alpha >0$ and $f_1(\infty) = - \infty$, $f_1(z)$ has two roots $z_1$ and $\hat{z}_1$ such that $0 < z_1 < 1 < \hat{z}_1$. We have $$\begin{aligned}
z_1 & = \frac{ \lambda + \mu + (c-1) \alpha - \sqrt{ (\lambda + \mu + (c-1) \alpha)^2 - 4 \lambda \mu } }{2 \lambda}, \\
\hat{z}_1 & = \frac{ \lambda + \mu + (c-1) \alpha + \sqrt{ (\lambda + \mu + (c-1) \alpha)^2 - 4 \lambda \mu } }{2 \lambda}.\end{aligned}$$ Substituting $z=z_1$ into (\[pi1z:eq\]), we obtain $$\label{pi1c:eq}
\pi_{1,c} = \frac{ c \alpha \widehat{\Pi}_0 (z_1) + \lambda \pi_{1,c-1} z_1^{c} }{ \mu z_1^{c-1} }.$$ We derive a recursive scheme to determine $\pi_{1,j}$ ($j = 2,3,\dots,c$). Indeed, rewriting (\[pi1c:eq\]) yields $$\pi_{1,c} = a^{(1)}_c + b^{(1)}_c \pi_{1,c-1},$$ where $$\label{a1cb1c:eq}
a^{(1)}_c = \frac{c \alpha \widehat{\Pi}_0 (z_1)}{\mu z_1^{c-1}}, \qquad b^{(1)}_c = \frac{\lambda z_1}{\mu}.$$ Using mathematical induction, we obtain the following lemma.
\[lemma1:lem\] $$\label{pi1j:eq}
\pi_{1,j} = a^{(1)}_j + b^{(1)}_j \pi_{1,j-1}, \qquad 2 \leq j \leq c,$$ where $$\label{a1j:b1j:eq}
a^{(1)}_j = \frac{ j \alpha \pi_{0,j} }{ \lambda + \mu + (j-1) \alpha - \mu b^{(1)}_{j+1} } , \qquad b^{(1)}_j = \frac{ \lambda }{ \lambda + \mu + (j-1) \alpha - \mu b^{(1)}_{j+1} },$$ for $j = c-1,c-2,\dots,1$. Furthermore, we have $$0 < a^{(1)}_j, \qquad 0 < b^{(1)}_j < \frac{\lambda}{\mu}, \qquad j = 1,2\dots,c.$$ The generating function $\widehat{\Pi}_1 (z)$ is explicitly obtained as follows. $$\label{pi_1hat(z):eq}
\widehat{\Pi}_1 (z) = z^{c-1} \left( \frac{A_{1,0}}{\hat{z}_0 - z} + \frac{A_{1,1}}{\hat{z}_1 - z} \right),$$ where $$A_{1,0} = \frac{A_{0,0} \hat{z}_0}{f_1(\hat{z}_0)}, \qquad A_{1,1} = -\frac{A_{0,0} \hat{z}_0}{f_1(\hat{z}_0)} + \pi_{1,c-1}.$$
We use mathematical induction for the proof of this lemma. First, we prove (\[pi1j:eq\]). It is clear that (\[pi1j:eq\]) is true for $j=c$ due to (\[a1cb1c:eq\]). Assuming that (\[pi1j:eq\]) is true for $j+1$, i.e., $$\pi_{1,j+1} = a^{(1)}_{j+1} + b^{(1)}_{j+1}\pi_{1,j},$$ for some $j \leq c-1$. Substituting this expression into (\[pi1j:eq\_original\]) and rearranging the result we obtain (\[pi1j:eq\]). Next, we also prove the inequalities. It is clear that Lemma \[lemma1:lem\] is true for $j=c$ since $$0 < a^{(1)}_c, \qquad 0 < b^{(1)}_c < \frac{\lambda}{\mu},$$ because $0 < z_1 < 1$. Assuming that $0 < b^{(1)}_{j+1} < \frac{\lambda}{\mu}$ and $a^{(1)}_{j+1} > 0$, we have $$\mu + (c-1)\alpha < \lambda + \mu + (c-1) \alpha - \mu b^{(1)}_{j+1} < \lambda + \mu + (c-1) \alpha,$$ which together with (\[a1j:b1j:eq\]) yield $$0 < \frac{\lambda}{\lambda + \mu + (j-1) \alpha} < b^{(1)}_j < \frac{\lambda}{\mu + (j-1) \alpha} < \frac{\lambda}{\mu}.$$ and $$0 < \frac{j \alpha \pi_{0,j}}{\lambda + \mu + (j-1)\alpha} < a^{(1)}_j.$$
Substituting (\[pi1c:eq\]) to (\[pi1z:eq\]) and dividing both sides by $(z-z_1)$, we obtain (\[pi\_1hat(z):eq\]) after some rearrangement. It should be noted that (\[decompose:formula\]) is used to decompose $\Pi_1(z)$ into simple form. $$\label{decompose:formula}
\frac{1}{(a-z)(b-z)} = \frac{1}{b-a} \left( \frac{1}{a-z} - \frac{1}{b-z} \right), \qquad \forall \ a \neq b.$$
At this moment, $\pi_{1,j}$ ($j \geq 1$) is expressed in terms of $\pi_{0,0}$. Thus, $\pi_{2,2}$ is also expressed in terms of $\pi_{0,0}$ due to the following formula representing the balance between the rates in and out the set $\{(i,j); i = 0,1; j \geq i \}$, i.e., $$2\mu \pi_{2,2} = \sum_{j=2}^\infty \min(j-1,c-1)\alpha \pi_{1,j}.$$
We are interested in finding the factorial moments. Taking the derivative of (\[pi1z:eq\]) $n$ times yields $$\begin{aligned}
\lefteqn{ f_1(z) \widehat{\Pi}_1^{(n)} (z) + n f_1^\prime (z) \widehat{\Pi}_1^{(n-1)} (z) + \frac{n(n-1)}{2} f_1^{\prime \prime} (z) \widehat{\Pi}_1^{(n-2)} (z) = } \nonumber \\
&& c \alpha \widehat{\Pi}_0^{(n)} (z) + \lambda \pi_{1,c-1} (c-n+1)_n z^{c-n} - \mu \pi_{1,c} (c-n)_n z^{c-1-n}.\end{aligned}$$ Putting $z=1$ into this equation yields, $$\begin{aligned}
\label{widehatPi1:eq}
\widehat{\Pi}_1^{(n)} (1) & = & \frac{c}{c-1} \widehat{\Pi}_0^{(n)} (1) + \frac{ n (\lambda -\mu - (c-1) \alpha) \widehat{\Pi}_1^{(n-1)} (1) + \lambda n(n-1) \widehat{\Pi}_1^{(n-2)} (1) }{(c-1)\alpha} \nonumber \\
& & \mbox{} + \frac{\lambda \pi_{1,c-1} (c-n+1)_n - \mu \pi_{1,c} (c-n)_n }{(c-1)\alpha}, \end{aligned}$$ which is a recursive formula for computing $\widehat{\Pi}_1^{(n)} (1)$ ($n \in {\mathbb{N}}$). It should be noted that $\widehat{\Pi}_0^{(n)} (1)$ is explicitly obtained from (\[widehatp0z:eq\]). Thus, from (\[widehatPi1:eq\]) we obtain the factorial moments ${\Pi}_1^{(n)} (1)$.
Now, we consider general case where $i = 2,3,\dots,c-1$. The balance equations are as follows. $$\begin{aligned}
\label{pi_ii_jleqc:eq}
(\lambda + i \mu) \pi_{i,i} & = \alpha \pi_{i-1,i} + i\mu \pi_{i,i+1} + (i+1)\mu \pi_{i+1,i+1}, \quad j = i \\
\label{pi_ij_jleqc:eq}
(\lambda + i \mu + (j-i) \alpha) \pi_{i,j} & = \lambda \pi_{i,j-1} + (j-i+1) \alpha \pi_{i-1,j} + i \mu \pi_{i,j+1}, \qquad i+1 \leq j \leq c-1, \\
\label{pi_ij_jgeqc:eq}
(\lambda + i\mu + (c-i) \alpha) \pi_{i,j} & = \lambda \pi_{i,j-1} + (c-i+1) \alpha \pi_{i-1,j} + i\mu \pi_{i,j+1}, \qquad j \geq c.\end{aligned}$$ We define the generating function $\widehat{\Pi}_i (z) = \sum_{j=c}^\infty \pi_{i,j-i} z^{j-i}$. We then have $\Pi_i (z) = \sum_{j=i}^{c-1} \pi_{i,j} z^{j-i} + \widehat{\Pi}_i (z) $. Multiplying (\[pi\_ij\_jgeqc:eq\]) by $z^{j-i}$ and summing over $j \geq c$, we obtain $$\begin{aligned}
(\lambda + i \mu + (c-i) \alpha) \widehat{\Pi}_i (z) & = & \lambda \pi_{i,c-1} z^{c-i} + \lambda z \widehat{\Pi}_i (z) + \frac{(c-i+1) \alpha}{z} \widehat{\Pi}_{i-1} (z) \nonumber \\
& & \mbox{} + \frac{i \mu }{ z} (\widehat{\Pi}_i (z) - \pi_{i,c} z^{c+1-i}). \end{aligned}$$ Rearranging this equation, we obtain $$\begin{aligned}
\label{Pi(z):eq}
[(\lambda + i \mu + (c-i) \alpha) z - \lambda z^2 - i \mu ] \widehat{\Pi}_i (z) = (c-i+1) \alpha \widehat{\Pi}_{i-1} (z) + \lambda \pi_{i,c-1} z^{c-i+1} - i \mu \pi_{i,c} z^{c-i}.\end{aligned}$$ Let $f_i (z) = (\lambda + i \mu + (c-i) \alpha) z - \lambda z^2 - i \mu $. Because $f_i (0) = -i \mu < 0$, $f_i (1) = (c-i) \alpha > 0$ and ($f_i (\infty) = - \infty$), there exists some $0 < z_i < 1 < \hat{z}_1$ such that $f_i (z_i) = f_i (\hat{z}_i) = 0$. In particular, we have $$\begin{aligned}
z_i & = \frac{ \lambda + i \mu + (c-i) \alpha - \sqrt{ (\lambda + i \mu + (c-i) \alpha)^2 - 4 i \lambda \mu } }{2 \lambda}, \\
\hat{z}_i & = \frac{ \lambda + i \mu + (c-i) \alpha + \sqrt{ (\lambda + i \mu + (c-i) \alpha)^2 - 4 i \lambda \mu } }{2 \lambda}.\end{aligned}$$ Putting $z= z_i$ into (\[Pi(z):eq\]) yields, $$\label{pi_ic:pi_ic_minus}
\pi_{i,c} = \frac{(c-i+1) \alpha \widehat{\Pi}_{i-1} (z_i) + \lambda \pi_{i,c-1} z_i^{c-i+1} }{i\mu z_i^{c-i} }$$ This equation together with (\[pi\_ij\_jleqc:eq\]) determine $\pi_{i,j}$ ($ i+1 \leq j \leq c$) as follows.
\[lemma32\] We have $$\pi_{i,j} = a^{(i)}_j + b^{(i)}_j \pi_{i,j-1}, \qquad j = i+1, i+2,\dots,c,$$ where $$a^{(i)}_c = \frac{ (c-i+1) \alpha \widehat{\Pi}_{i-1} (z_i)} { i\mu z_i^{c-i} }, \qquad b^{(i)}_c = \frac{\lambda z_i }{i \mu},$$ and for $j = c-1,\dots,i+1$, $$a^{(i)}_j = \frac{ (j-i+1)\alpha \pi_{i-1,j} + i\mu a^{(i)}_{j+1}}{ \lambda + i\mu + (j-i)\alpha - i\mu b^{(i)}_{j+1} }, \qquad b^{(i)}_j = \frac{\lambda}{\lambda + i\mu + (j-i)\alpha - i\mu b^{(i)}_{j+1}}.$$ Furthermore, we have $$0 < a^{(i)}_j, \qquad 0 < b^{(i)}_j < \frac{\lambda}{i\mu}.$$ In addition, the generating function $\widehat{\Pi}_i (z)$ ($i=2,\dots,c-1$) is explicitly obtained as follows. $$\label{explicit:Pi(z)}
\widehat{\Pi}_i (z) = z^{c-i} \left( \sum_{j=0}^{i} \frac{A_{i,j}} { \hat{z}_j - z} \right),$$ where $$A_{i,j}= \frac{A_{i-1,j} \hat{z}_j}{f_i (\hat{z}_j) }, \qquad A_{i,i} = -(c-i+1) \alpha \sum_{j=0}^{i-1} \frac{ A_{i-1,j} \hat{z}_j }{ f_i (\hat{z}_j) } + \pi_{i,c-1}.$$
The proof of Lemma \[lemma32\] proceeds in the same manner as used in Lemma \[lemma1:lem\]. We prove (\[explicit:Pi(z)\]) using mathematical induction. Indeed, substituting $$\widehat{\Pi}_{i-1} (z) = z^{c-i+1} \left( \sum_{j=0}^{i-1} \frac{A_{i-1,j}} { \hat{z}_j - z} \right),$$ into (\[Pi(z):eq\]), deleting $(z-z_i)$ from both sides and rearranging the result, we obtain (\[explicit:Pi(z)\]). It should be noted that (\[decompose:formula\]) is used to obtain (\[explicit:Pi(z)\]).
It should be noted that $\pi_{i,j}$ ($j \geq i$) is expressed in terms of $\pi_{0,0}$. Furthermore, $\pi_{i+1,i+1}$ is expressed in terms of $\pi_{i,j}$ ($j = i+1,i+2,\dots$) and then in terms of $\pi_{0,0}$ via the balance of the flows in and out the set of states $\{(k,j); 0 \leq k \leq i, j \geq k \}$, i.e., $$(i+1)\mu \pi_{i+1,i+1} = \sum_{j=i+1}^\infty \min(j-i,c-i) \alpha \pi_{i,j}.$$
Taking the derivative of (\[Pi(z):eq\]) $n$ times yields $$\begin{aligned}
\widehat{\Pi}_i^{(n)} (1) & = & \frac{c -i +1}{c-i} \widehat{\Pi}_{i-1}^{(n)} (1) + \frac{ n (\lambda -\mu - (c-i) \alpha) \widehat{\Pi}_i^{(n-1)} (1) + n(n-1) \lambda \widehat{\Pi}_i^{(n-2)} (1) }{(c-i)\alpha} \nonumber \\
& & \mbox{} + \frac{\lambda \pi_{i,c-1} (c -i +2 -n)_n - i \mu \pi_{i,c} (c-i + 1- n )_n }{(c-i)\alpha},\end{aligned}$$ which is a recursive formula to compute all the factorial moments $\widehat{\Pi}_i^{(n)} (1)$ ($n \in {\mathbb{N}}$). It should be noted that $\widehat{\Pi}_i^{(0)} (1) = \widehat{\Pi}_i (1)$ and $\widehat{\Pi}_{i-1}^{(n)} (1)$ ($n \in {\mathbb{N}}$) are already known.
Finally, the case $i = c $ needs some special treatment. Balance equations read as follows. $$\begin{aligned}
(\lambda + c\mu ) \pi_{c,c} & = & \alpha \pi_{c-1,c} + c\mu \pi_{c,c+1}, \qquad j = c, \\
\label{picj:eq}
(\lambda + c\mu ) \pi_{c,j} & = & \alpha \pi_{c-1,j} + \lambda \pi_{c,j-1} + c\mu \pi_{c,j+1}, \qquad j \geq c + 1.\end{aligned}$$ Defining $$\widehat{\Pi}_c ( z) = \sum_{j = c}^\infty \pi_{c,j} z^{j-c},$$ we have $\Pi_c (z) = \widehat{\Pi}_c ( z)$. Multiplying (\[picj:eq\]) by $z^{j-c}$ and summing up over $j \geq c$ yields $$\label{Pc(z):eq}
(\lambda + c\mu) \widehat{\Pi}_c(z) = \frac{\alpha}{z} \widehat{\Pi}_{c-1} (z) + \lambda z \widehat{\Pi}_c(z) + \frac{c\mu}{z} (\widehat{\Pi}_c(z) - \pi_{c,c}),$$ leading to $$f_c(z) \widehat{\Pi}_c(z) = \alpha \widehat{\Pi}_{c-1}(z) - c\mu \pi_{c,c},$$ or equivalently, $$\begin{aligned}
\label{pi_cz:eq}
\widehat{\Pi}_c(z) & = \frac{ \alpha \widehat{\Pi}_{c-1} (z) - c \mu \pi_{c,c} }{ z- 1} \frac{1}{c\mu - \lambda z} = \frac{ \alpha \left( \widehat{\Pi}_{c-1} (z) - \widehat{\Pi}_{c-1} (1) \right) }{ z- 1} \frac{1}{c\mu - \lambda z}.\end{aligned}$$ where $ f_c(z) = (\lambda + c\mu) z - \lambda z^2 - c\mu$ and $ \alpha \widehat{\Pi}_{c-1} (1) = c \mu \pi_{c,c}$ is used in the second equality of (\[pi\_cz:eq\]).
It should be noted that the numerator and denominator of the first term in the right hand side of (\[pi\_cz:eq\]) vanish at $z=1$. Thus, applying l’Hopital’s rule, we obtain $$\widehat{\Pi}_c(1) = \frac{ \alpha \widehat{\Pi}_{c-1}^\prime (1) }{ c\mu - \lambda}.$$ Substituting $\widehat{\Pi}_{c-1}(z)$ in the form of (\[explicit:Pi(z)\]) with $i=c-1$ into (\[pi\_cz:eq\]), we obtain $$\label{pi_c(z):decompose}
\widehat{\Pi}_c(z) = \sum_{j=0}^c \frac{A_{c,j}}{\hat{z}_j - z},$$ where $$\hat{z}_c = \frac{c\mu}{\lambda}, \qquad A_{c,j} = \frac{A_{c-1,j}}{\hat{z}_c - 1}, \qquad j = 0,1,\dots,c-1, \qquad A_{c,c}= - \sum_{j=0}^{c-1} \frac{A_{c-1,j} \hat{z}_j }{f_c(\hat{z}_j)}.$$
Taking the derivative of (\[Pc(z):eq\]) $n$ times and rearranging the result and then applying l’Hopital’s rule yields,
$$\Pi^{(n)}_c (1) = \frac{ \alpha \Pi_{c-1}^{(n+1)} (1) + \lambda n (n-1) \Pi_c^{(n-2)} (1) + 2\lambda n \Pi_c^{(n-1)} (1) }{ (n+1) (c\mu - \lambda)}.$$ It should be noted that $\Pi_{c-1}^{(n+1)} (1)$ and $\Pi_c^{(0)} (1) = \Pi_c (1)$ are already given.
At this moment, all the probabilities $\pi_{i,j}$ ($j \leq c$) and the generating functions $\widehat{\Pi}_i (z)$ ($i=0,1,\dots,c$) are expressed in terms of $\pi_{0,0}$ which is uniquely determined using the following normalization condition. $$\Pi_0 (1) + \Pi_1 (1) + \cdots + \Pi_c (1) = 1.$$
Since explicit expressions for the generating functions are available, we can easily obtain explicit results for the factorial moments and the joint stationary distribution using $A_{i,j}$ ($0 \leq i\leq j \leq c$) and $\hat{z}_i$ ($i=0,1,\dots,c$). In particular, it follows from (\[explicit:Pi(z)\]) and (\[pi\_c(z):decompose\]) that $\pi_{i,j}$ ($i=1,2,\dots,c$, $j \geq c$) is a linear combination of $1/\hat{z}_k^j$ ($k=0,1,\dots,i$).
It was shown in [@Levy_Yechiali:76] that $\hat{z}_i$ ($i=1,2,\dots,c-1$) are distinct. In the above analysis we implicitly assume that $\hat{z}_0 \neq \hat{z}_i$ ($i=1,2,\dots,c$) and $\hat{z}_c \neq \hat{z}_i$ ($i=0,1\dots,c-1$). In case where there exists some $i$ such that $\hat{z}_0 = \hat{z}_i$ ($i=1,2, \dots,c-1$) or (and) some $j$ such that $\hat{z}_j = \hat{z}_c$, we still have explicit expressions for the generating functions and the the joint stationary distribution after some minor modification. In particular, if $\hat{z}_0 = \hat{z}_i$ for some $i=1,2,\dots,c-1$, $\pi_{i,j}$ is a linear combination of $1/\hat{z}_k^j$ ($k=0,1,\dots,i-1$) and $j/\hat{z}_0^j$.
The computational complexity of the generating function approach is $O(c^2)$. Indeed, we need to calculate $A_{i,j}$ and $\pi_{i,j}$ ($i\leq j, 0 \leq j \leq c$) in the following order: $$(0,0) \rightarrow (0,1) \rightarrow \cdots \rightarrow (0,c) \rightarrow (1,1) \rightarrow (1,2) \rightarrow \cdots \rightarrow (1,c) \rightarrow \dots \rightarrow (c,c).$$ As a result, the complexity is of order $\sum_{i=0}^c i = c(c+1)/2 = O(c^2)$. It should be noted that the recursive procedure for $\pi_{i,j}$ ($0 \leq i \leq j \leq c$) is numerically stable since it involves only positive numbers, i.e., $a^{(i)}_j$ and $b^{(i)}_j$.
Conditional stochastic decomposition {#Conditional_Decomposition:sec}
------------------------------------
We have derived the following result. $$\begin{aligned}
\Pi_c(z) & = & \frac{ \alpha ( \Pi_{c-1} (z) - \pi_{c-1,c-1}) -c\mu \pi_{c,c} }{ (z-1) (c\mu -\lambda z) }, \\
\Pi_c(1) & = & \frac{ \alpha \Pi_{c-1}^\prime (1) }{c\mu - \lambda}.\end{aligned}$$ Let $Q^{(c)}$ denote the conditional queue length given that all $c$ servers are busy in the steady state, i.e., $$\mathbb{P} (Q^{(c)} = i) = \mathbb{P} (N(t) = i + c \ | \ C(t) = c).$$ Let $P_c(z)$ denote the generating function of $Q^{(c)}$. It is easy to see that $$\begin{aligned}
P_c(z) & = \frac{\Pi_c(z)}{\Pi_c(1)} \\
& = \frac{ \alpha ( \Pi_{c-1} (z) - \pi_{c-1,c-1}) -c\mu \pi_{c,c} }{\alpha \Pi_{c-1}^\prime(1) (z-1)} \frac{1- \rho}{1 -\rho z}\\
& = \frac{\Pi_{c-1} (z) - \Pi_{c-1}(1)}{\Pi_{c-1}^\prime(1)(z-1)} \frac{1- \rho}{1 -\rho z} \\
& = \frac{\sum_{j=1}^\infty \pi_{c-1,c-1+j} (z^j - 1) }{\Pi_{c-1}^\prime(1)(z-1)} \frac{1- \rho}{1 -\rho z} \\
& = \frac{\sum_{j=1}^\infty \pi_{c-1,c-1+j} \sum_{i=0}^{j-1} z^i }{\Pi_{c-1}^\prime(1)} \frac{1- \rho}{1 -\rho z} \\
& = \frac{ \sum_{i=0}^\infty \left(\sum_{j = i+1}^\infty \pi_{c-1,c-1+j} \right) z^i }{\Pi_{c-1}^\prime(1)} \frac{1- \rho}{1 -\rho z},\end{aligned}$$ where we have used $c\mu \pi_{c,c} = \alpha (\Pi_{c-1} (1) - \pi_{c-1,c-1}) $ in the second equality.
It should be noted that $(1-\rho)/(1-\rho z)$ is the generating function of the number of waiting jobs in the conventional M/M/$c$ system without setup times under the condition that $c$ servers are busy. We denote this random variable by $Q^{(c)}_{ON-IDLE}$. It should be noted that $Q^{(c)}_{ON-IDLE}$ can also be interpreted as the number of jobs in the M/M/1 queue without vacation where the arrival rate and the service rate are $\lambda$ and $c\mu$, respectively. We give a clear interpretation for the generating function $$\frac{ \sum_{i=0}^\infty \left(\sum_{j = i+1}^\infty \pi_{c-1,c-1+j} \right) z^i }{\Pi_{c-1}^\prime(1)}.$$ For simplicity, we define $$\begin{aligned}
p_{c-1,i} = \frac{\sum_{j = i+1}^\infty \pi_{c-1,c-1+j}}{\Pi_{c-1}^\prime(1)}, \qquad i \in {\mathbb{Z}}_+.\end{aligned}$$ We have $$\sum_{j=i+1}^\infty \pi_{c-1,c-1 + j} = \mathbb{P} ( N(t) - C(t) > i \ | \ C(t) = c-1) \mathbb{P} (C(t) = c-1),$$ and $$\Pi_{c-1}^\prime(1) = \mathbb{E} [N(t) - C(t) \ | \ C(t) = c-1] \mathbb{P} (C(t) = c-1).$$ Thus, we have $$p_{c-1,i} = \frac{\mathbb{P} ( N(t) - C(t) > i \ | \ C(t) = c-1)}{ \mathbb{E} [N(t) - C(t) \ | \ C(t) = c-1] }.$$
It should be noted that $N(t)-C(t)$ is the number of jobs in the system that are waiting for the last server (in setup mode) to be active. Thus, $p_{c-1,i}$ ($i=0,1,2,\dots$) represents distribution of the number of waiting customers in front of an arbitrary waiting customer (not being served) under the condition that $c-1$ servers are active and the last server is in setup mode (see Burke [@Burke]). Let $Q_{Res}$ denote the random variable with the distribution $p_{c-1,i}$ ($i=0,1,2,\dots$). Our decomposition result is summarized as follows. $$\label{conditional_decomposition}
Q^{(c)} \,{\buildrel d \over =}\ Q^{(c)}_{ON-IDLE} + Q_{Res}.$$ We observe that $Q_{Res}$ represents the number of extra jobs due to the setup time.
The conditional decomposition (\[conditional\_decomposition\]) is not explicit in the sense that $Q^{(c)}_{ON-IDLE}$ is not an explicit random variable. However, it is useful for understanding the behavior of the system. This situation is the same in the decomposition of M/M/1 queue with working vacation (M/M/1/WV) by Servi and Finn [@Servi02]. The reason for the “implicit" stochastic decomposition is that the service is continued during the working vacation.
Tian et al. [@Tian99; @Tian03b; @Tian_Zhang] obtain a similar result for a multiserver model with vacation. However, the random variable with the distribution $p_{c-1,i}$ here is not given a clear physical meaning in [@Tian99; @Tian03b; @Tian_Zhang].
Matrix Analytic Methods {#matrix_ana:sec}
=======================
In this section we present an analysis of the model based on a quasi-birth-and-dearth process (QBD) approach.
QBD formulation
---------------
The infinitesimal of $\{X(t) \}$ is given by $$Q = \left(
\begin{array}{ccccc}
Q^{(0)}_0 & Q^{(0)}_{1} & O & O & \cdots \\
Q^{(1)}_{-1} & Q^{(1)}_0 & Q^{(1)}_1 & O & \cdots \\
O & Q^{(2)}_{-1} & Q^{(2)}_0 & Q^{(2)}_1 & \cdots \\
O & O & Q^{(3)}_{-1} & Q^{(3)}_0 & \cdots \\
\vdots & \vdots & \vdots & \vdots & \ddots \\
\end{array}
\right),$$ where $O$ denotes the zero matrix with an appropriate dimension. A Markov chain with this type of block tridiagonal matrix is called a level dependent quasi-birth-and-death process for which some efficient algorithms are available [@bright95; @matrix-conf]. The block matrices $Q^{(i)}_{-1}$ ($i \geq c+1$), $Q^{(i)}_0$ ($i \geq c$) and $Q^{(i)}_1$ ($i \geq c$) are independent of $i$ and are explicitly given as follows. $$Q^{(i)}_{-1} = Q_{-1} = diag (0, \mu, \dots, c \mu), \qquad Q^{(i)}_1 = Q_{1} = \lambda I.$$ $$\begin{aligned}
Q^{(i)}_0 = Q_0 & = \left(
\begin{array}{cccccc}
-q_0 & c \alpha & 0 & \cdots & \cdots & 0 \\
0 & -q_1 & (c-1) \alpha & \ddots & & \vdots \\
0 & 0 & -q_2 & \ddots & \ddots & \vdots \\
\vdots & \ddots & \ddots & \ddots & \ddots & 0 \\
\vdots & & \ddots & \ddots & -q_{c-1} & \alpha \\
0 & \cdots & \cdots & 0 & 0 & -q_{c}
\end{array}
\right),
$$ where $q_j = \lambda + (c-j) \alpha + j \mu$. Furthermore, $Q^{(i)}_2$ ($i \leq c$), $Q^{(i)}_1$ ($i \leq c-1$) and $Q^{(i)}_0$ ($i \leq c$) are $(i+1) \times i$, $(i+1) \times (i+1)$ and $(i+1) \times (i+2)$ matrices whose contents are given as follows. $$\begin{aligned}
Q^{(i)}_1 & = \left(
\begin{array}{ccccc}
\lambda & 0 & \cdots & 0 & 0 \\
0 & \lambda & \ddots & \vdots & \vdots \\
\vdots & \ddots & \ddots & 0 & 0 \\
0 & \cdots & 0 & \lambda & 0
\end{array}
\right), \qquad
Q^{(i)}_{-1} = \left(
\begin{array}{cccccc}
0 & 0 & \cdots & \cdots & 0 \\
0 & \mu & \ddots & \ddots & \vdots \\
0 & 0 & & \ddots & \vdots \\
\vdots & \ddots & \ddots & \ddots & 0 \\
\vdots & & \ddots & \ddots & (i-1) \mu \\
0 & \cdots & \cdots & 0 & i \mu
\end{array}
\right), \\
Q^{(i)}_0 & = \left(
\begin{array}{cccccc}
-q^{(i)}_0 & i \alpha & 0 & \cdots & \cdots & 0 \\
0 & -q^{(i)}_1 & (i-1) \alpha & \ddots & & \vdots \\
0 & 0 & -q^{(i)}_2 & \ddots & \ddots & \vdots \\
\vdots & \ddots & \ddots & \ddots & \ddots & 0 \\
\vdots & & \ddots & \ddots & -q^{(i)}_{i-1} & \alpha \\
0 & \cdots & \cdots & 0 & 0 & -q^{(i)}_{i}
\end{array}
\right), \end{aligned}$$ where $q^{(i)}_j = (i-j) \alpha + j \mu$ ($j=0,1,\dots,i$). Let $$\begin{aligned}
\vc{\pi}_i & = (\pi_{0,i}, \pi_{1,i}, \dots, \pi_{\min(i,c),i}), \qquad i \in {\mathbb{Z}}_+, \qquad \vc{\pi} = (\vc{\pi}_0,\vc{\pi}_1,\dots).\end{aligned}$$ The stationary distribution $\vc{\pi}$ is the unique solution of $$\vc{\pi} Q = \vc{0}, \qquad \vc{\pi} \vc{e} = 1,$$ where $\vc{0}$ and $\vc{e}$ represent a row vector of zeros and a column vector of ones with an appropriate size. According to the matrix analytic method [@Neuts81; @Rdef], we have $$\vc{\pi}_i = \vc{\pi}_{i-1} R^{(i)}, \qquad i \in {\mathbb{N}},$$ and $\vc{\pi}_0$ is the solution of the boundary equation $$\vc{\pi}_0 ( Q^{(0)}_0 + R^{(1)} Q^{(1)}_{-1} ) = \vc{0}, \qquad \vc{\pi}_0 ( I + R^{(1)} + R^{(1)} R^{(2)} + \cdots ) \vc{e} = 1.$$ Here $\{ R^{(i)}; i \in {\mathbb{N}}\}$ is the minimal nonnegative solution of the following equation $$\label{ri_ri_plus:eq}
Q^{(i-1)}_1 + R^{(i)} Q^{(i)}_0 + R^{(i)} R^{(i+1)} Q^{(i+1)}_{-1} = O.$$
Homogeneous part
----------------
### The rate matrix
It should be noted that $Q^{(i-1)}_1 = Q_1$ ($i \geq c$), $Q^{(i)}_0 = Q_0$ ($i \geq c$) and $Q^{(i)}_{-1} = Q_{-1}$ ($i \geq c+1$). Thus, we have $R^{(i)} = R$ for $i \geq c+1$ and $R$ is the minimal nonnegative solution of the following equation. $$\label{homogeneous:Rmatrix}
Q_{1} + R Q_{0} + R^2 Q_{-1} = O.$$ We know that $R$ is an upper diagonal matrix, i.e., $R(i,j) = r_{i,j}$ ($j \geq i$) and $R(i,j) = 0$ if $j < i$ because $Q_{-1}, Q_0, Q_1$ are upper diagonal matrix. A similar structure is also found in the model in [@Neuts81; @Perel_Yechiali]. Furthermore, this type of QBD is considered in more general contexts in [@Benny_Johan11]. Comparing the diagonal part of the quadratic equation above, we obtain $$\lambda - (\lambda + i\mu + (c-i)\alpha ) r_{i,i} + i \mu r_{i,i}^2 = 0, \qquad i = 0,1,\dots,c-1,c.$$ which has two roots. Because $R$ is the minimal nonnegative solution of (\[homogeneous:Rmatrix\]), we must choose the smallest root for $r_{i,i}$. Thus, we have $$r_{i,i} = \frac{ \lambda + i \mu + (c-i) \alpha - \sqrt{ (\lambda + i\mu + (c-i) \alpha)^2 - 4 i \lambda \mu } }{ 2 i \mu}, \qquad i = 1,2,\dots,c-1,$$ and $$r_{0,0} = \frac{\lambda}{\lambda + c \alpha}, \qquad r_{c,c} = \frac{\lambda}{c \mu} < 1.$$ Next, we shift to the non-diagonal elements, i.e., $r_{i,j}$ ($j > i$). Comparing the $(i,j)$ element in the quadratic equation, we obtain $$(c-j+1) \alpha r_{i,j-1} - (\lambda + (c-j) \alpha + j \mu) r_{i,j} + j \mu \sum_{k=i}^j r_{i,k} r_{k,j} = 0.$$ For $j = i+1$, we obtain $$(c-i) \alpha r_{i,i} - (\lambda + (c-i-1) \alpha + (i+1) \mu) r_{i,i+1} + (i+1) \mu (r_{i,i} r_{i,i+1} + r_{i,i+1} r_{i+1,i+1} ) = 0.$$ Thus, $$\begin{aligned}
r_{i,i+1} = & \frac{ (c-i) \alpha r_{i,i} }{ \lambda + (c-i-1) \alpha + (i+1) \mu - (i+1) \mu (r_{i,i} + r_{i+1,i+1}) }, \\
& i = 0,1,\dots,c-1.\end{aligned}$$ It should be noted that the right hand side contains only known quantities obtained in previous steps. For the general case, we have $$r_{i,j} = \frac{(c-j+1)\alpha r_{i,j-1} + j\mu \sum_{k=i+1}^{j-1} r_{i,k} r_{k,j} } { \lambda + (c-j) \alpha + j\mu - j \mu (r_{i,i} + r_{j,j}) }, \qquad j > i.$$ We can rewrite this formula as follows. $$\begin{aligned}
r_{i,i+h+1} = & \frac{(c-i-h)\alpha r_{i,i+h} + (i+h+1) \mu \sum_{k=i+1}^{i+h} r_{i,k} r_{k,i+h+1} } { \lambda + (c-i-h-1) \alpha + (i+h+1)\mu - (i+h+1) \mu (r_{i,i} + r_{i+h+1,i+h+1}) }, \\
& i = 0,1,\dots,c-h-1, \qquad h = 0,1,\dots,c-1.\end{aligned}$$ From these recursive formulae, we can calculate the elements of the rate matrix from the diagonal part and then the upper diagonal parts consequently.
### Non-homogeneous part {#Non-homogeneous:part}
Because $R^{(i)} = R$ ($i = c+1,c+2,\dots$) which has been explicitly obtained, we only need to find $R^{(i)}$ ($i=c,c-1,\dots,1$). Indeed, $R^{(i)}$ ($i=c,c-1,\dots,1$) is easily obtained using the following backward formula. $$R^{(i)} = - Q^{(i-1)}_1 \left( Q^{(i)}_0 + R^{(i+1)} Q^{(i+1)}_{-1} \right)^{-1}, \qquad i = c,c-1,\dots,1.$$ This is equivalent to solving the following system of linear equations. $$R^{(i)} \left( Q^{(i)}_0 + R^{(i+1)} Q^{(i+1)}_{-1} \right) = - Q^{(i-1)}_1 , \qquad i = c,c-1,\dots,1.$$
Due to the special structure of the rate matrices, i.e., they are upper diagonal matrices, this system of linear equations can be efficiently solved as follows. In this case, we need to solve the following equation $$\label{backward:eq}
X A = - Q^{(i-1)}_0,$$ where $A = Q^{(i)}_0 + R^{(i+1)} Q^{(i+1)}_{-1} $ is an upper diagonal matrix of size $(i+1) \times (i+1)$ and and $X$ is also an upper diagonal matrix of size $i \times (i+1)$ matrix. Let $\vc{x}_j = (0,0,\dots,x_{j,j}, x_{j,j+1},\dots,x_{j,i})$ ($j=0,1,\dots,i-1$) denote the $j$-th row vector of $X$. The above equation is equivalent to $$\vc{x}_j A = (0,0,\dots, - \lambda, 0, \dots, 0), \qquad j = 0,1,\dots,i-1,$$ where the $- \lambda$ is the $(j+1)$-th entry of the vector in the right hand side. The solution of this equation is given by $$\begin{aligned}
\label{recursive:eq_xij}
x_{j,j} & = - \frac{\lambda}{a_{j,j}}, \qquad
x_{j,l} = - \frac{\sum_{k=j}^{l-1} x_{j,k} a_{k,l}}{a_{l,l}}, \qquad l = j+1,j+2,\dots,i,\end{aligned}$$ where $a_{i,j}$ is the $(i,j)$ entry of $A$.
The computational complexity of (\[recursive:eq\_xij\]) is $i-j$ and thus, the computational complexity for (\[backward:eq\]) is $O(i^2) = \sum_{j=0}^{i-1} (i-j)$ instead of $O(i^3)$ by a direct inversion of $X$. Therefore, the computational complexity for obtaining the rate matrices $R^{(i)}$ ($i = 1,2,\dots,c$) in the non-homogeneous part is of the order of $O(c^3)$ because $\sum_{j=1}^c i^2 = O(c^3)$. It should be noted that if we solve (\[backward:eq\]) by a direct inversion of the $X$, the computational complexity for $R^{(i)}$ is $i^3$ and thus the computational complexity for all the rate matrices in the non-homogeneous part ($R^{(i)}$, $i = 1,2,\dots,c$) is $O(c^4)$.
The $G$-matrix {#homo_part_gmatrix}
--------------
In this section, we derive explicit expressions for the $G$-matrix of our QBD process. It should be noted that $G$-matrix records the first passage probabilities to one level left in the homogeneous part (i.e., the number of jobs in the system is greater than $c$). These probabilities are also obtained using the recursive renewal reward approach by [@Gandhi13; @Gandhi14]. The $G$-matrix is the minimal and nonnegative solution of the following equation [@Neuts81]. $$\label{equationforG:eq}
Q_{-1} + Q_0 G + Q_1 G^2 = O.$$ From the physical interpretation of $G$, we see that $G$ is also an upper diagonal matrix. Using a similar method as in the case of $R$-matrix, we are able to obtain explicit expressions for all the elements of $G$. Let $g_{i,j}$ ($i,j=0,1,\dots,c$) denote the $(i,j)$ element of $G$. Comparing the element $(0,0)$ in both sides of (\[equationforG:eq\]) yields, $$-(\lambda + c\alpha) g_{0,0} + \lambda g_{0,0}^2 = 0.$$ Since $0 \leq g_{0,0} \leq 1$, we obtain $g_{0,0} = 0$. Equating the $(i,i)$ ($i = 1,2,\dots,c-1$) elements in both sides of (\[equationforG:eq\]), we obtain $$i\mu - (\lambda + (c-i) \alpha + i\mu) g_{i,i} + \lambda g_{i,i}^2 = 0, \qquad i = 1,2,\dots,c-1.$$ Combining with the condition that $0 \leq g_{i,i} \leq 1$, we obtain $$g_{i,i} = \frac{ \lambda + i \mu + (c-i) \alpha - \sqrt{ (\lambda + i \mu + (c-i) \alpha)^2 - 4 i \lambda \mu } }{2 \lambda},$$ which is identical to $z_i$. Finally, comparing the $(c,c)$ elements in both sides of (\[equationforG:eq\]) we obtain $$c\mu -(\lambda + c\mu )g_{c,c} + \lambda g_{c,c}^2 = 0,$$ which has two roots $1$ and $\lambda/(c\mu)$. Because $g_{c,c}$ is the minimal solution of this equation, we have $g_{c,c} = \lambda/(c\mu)$. We have obtained all the diagonal elements of the $G$-matrix. Using the same manner as for $R$-matrix, we also recursively obtain the upper diagonal elements. First, we obtain the upper diagonal elements $g_{i,i+1}$ ($i=0,1,\dots,c-1$). Indeed, comparing the elements $(i,i+1)$ in both sides of (\[equationforG:eq\]), we obtain $$-q_i g_{i,i+1} + (c-i)\alpha g_{i+1,i+1} + \lambda (g_{i,i} g_{i,i+1} + g_{i,i+1} g_{i+1,i+1}) = 0,$$ leading to $$g_{i,i+1} = \frac{(c-i)\alpha g_{i+1,i+1}}{\lambda + (c-i)\alpha + i \mu - \lambda (g_{i,i} + g_{i+1,i+1})}, \qquad i = 0,1,\dots,c-1.$$ It should be noted that the quantities in the left hand side are given. Furthermore, comparing elements $(i,j)$ in both sides of (\[equationforG:eq\]) and rearranging the result, we obtain $$g_{i,j} = \frac{ (c-i) \alpha + \lambda \sum_{k=i+1}^{j-1} g_{i,k} g_{k,j} }{ \lambda + (c-i)\alpha + i\mu - \lambda (g_{i,i}+g_{j,j})}, \qquad i + 1 < j \leq c.$$
Once $G$ is given, we obtain other $G^{(n)}$ ($n = 1,2,\dots,c$) matrices using the following backward formula. $$G^{(n)} = \left( -Q^{(n)}_0 - Q^{(n)}_1 G^{(n+1)} \right)^{-1} Q^{(n)}_{-1} , \qquad n = c,c-1,\dots,1.$$
Comparison of Several Approaches {#comparison_approach:sec}
================================
In this section, we present a comparison between several approaches that can be used to solve our M/M/c/Setup model.
We observe that the generating function approach and the matrix analytic method are equivalent in the following sense. Indeed, the homogeneous part in the QBD formulation corresponds to $\widehat{\Pi}_i (z)$ ($i = 0,1,\dots,c$) in the generating function approach. The non-homogeneous part in the matrix analytic method corresponds to the boundary part, i.e., $\{ (i,j); j = i = 0,1,\dots, c, i \leq j \leq c \}$ in the generating function approach. The advantage of the matrix analytic method is that it directly implies a recursive formula for computing the rate matrix. In our case, the generating function approach yields the exact closed form solution for the joint stationary distribution.
In general, in case the stationary distribution is exactly obtainable, generating function gives detailed information of the model. On the other hand, when such an analytical solution does not exit, matrix analytic approach provides a unify approach for numerical calculation.
The matrix analytic method here shares many spirits with the recursive renewal approach. In particular, both methods are based on probabilistic arguments. For example, the quantity $p^L_{i \to d}$ in [@Gandhi13; @Gandhi14] is identical to $g_{i,d}$ in Section \[homo\_part\_gmatrix\]. It should be noted that matrix $R$ could be obtained from matrix $G$. From this point of view, the matrix analytic method and the recursive renewal approach are equivalent. The difference in both approaches is that while the matrix analytic method aims at a direct computation of the queue length distribution, the recursive renewal reward approach could be used to obtain any quantity of interest such as the generating function of the queue length.
Van Houdt and Leeuwaarden [@Benny_Johan11] analyze a more general models, i.e., M/G/1-type and GI/M/1-type Markov chains. In [@Benny_Johan11] there is only one boundary level, i.e. level 0 and thus the focus is put on the explicit expression for the $G$-matrix (or $R$-matrix). In principle, our model falls to the framework of [@Benny_Johan11] by considering the non-homogeneous part as a single macro level. However, if we do so, the computational complexity in the boundary is dominant (i.e. order of $O(c^6)$) while the complexity of matrix $G$ or $R$ is only $O(c^2)$. It should be noted that special structure of non-homogeneous part is not taken into account in [@Gandhi13; @Gandhi14], thus the computational complexity is also $O(c^6)$.
Some Variant Models
===================
In [@Gandhi13], some variants of the M/M/$c$/Setup queue are presented and analyzed. The first variant model is the M/M/$c$/Setup/Sleep where a set of $s \leq c$ servers is set to “sleep" when idle whereas the rest $c-s$ servers are turned off when idle. The characteristic of the sleep state is that it takes a shorter setup time than the off state. The second variant is the M/M/$c$/Setup/Delayoff where a server stays idle for a while after completing a service but not yet having a job to serve. We confirm that the non-homogeneous part (the number of jobs in the system is greater than $c$) has the same structure with that of the M/M/$c$/Setup queue in this paper. In comparison with the original model, the boundary part M/M/$c$/Setup/Sleep has the same structure while that of M/M/$c$/Setup/Delayoff is different. The QBD formulation allows to obtain explicit rate matrix for the homogeneous part for both models using which we can recursively obtain the stationary distribution. The generating function approach in this paper can be applied to the M/M/$c$/Setup/Sleep directly while some further modifications are needed for the M/M/$c$/Setup/Delayoff model.
Performance Measures and Numerical Examples {#numerical:sec}
===========================================
Performance measures
--------------------
Let $\pi_i$ denote the stationary probability that there are $i$ active servers, i.e., $\pi_i = \sum_{j = i}^\infty \pi_{i,j}$. Let $\mathbb{E} [A] $ and $\mathbb{E} [S] $ denote the mean number of active servers and that in setup mode, respectively. We have $$\mathbb{E} [A] = \sum_{i = 1}^c i \pi_i , \qquad \mathbb{E} [S] = \sum_{i = 0}^c \sum_{j = i}^\infty \min(j-i, c-i) \pi_{i,j}.$$ Let $\mathbb{E} [S_r]$ denote the switching rate from OFF to ON in the steady state (mean number of switches from OFF to ON per unit time). We then have $$\begin{aligned}
\mathbb{E} [S_r] & = \sum_{i=0}^{c-1} \sum_{j = i+1}^\infty \min(c-i,j-i) \alpha \pi_{i,j} = \sum_{i = 1}^c i \mu \pi_{i,i}, \end{aligned}$$ where the second equality is due to the fact that the switching rate from OFF to on is equal to that from ON to OFF in the steady state. Furthermore, let $\mathbb{E} [L]$ denote the mean number of jobs in the systems, i.e., $$\mathbb{E} [L] = \sum_{j=0}^\infty \vc{\pi}_j \vc{e} j,$$ where $\vc{\pi}_j \vc{e}$ is the probability that there are $j$ customers in the system. We define a cost function for the model. $$Cost_{on-off} = C_a \mathbb{E} [A] + C_s \mathbb{E} [S].$$ where $C_a$ and $C_s$ are the cost per time unit for an active server and a server in setup mode, respectively.
For comparison, we also define the cost of the corresponding ON-IDLE model, i.e., M/M/$c$ without setup times. It is easy to see that the power consumption for this model is given as follows. $$Cost_{on-idle} = c \rho C_a + c (1 - \rho) C_i,$$ where $C_i$ is the power consumption of an idle server.
If each time of turning ON and turning OFF a server needs a cost of $C_{sw}$, we could also consider the following cost function [@Maccio13]. $$TotalCost_{on-off} = C_a \mathbb{E} [A] + C_s \mathbb{E} [S] + C_{sw} \mathbb{E} [S_r].$$
Numerical examples
------------------
In this section, we show some numerical examples. It should be noted that some of them are also presented in [@Gandhi10; @Gandhi13; @Gandhi14]. The numerical results are presented to show the feasibility of our computational procedure. Furthermore, we complement numerical results in [@Gandhi10; @Gandhi13; @Gandhi14] by taking the switching rate between ON and OFF into account.
In all the numerical examples, we fix $\mu = 1$, $C_a = C_s = 1$ and $C_i = 0.6 C_a$. The evidence for $C_i = 0.6 C_a$ is that an idle server still consumes about 60% of its peak processing a job [@Barroso07]. We will investigate the cost function with respect to the setup cost $C_s$ in Section \[cost\_vs\_setup\_cost:sec\].
All the numerical results in this section are obtained using the matrix analytic method presented in Section \[matrix\_ana:sec\]. The same numerical results can be also obtained using the procedure presented in Section \[components:sec\].
### Effect of the setup rate {#setup_rate:subsec}
Section \[setup\_rate:subsec\] investigates the effect of the setup rate on the power consumption ($Cost_{on-off}$, $Cost_{on-idle}$) and the mean number of jobs in the system. Figures \[Power\_consump\_vs\_nu\_c20:fig\] and \[Power\_consump\_vs\_nu\_c30:fig\] represent the power consumption against the setup rate for the case $c=20$ and 30, respectively. We observe that the power consumption decreases as the setup rate increases. For comparison, we also plot the power consumption for the corresponding M/M/$c$ model without setup times. We find that there exists some $\alpha_{\rho, c}$ such that the ON-OFF policy outperforms the ON-IDLE policy for $\alpha > \alpha_{\rho, c}$ while the latter is more power-saving for the case $\alpha < \alpha_{\rho, c}$. Furthermore, $\alpha_{\rho, c}$ increases as $\rho$ increases.
Figures \[Power\_consump\_vs\_nu\_switching\_cost\_c20:fig\] and \[Power\_consump\_vs\_nu\_switching\_cost\_c30:fig\] investigate the total energy consumption taking into account the switching cost, i.e., $TotalCost_{on-off}$ ($C_{sw} = 1$) against the setup rate $\alpha$ for $\rho = 0.3, 0.5$ and 0.7. We observe that the total power consumption does not always monotonically decreases as the setup rate increases as in Figures \[Power\_consump\_vs\_nu\_c20:fig\] and \[Power\_consump\_vs\_nu\_c30:fig\]. This is because when the setup rate $\alpha$ is large the number of switches per time unit increases leading to the increase in the cost function. We observe in the curves of $\rho = 0.5$ that there exist two points $\alpha_{min} $ and $\alpha_{max} $ such that the ON-IDLE policy outperforms the ON-OFF policy for $\alpha < \alpha_{min}$ and $\alpha > \alpha_{max}$. An interesting observation is that three curves for $\rho = 0.3, 0.5$ and 0.7 are the same when the setup rate is extremely low. The reason is that all the servers are in setup mode for almost the time when the setup time is extremely long.
[cc]{}
![Power consumption vs. $\alpha$ ($c=30$)[]{data-label="Power_consump_vs_nu_c30:fig"}](Power_consump_vs_nu_c20.eps)
![Power consumption vs. $\alpha$ ($c=30$)[]{data-label="Power_consump_vs_nu_c30:fig"}](Power_consump_vs_nu_c30.eps)
[cc]{}
![Total Pow. consump. vs. $\alpha$ ($c=30$)[]{data-label="Power_consump_vs_nu_switching_cost_c30:fig"}](Power_consump_vs_nu_switching_cost_c20_revised.eps)
![Total Pow. consump. vs. $\alpha$ ($c=30$)[]{data-label="Power_consump_vs_nu_switching_cost_c30:fig"}](Power_consump_vs_nu_switching_cost_c30_revised.eps)
Figures \[System\_num\_vs\_nu\_c10:fig\] and \[System\_num\_vs\_nu\_c30:fig\] represent the mean number of jobs in the system ($\mathbb{E} [L]$) against the setup rate $\alpha$. We observe that $\mathbb{E} [L]$ decreases as the setup rate increases. We also observe that $\mathbb{E} [L]$ converges to that of the ON-IDLE model as $\alpha \to \infty$ which agrees with intuition.
[cc]{}
![$\mathbb{E} [L]$ vs. $\alpha$ ($c=30$)[]{data-label="System_num_vs_nu_c30:fig"}](System_num_vs_nu_c10.eps)
![$\mathbb{E} [L]$ vs. $\alpha$ ($c=30$)[]{data-label="System_num_vs_nu_c30:fig"}](System_num_vs_nu_c30.eps)
### Effect of the number of servers
In this subsection, we investigate the effect of the number of servers on the power consumption ($Cost_{on-off}$) while keeping the traffic intensity for each server, i.e., $\rho = \lambda/(c\mu)$ constant. Figures \[Power\_consump\_vs\_serversnum\_rho05:fig\] and \[Power\_consump\_vs\_serversnum\_rho07:fig\] represent the case $\rho = 0.5$ and $\rho = 0.7$, respectively. We observe in both figures that the ON-OFF policy is always more power-saving than the ON-IDLE policy for $\alpha = 1$ while the latter always outperforms the former for the case $\alpha = 0.01$. For the case $\alpha = 0.1$, we observe in Figure \[Power\_consump\_vs\_serversnum\_rho05:fig\] that there exists some $c_{\alpha = 0.1}$ such that the ON-OFF policy outperforms the ON-IDLE one for $c > c_{\alpha = 0.1}$ while the latter is more power-saving than the former for $c < c_{\alpha = 0.1}$. Thus, for $\alpha=0.1$ and $\rho = 0.5$, the ON-OFF policy is more effective than the ON-IDLE system if the scale of the system is large enough.
[cc]{}
![Power consumption vs. $c$ ($\rho = 0.7$).[]{data-label="Power_consump_vs_serversnum_rho07:fig"}](Power_consump_vs_serversnum_rho05.eps)
![Power consumption vs. $c$ ($\rho = 0.7$).[]{data-label="Power_consump_vs_serversnum_rho07:fig"}](Power_consump_vs_serversnum_rho07.eps)
### Effect of traffic intensity
In this section, we show the effect of the traffic intensity on the power consumption ($Cost_{on-off}, Cost_{on-idle},$) for the cases $c=20$ and $c=50$ in Figure \[Power\_consump\_vs\_traffic\_intensity\_c20:fig\] and Figure \[Power\_consump\_vs\_traffic\_intensity\_c50:fig\], respectively. In each figure, we plot three curves with $\alpha = 1, 0.1$ and 0.01. For comparison, we also plot the power consumption for the corresponding model without setup times. We observe in both figures that the ON-OFF policy with $\alpha = 1$ always outperforms that of ON-IDLE policy. However, for the cases $\alpha = 0.1$ and 0.01, we observe that there exists some $\rho_\alpha$ for which the ON-OFF policy outperforms the ON-IDLE one for $\rho < \rho_\alpha$ while the latter is more power-saving than the former for the case $\rho > \rho_\alpha$.
[cc]{}
![Power consumption vs. $\rho$ ($c = 50$).[]{data-label="Power_consump_vs_traffic_intensity_c50:fig"}](Power_consump_vs_traffic_intensity_c20.eps)
![Power consumption vs. $\rho$ ($c = 50$).[]{data-label="Power_consump_vs_traffic_intensity_c50:fig"}](Power_consump_vs_traffic_intensity_c50.eps)
### Effect of the setup cost {#cost_vs_setup_cost:sec}
Figure \[pefor\_vs\_SetupCostofOneSwitch:fig\] show the sensitivity of the cost of a setting up server on the power consumption $Cost_{on-off}$ where $C_a = 1$. Letting $r = C_s/C_a$, we observe that there exists some $r_{\rho}$ such that the ON-IDLE policy outperforms the ON-OFF policy for $r > r_\rho$ while former outperforms the latter for the case $r < r_\rho$. We also observe that $r_\rho$ decreases with the increase of $\rho$. This agrees with intuition.
Figure \[Total\_Power\_consump\_vs\_traffic\_intensity\_c50:fig\] represents the total power consumption ($TotalCost_{on-off}$ with $C_{sw} = 1$) against the traffic intensity. We observe in the curves of $\alpha = 0.01, 0.1$ and 1 that the total power consumption monotonically increases as the traffic intensity increases. Interestingly, we observe that for the case $\alpha = 10$ and 100, the total power consumption increases as $\rho$ increases (for a relatively small $\rho$) and then decreases as $\rho$ increases (for a relatively large $\rho$). At the first glance, it may not be intuitive that the total power consumption decreases with the increase in $\rho$. However, this is due to the relation of $\mathbb{E} [S_r]$ and $\rho$ which will be investigated in detail in Figure \[Number\_of\_Switches\_c4050:fig\].
[cc]{}
![Total Pow. consump. vs. $\rho$ ($c = 50$).[]{data-label="Total_Power_consump_vs_traffic_intensity_c50:fig"}](pefor_vs_SetupCostofOneSwitch.eps)
![Total Pow. consump. vs. $\rho$ ($c = 50$).[]{data-label="Total_Power_consump_vs_traffic_intensity_c50:fig"}](Total_Power_consump_vs_traffic_intensity_c50.eps)
### Mean number of switches
In this section, we investigate the property of the switching rate $\mathbb{E} [S_r]$, i.e., the mean number of switches per a time unit. In particular, Figure \[Number\_of\_Switches\_c4050:fig\] shows the switching rate against the traffic intensity. We observe that the switching rate increases with the traffic intensity under a light traffic regime while it decreases with $\rho$ in relatively heavy traffic regime. The reason is as follows. Almost all the servers are OFF in light traffic regime while a large percent of servers are ON in heavy traffic. Thus, in light traffic regime, increasing the traffic intensity implies the increase in the number of switches from OFF to ON. However, in heavy traffic regime almost all the servers are already ON. As a result, increasing the traffic intensity does not lead to further increase in the switching rate. This suggests that from the switching rate point of view, the ON-OFF policy is preferable in a relatively light traffic regime or a relatively heavy traffic one.
Figure \[Number\_of\_Switches\_vs\_serversnum\_rho07:fig\] shows the switching rate against the number of servers. We observe that the switching rate increases with the number of servers. Moreover, the curves for the case $\alpha = 0.1$ and $\alpha = 0.01$ are almost linear while that for the case $\alpha = 1$ is not linear.
[cc]{}
![Switching rate vs. the $c$ ($\rho = 0.7$).[]{data-label="Number_of_Switches_vs_serversnum_rho07:fig"}](Number_of_Switches_c4050.eps)
![Switching rate vs. the $c$ ($\rho = 0.7$).[]{data-label="Number_of_Switches_vs_serversnum_rho07:fig"}](Number_of_Switches_vs_serversnum_rho07.eps)
Conclusion and future works {#conclusion:sec}
===========================
In this paper, we have presented a detailed analysis for the M/M/$c$/Setup model with ON-OFF policy for data centers. Using a generating function approach, we have derived explicit solutions for the generating functions from which we have obtained recursive formulae for the factorial moments. The generating function approach yields a conditional decomposition for the queue length. We also have observed that the model belongs to a special QBD class where the rate matrix of the homogeneous part is explicitly obtained. The boundary part also possesses some special structure allowing us to obtain the joint stationary distribution with the complexity of $O(c^2)$ by generating function approach and $O(c^3)$ by the matrix analytic method. Our numerical results have provided some insights into the performance of the system. We have found the range of the parameters under which the ON-OFF policy outperforms the ON-IDLE policy. We have pointed out the equivalence between the two methodologies.
In real world data center, in order to reduce the waiting time, a fixed number of servers may be kept ON all the time. The extension of the current model to this case may be worth to investigate. Other extensions include a threshold policy which turns ON and OFF the servers according to the load of the system.
Acknowledgements {#acknowledgements .unnumbered}
================
The author would like to thank two anonymous referees and the associate editor whose comments helped to improve the presentation of the paper. The author would like to thank Professor Herwig Bruneel of Ghent University and Professor Onno Boxma of Eindhoven University of Technology for useful remarks on the conditional decomposition. This research was supported in part by Japan Society for the Promotion of Science, JSPS Grant-in-Aid for Young Scientists (B), Grant Number 2673001.
[99]{}
Artalejo, J. R., Economou, A. and Lopez-Herrero, M. J. (2005). Analysis of a multiserver queue with setup times. Queueing Systems, 51(1-2), 53-76.
Barroso, L. A. and Holzle, U. (2007). The case for energy-proportional computing. Computer, 40 (12), 33-37.
Bischof W. (2001). Analysis of M/G/1-Queues with Setup Times and Vacations under Six Different Service Disciplines, Queueing Systems: Theory and Applications, 39(4), 265-301, 2001.
Bright, L.W. and Taylor P.G. (1995). Calculating the equilibrium distribution in level dependent quasi-birth-and-death processes. Stochastic Models, 11, 497–525.
Burke, P. J. (1975). Delays in single-server queues with batch input. Operations Research, 23, 830-833.
Choudhury, G. (1998). On a batch arrival Poisson queue with a random setup time and vacation period, Computers and Operations Research, 25(12), 1013-1026, 1998.
Choudhury, G. (2000). An MX/G/1 queueing system with a setup period and a vacation period, Queueing Systems: Theory and Applications, 36(1/3), 23-38, 2000.
Fuhrmann, S. W. and Cooper, R. B. (1985). Stochastic decompositions in the M/G/1 queue with generalized vacations. Operations research, 33(5), 1117-1129.
, 153-166, ACM.
Queueing Systems, 77(2), 177-209.
.
Mitrani, I. (2013). Managing performance and power consumption in a server farm. Annals of Operations Research 202(1), 121-134, 2013.
Neuts M.F.(1981), Matrix-Geometric Solutions in Stochastic Models: An Algorithmic Approach (Johns Hopkins University Press, Baltimore, 1981.
Perel, N. and Yechiali, U. (2013). The Israeli Queue with Priorities. Stochastic Models, 29, 353–379.
Levy, Y. and Yechiali, U. (1976). An M/M/s queue with servers’vacations. Infor, 14(2), 153-163.
Phung-Duc, T, Masuyama, H., Kasahara, S., Takahashi, Y. (2010). A simple algorithm for the rate matrices of level-dependent QBD processes, in [Proceedings of QTNA2010, Beijing, China]{}, 46–52. ACM New York, USA.
Phung-Duc, T. (2014). Impatient customers in power-saving data centers. Lecture Notes in Computer Science, LNCS 8499, 185–199.
Phung-Duc, T. (2015). Multiserver Queues with Finite Capacity and Setup Time. Lecture Notes in Computer Science, LNCS 9081, 173-187.
Ramaswami, V. and Taylor, P.G. (1996). Some properties of the rate operations in level dependent quasi-birth-and-death processes with countable number of phases. [Stochastic Models]{}, 12, 143–164.
Servi, L. D., Finn, S. G. (2002). M/M/1 queues with working vacations (M/M/1/WV). Performance Evaluation, 50(1), 41-52.
Schwartz, C., Pries, R. and Tran-Gia, P. (2012). A queuing analysis of an energy-saving mechanism in data centers. In Proceedings of IEEE 2012 International Conference on Information Networking (ICOIN), 70-75.
Takagi, H. (1990). Priority queues with setup times, Operations Research, 38(4), 667-677.
Tian N., Li Q. L. and Gao J. (1999). Conditional stochastic decompositions in the M/M/$c$ queue with server vacations. Stochastic Models, 15, 367-377.
Zhang, Z. G. and Tian N. (2003). Analysis of queueing systems with synchronous single vacation for some servers. Queueing systems, 45, 161-175.
Tian, N. and Zhang, Z. G. (2006). Vacation Queueing Models: Theory and Applications, Springer.
Van Houdt B. and van Leeuwaarden J. S.H. (2011), Triangular M/G/1-type and tree-like QBD Markov chains, INFORMS Journal on Computing, 23(1), 165-171, 2011.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
It is shown that the addition of a topological invariant (Gauss-Bonnet term) to the anti-de Sitter (AdS) gravity action in four dimensions recovers the standard regularization given by holographic renormalization procedure. This crucial step makes possible the inclusion of an odd parity invariant (Pontryagin term) whose coupling is fixed by demanding an asymptotic (anti) self-dual condition on the Weyl tensor. This argument allows to find the dual point of the theory where the holographic stress tensor is related to the boundary Cotton tensor as $T_{j}^{i}=\pm (\ell ^{2}/8\pi G)C_{j}^{i}$, which has been observed in recent literature in solitonic solutions and hydrodynamic models.
A general procedure to generate the counterterm series for AdS gravity in any even dimension from the corresponding Euler term is also briefly discussed.
author:
- Olivera Mišković
- Rodrigo Olea
title: |
**Topological regularization and self-duality\
in four-dimensional anti-de Sitter gravity**
---
Introduction
============
In the gravity side of the anti-de Sitter/Conformal Field Theory (AdS/CFT) correspondence, the relevant information that realizes this duality is encoded in the finite part of the boundary stress tensor [@Maldacena]. That identification requires the cancelation of the infrared divergences in the bulk theory made by holographic renormalization procedure [@dHSS], which is based on an asymptotic analysis of the metric in Fefferman-Graham (FG) coordinate system [@Fefferman-Graham] $$ds^2=\frac{\ell^2}{4\rho^2}\,d\rho^2+\frac{1}{\rho}\,g_{ij}\,dx^{i}dx^{j}, \label{FG}$$where $h_{ij}=g_{ij}/\rho$ corresponds to the boundary metric. For asymptotically AdS (AAdS) spaces, $g_{ij}(x,\rho )$ accepts a regular expansion near the boundary $\rho =0$ , i.e., $ g_{ij}(x,\rho )=g_{(0)ij}+\rho g_{(1)ij}+\cdots $. Solving the Einstein equations in this frame leads to the holographic reconstruction of the spacetime from a given boundary data $g_{(0)ij}$, what is essential to determine the series of intrinsic counterterms ${\cal L}_{ct}$ which renders finite the boundary stress tensor [@Ba-Kr].
However, the algorithm which produces ${\cal L}_{ct}$ becomes extremely complex as the spacetime dimension increases, such that there is no a closed formula for counterterms for an arbitrary dimension. This argument motivates the search for alternative approaches.
On the other hand, any other regularization scheme, even if properly removes the asymptotic divergences, might spoil the holographic interpretation of the theory within the AdS/CFT framework because of different boundary conditions.
In particular, a regularization mechanism for AdS gravity in any dimension which consists in the addition of counterterms that depend on the extrinsic curvature $K_{ij}$ (Kounterterms method) has been recently proposed [@OleaJHEP; @OleaKT]. In this case, the on-shell variation of the regularized action $I_{reg}$ contains terms of the type $\delta K_{ij}$, what makes a definition of quasilocal stress tensor more elusive. But one knows that in AAdS spacetimes the leading order of the asymptotic expansion in $K_{ij}$ coincides with the leading order of the induced metric $h_{ij}$ , i.e., $$K_{ij}=\frac{1}{\ell} \frac{g_{(0)ij}}{\rho}+{\cal O}(\rho).$$ The above relation inspires a reformulation of holographic renormalization in terms of an expansion of the extrinsic curvature [@Papadimitriou-Skenderis]. This suggests it might be still possible to obtain a regularized stress tensor $\langle
T_{ij} \rangle$ associated to $g_{(0)ij}$ even though Kounterterms regularization does not lend itself to a Brown-York stress tensor definition $T_{ij}={2\over\sqrt{-h}}{\delta I_{reg}\over\delta h^{ij}}$. It also motivates a direct comparison with the standard procedure what, until now, has been performed in Einstein gravity only in three dimensions [@Miskovic-Olea]. For four and higher even dimensions, this is carried out below.
Gauss-Bonnet invariant in 4-dimensional AdS gravity
===================================================
Let us consider the Einstein-Hilbert action with negative cosmological constant in four dimensions supplemented by the Gauss-Bonnet (GB) term $\mathcal{E}_{4}$ with an arbitrary coupling constant $\alpha $$$I=\int\limits_{M}d^{4}x \sqrt{-{\cal G}}\left[\frac{1}{16\pi
G}\left( R-2\Lambda \right) +\alpha (R_{\mu \nu \alpha \beta }R^{\mu
\nu \alpha \beta }-4R_{\mu \nu }R^{\mu \nu }+R^{2})\right],
\label{I4}$$where $\Lambda =-3/\ell ^{2}$ is the cosmological constant in terms of the AdS radius $\ell $. It was shown in ref.[@ACOTZ4] that a well-posed action principle for gravity with AdS asymptotics removes the arbitrariness in the GB coupling. Since $\mathcal{E}_{4}$ is a topological invariant, it does not modify the field equations. However, it still contributes to the surface term when the total action is varied $$\delta I=\int\limits_{M}EOM+\int\limits_{\partial M}d^{3}x\,%
\sqrt{-h}\,n_{\sigma }\delta _{\lbrack \gamma \delta \alpha \beta
]}^{[\sigma \lambda \mu \nu ]}\mathcal{G}^{\delta \varepsilon
}\delta \Gamma _{\lambda \varepsilon }^{\gamma } \left(
\frac{1}{64\pi G}\delta _{\lbrack \mu \nu ]}^{[\alpha \beta
]}+\alpha R_{\mu \nu }^{\alpha \beta }\right) , \label{deltaI4}$$where $n_{\mu }$ is the normal vector to the boundary [@delta]. The total action is rendered stationary demanding $\delta I=0$ on-shell for a given boundary condition. For asymptotically locally AdS spacetimes, i.e., $R_{\mu \nu }^{\alpha \beta }+\frac{1}{\ell ^{2}}\delta _{\lbrack
\mu \nu ]}^{[\alpha \beta ]}=0$ at $\partial M$, the variational principle fixes the coupling constant as $%
\alpha =\ell ^{2}/(64\pi G)$, which produces finite Noether charges [@ACOTZ4]. Surprisingly enough, the same value of $%
\alpha $ regularizes the Euclidean action in a background-independent way [OleaJHEP]{} and casts eq.(\[I4\]) into MacDowell-Mansouri form [Mac-Man]{}$$I_{4}=\frac{\ell ^{2}}{256\pi G}\int\limits_{M}d^{4}x\,\sqrt{-%
\mathcal{G}}\,\delta _{\lbrack \gamma \delta \alpha \beta ]}^{[\sigma
\lambda
\mu \nu ]}\left( R_{\sigma \lambda }^{\gamma \delta }+\frac{1}{\ell ^{2}}%
\,\delta _{\lbrack \sigma \lambda ]}^{[\gamma \delta ]}\right)
\left( R_{\mu \nu }^{\alpha \beta }+\frac{1}{\ell
^{2}}\,\delta _{\lbrack \mu \nu ]}^{[\alpha \beta ]}\right) .
\label{MacMan}$$ Using the field equations, one proves that the Weyl tensor is $$W_{\mu \nu }^{\alpha \beta }=R_{\mu \nu }^{\alpha \beta }+\frac{1}{\ell ^{2}}%
\,\delta _{\lbrack \mu \nu ]}^{[\alpha \beta ]}, \label{weyl4}$$where the r.h.s. is the curvature of the AdS group (the rest corresponds to the torsion, which vanishes in Riemann gravity). This fact implies that the action (\[MacMan\]) is on-shell equivalent to conformal gravity $$I_{4}=\frac{\ell ^{2}}{64\pi G}\int\limits_{ M}d^{4}x\,\sqrt{-%
\mathcal{G}}\,W_{\mu \nu \alpha \beta }W^{\mu \nu \alpha \beta },
\label{IregWeyl}$$because any trace of $W_{\mu \nu \alpha \beta }$ is identically zero [@MPT].
In what follows, we show that the addition of a topological invariant of the Euler class recovers the standard counterterm regularization and holographic stress tensor, by considering its equivalent boundary formulation.
Boundary formulation
=====================
In a four-dimensional manifold without boundaries, the integration of the GB term is proportional to the Euler characteristic $\chi
(M)$. When a boundary is introduced, a correction to $\chi (M)$ is required, such that the Euler theorem reads $$\int\limits_{M}d^{4}x\,\mathcal{E}_{4}=32\pi ^{2}\chi
(M)+\int\limits_{\partial M}d^{3}x\,B_{3}, \label{EulerTh}$$where $B_{3}$ is a boundary term known as Second Chern Form. If the spacetime is foliated using Gaussian (radial) coordinates $ds^{2}=N^{2}(\rho )d\rho ^{2}+h_{ij}(\rho ,x)dx^{i}dx^{j}$, the term $B_{3}$ is given as a polynomial in the extrinsic curvature $K_{ij}=-\frac{%
1}{2N}\,\partial _{\rho }h_{ij}$ and the intrinsic curvature $\mathcal{R}%
_{kl}^{ij}(h)$ as [@OleaJHEP]$$B_{3}=4\sqrt{-h}\ \delta _{\left[ j_{1}j_{2}j_{3}\right] }^{\left[
i_{1}i_{2}i_{3}\right] }K_{i_{1}}^{j_{1}}\left( \frac{1}{2}\,\mathcal{R}%
_{i_{2}i_{3}}^{j_{2}j_{3}}(h)-\frac{1}{3}\,K_{i_{2}}^{j_{2}}K_{i_{3}}^{j_{3}}%
\right) . \label{B3KR}$$There is a reason why to consider the boundary formulation of topological invariants beyond the purpose of comparison with the counterterm regularization. The boundary dynamics does not tell between the Euler and boundary term $B_{3}$, as they are locally equivalent. However, computations of the Euclidean action show that the Euler term shifts the black hole entropy $S$ by a constant proportional to $\chi (M)$ [@OleaJHEP], what can also be obtained using Wald’s entropy formula. Thus, $S$ may take negative values for topological black holes with hyperbolic spatial section, what can only be avoided by supplementing the action with the Kounterterm $B_{3}$ instead.
In order to compare to the standard regularization procedure, one can simply add and subtract the Gibbons-Hawking term from the Einstein-Hilbert action plus the boundary term $B_{3}$, $$I_{4}=I_{EH}-\frac{1}{8\pi G}\int\limits_{\partial M}d^{3}x\sqrt{-h}%
\,K+\int\limits_{\partial M}d^{3}x\,\mathcal{L}_{ct}.
\label{addandsubGH}$$The first two terms define the Dirichlet problem in gravity, while the quantity $\mathcal{L}_{ct}$ is given by $$\mathcal{L}_{ct}=\frac{\ell ^{2}}{16\pi G}\sqrt{-h}\delta _{\left[
j_{1}j_{2}j_{3}\right] }^{\left[ i_{1}i_{2}i_{3}\right]
}K_{i_{1}}^{j_{1}} \left( \frac{1}{2}\,\mathcal{R}_{i_{2}i_{3}}^{j_{2}j_{3}}(h)-\frac{1}{3}\,%
K_{i_{2}}^{j_{2}}K_{i_{3}}^{j_{3}}+\frac{1}{\ell ^{2}}\,\delta
_{i_{2}}^{j_{2}}\delta _{i_{3}}^{j_{3}}\right) . \label{Lct4K}$$For the boundary metric $h_{ij}=g_{ij}/\rho$, the intrinsic curvature and the determinant rescale as $\mathcal{R}%
_{kl}^{ij}(h)=\rho \mathcal{R}_{kl}^{ij}(g)$ and $\sqrt{-h}=\sqrt{-g}/\rho ^{3/2}$, respectively. This also implies $$K_{i}^{j}=K_{ik}h^{kj}= \frac{1}{\ell }\left(\delta _{i}^{j}-\rho
k_{i}^{j}\right)$$for the extrinsic curvature, with the definition $k_{i}^{j}=g^{jk}\partial _{\rho }g_{ki}$. Expanding eq.(\[Lct4K\]) in FG form, one notices that $k_{j}^{i}$ is absent from the divergent terms, $$\mathcal{L}_{ct}=\frac{1}{8\pi G}\frac{\sqrt{-g}}{\rho ^{3/2}}\left( \frac{2%
}{\ell }+\frac{\ell }{2}\,\rho \mathcal{R}(g)\right) +\mathcal{O}(\rho
^{1/2}) \label{Lct}$$ such that one recovers the Balasubramanian-Kraus local counterterms$$\mathcal{L}_{ct}=\frac{1}{8\pi G}\sqrt{-h}\left( \frac{2}{\ell
}+\frac{\ell }{2}\mathcal{R}(h)\right) . \label{LctBK}$$The agreement with the standard holographic renormalization can be also seen from the on-shell variation of the action (\[deltaI4\]), which for the radial foliation and the value $\alpha =\ell ^{2}/(64\pi G)$ adopts the form$$\delta I_{4}=\frac{\ell^{2}}{32\pi G}\int\limits_{\partial M}d^{3}x\sqrt{-h}%
\delta _{\left[ mnp\right] }^{\left[ jkl\right] } \left( \delta K_{j}^{m}+%
\frac{1}{2}K_{i}^{m}\left( h^{-1}\delta h\right) _{j}^{i}\right)
\left( R_{i_{2}i_{3}}^{j_{2}j_{3}}+\frac{1}{\ell ^{2}}\delta
_{\lbrack i_{2}i_{3}]}^{[j_{2}j_{3}]}\right) . \label{varIT}$$Expanding the fields in the FG frame, the first term vanishes at the boundary, whereas the second gives a stress tensor$$\tau _{i}^{j}=\frac{\ell^{2}}{32\pi G}\,\delta _{\left[ mnp\right] }^{\left[ jkl%
\right] }K_{i}^{m}\left( R_{kl}^{np}+\frac{1}{\ell ^{2}}\delta
_{\lbrack kl]}^{[np]}\right). \label{stressK}$$Note that any conserved quantity constructed with this stress tensor will vanish for spacetimes which are globally of constant curvature, as AdS vacuum. Using Gauss-Codazzi relations, one might also notice that $\tau _{i}^{j}$ contains higher powers in the extrinsic curvature. However, it is straightforward to prove that $\tau _{i}^{j}$ coincides up to the relevant order in $\rho $ with the Balasubramanian-Kraus stress tensor $T_{i}^{j}$, when it is appropriately rewritten $$\begin{aligned}
T_{i}^{j} &=& \frac{1}{8\pi G}\left( K_{i}^{j}-\delta _{i}^{j}K+\frac{2}{\ell }%
\delta _{i}^{j}-\left( \mathcal{R}_{i}^{j}(h)-\frac{1}{2}\delta _{i}^{j}%
\mathcal{R}(h)\right) \right) \nonumber \\
&=& \frac{\rho \ell }{32\pi G}\,\delta _{\left[ inp\right] }^{\left[ jkl%
\right] }\left( \mathcal{R}_{kl}^{np}(g)+\frac{4}{\ell^{2}}\delta
_{k}^{n}k_{l}^{p}\right). \label{BKtensor}\end{aligned}$$ The above derivation shows that the divergence cancelation provided by the counterterm series can be regarded as a [*topological*]{} regularization, since it comes from the addition of the GB term with a coupling such that the regularized action takes the MacDowell-Mansouri form (\[MacMan\]).
In the holographic renormalization framework, the information on the holographic stress tensor in four dimensions is carried by the coefficient $g_{(3)}$ in FG expansion $$g_{ij}(x,\rho )=g_{(0)ij}+\rho g_{(1)ij}+\rho^{3/2} g_{(3)ij}+\cdots\,.$$ It is just after solving the Einstein equations order by order in $\rho$ that the vanishing of the Weyl anomaly comes from the zero trace of $g_{(3)}$ [@dHSS]. On the other hand, the anomaly $\cal{A}$ can be also read off from a Weyl transformation with infinitesimal parameter $\sigma$ on the regularized action, that is, $\delta_{\sigma}I_{reg}=\int_{\partial M} d^{D-1}x \sqrt{g_{(0)}}\,
\sigma\cal{A} $. This means that one might have concluded the same by simple inspection of the eq.(\[IregWeyl\]), since it is manifestly invariant under conformal transformations.
Up to the relevant order, the stress tensor (\[stressK\]) can be rewritten as $$\tau _{i}^{j} =\frac{\ell }{8\pi G}W_{\,\mu i\nu }^{j}\,n^{\mu
}n^{\nu },$$ using the traceless property and index symmetries of the Weyl tensor.
The conformal completion technique [Ashtekar-Magnon-Das]{} defines an AAdS spacetime in such a way that the metric ${\cal G}_{\mu \nu }$ which obeys the Einstein equations can be conformally mapped into an *unphysical* one ${\cal
\tilde{G}}_{\mu \nu}= \Omega^{-2}{\cal
{G}}_{\mu \nu}$ by a smooth conformal factor $\Omega$ which satisfies precise fall-off conditions. The procedure gives rise to a background-independent conserved charge for every asymptotic symmetry $\xi ^{i}$ as the integral on the spatial section $\tilde{\Sigma}$ of the boundary $${\cal H}_{\xi }=\frac{\ell }{8\pi G}\int\limits_{\tilde{\Sigma}}\tilde{E}%
_{i}^{j}\xi ^{i}\tilde{u}_{j}\,d\tilde{\Sigma}, \label{AMD}$$ where $\tilde{E}_{i}^{j}=\Omega ^{3-D}\tilde{W}_{\,\mu i\nu }^{j}\,%
\tilde{n}^{\mu }\tilde{n}^{\nu }/(D-3)$ is the [*electric*]{} part of the unphysical Weyl tensor, $d\tilde{\Sigma}$ is the integration element on $\tilde{\Sigma}$ and $\tilde{u}_{j}$ is the unit timelike normal to $\tilde{\Sigma}$. Rescaling all the quantities into the ones of the spacetime metric, it is easy to prove that the conserved quantities $Q_{\xi }\equiv\int_{\Sigma }\tau
_{i}^{j}\xi ^{i}u_{j}\,d\Sigma$ coming from Kounterterms regularization in $D=4$ are the same as the Ashtekar-Magnon-Das formula (\[AMD\]).
Pontryagin term and self-dual solutions
=======================================
In four dimensions there exists an additional (odd parity) topological invariant known as Pontryagin term $\mathcal{P}_{4}$, which is locally equivalent to the derivative of the gravitational Chern-Simons term $$\mathcal{P}_{4}=-\frac{1}{4}\epsilon ^{\mu \nu \alpha \beta }R_{\mu
\nu }^{\sigma \lambda }R_{\sigma \lambda \alpha \beta }=
\epsilon^{\mu \nu \alpha \beta }\partial_{\mu}\left( \Gamma
_{\nu\lambda }^{\sigma}\partial_{\alpha}\Gamma _{\beta\sigma
}^{\lambda }+\frac{2}{3}\Gamma _{\nu\lambda }^{\sigma}\Gamma
_{\alpha\varepsilon }^{\lambda }\Gamma _{\beta \sigma }^{\varepsilon
}\right), \label{pontryagin}$$where $\epsilon ^{\mu \nu \alpha \beta }$ is the constant Levi-Civita tensor density.
The Pontryagin term $F\wedge F$ in four-dimensional Maxwell electromagnetism modifies the dynamics such that the Lorentz boost and the parity invariance are lost when it is coupled through an external, fixed quantity.
We will consider here the addition of the Pontryagin term with a constant coupling $\beta$ to the regularized action, i.e., $$\tilde{I}=I_{EH}+\frac{\ell^2}{64\pi G} \int\limits_{M}
d^4x\,{\cal{E}}_{4}+\beta \int\limits_{M} d^4x\,{\cal{P}}_{4}\,,
\label{Itilde}$$ with Euclidean signature. Therefore, in a similar fashion to the case of the addition of the Euler term, the bulk dynamics cannot fix the Pontryagin coupling. However, one may expect that again the variational principle would provide a criterion to remove the arbitrariness in $\beta$.
The on-shell variation of the total action produces $$\delta \tilde{I} =\int\limits_{\partial
M}d^{3}x\,\sqrt{\cal{G}}\,\frac{n_{\sigma }}{N}\,\delta
\Gamma _{\varepsilon \lambda }^{\gamma }\left( \frac{\ell ^{2}}{64\pi G}\,%
\delta _{\lbrack \gamma \delta \alpha \beta ]}^{[\sigma \lambda \mu \nu ]}%
{\cal G}^{\delta \varepsilon }W_{\mu \nu }^{\alpha \beta } + \beta
\,\frac{\epsilon ^{\sigma \lambda \mu \nu }}{\sqrt{\cal{G}}}\,{\cal
G}_{\gamma \tau }W_{\mu \nu }^{\varepsilon \tau }\right) ,
\label{varItilde}$$where in the last term the part along $\delta_{[\mu \nu]
}^{[\varepsilon \tau] }$ is identically zero. The total surface term must vanish identically for certain boundary conditions. The argument here is different from the one used to fix the GB coupling in eq.(\[I4\]). In that case, $\alpha $ is also determined from the cancellation of the leading-order divergences in the Euclidean action what can be seen, e.g., from evaluating it for Schwarzschild-AdS black hole $$-TI_{SAdS}=TS-\frac{\pi
r^3}{4G\ell^2}\left(1-\frac{64\pi G}{\ell^2}\,\alpha\right)
-\frac{M}{2}\left(1+\frac{64\pi G}{\ell^2}\,\alpha\right),
\label{IESAdS}$$ where $S$ and $M$ are the black hole entropy and mass, respectively, and $T$ is the Hawking temperature. It is clear that the correct black hole thermodynamics is reproduced only by the same value of $\alpha$ as before. Moreover, for a given cosmological constant, it is not possible to express the variation eq.(\[varItilde\]) only in terms of the Weyl tensor unless $\alpha$ takes the value fixed in the previous sections.
The result (\[IESAdS\]) remains unchanged when ${\cal{P}}_{4}$ is added to the action, as it vanishes for static AdS$_{4}$ black holes. In general, it can be shown that the contribution of the Pontryagin term to the action is at most finite.
This means that we should look for asymptotic conditions in the next-to-leading order in the curvature of the AdS group (\[weyl4\]). Considering (anti) self-duality in the Weyl tensor $$W_{\mu \nu \alpha \beta }=\pm \frac{1}{2}\sqrt{\cal{G}}\,\epsilon
_{\mu \nu \lambda \sigma }W_{\alpha \beta }^{\lambda \sigma }
\label{selfdualC}$$ in the asymptotic region, we can fix the coupling constant of $\mathcal{P}_{4}$ as $$\beta =\pm \frac{\ell ^{2}}{32\pi G}\,,
\label{betadual}$$ demanding a well-posed action principle.
For arbitrary $\beta$, the variation of the action (\[Itilde\]) projected to the boundary indices defines a total stress tensor ${\cal T}_{j}^{i}$ $$\begin{aligned}
\delta \tilde{I}&=&
\frac{1}{2}\int\limits_{\partial
M}d^{3}x\,\sqrt{h}\,{\cal T}_{j}^{i} (h^{-1}\delta h)_{i}^{j}
\nonumber \\
&=& \frac{1}{2}\int\limits_{\partial M}d^{3}x\,\sqrt{h}
\left( T_{j}^{i}+\beta \, C_{j}^{i}\right) (h^{-1}\delta h)_{i}^{j}\,,
\label{totalstress}\end{aligned}$$ where $T_{j}^{i}$ is the stress tensor (\[BKtensor\]) and $C^{i}_{j}=\frac{1}{\sqrt{h}}\,\epsilon^{ikl}\nabla _{k}\left( \mathcal{R}_{lj}-\frac{%
1}{4}h_{lj}\mathcal{R}\right) $ is the Cotton tensor, obtained from the functional variation of the gravitational Chern-Simons term respect to the induced metric $h_{ij}$. The Cotton tensor is symmetric, traceless and covariantly conserved, and contributes as above to the total stress tensor of the theory when $h_{ij}$ is held fixed on the boundary (Dirichlet problem).
The term ${\cal{P}}_{4}$ does not modify the AdS asymptotics, such that we can use FG expansion and find the finite part of eq.(\[totalstress\]), which is given by $$\delta \tilde{I}=\frac{1}{2}\int\limits_{\partial M}d^{3}x\,\sqrt{g_{(0)}}
\left( -\frac{3}{16\pi G\ell }\,g_{(3)i}^{j}+\beta \, C_{j}^{i}(g_{(0)})\right)
(g_{(0)}^{-1}\delta g_{(0)})_{i}^{j}\,.
\label{totalholstress}$$
In a similar fashion, (anti) self-duality reads $$\begin{aligned}
&&\rho\, {\cal W}_{kl}^{np}+\frac{3\rho ^{3/2}}{2\ell ^{2}}\left(
g_{(3)k}^{n}\delta _{l}^{p}-g_{(3)l}^{n}\delta _{k}^{p}+\delta
_{k}^{n}g_{(3)l}^{p}-\delta _{l}^{n}g_{(3)k}^{p}\right)+\mathcal{O}(\rho ^{2}) \nonumber \\
&&\quad\qquad=\mp \,\frac{\rho ^{3/2}}{\ell \sqrt{g_{(0)}}}\,\epsilon ^{npm}
\left(\nabla _{(0)k}g_{(1)ml}-\nabla _{(0)l}g_{(1)mk}\right)+\mathcal{O}(\rho ^{2})\,,\end{aligned}$$ where ${\cal W}$ is the Weyl tensor of $g_{(0)}$.
As a consequence, when the condition (\[selfdualC\]) holds, the value $\beta=\pm\ell^2/(32\pi G)$ corresponds to the self-dual point where the total stress tensor vanishes identically, i.e., ${\cal T}_{j}^{i}=0$.
This reproduces the relation between the holographic stress tensor $T_{j}^{i}$ and the Cotton tensor $$T_{j}^{i}=\pm\frac{\ell ^{2}}{8\pi G}\,C_{j}^{i}\,, \label{TCduality}$$ which has been observed in recent literature for solitonic solutions [@deHaro-Petkou], electric-magnetic transformations in the fields in first-order gravity [@MPT] and axial-polar perturbations in hydrodynamic models in AdS$_{4}$ [@Bakas].
The full duality between the renormalized stress tensor and Cotton tensor has been obtained in [@deHaro] by relating two dual boundary CFTs which correspond to Dirichlet and Neumann boundary conditions (for a related work on boundary conditions, see, [@Compere-Marolf]). The two descriptions are mapped one into another by a Legendre transformation generated by a gravitational Chern-Simons term.
The total action for the particular value of $\beta$ which realizes the relation (\[TCduality\]) can be written in tetrad formalism as $$\tilde{I}=\frac{\ell ^{2}}{64\pi G}\int\limits_{M}\left( \epsilon
_{ABCD}W^{AB}W^{CD}\mp 2W^{AB}W_{AB} \right), \label{Iforms}$$ with the Weyl 2-form $W^{AB}=W^{\alpha \beta}_{\mu \nu}
e^{A}_{\alpha} e^{B}_{\beta} dx^{\mu}dx^{\nu}$ in terms of the local orthonormal basis $e^{A}=e^{A}_{\mu}dx^{\mu}$. In this notation the (anti) self-duality condition (\[selfdualC\]) reads $W_{AB}=\pm \ast W_{AB}=\pm \frac{1}{2}\epsilon _{ABCD}W^{CD}$, with $\ast \ast =+1$ for Euclidean signature.
Using the identity $\epsilon_{ABCD}W^{AB}W^{CD}=\frac{1}{2}\epsilon
_{ABCD}\left( W^{AB}W^{CD}+\ast W^{AB}\ast W^{CD}\right)$ and also that $\epsilon _{ABCD}W^{AB}\ast W^{CD}=-2{\cal{P}}_{4}d^4x$ (in an analogous way as in Yang-Mills theory), the total action (\[Iforms\]) can be cast into the form $$\begin{aligned}
\tilde{I}&=&\frac{\ell ^{2}}{128\pi G}\int\limits_{M}\epsilon _{ABCD}\left(
W^{AB}\mp \ast W^{AB}\right) \left( W^{CD}\mp \ast
W^{CD}\right) \nonumber\\
&=& \frac{\ell ^{2}}{16\pi G} \int\limits_{M} \sqrt{\det \left(W^{AB}\mp \ast W^{AB}\right)}.
\label{dual-anti-action}\end{aligned}$$ It is evident from the form of eq.(\[MacMan\]) that the value of the action reaches an absolute minimum for spacetimes which are globally of constant curvature (vacuum states of AdS gravity). The action (\[dual-anti-action\]) naturally generalizes this property to states which are globally (anti) self-dual in AdS gravity.
Conclusions
===========
We have shown that the standard regularization of AdS gravity with counterterms is indeed topological, as it can be obtained from the addition of the Gauss-Bonnet invariant or the corresponding boundary term.
We have also considered the odd parity Pontryagin invariant, which accounts for viscosity in hydrodynamic models and for *magnetic* properties of solitonic solutions in AdS$_{4}$ (by analogy to the charge formula (\[AMD\]) which involves the electric part of the Weyl tensor due to the addition of the Euler term). It is shown that the inclusion of this term is consistent assuming an asymptotic (anti) self-dual condition on the Weyl tensor. This reasoning explains the holographic stress tensor/Cotton tensor relation (\[TCduality\]) recently found in different setups in the literature, and interprets it as coming from a duality between topological invariants.
The addition of topological invariants of the Euler class to the Einstein-Hilbert gravity action in $D=2n$ dimensions was studied in ref.[@ACOTZ2n], with the purpose of rendering finite the Noether charges for AAdS spacetimes. The variational principle singles out the value of the Euler coupling which produces a regularizing effect. One can instead consider the action supplemented by a boundary term $I_{2n}=I_{EH}+c_{2n-1}\int_{\partial M}d^{2n-1}x B_{2n-1}$, where $c_{2n-1}$ is a constant. In ref.[@OleaJHEP], it is claimed that the term $B_{2n-1}$ which solves the regularization problem in even-dimensional AdS gravity is always prescribed by the Euler theorem and written using a parametric integration as a polynomial in the intrinsic and extrinsic curvatures $$\begin{aligned}
B_{2n-1} &=&2n\sqrt{-h}\int\limits_{0}^{1}dt\ \delta _{\left[
j_{1}\cdots j_{2n-1}\right] }^{\left[ i_{1}\cdots
i_{2n-1}\right] }K_{i_{1}}^{j_{1}}\left( \frac{1}{2}\mathcal{R}%
_{i_{2}i_{3}}^{j_{2}j_{3}}-t^{2}K_{i_{2}}^{j_{2}}K_{i_{3}}^{j_{3}}\right)
\times \cdots \notag \\
&&\cdots \times \left( \frac{1}{2}\mathcal{R}%
_{i_{2n-2}i_{2n-1}}^{j_{2n-2}j_{2n-1}}-t^{2}K_{i_{2n-2}}^{j_{2n-2}}K_{i_{2n-1}}^{j_{2n-1}}\right)
\label{B2n-1}\end{aligned}$$with a coupling constant $c_{2n-1}=(-\ell
^{2})^{n-1}/(16\pi Gn(2n-2)!)$. On purpose, we have not absorbed the constant in the boundary term, in order to stress the geometrical origin of the Kounterterm $B_{2n-1}$, as it is linked to topological invariants.
Furthermore, it has been proved that the term (\[B2n-1\]) regulates the Euclidean action and conserved quantities in any gravity theory of Lovelock type with AdS asymptotics – including Einstein-Gauss-Bonnet AdS–, and where the information on a particular theory is contained only in its coupling constant [@EGB-KO].
As in the four-dimensional case, we add and subtract the Gibbons Hawking term, i.e., $%
I_{2n}=I_{EH}-\frac{1}{8\pi G}\int_{\partial M}d^{2n-1}x\sqrt{-h}%
K+\int_{\partial M}d^{2n-1}x\mathcal{L}_{ct}$, where $\mathcal{L}_{ct}$ is given by$$\begin{aligned}
\mathcal{L}_{ct} &=&\frac{(-\ell ^{2})^{n}}{8\pi G(2n-2)!}\sqrt{-h}\delta _{%
\left[ j_{1} \cdots j_{2n-1}\right] }^{\left[ i_{1}\cdots i_{2n-1}
\right] }K_{i_{1}}^{j_{1}} \int\limits_{0}^{1}dt\ \left[ \left( \frac{1}{2}\mathcal{R}%
_{i_{2}i_{3}}^{j_{2}j_{3}}-t^{2}K_{i_{2}}^{j_{2}}K_{i_{3}}^{j_{3}}\right)
\cdots \right. \notag
\\
&& \quad \cdots \left. \left( \frac{1}{2}\mathcal{R}%
_{i_{2n-2}i_{2n-1}}^{j_{2n-2}j_{2n-1}}-t^{2}K_{i_{2n-2}}^{j_{2n-2}}K_{i_{2n-1}}^{j_{2n-1}}\right) +%
\frac{(-1)^{n}}{\ell ^{2n-2}}\delta _{i_{2}}^{j_{2}}\cdots \delta
_{i_{2n-1}}^{j_{2n-1}}\right] .\end{aligned}$$When all the fields are expanded in FG frame, one can collect terms as a power series in $\rho $ and perform explicitly the parametric integration. It is useful to express the extrinsic curvature expansion as $K_{j}^{i}=\frac{1}{\ell }\delta _{j}^{i}-\rho \ell S_{j}^{i}(g)+\mathcal{O}%
(\rho ^{2})$, where $S_{j}^{i}(g)=\frac{1}{D-3}(\mathcal{R}_{j}^{i}(g)-\frac{1}{2(D-2)}%
\delta _{j}^{i}\mathcal{R}(g))$ is the Schouten tensor of the metric $%
g_{ij}(x,\rho )$. Owing to the rescaling properties of the boundary Riemann tensor, the result can be written as a series of intrinsic counterterms
$$\begin{aligned}
\mathcal{L}_{ct} &=&\frac{\sqrt{-h}}{8\pi G}\left[ \frac{(2n-2)}{\ell }+%
\frac{\ell }{2(2n-3)}\,\mathcal{R}+\right. \nonumber \\
&&\hspace{-5mm}\left. +\frac{\ell ^{3}}{2(2n-3)^{2}(2n-5)}\left( 2\mathcal{R}^{ij}%
\mathcal{R}_{ij}-\frac{(2n+1)}{4(2n-2)}\mathcal{R}^{2}-\frac{(2n-3)}{4}%
\mathcal{R}^{ijkl}\mathcal{R}_{ijkl}\right) +\cdots \right] .\end{aligned}$$
which includes a rather unusual $(Riemann)^{2}$ contribution. However, the fall-off conditions for AAdS solutions imply that the Weyl tensor is such that $\sqrt{-h}%
W^{ijkl}W_{ijkl}\sim \frac{1}{r^{D-1}}$ in Schwarzschild-like coordinates (see also [@Ashtekar-Magnon-Das]). Using this property for $D\geq6$, we trade off the Riemman-squared term for the other curvature-squared terms, i.e., $\mathcal{R}^{ijkl}\mathcal{R}_{ijkl}=\frac{4}{(2n-3)}(\mathcal{%
R}^{ij}\mathcal{R}_{ij}-\frac{1}{2(2n-2)}\mathcal{R}^{2})$. Remarkably enough, the series $\mathcal{L}_{ct}$ adopts the form of standard counterterms obtained by holographic renormalization$$\begin{aligned}
\mathcal{L}_{ct} &=&\frac{\sqrt{-h}}{8\pi G}\left[ \frac{(2n-2)}{\ell }+%
\frac{\ell }{2(2n-3)}\mathcal{R}+\right. \nonumber \\
&&\left. +\frac{\ell ^{3}}{2(2n-3)^{2}(2n-5)}\left( \mathcal{R}^{ij}\mathcal{%
R}_{ij}-\frac{(2n-1)}{4(2n-2)}\mathcal{R}^{2}\right) +\cdots \right]
,\end{aligned}$$where cubic terms in the curvature are required by the regularization problem only for $D\geq 8$ dimensions.
We will provide the details of this derivation in a forthcoming publication.
Acknowledgments {#acknowledgments .unnumbered}
===============
We would like to thank A. Anabalón, D. Klemm, M. Leoni and S. Theisen for interesting discussions and S. de Haro for useful correspondence. R.O. also thanks Prof. S. Theisen for kind hospitality at AEI, Golm, during the completion of this work. O.M. is supported by FONDECYT grant 11070146 and the PUCV through the project 123.797/2007. The work of R.O. was funded in part by AEI-MPG.
[99]{}
J.M. Maldacena, Adv. Theor. Math. Phys. **2**, 231 (1998); E. Witten, Adv. Theor. Math. Phys. **2**, 253 (1998).
M. Henningson and K. Skenderis, J. High Energy Phys. **07**, 023 (1998); S. de Haro, K. Skenderis and S. Solodukhin, Commun. Math. Phys. **217**, 595 (2001).
C. Fefferman and R. Graham, in Elie Cartan et les Mathématiques d’aujourd’hui (Astérisque, Paris, 1985), p. 95.
V. Balasubramanian and P. Kraus, Commun. Math. Phys. **208**, 413 (1999); R. Emparan, C.V. Johnson and R.C. Myers, Phys. Rev. D **60**, 104001 (1999).
R. Olea, J. High Energy Phys. **06**, 023 (2005).
R. Olea, J. High Energy Phys. **04**, 073 (2007).
I. Papadimitriou and K. Skenderis, arXiv: hep-th/0404176; J. High Energy Phys. **08**, 004 (2005).
O. Mišković and R. Olea, Phys. Lett. B **640**, 101 (2006).
R. Aros, M. Contreras, R. Olea, R. Troncoso and J. Zanelli, Phys. Rev. Lett. **84**, 1647 (2000).
Totally antisymmetric delta symbol is defined as the determinant of Kronecker deltas $\delta^{\mu}_{\nu}$ with normalization equal to unity.
S.W. MacDowell and F. Mansouri, Phys. Rev. Lett. **38**, 739 (1997); Erratum-ibid: **38**, 1376 (1977).
D.S. Mansi, A.C. Petkou and G. Tagliabue, Class. Quantum Grav. **26**, 045008 (2009); Class. Quantum Grav. **26**, 045009 (2009).
A. Ashtekar and A. Magnon, Class. Quantum Grav. **1**, L39 (1984); A. Ashtekar and S. Das, Class. Quantum Grav. **17**, L17 (2000).
S. de Haro and A.C. Petkou, J. Phys. Conf. Ser. **110**, 102003 (2008).
I. Bakas, J. High Energy Phys. **01**, 003 (2009).
S. de Haro, J. High Energy Phys. **01**, 042 (2009).
G. Compere and D. Marolf, Class. Quantum Grav. **25**, 195014 (2008).
R. Aros, M. Contreras, R. Olea, R. Troncoso and J. Zanelli, Phys. Rev. D **62**, 044002 (2000).
G. Kofinas and R. Olea, Phys. Rev. D **74**, 084035 (2006); J. High Energy Phys. **11**, 069 (2007).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Leonardo Bagaglini[^1]'
title: '[Non orientable three-submanifolds of ${G_2}-$manifolds]{}'
---
Introduction {#introduction .unnumbered}
============
The compact, simply connected, exceptional, real, Lie group ${G_2}$, as subgroup of ${{\rm SO}}(7)$, acts on the grassmannians $\mathrm{Gr}_k({{\mathbb R}}^7)$ of $k-$planes. This action in transitive unless $k=3,4$ (see [@Bry2]). It happens that, in $k=3,4$ cases, the action has cohomogeneity one with principal isotropy a reducible representation of ${{\rm SO}}(3)$. In the oriented case, the remaining two special orbits are isomorphic and have isotropy representation given by a reducible, eight dimensional, ${{\rm SO}}(4)-$module; these planes correspond to associative and co-associative ones. But if we consider the not-oriented case, a different special orbit arises, with isotropy type ${{\rm SO}}(3)\times{\mathbb Z}_2$. Its planes are the only ones reversed by ${G_2}$. Therefore it is natural to ask whether there exist not-orientable submanifolds, of almost ${G_2}-$manifolds, modelled on that orbit. The answer is positive, indeed we prove that such class of manifolds is rich in examples and shows interesting properties.
The paper is structured as follows. First, in Section §\[sec1\], we analyse the action of ${G_2}$ on grassmannians, proving the structure theorem of the orbit space. In Section §\[sec2\] we introduce the main definition of *$\varphi-$planes*, which determines our local model. Next, in Section §\[sec3\], we recall fundamental concepts of Cartan-Kähler theory and in Section §\[sec4\] we prove that some closed, analytic, three-manifold can be presented as *$\varphi-$*submanifold of an open ${G_2}-$manifold (Theorem \[thm\] and Corollary \[cor1\]). In Section §\[sec5\] we classify all the homogeneous not-orientable $\varphi-$manifolds in ${\mathbb{RP}}^7$, when equipped with the canonical nearly parallel structure (Theorem ).
Generally we refer to manifolds equipped with a ${G_2}-$structure as ${G_2}-$manifolds in the torsion-free case, otherwise as *almost* ${G_2}-$manifolds.
${G_2}$ actions on Grassmannians {#sec1}
================================
In this section we describe the action of ${G_2}$ on grassmannians of three and four-planes of ${{\mathbb R}}^7$, both oriented and non oriented. In the sequel if $Z$ is a compact manifold on which a compact Lie group $G$ acts with cohomegenity one, by notation $Z/G=[G/H_1|G/K|G/H_2]$ we mean that the orbits space $Z/G$ is diffeomorphic to the closed interval and $G/H_1$, $G/H_2$, $G/K$ are models for special orbits and generic one respectively. We refer to [@Bre] for an exhaustive treatment of general theory of compact Lie groups.
Let $X$ be the grassmannian $\mathrm{Gr}^+_4({{\mathbb R}}^7)$ of oriented four-planes of ${{\mathbb R}}^7$. We often denote planes $\xi\in X$ by $u_1\wedge u_2\wedge u_3\wedge u_4$, for ordered linear independent vectors $u_1,u_2,u_3,u_4\in\xi$. Identify ${{\mathbb R}}^7$ with the reducible ${{\rm SO}}(4)-$module $$\label{mod1}
\quad{{\mathbb R}}^4\oplus\Lambda^2_-{{\mathbb R}}^4,$$ where $\Lambda^2_-{{\mathbb R}}^4$ represents the pre-dual space of anti-self dual two-forms on ${{\mathbb R}}^4$.\
Let $(x^1,x^2,x^3,r^0)$ be standard coordinates on ${{\mathbb R}}^4$ so that the two-forms $$\omega_1=dr^{0}\wedge dx^1+dx^2\wedge dx^3,\quad\omega_2=dr^{0}\wedge dx^2-dx^1\wedge dx^3,\quad\omega_3=dr^{0}\wedge dx^3+dx^1\wedge dx^2,$$ are a basis of $\Lambda^2_-({{\mathbb R}}^4)^*$. Consider $(r^1,r^2,r^3)$ the dual basis of $(\omega_1,\omega_1,\omega_3)$ in $\Lambda^2_-{{\mathbb R}}^4$. Then the three-form $$\label{stdcoo}
\varphi=\omega_1\wedge dr^1+\omega_2\wedge dr^2+\omega_3\wedge dr^3-dr^{123},$$ is stable[^2] and positive[^3] (see [@Hit]), therefore the action of its stabiliser, in ${\mathrm{GL}(7,{{\mathbb R}})}$, on ${{\mathbb R}}^7$ corresponds to the seven dimensional irreducible representation ${G_2}\subset{{\rm SO}}(7),$ where $(\underline{x},\underline{r})$ are orthonormal and positive oriented coordinates. We refer to $\varphi$ as the *fundamental* three-form associated to ${G_2}$[^4].
We call $(\underline{x},\underline{r})$ *Cayley coordinates* if $\varphi$ is given by $\eqref{stdcoo}$.
Observe that the Hodge dual $\phi$ of $\varphi$, also known as the *fundamental* four-form, is given by $$\phi=-\omega_1\wedge dr^{12}+\omega_2\wedge dr^{13}-\omega_3\wedge dr^{12} +dr^0\wedge dx^1 \wedge dx^2 \wedge dx^3.$$ with respect to Cayley coordinates.
[@Fri] The action of ${G_2}$ on $X$ has cohomogeneity one with principal isotropy ${{\rm SO}}(3)\subset{{\rm SO}}(4)$, via representation . There are two ${G_2}$-isomorphic special orbits, each one of isotropy type ${{\rm SO}}(4)$, which are through $$\xi_+=\frac{\partial}{\partial r^0}\wedge\frac{\partial}{\partial x^1}\wedge\frac{\partial}{\partial x^2}\wedge\frac{\partial}{\partial x^3}\quad\text{and}\quad \xi_-=-\xi_+,$$ respectively.
A path $\xi$, parametrizing the orbits, starting from $\xi_-$ and ending at $\xi_+$, is given by $$\begin{gathered}
\xi_\theta=\left(\mathrm{sin}(\theta)\frac{\partial}{\partial r^0}+\mathrm{cos}(\theta)\wedge \frac{\partial}{\partial r^1}\right)\wedge\frac{\partial}{\partial x^1}\wedge\frac{\partial}{\partial x^2}\wedge\frac{\partial}{\partial x^3},\quad \forall \theta\in\left[-\frac{\pi}{2},\frac{\pi}{2}\right].\end{gathered}$$ Observe that $$\phi|_{\xi_\theta}=\mathrm{sin}(\theta)(\mathrm{sin}(\theta)dr^0+\mathrm{cos}(\theta)dr^1)\wedge dx\wedge dy\wedge dz,$$ thus there exists only one (principal) orbit $\mathcal{O}_0$, through $\xi_0$, such that $$\mathcal{O}_0=\left\lbrace\sigma\in X\;|\;\phi|_{\sigma}=0\right\rbrace.$$ Hence the following holds.
${G_2}$ can reverse planes lying in $\mathcal{O}_0$.
The corollary follows since condition $\phi|_\sigma=0$, if $\sigma\in X$, does not depend on the orientation, hence both $\sigma$ and $-\sigma$ belong to the same orbit.
Let $$\varepsilon:X\longrightarrow X,$$ be the map which reverses the orientations, and $$p:X\longrightarrow Y,$$ be the quotient map over $Y=X/\varepsilon$, the grassmannian $\mathrm{Gr}_4({{\mathbb R}}^7)$ of non oriented planes.\
Then $$\varepsilon(\mathcal{O}_{\pm})=\mathcal{O}_{\mp}\quad\text{and}\quad\varepsilon(\mathcal{O}_0)=\mathcal{O}_0.$$
The action of ${G_2}$ on $Y$ has cohomogeneity one with principal isotropy ${{\rm SO}}(3)$. There are two, not equivalent, special orbits, of isotropy types ${{\rm SO}}(4)$ and ${{\rm SO}}(3)\times\mathbb{Z}_2$ respectively.
It is easy to see that there exists an open, connected, ${G_2}-$invariant neighbourhood $\mathcal{V}$ of $\mathcal{O}_+$ such that $\varepsilon(\mathcal{V})\cap\mathcal{V}=\emptyset$. Hence the restriction of $p$ $$p|_\mathcal{V}:\mathcal{V}\longrightarrow p(\mathcal{V}),$$ is a ${G_2}-$equivariant diffeomorphism onto an open ${G_2}-$invariant neighbourhood of $p(\mathcal{O}_+)$. Then it follows that the action has cohomogeneity one, principal isotropy ${{\rm SO}}(3)$ and one special orbit $p(\mathcal{O_+})$ of isotropy type ${{\rm SO}}(4)$. Now let $H$ and $K$ be the stabilisers of $p(\xi_0)\in Y$ and $\xi_0\in X$ respectively. Since $$hKh^{-1}\subseteq K,\quad\forall h\in H,$$ and $p$ is a double cover, $K$ is a normal subgroup of $H$ with index $2$. Thus $H=K \times\mathbb{Z}_2$ is the isotropy of the special orbit $p(\mathcal{O}_0)$ through $p(\xi_0)$. Summarizing $$Y/{G_2}=\left[{G_2}/SO(4)|{G_2}/{{\rm SO}}(3)|{G_2}/{{\rm SO}}(3)\times\mathbb{Z}_2\right].$$
The analysis we have just completed turns out to describe also the orbits space of both $\mathrm{Gr}_3^+({{\mathbb R}}^7)$ and $\mathrm{Gr}_3({{\mathbb R}}^7)$, as shown by the following remark.
Since $$\mathrm{Gr}_4^+({{\mathbb R}}^7)\cong\mathrm{Gr}_3^+({{\mathbb R}}^7)\quad\text{and}\quad \mathrm{Gr}_4({{\mathbb R}}^7)\cong\mathrm{Gr}_3({{\mathbb R}}^7),$$ by ${{\rm SO}}(7)-$equivariant isomorphisms, such manifolds are isomorphic as ${G_2}-$spaces too.
Local model {#sec2}
===========
In this section we investigate the local model of three-planes lying in $\mathcal{O}_0$.
By analogy with the previous section let $$p:\mathrm{Gr}_3^+({{\mathbb R}}^7)\rightarrow \mathrm{Gr}_3({{\mathbb R}}^7),$$ be the two-fold covering which forgets the orientations.\
Planes lying in $p(\mathcal{O}_0)\subset \mathrm{Gr}_3({{\mathbb R}}^7)$ are called $\varphi-$planes.
$\varphi-$planes are characterized by the the vanishing of the three-form $\varphi$. In fact the path $$\xi_{\theta}=\left(\mathrm{sin}(\theta)\frac{\partial}{\partial t^1}+\mathrm{cos}(\theta)\frac{\partial}{\partial x^1}\right)\wedge\frac{\partial}{\partial x^2}\wedge\frac{\partial}{\partial x^3},\quad \theta\in\left[-\frac{\pi}{2},\frac{\pi}{2}\right],$$ parametrizes the orbits and meets $\mathcal{O}_0$ in $\theta=0$.
Now, consider Cayley coordinates $(\underline{x},\underline{r})$ and let $\xi=\frac{\partial}{\partial x^1}\wedge\frac{\partial}{\partial x^2}\wedge\frac{\partial}{\partial x^3}\in \mathcal{O}_0$. Its stabiliser in ${G_2}$ is $$K=\left\{\left(\begin{matrix}
a&0&0\\
0&1&0\\
0&0&a\\
\end{matrix}\right)\;|\;a\in{{\rm SO}}(3)\right\}\cong {{\rm SO}}(3).$$ While the stabiliser of its projection $p(\xi)$ is $H=K\times {\mathbb Z}_2,$ where $${\mathbb Z}_2=\left\lbrace\left(\begin{matrix}
1_{{{\rm SO}}(3)}&0&0\\
0&1&0\\
0&0&1_{{{\rm SO}}(3)}\\
\end{matrix}\right),\left(\begin{matrix}
-1_{{{\rm SO}}(3)}&0&0\\
0&-1&0\\
0&0&1_{{{\rm SO}}(3)}\\
\end{matrix}\right)\right\rbrace.$$ Consider $\xi^\perp$, explicitly $\xi^{\perp}=\frac{\partial}{\partial r^0}\wedge\frac{\partial}{\partial r^1}\wedge\frac{\partial}{\partial r^2}\wedge\frac{\partial}{\partial r^3},$ and decompose it as orthogonal sum $\xi^\perp=\mu\oplus\lambda,$ where $\mu=\mathrm{Span}(\frac{\partial}{\partial r^0})$ and $\lambda=\mathrm{Span}(\frac{\partial}{\partial r^1},\frac{\partial}{\partial r^2},\frac{\partial}{\partial r^3})$. Let $H_\xi$ and $H_{\xi^{\perp}}$ be the groups $\mathrm{O}(\mu)\times{{\rm SO}}(\lambda)$ and $\mathrm{O}(\xi^{\perp})$ respectively. Then the following proposition holds.
\[rapp\] There exist omomorphisms $f_1$ and $f_2$ such that the diagrams $$\begin{tikzcd}
H\arrow{r}{}\arrow{d}{f_1}&\mathrm{GL}(\xi)\\
H_\xi
\arrow{ru}{}&
\end{tikzcd}
\quad\text{and}\quad
\begin{tikzcd}
H\arrow{r}{}\arrow{d}{f_2}&\mathrm{GL}(\xi^\perp)\\
H_{\xi^\perp}\arrow{ru}{}&
\end{tikzcd}$$are commutative. Moreover the representations of $H$ on $\xi^{\perp}=\mu\oplus\lambda$ and $\Lambda^3\xi^*\oplus\Lambda^2\xi^*$ are equivalent, with an explicit equivariant isomorphism given by $$\begin{CD}
\mu\oplus\lambda@>>>\Lambda^3\xi^*\oplus\Lambda^2\xi^*,\\
(u,v)@>>> (i_u\phi|_\xi,i_v\varphi|_\xi).
\end{CD}$$
The proof is a straightforward computation performed in Cayley coordinates.
Cartan-Kähler theory {#sec3}
====================
For the sake of clarity we recall the key concepts and theorems of Cartan-Kähler theory and its applications to ${G_2}-$geometry. For a detailed treatment we refer to [@BryB] and [@Bry].
Let $N$ be a manifold and ${\mathcal{I}}$ be a differential ideal of the ring $\Omega^*(N)$. Denote by ${\mathcal{I}}^n$ the intersection ${\mathcal{I}}\cap\Omega^n(N)$ and suppose that ${\mathcal{I}}^0$ is empty.\
Let $E$ be a $n-$dimensional subspace of some tangent space, $T_xN$, of $N$. We say $E$ an *integral element* of ${\mathcal{I}}$, equivalently $E\in V^n({\mathcal{I}})$, if every $n-$form lying in ${\mathcal{I}}$ vanishes when restricted to $E$. An integral element is said to be *ordinary* if, locally, $V^n({\mathcal{I}})\subset\mathrm{Gr}_n(TN)$ appears as the zero locus of some non zero functions with linear independent differentials.\
If $E$ is an integral element of ${\mathcal{I}}$ we define its polar space $H(E)$ as the set of $n+1-$dimensional integral extensions of $E$, explicitly $$H(E)=\left\{v\in T_xX\,|\,(i_v\delta)|_E=0,\;\forall \delta\in{\mathcal{I}}^{n+1}\right\}.$$ The *extension rank* of $E$ is the integer $r(E)=\mathrm{dim}H(E)-n-1$. Observe that $E$ is maximal if and only if $r(E)=-1$. We say $E$ *regular* if it is ordinary and $r$ is locally constant around it.\
An integral manifold $Y$ of ${\mathcal{I}}$ is a submanifold of $N$ whose tangent spaces are all integral elements. It is said to be ordinary, or regular, if its tangent spaces are. Now we are ready to state the Cartan-Kähler Theorem.
Let $N$ be an analytic manifold, ${\mathcal{I}}\subset\Omega^*(N)$ be a analytic, differential ideal and $X$ be one of its analytic, integral, $n-$dimensional manifolds. Suppose $X$ is regular, with extension rank $r\geq 0$, and let $Z$ be an analytic submanifold, of codimension $r$, containing $X$ and transversal to each of its polar spaces.\
Then there exists an analytic, integral, $(n+1)-$dimensional manifold $Y$ satisfying $$X\subset Y\subset Z.$$ Moreover, if $Y'$ is a manifold with the same properties, then $Y\cap Y'$ is still an integral $(n+1)-$dimensional manifold.
In order to verify regularity of an integral element we will need the following result, known as Cartan’s test of regularity. But first we give some other definitions.
Let $E$ be a $n-$dimensional, integral element of ${\mathcal{I}}$. An integral flag $(E_j)_j$, of length $n$ with terminus $E$, is an increasing filtration of $n+1$ vector spaces verifying the followings: $$E_0\subset E_1\subset\dots\subset E_n= E,\quad E_j\in V^j({\mathcal{I}}),\quad \mathrm{dim}E_j=j\quad j=0,\dots,n.$$ Let $(E_j)_j$ be an integral flag of length $n$ and $c_j$ be the codimension of $H(E_j)$ in the appropriate tangent space, for each $j$. Call $(E_j)_j$ regular if $E_j$, $j<n$, is regular and denote by $C$ the sum over $j$ of each $c_j$.
Let $(E_j)_j$ be a regular integral flag of length $n$.\
Then, locally around $E_n$, $V^n({\mathcal{I}})$ lies in a codimension $C$ submanifold of the grassmannian $\mathrm{Gr}_n(TN)$. Moreover the flag is regular if and only if, near $E_n$, $V^n({\mathcal{I}})$ is a smooth manifold of codimension $C$.
Let $N$ be a seven dimensional manifold and suppose its frame bundle[^5] ${\mathcal{F}}\xrightarrow{P}N$, where $${\mathcal{F}}=\left\{\xi_x\;|\;\xi_x:{{\mathbb R}}^7\rightarrow T_{x}N,\;x\in N\right\},$$ can be reduced to a principal ${G_2}-$bundle ${\mathcal{P}}$. This is equivalent to the existence of a global section of the fiber bundle ${\mathcal{F}}/{G_2}\xrightarrow{f} N$ defined by the diagram $$\begin{tikzcd}
{\mathcal{F}}\arrow{r}{Q}\arrow{dr}{P}&{\mathcal{F}}/{G_2}\arrow{d}{f}\\
&N\\
\end{tikzcd}$$ or, similarly, to the existence of a stable and positive (locally given by ), three-form $\varphi$.
Let us denote the total space ${\mathcal{F}}/{G_2}$ by $S$. Following Bryant (see [@Bry]) we introduce differential ideals ${\mathcal{I}}$’s on ${\mathcal{F}}$ and $S$ (labelled by the same letter), related by the projection $Q$, as follows. Let $\Lambda^*({{\mathbb R}}^7)^{{G_2}}$ be the ring of ${G_2}-$invariant, constant coefficients, differential forms on ${{\mathbb R}}^7$, and, for each $\delta_0\in\Lambda^*({{\mathbb R}}^7)^{{G_2}}$, consider $$\tilde{\delta}\in\Omega^*({\mathcal{F}}),\quad\text{where}\quad \delta_0\left(\xi^*\left(P_*(.)\right)\right)=\tilde{\delta}_{\xi}(.),\quad\forall \xi\in{\mathcal{F}}.$$ Now define $${\mathcal{I}}=<\left\{ d(\tilde{\delta})\;|\;\delta_0\in\Lambda^*({{\mathbb R}}^7)^{{G_2}}\right\}>.$$
The role of ${\mathcal{I}}$ in the study of ${G_2}-$structures is well explained by the following theorem. Recall that a ${G_2}$-structure is said to be *torsion-free* if the Levi-Civita connection restricts to ${\mathcal{P}}$.[^6].
Let $V^7({\mathcal{I}},f)$ be the set of seven dimensional integral elements of ${\mathcal{I}}$ which are transversal to the fibers of $f:S\rightarrow N$. Then $V^7({\mathcal{I}},f)$ consists of tangent spaces to graphs of local sections corresponding to torsion-free structures.
\[rem1\] Let $F$ be the projection ${\mathcal{F}}\rightarrow N$. Then $$\mathrm{codim}(V^7({\mathcal{I}},F),\mathrm{Gr}_7(T{\mathcal{F}}))=\mathrm{codim}(V^7({\mathcal{I}},f),\mathrm{Gr}_7(TS)).$$
Existence results {#sec4}
=================
In this section we prove existence of closed, connected, three-submanifolds of (open) ${G_2}-$manifolds, modelled on $\varphi-$planes.
A submanifold $X$ of an almost ${G_2}-$manifold $(N,\varphi)$ is said to be *$\varphi-$*manifold if all its tangent spaces are $\varphi-$planes or, equivalently, if the pullback of $\varphi$ to $X$ vanishes.
Any submanifold $X$ of a co-associative one, say $Y$, is, by definition, $\varphi-$manifold. Moreover, since for any $\varphi-$plane there is a unique direction defining a co-associative extension, such $X$ defines $TY|_X$.
Before proving the main theorem of this section we need the following lemma.
\[id\] Let $X$ be a $\varphi-$manifold of some, connected, almost ${G_2}$ manifold $(N,\varphi)$. Assume that there exist an isometry $\tau$ of $N$ and a point $x\in X$ such that $$\tau(x)=x,\quad (\tau^*)\varphi_x=\varphi_x,\quad \tau^*|_{T_x^*X}=\mathrm{Id}_{T_x^*X}.$$ Then $\tau=\mathrm{Id}_N$. In particular if $\tau_1$ and $\tau_2$ are isometries of $N$ preserving the structure, and they agree on some open set of $X$, then $\tau_1=\tau_2$.
Consider Cayley coordinates around $x$. Since $\tau^*_x$ preserves $\varphi_x$ its matrix representation lies in ${G_2}$; explicitly $$\tau^*_x=\left(\begin{matrix}
\pm t&0&0\\
0&\pm 1&0\\
0&0&t
\end{matrix} \right),\quad\text{for some}\quad t\in{{\rm SO}}(3).$$ But $\pm t$ represents $\tau^*_x|_{T_x^*X}$, which is the identity, hence $\tau^*_x=\mathrm{Id}_{T_xN}$. Since an isometry fixing $x$ and acting identically on $T_xN$ must be $\mathrm{Id}_N$ ($N$ is connected), it follows $\tau=\mathrm{Id}_N$.
Recall that, as stressed by Bryant in [@Bry], a closed, orientable, analytic, Riemannian three-manifold always admits an analytic parallelization. This follows since every orientable three-manifold is parallelizable (see [@WU]) and analytic differential forms on closed, analytic manifolds, are dense in the space of smooth differential forms (see [@Bo]).
\[thm\] Let $X$ be a closed, connected, orientable, analytic, Riemannian, three-manifold. Then $X$ can be isometrically embedded into an open, analytic, ${G_2}-$manifold $N$ as $\varphi-$manifold contained in a co-associative one, the last isometric to $X\times S^1$. Moreover if there exists an analytic, non trivial, involutive isometry $\tau\in \mathrm{Iso}(X)$ and an orthonormal co-frame $(\alpha_1,\alpha_2,\alpha_3)$ such that $\tau^*\alpha_j=\alpha_j$, if $j=1,2$, and $\tau^*\alpha_3=-\alpha_3$, then $\tau$ can be extended to an unique involutive isometry, which preserves the structure, on whole $N$.
Let $\tau$ be as in the second part of the statement, otherwise put $\tau=\mathrm{Id}_X$. Let $\eta=(\alpha_1,\alpha_2,\alpha_3)$ be an orthonormal global co-frame and $f\in\mathrm{O}(3)$ defined by $$f.\eta =(\tau^*\eta).$$ Consider the section $\xi$ of the the frame bundle ${\mathcal{F}}$, over $M=X\times S^1\times {{\mathbb R}}^3$, defined by the co-frame $(\alpha_1,\alpha_2,\alpha_3,dr^0,dr^1,dr^2,dr^3)$, where $r^0$ is the angle coordinate on $S^1$ and $(r^1,r^2,r^3)$ are coordinates on ${{\mathbb R}}^3$. Now define $\underline{\tau}\in\mathrm{Diff}(M)$ as $$\underline{\tau}(x,r^0,r^1,r^2,r^3)=(\tau(x),\mathrm{det}(f)r^0,(\mathrm{det}(f)f).(r^1,r^2,r^3)),\quad\forall (x,\underline{r})\in M.$$ Thanks to the following identification of principal bundles $$\begin{CD}
M\times{\mathrm{GL}(7,{{\mathbb R}})}@>>>{\mathcal{F}},\\
(x,\underline{r};a)@>>>\xi_{(x,\underline{r})}.a.
\end{CD}$$ the action of $\underline{\tau}$ on ${\mathcal{F}}$ is given by $$\underline{\tau}_*(x,\underline{r};a)=(\underline{\tau}(x,\underline{r});ta),\quad\forall (x,\underline{r})\in M,$$ where $t=t^{-1}\in{G_2}$ is $$t=\left(
\begin{matrix}
f&0&0\\
0&\mathrm{det}(f)&0\\
0&0&\mathrm{det}(f)f\\
\end{matrix}
\right).$$ In fact, if $m=(x,\underline{r})\in M$ and $a\in{\mathrm{GL}(7,{{\mathbb R}})}$, the following diagram turns out to be commutative $$\begin{tikzcd}
{{\mathbb R}}^7\arrow{rr}{a}&&{{\mathbb R}}^7\arrow{rr}{\xi_{m}}\arrow{dd}{t_m}\arrow{ddrr}{(\tau_*\xi)_{\tau(m)}}&&T_{m}M\arrow{dd}{\tau_*}\\
&&&&\\
&&{{\mathbb R}}^7\arrow{rr}{\xi_{\tau(m)}}&&T_{\tau(m)}M,
\end{tikzcd}$$ hence $\tau_*(\xi_m .a)=\xi_{\tau(m)}.t_m a$.\
Let $S$ be the total space of the ${G_2}-$structure bundle associated to ${\mathcal{F}}$ defined by $$Q:{\mathcal{F}}\longrightarrow {\mathcal{F}}/{G_2}=S.$$ The map $\underline{\tau}_*$ descends to a well defined diffeomorphism $[\underline{\tau}_*]$ of $S$, given by[^7] $$[\underline{\tau}_*][x,\underline{r};a]=[\underline{\tau}(x,\underline{r});ta].$$ Now, the form $\tilde{\varphi}\in \Omega^3(M)$ $$\tilde{\varphi}=(dr^0\wedge \alpha_1+\alpha_2\wedge\alpha_3)dr^1+(dr^0\wedge \alpha_2-\alpha_1\wedge\alpha_3)dr^2+(dr^0\wedge \alpha_3+\alpha_1\wedge\alpha_2)dr^3-dr^1\wedge dr^2\wedge dr^3,$$ is stable, positive, $\underline{\tau}-$invariant and related to $\sigma\in\Gamma(M,S)$ $$\sigma(x,\underline{r})=[x,\underline{r};1],\quad \forall (x,\underline{r})\in M.$$ Let $\underline{\varphi}$ be its pullback on ${\mathcal{F}}$.
Before proceeding further fix $(x,\underline{r})\in M$. Observe that $f$ leaves unchanged a complete flag of ${{\mathbb R}}^3$ $$\left\{0\right\}=L_0\subset L_1\subset L_2\subset L_3={{\mathbb R}}^3.$$ Then we may consider the complete flag $(F_k)_k$ of ${{\mathbb R}}^7$ given by $$\left\{0\right\}={{\mathbb R}}\subset {{\mathbb R}}^2\subset{{\mathbb R}}^3\subset {{\mathbb R}}^4\subset{{\mathbb R}}^4\oplus L_1\subset {{\mathbb R}}^4\oplus L_2\subset {{\mathbb R}}^4\oplus L_3={{\mathbb R}}^7.$$ Consider a seven dimensional integral element $E_7\subset T{\mathcal{F}}$ of the ideal ${\mathcal{I}}$, intrudced in §\[sec3\], transverse to the fiber over $(x,\underline{r})$. If $\theta=(\theta_k)_k$ represents the tautological one-form on ${\mathcal{F}}$ with respect to $\xi$[^8], then $E_7$ is the terminus of the complete integral flag $(E_k)_k$ given by $$\begin{cases}
E_k=\left\{e\in E_7\,|\,\theta_j(e)=0,\;k<j \right\},&\text{if}\quad 0\leq k\leq 4,\\
E_5=\left\{e\in E_7\,|\,\theta_5(e)+\theta_6(e)+\theta_7(e)\in L_1\right\},\\
E_6=\left\{e\in E_7\,|\,\theta_5(e)+\theta_6(e)+\theta_7(e)\in L_2\right\}.
\end{cases}$$
In order to compute the polar spaces of $E_k$ identify ${{\mathbb R}}^k$ with $F_k$, define, for $0\leq k\leq 7$, $\iota_k:{{\mathbb R}}^k\rightarrow{{\mathbb R}}^7$, and consider the decreasing filtration $({\mathfrak{h}}_k)_k$ of vector spaces given by $$\mathfrak{{\mathfrak{h}}}_k=\left\{A\in{\mathfrak{gl}(7,{{\mathbb R}})}\;|\;\iota_k^*(A^*\delta)=0\quad\forall \delta\in{\left(\Lambda^* {{\mathbb R}}^7\right)}^{{G_2}} \right\}.$$ Observe that, if $c_k=\mathrm{codim}({\mathfrak{h}}_k,{\mathfrak{gl}(7,{{\mathbb R}})})$, it turns out that $$(c_0,\dots,c_7)=(0,0,0,1,5,15,28,35).$$ In fact the computation is straightforward for $k<5$, and, if $k\geq 5$, there exists a ${G_2}-$isometry turning $F_k$ into any given $k-$plane: if we choose one of them a computation shows the claim.
Then, identifying ${\mathrm{GL}(7,{{\mathbb R}})}$ with the fiber, it turns out $$H(E_k)={\mathfrak{h}}_k+E_7.$$ Thus, by Cartan’s test, $(E_k)_k$ is a regular integral flag of ${\mathcal{I}}$. Moreover, by Remark \[rem1\], also $([E_k])_k$ is an integral regular flag of ${\mathcal{I}}$ on $S$, still transverse to the fiber.
For future reference observe that, since $\iota_k({{\mathbb R}}^k)$, for $k\geq 4$, is a $t-$module, $t$ acts on ${{\mathbb R}}^k$ and $$\iota_k^*t^*=t^*\iota_k^*,\quad\text{on}\quad \Lambda^*{{\mathbb R}}^7.$$ As consequence, if $k\geq 4$, ${\mathfrak{h}}_k$ turns out to be $\mathrm{Ad}(t)-$invariant: $$\iota_k^*((tAt)^*\delta)=\iota_k^*(t^{*}A^*t^*\delta)=t ^{*}\iota_k^*(A^*\delta)=0,\quad\forall A\in{\mathfrak{h}}_k,\delta\in{\left(\Lambda^* {{\mathbb R}}^7\right)}^{{G_2}}.$$ By hypotheses we can equip ${\mathfrak{gl}(7,{{\mathbb R}})}$ with an $\mathrm{Ad}(t)-$invariant metric and define the increasing filtration $(W_k)_k$ of invariant subspaces given by $$W_k={\mathfrak{h}}_k^{\perp},\quad k\geq 4.$$ Now, for each $k\geq 4$, let $U_k$ be an $\mathrm{Ad}(t)-$invariant open neighbourhood of $0\in W_k$ such that the map $$\begin{CD}
U_k\times{G_2}@>>>{\mathrm{GL}(7,{{\mathbb R}})},\\
(u,g)@>>> e^ug,
\end{CD}$$ is an embedding. It exists since $W_k$ does not intersect ${\mathfrak{h}}_k$, which contains ${\mathfrak{g}}$. With no loss of generality we may suppose $U_k\subset U_{k+1}$.
Finally we are ready to apply the Cartan-Käheler Theorem to produce integral manifolds of ${\mathcal{I}}$.\
First, define $X_4$ as $$X_4=\left\{[x,r^0,\underline{0};1]\;|\;x\in X,\;r^0\in{{\mathbb R}}\right\rbrace.$$ Obviously $X_4$ is $[\underline{\tau}_*]-$invariant. Moreover it is a $4-$dimensional integral manifold of ${\mathcal{I}}$ (since $(Q_*\underline{\varphi})|_{X_4}=0$ and $d(Q_*\underline{\phi})|_{X_4}=0$) whose tangent spaces are all regular elements of type $[E_4]$ with respect to a regular flag introduced before. Consequently $X_4$ has extension rank $r(X_4)=32$.\
Now consider the $10-$dimensional manifold $$Z_4=\left\lbrace [x,r^0,s;e^u]\;|\;x\in X,\;r^0\in{{\mathbb R}},\;s\in L_1,\;u\in U^4\right\rbrace.$$ $Z_4$ is $[\underline{\tau}_*]-$invariant, in fact $$\begin{aligned}
[\underline{\tau}_*]\left[x,{r^0},s;e^u\right]=&\left[\underline{\tau}(x,{r^0},s);te^u \right]\\
=&\left[\underline{\tau}(x,{r^0},s);e^vt\right],\\
=&\left[\underline{\tau}(x,r^0,s);e^v\right]\\
&\text{for some $v\in U^4$}.\end{aligned}$$ Moreover its tangent spaces are transversal to the polar spaces of type $H([E_4])$, and, by the Cartan-Kähler theorem, there exists a $5-$dimensional integral manifold $Y_4$ of ${\mathcal{I}}$, verifying $$X_4\subset Y_4\subset Z_4.$$ By invariance of $Z_4$ and uniqueness also $X_5=Y_4\cap [\underline{\tau_*}]Y_4$ is a $5-$dimensional integral manifold of ${\mathcal{I}}$ with the same property.\
Replacing $X_5$ with a neighbourhood of $X_4$ if necessary we may assume that $X_5$ is connected, $[\underline{\tau_*}]-$invariant, and the graph of a section over an open neighbourhood of $\left\{(x,r^0,0,0,0)\right\}$ in $\left\{(x,r^0,s)\,|\,s\in L_1\right\}$. Since it is a graph, its tangent spaces are regular of type $E_5$.
Now it would be clear what strategy we are following. The rest of the proof will proceed as we have just seen.
Define the $21-$dimensional manifold $Z_5$ as follows $$Z_5=\left\lbrace [x,{r^0},s;e^u]\;|\;x\in X,\;r^0,\in{{\mathbb R}},\;s\in L_2,\;u\in U^5\right\rbrace.$$ It is $[\underline{\tau_*}]-$invariant, meets the polar spaces of $X_5$ transversally, and its codimension equals the extension rank of $X_5$.\
Thus there exists a $6-$dimensional integral manifold $Y_5$ of ${\mathcal{I}}$ satisfying $$X_5\subset Y_5\subset Z_5.$$ Defining $X_6$ to be $Y_5\cap [\underline{\tau_*}] Y_5$, and replacing it with a suitable neighbourhood of $X_5$, we may assume that $X_6$ is a connected graph of a section over an open neighbourhood of $\left\{(x,{r^0},s)\,|\,s\in L_1\right\}$ in $\left\{(x,{ r^0},s)\,|\,s\in L_2\right\}$, hence also regular.
Define the $35-$dimensional manifold $Z_6$ as follows $$Z_6=\left\lbrace [x, r^0,\underline{r};e^u]\;|\;x\in X,\;r^0,r^1,r^2,r^3\in{{\mathbb R}},\;u\in U^6\right\rbrace.$$ It is $[\underline{\tau_*}]-$invariant, meets the polar spaces of $X_6$ transversally, and its codimension equals the extension rank of $X_6$.\
Thus there exists a $7-$dimensional integral manifold $Y_6$ of ${\mathcal{I}}$ satisfying $$X_6\subset Y_6\subset Z_6.$$ Finally defining $Y$ as $Y_6\cap [\underline{\tau_*}]Y_6$, and replacing it with a suitable neighbourhood of $X_6$, we may assume that $Y$ is a connected graph of a section, say $\varsigma$, over an open ($\underline{\tau}-$invariant) neighbourhood $N$ of $\left\{(x, r^0,s)\,|\,s\in L_2\right\}$ in $M$. Such $\varsigma$ defines a torsion-free ${G_2}-$structure on $N$, which agrees with that one induced by ${\varphi}$ on $X$. Hence $X$ turns out to be a $\varphi-$manifolds contained in the, compact, co-associative, $X\times S^1$. Finally the restriction of $\underline{\tau}$ is the unique isometry, by Lemma \[id\], which extends $\tau$.
The previous theorem allows us to prove the following corollary, on existence of non-orientable $\varphi-$manifolds.
\[cor1\] Let $X$ be a closed, connected, non orientable, analytic, Riemannian three-manifold, and $\pi:X'\rightarrow X$ its Riemannian orientation covering. Suppose there exist two orthonormal one-forms, $a_1$ and $a_2$, on $X$. Then there exist two open ${G_2}-$manifolds $N'$ and $N$ containing $X'$ and $X$ as $\varphi-$manifolds respectively, and a two-fold covering $\underline{\pi}:N'\rightarrow N$ preserving the structures and extending $\pi$. Moreover $X'$ and $X$ are contained in $\underline{\pi}$-related, closed, co-associative submanifolds $Y'$, $Y$, isometric to $X'\times S^1$ and $(X'\times S^1)/{\mathbb Z}_2$ respectively. In particular $X$ defines a non trivial class of $H^1(Y,{\mathbb Z}_2)$ (see [@BW]).
Let $\tau$ the non trivial deck transformation of $\pi$. Fix an orientation and a $\tau-$invariant metric on $X'$ and define $\alpha_j=\pi^*a_j$, for $j=1,2$, and $\alpha_3=*(\alpha_1\wedge\alpha_2)$. Obviously $(\alpha_1,\alpha_2,\alpha_3)$ satisfies the hypothesis of Theorem \[thm\], therefore there exists an open ${G_2}-$manifold $N'$, containing $X'$ as $\varphi-$manifold (inducing the same Riemannian structure). Moreover there exists a unique isometry $\underline{\tau}$, extending $\tau$ and preserving the structure. By Lemma \[id\] such extension verifies $\underline{\tau}^2=\mathrm{Id}_M$.
Since the group generated by $\underline{\tau}$ acts freely we can consider a $\underline{\tau}-$invariant tubular neighbourhood $N(X'\times S^1)$, of $X'\times S^1$ in $N'$, on which the restriction of $\underline{\tau}$ has no fixed points. Consequently the space $N=N(X'\times S^1)/\underline{\tau}$ turns out to be a manifold. Furthermore $N$ inherits a torsion-free ${G_2}-$structure and its submanifold $X'/\underline{\tau}$, naturally isometric to $X$, satisfies the condition of being a $\varphi-$manifold. Observe that the last is contained in the compact co-associative submanifold $(X'\times S^1)/\underline{\tau}$.
The following example shows a manifold obtained with trivial applications of the Cartan-Kähler argument.
\[flat torus\] Let $X'={{\mathbb R}}^3/{\mathbb Z}^3$ be the three-torus and $\tau$ be the involution $$\tau(x^1,x^2,x^3)=\left(x^2,x^1,x^3+\frac{1}{2}\right), \quad \forall(x^1,x^2,x^3)\in X'.$$ Then $\tau$ is the non trivial deck transformation of an orientation covering $\pi:X'\rightarrow X$, since it has no fixed points. Now consider $\left\{dx^1,dx^2,dx^3\right\}$ on $X'$ and let $T^4$ be the four-torus[^9] and $f\in\mathrm{Diff}(T^4)$, $f_\tau\in\mathrm{Diff}(X'\times T^4)$ given by $$f(r^0,r^1,r^2,r^3)=(-r^0,-r^2,-r^1,-r^3),\quad f_\tau=\tau\times f.$$ Then the form $$\begin{aligned}
\tilde{\varphi}=(dr^0\wedge dx^1+dx^2\wedge dx^3)dr^1+(dr^0\wedge dx^2-dx^1\wedge dx^3)dr^2+\nonumber\\(dr^0\wedge dx^3+dx^1\wedge dx^2)dr^3-dr^1\wedge dr^2\wedge dr^3.\end{aligned}$$ descends to a stable, positive, closed and co-closed, three-form $\varphi$ on $M=(X'\times T^4)/f_\tau$.
Observe that the submanifolds $$\begin{cases}
X_4=\left(X'\times T^1\times\left\{(0,0,0)\right\}\right)/f_\tau,\\
X_5=\left(X'\times T^1\times\left\{(r,r,0)\,|\, r\in {{\mathbb R}}\right\}\right)/f_\tau,\\
X_6=\left(X'\times T^3\times\left\{0\right\}\right)/f_\tau,\\
\end{cases}$$ satisfy $$X\subset X_4\subset X_5 \subset X_6 \subset M.$$
The next example shows that co-associative manifolds also arise in non trivial torsion classes of ${G_2}-$structures.
\[Ex2\] Consider ${\mathfrak{su}}(2)$ spanned by the Pauli matrices $\sigma_1,\sigma_2,\sigma_3$. Since the constant structures are $2\epsilon_{ijk}$ (sign of $(ijk)\in\mathbb{S}_3$) the isomorphism $$f_*=
\left(\begin{matrix}
0&0&-1\\
0&-1&0\\
-1&0&0\\
\end{matrix}\right),$$ lies in $\mathrm{Aut}({\mathfrak{su}}(2))$. Hence it defines an (involutive) automorphism $f$ of ${{\rm SU}}(2)$. Denote by $r_1,r_2,r_3$ the left-invariant one-forms defined by the generators of ${\mathfrak{su}}(2)$. Now let $S^1$ be the unit circle equipped with the angle coordinate $r^0$ and define $f_0\in \mathrm{Diff}(S^1)$ as $f_0(e^{2\pi i r^0})=e^{-2\pi i r^0}$. Obviously $f_0^*dr^0=-dr^0$.
If $X$ is a closed, non orientable, three-manifold let $X'\xrightarrow{\pi}X$ be its orientation covering with not trivial deck transformation $\tau$. Suppose there is a global co-frame $(\alpha_1,\alpha_2,\alpha_3)$ verifying $$\tau^*\alpha_1=\alpha_3,\quad\tau^*\alpha_2=\alpha_2.$$ Now, on $M'=X'\times S^1\times {{\rm SU}}(2)$, define the stable and positive three-form $$\begin{aligned}
\tilde{\varphi}=(dr^0\wedge \alpha_1+\alpha_2\wedge\alpha_3)r_1+(dr^0\wedge \alpha_2-\alpha_1\wedge\alpha_3)r_2+(dr^0\wedge \alpha_3+\alpha_1\wedge\alpha_2)r_3+\\\nonumber-r_1\wedge r_2\wedge r_3.\end{aligned}$$ Such $\tilde{\varphi}$ is invariant under the action of the involutive, with no fixed points, diffeomorphism $$f_\tau=\tau\times f_0\times f\in\mathrm{Diff}(M'),$$ so that it defines a stable and positive three-form $\varphi$ on the compact manifold $M=M'/f_\tau$. Moreover, identifying $X$ with the submanifold $(X'\times\left\{1\right\}\times\left\{1_{{{\rm SU}}(2)}\right\})/f_\tau$ or $(X'\times\left\{-1\right\}\times\left\{1_{{{\rm SU}}(2)}\right\})/f_\tau$, it turns out that $\varphi|_X=0$.
Now if we consider $X'$ to be the flat torus and $\alpha_1,\alpha_2,\alpha_3$ as in Example \[flat torus\], defining[^10] $$2\chi_3=[r_1\wedge r_2\wedge r_3]\quad\text{hence}\quad 2*\chi_3=[-dr^0\wedge\alpha_1\wedge\alpha_2\wedge\alpha_3],$$ the ${G_2}-$structure defined above satisfies $$\begin{cases}
d{\varphi}=\frac{1}{2}{\phi}+*\chi_3,\\
d\phi=0.
\end{cases}$$ In particular the submanifold $Y=(X'\times S^1\times\left\{1_{{{\rm SU}}(2)}\right\})/f_\tau$ is calibrated by $\phi$, thus it is volume-minimizing in its homological class.
An homogeneous classification {#sec5}
=============================
In this section we classify all the $G-$homogeneous not-orientable $\varphi-$manifolds of ${\mathbb{RP}}^7$, where $G$ is a closed subgroup of ${\mathrm{Spin}}(7)$ and ${\mathbb{RP}}^7$ is equipped with the canonical nearly parallel ${G_2}-$structure.
Let $G$ be a closed and connected subgroup of ${\mathrm{Spin}}(7).$ We refer to a three-dimensional $G-$homogeneous $\varphi-$manifold $X$ of ${\mathbb{RP}}^7$, or $S^7$, as *homogenous $\varphi-$manifold*.
In the previous sections we have seen examples of $\varphi-$manifolds arising as three-submanifolds of some co-associative ambient. In ${\mathbb{RP}}^7$ co-associative manifolds does not exist, therefore, in this setting, $\varphi-$manifolds cannot be extended, neither locally, to co-associative ones.
We are able to prove the following theorem.
\[homogphiman\] There exists a non-orientable, homogeneous, $\varphi-$manifold $X$ in ${\mathbb{RP}}^7$. Moreover any other submanifold sharing the same properties can be turned into $X$ by an element of ${\mathrm{Spin}}(7)$.
First let us recall how the nearly parallel ${G_2}-$structure is defined. Let ${{\mathbb R}}^8$ be the fundamental ${\mathrm{Spin}}(7)-$module and $\Phi_0\in\Lambda^4({{\mathbb R}}^8)^*$ the ${\mathrm{Spin}}(7)-$invariant form given by $$\Phi_0=\frac{1}{2}\omega_0^2+\Re(\Psi_0),$$ where $2i\omega_0=dz_1d\bar{z_1}+dz_2d\bar{z_2}+dz_3d\bar{z_3}+dz_4d\bar{z_4}$ and $\Psi_0=dz_1dz_2dz_3dz_4$ with respect to complex coordinates $(z_1,z_2,z_3,z_4)$ of ${{\mathbb C}}^4={{\mathbb R}}^8$. Let us identify ${{\mathbb R}}^8\setminus\left\{0\right\}$ with ${{\mathbb R}}^*\times S^7$ via $${{\mathbb R}}^*\times S^7\ni(r,x)\mapsto rx\in{{\mathbb R}}^8\setminus\left\{0\right\}.$$ Then there exists a unique $\varphi\in\Omega^3(S^7)$ such that $$\Phi_0=r^3dr\wedge\varphi+r^4*_7\varphi.$$ Such $\varphi$ defines a ${\mathrm{Spin}}(7)-$invariant, nearly parallel ($d\varphi=4*_7\varphi$), ${G_2}-$structure on $S^7$, therefore on ${\mathbb{RP}}^7$. For a more exhaustive treatment see [@Lot].
Clearly any $X\subset{\mathbb{RP}}^7$ arises from some ${\mathbb Z}_2-$invariant $X'\subset S^7$. Therefore the classification of not-orientable homogeneous $\varphi-$manifolds in ${\mathbb{RP}}^7$ reduces to the classification of homogeneous $\varphi-$manifolds in $S^7$ on which $-1\in{\mathbb Z}_2$ acts as an orientation reversing map. By abuse of notation we continue to denote $X'$ by $X$.
The following result, which [easily]{} follows from standard representation theory, represents the main tool to prove Theorem .
\[subgroups\] If $K$ is a three-dimensional closed and connected subgroup of ${\mathrm{Spin}}(7)$ then it is conjugated to one of the following, listed together with their defining representations on ${{\mathbb R}}^8$ and the dimensions $d$ of their centralizers[^11],
1. ${\mathrm{U}}(1)^3\subset{{\rm SU}}(4)$ via $[{{\mathbb C}}^4]$ with $d=0$;
2. ${{\rm SU}}(2)\subset{{\rm SU}}(4)$ via $[S^3{{\mathbb C}}^2]$ with $d=1$;
3. ${{\rm SU}}(2)\subset{{\rm SU}}(4)$ via $[{{\mathbb C}}^2\oplus{{\mathbb C}}^2]$ with $d=6$;
4. ${{\rm SU}}(2)\subset{{\rm SU}}(3)\subset{G_2}$ via $[{{\mathbb C}}^2]\oplus{{\mathbb R}}\oplus{{\mathbb R}}\oplus{{\mathbb R}}\oplus{{\mathbb R}}$ with $d=6$;
5. ${{\rm SU}}(2)\subset{{\rm SO}}(4)\subset{G_2}$ via $[[S^2{{\mathbb C}}^2]]\oplus[{{\mathbb C}}^2]\oplus{{\mathbb R}}$ with $d=3$;
6. ${{\rm SO}}(3)\subset{{\rm SU}}(3)\subset{G_2}$ via ${{\mathbb R}}^3\oplus{{\mathbb R}}^3\oplus{{\mathbb R}}\oplus{{\mathbb R}}$ with $d=1$;
7. ${{\rm SO}}(3)\subset{G_2}$ via $[[S^6{{\mathbb C}}^2]]\oplus{{\mathbb R}}$ with $d=0$.
Clearly, by compactness, either $K$ is abelian or it is locally isomorphic to ${{\rm SU}}(2)$. In the first case $K$ is a maximal torus of ${\mathrm{Spin}}(7)$ and the proposition follows, so let us assume the second holds.
Firstly we prove that the above representations are the only admissible ones. Denote by ${\mathfrak{k}}$ the Lie algebra of $K$, which is isomorphic to ${\mathfrak{su}}(2)$. The only irreducible real ${\mathfrak{k}}-$modules of dimension less than or equal to $8$ are $V_k$ for $k=\mathrm{dim}(V_k)=1,3,4,5,7$ or $8$, and the following are not allowed:
1. $V_5\oplus V_3$;
2. $V_5\oplus V_1\oplus V_1\oplus V_1$;
3. $V_3\oplus V_1\oplus V_1\oplus V_1\oplus V_1\oplus V_1$.
Indeed if one of these occurred, ${\mathfrak{k}}$ should be contained in the algebra of a stabiliser of a three-plane of ${{\mathbb R}}^8$, which is isomorphic to ${\mathfrak{so}}(4)={\mathfrak{su}}(2)\oplus{\mathfrak{su}}(2)$ contained in some ${\mathfrak{g}}_2$. But the only three-dimensional simple subalgebras of ${\mathfrak{so}}(4)$, up to inner automorphism, are three not-equivalent ${\mathfrak{so}}(3)$: namely $\left\{(X,0)\right\}$, $\left\{(0,X)\right\}$ and $\left\{(X,X)\right\}$ with representations on the correspondent ${{\mathbb R}}^7$ given by ${{\mathbb R}}^3\oplus[{{\mathbb C}}^2]$, ${{\mathbb R}}^3\oplus{{\mathbb R}}\oplus{{\mathbb R}}\oplus{{\mathbb R}}\oplus{{\mathbb R}}$ and ${{\mathbb R}}^3\oplus{{\mathbb R}}^3\oplus{{\mathbb R}}$ respectively.
Now we prove that any previous representation is given by a unique, up to ${\mathrm{Spin}}(7)-$action, Lie subgroup. This essentially follows from Dynkin’s classification of $\mathfrak{sl}(2,{{\mathbb C}})$ subalgebras of classical simple Lie algebras (see [@OniVil] and the references therein).
Clearly if $K$ represents a maximal torus the claim is straightforward. Even if $K$ lies in some ${G_2}$ there is nothing to prove: any subgroup of ${G_2}$, which is locally isomorphic to ${{\rm SU}}(2)$, must be conjugated to one of those listed above (see also [@Ma]), and all the subgroups of ${\mathrm{Spin}}(7)$ isomorphic to ${G_2}$ are conjugated.
Let us suppose that $K$ corresponds to one of the remain representations, that are (2) and (3). Clearly $-\mathrm{id}_{{{\mathbb R}}^8}$ belongs to $K$, therefore the projection of $K$ into ${{\rm SO}}(7)$, say $H$, is a copy of ${{\rm SO}}(3)$. Since the only irreducible real ${{\rm SO}}(3)-$modules of dimension less than or equal to $7$ are $V_k$ with $k$ equal to $1,3,5$ or $7$ there are few admissible type-decompositions of ${{\mathbb R}}^7$. Moreover the restriction of the standard metric on each factor coincide, up to a scalar multiple, to the unique $H$-invariant metric, since the modules are of real type.
One can observe that $V_3\oplus V_3\oplus V_1$ and $V_7$ are modules realized by subgroups of some ${G_2}\subset{{\rm SO}}(7)$ and thus they cannot occur in our hypothesis, but this is not relevant. In each case $H$ must be cojugated, by an element of $\mathrm{O}(7)$, to the standard ${{\rm SO}}(3)$ defined by the correspondent representation; actually in ${{\rm SO}}(7)$, combing with $-\mathrm{id}_{{{\mathbb R}}^7}$. Therefore also $K$ will be conjugated, by an element of ${\mathrm{Spin}}(7)$, to the appropriate ${{\rm SU}}(2)$.
Finally let ${\mathfrak{z}}$ be the Lie algebra of the centralizer of $K$ in ${\mathrm{Spin}}(7)$.
In case 1 $d=0$.
In case 2 it is well known that ${\mathfrak{spin}}(7)={\mathfrak{su}}(4)\oplus{\mathfrak{m}}$, where ${\mathfrak{m}}\cong[[\Lambda^2{{\mathbb C}}^4]]$ as ${{\rm SU}}(4)-$modules. Therefore $${\mathfrak{spin}}(7)=\underbrace{{\mathfrak{su}}(2)\oplus{\mathfrak{n}}}_{{\mathfrak{su}}(4)}\oplus[[S^4{{\mathbb C}}^2]]\oplus[[{{\mathbb C}}]],$$ with $\mathrm{Ad}({{\rm SU}}(2)){\mathfrak{n}}={\mathfrak{n}}$, and consequently ${{\mathbb R}}\cong{\mathfrak{z}}$.
For the following cases let us recall that ${\mathfrak{spin}}(7)={\mathfrak{g}}_2\oplus{{\mathbb R}}^7$ as ${\mathfrak{g}}_2-$modules and that ${\mathfrak{g}}_2={\mathfrak{su}}(3)\oplus[{{\mathbb C}}^3]$ as ${{\rm SU}}(3)-$modules.
In case 3 $${\mathfrak{spin}}(7)=\underbrace{{\mathfrak{su}}(2)\oplus{{\mathbb R}}^3\oplus [S^2{{\mathbb C}}^2]\oplus[{{\mathbb C}}]\oplus{{\mathbb R}}}_{{\mathfrak{su}}(4)}\oplus [{{\mathbb C}}^2\oplus{{\mathbb C}}^2\oplus{{\mathbb C}}]\oplus[[{{\mathbb C}}]],$$ and thus ${\mathfrak{z}}\cong{{\mathbb R}}^6$.
In case 4 $${\mathfrak{spin}}(7)=\underbrace{\underbrace{{\mathfrak{su}}(2)\oplus[{{\mathbb C}}^2]\oplus{{\mathbb R}}}_{{\mathfrak{su}}(3)}\oplus[{{\mathbb C}}^2]\oplus{{\mathbb R}}\oplus{{\mathbb R}}}_{{\mathfrak{g}}_2}\oplus[{{\mathbb C}}^2]\oplus{{\mathbb R}}\oplus{{\mathbb R}}\oplus{{\mathbb R}},$$ and consequently ${\mathfrak{z}}\cong{{\mathbb R}}^6$.
In case 5 $${\mathfrak{spin}}(7)=\underbrace{\underbrace{{\mathfrak{su}}(2)\oplus{{\mathbb R}}\oplus{{\mathbb R}}\oplus{{\mathbb R}}}_{{\mathfrak{so}}(4)}\oplus[{{\mathbb C}}^2\oplus {{\mathbb C}}^2]}_{{\mathfrak{g}}_2}\oplus{{\mathbb R}}^3\oplus[{{\mathbb C}}^2],$$ which means ${\mathfrak{z}}\cong{{\mathbb R}}^3$.
In case 6 $${\mathfrak{spin}}(7)=\underbrace{\underbrace{{\mathfrak{so}}(3)\oplus{{\mathbb R}}^5}_{{\mathfrak{su}}(3)}\oplus{{\mathbb R}}^3\oplus{{\mathbb R}}^3}_{{\mathfrak{g}}_2}\oplus{{\mathbb R}}^3\oplus{{\mathbb R}}^3\oplus{{\mathbb R}},$$ which gives ${\mathfrak{z}}\cong{{\mathbb R}}^1$.
Finally, in case 7, a direct computation shows that $d=0$.
\[isotropy\] Let $G$ be a closed subgroup of ${\mathrm{Spin}}(7)$ and $X=G/H$ be an homogeneous not-associative three-submanifold of $S^7$. Then $\mathrm{dim}(H)\leq 3$. Moreover if $\mathrm{dim}(H)=1$ then $H^0$ must fix at least four directions in ${{\mathbb R}}^8$.
Let $x$ be the origin point of $X$ and $\rho:H\rightarrow T_xS^7$ be the isotropy representation. Clearly $\rho$ is injective and from Section §\[sec1\] it follows that $$\rho(H^0)\subseteq \mathrm{Stab}(\varphi_x)\cap\mathrm{Stab}(T_xX)^0\cong{{\rm SO}}(3),$$ where ${{\rm SO}}(3)$ acts on $T_xS^7$ as ${{\mathbb R}}^3\oplus{{\mathbb R}}^3\oplus{{\mathbb R}}$. Then the Lemma follows.
Now we are ready to prove the main theorem of this section.
Let $G$ be a closed subgroup of ${\mathrm{Spin}}(7)$ whose three-dimensional ${\mathbb Z}_2-$invariant orbit $X$ trough $x\in S^7$ satisfies $\varphi|_X=0$ and let $H$ be the stabiliser of $x$ in $G$. Let us denote the Lie algebras of $G$ and $H$ by ${\mathfrak{g}}$ and ${\mathfrak{h}}$ respectively.
First suppose that $\mathrm{dim}({\mathfrak{g}})=3$. Then by Proposition $G$ belongs to one of seven conjugacy classes. If $X/{\mathbb Z}_2$ is not-orientable, by compactness, its first Betti number cannot vanish, as well as that of $G$. Therefore $G$ would be conjugated to the assigned maximal torus ${\mathrm{U}}(1)^3$ of ${\mathrm{Spin}}(7)$; so let us assume they are equal. It is easy to see that ${\mathbb Z}_2$ acts trivially on its Lie algebra, and therefore any ${\mathrm{U}}(1)^3-$orbit in ${\mathbb{RP}}^7$ is orientable.
Suppose $\mathrm{dim}({\mathfrak{g}})>3$. Lemma implies that $1\leq\mathrm{dim}({\mathfrak{h}})\leq 3$. More precisely either $\mathrm{dim}({\mathfrak{h}})=1$ or $\mathrm{dim}({\mathfrak{h}})=3$; indeed ${{\rm SO}}(3)$ has rank one. In the second case $X$ should be diffeomorphic to $S^3$ and thus have vanishing first Betti number; in particular $X/{\mathbb Z}_2$ will be orientable. Therefore assume $\mathrm{dim}({\mathfrak{h}})=1$ and consequently $\mathrm{dim}({\mathfrak{g}})=4$.
Let $G_s\ltimes T$ be the Levi decomposition of $G$, where $G_s$ is a semisimple closed subgroup, with Lie algebra ${\mathfrak{g}}_s$, and $T$ a torus, with Lie algebra ${\mathfrak{t}}$, lying in its centralizer in ${\mathrm{Spin}}(7)$. Since ${\mathrm{Spin}}(7)$ has rank three $G_s$ must be three-dimensional and therefore conjugated to one of the simple Lie groups listed in Proposition different from (7).
Let ${\mathfrak{h}}_s$ and ${\mathfrak{h}}_t$ be the projections of ${\mathfrak{h}}$ on ${\mathfrak{g}}_s$ and ${\mathfrak{t}}$ respectively. If ${\mathfrak{h}}_t$ was one-dimensional then $\left\{A.x\;|\;A\in{\mathfrak{g}}_s\right\}$ would generate all $T_xX$ and consequently $X$ should be $G_s-$homogeneous as well, and thus described by one of the previous cases. Therefore we can suppose that ${\mathfrak{h}}\subset{\mathfrak{g}}_s$.
Then it is easy to see that the only two simple subgroups which admit non-trivial stabilisers are (5) and (6). But one can observe that in (5) any one-dimensional subspace generates a connected subgroup which fixes two lines only, in contrast with Lemma . Therefore the only admissible groups are conjugated to (6).
Let $(x_0,\dots,x_7)$ be standard coordinates on ${{\mathbb R}}^8$. With no loss of generality we can assume that $G_s$ acts on $x_0=x_7=0$: in fact ${\mathrm{Spin}}(7)$ is transitive on the Grassmannian of three-planes. Then a generic element $A\in {\mathfrak{g}}$ can be written as $$A=\sum_{j=0}^3 a_jE_j=
\left( \begin {array}{cccccccc} 0&0&0&0&0&0&0&3\,a_{{0}}
\\ \noalign{\medskip}0&0&a_{{1}}&a_{{2}}&0&0&-a_{{0}}&0
\\ \noalign{\medskip}0&-a_{{1}}&0&-a_{{3}}&0&a_{{0}}&0&0
\\ \noalign{\medskip}0&-a_{{2}}&a_{{3}}&0&a_{{0}}&0&0&0
\\ \noalign{\medskip}0&0&0&-a_{{0}}&0&a_{{3}}&a_{{2}}&0
\\ \noalign{\medskip}0&0&-a_{{0}}&0&-a_{{3}}&0&a_{{1}}&0
\\ \noalign{\medskip}0&a_{{0}}&0&0&-a_{{2}}&-a_{{1}}&0&0
\\ \noalign{\medskip}-3\,a_{{0}}&0&0&0&0&0&0&0\end {array} \right),$$ where $E_0$ and $E_1,E_2,E_3$ span ${\mathfrak{t}}$ and ${\mathfrak{g}}_s$ respectively. Observe that $T$ acts as a diagonal ${{\rm SO}}(2)$ on $(x_0,x_7,x_6,x_1,x_2,x_5,x_4,x_3)$; in particular ${\mathbb Z}_2\subset T$ and $G_s\cap T=\left\{1\right\}$, whereas $G_s$ acts as a diagonal ${{\rm SO}}(3)$ on $(x_1,x_2,x_3,-x_6,x_4,x_5)$. Therefore $x$ must satisfy $(x_1,x_2,x_3)\wedge(-x_6,x_4,x_5)=0$ to be fixed by some non-trivial subgroup, and $x_0^2+x_7^2<1$ to have three-dimensional orbit. Clearly, acting with $G$, we can ensure that $x=(x_0,x_1,0,0,0,0,x_6,0)$.
Since the tangent space to the orbit at $x$ is three-dimensional and generated by $$v_1=(8x_0^2+1)^{-\frac{1}{2}}E_0x,\;v_2=(x_1^2+x_6^2)^{-\frac{1}{2}}E_1x,\;v_3=(x_1^2+x_6^2)^{-\frac{1}{2}}E_2x,$$ it turns out that $$|\varphi_x(v_1,v_2,v_3)|=|\Phi_0(x,v_1,v_2,v_3)|=4\,{\frac {|x_{{0}}x_{{6}} \left( 3\,{x_{{1}}}^{2}-{x_{{6}}}^{2}
\right) |}{ \left( {x_{{1}}}^{2}+{x_{{6}}}^{2} \right) \sqrt {8\,{x_{{0
}}}^{2}+1}}}
.$$ We therefore see that the homogeneous $\varphi-$manifolds are those given by either $x_0x_6=0$ or $3x_1^2=x_6^2$. Let us describe the topology of these orbits. Consider a generic element $at\in H$, where $a\in G_s$ and $t\in T$. Then $$x_0=(atx)_0=(tx)_0=cx_0,\quad c\in[-1,1].$$ Therefore, if $x_0\neq 0$, $t$ equals the identity. Thus $H=H^0\cong{{\rm SO}}(2)$. In this case the orbit is isomorphic to $({{\rm SO}}(3)/{{\rm SO}}(2))\times T\cong S^2\times S^1$ and the action of ${\mathbb Z}_2$ on $X$ equals the antipodal map on $S^1$. Otherwise, if $x_0=0$, from $atx=x$ it follows that $$a=\left(\begin{matrix}
a_{11}&0&0\\
0&a_{22}&a_{23}\\
0&a_{32}&a_{33}\\
\end{matrix}\right),\;t=\left(\begin{matrix}
c&-s\\
s&c
\end{matrix}\right),\; a_{11}c=1.$$ Consequently $$H=\mathrm{O}(2)\tilde{\times}{\mathbb Z}_2=(\mathrm{O}(2)\times{\mathbb Z}_2)/((1_{\mathrm{O}(2)},1_{{\mathbb Z}_2})\sim(-1_{\mathrm{O}(2)},-1_{{\mathbb Z}_2}))\subset{{\rm SO}}(3)\times T,$$ and the action of ${\mathbb Z}_2$ on the orbit is given by $$G/H\ni g H\mapsto (-1_{{\mathbb Z}_2}) g H\in G/H$$ Thus $$\left(G/H\right)/{\mathbb Z}^2=\left({{\rm SO}}(3)\times T\right)/\left(\mathrm{O}(2)\times {\mathbb Z}_2\right)\cong\left({{\rm SO}}(3)/\mathrm{O}(2)\right)\times\left(T/ {\mathbb Z}_2\right)\cong {\mathbb{RP}}^2\times S^1.$$ Acting with $T$ we can ensure that $x_6=0$ and then see that the orbit is unique, passing through $x=(0,1,0,0,0,0,0,0)$ and defined by the intersection of the totally geodesic ${\mathbb{RP}}^5$, of equations $x_0=x_7=0$, and three quadrics $$\left\{x_1x_5+x_2x_6=0\right\}\cap\left\{x_1x_4+x_3x_6=0\right\}\cap\left\{x_2x_4-x_3x_5=0\right\}.$$
Acknowledgements {#acknowledgements .unnumbered}
================
The author is grateful to Prof. F. Podestà for all the useful suggestions and conversations he shared with him, to Prof. A. Fino for her constant interest in his work and finally to ‘‘Università degli studi di Firenze" for all the support he received.
[9]{}
S. Bochner, *Analytic mapping of compact Riemannian spaces into Euclidean space*, Duke Math. J. 3 (1937), 339–354.
Glen E. Bredon and John W. Wood, *Non-Orientable Surfaces in Orientable 3-Manifolds*, Inventiones math. 7, 83 - 110 (1969).
Glen E. Bredon, *Introduction to compact trasformation groups*. Academic Press (1972).
Robert L. Bryant, *Metrics with exceptional holonomy*, Ann. of Math. 126 (1987), 525 – 576.
Robert L. Bryant, *Exterior Differential Systems*. Springer-Verlag, New York (1991).
Robert L. Bryant, *Calibrated embeddings in the special lagrangian and co-associative cases*. Annals of Global Analysis and Geometry 18, Issue 3, 405-435 (2000).
T. Friedrich, I. Kath, A. Moroianu, U. Semmelmann. *On nearly parallel G2-structures*. Journal of Geometry and Physics, Volume 23, Issue 3-4, p. 259-286 (1997).
N.J. Hitchin, *Stable forms and special metrics*. In Global Differential Geometry: The Mathematical Legacy of Alfred Gray. Contemporary Mathematics 288. (Bilbao, 2000).
J. Lotay, *Associative Submanifolds of the 7-Sphere*, Proceedings of the London Mathematical Society 105(6) (June 2010).
K. Mashimo, *Homogeneous totally real submanifolds of $S^6$*, Tsukuba J. Math. 9 (1985) 185–202.
Robert C. McLean, *Deformations Of Calibrated Submanifolds*. lnventiones math. 7, 83 - 110 (1969) Commun. Analy. Geom. 6, 705 - 747 (1996).
A.L. Onishchik E. B. Vinberg (Eds.), *Lie Groups and Lie Algebras III. Structure of Lie Groups and Lie Algebras*. Encyclopaedia of Mathematical Sciences Vol 41. 1994. Springer-Verlag.
J. Milnor and J. Stasheff, *Characteristic Classes*, Annals of Math Studies 76, Princeton University Press, Princeton (1974).
[^1]: Dipartimento di Matematica e Informatica [“]{}Ulisse Dini["]{}. Università degli Studi di Firenze. Viale Morgagni, 67/a - 50134, Firenze, Italy. E-mail address: [email protected].
[^2]: Its ${\mathrm{GL}(7,{{\mathbb R}})}-$orbit is open.
[^3]: It induces a positive definite metric.
[^4]: Such form is defined by (the irreducible seven dimensional representation of) ${G_2}$ up to constant. It can be fixed by choosing an orientation and imposing $||\varphi||^2=7$.
[^5]: The action of $a\in{\mathrm{GL}(7,{{\mathbb R}})}$ on $\xi\in{\mathcal{F}}$ is given by $\xi.a=\xi\circ a$.
[^6]: Other equivalent conditions are: $\varphi$ is parallel; $\varphi$ is both closed and co-closed
[^7]: In the sequel square brackets denote points in $S$ as follows $\left[x,\underline{r};a\right]=\left\{(x,\underline{r};ab)\;|\;b\in{G_2}\right\}$.
[^8]: Explicitly $\theta_k(\xi^h)=\delta_k^h$.
[^9]: We could use ${{\mathbb R}}^4$ as well.
[^10]: Here, if $\delta\in\Omega^*(M')$ is invariant under $f_\tau^*$, then $[\delta]$ denotes the correspondent differential form on $M$.
[^11]: In the following if $W$ is a complex $K-$module then $[W]$ denotes the real module equal to $W$ whereas $[[W]]$ denotes the real module such that ${{\mathbb C}}\otimes[[W]]=W$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We derive the splitting kernels for partons produced in large $Q^2$ scattering processes that subsequently traverse a region of strongly-interacting matter using a recently-developed effective theory . We include all corrections beyond the small-$x$ approximation, consistent with the power counting of . We demonstrate how medium recoil, geometry and expansion scenarios, and phase space cuts can be implemented numerically for phenomenological applications. For the simplified case of infinite transverse momentum kinematics and a uniform medium, we provide closed-form analytic results that can be used to validate the numerical simulations.'
author:
- Grigory Ovanesyan
- Ivan Vitev
title: |
Medium-induced parton splitting kernels from Soft Collinear Effective Theory\
with Glauber gluons
---
Introduction {#introsec}
============
The suppression in the production rate of energetic leading particles and particle correlations due to final-state interactions in reactions with ultra-relativistic nuclei is among the best-known experimental discoveries at the Relativistic Heavy Ion Collider (RHIC) [@Arsene:2004fa]-[@Adcox:2004mh], and now at the Large Hadron Collider (LHC) [@Aamodt:2010jd]-[@Dainese:2011vb]. This jet quenching phenomenon also provides one of the strongest pieces of evidence for the creation of dense strongly-interacting matter in such collisions [@Gyulassy:2003mc]. Recent advances in understanding the modification of partons and parton showers in Quantum Chromodynamics (QCD) media have come from the inclusion of jets in the theoretical [@Vitev:2009rd]-[@Neufeld:2011fh] and experimental [@Salur:2009vz]-[@Chatrchyan:2011ua] analyses. Jet observables are more sensitive to the underlying theoretical model assumptions and the properties of the QCD medium when compared to leading particle measurements [@Vitev:2008rz].
In recent years effective field theories (EFT) have become a powerful modern tool for jet physics. In particular, Soft Collinear Effective Theory [@Bauer:2000ew; @Bauer:2000yr; @Bauer:2001ct; @Bauer:2001yt] (SCET) is an effective theory for QCD that describes the dynamics of highly energetic partons. It has been successfully applied to improve the theoretical accuracy in the evaluation of high energy cross sections at lepton [@Fleming:2007xt]-[@Abbate:2010xh] and hadron colliders [@Becher:2007ty]-[@Becher:2009th].
The first step in constructing an effective theory for jets propagating in a QCD medium was done in Ref. [@Idilbi:2008vm], where the SCET Lagrangian was extended by adding a term that describes the interaction of a quark jet with gluons that have momentum purely transverse to it, traditionally referred to in the literature as Glauber gluons. As an application of effective Lagrangian derived in [@Idilbi:2008vm], in Ref. [@D'Eramo:2010ak] the probability density of quark jet broadening [@Gyulassy:2002yv; @Qiu:2003pm] was re-derived as an expectation value of Wilson lines, which later the authors evaluate using AdS/CFT correspondence. In Ref. [@Ovanesyan:2011xy] the Yang-Mills part of the collinear SCET Lagrangian was coupled to Glauber gluons, which allowed to perform calculations for parton splitting processes in the medium. The resulting effective theory was called $\text{\SCETG}$, where “G" stands for Glauber gluons. In that paper a detailed connection was made between calculations in $\text{\SCETG}$ and the evaluation of jet broadening [@Gyulassy:2002yv; @Qiu:2003pm] and medium-induced quark energy loss in the Gyulassy-Levai-Vitev approach [@Gyulassy:2002yv; @Vitev:2007ve]. The gauge invariance of the physics results was explicitly demonstrated for three different gauge choices. One of the medium-induced splittings, namely $q\rightarrow qg$ has been calculated in Ref. [@Ovanesyan:2011xy] beyond the soft emitted gluon approximation. There are three additional splittings: $g\rightarrow gg, g\rightarrow q\bar{q}$ and $q\rightarrow gq$. To complete the derivation of all medium-induced branching processes without the assumption of a soft final-state parton and to understand the correction that arise from the finite parton scattering kinematics, branching kinematics, and recoil of the constituents of the QCD medium is the main goal of this Letter.
The rest of this Letter is organized as follows: in section \[framework\] we discuss the theoretical framework $\text{\SCETG}$ for our calculation and demonstrate how the vacuum Altarelli-Parisi splitting kernels can be derived in Soft Collinear Effective Theory. The derivation of the full splitting kernels for in-medium jet production with final-state interactions is discussed in section \[medsplit\]. We elucidate the relation to early soft gluon approximation results and provide analytic formulas for simplified kinematics and medium geometry scenarios. Numerical control over the newly-derived medium-induced splitting intensities is demonstrated in section \[numericssec\]. In this section we also quantify the effects of large-$x$ corrections, finite kinematics, and medium recoil. A brief summary and outlook is presented in section \[conclusions\].
Theoretical Framework {#framework}
=====================
An effective theory, well-suited to describing the propagation of jets in the medium, has been recently developed in Ref. [@Ovanesyan:2011xy]. The Lagrangian of this EFT is given by the sum of the SCET Lagrangian [@Bauer:2000ew; @Bauer:2000yr; @Bauer:2001ct; @Bauer:2001yt] and a term that specifies the interactions of collinear partons in QCD matter: $$\begin{aligned}
&& \!\!\!\!\!\! \mathcal{L}_{\text{\SCETG}}(\xi_n, A_n, A_G)=\mathcal{L}_{\text{SCET}}(\xi_n, A_n)+
\mathcal{L}_{\text{G}}\left(\xi_n, A_n, A_G\right),\nonumber\\
&& \!\!\!\!\!\! \mathcal{L}_{\text{G}}\left(\xi_n, A_n, A_G\right)=\sum_{p,p'}{\mathrm{e}}^{-i(p-p')x}\Big(\bar{\xi}_{n,{p'}}
\Gamma^{\mu,a}_{\rm qqA_G}\frac{\bnslash}{2}\xi_{n,p}\nonumber\\
&& \!\!\!\!\!\! \qquad\qquad\qquad-i \Gamma^{\mu\nu\lambda,abc}_{\rm ggA_G}\,
\left({A}^{b}_{n, p'}\Big)_{\nu}\left({A}^{c}_{n, p}\right)_{\lambda}\right)\,
A_{{\rm G}\, \mu, a}(x)\label{LGdef0}\, . \qquad\nonumber\\\end{aligned}$$ In Ref. [@Ovanesyan:2011xy] the vertexes $\Gamma^{\mu,a}_{\rm qqA_G}, \Gamma^{\mu\nu\lambda,abc}_{\rm ggA_G}$ have been derived for three types of gauge-fixing conditions: covariant, light-cone and hybrid gauges. In the first case we gauge-fix both the physical collinear gluons as well as the Glauber gluons in the covariant gauge. The second choice corresponds to gauge-fixing both fields using the light-cone gauge. The third choice, which appears to be the most convenient from the practical point of view, corresponds to a light-cone gauge for collinear gluons and a covariant gauge for the Glauber gluons. This is a legitimate choice from effective theory point of view, since we are allowed to gauge-fix separate gauge sectors independently. Another way of justifying this gauge choice is factorization between the splitting and the elastic scattering. In this hybrid case both the collinear Wilson line $W$ and the transverse gauge link $T$ [@Ji:2002aa; @Idilbi:2010im; @GarciaEchevarria:2011md] vanish. Gauge invariance of the physics results for the in-medium elastic scattering and radiative energy loss was demonstrated in [@Ovanesyan:2011xy], providing a cross-check on the approach and the newly-derived Feynman rules. It is interesting to note that the same effective theory $\text{\SCETG}$ is relevant for describing the Drell-Yan process, as shown in Ref. [@Bauer:2010cc].
We start from amplitudes for the parton splitting processes: $$\begin{aligned}
A_{q\rightarrow q g}&=&{\left\langle q(p) g(k)\right\rvert}T\,{\mathrm{e}}^{iS}\,\bar{\chi}_n(x_0){\left\lvert q(p_0)\right\rangle},\label{A1def}\\
A_{g\rightarrow gg}&=&{\left\langle g(p) g(k)\right\rvert}T\,{\mathrm{e}}^{iS}\,\mathcal{B}^{\lambda c}(x_0){\left\lvert g(p_0)\right\rangle},\label{A3def}\\
A_{g\rightarrow q \bar{q}}&=&{\left\langle q(p) \bar{q}(k)\right\rvert}T\,{\mathrm{e}}^{iS}\,\mathcal{B}^{\lambda c}(x_0){\left\lvert g(p_0)\right\rangle},\label{A2def}\\
A_{q\rightarrow g q}&=&{\left\langle g(p) q(k)\right\rvert}T\,{\mathrm{e}}^{iS}\,\bar{\chi}_n(x_0){\left\lvert q(p_0)\right\rangle},\label{A4def}\end{aligned}$$ where $\chi, \mathcal{B}$ are collinear gauge invariant SCET fields [@Arnesen:2005nk; @Bauer:2008qu] and the momentum four-vectors, such as $p_0=p+k$, are parametrized in the standard way, consistent with energy momentum conservation and the on-shell condition $p^2=k^2=0$: $$\begin{aligned}
p_0&=&\left[p_0^+, \frac{{\bf{k}}_{\perp}^2}{x(1-x)p_0^+}, {\bf{0}}_{\perp}\right],\\
p&=&\left[(1-x)p_0^+, \frac{{\bf{k}}_{\perp}^2}{(1-x)p_0^+}, -{\bf{k}}_{\perp}\right],\\
k&=&\left[x p_0^+, \frac{{\bf{k}}_{\perp}^2}{x p_0^+}, {\bf{k}}_{\perp}\right].\end{aligned}$$ We use square brackets to indicate the light-cone notation, which we define for arbitrary four-vector $q$ in the following way: $q\equiv\left[q^{+},q^{-},{\boldsymbol{q}}_{\perp}\right]=\left[\bar{n}{\!\cdot\!}q, n{\!\cdot\!}q, {\boldsymbol{q}}_{\perp}\right]$ and $n^{\mu}=\left(1,0,0,1\right), \bar{n}^{\mu}=\left(1,0,0,-1\right)$. The action in [Eq. ]{}-[Eq. ]{} is given by Lagrangian of $\text{\SCETG}$: $$\begin{aligned}
S=i\int \text{d}^{4}x\, \mathcal{L}_{\text{\SCETG}}.\end{aligned}$$ The Lagrangian of $\text{\SCETG}$ [@Idilbi:2008vm; @Ovanesyan:2011xy] is given in [Eq. ]{} and it evolves the created jet and describes the parton splitting processes and the interaction of the parton shower in the medium. The amplitude with $q\bar{q} \leftrightarrow \bar{q}q$ is not shown explicitly.
Restricting ourselves to the SCET Lagrangian without Glauber gluons, we first verify that at tree level we recover the Altarelli-Parisi splitting kernels [@Altarelli:1977zs], which have been originally calculated in full QCD: $$\begin{aligned}
\label{qqg}
\left(\frac{dN}{dx \,d^2{\bf{k}}}_{\perp}\right)_{q\rightarrow qg}&=&
\frac{\alpha_s}{2\pi^2} C_F \frac{1+(1-x)^2}{x}\frac{1}{{\bf{k}}_{\perp}^2},\\
\label{ggg}
\left(\frac{dN}{dx \,d^2{\bf{k}}}_{\perp}\right)_{g\rightarrow gg}&=& \frac{\alpha_s}{2\pi^2}
2 C_A \Big(\frac{1-x}{x}+\frac{x}{1-x} \nonumber\\
&& \qquad\qquad +x(1-x) \Big)\frac{1}{{\bf{k}}_{\perp}^2},\\
\label{gqq}
\left(\frac{dN}{dx \,d^2{\bf{k}}}_{\perp}\right)_{g\rightarrow q\bar{q}}&=
& \frac{\alpha_s}{2\pi^2} T_R\ \left( x^2+(1-x)^2 \right)\frac{1}{{\bf{k}}_{\perp}^2},\\
\left(\frac{dN}{dx \,d^2{\bf{k}}}_{\perp}\right)_{q\rightarrow gq}&=&
\label{qgq}
\left(\frac{dN}{dx \,d^2{\bf{k}}}_{\perp}\right)_{q\rightarrow qg} (x\rightarrow 1-x).\nonumber\\
\end{aligned}$$ We note that we are interested in real splitting processes away from the singular end points $x=0$ and $x=1$. In all expressions above the transverse momentum ${\boldsymbol{k}}_{\perp}$ and the lightcone momentum fraction $x =k^+/p_0^+ = k^+/(p^+ + k^+)$ are for the second final-state parton. The parent parton has no net transverse momentum and ${\boldsymbol{k}}_{\perp}=-{\boldsymbol{p}}_{\perp}$. Note that Eq. (\[qqg\]) and Eq. (\[qgq\]) are interchangeable under $x\rightarrow 1-x$, whereas Eq. (\[ggg\]) and Eq. (\[gqq\]) are symmetric under this substitution. The same symmetries hold for the medium-induced splittings that we derive in section \[medsplit\].
In this paper we use the following terminology: the double differential distribution $dN/dx d^2{\bf{k}}_{\perp}$ we call a splitting kernel, $x dN/dx$ we call a splitting intensity and ${dN}/{dx}$ we call a differential emitted parton number distribution. This terminology applies to both vacuum and medium-induced splittings. The $x-$dependent part of the vacuum splitting kernel we call a splitting function. Since the medium-induced kernel has a more complicated ${\boldsymbol{k}}_{\perp}, x$ correlation structure compared to the simple factorized form in Eq. (10) – Eq. (13) we avoid definition of a similar term in the medium.
![Feynman diagrams contributing to medium-induced splittings at first order in opacity. Red lines corresponds to Glauber gluons. The kinematics and topology are common to all splitting processes: $q\rightarrow qg$, $g\rightarrow gg$, $g\rightarrow q \bar{q}$, $q\rightarrow gq$.[]{data-label="fig:diagrams"}](MEDiag.eps){width="255pt"}
Medium-induced parton splittings {#medsplit}
==================================
To describe the collisional and radiative processes for partons propagating in QCD matter, both single and double Glauber gluon exchanges between the jets and the constituents of the medium must be considered [@Gyulassy:2000er; @Vitev:2007ve; @Ovanesyan:2011xy]. The calculation to first order in opacity, which takes into account the contribution from the splitting induced by the interactions along the trajectory of the parent parton and the dominant interference with the splitting induced by the large $Q^2$ process, is illustrated in figure \[fig:diagrams\]. We do not specify the parent and daughter parton flavors since the topology and kinematics are the same for the splitting processes enumerated in Eqs. (\[qqg\]) - (\[qgq\]). Consequently, all results can be expressed in terms of universal transverse momentum vectors ${\boldsymbol{A}}_{\perp}, {\boldsymbol{B}}_{\perp}, {\boldsymbol{C}}_{\perp}, {\boldsymbol{D}}_{\perp}$ and interference phases $\Omega_1,...,\Omega_5$, defined in [@Ovanesyan:2011xy]: $$\begin{aligned}
&&{\boldsymbol{A}}_{\perp}={\boldsymbol{k}}_{\perp},\,\, {\boldsymbol{B}}_{\perp}={\boldsymbol{k}}_{\perp} + x {\boldsymbol{q}}_{\perp} , \,\,
{\boldsymbol{C}}_{\perp}={\boldsymbol{k}}_{\perp} - (1-x){\boldsymbol{q}}_{\perp},\nonumber\\[1ex]
&& {\boldsymbol{D}}_{\perp}={\boldsymbol{k}}_{\perp}-{\boldsymbol{q}}_{\perp},\,\, \\
&&\Omega_1-\Omega_2=\frac{{\boldsymbol{B}}_{\perp}^2}{p_0^+ x(1-x)}, \,\Omega_1-\Omega_3=\frac{{\boldsymbol{C}}_{\perp}^2}{p_0^+x(1-x)},
\nonumber\\
&& \Omega_2-\Omega_3=\frac{{\boldsymbol{C}}_{\perp}^2-{\boldsymbol{B}}_{\perp}^2}{p_0^+x(1-x)}, \,\,
\Omega_4=\frac{{\boldsymbol{A}}_{\perp}^2}{p_0^+x(1-x)}, \nonumber \\
&& \Omega_5=\frac{{\boldsymbol{A}}_{\perp}^2-{\boldsymbol{D}}_{\perp}^2}{p_0^+x(1-x)},\end{aligned}$$ where $ p_0^+ = p^+ + k^+ $ and the parent parton has no net transverse momentum.
For completeness, we first present below the result for the $q\rightarrow q g$ splitting, calculated in [@Ovanesyan:2011xy] and shown to be gauge invariant:
$$\begin{aligned}
&& \left( \frac{dN}{ dxd^2{\boldsymbol{k}}_{\perp} }\right)_{q\rightarrow qg} = \frac{\alpha_s}{2\pi^2}
C_F \frac{1+(1-x)^2}{x}
\int \frac{d\Delta z}{\lambda_g(z)}
\int d^2{\bf q}_\perp \frac{1}{\sigma_{el}} \frac{d\sigma_{el}^{\; {\rm medium}}}{d^2 {\bf q}_\perp} \;
\Bigg[ \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} {\!\cdot\!}\left( \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} - \frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2} \right)
\nonumber \\
&& \qquad \qquad
\times \big( 1-\cos[(\Omega_1 -\Omega_2)\Delta z] \big) + \frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2} {\!\cdot\!}\left( 2 \frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2}
- \frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2} - \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} \right) \big(1- \cos[(\Omega_1 -\Omega_3)\Delta z] \big) \nonumber \\
&&
\qquad \qquad + \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} {\!\cdot\!}\frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2}
\big( 1 - \cos[(\Omega_2 -\Omega_3)\Delta z] \big)
+ \frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2} {\!\cdot\!}\left( \frac{{\boldsymbol{D}}_{\perp}}{{\boldsymbol{D}}_{\perp}^2} - \frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2} \right)
\big(1-\cos[\Omega_4\Delta z]\big) \nonumber \\
&& \qquad \qquad -\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2} {\!\cdot\!}\frac{{\boldsymbol{D}}_{\perp}}{{\boldsymbol{D}}_{\perp}^2}\big(1-\cos[\Omega_5\Delta z]\big)
+ \frac{1}{N_c^2} \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} {\!\cdot\!}\left( \frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2} -
\frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} \right)
\big( 1-\cos[(\Omega_1 -\Omega_2)\Delta z] \big) \Bigg] \, ,
\label{CohRadSX1} \end{aligned}$$
where $\lambda_g(z)$ is the scattering length of a gluon in the medium and $\left(1/\sigma_{el}\right) \,{d\sigma_{el}^{\; {\rm medium}}}/{d^2 {\bf q}_\perp}$ stands for normalized elastic scattering cross section of a parton in the medium. Even though this quantity varies when parton is a quark or a gluon, in the high energy limit, when the $t-$ channel dominates the elastic scattering, this normalized cross section does not change significantly.
Using the Feynman rules of in the hybrid gauge and the Feynman diagrams exactly analogous to the case of $q\rightarrow q g$ splitting considered in [@Ovanesyan:2011xy] and shown in figure \[fig:diagrams\], we derive the remaining parton splittings in the medium. The calculations are non-trivial and facilitated by intermediate results in [@Ovanesyan:2011xy]. As discussed in section \[framework\], the medium-induced splitting for $q \rightarrow gq$ can be obtained from Eq. (\[CohRadSX1\]) with the substitution $x\rightarrow 1-x$. Here, we skip the explicit expression for brevity. The remaining two splittings from a parent gluon are as follows:
$$\begin{aligned}
&& \left( \frac{dN}{ dxd^2{\boldsymbol{k}}_{\perp} }\right)_{ \left\{ \begin{array}{c} g \rightarrow gg\\ g\rightarrow q\bar{q} \end{array} \right\} }
=
\left\{ \begin{array}{c} \frac{\alpha_s}{2\pi^2} \, 2 C_A \left(\frac{x}{1-x}+\frac{1-x}{x}+x(1-x) \right) \\[1ex]
\frac{\alpha_s}{2\pi^2} T_R \left( x^2+(1-x)^2 \right) \end{array} \right\}
\int {d\Delta z} \left\{ \begin{array}{c} \frac{1}{\lambda_g(z)} \\[1ex] \frac{1}{\lambda_q(z)} \end{array} \right\}
\int d^2{\bf q}_\perp \frac{1}{\sigma_{el}} \frac{d\sigma_{el}^{\; {\rm medium}}}{d^2 {\bf q}_\perp} \; \nonumber \\
&& \qquad \qquad
\times \Bigg[ 2\, \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2} {\!\cdot\!}\left(\frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2}-\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}\right) \big( 1-\cos[(\Omega_1 -\Omega_2)\Delta z] \big)
+2\, \frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2} {\!\cdot\!}\left(\frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2}-\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}\right) \big( 1-\cos[(\Omega_1 -\Omega_3)\Delta z] \big)
\nonumber \\
&&\qquad \qquad + \left\{ \begin{array}{c} - \frac{1}{2} \\[1ex] \frac{1}{N_c^2-1} \end{array} \right\}
\Bigg(2 \frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2}{\!\cdot\!}\left(\frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2}-\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}\right)\big(1-\cos[(\Omega_1-\Omega_2)\Delta z ]\big)
\nonumber\\
&&\qquad \qquad +2\,\frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2}{\!\cdot\!}\left(\frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2}-\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}\right)\big(1-\cos[(\Omega_1-\Omega_3)\Delta z]\big)-2\,\frac{{\boldsymbol{C}}_{\perp}}{{\boldsymbol{C}}_{\perp}^2}{\!\cdot\!}\frac{{\boldsymbol{B}}_{\perp}}{{\boldsymbol{B}}_{\perp}^2}\big(1-\cos[(\Omega_2-\Omega_3)\Delta z]\big)
\nonumber\\
&&\qquad \qquad +2\,\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}{\!\cdot\!}\left(\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}-\frac{{\boldsymbol{D}}_{\perp}}{{\boldsymbol{D}}_{\perp}^2}\right)\big(1-\cos[\Omega_4\Delta z]\big)+2\,\frac{{\boldsymbol{A}}_{\perp}}{{\boldsymbol{A}}_{\perp}^2}{\!\cdot\!}\frac{{\boldsymbol{D}}_{\perp}}{{\boldsymbol{D}}_{\perp}^2}\big(1-\cos[\Omega_5\Delta z]\big)\Bigg) \Bigg] \, ,
\label{CohRadSX2} \end{aligned}$$
where $\lambda_q(z)$ is the scattering length of a quark in the medium and same comment applies to the quantity $\left(1/\sigma_{el}\right) \,{d\sigma_{el}^{\; {\rm medium}}}/{d^2 {\bf q}_\perp}$ as after [Eq. ]{}. Note that up to the overall vacuum-like splitting functions and color factors reflected both in the mean free paths (quark versus gluon) and the corrections relevant beyond the small-$x$ approximation, the structure of the answers is very similar. The symmetry of $g\rightarrow gg, g\rightarrow q\bar{q}$ splitting kernels under $x \rightarrow 1-x $ is most easily verified explicitly by realizing that the parton scattering cross section in the medium is invariant under ${\boldsymbol{q}}_\perp \rightarrow -{\boldsymbol{q}}_\perp$.
The basic features of the medium-induced parton splitting kernels are:
- In QCD, for parent quark they factorize from the hard scattering cross section up to a standard integral convolution [@Ovanesyan:2011xy]. For parent gluons non-trivial spin correlation are present analogous to the vacuum case [@Catani:1998nv].
- They are proportional to their vacuum Altarelli-Parisi splitting functions [@Altarelli:1977zs].
- The in-medium splittings are gauge-invariant, as they should be, since the underlying jet production process itself is gauge-invariant [@Ovanesyan:2011xy].
- The splitting kernels depend on the properties of the QCD matter and vanish when the size or density of the medium vanish. The functions derived here are only valid for final-state interactions [@Vitev:2007ve].
It is instructive to verify that in the small-$x$ limit only two of the four medium-induced splitting intensities survive and this allows for the standard energy loss interpretation of jet quenching: $$\begin{aligned}
&& \!\!\!\!\!\!\! x \left( \frac{dN}{ dx}\right)_{ \left\{ \begin{array}{c} q \rightarrow qg \\
g\rightarrow gg \\ g \rightarrow q\bar{q} \\ q \rightarrow gq\\ \end{array} \right\} } = \frac{\alpha_s}{\pi^2}
\left\{ \begin{array}{c} C_F[ 1+ {\cal O}(x) ] \\ C_A[ 1+ {\cal O}(x) ] \\
T_R [0 + \frac{x}{2} + {\cal O}(x^2)] \\ C_F [0 + \frac{x}{2} + {\cal O}(x^2)]
\end{array} \right\} \nonumber \\
&& \qquad \times \int {d\Delta z} \left\{ \begin{array}{c} \frac{1}{\lambda_g(z)} \\[1ex] \frac{1}{\lambda_g(z)} \\
\frac{1}{\lambda_q(z)} \\ \frac{1}{\lambda_q(z)} \end{array} \right\}
\int d^2{\bf k}_\perp d^2{\bf q}_\perp \frac{1}{\sigma_{el}} \frac{d\sigma_{el}^{\; {\rm medium}}}{d^2 {\bf q}_\perp} \; \nonumber \\
&& \qquad \times \frac{2 {\boldsymbol{k}}_{\perp} \cdot {\boldsymbol{q}}_{\perp} }{{\boldsymbol{k}}_{\perp}^2 ({\boldsymbol{k}}_{\perp}-{\boldsymbol{q}}_{\perp})^2}
\left [ 1-\cos \frac{ ({\boldsymbol{k}}_{\perp}-{\boldsymbol{q}}_{\perp})^2}{xp^+_0} \Delta z \right].
\label{smallx}\end{aligned}$$ In this limit the interference structure for all medium-induced splitting intensities is the same. Furthermore, in the small-$x$ limit the last two splittings are suppressed (${\mathcal {O}}(x)$) relative to the first two. We keep the first correction for numerical comparison only. The color structure for the in-medium interactions also simplifies in this limit and is determined by the flavor of the small-$x$ parton in the final state. Specifically, the first two in-medium splittings are proportional to $1/\lambda_g$ and the second two are proportional to $1/\lambda_q$. In deriving these results, we have used relation: $\lambda_q/\lambda_g = C_A/C_F$, which follows from the leading order perturbation theory approximation. As expected, in the small-$x$ emission limit our results coincide exactly with the intensity derived (or neglected when the leading term is 0) in [@Gyulassy:2000er].
In section \[numericssec\] we will study numerically the in-medium splittings derived here with an emphasis on going beyond the traditional small-$x$ approximation and on including medium recoil. The remaining part of the current section is devoted to deriving analytic formulas for the inclusive splitting intensity $x (dN/dx) $ under certain idealized assumptions. This will, in turn, allow us to obtain fully analytic formulas that can be used to benchmark the realistic numerical calculation.
A useful starting point for integrating the splitting kernels analytically over the transverse momenta is the following master formula: $$\begin{aligned}
&&\int d^2{\boldsymbol{k}}_{\perp}d^2{\boldsymbol{q}}_{\perp} \frac{1}{\sigma_{el}} \frac{d\sigma_{el}}{d^2 {\bf q}_\perp}
\frac{2 {\boldsymbol{k}}_{\perp} \cdot {\boldsymbol{q}}_{\perp} }{{\boldsymbol{k}}_{\perp}^2 ({\boldsymbol{k}}_{\perp}-{\boldsymbol{q}}_{\perp})^2}
\nonumber\\
&& \qquad \qquad \times \left [ 1-\cos(\alpha({\boldsymbol{k}}_{\perp}-{\boldsymbol{q}}_{\perp})^2) \right] =f[\alpha\mu^2], \qquad
\label{masterformula}\end{aligned}$$ where a specific form for elastic cross section was assumed as explained below and shown in [Eq. ]{}, and function $f[x]$ equals: $$\begin{aligned}
f[x]&=& 2\pi\Big[ \gamma_{\text{E}}+\ln(x)+\frac{\pi}{2}\sin(x)-\cos(x)\text{Ci}(x) \nonumber \\
&& -\sin(x)\text{Si}(x) \Big].
\label{comfundef}\end{aligned}$$ Two assumptions are already made at this level. First, we took the limits of integration on ${\boldsymbol{k}}_{\perp}, {\boldsymbol{q}}_{\perp}$ to infinity. In reality, phase space cuts affect the cross section and we study this effect numerically in the next section. Second, we neglected the recoil effect in the medium. In that approximation the normalized cross section equals: $$\begin{aligned}
\frac{1}{\sigma_{el}} \frac{d\sigma_{el}}{d^2 {\bf q}_\perp}
=\frac{\mu^2}{\pi({\boldsymbol{q}}_{\perp}^2+\mu^2)^2}.\label{infsigmael}\end{aligned}$$ The effects of finite medium recoil are also studied in the next section numerically. It turns out that using [Eq. ]{} it is possible to calculate ${\boldsymbol{k}}_{\perp}, {\boldsymbol{q}}_{\perp}$ integrals in all in-medium splittings [Eq. ]{}-[Eq. ]{}. The result is rather compact and can be expressed in terms of the function $f[x]$ defined in [Eq. ]{}:
$$\begin{aligned}
&&x\left(\frac{dN}{dx}\right)^{\infty}_{q\rightarrow qg}= x\frac{\alpha_s}{2\pi^2} C_F
\frac{ 1 + (1-x)^2}{x} \int\frac{d\Delta z}{\lambda_g(z)}
\frac{f[\beta]+f[\beta(1-x)^2]-\frac{1}{N_c^2}f[\beta x^2]}{2}, \label{split1analytics}\\
&&x\left(\frac{dN}{dx}\right)^{\infty}_{g\rightarrow gg}= x\frac{\alpha_s}{2\pi^2}
2C_A \left(\frac{x}{1-x}+\frac{1-x}{x}+x(1-x)\right)
\int\frac{d\Delta z}{\lambda_g(z)}\frac{f[\beta x^2]+f[\beta(1-x)^2]+f[\beta]}{2},\\
&&x\left(\frac{dN}{dx}\right)^{\infty}_{g\rightarrow q\bar{q}}= x\frac{\alpha_s}
{2 \pi^2} T_R \left(x^2+(1-x)^2\right)\int\frac{d\Delta z}{\lambda_q(z)}
\left[{\frac{N_c^2}{N_c^2-1}\left(f[\beta x^2]+f[\beta(1-x)^2]\right)-\frac{1}{N_c^2-1}f[\beta]}\right],\\
&&x\left(\frac{dN}{dx}\right)^{\infty}_{q\rightarrow gq}=x \left(\frac{dN}{dx}\right)^{\infty}_{q\rightarrow qg}(x\rightarrow 1-x), \label{split4analytics}\end{aligned}$$
where the superscript $\infty$ stands for infinite limits of integrations for ${\boldsymbol{k}}_{\perp}, {\boldsymbol{q}}_{\perp}$, and: $$\beta\equiv\frac{\mu^2 \Delta z}{p_0^+x(1-x)}.$$
\
In order to perform the remaining $\Delta z$ integral, one has to specify the geometry and an expansion scenario for the QCD medium that the partons traverse. Even the simplest realistic model of the medium, which includes the Glauber nuclear geometry and Bjorken expansion, requires numerical evaluation of the in-medium splitting kernels. One can validate the numerical simulation techniques by comparing the results to closed-form analytic formulas for uniform QCD matter, where the parton mean free paths $\lambda_q,\, \lambda_g$ and trajectory length $L$ are fixed. Thus, it is instructive to have analytical formulas for splitting intensity for uniform QCD matter. By assuming constant $\lambda_{q, g}$ as a function of $\Delta z$, and integrating expressions in [Eq. ]{}-[Eq. ]{} over $0<\Delta z<L$, we obtain the following differential splitting intensities:
$$\begin{aligned}
\label{finalanalytics01}
&&x\left(\frac{dN}{dx}\right)^{\infty, \text{static}}_{q\rightarrow qg}=
x\frac{\alpha_s}{2\pi} C_F \frac{ 1+(1- x)^2 }{x}
\frac{ L}{\lambda_g} \left( g[\gamma]+g[\gamma(1-x)^2]-\frac{1}{N_c^2}g[\gamma x^2] \right),\\
&&x\left(\frac{dN}{dx}\right)^{\infty, \text{static}}_{g\rightarrow gg}= x\frac{\alpha_s}{2\pi} 2 C_A
\left(\frac{x}{1-x}+\frac{1-x}{x}+x(1-x)\right) \frac{ L}{\lambda_g}
\left( g[\gamma x^2]+g[\gamma(1-x)^2]+g[\gamma] \right) ,
\label{finalanalytics02}\\
&&x\left(\frac{dN}{dx}\right)^{\infty, \text{static}}_{g\rightarrow q\bar{q}}
=x\frac{\alpha_s}{2\pi} T_R \left(x^2+(1-x)^2\right) \frac{ L}{\lambda_q} \left[{\frac{2N_c^2}{N_c^2-1}
\left(g[\gamma x^2]+g[\gamma(1-x)^2]\right)-\frac{2}{N_c^2-1}g[\gamma]}\right] ,
\label{finalanalytics03}\end{aligned}$$
where the function $g$ is given by: $$\begin{aligned}
w \, g[w]&\equiv& \frac{\pi}{2}(1-\cos(w))+(\gamma_{\text{E}}-1)w+w \ln(w) \nonumber \\
&&+\text{Si}(w)\cos(w)-\text{Ci}(w)\sin (w) ,
\label{finalanalytics1}\end{aligned}$$ and $\gamma$ is defined as: $$\begin{aligned}
\gamma &\equiv& \frac{\mu^2 L}{p^+_0 x(1-x)}.\end{aligned}$$ The intensity spectrum for the last splitting $q\rightarrow gq$ can be obtained from substitution $x\rightarrow 1-x$ in the $q\rightarrow qg$ splitting and is given in [Eq. ]{}.
Numerical results {#numericssec}
=================
In this section we study the effects of kinematic cuts and recoil of the medium by evaluating $dN/dx$ numerically. In so doing, we demonstrate control over the numerical evaluation, keeping in mind that future applications will require such approach to incorporate the finite kinematics, the spatially non-uniform and time-dependent density of the QCD matter, and the recoil of the in-medium partons. For each splitting we consider the full result given by [Eq. ]{} - [Eq. ]{} and compare it to the small-$x$ limit presented in [Eq. ]{}. In this paper we consider a medium of uniform density for simplicity and set the parameters of the simulation as follows: the typical inverse range of the parton scattering in the medium is $\mu=0.75{~\mathrm{GeV}}$, the size of the QCD medium is $L=5\text{ fm}$, the gluon mean free path in matter is $\lambda_g=1\text{ fm}$, and the parent parton energy is $E_0 = p^+_0/2=100\ {~\mathrm{GeV}}$.
For infinite limits of the ${\boldsymbol{k}}_{\perp}, {\boldsymbol{q}}_{\perp}$ integrations, ignoring the medium recoil effects and assuming static QCD matter, we checked numerically our analytic formulas in [Eq. ]{} - [Eq. ]{}. We found perfect agreement that validates the numerical integration methods and the analytic results. This can be seen from figure \[fig:numerics\]. Solid lines represents the analytic results of [Eq. ]{} - [Eq. ]{}. Dashed lines represent numerical results. Our conclusions are valid for both the full in-medium splitting intensity $x(dN/dx)$ and its small-$x$ limit. Note that for such comparison to be possible we have retained the subleading ${\cal O} (x)$ term for the $g \rightarrow q\bar{q}$ and $q \rightarrow gq$ processes. As expected, the deviation between the full in-medium splittings (red and blue lines) and their small-$x$ approximation (green and black lines) is the largest as $x \rightarrow 1$. For intermediate $x \sim 0.5$ the deviation is on the order of a factor of 2 and changes sign.
\
In figure \[fig:numerics2\] we present the comparison of the splitting intensities without transverse momentum cuts and without parton recoil in the medium (solid red curve) to three different cut scenarios. In all three cases we use the same cut on $k_{\text{max}}=\sqrt{Q^2 x(1-x)} $, which is unambiguous, and we choose $Q=E_0$. The three scenarios for the ${\boldsymbol{q}}_{\perp}$ cut are: *a*) the dashed green curve corresponds to $q_{\text{max}}=\sqrt{\mu E_0/2}$, *b*) the dashed blue curve corresponds to $q_{\text{max}}=\sqrt{2\mu E_0}$, *c*) the dot-dashed black curve corresponds to the exact phase space, given by $2\rightarrow2$ scattering available phase space. Finally, the solid black curve includes the recoil effect which is calculated by substituting the normalized cross section in [Eq. ]{} by the $2\rightarrow 2$ $t$-channel differential cross section, which can be found in Eqs. (3.2-3.3) in Ref. [@Ovanesyan:2011xy]. From the definition of cut on ${\boldsymbol{k}}_{\perp}$, for small $x$ we have $k_{\text{max}}\sim Q\sqrt{x}\rightarrow 0$. From figure \[fig:numerics2\] one can see that for small $x$ the cut on ${\boldsymbol{k}}_{\perp}$ is the only one that affects the splitting intensity, since all three types of cuts on ${{\boldsymbol{q}}_{\perp}}$ give practically identical results. As far as the intermediate $x$ region is concerned, the cut on ${\boldsymbol{k}}_{\perp}$ does not play a significant role since from the definition $k_{\text{max}}(x\sim1/2)\sim Q/2$, thus the observable difference must be attributed to the cut on ${\boldsymbol{q}}_{\perp}$ for this region. The kinematic cut on ${\boldsymbol{q}}_{\perp}$, however, can lead to a factor of 2 variation of the in-medium parton splitting intensities at intermediate $x$. Note that for intermediate $x$ the third cut on ${\boldsymbol{q}}_{\perp}$, which corresponds to full kinematics but retains the $1/{\boldsymbol{q}}_{\perp}^4$ dependence of the scattering cross section [Eq. ]{}, agrees perfectly with the uncut solid red curve. In this case, cuts alone (in the sense of full kinematics) do not affect significantly the in-medium branching processes. We find that what affects the splitting intensity is the deviation between the exact scattering cross section from [@Ovanesyan:2011xy] and the approximate power-law form in [Eq. ]{}. This is illustrated in figure \[fig:numerics2\] by the solid black curve that pushes the intensity of the medium-induced branching processes down when compared to the dot-dashed black curve. We finally note that if one wishes to simplify the calculation and use the approximate form [Eq. ]{} for in-medium parton scattering the most adequate transverse momentum cut would be $q_{\text{max}}=\sqrt{\mu E_0}$. Finally, for $x\rightarrow 1$, we find that $k_{\text{max}}\sim Q\sqrt{1-x}\rightarrow 0$, and all the splitting intensities with phase space cuts turn over at large enough $x$, which is not visible in figure \[fig:numerics2\] because we do not plot values of $x$ very close to 1.
Numerically, all effects: finite $x$, phase space cuts, recoil effect are of the same order at high energies. In addition, we observe that parton recoil, similar to finite $x$ corrections appears at intermediate $x$, while phase space cuts play role both for small $x$ and intermediate $x$.
Conclusions
===========
In this Letter we derived the medium-induced parton splittings for quarks and gluons produced in large $Q^2$ scattering processes that subsequently traverse a region of dense QCD matter and undergo final-state interactions using a recently constructed effective theory $\text{\SCETG}$ [@Ovanesyan:2011xy]. Our results include both the contributions from the in-medium parton scattering and their dominant interference with the vacuum-like branching processes [@Gyulassy:2000er; @Vitev:2007ve]. These formulas are valid for an arbitrary splitting parameter $x$ and include all leading terms consistent with the power counting of $\text{\SCETG}$ [@Ovanesyan:2011xy]. Our results for the $q\rightarrow qg,\, q\rightarrow gq,\, g\rightarrow gg, \, g\rightarrow q\bar{q}$, including Landau-Pomeranchuck-Migdal coherence and interference effects, are presented in [Eq. ]{}-[Eq. ]{} and are the main new findings reported in this Letter. We verified in [Eq. ]{} that in the small-$x$ approximation our formulas simplify considerably and reduce to the soft gluon emission results of the Gyulassy-Levai-Vitev approach to inelastic parton interactions in dense QCD matter [@Gyulassy:2000er; @Vitev:2007ve]. In this case, to leading power in $1/x$, only two medium-induced splitting kernels survive that do not change the flavor of the leading parton and have a natural interpretation in terms of parton energy loss.
Neglecting the recoil of the partons in the QCD medium and ignoring the phase space cuts, we derived fully analytic formulas for the in-medium splitting intensity, given in [Eq. ]{}-[Eq. ]{}. We see the main utility of these formulas as a convenient cross check for the numerical simulations, which we have demonstrated in this Letter. Our full results provide a basis for further improvements in the jet quenching phenomenology [@Neufeld:2010fj]-[@Mironov:2011zz] by including the following effects: $a)$ finite $x$ corrections in the $q\rightarrow qg$ and $g\rightarrow gg$ splittings, consistent with the power counting of , $b)$ new splittings for $g\rightarrow q\bar{q}$, $q\rightarrow gq$ from coherent final-state interactions, $c)$ exact parton recoil kinematics and, $d)$ exact phase space cuts. All of the above effects can be of the same order, as one can see from our numerical simulation results in section \[numericssec\]. This program of improving the theoretical accuracy of jet quenching simulations is especially interesting in light of recent RHIC and LHC heavy ion results [@Aamodt:2010jd]-[@Dainese:2011vb], [@Salur:2009vz]-[@Chatrchyan:2011ua] and we plan to show first phenomenological applications in the near future.
This research is supported by the US Department of Energy, Office of Science, under Contract No. DE-AC52-06NA25396 and in part by the LDRD program at LANL and the JET topical collaboration.
[99]{}
I. Arsene [*et al.*]{} \[BRAHMS Collaboration\], Nucl. Phys. A [**757**]{}, 1 (2005) \[arXiv:nucl-ex/0410020\].
B. Back [*et al.*]{} \[ PHOBOS Collaboration \], Nucl. Phys. A [**757**]{}, 28 (2005)
J. Adams [*et al.*]{} \[STAR Collaboration\], Nucl. Phys. A [**757**]{}, 102 (2005) \[arXiv:nucl-ex/0501009\].
K. Adcox [*et al.*]{} \[PHENIX Collaboration\], Nucl. Phys. A [**757**]{}, 184 (2005) \[arXiv:nucl-ex/0410003\].
K. Aamodt [*et al.*]{} \[ALICE Collaboration\], Phys. Lett. B [**696**]{}, 30 (2011) \[arXiv:1012.1004 \[nucl-ex\]\].
A. S. Yoon for the CMS collaboration, arXiv:1107.1862 \[nucl-ex\].
A. Dainese for the ALICE Collaboration, arXiv:1106.4042 \[nucl-ex\]. M. Gyulassy, I. Vitev, X. N. Wang and B. W. Zhang, arXiv:nucl-th/0302077.
I. Vitev and B. W. Zhang, Phys. Rev. Lett. [**104**]{}, 132001 (2010) \[arXiv:0910.1090 \[hep-ph\]\].
R. B. Neufeld, I. Vitev and B. W. Zhang, Phys. Rev. C [**83**]{}, 034902 (2011) \[arXiv:1006.2389 \[hep-ph\]\].
I. P. Lokhtin, A. V. Belyaev and A. M. Snigirev, Eur. Phys. J. C [**71**]{}, 1650 (2011) \[arXiv:1103.1853 \[hep-ph\]\]. G. Y. Qin and B. Muller, Phys. Rev. Lett. [**106**]{}, 162302 (2011) \[arXiv:1012.5280 \[hep-ph\]\].
Y. He, I. Vitev and B. W. Zhang, arXiv:1105.2566 \[hep-ph\].
R. B. Neufeld and I. Vitev, arXiv:1105.2067 \[hep-ph\].
C. Mironov, M. Castro, P. Constantin, G. J. Kunde and R. Vogt, J. Phys. G [**38**]{}, 065002 (2011).
R. B. Neufeld, arXiv:1108.6297 \[nucl-th\].
S. Salur, Nucl. Phys. A [**830**]{}, 139C (2009) \[arXiv:0907.4536 \[nucl-ex\]\].
M. Ploskon \[STAR Collaboration\], Nucl. Phys. A [**830**]{}, 255C (2009) \[arXiv:0908.1799 \[nucl-ex\]\].
J. Jia, Nucl. Phys. A [**855**]{}, 92 (2011) \[arXiv:1012.0858 \[nucl-ex\]\].
A. Angerami and A. Collaboration, arXiv:1108.5191 \[nucl-ex\].
S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], Phys. Rev. C [**84**]{}, 024906 (2011) \[arXiv:1102.1957 \[nucl-ex\]\]. G. Aad [*et al.*]{} \[Atlas Collaboration\], Phys. Rev. Lett. [**105**]{}, 252303 (2010) \[arXiv:1011.6182 \[hep-ex\]\]. S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], Phys. Rev. Lett. [**106**]{}, 212301 (2011) \[arXiv:1102.5435 \[nucl-ex\]\].
I. Vitev, S. Wicks and B. W. Zhang, JHEP [**0811**]{}, 093 (2008) \[arXiv:0810.2807 \[hep-ph\]\].
C. W. Bauer, S. Fleming and M. E. Luke, Phys. Rev. D [**63**]{}, 014006 (2000) \[arXiv:hep-ph/0005275\]. C. W. Bauer, S. Fleming, D. Pirjol and I. W. Stewart, Phys. Rev. D [**63**]{}, 114020 (2001) \[arXiv:hep-ph/0011336\].
C. W. Bauer, D. Pirjol and I. W. Stewart, Phys. Rev. D [**65**]{}, 054022 (2002) \[arXiv:hep-ph/0109045\].
C. W. Bauer and I. W. Stewart, Phys. Lett. B [**516**]{}, 134 (2001) \[arXiv:hep-ph/0107001\].
S. Fleming, A. H. Hoang, S. Mantry and I. W. Stewart, Phys. Rev. D [**77**]{}, 114003 (2008) \[arXiv:0711.2079 \[hep-ph\]\].
T. Becher and M. D. Schwartz, JHEP [**0807**]{}, 034 (2008) \[arXiv:0803.0342 \[hep-ph\]\].
A. Hornig, C. Lee and G. Ovanesyan, JHEP [**0905**]{}, 122 (2009) \[arXiv:0901.3780 \[hep-ph\]\].
R. Abbate, M. Fickinger, A. H. Hoang, V. Mateu and I. W. Stewart, Phys. Rev. D [**83**]{}, 074021 (2011) \[arXiv:1006.3080 \[hep-ph\]\].
T. Becher, M. Neubert and G. Xu, JHEP [**0807**]{}, 030 (2008) \[arXiv:0710.0680 \[hep-ph\]\].
I. W. Stewart, F. J. Tackmann and W. J. Waalewijn, Phys. Rev. D [**81**]{}, 094035 (2010) \[arXiv:0910.0467 \[hep-ph\]\].
T. Becher and M. D. Schwartz, JHEP [**1002**]{}, 040 (2010) \[arXiv:0911.0681 \[hep-ph\]\].
A. Idilbi and A. Majumder, Phys. Rev. D [**80**]{}, 054022 (2009) \[arXiv:0808.1087 \[hep-ph\]\].
F. D’Eramo, H. Liu and K. Rajagopal, arXiv:1006.1367 \[hep-ph\].
M. Gyulassy, P. Levai and I. Vitev, Phys. Rev. D [**66**]{}, 014005 (2002) \[arXiv:nucl-th/0201078\].
J. w. Qiu and I. Vitev, Phys. Lett. B [**570**]{}, 161 (2003) \[arXiv:nucl-th/0306039\].
G. Ovanesyan and I. Vitev, JHEP [**1106**]{}, 080 (2011) \[arXiv:1103.1074 \[hep-ph\]\]. I. Vitev, Phys. Rev. C [**75**]{}, 064906 (2007) \[arXiv:hep-ph/0703002\].
X. d. Ji and F. Yuan, Phys. Lett. B [**543**]{}, 66 (2002) \[arXiv:hep-ph/0206057\].
A. Idilbi and I. Scimemi, Phys. Lett. B [**695**]{}, 463 (2011) \[arXiv:1009.2776 \[hep-ph\]\].
M. Garcia-Echevarria, A. Idilbi and I. Scimemi, Phys. Rev. D [**84**]{}, 011502 (2011) \[arXiv:1104.0686 \[hep-ph\]\].
C. W. Bauer, B. O. Lange and G. Ovanesyan, JHEP [**1107**]{}, 077 (2011) \[arXiv:1010.1027 \[hep-ph\]\].
C. M. Arnesen, J. Kundu and I. W. Stewart, Phys. Rev. D [**72**]{}, 114002 (2005) \[arXiv:hep-ph/0508214\].
C. W. Bauer, O. Cata and G. Ovanesyan, arXiv:0809.1099 \[hep-ph\].
G. Altarelli and G. Parisi, Nucl. Phys. B [**126**]{}, 298 (1977).
S. Catani and M. Grazzini, Phys. Lett. B [**446**]{}, 143 (1999) \[arXiv:hep-ph/9810389\].
M. Gyulassy, P. Levai and I. Vitev, Nucl. Phys. B [**594**]{}, 371 (2001) \[arXiv:nucl-th/0006010\].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We develop the theory of interstellar scintillation as caused by an irregular plasma having a power-law spatial density spectrum with a spectral exponent of $\beta = 4$ corresponding to a medium with abrupt changes in its density. An “outer scale” is included in the model representing the typical scale over which the density of the medium remains uniform. Such a spectrum could be used to model plasma shock fronts in supernova remnants or other plasma discontinuities. We investigate and develop equations for the decorrelation bandwidth of diffractive scintillations and the refractive scintillation index and compare our results with pulsar measurements. We consider both a medium concentrated in a thin layer and an extended irregular medium. We conclude that the $\beta = 4$ model gives satisfactory agreement for many diffractive measurements, in particular the VLBI meaurements of the structure function exponent between 5/3 and 2. However, it gives less satisfactory agreement for the refractive scintillation index than does the Kolmogorov turbulence spectrum. The comparison suggests that the medium consists of a pervasive background distribution of turbulence embedded with randomly placed discrete plasma structures such as shocks or HII regions. This can be modeled by a composite spectrum following the Kolmogorov form at high wavenumbers and steepening at lower wavenumbers corresponding to the typical (inverse) size of the discrete structures. Such a model can also explain the extreme scattering events. However, lines of sight through the enhanced scattering prevalent at low galactic latitudes are accurately described by the Kolmogorov spectrum in an extended medium and do not appear to have a similar low-wavenumber steepening.'
author:
- 'H. C. Lambert'
- 'B. J. Rickett'
- 'DRAFT:'
title: RADIO SCINTILLATION DUE TO DISCONTINUITIES IN THE INTERSTELLAR PLASMA DENSITY
---
INTRODUCTION
============
It is well known that radio waves propagating in the interstellar medium (ISM) are scattered by the irregularities in the Galactic electron density. The scattering in turn gives rise to a number of observable phenomena. Among others, these include angular broadening and intensity fluctuations (both in time and frequency) of compact radio sources. While a nuisance in many radio astronomical observations, these phenomena can be used to investigate the nature of the irregularities in the interstellar plasma density. These density irregularities are in turn believed to follow the fluctuations in the interstellar kinetic and magnetic energies. Ideally one would like to invert observations of radio scintillation and scattering to determine the statistics of the plasma density. As noted by Narayan (1992) this inverse problem is not well posed and one must rely on modeling methods. A complete prediction of scintillation observables requires an [*a priori*]{} knowledge of both the form of the spatial power spectrum of the electron density fluctuations and the distribution of the scattering material along the line of sight. For a given profile of the distribution of the scattering material, one may compare observations and predictions and so constrain the functional form of the spectrum. The power spectrum provides useful insight into the physics of the plasma irregularities. Hence a knowledge of the form of the density spectrum becomes central in both predicting scintillation phenomena and understanding the physics of the interstellar plasma. In this paper, we revisit the investigation of the form of the density spectrum.
A commonly used model for the density spectrum has been based on a power-law model with a large range between “inner” and “outer” scales (e.g. Rickett 1977): $$P_{N_e}(q) =
\frac{C_{N_e}^2(z)}{(q^2 + \kappa_o^2)^\beta}
\exp\left[ -\frac{q^2}{4\kappa_i^2} \right]
\, \mbox{.}
\label{eq:extpowerlaw}
$$ Here $q$ is the magnitude of the three-dimensional wavenumber $\vec{q}$.
$C_{N_e}^2(z)$ denotes the strength of fluctuations (with a weak dependence on distance $z$). $\beta$ is the spectral exponent, and $\kappa_i^{-1} = L_i$ and $\kappa_o^{-1} = L_o$ are the inner and outer scales respectively. For $\kappa_o << q << \kappa_i$, we obtain the [*simple*]{} power-law model: $P_{N_e}(q) = C_{N_e}^2 q^{-\beta}$. Armstrong, Rickett, & Spangler (1995) have constructed an empirical density spectrum by combining radio scintillation observations in the local ISM ($\simless$ 1 kpc) with measurements of the differential Faraday rotation angles and large-scale electron density gradients. They have shown the power spectrum to be consistent with a simple Kolmogorov power-law model ($\beta=11/3$) over an astronomical 10 orders of magnitude in wavenumber scale ($10^{-18}$m$^{-1} < q < 10^{-8}$m$^{-1}$). The Kolmogorov spectrum in density suggests a turbulent cascade in the magnetic and kinetic energies. This has lead to several theoretical investigations of the generation and maintenance of hydromagnetic turbulence in the ISM (e.g. Pouquet, 1978; Higdon 1984 & 1986; Biskamp, 1993: Sridhar & Goldreich 1994; Goldreich & Sridhar 1995 and 1997). However, the Armstrong et al. study combined observations from many lines of sight and the scatter among them leaves a substantial uncertainty in the exponent $\beta$. A list of symbols is given in Table 3.
In spite of the positive evidence for the simple Kolmogorov spectrum, substantial observational inconsistencies remain. For instance, long-term refractive intensity scintillations of some pulsars have modulation indices as much as a factor of 2 larger than predicted by the simple Kolmogorov model (cf.Gupta, Rickett, & Coles 1993). Other discrepancies are revealed in the diffractive dynamic spectra of pulsars. On some occasions, periodic fringes are observed, which are not predicted by the simple Kolmogorov model (cf. Roberts & Ables 1982; Cordes & Wolszczan 1986; Rickett, Lyne, & Gupta 1997; Gupta, Bhat & Rao, 1999); in addition, sloping bands in the dynamic spectra often persist longer than predicted by the model (cf.Gupta, Rickett, & Lyne 1994; Bhat, Rao & Gupta, 1999b Bhat, Gupta & Rao, 1999c). Further for some pulsars, the decorrelation bandwidth has larger amplitude variations than predicted for the Kolmogorov spectrum (Bhat, Gupta & Rao, 1999c). Such anomalies suggest the presence of large refractive structures giving rise to the focusing and defocusing of the scattered ray bundles. The interference of the ray bundles can also explain the occasional fringes observed in the dynamic spectra of some pulsars. Given the relatively frequent occurrence of such events, one can ask whether they can be considered as mere occasional anomalies or should be considered as a widespread phenomenon intrinsic to the spectrum on a grand scale.
The observational inconsistencies suggest the need for an enhancement in the power on the large “refractive” (10$^{11}$ m to 10$^{12}$ m) spatial scales relative to the power on the small “diffractive” (10$^7$ m to 10$^8$ m) scales. There are several means by which this ratio may be enhanced. One is to include the inner scale cut-off in the density spectrum, which reduces the power at small scales (Coles et al. 1987); these authors proposed inner scales of $10^8-10^9$ m., though this does not correspond to any obvious physical scale. Physically, the inner scale corresponds to the scale at which the turbulent cascade dissipates and becomes a source of heating for the plasma (Spangler, 1991). The value of the inner scale is largely unknown. Using different methods, Spangler & Gwinn (1990), Kaspi & Stinebring (1992), and Gupta et al. (1993) have reported values for the inner scale ranging from $10^4$ to $10^9$ meters. In proposing the smaller values, Spangler & Gwinn (1990) argued that the inner scale is the larger of the ion inertial length, $L_i \equiv V_A/\Omega_i$ (where $V_A$ is the Alfvén speed, and $\Omega_i$ is the ion cyclotron frequency), and the ion Larmor radius, $r_i \equiv v_{th}/\Omega_i$ (where $v_{th}$ is the ion thermal speed); they obtained parameters for the warm ionized medium in reasonable agreement with observations. In a recent discussion, Minter & Spangler (1997) have suggested ion-neutral collisional damping and wave-packet steepening as possible dissipation mechanisms for the turbulence in the diffuse ionized gas, which would make the mean-free path for ion-neutral collisions a possible value for the inner scale. However, this is thought to be larger than the maximum values proposed to explain the observations, making it a less convincing dissipation mechanism. Observationally, the Kolmogorov model with a large inner scale predicts refractive modulation indices consistent with pulsar measurements (Gupta et al. 1993). It has also been proposed to explain the occasional periodic fringes, with an inner scale on the order of the Fresnel scale (Cordes, Pidwerbetsky, & Lovelace 1986; Goodman et al. 1987). However, Rickett et al. (1997) reported a fringe event for the pulsar B0834+06 that could not be explained as the effect of a large inner-scale spectrum. The event requires similar conditions to those needed to explain the extreme scattering events (Fiedler et al. 1987).
Another way to enhance the ratio of the power between the refractive and diffractive scales in the spectrum is to steepen the spectrum—with spectral exponents $\beta > 4$ (Blandford & Narayan 1985; Goodman & Narayan 1985; Romani, Narayan, & Blandford 1986). While power-law spectra with $\beta \sim$ 11/3 have a turbulence connotation, spectra with $4<\beta<6$ might involve some forms of turbulence, but are also consistent with a distribution of non-turbulent structures with a range of spatial scales. Such steep spectra with $4 < \beta < 6$ predict refractive modulation indices close to unity (Goodman & Narayan 1985), which is substantially larger than the range 30 to 40 % observed from the nearby pulsars. On this basis Rickett & Lyne (1990) and Armstrong et al. (1995) have rejected spectra steeper than 4 for the interstellar plasma. However, the special case of $\beta=4$ has been given little attention. We can conceive of a power-law model with spectral exponent $\beta=4$ given by: $$P_{N_e}(q;z) = \frac{C_{N_e}^2} {\left(q^2+\kappa_o^2\right)^2}
\, \mbox{.}
\label{eq:beta4}
$$ Hereafter, we refer to this model as the “$\beta=4$ model.” Blandford & Narayan (1985) briefly discussed this special case without including a cut-off at low wavenumbers. It is interesting to note that even though the Kolmogorov spectral exponent $\beta=11/3$ is very close to 4, the $\beta=4$ model has a very different physical implication. Its physical origin has been rarely discussed, and it has not been formally compared with observations. Physically, this spectrum suggests the random distribution in location and orientation of discrete discontinuous objects across the line of sight. An “outer” scale, $L_o=\kappa_o^{-1}$, is included to account for the typical size of such objects. The “inner” scale here would correspond to the scale of the sharpness of a typical discontinuity. We assume the “inner” scale to be smaller than the diffractive scale of scintillations, and hence has no significant effect on the scintillations. The $\beta=4$ model could characterize stellar wind boundaries, supernova shock fronts, sharp boundaries of HII regions at the Str[ö]{}mgren radius, or any plasma “cloud” with sufficiently sharp boundaries (transition regions shorter than the diffractive scintillation scale) which may cross the line of sight. Note that, though turbulence is not necessarily implied by the model, strong turbulence which has steepened to form shocks would also be described by the $\beta=4$ model. In these models a single discrete object crossing the line of sight could also explain the “extreme scattering events” observed in the flux density variations of extra-galactic sources (e.g. Fiedler et al. 1987). We note that the analysis of such events have been primarily based on geometrical optics involving a single “cloud” (Goodman et al. 1987; Cordes et al. 1986; Cordes & Wolszczan 1986; Roberts & Ables 1982; Ewing et al. 1970). Our analysis of the $\beta=4$ model includes the “wave optics” effects when the line of sight passes through very many such clouds. The ISM is assumed to consist of a random assembly of discrete structures with abrupt density steps which may be independent of each other. If the $\beta=4$ model were compatible with all of the scintillation observations, it would remove the implication of interstellar plasma turbulence with an inertial range spanning as much as 10 orders of magnitude in scale, which has become the canonical model for ISS phenomena. It would, however, still be consistent with turbulence that has steepened into shocks.
We start with a simple derivation of the $\beta=4$ model in section \[sec:deriv\]. In section \[sec:structfn\], we give equations for the wave structure function for the $\beta=4$ model and compare predictions In section \[sec:decbw\], the variation of the diffractive decorrelation bandwidth with frequency is used to test the $\beta=4$ model against the simple Kolmogorov model and the Kolmogorov model with an inner-scale (hereafter, the “inner-scale” model). In section \[sec:mr\], the observed variation of the refractive scintillation index with the normalized diffractive decorrelation bandwidth is compared with theoretical predictions for the simple Kolmogorov, inner-scale, and $\beta=4$ models. We use theoretical results from a previous paper (Lambert & Rickett 1999; hereafter, LR99), in which we developed the theory of diffractive scintillations in a medium modeled by these spectra, and we add details of the theory for refractive scintillation in Appendix A. In Appendix B, we discuss the use of the square of the second moment to approximate the intensity correlation function for diffractive scintillation with $\beta=4$. In section \[sec:conclusion\] we give a discussion and our conclusions.
THE SPECTRUM FOR RANDOM DISCONTINUITIES IN DENSITY {#sec:deriv}
==================================================
In this section, we give a simple derivation of the $\beta=4$ model. Consider first a random distribution in space of identical plasma structures (blobs). The resulting electron density may be described as the convolution of a three-dimensional Poisson point process with the density profile of one individual blob. This is the spatial analog of shot noise, in which each charge carrier contributes the same temporal profile of current arriving at randomly distributed times. In the wavenumber domain, the power spectrum of the electron density is then given by the spectrum of the Poisson point process multiplied by the squared magnitude of the Fourier transform of one blob. The spectrum of the Poisson point process is a constant equal to the number density, $n_0$, of the blobs in space (Papoulis 1991). Thus the power spectrum of the medium follows the same shape as that of an individual blob. A similar description applies to water droplets in a fog and many naturally occurring media, as discussed by Ratcliffe (1956). If the blobs are asymmetrical and randomly oriented, one must also average the spectrum over the possible angles of orientation.
The basic feature of structures with discontinuous boundaries is that their power spectrum has an asymptotic behavior at large wavenumbers that varies as (wavenumber)$^{-4}$. This is the spatial analog of the idea that the temporal spectrum of any pulse shape with an abrupt rise or fall has a high-frequency asymptote as (frequency)$^{-2}$. The prime example is a rectangular pulse, for which the spectrum is a sinc-squared function, whose high-frequency envelope follows this law. The simplest spatial example is spherical blobs of radius $a$ with uniform density $f_0$ inside and zero outside; clearly, an added uniform background density does not affect the power spectrum. The three-dimensional spatial Fourier transform of an *isotropic function $f(r)$ of radial distance $r$ is given by (Tatarskii 1961): $$F(q) = \frac{1}{2\pi^2q}
\int_0^{\infty} \, f(r) \sin(q r) r \, dr
\, \mbox{,}
\label{eq:isofourier}
$$ where $q$ is the magnitude of the wavenumber. For any function $f(r)$ that falls to zero beyond a radius $a$, the integral will vary as $1/q$ in the high-wavenumber limit $q a \gg 1$, making $F(q) \propto 1/q^2$. For the spherical blobs of size $a$ we have: $$F(q) = \frac{f_0}{2\pi^2 q}
\left\{
- \frac{a \cos(q a)}{q} +
\frac{\sin(q a)}{q^2}
\right\}
\, \mbox{.}
\label{eq:spherefourier}
$$ For the squared magnitude of the Fourier transform, for very small and large values of $q a$: $$| F(q) |^2 =
\left\{
\begin{array}{lr}
{\displaystyle
f_0^2 a^6 /(36 \pi^4) \, \mbox{,}
} &
q a << 1 \, \mbox{,}
\\ [0.25in]
{\displaystyle
q^{-4} \; a^2 f_0^2 /(8\pi^4) \, \mbox{,}
} &
q a >> 1 \, \mbox{.}
\\
\end{array}
\right.
\label{eq:magsqspherefourier}
$$ The spectrum of the electron density is then $n_0 |F(q)|^2$. We can approximate this by equation (\[eq:beta4\]), which exactly matches equation (\[eq:magsqspherefourier\]) at small and large values of $qa$, with $C_{N_e}^2 = n_0 a^2 f_0^2 / (8 \pi^4)$ and $\kappa_o^{-1} = L_o = (2/9)^{0.25} a$.*
There are evident generalizations to make the model more like a real medium. The spheres could have a probability distribution for their radii, which would weight the average of the square of equation (\[eq:spherefourier\]). The spheres could have smoothly varying density inside and an abrupt boundary at radius $a$. As with the pulse example, the high-wavenumber behavior would not change. More realistic models would include anisotropic structures with random orientations and a distribution of scales. For example, consider a circular cylinder as a model for an anisotropic blob, with a particular radius and thickness. Its Fourier transform can be computed with respect to its major axes. This can then be transformed to tilted axes and the result averaged over an isotropic distribution of angles of tilt. Evidently the result will also be isotropic; we have solved this approximately and find a spectrum with the same high-wavenumber *asymptotic form as equation (\[eq:beta4\]), where the “outer scale,” $L_o$, is the smaller of the thickness and radius. This result could be averaged over a probability distribution of radii and thicknesses and a similar conclusion reached, with $L_o$ as a weighted average of the smallest dimension. Planar sheets at random angles are modeled as very flat cylinders and provide a crude representation of a distribution of shocks. Another simplified model would be ellipsoids of different sizes, eccentricities, and orientations, and again we expect a similar result.*
We conclude that the $\beta=4$ spectrum is an approximate description of the power spectrum for a medium with a random distribution of discontinuous structures, in the limit for wavenumbers greater than the reciprocal of the smallest dimension across the structures. We note that for wavenumbers at or below about $\kappa_o = L_o^{-1}$ the functional form depends on the detailed shapes, though it will be much less steep than $\kappa^{-4}$. Thus equation (\[eq:beta4\]) will not be generally applicable at the lowest wavenumbers. Similarly, in practice there is a finite scale over which the density jump occurs (e.g. shock thickness). This can be modeled by convolving the idealized discontinuous structures with a suitable narrow, say, Gaussian function which thus provides a very high-wavenumber cut-off beyond which the spectrum falls rapidly to zero. Consequently, we consider a spectrum with an inverse fourth power law between inner and outer scales, though these do not have the same connotations as inner and outer scales in a turbulent cascade. We do not discuss the physical origin of the supposed discontinuous plasma structures beyond the ideas formulated in the Introduction, such as supernova shock fronts, stellar wind boundaries or sharp boundaries of HII regions at the Str[ö]{}mgren radius. There needs to be no physical coupling between the individual blob structures, and there is no turbulent cascade implied by the $\beta=4$ spectrum. However, we note that shocks developing from the steepening of very strong turbulence could also be described by the model, and so if successful, the model would not rule out such strong turbulence, though in this case the $\beta=4$ range does not correspond to an inertial cascade as does the 11/3 spectrum.
In the following sections, we give the necessary scintillation theory needed to develop equations for the scintillation observables corresponding to the $\beta=4$ model, which we then compare with ISS observations. We concentrate on measurements which are sensitive to the form of the density spectrum and are relatively insensitive to the distribution of scattering material along the line of sight.
PHASE STRUCTURE FUNCTION {#sec:structfn}
========================
Central to the description of the second moment and the coherence function for intensity is the phase structure function. The longitudinal gradient of the phase structure function is related to the electron density spectrum through (cf. Coles et al. 1987): $$\begin{aligned}
D_{\phi}'(s,z;\nu) & = & \frac{8\pi^2 r_{\rm e}^2 c^2}{\nu^2}
\int_0^{\infty} P_{N_e}(\kappa, q_z=0;z) \nonumber \\
& & \times \left[1-J_0(\kappa s)\right] \kappa d\kappa
\, \mbox{.}
\label{eq:long_grad_phase_strucfunc}
$$ Here $c$ is the speed of light, $r_{\rm e}$ is the classical electron radius, and $\nu$ is the radio frequency. For a plane wave incident on a scattering medium, the line of sight integral of $D_{\phi}'(s,z;\nu)$ gives the structure function of the geometric optics phase, also called the [*wave*]{} structure function, $D_{\phi}(s)$. For the spherical wave geometry applicable to a pulsar at distance $L$, the separation $s$ at the observer’s plane is projected back to the point of integration, giving $D_{\rm S}(s) = \int_0^L D_{\phi}'(sz/L,z;\nu) \, dz$. The integrals are evaluated explicitly in equations (13), (17), and (21) of LR99 for a uniform scattering medium with the simple power-law, inner-scale, and $\beta=4$ spectral models, respectively. In the extended scattering geometry, the field coherence length at the observer, $s_0$, is defined by: $D_{\rm S}(s_0)=1$. For a screen of thickness $\Delta z$ at distance $z_p$ from the pulsar, the wave structure function at the observer becomes $D_{\rm S}(s) = D_{\phi}'(sz_p/L,z_p;\nu) \Delta z$, which has an explicit dependence on the screen location. In LR99, we found it useful to define a coherence scale $s_{\rm 0,scr}$ for the screen by its phase structure function $D_{\phi}(s_{\rm 0,scr})= D_{\phi}'(s_{\rm 0,scr},z_p;\nu) \Delta z = 1$, which is independent of the screen location. Note that in applying equation (\[eq:long\_grad\_phase\_strucfunc\]), we are assuming that the screen thickness, $\Delta z$, is much greater than the outer scale $L_o$ and small compared to total distance $L$.
Here, we repeat the equation for the phase structure function of the $\beta=4$ model in a thin layer (screen). Substituting the $\beta=4$ model for the density spectrum into equation (\[eq:long\_grad\_phase\_strucfunc\]), integrating, and multiplying by the screen thickness, $\Delta z$, we obtain: $$D_{\phi}(s) = \frac{4 \pi^2 r_{\rm e}^2 c^2 SM}{\kappa_o^2 \nu^2}
\left[1-(\kappa_o s)K_1(\kappa_o s)\right]
\, \mbox{,}
\label{eq:struc4exact}
$$ where $K_1$ is the first order modified Bessel function of the second kind, and $SM = C_{N_e}^2 \Delta z$ is the scattering measure. Equation (\[eq:struc4exact\]) may be approximated by the following logarithmic expression: $$D_{\phi}(s) = \frac{\pi^2 r_{\rm e}^2 c^2 SM}{\nu^2}
s^2 \ln\left[1+\frac{4}{(\kappa_o s)^2}\right]
\, \mbox{,}
\label{eq:struc4}
$$ which matches the full Bessel expression at both large and small values of the argument. The detailed shape, near where the structure function flattens, is governed by the shape of the low-frequency turnover in our model spectrum, equation (\[eq:beta4\]); however, as noted above, the turnover shape depends on details of the density profiles in the plasma blobs which are not constrained in our density model.
The logarithmic structure function for the $\beta=4$ model has far-reaching consequences. As the outer scale becomes larger, the structure function approaches a square-law. However, due to the presence of the logarithm, this occurs only slowly; that is, only at the limit as $s$ goes to zero does the $\beta=4$ model structure function become exactly square-law. A square-law structure function for the medium is a convenient mathematical model, but as noted in LR99, in a real medium the structure function must eventually saturate at very large separations. Nevertheless, the square-law structure function has been widely used since it provides a valid approximation for scales smaller than the reciprocal of the wavenumber at which the spectrum is cut-off. The fact that the $\beta=4$ model structure function deviates from a pure square-law, even for very large values of the outer scale, makes this model interesting and an independent investigation worthwhile.
In Figure \[fig:slope\], we have plotted the structure function of the $\beta=4$ model and its effective local logarithmic slope, $\alpha(s)$, versus the transverse spatial lag, $s$, for various values of the outer scale, $L_o$ (see equation 20 of LR99). To illustrate the shapes, we have chosen $s_{\rm 0,scr}=1000$ km for all curves. The thin dashed lines correspond to the simple Kolmogorov model with logarithmic slope of 5/3, and the thin dotted lines correspond to the square-law structure function. The plot shows how slowly the local slope approaches 2; even for $s/L_o \sim 10^{-8}$, the slope is 1.95–distinctly less than 2.
The points in the upper panel correspond to measured local exponent of the wave structure function obtained from VLBI observations, as tabulated by Spangler & Gwinn (1990). In principle, this method of comparison provides a good test for the form of the density spectrum. These authors presented a similar plot corresponding to the inner-scale model, for which $\alpha(s)$ is close to 2.0 for $s$ less than the inner scale and close to 5/3 for $s$ greater than the inner scale, with a transition occurring over about a decade in $s$. They derived estimates of the inner scale ranging from 50,000 to 200,000 meters, for the highly scattered sources that they studied. Whereas there is some uncertainty in some of the estimates of $\alpha$, there are several examples in which values greater than 5/3 are reliably observed (e.g. Trotter et al., 1998). As can be seen, there is also reasonable agreement with the $\beta=4$ model. Assuming the model to be correct, a separate model-fitting to each observation yields an outer scale in the range $L_o \sim 3 \times 10^7$ to $10^8$ meters. These values are much smaller than the parsec scales of supernova remnants and interstellar clouds. However based on this comparison alone, we are unable to discriminate between the inner-scale and $\beta=4$ models since they both provide equally satisfactory agreement with the observations. Hence further independent tests are needed, and in the following sections, we present comparisons with two other scintillation observables which in the end argue against the $\beta=4$ model as a universal description of the ionized ISM.
DIFFRACTIVE DECORRELATION BANDWIDTH {#sec:decbw}
===================================
The diffractive decorrelation bandwidth, $\Delta\nu_d$, is perhaps the easiest ISS observable to measure. It is the frequency-difference for a decorrelation to 50% of the correlation function for diffractive intensity scintillations. These are usually recorded as a dynamic spectrum centered on a given radio frequency, $\nu$, for a pulsar. Observations of $\Delta\nu_d$ for one pulsar at a wide range of radio frequencies have provided an important test of power-law models for the interstellar density spectrum on that particular line of sight. When the diffractive scale is far from the inner and outer scales, one expects $\Delta\nu_d \propto \nu^{2\beta/(\beta-2)}$, which for the simple Kolmogorov model is $\nu^{4.4}$, in both screen and extended medium geometries. In this section we re-examine the published measurements and compare them with theory for the three different spectral models. Since the scaling laws are all so steep, in Figures \[fig:vela\] and \[fig:other\], we have plotted the observations logarithmically as $\Delta\nu_d/\nu^4$. Overplotted are lines giving the theoretical predictions for our three spectral models, with the simple Kolmogorov model as a line of slope 0.4. Before giving our conclusions, we must discuss the error bars and various corrections to the data and also the method for deriving the theoretical predictions.
Observations {#observations .unnumbered}
------------
Decorrelation bandwidth data, collected together by Cordes, Weisberg, & Boriakoff (1985; hereafter, CWB), for pulsars PSR B0833$-$45 (Vela), PSR B0329$\-+$54, PSR B1642$\--$03, PSR B1749$-$28, and PSR B1933$\-+$16 are shown in Figures \[fig:vela\] and \[fig:other\]. We have also included recently measured points by Johnston et al. (1998) for the Vela pulsar. These are $(\nu, \Delta\nu_d) = (8.4\mbox{GHz}, 13.9\mbox{MHz})$ and $(13.7\mbox{GHz}, 58.2\mbox{MHz})$. The data are shown as solid circles if derived from dynamic spectra and as solid stars if derived from pulse broadening measurements. The latter become easier to estimate at low frequencies, where the resolution bandwidth required for the former becomes too narrow for an adequate signal-to-noise ratio. The conversion of an estimate of a pulse broadening time to a decorrelation bandwidth relies on the uncertainty relation $2 \pi \Delta\nu_d \tau_d = C_1$. Unfortunately, the “constant” $C_1$ takes on different values for different geometries and spectral models. For the $\beta=4$ model, it is also very weakly dependent on frequency. In Table 1 of LR99, we gave numerical values for $C_1$ for the three spectral models with spherical waves in both a screen and an extended medium geometry. For a given geometry, the smallest $C_1$ is for the Kolmogorov spectrum and the largest (by about 50%) is for the square-law structure function. Thus in Figures \[fig:vela\] and \[fig:other\] we plot, as solid and open stars, the values converted using $C_1$ for the Kolmogorov and square-law structure function models, respectively. We display the predictions for extended medium and note that the screen values are only about 20% less for each spectrum. Thus in comparing data with the theory for the $\beta = 4$ and inner-scale models, the data should lie between the extremes of the open and solid stars, depending on the outer and inner scales, respectively. Note that the value of $C_1$ in equation (7) of the Taylor et al. (1993) catalog is more than 50% greater than values computed by LR99, since it concerns the mean pulse delay rather than the 1/e decay time of the pulse.
The second correction to the data concerns the effects of refractive scintillation at the higher radio frequencies, which are closer to the transition to weak scintillation. Gupta et al. (1994) gave a heuristic theory of the effect and expressions for the bias to the diffractive $\Delta\nu_d$ due to refractive shifts; LR99 discussed the effect from the work of Codona et al. (1986; hereafter, CCFFH). In Figures \[fig:vela\] and \[fig:other\], we have plotted as open circles $\Delta\nu_d$ corrected for this effect using equation (D6) of Gupta et al. (1994), who also estimated the variability in estimates of $\Delta\nu_d$ due to the refractive modulation. We used this to estimate an error bar on the open circles, which is typically larger than the error bars quoted by the original observers, plotted on the solid circles.
In summary, the open circles and their error bars provide the best direct estimates of $\Delta\nu_d$ and the indirect estimates lie between the open and solid stars.
Theory {#theory .unnumbered}
------
The theory of diffractive scintillations has been discussed by many authors; we will use the results of section 6 in LR99 for a point source (spherical waves) either in an extended scattering medium or with a screen at $z_p$ from the pulsar and $z_o$ from the observer (with $z_p+z_o = L$). Following usual practice, we assume the strong scattering limit, in which the frequency decorrelation function for intensity is the squared magnitude of the diffractive component of the two-frequency second moment of the field. The validity of this approximation is examined in Appendix B, where we find that it introduces a small error for the $\beta=4$ model, which however is negligible compared to the errors in the observations. The decorrelation bandwidth, $\Delta\nu_d$, can be defined in terms of a normalized bandwidth, $v_d$: $$\Delta \nu_d = \frac{v_d \nu}{u^2}
\;\;\;{\rm where}\;\; u^2 = \frac{z_{\rm scatt} c}{2 \pi \nu s_{0}^2}
\; .
\label{eq:Delta_nu_d}
$$ Here $\nu$ is the (geometric) mean of the two frequencies, and the parameter $u$ determines the strength of scattering; $c$ is the speed of light and $z_{\rm scatt}$ is the effective scattering distance. For a uniform scattering medium, $z_{\rm scatt} = L$, and for a screen geometry, $z_{\rm scatt} = z_e = z_o z_p/L$. $s_{0}$ is the field coherence scale where the appropriate phase structure function equals unity; for the uniform scattering medium, it is defined at the observer, and for a screen, it is $s_{\rm 0,scr}$ which is independent of the distances to the pulsar and screen, and leads to $u_{\rm scr}$ as the appropriate strength of scattering parameter.
The computations presented in LR99 include a tabulation of the normalized bandwidth $v_d$ for the various models under discussion, and associated plots of the intensity decorrelation function itself. For the simple Kolmogorov model, a constant value for $v_d$ is obtained regardless of the frequency or distance (0.773 and 0.654 for the extended medium and screen geometries, respectively). Thus the frequency dependence of $\Delta\nu_d$ reflects the frequency dependence of $s_0$, giving $\Delta\nu_d \propto \nu^{2\beta/(\beta-2)}$. CWB used this to estimate $\beta$ from observations of $\Delta\nu_d$. We now compare the observations with theoretical scaling laws for the two other spectral models. There are, however, some complications. The quantity $v_d$ is no longer exactly independent of frequency, and $s_0$ depends on both frequency and the outer scale, $L_0$, or the inner scale, $L_i$, that parameterize the other two models.
Consider the details for the discontinuity spectrum. In Figure \[fig:vela\], we plot $\Delta \nu_d/\nu_{\rm GHz}^4$ versus $\nu_{\rm GHz}$ as observed for the Vela pulsar. In Figure \[fig:vela\]a, we have plotted theoretical curves for $\beta=4$ model with various outer scales. We use equation (\[eq:Delta\_nu\_d\]) for the models and so need the variation of both $s_0$ and $v_d$ with $\nu$.
For a screen, we determine $s_{\rm 0,scr}$ from equation (\[eq:struc4\]) for the $\beta=4$ model, and eliminate the scattering measure $SM$ using the same equation for $s_{\rm 0,1GHz}$ at a reference frequency of 1 GHz. This gives a relation between $s_{\rm 0,scr}/s_{\rm 0,1 GHz}$ and $\nu_{\rm GHz}$, with $L_o/s_{\rm 0,1 GHz}$ as a parameter. For an extended scattering medium, we use equation (21) of LR99 to determine $s_0$ at the observer, and the same method as for the screen to eliminate $SM$. In Figure \[fig:vd\], we show the variation of $v_d$ with $L_o/s_{0}$ at a fixed frequency (computed by Lambert 1998 and described in LR99). This is combined with the $s_0$–$\nu_{\rm GHz}$ relation to obtain the relation of $v_d$ to $\nu_{\rm GHz}$. For the curves in Figure \[fig:vela\], $s_{\rm 0,1GHz}$ is determined so that, for each value of $L_o$, the model $\Delta \nu_d$ fits the measured values near 1GHz. For most of the models of interest $L_o \gg s_{0}$, and then the logarithm functions vary much more slowly than the $s_0^2$ term in equation (\[eq:struc4\]). In which case there is nearly a linear relation $s_0/s_{0,\rm ref} = \nu/\nu_{\rm ref}$. Furthermore, Figure \[fig:vd\] shows that as $s_0$ changes, the parameter $v_d$ changes very slowly. Consequently, for a very wide range of the outer scale, a good approximation is $\Delta \nu_d \propto \nu^4$, which reflects the fact that the underlying structure function then approaches a square-law behavior.
Consider the plot for the Vela pulsar PSR B0833$-$45 (Figure \[fig:vela\]) for which the data are most extensive. Although we have computed curves and data corrections for both the screen and extended scattering geometries, we only show the extended medium results in Figure \[fig:vela\] because the two plots are so similar. In the comparison with the screen model, the star points are lowered about 0.08 vertical units (20%), and the $\beta=4$ model curves are very slightly steeper functions of frequency. However, the scatter among the observations is greater than the differences in the models between screen and extended geometries. In the Vela plot, we see that the $\beta=4$ model agrees somewhat better with the observations than does the simple Kolmogorov model. This pulsar is known to lie in a highly scattered region, and hence the presence of discontinuities, as incorporated in the $\beta=4$ model, is perhaps reasonable. However, the conclusion is not strong, since the data are also reasonably fitted by an inner scale in the range $100$ km to $500$ km, as can be found from Figure \[fig:vela\]b.
For the other pulsars (Figure \[fig:other\]), the data show a stronger frequency dependence than predicted by both the simple Kolmogorov and $\beta=4$ models. However, the inconsistencies between the measurements at different frequencies are even greater, and better observations are needed, before such comparisons could discriminate between the models. A recent series of measurements at 327 MHz (Bhat et al. 1999a,b,c) documents the variability of $\Delta\nu_d$, and a similar long sequence of such measurements at other frequencies is needed on the same pulsars, before reliable conclusions can be reached from the frequency scaling observations.
REFRACTIVE SCINTILLATION INDEX {#sec:mr}
==============================
Slow variations in the flux of pulsars are caused by refractive interstellar scintillation (RISS), and can be characterized by the rms deviation in flux density normalized by its mean, or scintillation index, $m_R$. RISS is due to inhomogeneities in the interstellar electron density on scales much larger than those responsible for diffractive scintillation, as characterized by the decorrelation bandwidth. There are now measurements of both phenomena on a substantial number of pulsars, from which we can constrain the density spectrum on each line of sight. The relation of the refractive scintillation index, $m_R$, to the normalized diffractive decorrelation bandwidth, $\Delta\nu_d/\nu$, depends on the ratio of the power in the density spectrum at the large refractive scales ($10^{11}$ m to $10^{12}$ m) to the power at the smaller diffractive scales ($10^7$ m to $10^8$ m).
We compare pulsar measurements gathered from the literature with theoretical predictions by plotting $m_R$ versus $\Delta\nu_d/\nu$ (cf. Rickett & Lyne 1990 and Gupta et al. 1993). Related tests have been made by Armstrong et al. (1995), Bhat et al. (1999a) and Smirnova, Shishov, & Stinebring (1998). $m_R$ measurements at 610 MHz are listed in Table \[table:mr610\] and include the results of long-term monitoring observations by Smirnova et al. (1998), and older measurements near 100 MHz are listed in Table \[table:mr100\]. The tables also include the decorrelation bandwidth obtained from diffractive scintillation, typically observed at a different radio frequency, scaled to the observing frequency for $m_R$. We used the Kolmogorov scaling law $\Delta\nu_d \propto \nu^{4.4}$, and we note that the minor differences that result from changes in the scaling law for other spectral models are insignificant in the comparison. The 100 MHz data consist of $m_R$ measurements at 73.8 MHz, 81.5 MHz, and 156 MHz. The pulsars observed at 610 MHz are primarily located at distances $\simgreat 1$ kpc, whereas the pulsars observed at 100 MHz are nearer at distances $\simless 1$ kpc. Figures \[fig:mr4\] and \[fig:mrin\] show plots of the theoretical and measured refractive scintillation index versus the normalized decorrelation bandwidth in separate panels for the two sets of measurements. Theoretical curves are also shown in the Figures for different spectral models in both the screen and extended medium geometries. The dashed line corresponds to the simple Kolmogorov model ($\alpha=5/3$), and the solid lines in Figures \[fig:mr4\] and \[fig:mrin\] correspond to the $\beta=4$ and inner-scale models, respectively.
In section \[sec:decbw\] we noted that variable refraction tends to decrease the measured decorrelation bandwidth and used an expression from Gupta et al. (1994) to apply a nominal correction to measured values. In the table of data from near 100 MHz, we have made use of the recently published measurements of Bhat et al. (1999a). They monitored the apparent decorrelation bandwidth near 327 MHz for 20 nearby pulsars, many of which have $m_R$ measurements in Table \[table:mr100\]. They found the bandwidth to vary by factors 3-5 and discussed the influence of varying refraction as a refractive bias. From their measurements, they derived a corrected decorrelation bandwidth, which we have scaled (using the $\nu^{4.4}$ scaling law) to the frequency at which $m_R$ was observed. These gave values typically 2–5 times bigger than obtained from earlier measurements (e.g. Cordes 1986). The pulsar B0809+74 was not observed by Bhat et al. (1999a); however, the recent weak scintillation observations of this pulsar by Rickett et al. (1999) similarly suggests that earlier decorrelation bandwidth measurements overestimated the strength of scattering. The effect of these changes is to shift the plots to the right in Figures \[fig:mr4\] and \[fig:mrin\] by about half a decade, compared to Figure 5 of Gupta et al. (1993). There is only one pulsar (B0329+54) in Table \[table:mr610\] common to the Bhat et al. (1999a) observations, and its decorrelation bandwidth has also been corrected for the refractive bias. Since the pulsars observed at 610 MHz were mostly more heavily scattered, the refractive bias correction is substantially smaller (see Gupta et al. 1994) and has been ignored.
Theory {#theory-1 .unnumbered}
------
The theory of refractive scintillations has been known since the 1970s and was applied to pulsar flux variations by Rickett et al. (1984). For example, Prokhorov et al. (1975) described how the modulation index for intensity can exceed unity in strong scattering in their equations (4.53) *et. seq. We use the notation of Coles et al. (1987) and confine our discussion to spherical wave sources, propagating in a scattering plasma which is either concentrated in a screen or extended uniformly between source and observer. The “low-frequency” approximation for the intensity covariance function is given by equation (10) of Coles et al. (1987), from which we obtain the normalized refractive variance, $m_R^2$, by setting the spatial offset equal to zero: $$\begin{aligned}
m_R^2 & = &
\frac{16 \pi^2 r_{\rm e}^2 c^2 L}{\nu^2}
\int_0^1 \, \int_0^{\infty}
P_{N_e}(\kappa, q_z=0;z) \times \nonumber \\
& & \exp\left\{
- L \int_0^1 D_{\phi}'\left[\kappa r_{{\rm F},L}^2 h(x,y)\right] dy
\right\} \times
\nonumber \\
& &
\sin^2\left[ 0.5 \kappa^2 r_{{\rm F},L}^2 x(1-x)\right]
\, \kappa d\kappa \, dx
\, \mbox{,}
\label{eq:crextmed}
$$ where: $$\begin{aligned}
h(x,y) & = & {\rm min}\left[y(1-x), x(1-y) \right] \\
r_{{\rm F},L}& = & \sqrt{Lc/2 \pi \nu} \mbox{.} \nonumber
\label{eq:h}
$$ Here the transverse wavenumber is $\kappa = \sqrt{q_x^2 + q_y^2}$, and $x=z/L$, where $z$ is the distance along the line of sight measured from the source. For the screen geometry, both the density spectrum and the gradient in the phase structure function $D_{\phi}'$ are concentrated in a thin layer of thickness $\Delta z$ at distance $z_p$ from the source and $z_o$ from the observer. Thus the $x$ and $y$ integrations in equation (\[eq:crextmed\]) give $x=y=z_p/L$, and we find it useful to define an equivalent Fresnel scale $r_{\rm F,scr}$: $$r_{\rm F,scr} = \sqrt{z_{\rm e}c/2 \pi \nu} \;\;
\mbox{where} \;\; z_{\rm e}=z_p z_o/L
\, \mbox{.}
\label{eq:rfscr}
$$ The $\kappa$ integration in equation (\[eq:crextmed\]) involves the product of the density spectrum, decreasing as a steep power of $\kappa$, times the high-pass Fresnel filter $\propto \kappa^4$ up to $r_{\rm F,scr}^{-1}$, times the low-pass exponential term that cuts off wavenumbers above $s_R^{-1}$, where $s_R = u_{\rm scr} r_{\rm F,scr}$ is the radius of the scattering disc. Hence in strong scattering ($u_{\rm scr} > 1$) we only consider the $\kappa^4$ part of the Fresnel filter. Consequently, the integration depends on the ratio $L_o/s_R$ or $L_i/s_R$. For an extended scattering medium the line-of-sight integration softens the exponential cut-off in the $\kappa$ integration, but the basic relationships remain the same. As others have noted the level of refractive scintillation is greater in an extended medium than in a screen with the same observed diffractive scintillation.*
The canonical spectral model in ISS studies of pulsars has been the simple Kolmogorov spectrum $P_{N_e}(q,z) = C_{N_e}^2(z) q^{-11/3}$. By fitting this model to diffractive scintillation observations, observers have estimated the scattering measure, $SM = \int_0^L C_{N_e}^2(z) dz$, toward many pulsars. When divided by the pulsar distance, this gives a line-of-sight average of $C_{N_e}^2(z)$, which is found to vary greatly from one direction to another and to increase dramatically for distant pulsars seen at low galactic latitudes (e.g. CWB). Taylor & Cordes (1993) have developed a smoothed model for the Galactic plasma density distribution, which includes enhanced scattering in spiral arms and toward the Galactic center. However, there are also large random variations on much finer spatial scales, which would produce a scatter in a plot of $m_R$ against distance of dispersion measure. For a simple power-law spectrum both $m_R$ and $\Delta\nu_d/\nu$ depend on $SM$ through a single strength of scattering parameter. Thus for the Kolmogorov spectrum the variation of $m_R$ with $\Delta\nu_d/\nu$ is independent of distance or frequency and is given by a single dashed line in each plot in Figures \[fig:mr4\] and \[fig:mrin\]. If the scattering medium is uniform, the strength of scattering is $u = r_{{\rm F},L}/s_0$. For the screen it becomes $u_{\rm scr} = r_{\rm F,scr}/s_{\rm 0,scr}$, and in that case, the dashed theoretical line is independent of the location of the screen between the source and the observer. This point is substantiated in Appendix A, where the theory is laid out in more detail. According to this screen model, along each line of sight there is a scattering layer with a certain $SM$, which determines both $m_R$ and $\Delta\nu_d/\nu$, but $SM$ is not necessarily related to the pulsar distance. The theoretical curve with $SM$ as the variable is independent of where the layer is located along the line of sight.
For the $\beta=4$ and inner-scale models, the theoretical curves for $m_R$ versus $\Delta\nu_d/\nu$ depend on the extra parameter $L_o$ or $L_i$, respectively. Details of the theory are given in Appendix A, where the relevant parameters are shown to be $L_o/s_{\rm 0,scr}$ and $L_i/s_{\rm 0,scr}$. Since $s_{\rm 0,scr}$ depends on frequency, $SM$, and distance, then $m_R$ also depends somewhat on frequency and distance. In order to fix the frequency dependence of the theoretical $m_R$ values, we have separated the measurements into two groups (at 610 MHz and near 100 MHz). The distance dependence is dealt with by assuming that $SM \propto$ distance. Whereas this is clearly appropriate for the extended scattering medium, it is less clear for the screen model, since the screen model supposes that $SM$ is not necessarily related to distance. However, it is reasonable that, if on a long line of sight there is a single region that dominates the scattering, its $SM$ value will statistically increase with line-of-sight distance. Indeed the experimentally derived scattering measure increases faster with distance than if the medium were uniform (see Figure 1 of Cordes et al. (1991).
In Figure \[fig:mr4\] we show the theoretical curves for $m_R$ in the $\beta=4$ model, which are relatively flat where $L_o > s_R$. As $u$ increases ($\Delta\nu_d/\nu$ decreases), $s_R$ increases, and when $s_R > L_o$ the scintillations are suppressed. In a screen this occurs when $\Delta\nu_d/\nu \simless (r_{\rm F,scr}/L_o)^2$.
Inspection of Figure \[fig:mr4\] shows that the measured $m_R$ values at 100 MHz are above the prediction of the simple Kolmogorov model for the extended medium geometry. These measurements can, however, be modeled with the $\beta=4$ spectrum with suitable specific choices of outer scale, $L_o$. The screen geometry is inconsistent with the measurements at 100 MHz for both the simple Kolmogorov and $\beta=4$ models. For the 610 MHz data, the measured $m_R$ values are mostly in agreement with the prediction of the simple Kolmogorov model for the extended medium geometry; however, they lie below the predictions of the $\beta=4$ model. For the screen geometry, the measured $m_R$ values are above the simple Kolmogorov model curve. However, for the $\beta=4$ model, though less convincingly, an agreement with the measurements can be found by suitable choices of the outer scale, $L_o$. The most striking feature of the Figure is the good agreement with the uniform extended Kolmogorov model for most of the pulsars. This result, relies most heavily on the excellent data from Smirnova et al. (1998), who also note this agreement. We are persuaded by this Figure that the $\beta=4$ model cannot be viewed as a global alternative to the Kolmogorov spectrum. We also conclude that for the medium-distance pulsars measured at 610 MHz, $m_R$ does not agree with the screen geometry. It appears that even if the scattering medium is not uniform on scales of a few kpc, there is not a single region that dominates the scattering. Computations of $m_R$ due to a patchy distribution along the line of sight could test what distribution would start to approximate the uniformly extended medium.
Turning to a comparison with the alternative inner-scale spectrum in Figure \[fig:mrin\], the 100 MHz observations could be explained by very large values of $L_i$ ($10^{10}$m to $10^{11}$m) in a screen configuration or by more modest, but still large, values of $L_i$ ($10^8$m) in an extended medium. The latter was essentially the proposal made by Coles et al. (1987) for the nearby lines of sight. The 610 MHz points are relatively lower than those at 100 MHz. For an inner scale spectrum in a screen, large values of $L_i$ ($10^8$m to $10^{10}$m) would be necessary; and in an extended medium 60% of the points lie near the simple Kolmogorov model, with the rest requiring inner scale values ($10^7$m to $10^8$m). The 610 MHz data in our analysis come from RISS observations of 21 pulsars made by Stinebring and are described in more detail by Smirnova et al. (1998), who also compare the results with various models for the spectrum and spatial distribution of the electron density. They note that 4 pulsars which are seen through known HII regions or supernovae remnants show relatively elevated values for $m_R$, and find 3 other pulsars with similar behavior. These are the points which lie above the Kolmogorov line in Figure \[fig:mrin\]b. Their interpretation is that for these objects the scattering is concentrated into regions either near the pulsar or near the Earth, and that these regions are characterized by an inner scale near $3\times 10^8$m. Their spectrum model is very similar to our inner-scale model, except that the cut-off is characterized by a steep power-law rather than by an exponential function. We find a somewhat larger numerical value for the inner scale required to match those objects. We also note that the theoretical $m_R$ values are consistently higher for an extended medium, and so require a more modest inner scale.
However, an important result of our analysis is an alternative explanation for the relatively high $m_R$ values seen for some pulsars. We suggest that pulsar lines of sight can pass through discrete clouds with increased plasma density on large scales, that steepen the low-wavenumber spectrum as opposed to cutting off the high-wavenumbers. A discontinuity spectrum ($\beta=4$) is one way that the spectrum can be steepened, but smoother structures on scales larger than $10^{11}$m ($\sim 1$ AU) would also boost the low-wavenumbers and could cause similar enhancements of refractive compared to diffractive scintillations. Such enhancements are likely to be associated with HII regions or supernova remnants. A similar idea was also discussed by Lestrade, Rickett, & Cognard (1998) in the context of extreme scattering events in pulsar timing measurements. To take this idea further, a composite between the Kolmogorov and the $\beta=4$ or steeper spectra should be investigated, particularly one in which the line of sight is not uniformly weighted. Smirnova et al. (1998) also note that the strength of scattering increases very much faster with distance and dispersion measure than if the medium were statistically uniform. The distant pulsars in their sample are mostly observed at low galactic latitudes and so are subject to the enhanced density and turbulence described by the Taylor & Cordes (1993) model. It appears that most distant pulsars follow the uniform extended Kolmogorov model with inner scale smaller than $10^{6}$m. Thus although these lines of sight are subject to enhanced scattering in the inner galactic plane, the spectrum effectively follows the Kolmogorov law and the plasma is dispersed enough to approximate a uniform scattering medium. There are presumably discrete refracting structures present along these lines of sight, but their contribution to the density spectrum is masked by the higher densities of the inner Galaxy, which still follow an apparently turbulent spectral form.
[llr@[.]{}lr@[.]{}lr@[.]{}lr@[.]{}lr@[.]{}l]{}
Rickett & Lyne (1990) & B0531+21 & 2&0 & 610&0 & 2&96$\times 10^{-2}$ & 0&327 & 0&052
& B0531+21 & 2&0 & 610&0 & 2&96$\times 10^{-2}$ & 0&402 & 0&095
Kaspi & Stinebring (1992) & B0329+54 & 1&4 & 610&0 & 0&44 & 0&39 & 0&07
& B0833$-$45 & 0&55 & 610&0 & 1&29$\times 10^{-4}$ & 0&11 & 0&01
& B1749$-$28 & 1&2 & 610&0 & 5&13$\times 10^{-2}$ & 0&26 & 0&05
& B1911$-$04 & 2&29 & 610&0 & 1&91$\times 10^{-3}$ & 0&20 & 0&03
& B1933+16 & 7&8 & 610&0 & 1&82$\times 10^{-3}$ & 0&18 & 0&04
& B2111+46 & 5&22 & 610&0 & 2&69$\times 10^{-3}$ & 0&16 & 0&05
& B2217+47 & 2&31 & 610&0 & 0&26 & 0&21 & 0&02
Stinebring et al. (1999) & B0136$+$57 & 2&9 & 610&0 & 5&7$\times 10^{-3}$ & 0&15 & 0&02
& B0329$+$54 & 1&4 & 610&0 & 0&350 & 0&37 & 0&02
& B0525$+$21 & 2&3 & 610&0 & 0&30 & 0&31 & 0&01
& B0531$+$21 & 2&0 & 610&0 & 4&0$\times 10^{-2}$ & 0&32 & 0&01
& B0736$-$40 & 2&1 & 610&0 & 9&0$\times 10^{-6}$ & 0&03 & 0&01
& B0740$-$28 & 1&9 & 610&0 & 3&5$\times 10^{-3}$ & 0&13 & 0&01
& B0818$-$13 & 2&5 & 610&0 & 0&061 & 0&23 & 0&01
& B0833$-$45 & 0&5 & 610&0 & 1&50$\times 10^{-4}$ & 0&24 & 0&01
& B0835$-$41 & 4&2 & 610&0 & 7&0$\times 10^{-4}$ & 0&21 & 0&02
& B1641$-$45 & 4&6 & 610&0 & 7&0$\times 10^{-7}$ & $\sim$0&1 & 0&05
& B1642$-$03 & 0&5 & 610&0 & 0&770 & 0&46 & 0&04
& B1749$-$28 & 1&5 & 610&0 & 6&0$\times 10^{-2}$ & 0&25 & 0&02
& B1818$-$04 & 1&6 & 610&0 & 4&0$\times 10^{-4}$ & $\sim$0&1 & 0&05
& B1859$-$03 & 8&1 & 610&0 & 5&0$\times 10^{-6}$ & $\sim$0&05 & 0&03
& B1911$-$04 & 3&2 & 610&0 & 8&0$\times 10^{-4}$ & 0&22 & 0&01
& B1933$+$16 & 3&5 & 610&0 & 2&0$\times 10^{-3}$ & 0&24 & 0&02
& B1946$+$35 & 7&9 & 610&0 & 3&0$\times 10^{-5}$ & $\sim$0&05 & 0&03
& B2111$+$46 & 5&0 & 610&0 & 3&0$\times 10^{-3}$ & 0&15 & 0&01
& B2217$+$47 & 2&5 & 610&0 & 0&20 & 0&30 & 0&01
[llr@[.]{}lr@[.]{}lr@[.]{}lr@[.]{}lr@[.]{}l]{}
Cole et al. (1970) & B0809+74 & 0&31 & 81&5 & 1&38$\times 10^{-3}$ & 0&45 & 0&18
& B0834+06 & 0&72 & 81&5 & 1&3$\times 10^{-3}$ & 0&39 & 0&11
& B1919+21 & 0&66 & 81&5 & 1&2$\times 10^{-3}$ & 0&53 & 0&18
Helfand et al. (1977) & B0329+54 & 1&4 & 156&0 & 5&7$\times 10^{-3}$ & 0&35 & 0&08
& B0823+26 & 0&38 & 156&0 & 1&0$\times 10^{-2}$ & 0&46 & 0&11
& B1133+16 & 0&27 & 156&0 & 2&9$\times 10^{-2}$ & 0&50 & 0&08
& B1508+55 & 1&93 & 156&0 & 7&0$\times 10^{-3}$ & 0&41 & 0&07
& B1919+21 & 0&66 & 156&0 & 1&4$\times 10^{-2}$ & 0&50 & 0&07
& B2217+47 & 2&31 & 156&0 & 6&45$\times 10^{-4}$ & 0&41 & 0&11
Gupta et al. (1993) & B0329+54 & 1&4 & 73&8 & 3&1$\times 10^{-4}$ & $>$0&15 &
& B0809+74 & 0&31 & 73&8 & 8&90$\times 10^{-4}$ & 0&34 & 0&06
& B0834+06 & 0&72 & 73&8 & 8&6$\times 10^{-4}$ & 0&16 & 0&03
& B0950+08 & 0&12 & 73&8 & 0&17 & 0&45 & 0&05
& B1133+16 & 0&27 & 73&8 & 1&6$\times 10^{-3}$ & 0&18 & 0&02
& B1237+25 & 0&56 & 73&8 & 1&8$\times 10^{-3}$ & 0&30 & 0&06
& B1508+55 & 1&93 & 73&8 & 3&8$\times 10^{-4}$ & 0&28 & 0&09
& B1919+21 & 0&66 & 73&8 & 7&5$\times 10^{-4}$ & $>$0&21 &
SUMMARY AND CONCLUSION {#sec:conclusion}
======================
In this paper, the theory of the $\beta=4$ model for the electron density spectrum was derived for discontinuous density structures and compared with pulsar observations. A new feature of our analysis is the inclusion of an “outer scale” needed in any realistic model. The model is characterized by an effective exponent $\alpha(s)$ of the structure function, which remains between 1.95 and 1.6 over a very wide range of $L_o$ values (cf. Figure \[fig:slope\]). This at first seems a promising explanation for the spread in the estimates of $\alpha$ derived from VLBI observations of the angular broadening profile, as observed on heavily scattered lines of sight.
As discussed in section \[sec:decbw\], from Figures \[fig:vela\] and \[fig:other\], we find that the $\beta=4$ model provides a somewhat better agreement with the measurements of the diffractive decorrelation bandwidth versus frequency for pulsar PSR B0833$-$45 (Vela) than does the simple Kolmogorov model. This might arise from refractive scattering effects caused in the supernova remnant associated with the Vela pulsar. Four other pulsars with decorrelation bandwidths measured against frequency show an appreciably stronger frequency dependence than the predictions of both the simple Kolmogorov and $\beta=4$ models. However, there are substantial inconsistencies among the measurements and better observations are clearly needed, especially in view of the variability in $\Delta \nu_d$ documented by Bhat at al. (1999c).
The predictions of the $\beta=4$ model for the variation of the refractive scintillation index with the diffractive decorrelation bandwidth are in partial agreement with the observations. As discussed in section \[sec:mr\], the values of $m_R$ measured near 100 MHz are above the prediction of the simple Kolmogorov model and had previously been explained as the effect of an inner scale, substantially larger than the values invoked by Spangler & Gwinn 1990. However, the measurements are also consistent with the $\beta=4$ model with suitable choices of the outer scale. For the 610 MHz data, most of the measured $m_R$ values are in good agreement with the prediction of the simple Kolmogorov model for the extended medium, and they lie below the curves of the $\beta=4$ model. However, there are a significant number of good-quality observations, which lie somewhat above the Kolmogorov line. We suggest an alternative to a large inner scale on those lines of sight, that they pass through regions of enhanced density, which cause enhanced refractive scattering; these regions must have less small-scale substructure than in a turbulent medium and could include discontinuities.
Based on the above considerations, we reject the $\beta=4$ model as a universal spectral model for the interstellar electron density fluctuations. The corollary is to strengthen the evidence for the Kolmogorov density spectrum, which in turn suggests a turbulent process in the interstellar plasma. However, the simple Kolmogorov spectrum is not a universal model either, since it disagrees with several of the $m_R$ observations. Since the $\beta=4$ model provides reasonable agreement for many of these discrepant observations, we propose that enhancements in the large scale part of the spectrum (which need not be described by discontinuities in density) occur on these lines of sight. With such enhancements causing the increase in refractive scintillation, there is no need to invoke the relatively large inner scales proposed by Coles et al. (1987). As proposed by Spangler & Gwinn (1990), a relatively small inner scale is then likely, controlled by the ion inertial length or Larmor radius.
--------------------------------------------------------- -----------------------------------------------------------------------------------
Symbols Definition
$\alpha$, $\alpha(s)$ Exponent of separation $s$ in structure function
$\beta$ Exponent of wavenumber in density power spectrum
$\mbox{\boldmath $\beta$}$, $\mbox{\boldmath $\beta$}'$ Vector spatial offsets in Appendix B
$c$ speed of light
$C_1$ Constant in uncertainty relation, $2\pi\Delta \nu_d\tau_d = C_1$
$C_{N_e}^2$ Coefficient in electron-density wavenumber spectrum (m$^{-20/3}$)
$D_{\phi}$,$D_{\phi}'$ Wave/phase structure function and its longitudinal gradient Equation (6)
$D_{\rm S}$ Wave/phase structure function for a point (spherical wave) source
$\Delta z$ Screen or layer thickness
$\Delta \nu_d$ Diffractive Decorrelation Bandwidth
$\epsilon$ Fractional frequency difference $|k_1-k_2|/(k_1+k_2)$
$f(r), f_0$ Radial profile function for electron density, interior density
$F(q)$ Three-dimesnsional Fourier Transform of $f(r)$ Equation (3)
$g(\eta), g_1(\eta)$ Intermediate functions of $\eta = \kappa^2$
$J_0$ Zero-order Bessel function of the first kind
$\kappa$ Transverse (2-dimensional) wavenumber
$\kappa_i,\kappa_o$ High and low wavenumber cutoffs Equation (1)
$k=2\pi/\lambda, k_1, k_2$ Radio wavenumbers
$k_m$ Geometric mean of $k_1,k_2$
$L$ Pulsar distance
$L_i,L_o$ Inner and Outer scales
$\lambda$ Radio wavelength
$m_R$ Scintillation index (rms/mean) for refractive scintillation Equation (10)
$m_{\rm int}$ Scintillation index due to scales intermediate between diffractive and refractive
$n_0$ Spatial number density of discontinuous objects
$\nu$ Radio frequency
$P_{N_e}(q)$ Wavenumber spectrum for the electron density Equation (1)
$q$ Three-dimensional wavenumber
$r_e$ Classical electron radius
$r_{{\rm F},L}$ Fresnel scale ($\sqrt{L/k}$) Equation (11)
$r_{\rm F,scr}$ Fresnel scale for a screen ($\sqrt{z_e/k}$) Equation (12)
$s$ Transverse separation
$SM$ Scattering measure = Line of sight integral of $C_{N_e}^2$
$s_0$ Field coherence scale
$s_R$ Refractive scintillation scale
$\tau_d$ Pulse decay time to 1/e
$u$ Strength of scattering parameter ($=r_{{\rm F},L}/s_0$) Equation (9)
$v_d$ Normalized decorrelation bandwidth Equation (9)
$V_4$ Sum of wave structure functions Equation (B1)
$x, y$ Normalized distances
$z, z_p$ Distance from pulsar along line of sight toward observer
$z_o$ Observer-screen distance
$z_e$ Effective screen observer-distance $=z_o z_p/L$
$\zeta$ $(s_R \kappa_o)^2$
--------------------------------------------------------- -----------------------------------------------------------------------------------
It appears that different spectral models need to be considered for different lines of sight. A widely distributed turbulent plasma with occasional large ionized structures that increase the effective average power density, $P_{N_e}$, at low wavenumbers (large scales: $10^{11}$ m to $10^{14}$ m) is thus a model that needs further formal investigation. Very similar conclusions have been reached by Lestrade et al. (1998) and by Bhat et al. (1999b,c). This model could also explain the occasional “extreme scattering events” and episodes of fringes in dynamic spectra when a line of sight passes through a particular discrete density enhancement. New theoretical work is needed to quantify the expected statistics of these propagation events. It is likely that numerical modeling will be necessary to model non-stationary scattering media.
**Acknowledgements: this work was supported by the NSF under grant AST-9414144.**
THEORY FOR REFRACTIVE SCINTILLATION INDEX
=========================================
Screen Geometry {#screen-geometry .unnumbered}
---------------
Here we derive expressions for the refractive modulation index, $m_R$, as a function of the normalized diffractive decorrelation bandwidth, $\Delta\nu_d/\nu$, for the inner-scale and $\beta=4$ models. The simple Kolmogorov model is included as a special case where the inner scale goes to zero. A general integral expression for the refractive scintillation index is given by equation (\[eq:crextmed\]). With the medium concentrated into a thin layer (thickness $\Delta z$ at $z_p$ from the source) and an assumption of isotropy in the density spectrum, the equation becomes: $$\begin{aligned}
m_R^2 & = &\frac{16 \pi^2 r_{\rm e}^2 c^2 \Delta z}{\nu^2}
\int_0^{\infty}
P_{N_e}(\kappa)
\exp\left[-D_{\phi}(r_{\rm F,scr}^2 \kappa)\right] \nonumber \\
& & \times \sin^2\left(\frac{r_{\rm F,scr}^2 \kappa^2}{2}\right)
\kappa \, d\kappa
\, \mbox{.}
\label{eq:app:mrscreen}
$$ For both the inner-scale and $\beta=4$ models, the exponential function in equation (\[eq:app:mrscreen\]) may be approximated by the Gaussian (Coles et al. 1987): $$\exp\left[ -D_{\phi}(r_{\rm F,scr}^2 \kappa) \right]
\simeq
\exp\left[ -(r_{\rm F,scr} u_{\rm scr})^2 \kappa^2 \right]
\, \mbox{,}
\label{eq:app:gauss}
$$ where $r_{\rm F,scr}$ is the Fresnel scale and $u_{\rm scr}$ is the strength of scattering, as defined in section \[sec:mr\]. We now substitute models for the density spectrum $P_{N_e}(\kappa)$, given by equations (\[eq:extpowerlaw\]) (with $\kappa_o = 0$) and (\[eq:beta4\]) and obtain relations between $m_R$ and the scattering measure $SM = C_{N_e}^2 \Delta z$. To these we add the connections between $SM$ and $s_{\rm 0,scr}$, for each spectral model, from the equations for the phase structure functions (17), (11), and (21) of LR99.
For the inner-scale model, we obtain: $$\begin{aligned}
\frac{8 \pi^2 r_{\rm e}^2 c^2 SM}{\nu^2} & = &
\frac{\alpha 2^{\alpha}} {s_{\rm 0,scr}^{\alpha}}
\frac{\Gamma\left(1 + \alpha/2 \right)}
{\Gamma\left(1 - \alpha/2 \right)} \times \nonumber \\
& & \left[
1 + \left(\frac{L_i}{s_{\rm 0,scr}}\right)^{\alpha} \mu_{\alpha}
\right]^{(2-\alpha)/\alpha}
\, \mbox{,}
\label{eq:app:sminscale}
$$ where $$\mu_{\alpha} =
\left[
\frac{\alpha}{2}
\Gamma\left(1 + \frac{\alpha}{2}\right)
\right]^{\alpha/(\alpha-2)}
\, \mbox{.}
\label{eq:app:mualpha}
$$ Then substituting for $SM$ we obtain an integral for $m_R^2$, which can be solved analytically using standard techniques (see e.g. Appendix 2 of Rickett 1973): $$\begin{aligned}
m_R^2 & = &
2^{\alpha}
\Gamma\left( 1 + \frac{\alpha}{2}\right)
\left[
1 + \left(\frac{L_i}{s_{\rm 0,scr}}
\right)^{\alpha} \mu_{\alpha}
\right]^{(2-\alpha)/\alpha} \times \nonumber \\
& & \left[ (\cos\psi)^{-\alpha/2}
\cos(\psi\alpha/2) - 1
\right] \, \times \nonumber \\
& &
\left[ (L_i/2s_{\rm 0,scr})^2 + u_{\rm scr}^4 \right]^{\alpha /2}
\, \mbox{,}
\label{eq:app:mrscreeninscale}
$$ where $u_{\rm scr} = r_{\rm F,scr}/s_{\rm 0,scr}$, and $\psi$ is given by: $$\psi = \cot^{-1} \left[
\left( \frac{L_i}{2 s_{\rm 0,scr}u_{\rm scr}} \right)^2
+ u_{\rm scr}^2 \right]
\, \mbox{.}
\label{eq:app:thetae}
$$ We see that $m_R$ depends on $u_{\rm scr}$ with $L_i/s_{\rm 0,scr}$ as a parameter. Similarly the diffractive decorrelation bandwidth, $\Delta\nu_d/\nu$, depends on $u_{\rm scr}$ through equation (\[eq:Delta\_nu\_d\]) with $v_d$ as a parameter, which is a very slow function of $L_i/s_{\rm 0,scr}$ as discussed in section \[sec:decbw\].
For the simple power-law model we set $L_i = 0$, and in the strong scattering limit ($u_{\rm scr}>>1$), equation (\[eq:app:thetae\]) gives $\psi \approx u_{\rm scr}^{-2}$, and we obtain the asymptotic expression: $${\displaystyle
m_R^2 = \frac{\alpha 2^{\alpha}}{4}
\left(1 - \frac{\alpha}{2}\right)
\Gamma\left(1 + \frac{\alpha}{2}\right)
\left(\frac{1}{u_{\rm scr}^2}\right)^{2-\alpha}
}
\, \mbox{.}
\label{eq:app:mrscreenkol}
$$ Putting this together for the simple Kolmogorov model with $\alpha=5/3$, we have the simple relation: $m_R = 0.459 (\Delta\nu_d/\nu)^{0.167}$, plotted as the thick dashed lines in the screen plots in Figures \[fig:mr4\] and \[fig:mrin\].
For the general inner-scale model, with $L_i$ non-zero, we show computed curves in Figure \[fig:mrin\]. We can recognize two regimes. Consider first the case of small $L_i$. With sufficiently strong scattering, $s_{\rm 0,scr} < L_i < r_{\rm F,scr}$, $m_R^2 \propto (L_i/u_{\rm scr} r_{\rm F,scr})^{2-\alpha}$. However, there is a complication in portraying this behavior in a plot versus $\Delta \nu_d/\nu$ for fixed inner scale, because of the variation of $r_{\rm F,scr}$ with distance, which is not specified by the horizontal variable $\Delta \nu_d/\nu$. We deal with this by obtaining an approximate scaling of $r_{\rm F,scr}$ with $\Delta \nu_d/\nu$. When $s_{\rm 0,scr} < L_i$, equation (\[eq:app:sminscale\]) relates $SM \propto s_{\rm 0,scr}^{-2}$, or $SM r_{\rm F,scr}^2 \propto u_{\rm scr}^{2}$. We now argue that on average $SM \propto$ pulsar distance $\propto r_{\rm F,scr}^2$, and eliminating $SM$, we see that, $r_{\rm F,scr} \propto u_{\rm scr}^{0.5}$. With these scalings and $s_{\rm 0,scr} < L_i < r_{\rm F,scr}$, we see: $$\begin{aligned}
m_R^2 & \propto & (L_i/u_{\rm scr} r_{\rm F,scr})^{2-\alpha}
\propto (L_i/r_{\rm F,ref} u_{\rm scr}^{1.5} )^{2-\alpha} \nonumber \\
& \propto & u_{\rm scr}^{-0.5} \propto (\Delta \nu_d/\nu)^{0.25}
\, \mbox{.}
\label{eq:mr.asymptote1}
$$ Here $r_{\rm F,ref}$ is the Fresnel scale for a “reference” pulsar, and the last version in equation (\[eq:mr.asymptote1\]) comes from using the Kolmogorov exponent $\alpha = 5/3$, which gives the asymptotic slope of 1/8 for curves at the lower left of Figure \[fig:mrin\]. Now with $ L_i$ and $r_{\rm F,scr}$ fixed, let $s_{\rm 0,scr}$ increase (i.e. less scattering) until $L_i < s_{\rm 0,scr} < r_{\rm F,scr}$. The inner scale is no longer important, and we get the same relation as for the simple Kolmogorov spectrum: $m_R^2 \propto u_{\rm scr}^{-2(2-\alpha)}
\propto (\Delta \nu_d/\nu)^{1/3}$; visible where the curves steepen with increasing $\Delta \nu_d/\nu$.
For larger inner scales, the curves in Figure \[fig:mrin\] show a pronounced peak. These occur for inner scales greater than the Fresnel scale and will be associated with focusing and caustics. Since our treatment only includes the first order term in the low wavenumber expansion, it is not reliable in the region of the peak. Goodman et al. (1987) have discussed caustics at length for the same inner-scale spectrum. They note that when $s_{\rm 0,scr} < r_{\rm F,scr} < L_i < s_R$, the scintillation power spectrum starts to fill in at scales intermediate between the diffractive and refractive wavenumbers. Their equation (2.5.7) gives an estimate of the variance in this extra term as: $$\begin{aligned}
m_{\rm int}^2 & \sim & 2 (L_i/s_R)^2 \ln (L_i/s_{\rm 0,scr}) \nonumber \\
& \sim & 2 (L_i/u_{\rm scr} r_{\rm F,scr})^2
\ln (L_i u_{\rm scr}/r_{\rm F,scr})
\mbox{.}
\label{eq: mr.asymptote2}
$$ With fixed $L_i$, $m_{\rm int}^2$ increases much more steeply with decreasing $u_{\rm scr}$ than does $m_R^2$. Thus as $u_{\rm scr} r_{\rm F,scr}$ decreases, there are fewer and fewer independent phase perturbations across the scattering disc. When $u_{\rm scr} r_{\rm F,scr}
\sim L_i$, focusing represented by $m_{\rm int}^2
\sim 4 \ln(L_i/r_{\rm F,scr}) > 1$ occurs. We note that the theoretical $m_R$ versus $\nu_d/\nu$ plots in Gupta et al. (1993) omit the focusing condition and are wrong for inner scales greater than the Fresnel scale. The effect of higher order terms in the low-wavenumber expansion has been studied by Dashen & Wang (1993). They obtain a more efficient expansion scheme that gives improved accuracy near the peak in scintillation index. Nevertheless, it seems that a reliable prediction for the behavior near the peak in scintillations requires numerical evaluation. This becomes even more necessary in treating an extended scattering medium.
We also note that the drop in $m_R$ as scattering gets weaker past the peak in Figure \[fig:mrin\] is real. It represents the fact that when a square-law structure function applies for scales from $s_{\rm 0,scr}$ up to the scattering disc size, there is insufficient phase curvature and the scintillations remain weak even though $u_{\rm scr} > 1$. In such circumstances the “scattering disc” is a misnomer, since an observer would see only a single angle of arrival, that could wander over a region of scale $u_{\rm scr} r_{\rm F,scr}$.
Turning to the screen analysis of the $\beta=4$ model, we relate $s_{\rm 0,scr}$ to $SM$ using equation (21) from LR99, where the phase structure function equals one. This gives the following analog of equation (\[eq:app:sminscale\]) for the inner-scale model: $$\frac{\pi^2 r_{\rm e}^2 c^2 SM}{\nu^2} =
\left\{
s_{\rm 0,scr}^2 \ln\left[1 + 4 \left(
\frac{L_o}{s_{\rm 0,scr}}\right)^2 \right]
\right\}^{-1}
\mbox{.}
\label{eq:app:smbeta4}
$$ Substituting the $\beta=4$ model for the density spectrum in equation (\[eq:app:mrscreen\]) and letting $\eta = \kappa^2$, we obtain: $$\begin{aligned}
m_R^2 \; = \; \frac{8 \pi^2 r_{\rm e}^2 c^2 SM}{\nu^2} \;
\int_0^{\infty}
\frac{1}{(\eta + \kappa_o^2)^2} \; \times \nonumber \\
\exp [ -(r_{\rm F,scr} u_{\rm scr})^2\eta ] \;
\sin^2 ( r_{\rm F,scr}^2 \eta/2 ) \; d\eta
\label{eq:app:mrscreenb4_1}
$$ In order to evaluate this integral, we let $g(\eta)$ represent the integrand, and we let $g_1(\eta)$ be the same as $g(\eta)$ but with $\kappa_o$ set equal to zero. We can then rewrite the integral in equation (\[eq:app:mrscreenb4\_1\]) as: $$\begin{aligned}
& & m_R^2 \;\; = 8
\left\{
s_{\rm 0,scr}^2 \ln\ [1 + (2 L_o/s_{\rm 0,scr})^2 ]
\right\}^{-1} \times \nonumber \\
& & \left\{
\int_0^{\infty} g_1(\eta) d\eta -
\int_0^{\infty} [g_1(\eta) - g(\eta) ] d\eta
\right\}
\label{eq:app:mrscreenb4_2}
$$ The first integral can be evaluated analytically (see e.g. & 1965). In strong scattering, the exponential term cuts off the oscillations of the sine term, which can be approximated by its argument, and the $g_1(\eta) - g(\eta)$ becomes negligible for values of $\eta$ larger than $\kappa_o^2$. With these approximations we can also do the second integral. Putting these together, we obtain: $$\begin{aligned}
& & m_R^2 =
\left\{ \ln [ 1 + (2 L_o/s_{\rm 0,scr})^2 ] \right\}^{-1} \times \nonumber \\
& & \left\{ 4 u_{\rm scr}^2 \tan^{-1}(u_{\rm scr}^{-2}) -
2 u_{\rm scr}^4\ln (1 + u_{\rm scr}^{-4} ) - \right. \nonumber \\
& & \left. 2 \zeta \, [(2 + \zeta) \, e^{\zeta} \, {\rm E_1}(\zeta) - 1 ]
\right\}
\, \mbox{,}
\label{eq:app:mrscreenb4}
$$ where ${\rm E_1}$ is the exponential integral, $\zeta = (s_0\kappa_o)^2 u_{\rm scr}^4 = (s_R \kappa_o)^2$. In the latter form $s_R = r_{\rm F,scr} u_{\rm scr}$ is the refractive scale (equal to scattering disk radius). Again, $m_R$ is related to $\Delta\nu_d/\nu$ through: $\Delta\nu_d/\nu = v_d/u_{\rm scr}^2$. The resulting curves are shown in the screen panels of Figure \[fig:mr4\]. Consider the asymptotic behavior for $m_R^2$ in equation (\[eq:app:mrscreenb4\]) as a function of $u_{\rm scr}$. As the strength of scattering $u_{\rm scr}$ increases, $\Delta\nu_d/\nu$ decreases ($\propto u_{\rm scr}^{-2}$). In equation (\[eq:app:mrscreenb4\_1\]), the exponential term cuts off the integral (at $1/s_R^2$) before the oscillations of the sin$^2$ Fresnel filter, which then approximates $\eta^2 r_{\rm F,scr}^4/4$. If also $\zeta = (s_R \kappa_o)^2 \ll 1$, we can ignore $\kappa_o^2$ in the denominator and the remaining $\eta^{-2}$ cancels the $\eta^2$ from the Fresnel filter, and the integral depends only on $1/s_R$. In this approximation, $m_R^2$ is then simply proportional to the slowly varying logarithmic term, which explains the relatively flat part of the curves in Figure \[fig:mr4\]; under these conditions, in equation (\[eq:app:mrscreenb4\]) the first two terms in the curly brackets sum to 2 and the last term is negligible. With $L_o$ fixed, now let $u_{\rm scr}$ increase, making $s_R$ increase. Eventually $\zeta$ becomes greater than one when the scattering disk becomes greater than the outer scale. At this point the exponential term cuts off the integral below $\kappa_o^2$, where the spectrum flattens. As $u_{\rm scr}$ increases even further, the integral decreases steeply, causing the down-turn at very small $\Delta\nu_d/\nu$. We again note that our expressions rely on the first order of an expansion and will not be reliable near the peak in the scintillation index. However, there is not the same focusing condition that applied for very large inner scales.
Extended Scattering Medium {#extended-scattering-medium .unnumbered}
--------------------------
In order to obtain expressions for $m_R$ for the two spectrum models in the *uniform extended medium geometry, we must complete the line-of-sight integrals in equation (\[eq:crextmed\]) in addition to following the steps used in the screen geometry. For each distance $x = z/L$ in the line of sight, there is also an integration over variable $y$ in the exponential cut-off. If $D'(s) \propto s^{\alpha}$, this $y$-integration yields $L D'(\kappa r_{F,L}^2 x(1-x))/(\alpha+1)$, where $r_{{\rm F},L}= \sqrt{Lc/(2\pi\nu)}$. This again provides a low-pass cut-off at the reciprocal of the radius of the effective scattering disk, where $\kappa \sim (u_{S}r_{{\rm F},L})^{-1}$. We define the scattering strength by $u_S = r_{{\rm F},L}/s_{0_S}$, with the field coherence scale $s_{0_S}$, as in LR99, defined where the spherical wave structure function equals unity, measured in the observing plane. For the other spectrum models there is not such a simple relation for the $y$-integration, but there is still an effective cutoff given by a similar equation. When the $x$-integration is completed, the effective Fresnel scale is actually smaller than $r_{{\rm F},L}$ due to averaging over the $\sqrt{x (1-x)}$.*
In analogy with the screen geometry, we make use of identities similar to those given by equations (\[eq:app:sminscale\]) and (\[eq:app:smbeta4\]). For the extended medium, these identities are in turn derived from the [*wave*]{} structure functions for the inner-scale and $\beta=4$ models (cf. LR99). The identities obtained thus will be similar to those for the screen geometry, except that for the inner-scale model, (1) there will be a factor of 3 on the right side of equation (\[eq:app:sminscale\]), and (2) the 1 in square brackets is replaced by $[3 / (1+\alpha)]^{\alpha/(\alpha-2)} \approx 1.8$ for the Kolmogorov exponent. For the $\beta=4$ model, the only change will be a factor of 3 on the right side of equation (\[eq:app:smbeta4\]). We use all of the aforementioned identities for the inner-scale and $\beta=4$ models and compute the $x$ integral numerically, since it cannot be carried out analytically.
The shapes of the curves bear a close relationship to the screen results, though the extended medium values generally lie above the associated screen values at the same $\Delta \nu_d/\nu$.
Diffractive Intensity Correlation Function
==========================================
The second moment of intensity is needed to describe the fluctuations of intensity. Under strong scintillation conditions, separate forms can be used for refractive and diffractive fluctuations, since their spatial scales differ by several orders of magnitude. In LR99, as elsewhere in the ISS literature, the correlation of diffractive scintillations is approximated by the squared magnitude of the second moment of the field, leading to the simple result that the spatial scale of the diffractive scintillations is equal to the scale where the phase structure function equals unity ($s_0$). However, Goodman and Narayan (1985) showed that for steep spectra ($\beta>4$) this is no longer the case and the diffractive scale can be larger than $s_0$. Here we examine this question for the $\beta=4$ spectrum. We give the details for a phase screen with plane wave source, which are readily generalized to a spherical wave source.
The two-frequency intensity cross-spectrum at wavenumber [$ \kappa$]{} for a screen at distance $L$ is given by the Fourier-like integral equation (17) of CCFFH. This depends on the combination of structure functions $V_4$, which for a plasma screen can be written as: $$\begin{aligned}
V_4 = \frac {k_m^2}{k_1^2} D_{\phi}(\mbox{\boldmath $\kappa$} \frac{L}{k_1})
+ \frac {k_m^2}{k_2^2} D_{\phi}(\mbox{\boldmath $\kappa$} \frac{L}{k_2})
- D_{\phi}( \frac{\mbox{\boldmath$\kappa$}L}{k_o} + \mbox{\boldmath$\beta $}')\nonumber \\
- D_{\phi}( \frac{\mbox{\boldmath$\kappa$}L}{k_o} - \mbox{\boldmath$\beta $}')
+ D_{\phi}(\mbox{\boldmath$\beta $}' + \frac{\mbox{\boldmath$\kappa$} \epsilon L}{k_o})
+ D_{\phi}(\mbox{\boldmath$\beta $}' - \frac{\mbox{\boldmath$\kappa$} \epsilon L}{k_o}) \, ,
\label{eq:v4}\end{aligned}$$ where $k_1$ and $k_2$ are the two radio wavenumbers, $k_m$ is their geometric mean, $\bar{k}$ is their arithmetic mean, $k_o = k_m^2/\bar{k}$, $\epsilon=|k_1-k_2|/2\bar{k}$, and $D_{\phi}$ is evaluated at $k_m$; $\mbox{\boldmath$\beta $}'$ is a spatial offset which is the variable of integration.
Consider first the single frequency case $k_1=k_2 (\epsilon=0$). In the limit of very large $\kappa$, the first four structure functions saturate and sum to zero. The last two are equal and $V_4 \approx 2 D_{\phi}(\beta ')$, which gives the simple diffractive limit mentioned above. This is the zero-order term of an expansion, which is obtained in terms of the sum of the first four terms as a small quantity. The zero order result requires full saturation, which requires $\kappa L/k_m \simgreat L_o$. In diffractive scintillation $\kappa \sim 1/s_0$; hence, the condition becomes that the refractive scale $s_R = L/(k_m s_0) \simgreat L_o$. Our concern here is to consider what happens when the diffractive $\kappa$ is not large enough for saturation of $D_{\phi}$. For shallow density spectra ($\beta < 4$), small argument approximations to the structure function follow an exponent $\beta-2 < 2$, and the zero-order term gives a good approximation even when $s_R < L_o$. However, for steep spectra Goodman and Narayan (1985) showed that the leading term in the structure function follows a square law, which exactly cancels in $V_4$; the result is that the high wavenumber limit depends on the next term in the structure function expansion, which yields a diffractive scale that is greater than the scale $s_0$ (defined by the square law term).
Now we consider the case for $\beta=4$ model, when the scattering disc $s_R$ is smaller than the outer scale $L_o$. Here we can approximate equation (\[eq:struc4\]) by: $$D_{\phi}(s) = \frac {s^2}{s_0^2} - \frac {s^2 \ln(s^2/s_0^2)}
{s_0^2 \ln(4/s_0^2\kappa_o^2)}
\, \mbox{.}
\label{eq:Dapprox}$$ As for the steep spectra, the leading term in the structure function follows a square law, which cancels when substituted into equation (\[eq:v4\]). $V_4$ can then be approximated for large $\kappa$ by expanding in $\beta ' k_m/(\kappa L)$. The result is: $$V_4 \sim \frac {2 \beta '^2[\ln(u^4) + 1 + 2 \cos^2\theta - \ln(\beta '^2/s_0^2)]}
{s_0^2 \ln(4/s_0^2\kappa_o^2)}
\, \mbox{.}
\label{eq:v4approx}$$ Here $u$ is the strength of scattering defined in equation (\[eq:Delta\_nu\_d\]); $\theta$ is the angle between vectors [$\kappa$]{} and $\mbox{\boldmath $\beta$}'$. We note that for a Kolmogorov spectrum in the high-wavenumber limit, $V_4$ also includes terms in $\cos^2\theta$, which would be accounted for in the higher order terms of the expansion. A result similar to equation (\[eq:v4approx\]) is given by Dashen & Wang (1993), though for a 1-dimensional phase screen. In considering the spectrum of intensity fluctuations (eq. 17 of CCFFH), one can show that the dominant wavenumber is approximately $\kappa \sim \beta '^{-1}$ With these substitutions in $V_4$, which we then set $=2$, we solve for $\beta '$; this gives an approximate equation for the diffractive spatial scale $s_d$: $$s_d^2 \sim s_0^2 \frac {\ln(4/s_0^2\kappa_o^2)}{\ln(u^4)+ 2}
\, \mbox{.}$$ Under the condition assumed in this approximation, $s_R \ll L_o$, we find $s_d > s_0$. However, in practice the ratio $s_d/s_0$ never becomes large. With a large outer scale, say $L_o = 3$ pc, and typical observing conditions $s_0 \sim 10^8$ m and $u \simgreat 100$, we find $s_d \simless 1.7 s_0$. Thus the diffractive scale could be 70% greater than $s_0$ and would slowly approach $s_0$ for smaller outer scales.
Turning to the two-frequency intensity correlation ($0 < \epsilon \ll 1$) in the diffractive limit of large $\kappa$, the results of CCFFH still apply. Namely, that the last two terms of equation (\[eq:v4\]) largely control the decorrelation versus frequency. They group the remaining terms into a filter that depends only on $\kappa$ and a smaller term that becomes the basis of the expansion. The filter term was discussed by LR99 and shown to be important only as the strength of scattering decreases. It is the last two terms in $V_4$ that determine the zero-order result, so we looked at the effect of the higher order terms. The quantity that we are ultimately concerned with is the cross-correlation of intensity at offset frequencies at the same observing point. This comes from the integral of the cross-spectrum. Equations (31) through (34) of CCFFH give the zero and first order terms of the cross spectrum in terms of the spectrum of refractive index fluctuations in the layer. For the $\beta=4$ spectrum we reduced these to a sum of confluent hypergeometric functions, which can be explicitly computed. For a sample observing condition we found that the higher order terms for the cross-spectrum itself were significant compared to the zero-order term; however when integrated to give intensity cross-correlation, they only had a minor effect on the decorrelation bandwidth itself ($\simless 5$% increase). The reason for this appears to be the dominant effect of the last two terms in the $V_4$ summation with unequal frequencies.
To summarize we find a modest (logarithmic) increase in the diffractive scale relative to the field coherence scale $s_0$, but that this remains less than a factor of 1.7 for the likely ISS parameters. This is accompanied by a smaller increase in the decorrelation bandwidth relative to the calculations of LR99, which relied on the normal zero-order expansion at high wavenumbers. This small offset in the decorrelation bandwidth is negligible compared to the measurement errors for the observations under consideration. We assume that the conclusions reached here for a screen would also apply for an extended scattering medium.
Armstrong, J. W., Rickett, B. J., & Spangler, S. R. 1995, , 443, 209
Bhat, N. D. R., Gupta, Y., & Rao, A. P. 1998, , 500, 262
Bhat, N. D. R., Rao, A. P., & Gupta, Y. 1999a, , 121, 483
Bhat, N. D. R., Rao, A. P., & Gupta, Y. 1999b, , 514, 249
Bhat, N. D. R., Gupta, Y., & Rao, A. P. 1999c, , 514, 272
Biskamp, D. 1993, *Nonlinear Magnetohydrodynamics , Cambridge Univ. Press, Cambridge, UK*
Blandford, R., & Narayan, R. 1985, , 213, 591
Codona, J. L., Creamer, D. B., Flatté, S. M., Frehlich, R. G., & Henyey, F. S. 1986, Radio Science, 21, No. 5, 805 \[CCFFH\]
Cole, T. W., Hesse, H. K., & Page, C. G. 1970, Nature, 225, 712
Coles, W. A., Frehlich, R. G., Rickett, B. J., & Codona, J. L. 1987, , 315, 666
Cordes, J. M., Weisberg, J. M., & Boriakoff, V. 1985, , 288, 221 \[CWB\]
Cordes, J. M., Pidwerbetsky, A., & Lovelace, R. V. E. 1986, , 310, 737
Cordes, J. M., & Wolszczan, A. 1986, , 307, L27
Cordes, J. M., Weisberg, J. M., Frail, D. A., Spangler, S. R., & Ryan, M. 1985, , 288, 221
Dashen, R., & Wang, G. Y. 1993, J. Opt. Soc. Am. A, 10, 1219
Fiedler, R. L., Dennison, B., Johnston, K. J., & Hewish, A. 1987, Nature, 326, 675
Goldreich, P., & Sridhar, S. 1995, , 438, 763
Goldreich, P., & Sridhar, S. 1997, , 485, 680
Goodman, J., & Narayan, R. 1985, , 214, 519
Goodman, J., Romani, R. W., Blandford, R. D., & Narayan, R. 1987, , 229, 73
Gradshteyn, I. S., & Ryzhnik, I. M. 1965, Table of Integrals, Series, and Products (New York: Academic Press)
Gupta, Y., Rickett, B. J., & Coles, W. A. 1993, , 403, 183
Gupta, Y., Rickett, B. J., & Lyne, A. G. 1994, , 269, 1035
Gupta, Y., Bhat, N. D. R., & Rao, A. P. 1999 , in press
Helfand, D. J., Fowler, L. A., & Kuhlman, J. V. 1977, , 82, 701
Higdon, J. C. 1984, , 285, 109
Higdon, J. C. 1986, , 309, 342
Johnston, S., Nicastro, L., & Koribalski, B. 1998, , 297, 108
Kaspi, V. M., & Stinebring, D. R. 1992, , 392, 530
Lambert, H. C. 1998, Interstellar Electron Density Spectra, PhD Thesis, University of California, San Diego
Lambert, H. C., & Rickett, B. J. 1999, submitted to
Lestrade, J-F, Rickett, B. J., & Cognard, I. 1998, A&A, 334, 1068
Minter, A. H., & Spangler, S. R. 1997, , 485, 182
Narayan, R. 1992, Phil. Trans. R. Soc. London A, 341, 151
Papoulis, A. 1991, Probability, Random Variables, and Stochastic Processes, Third Edition (New York: McGraw-Hill Inc.)
Pouquet, A. 1978, J. Fluid Mech., 88, 1
Prokhorov, A. M., Bunkin, F. V., Gochelashvily, K. S., & Shishov, V. I. 1975, Proc. IEEE, 63, 790
Ratcliffe, J. A. 1956, Reports on Progress in Physics 19, 188
Rickett, B. J. 1973, , 78, 1543
Rickett, B. J. 1977, , 15, 479
Rickett, B. J., Coles, W.A., & Bourgois, G. 1984, , 134, 390
Rickett, B. J., & Lyne, A. G. 1990, , 244, 68
Rickett, B. J., Lyne, A. G., & Gupta, Y. 1997, , 287, 739
Rickett, B. J., Coles, W. A., & Markkanen, J. 1999, submitted to
Roberts, J. A., & Ables, J. G. 1982, , 201, 1119
Romani, R. W., Narayan, R., & Blandford, R. 1986, , 220 19
Smirnova, T. V., Shishov, V. I., & Stinebring, D. R. 1998, Astronomical Reports, 42, 766
Spangler, S. R., & Gwinn, C. R. 1990, , 353, L29
Spangler, S. R. 1991, , 376, 540
Spangler, S. R. 1999, , 522, 879
Sridhar, S., & Goldreich, P. 1994, , 432, 612
Stinebring, D. R., Smirnova, T. V., Hovis, J., Kempner, J. C., Myers, E. B., Hankins, T. H., Kaspi, V. M., & Nice, D. J. 1996, in “Pulsars: Problems and Progress”, Ed. Johnston, S., Walker, M. A., & Bailes, M., Proceedings of IAU Colloquium 160, ASP Conference Series, 105, 455
Tatarskii, V. I. 1961, Wave Propagation in a Turbulent Medium (New York: Dover Publications)
Taylor, J. H., & Cordes, J. M. 1993, , 411, 674
Taylor, J. H., Manchester, R. N., & Lyne, A. G. 1993, , 88, 529
Trotter, A. S., Moran, J. M., & Rodríguez, L. F. 1998, , 493, 666
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper we consider the initial-boundary value problem for the time-dependent Maxwell-Schrödinger equations, which arises in the interaction between the matter and the electromagnetic field for the semiconductor quantum devices. A Crank-Nicolson finite element method for solving the problem is presented. The optimal energy-norm error estimates for the numerical algorithm without any time-step restrictions are derived. Numerical tests are then carried out to confirm the theoretical results.'
author:
- 'Chupeng Ma[^1]'
- 'Liqun Cao[^2]'
title: 'A Crank-Nicolson Finite Element Method and the Optimal Error Estimates for the modified Time-dependent Maxwell-Schrödinger Equations [^3]'
---
time-dependent Maxwell-Schrödinger equations, finite element method, Crank-Nicolson scheme, optimal error estimate.
65N30, 65N55, 65F10, 65Y05
Introduction {#sec-1}
============
When the characteristic size of the semiconductor device reaches the wavelength of an electron, the quantum effects become important even dominant and can not be neglected. The accurate electromagnetic theory for the case is quantum electrodynamics (QED), i.e. the second quantization for the matter and quantization for the electromagnetic field. However, so far it is extremely difficult even impossible to employ QED to analyze the interaction between the matter and the electromagnetic field for some complex systems. The semiclassical (or semi-quantum) electromagnetic models are widely used in the semiconductor quantum devices. The basic idea is that we use the Maxwell’s equations for the electromagnetic field while we use the Schrödinger equation of the non-relativistic quantum mechanics for the matter (see [@Ez; @Sch]). The Maxwell-Schrödinger coupled system (M-S) is written as follows: $$\label{eq:1-1}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle \mathrm{i}\hbar\frac{\partial \Psi(\mathbf{x},t)}{\partial t}=
\left\lbrace\frac{1}{2m}\left[\mathrm{i}\hbar\nabla +q\mathbf{A}(\mathbf{x},t)\right]^{2}
+ q \Phi(\mathbf{x},t)+V_{0} \right\rbrace\Psi(\mathbf{x},t),}\\[2mm]
{\displaystyle \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad(\mathbf{x},t)\in\Omega\times(0,T),}\\[2mm]
{\displaystyle -\frac{\partial}{\partial t}\nabla\cdot\big(\epsilon\mathbf{A}(\mathbf{x},t)\big)
-\nabla\cdot\big(\epsilon\nabla\Phi(\mathbf{x},t)\big) =q |\Psi(\mathbf{x},t)|^{2}, \,\,
(\mathbf{x},t)\in\Omega\times(0,T),} \\[2mm]
{\displaystyle \epsilon\frac{\partial ^{2}\mathbf{A}(\mathbf{x},t)}{\partial t^{2}}+\nabla\times
\big({\mu}^{-1}\nabla\times \mathbf{A}(\mathbf{x},t)\big)
+\epsilon \frac{\partial (\nabla \Phi(\mathbf{x},t))}{\partial t} =\mathbf{J}_{q}(\mathbf{x},t),}\\[2mm]
{\displaystyle \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad \qquad
(\mathbf{x},t)\in \Omega\times(0,T),}\\[2mm]
{\displaystyle \mathbf{J}_q=-\frac{\mathrm{i}q\hbar}{2m}\big(\Psi^{\ast}\nabla{\Psi}-\Psi\nabla{\Psi}^{\ast}\big)-\frac{\vert q\vert^{2}}{m}\vert\Psi\vert^{2}\mathbf{A},} \\[2mm]
{\displaystyle \Psi, \Phi, \mathbf{A} \,\, \mathrm{subject \ to \ the \ appropriate \ initial \ and\ boundary \ conditions}, }
\end{array}
\right.$$ where $ \Omega\subset \mathbb{R}^d $, $ d\geq 2 $ is a bounded Lipschitz polyhedral convex domain, $ \Psi^{\ast} $ denotes the complex conjugate of $ \Psi $, $\epsilon$ and $\mu$ respectively denote the electric permittivity and the magnetic permeability of the material and $V_0$ is the constant potential energy.
It is well-known that the solutions of the Maxwell-Schrödinger equations (\[eq:1-1\]) are not unique. In fact, for any function $\chi : \Omega\times (0,T)\rightarrow \mathbb{R}$, if $(\Psi, \Phi, \mathbf{A}) $ is a solution of (\[eq:1-1\]), then $(\exp(i\chi)\Psi, \Phi-\partial_{t}\chi, \mathbf{A}+\nabla \chi) $ is also a solution of (\[eq:1-1\]). It is often assumed that the further equations can be adjoined to the Maxwell-Schrödinger equations by means of a gauge transformation. In this paper we consider the M-S system (\[eq:1-1\]) under the temporal gauge (also called Weyl gauge), i.e. $\Phi=0 $.
In this paper we employ the atomic units, i.e. $\hbar=m=q=1 $. For simplicity, we also assume that $\epsilon=\mu=1$ without loss of generality. Hence, $\Psi$ and $\mathbf{A}$ satisfy the following Maxwell-Schrödinger equations : $$\label{eq:1-2}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}\frac{\partial \Psi}{\partial t}+
\frac{1}{2}\left(\mathrm{i}\nabla +\mathbf{A}\right)^{2}\Psi
+ V_{0}\Psi = 0 ,\,\, (\mathbf{x},t)\in
\Omega\times(0,T),}\\[2mm]
{\displaystyle \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}}+\nabla\times
(\nabla\times \mathbf{A}) +\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big)+\vert\Psi\vert^{2}\mathbf{A}=0,
\,\, (\mathbf{x},t)\in \Omega\times(0,T),}\\[2mm]
\end{array}
\right.$$ Here we omit the initial and boundary conditions for $\Psi$ and $\mathbf{A}$ temporarily.
Under the temporal gauge, the second equation in (\[eq:1-1\]) involving the divergence of $\mathbf{A}$ can be rewritten as $$\label{eq:1-2-0}
-\frac{\partial}{\partial t}\nabla\cdot\mathbf{A}(\mathbf{x},t) = |\Psi(\mathbf{x},t)|^{2}, \,\,(\mathbf{x},t)\in\Omega\times(0,T),$$ which can be derived from (\[eq:1-2\]) if the solutions of (\[eq:1-2\]) are sufficiently smooth and the initial datas are consistent.
Integrating with respect to $t$ on the both sides of (\[eq:1-2-0\]), we have $$\label{eq:1-3}
{\displaystyle -\nabla\cdot\mathbf{A}(\mathbf{x},t) + \nabla\cdot\mathbf{A}(\mathbf{x},0) - \int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau = 0 \, ,}$$ where $ \rho(\mathbf{x}, t) =|\Psi(\mathbf{x},t)|^{2} $.
For the purpose of theoretical analysis, we take the gradient of (\[eq:1-3\]), multiply it by a parameter $\gamma > 0$ and add it to the second equation of (\[eq:1-2\]), to obtain $$\label{eq:1-4}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}\frac{\partial \Psi}{\partial t}+
\frac{1}{2}\left(\mathrm{i}\nabla +\mathbf{A}\right)^{2}\Psi
+ V_{0}\Psi = 0 ,\,\, (\mathbf{x},t)\in
\Omega\times(0,T),}\\[2mm]
{\displaystyle \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}}+\nabla\times
(\nabla\times \mathbf{A}) -\gamma \nabla(\nabla \cdot \mathbf{A}) +\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big)+\vert\Psi\vert^{2}\mathbf{A} }\\[2mm]
{\displaystyle \quad\,+\, \gamma \nabla(\nabla \cdot \mathbf{A}(\mathbf{x},0)) - \gamma \nabla \int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau=0,
\,\, (\mathbf{x},t)\in \Omega\times(0,T).}\\[2mm]
\end{array}
\right.$$
The parameter $\gamma $ is referred to as the penalty factor. The choice of $\gamma$ depends on how much emphasis one places on the equality (\[eq:1-2-0\]). In this paper, we keep $\gamma$ fixed. To avoid the difficulty for integro-differential equations, assuming that the change of the density function $\rho(\mathbf{x}, t)$ is smooth with respect to $t$ for all $\mathbf{x} \in \Omega$, we give an approximation of $\int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau $ as follows.
First denoting by $t_j= jT/M, j=0,1,\cdots,M$, we divide the time interval $[0,T]$ into $M$ subintervals $[0, t_1], (t_1, t_2],\cdots, (t_{M-1}, T]$. For $t \in[0, t_1]$, $\rho(\mathbf{x}, t)$ is approximated by Taylor expansion and the initial conditions: $$\rho(\mathbf{x}, t) \approx \rho(\mathbf{x}, 0)+t\frac{\partial \rho}{\partial t}(\mathbf{x}, 0),\quad \forall x \in \Omega$$ and $$\int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau \approx t\rho(\mathbf{x}, 0)
+ \frac{1}{2}t^{2} \frac{\partial \rho}{\partial t}(\mathbf{x}, 0),\quad \forall x \in \Omega.$$
The computation of $\frac{\displaystyle \partial \rho}{\displaystyle \partial t}(\mathbf{x}, 0)$ involves $\frac{\displaystyle\partial \Psi}{\displaystyle\partial t}(\mathbf{x}, 0) $, the time derivative of initial wave function. Here we assume the initial conditions are consistent and so we can obtain $\frac{\displaystyle \partial \Psi}{\displaystyle\partial t}(\mathbf{x}, 0) $ from Schrödinger’s equation. Given an approximation of $\int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau $ in $ [0, t_1]$, we can solve the coupled differential equations (\[eq:1-4\]) in the subinterval $[0, t_1]$ and integrate the density function to obtain $\int_{0}^{t_1}\rho(\mathbf{x}, \tau) d\tau $. Then for $t \in (t_1, t_2]$, $\int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau $ can be calculated as follows: $$\begin{array}{@{}l@{}}
{\displaystyle \int_{0}^{t}\rho(\mathbf{x}, \tau) d\tau = \int_{0}^{t_1}\rho(\mathbf{x}, \tau) d\tau
+\int_{t_1}^{t}\rho(\mathbf{x}, \tau) d\tau} \\[2mm]
{\displaystyle\quad\quad
\quad \approx \int_{0}^{t_1}\rho(\mathbf{x}, \tau) d\tau + (t-t_1)\rho(\mathbf{x}, t_1)+\frac{1}{2}(t-t_1)^{2} \frac{\partial \rho}{\partial t}(\mathbf{x}, t_1), \quad \forall x \in \Omega}.
\end{array}$$
Now we solve the Maxwell-Schrödinger equations (\[eq:1-4\]) in the subinterval $(t_1, t_2]$. Repeating the above procedure, we can solve the Maxwell-Schrödinger equations (\[eq:1-4\]) in the subinterval $(t_2, t_3],\cdots, (t_{M-1}, T]$ successively. Therefore, we decompose the original system in $[0, T] $ into M system in $[0, t_1], (t_1, t_2],\cdots, (t_{M-1}, T]$, respectively. For $t\in (t_{j-1}, t_j], j = 1, 2,\cdots, M$, the Maxwell-Schrödinger equations (\[eq:1-4\]) can be rewritten as follows: $$\label{eq:1-5}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}\frac{\partial \Psi}{\partial t}+
\frac{1}{2}\left(\mathrm{i}\nabla +\mathbf{A}\right)^{2}\Psi
+ V_{0}\Psi = 0 ,\,\, (\mathbf{x},t)\in
\Omega\times(t_{j-1}, t_j],}\\[2mm]
{\displaystyle \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}}+\nabla\times
(\nabla\times \mathbf{A})-\gamma \nabla(\nabla \cdot \mathbf{A})+\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big)}\\[2mm]
{\displaystyle \quad\quad+\,\,\vert\Psi\vert^{2}\mathbf{A} =\mathbf{f}_{j}(\mathbf{x},t),\quad \,\, (\mathbf{x},t)\in \Omega\times(t_{j-1}, t_j],}\\[2mm]
\end{array}
\right.$$ where $\mathbf{f}_j(\mathbf{x},t)$ is the known function.
\[rem1-1\] We can get the modified Maxwell-Schrödinger equations (\[eq:1-5\]) under the assumption that the change of the density function $\rho(\mathbf{x}, t)$ is smooth with respect to $t$ for all $\mathbf{x} \in \Omega$. If the initial wave function $\Psi(\mathbf{x},0)$ is the eigenfunction of the stationary Schrödinger equation and the incoming electromagnetic field is weak and can be considered as a small perturbation to the quantum system, this assumption is reasonable. The choice of the number $M$ of subintervals depends on the initial wave function, the incoming electromagnetic field and $T$.
In this paper, we consider the following modified Maxwell-Schrödinger equations: $$\label{eq:1-6}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}\frac{\partial \Psi}{\partial t}+
\frac{1}{2}\left(\mathrm{i}\nabla +\mathbf{A}\right)^{2}\Psi
+ V_{0}\Psi = 0 ,\,\, (\mathbf{x},t)\in
\Omega\times(0,T),}\\[2mm]
{\displaystyle \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}}+\nabla\times
(\nabla\times \mathbf{A}) - \gamma\nabla(\nabla \cdot \mathbf{A}) +\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big)}\\[2mm]
{\displaystyle \quad\quad+\,\,\vert\Psi\vert^{2}\mathbf{A} =\mathbf{g}(\mathbf{x},t),\quad \,\, (\mathbf{x},t)\in \Omega\times(0,T).}\\[2mm]
\end{array}
\right.$$
The boundary conditions are $$\label{eq:1-7}
{\displaystyle \Psi(\mathbf{x},t)=0,\quad \mathbf{A}(\mathbf{x},t)\times\mathbf{n}=0, \quad \nabla \cdot \mathbf{A}(\mathbf{x},t) = 0, \quad (\mathbf{x},t)\in
\partial \Omega\times(0,T),}$$ and the initial conditions are $$\label{eq:1-8}
{\displaystyle \Psi(\mathbf{x},0) = \Psi_0(\mathbf{x}),\quad\mathbf{A}(\mathbf{x},0)=\mathbf{A}_{0}(\mathbf{x}),\quad
\mathbf{A}_{t}(\mathbf{x},0)=\mathbf{A}_{1}(\mathbf{x}),}$$ where $ \mathbf{A}_{t} $ denotes the derivative of $ \mathbf{A} $ with respect to the time $ t $, $ \mathbf{n}=(n_1, n_2, n_3) $ is the outward unit normal to the boundary $ \partial \Omega $. We assume that $\nabla \cdot \mathbf{A}_{0} = \nabla \cdot \mathbf{A}_{1}= 0$ on $\partial \Omega$.
\[rem1-2\] The boundary condition $
\Psi(x)=0 $ on $\partial \Omega $ implies that the particle is confined in a whole domain $\Omega$. The boundary condition $ \mathbf{A}(\mathbf{x},t)\times\mathbf{n}=0 $ on $\partial \Omega$ is referred to as the perfect conductive boundary condition. The boundary condition $\nabla \cdot \mathbf{A}(\mathbf{x},t) = 0$ on $\partial \Omega$ can be deduced from the boundary condition of $\,\Psi$ and (\[eq:1-3\]) if the initial conditions $\mathbf{A}_{0}$ and $\mathbf{A}_{1}$ satisfy $\nabla \cdot \mathbf{A}_{0} = \nabla \cdot \mathbf{A}_{1}= 0$ on $\partial \Omega$. As for the determination of the boundary conditions for the vector potential $\mathbf{A}$, we refer to [@Weng].
\[rem1-3\] The existence and uniqueness of the solution for the time-dependent Maxwell-Schrödinger equations (\[eq:1-2\]) have been investigated in [@Gin; @Guo; @Nak; @Nak-1; @Nak-2; @Shi; @Wa]. However, the results of the well-posedness of the problem were obtained only for the Cauchy problem in $\mathbf{R}^{d} $, $ d\geq 1 $ instead of the initial-boundary value problem. To the best of our knowledge, the existence and uniqueness of the solution for the Maxwell-Schrödinger equations in a bounded domain seem to be open. For the modified equations (\[eq:1-6\])-(\[eq:1-8\]), we will investigate the existence of solutions in another paper.
Many authors have discussed the numerical methods for the time-dependent Maxwell-Schrödinger equations. We recall some important studies about the problem. Sui and his collaborators [@Sui] used the finite-difference time-domain (FDTD) method to solve the Maxwell-Schrödinger equations and to simulate a simple electron tunneling problem. Pierantoni, Mencarelli and Rozzi [@Pi] applied the transmission line matrix method(TLM) to solve the Maxwell’s equations and employed the FDTD method to solve the Schrödinger equation, and did the simulation for a carbon nanotube between two metallic electrodes. Ahmed and Li [@Ah-1] used the FDTD method for the Maxwell-Schrödinger system to simulate plasmonics nanodevices. The numerical studies listed above all include a step where they extract the vector potential $\mathbf{A}$ and the scalar potential $\Phi$ from the electric field $\mathbf{E}$ and the magnetic field $\mathbf{H}$ after solving the Maxwell’s equations involving $\mathbf{E}$ and $\mathbf{H}$. Recently, Ryu [@Ryu] employed directly the FDTD scheme to discretize the Maxwell-Schrödinger equations (\[eq:1-1\]) under the Lorentz gauge and to simulate a single electron in an artificial atom excited by an incoming electromagnetic field. Other related studies on this topic have been reported in [@Oh; @Sa; @Tur] and the references therein.
There are few results on the finite element method (FEM) of the Maxwell-Schrödinger equations and the convergence analysis. In this paper we will present a Crank-Nicolson finite element method for solving the problem (\[eq:1-6\])-(\[eq:1-8\]), i.e. the finite element method in space and the Crank-Nicolson scheme in time. Then we will derive the optimal error estimates for the proposed method. Roughly speaking, compared with explicit algorithms such as the FDTD method, our method is more stable and suffers from less restriction in the time step-size since we use the Crank-Nicolson scheme in the time direction. Moreover, our method is more appropriate to deal with materials with discontinuous electromagnetic coefficients than the FDTD method. our work is motivated by [@Mu-2] in which Mu and Huang proposed an alternating Crank-Nicolson method for the time-dependent Ginzburg-Landau equations. The optimal error estimates were derived under the time step restrictive conditions $\Delta t \leq O(h^{\frac{11}{12}})$ for the two-dimension model and $\Delta t \leq O(h^{2})$ for the three-dimension model, where $h$ and $\Delta t$ are the spatial mesh size and the time step, respectively. The related convergence results associated with the time-dependent Ginzburg-Landau equations can be also given in [@Bu-1; @Chen-1; @Du-2; @Gao-1; @Mu-1]. It should be emphasized that although the time-dependent Ginzburg-Landau model is somehow formally similar to the time-dependent Maxwell-Schrödinger system, there exists the essential difference between them. The former is classified as a parabolic system and the latter is a hyperbolic system. The main key point in our work is how to avoid using the finite element inverse estimates when dealing with the nonlinear terms. The new ideas are to derive the energy-norm error estimates for the Schrödinger’s equation, and to employ some tricks to eliminate the nonlinear terms both in the Schrödinger’s equation and in Maxwell’s equations, respectively.
The remainder of this paper is organized as follows. In section \[sec-2\], a Crank-Nicolson scheme with the Galerkin finite element approximation for the modified Maxwell-Schrödinger equations (\[eq:1-6\])-(\[eq:1-8\]) is developed. In section \[sec-3\], the stability estimates are given. The optimal error estimates for the numerical solution without any restriction on time step are derived in section \[sec-4\]. Finally, the numerical testes are then carried out to confirm the theoretical results.
Throughout this paper, we denote by $C$ a generic positive constant independent of the mesh size and the time step without distinction.
A Crank-Nicolson Galerkin finite element scheme {#sec-2}
===============================================
In this section, we present a numerical scheme for the modified Maxwell-Schrödinger equations (\[eq:1-6\])-(\[eq:1-8\]) using Galerkin finite element method in space and the Crank-Nicolson scheme in time. To start with, here and afterwards, we assume that $\Omega $ is a bounded Lipschitz polygonal convex domain in $\mathbb{R}^{2}$ (or a bounded Lipschitz polyhedron convex domain in $\mathbb{R}^{3}$).
We introduce the following notation. Let $W^{s,p}(\Omega)$ denote the conventional Sobolev spaces of the real-valued functions. As usual, $W^{s,2}(\Omega)$ and $W^{s,2}_{0}(\Omega)$ are denoted by $H^{s}(\Omega)$ and $H^{s}_{0}(\Omega)$ respectively. We use $\mathcal{W}^{s,p}(\Omega)=\{u+\mathrm{i}v\,|\, u,v \in W^{s,p}(\Omega)\} $ and $\mathcal{H}^{s}(\Omega)=\{u+\mathrm{i}v\,|\, u,v \in H^{s}(\Omega)\}$ with calligraphic letters for Sobolev spaces of the complex-valued functions, respectively. Furthermore, let $\mathbf{W}^{s,p}(\Omega) =[W^{s,p}(\Omega)]^{d} $ and $\mathbf{H}^{s}(\Omega)=[H^{s}(\Omega)]^{d}$ with bold faced letters be Sobolev spaces of the vector-valued functions with $d$ components ($d$=2,3). $ L^{2}$ inner-products in $H^{s}(\Omega) $, $\mathcal{H}^{s}(\Omega)$ and $\mathbf{H}^{s}(\Omega)$ are denoted by $(\cdot,\cdot )$ without ambiguity.
In particular, we introduce the following subspace of $\mathbf{H}^{1}(\Omega)$: $$\mathbf{H}^{1}_{t}(\Omega)=\{\mathbf{A}\,|\,\mathbf{A}\in\mathbf{H}^{1}(\Omega),\,\,\mathbf{A}\times\mathbf{n}=0 \,\, \,{\rm on}\, \,\,\partial\Omega\}$$ The semi-norm on $\mathbf{H}^{1}_{t}(\Omega)$ is defined by $$\Vert \mathbf{u} \Vert_{ \mathbf{H}^{1}_{t}(\Omega)} : = \left[\Vert\nabla \cdot \mathbf{u}\Vert^{2}_{\mathbf{L}^{2}(\Omega)} + \Vert\nabla \times \mathbf{u}\Vert^{2}_{\mathbf{L}^{2}(\Omega)
}\right]^{\frac{1}{2}},$$ which is equivalent to the standard $\mathbf{H}^{1}(\Omega)$-norm $\Vert \mathbf{u} \Vert_{ \mathbf{H}^{1}(\Omega)}$, see [@Gir].
To take into account the time-independence, for a time $T > 0 $ fixed, let $L^p(0,T;X)$ be the Bochner space defined in [@Simon] for $p\in[1,\infty]$ and a Banach space $X$.
The weak formulation of the Maxwell-Schrödinger system (\[eq:1-6\])- (\[eq:1-8\]) can be specified as follows: given $\mathbf{g} \in {L}^2(0,T;\mathbf{L}^2(\Omega))$, find $(\Psi,\mathbf{A})\in \mathcal{H}_{0}^{1}(\Omega)\times \mathbf{H}_{t}^{1}(\Omega)$ such that $\forall t\in(0,T)$, $$\label{eq:2-1}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle
-\mathrm{i}(\frac{\partial \Psi}{\partial t},\varphi)+
\frac{1}{2}(\left(\mathrm{i}\nabla +\mathbf{A}\right)\Psi,\left(\mathrm{i}\nabla +\mathbf{A}\right)\varphi)
+ (V_{0}\Psi,\varphi) = 0 ,\quad \forall \varphi\in\mathcal{H}_{0}^{1}(\Omega)}\\[2mm]
{\displaystyle ( \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}},\mathbf{v})+(\nabla\times
\mathbf{A},\nabla\times\mathbf{v}) + \gamma(\nabla\cdot
\mathbf{A},\nabla\cdot\mathbf{v})+(\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big),\mathbf{v}) }\\[2mm]
{\displaystyle\quad \quad \quad + \,( |\Psi|^{2}\mathbf{A},\mathbf{v})= ( \mathbf{g},\mathbf{v}) ,\qquad \qquad \qquad\qquad\qquad\forall\mathbf{v}\in\mathbf{H}^{1}_{t}(\Omega),}
\end{array}
\right.$$ with the initial conditions $\Psi_{0} \in \mathcal{H}^{1}_{0}(\Omega)$, $\mathbf{A}_{0} \in \mathbf{H}^{1}_{t}(\Omega)$ and $\mathbf{A}_{t}(\cdot , 0) \in \mathbf{L}^{2}(\Omega)$.
Let M be a positive integer and let $\Delta t =T / M$ be the time step. For any k=1,2,$\cdots, M$, we introduce the following notation: $$\label{eq:2-2}
\begin{array}{@{}l@{}}
{\displaystyle \partial U^{k}= (U^{k}-U^{k-1})/\Delta t, \quad\partial^{2} U^{k}=(\partial U^{k}-\partial U^{k-1})/\Delta t,}\\[2mm]
{\displaystyle \overline{U}^{k} = (U^{k}+U^{k-1})/2, \quad \widetilde{U}^{k}=(U^{k}+U^{k-2})/2,}\\[2mm]
\end{array}$$ for any given sequence $\{U^{k}\}_{0}^{M}$ and denote $u^{k}=u(\cdot,t^{k})$ for any given functions $u\in C(0,T;\,X)$ with a Banach space $ X $.
Let $\mathcal{T}_{h}=\{K\}$ be a regular partition of $\Omega$ into triangles in $\mathbb{R}^{2}$ or tetrahedrons in $\mathbb{R}^{3}$ without loss of generality, where the mesh size $h= \mathrm{max}_{K\in\mathcal{T}_{h}}\{diam(K)\}$. For any $K \in \mathcal{T}_{h}$, we denote by $P_{r}(K)$ the spaces of polynomials of degree $r \;(r \geq 1)$ defined on $K$. We now define the standard Lagrange finite element space $$Y^{r}_{h} = \{u_{h} \in C(\overline{\Omega}):\; u_{h}|_{K} \in P_{r}(K), \; \forall \; K \in \mathcal{T}_{h}\}.$$ We have the following finite element subspaces of $H_{0}^{1}(\Omega)$, $\mathcal{H}_{0}^{1}(\Omega)$ and $\mathbf{H}^{1}_{\rm t}(\Omega)$ $$V^{r}_{h} = Y^{r}_{h} \cap H_{0}^{1}(\Omega), \quad \mathcal{V}^{r}_{h} = V^{r}_{h} \oplus {\rm i}V^{r}_{h},\quad \mathbf{V}^{r}_{h} = \big( Y^{r}_{h}\big)^{3}\cap \mathbf{H}^{1}_{\rm t}(\Omega).$$
We shall approximate the wave function $\Psi$ and the vector potential $\mathbf{A}$ by the functions in $\mathcal{V}_{h}^{r}$ and $\mathbf{V}_{h}^{r}$, respectively. Let $I_h$ and $ {\bm \pi}_h $ be the conventional pointwise interpolation operators on $\mathcal{V}_{h}^{r}$ and $\mathbf{V}_{h}^{r}$, respectively. For $0\leq s \leq m\leq r+1$, $m\geq 2$, $2\leq p \leq \infty$, standard finite element theory gives that [@Bre]: $$\label{eq:2-3}
\begin{array}{@{}l@{}}
{\displaystyle \Vert u - I_h u \Vert_{\mathcal{W}^{s}_{p}} \leq Ch^{m-s}\Vert u \Vert_{\mathcal{W}^{m}_{p}} \quad \forall \; u \in \mathcal{W}^{m}_{p}(\Omega) ,} \\[2mm]
{\displaystyle \Vert \mathbf{v} - {\bm \pi}_{h} \mathbf{v} \Vert_{\mathbf{W}^{s}_{p}} \leq Ch^{m-s}\Vert \mathbf{v} \Vert_{\mathbf{W}^{m}_{p}} \quad \forall \; \mathbf{v} \in \mathbf{W}^{m}_{p}(\Omega).}
\end{array}$$
For convenience, assume that the function $\mathbf{A}$ is defined in the interval $[-\Delta t, T]$ in terms of the time variable $t$. We can compute $ \mathbf{A}(\cdot,-\Delta t) $ by $$\label{eq:2-4}
{\displaystyle \mathbf{A}(\cdot,-\Delta t)=\mathbf{A}(\cdot,0)-\Delta t \frac{\partial \mathbf{A}}{\partial t}(\cdot,0)=\mathbf{A}_{0}-\Delta t\mathbf{A}_{1}},$$ which leads to an approximation to $\mathbf{A}^{-1}$ with second order accuracy.
A Crank-Nicolson Galerkin finite element approximation to the Maxwell-Schrödinger system (\[eq:2-1\]) is formulated as follows: $$\label{eq:2-5}
{\displaystyle \Psi_{h}^{0}=I_{h}\Psi_{0},\quad\mathbf{A}_{h}^{0}={\bm \pi}_h \mathbf{A}_{0},\quad
\mathbf{A}_{h}^{0}-\mathbf{A}_{h}^{-1}=\Delta t {\bm \pi}_h\mathbf{A}_{1},}$$ and find $(\Psi_{h}^{k},\mathbf{A}_{h}^{k})\in\mathcal{V}_{h}^{r}\times\mathbf{V}^{r}_{h}$ such that for $k=1,2 ,\cdots, M$, $$\label{eq:2-6}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle
-\mathrm{i}(\partial\Psi_{h}^{k},\varphi)+
\frac{1}{2}\left((\mathrm{i}\nabla +\overline{\mathbf{A}}_{h}^{k})\overline{\Psi}_{h}^{k},(\mathrm{i}\nabla +\overline{\mathbf{A}}_{h}^{k})\varphi\right)
+ (V_{0}\overline{\Psi}_{h}^{k},\varphi) = 0 ,\quad \forall \varphi\in\mathcal{V}_{h}^{r}}\\[2mm]
{\displaystyle (\partial^{2}\mathbf{A}_{h}^{k},\mathbf{v}) +\left(\frac{\mathrm{i}}{2}\big((\Psi_{h}^{k-1})^{\ast}\nabla{\Psi_{h}^{k-1}}
-\Psi_h^{k-1}\nabla{(\Psi_{h}^{k-1})}^{\ast}\big),\mathbf{v}\right) + (\nabla\times
\widetilde{\mathbf{A}}_{h}^{k},\nabla\times\mathbf{v})}\\[2mm]
{\displaystyle\quad +\gamma(\nabla\cdot
\widetilde{\mathbf{A}}_{h}^{k},\nabla\cdot\mathbf{v})+\big(|\Psi_{h}^{k-1}|^{2}\frac{\overline{\mathbf{A}}_{h}^{k}
+\overline{\mathbf{A}}_{h}^{k-1}}{2},\mathbf{v}\big)=( \mathbf{g}^{k-1}, \mathbf{v}), \quad \forall\mathbf{v}\in\mathbf{V}^{r}_{h},}
\end{array}
\right.$$ where $ \overline{\mathbf{A}}_h^k $, $ \overline{\Psi}_h^k $ and $ \widetilde{\mathbf{A}}_h^k $ have been defined in (\[eq:2-2\]).
For convenience, we define the following bilinear forms: $$\label{eq:2-7}
\begin{array}{@{}l@{}}
{\displaystyle B(\mathbf{A};\Psi,\varphi)=\left((\mathrm{i}\nabla+\mathbf{A})\Psi,(\mathrm{i}\nabla+\mathbf{A})\varphi\right),}\\[2mm]
{\displaystyle D(\mathbf{A},\mathbf{v})=\gamma(\nabla\cdot \mathbf{A},\nabla\cdot \mathbf{v})+
(\nabla\times\mathbf{A},\nabla\times\mathbf{v}),}\\[2mm]
{\displaystyle f(\Psi,\varphi)=\frac{\mathrm{i}}{2}(\varphi^{\ast}\nabla\Psi-\Psi\nabla\varphi^{\ast}).}
\end{array}$$
Then the variational forms of the modified Maxwell-Schrödinger equations and the discrete system can be written as follows: $$\label{eq:2-8}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}(\frac{\partial \Psi}{\partial t},\varphi)+\frac{1}{2}B(\mathbf{A};\Psi,\varphi)+(V_{0}\Psi,\varphi)=0,\quad \forall \varphi\in\mathcal{H}_{0}^{1}(\Omega),}\\[2mm]
{\displaystyle (\frac{\partial ^{2}\mathbf{A}}{\partial t^{2}},\mathbf{v})
+D(\mathbf{A},\mathbf{v})+( f(\Psi,\Psi),\mathbf{v})+( |\Psi|^{2}\mathbf{A},\mathbf{v})=( \mathbf{g}, \mathbf{v}), \quad \forall\mathbf{v}\in\mathbf{H}^{1}_{t}(\Omega),}
\end{array}
\right.$$ and for $ k=1,2,\cdots,M$, $$\label{eq:2-9}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle
-\mathrm{i}(\partial\Psi_{h}^{k},\varphi)+
\frac{1}{2}B(\overline{\mathbf{A}}_{h}^{k};\overline{\Psi}_{h}^{k},\varphi)
+( V_{0}\overline{\Psi}_{h}^{k},\varphi) = 0 ,\quad \forall \varphi\in\mathcal{V}_{h}^{r}}\\[2mm]
{\displaystyle (\partial ^{2}\mathbf{A}_{h}^{k},\mathbf{v})+D(\widetilde{\mathbf{A}}_{h}^{k},\mathbf{v})
+\left( f(\Psi_{h}^{k-1},\Psi_h^{k-1}),\mathbf{v}\right)}\\[2mm]
{\displaystyle\quad+\big( |\Psi_{h}^{k-1}|^{2}\frac{(\overline{\mathbf{A}}_{h}^{k}+\overline{\mathbf{A}}_{h}^{k-1})}{2},\mathbf{v}\big)=( \mathbf{g}^{k-1}, \mathbf{v}), \quad
\forall\mathbf{v}\in\mathbf{V}^{r}_{h}.}
\end{array}
\right.$$
Note that after discretization in time and space, the Maxwell equation and Schrödinger equation in the discrete system (\[eq:2-9\]) are decoupled. At each time step, we only need to solve the two discrete linear equations alternately.
In this paper we assume that the modified Maxwell-Schrödinger equations (\[eq:2-8\]) has one and only one weak solution $(\Psi,\mathbf{A})$ and the following regularity conditions are satisfied: $$\label{eq:2-10}
\begin{array}{@{}l@{}}
{\displaystyle
\Psi,\Psi_{t},\Psi_{tt} \in {L}^{\infty}(0, T; \mathcal{H}^{r+1}(\Omega)),\quad \Psi_{ttt} \in {L}^{\infty}(0, T; \mathcal{H}^{1}(\Omega)),}\\[2mm]
{\displaystyle\Psi_{tttt} \in L^{2}(0, T; \mathcal{L}^{2}(\Omega)),\quad
\mathbf{A},\mathbf{A}_{t},\mathbf{A}_{tt} \in {L}^{\infty}(0, T; \mathbf{H}^{r+1}(\Omega)),}\\[2mm]
{\displaystyle\mathbf{A}_{ttt} \in {L}^{\infty}(0, T; \mathbf{H}^{1}(\Omega)),\mathbf{A}_{tttt} \in L^{2}(0, T; \mathbf{L}^{2}(\Omega)).}
\end{array}$$
For the initial conditions $(\Psi_{0},\mathbf{A}_{0},\mathbf{A}_{1})$ and the right hand function $\mathbf{g}(\mathbf{x},t)$, we assume that $$\label{eq:2-11}
\begin{array}{@{}l@{}}
{\displaystyle \Psi_{0}\in \mathcal{H}^{r+1}(\Omega) \cap \mathcal{H}_{0}^{1}(\Omega), \,\,\, \mathbf{A}_{0},\mathbf{A}_{1}\in\mathbf{H}^{r+1}(\Omega)\cap\mathbf{H}_{t}^{1}(\Omega),\,\,
\,\mathbf{g}\in C(0, T; \mathbf{L}^{2}(\Omega)).}
\end{array}$$
We now give the main convergence result in this paper as follows:
\[thm2-1\] Suppose that $ \Omega\subset \mathbb{R}^d $, $ d\geq 2 $ is a bounded Lipschitz polyhedral convex domain. Let $(\Psi,\mathbf{A})$ be the unique solution of the modified Maxwell-Schrödinger equations (\[eq:2-8\]), and let $ (\Psi_h^k,\mathbf{A}_h^k)$ be the numerical solution of the full discrete scheme (\[eq:2-9\]) associated with (\[eq:2-8\]). Under the assumptions (\[eq:2-10\]) and (\[eq:2-11\]), we have the following error estimates $$\label{eq:2-12}
\begin{array}{@{}l@{}}
{\displaystyle \max_{1\leq k \leq M}\big\{\|\Psi_{h}^{k}-\Psi^{k}\|_{\mathcal{L}^2(\Omega)}^{2}
+\|\nabla(\Psi_{h}^{k}-\Psi^{k})\|_{\mathbf{L}^2(\Omega)}^{2}
+ \|\mathbf{A}_{h}^{k}-\mathbf{A}^{k}\|_{\mathbf{L}^2(\Omega)}^{2}}\\[2mm]
{\displaystyle \quad+\|\nabla\cdot(\mathbf{A}_{h}^{k}-\mathbf{A}^{k})\|_{{L}^2(\Omega)}^{2}
+\|\nabla\times(\mathbf{A}_{h}^{k}-\mathbf{A}^{k})\|_{\mathbf{L}^2(\Omega)}^{2}\big\}
\leq C \big\{h^{2r}+(\Delta t)^{4}\big\},}
\end{array}$$ where $ \Psi^k=\Psi(\cdot, t^k) $, $ \mathbf{A}^k=\mathbf{A}(\cdot, t^k) $, and $C$ is a constant independent of $h$, $\Delta t$.
The proof of Theorem \[thm2-1\] will be given in section \[sec-4\].
Stability estimates {#sec-3}
====================
In this section we derive some stability estimates for the numerical solutions of the full discrete system (\[eq:2-9\]), which play an important role in the error estimates in the next section.
For convenience, we list the following imbedding inequalities and interpolation inequalities in Sobolev spaces (see, e.g., [@Lad] and [@Gir]), and use them in the sequel: $$\label{eq:3-1}
{\displaystyle \|u\|_{L^p} \leq C \|u\|_{{H}^1}, \quad \|\mathbf{v}\|_{\mathbf{L}^p}\leq C \|\mathbf{v}\|_{\mathbf{H}^1},
\quad 1\leq p \leq 6 \,\,(d=2,3),}$$ $$\label{eq:3-2}
{\displaystyle \|\mathbf{v}\|_{\mathbf{H}^1} \leq C(\|\nabla\times\mathbf{v}\|_{\mathbf{L}^2}+\|\nabla\cdot\mathbf{v}\|_{{L}^2}),\quad
\mathbf{v}\in \mathbf{H}^{1}_{t}(\Omega),}$$ $$\label{eq:3-3}
{\displaystyle \|u\|_{{L}^3}^{2}\leq \|u\|_{{L}^2}\|u\|_{{L}^6},}$$ where $ \|u\|_{L^p}=\|u\|_{L^p(\Omega)} $, $ \|u\|_{{H}^1}=\|u\|_{{H}^1(\Omega)} $, $ \|\mathbf{v}\|_{\mathbf{L}^2}=\|\mathbf{v}\|_{\mathbf{L}^2(\Omega)} $ and $\|\mathbf{v}\|_{\mathbf{H}^1}=\|\mathbf{v}\|_{\mathbf{H}^1(\Omega)} $.
We first give the definition of the discrete energy functional of (\[eq:2-9\]) as follows: $$\label{eq:3-8}
\begin{array}{@{}l@{}}
{\displaystyle \mathcal{G}_{h}^{k}=\frac{1}{2}B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k},\Psi_{h}^{k})+\frac{1}{4}D(\mathbf{A}_{h}^{k},\mathbf{A}_{h}^{k})
+\frac{1}{4}D(\mathbf{A}_{h}^{k-1},\mathbf{A}_{h}^{k-1})+V_0\|\Psi_{h}^{k}\|_{\mathcal{L}^2}^{2} }\\[2mm]
{\displaystyle \quad+\frac{1}{2} \|\partial \mathbf{A}_{h}^{k}\|_{\mathbf{L}^2}^{2}=\frac{1}{2}B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k},\Psi_{h}^{k})
+\frac{1}{2}\overline{D(\mathbf{A}_{h}^{k},\mathbf{A}_{h}^{k})}+V_0\|\Psi_{h}^{k}\|_{\mathcal{L}^2}^{2} + \frac{1}{2}\|\partial \mathbf{A}_{h}^{k}\|_{\mathbf{L}^2}^{2},}
\end{array}$$ where $ \overline{D(\mathbf{A}_{h}^{k},\mathbf{A}_{h}^{k})}=\frac{\displaystyle 1}{\displaystyle 2}\Big[
D(\mathbf{A}_{h}^{k},\mathbf{A}_{h}^{k})+D(\mathbf{A}_{h}^{k-1},\mathbf{A}_{h}^{k-1})\Big] $.
\[lem3-2\] For the solution of (\[eq:2-9\]), for $k=1,2\cdots,M $, we have $$\label{eq:3-9}
{\displaystyle {\|\Psi_{h}^{k}\|}_{\mathcal{L}^2}^{2}={\|\Psi_{h}^{0}\|}_{\mathcal{L}^2}^{2},\quad \mathcal{G}_{h}^{k}\leq C,}$$ where C is a constant independent of k, $h$ and $\Delta t $.
Choosing $\varphi=\overline{\Psi}_{h}^{k}$ in $(\ref{eq:2-9})_1 $ and taking the imaginary part, we can complete the proof of $(\ref{eq:3-9})_1 $. Let us turn to the proof of $(\ref{eq:3-9})_2 $. It is obvious that $$\begin{array}{@{}l@{}}
{\displaystyle \mathrm{Re}\left[B\left(\overline{\mathbf{A}}_{h}^{k};\overline{\Psi}_{h}^{k},\partial \Psi_{h}^{k}\right)\right]
=\frac{1}{2}\partial B(\overline{\mathbf{A}}_{h}^{k};{\Psi}_{h}^{k},\Psi_{h}^{k}) }\\[2mm]
{\displaystyle \quad\quad+ \frac{1}{2\Delta t}\left[B(\overline{\mathbf{A}}_{h}^{k-1};{\Psi}_{h}^{k-1},\Psi_{h}^{k-1})
-B(\overline{\mathbf{A}}_{h}^{k};{\Psi}_{h}^{k-1},\Psi_{h}^{k-1})\right]}\\[2mm]
{\displaystyle \quad\quad+ \frac{1}{2 \Delta t}\mathrm{Re}\left[B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k-1},\Psi_{h}^{k})
-B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k},\Psi_{h}^{k-1})\right].}
\end{array}$$
By a direct computation, we get $$\label{eq:3-10}
\begin{array}{@{}l@{}}
{\displaystyle B(\mathbf{A};\psi,\varphi)=(\nabla\psi,\nabla\varphi)+(\mathbf{A}\psi,\mathbf{A}\varphi)+2(f(\psi,\varphi),\mathbf{A}),}\\[2mm]
{\displaystyle B(\mathbf{A};\psi,\varphi)-B(\hat{\mathbf{A}};\psi,\varphi)=\left((\mathbf{A}
+\hat{\mathbf{A}})\psi \varphi^{*},\mathbf{A}-\hat{\mathbf{A}}\right)
+2(f(\psi,\varphi),\mathbf{A}-\hat{\mathbf{A}}),}
\end{array}$$ and consequently $${\displaystyle\mathrm{Re}\left[B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k-1},\Psi_{h}^{k})-B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k},\Psi_{h}^{k-1})\right] = 0.}$$
We thus have $$\label{eq:3-11}
\begin{array}{@{}l@{}}
{\displaystyle \mathrm{Re}\left[B\left(\overline{\mathbf{A}}_{h}^{k};\overline{\Psi}_{h}^{k},\partial \Psi_{h}^{k}\right)\right]
=-\left(|\Psi_{h}^{k-1}|^{2}\frac{\overline{\mathbf{A}}_{h}^{k}+\overline{\mathbf{A}}_{h}^{k-1}}{2},\frac{\overline{\mathbf{A}}_{h}^{k}
-\overline{\mathbf{A}}_{h}^{k-1}}{\Delta t}\right)
}\\[2mm]
{\displaystyle \quad+\frac{1}{2}\partial B(\overline{\mathbf{A}}_{h}^{k};{\Psi}_{h}^{k},\Psi_{h}^{k})-\left(f(\Psi_{h}^{k-1},\Psi_{h}^{k-1}),\frac{\overline{\mathbf{A}}_{h}^{k}
-\overline{\mathbf{A}}_{h}^{k-1}}{\Delta t}\right).}
\end{array}$$
It is not difficult to check that $$\label{eq:3-12}
{\displaystyle \mathrm{Re}\left[V_0(\overline{\Psi}_{h}^{k},\partial \Psi_{h}^{k})\right]=\frac{V_{0}}{2}\partial(\Psi_{h}^{k},\Psi_{h}^{k}).}$$
We choose $\varphi=\partial\Psi_{h}^{k}$ in $ (\ref{eq:2-9})_1 $ and take the real part. Combining (\[eq:3-11\]) and (\[eq:3-12\]) gives $$\label{eq:3-13}
\begin{array}{@{}l@{}}
{\displaystyle \frac{1}{2}\partial\|\left(\mathrm{i}\nabla+\overline{\mathbf{A}}_{h}^{k}\right)\Psi_{h}^{k}\|_{\mathbf{L}^2}^{2}+V_{0}\partial \|\Psi_{h}^{k}\|_{\mathcal{L}^2}^{2} -\left(|\Psi_{h}^{k-1}|^{2}\frac{\overline{\mathbf{A}}_{h}^{k}+\overline{\mathbf{A}}_{h}^{k-1}}{2},\frac{\overline{\mathbf{A}}_{h}^{k}
-\overline{\mathbf{A}}_{h}^{k-1}}{\Delta t}\right) }\\[2mm]
{\displaystyle \quad -\left(f(\Psi_{h}^{k-1},\Psi_{h}^{k-1}),\frac{\overline{\mathbf{A}}_{h}^{k}-\overline{\mathbf{A}}_{h}^{k-1}}{\Delta t}\right)=0.}
\end{array}$$
Taking $\mathbf{v}=\frac{\displaystyle 1}{\displaystyle 2\Delta t}(\mathbf{A}_{h}^{k}-\mathbf{A}_{h}^{k-2})=\frac{\displaystyle 1}
{\displaystyle \Delta t}
(\overline{\mathbf{A}}_{h}^{k}-\overline{\mathbf{A}}_{h}^{k-1})=\frac{\displaystyle 1}{\displaystyle 2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})$ in $ (\ref{eq:2-9})_2 $, and combining with (\[eq:3-13\]), we get $$\begin{array}{@{}l@{}}
{\displaystyle \partial \left(\frac{1}{2} B(\overline{\mathbf{A}}_{h}^{k};\Psi_{h}^{k},\Psi_{h}^{k})+V_0\|\Psi_{h}^{k}\|_{\mathcal{L}^2}^{2}
+\frac{1}{2}\|\partial \mathbf{A}_{h}^{k}\|_{\mathbf{L}^2}^{2}\right)}\\[2mm]
{\displaystyle \quad+\partial\left(\frac{1}{4}D(\mathbf{A}_{h}^{k},\mathbf{A}_{h}^{k})
+\frac{1}{4}D(\mathbf{A}_{h}^{k-1},\mathbf{A}_{h}^{k-1})\right)=(\mathbf{g}^{k-1},\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})) .}
\end{array}$$ It follows that $$\label{eq:3-13-0}
\begin{array}{@{}l@{}}
{\displaystyle \partial \mathcal{G}_{h}^{k} \leq \Vert \mathbf{g}^{k-1} \Vert_{\mathbf{L}^{2}} \Vert
\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\Vert_{\mathbf{L}^{2}} }\\[2mm]
{\displaystyle \qquad \leq C\left(\Vert \mathbf{g}^{k-1} \Vert_{\mathbf{L}^{2}}^{2} + \Vert
\partial \mathbf{A}_{h}^{k}\Vert_{\mathbf{L}^{2}}^{2} + \Vert
\partial \mathbf{A}_{h}^{k-1}\Vert_{\mathbf{L}^{2}}^{2}\right).}
\end{array}$$ Multiply (\[eq:3-13-0\]) by $\Delta t$, sum $k=1,2,\cdots,M$, to discover $$\begin{array}{@{}l@{}}
{\displaystyle \mathcal{G}_{h}^{M} \leq \mathcal{G}_{h}^{0} + C\Delta t \sum_{k=0}^{M-1} \Vert \mathbf{g}^{k-1} \Vert_{\mathbf{L}^{2}}^{2} + C \Delta t \sum_{k=0}^{M}\mathcal{G}_{h}^{k}}.
\end{array}$$
Now $(\ref{eq:3-9})_2 $ follows from the discrete Gronwall’s inequality and thus we complete the proof of Lemma \[lem3-2\].
\[rem3-2\] Lemma \[lem3-2\] shows that the numerical scheme presented in this paper for the modified Maxwell-Schrödinger equations (\[eq:2-8\]) is stable in some senses.
\[thm3-1\] The solution of the full discrete system (\[eq:2-9\]) fulfills the following estimates $$\label{eq:3-14}
\begin{array}{@{}l@{}}
{\displaystyle\|\left(\mathrm{i}\nabla+\overline{\mathbf{A}}_{h}^{k}\right)\Psi_{h}^{k}\|_{\mathbf{L}^2}
+\|\Psi_{h}^{k}\|_{\mathcal{L}^2}+\|\mathbf{A}_{h}^{k}\|_{\mathbf{L}^2}+\|\partial \mathbf{A}_{h}^{k}\|_{\mathbf{L}^2}}\\[2mm]
{\displaystyle\quad +\|\nabla \times\mathbf{A}_{h}^{k}\|_{\mathbf{L}^2}
+\gamma\|\nabla \cdot\mathbf{A}_{h}^{k}\|_{{L}^2}\leq C,}
\end{array}$$ and $$\label{eq:3-15}
{\displaystyle\|\mathbf{A}_{h}^{k}\|_{\mathbf{H}^1}
+\|\mathbf{A}_{h}^{k}\|_{\mathbf{L}^6}
+\|\Psi_{h}^{k}\|_{\mathcal{H}^1}+\|\Psi_{h}^{k}\|_{\mathcal{L}^6}\leq C,}$$ where $ C $ is a constant independent of $ h $, $ \Delta t $.
(\[eq:3-14\]) is the direct result of Lemma \[lem3-2\]. Next we give the proof of (\[eq:3-15\]). Since the semi-norm in $\mathbf{H}_t^{1}(\Omega)$ is equivalent to $\mathbf{H}^{1}$-norm, from (\[eq:3-14\]) we get $$\label{eq:3-16}
\Vert \mathbf{A}^{k}_h\|_{\mathbf{H}^1} < C.$$ Then Sobolev’s imbedding theorem implies that $$\label{eq:3-17}
\Vert \mathbf{A}^{k}_h\|_{\mathbf{L}^p} < C ,$$ with $ \, 1\leq p \leq 6$ for $d = 3$ and $ 1\leq p < \infty $ for $ d=2$.
Using Young’s inequality and the interpolation inequalities (\[eq:3-3\]), we further prove $$\begin{array}{@{}l@{}}
{\displaystyle \|{\mathbf{A}}^{k}_{h}\Psi^{k}_{h}\|_{\mathbf{L}^2}\leq \|{\mathbf{A}}^{k}_{h}\|_{\mathbf{L}^6} \|\Psi^{k}_{h}\|_{\mathcal{L}^3} \leq C \|\Psi^{k}_{h}\|_{\mathcal{L}^3}
}\\[2mm]
{\displaystyle \quad\leq C\|\Psi^{k}_{h}\|_{\mathcal{L}^2}^{\frac{1}{2}}\|\Psi^{k}_{h}\|_{\mathcal{L}^6}^{\frac{1}{2}} \leq C\|\Psi^{k}_{h}\|_{\mathcal{L}^2}^{\frac{1}{2}}\|\nabla\Psi^{k}_{h}\|_{\mathbf{L}^2}^{\frac{1}{2}}
\leq C+\frac{1}{2}\|\nabla\Psi^{k}_{h}\|_{\mathbf{L}^2}}.
\end{array}$$
Hence we have $$\begin{array}{@{}l@{}}
{\displaystyle \|\nabla\Psi^{k}_{h}\|_{\mathbf{L}^2}\leq \|\left(\mathrm{i}\nabla
+{\mathbf{A}}^{k}_{h}\right)\Psi^{k}_{h}\|_{\mathbf{L}^2}
+ \|{\mathbf{A}}^{k}_{h}\Psi^{k}_{h}\|_{\mathbf{L}^2}
\leq C+\frac{1}{2}\|\nabla\Psi^{k}_{h}\|_{\mathbf{L}^2}.}
\end{array}$$ Consequently, we obtain $$\label{eq:3-18}
\|\Psi^{k}_{h}\|_{\mathcal{H}_{0}^1} + \|\Psi^{k}_{h}\|_{\mathcal{L}^6} \leq C.$$ Combining (\[eq:3-16\]), (\[eq:3-17\]) and (\[eq:3-18\]), we complete the proof of (\[eq:3-15\]). $\qquad$
The error estimates {#sec-4}
===================
In this section, we will give the proof of Theorem \[thm2-1\]. Let $(I_{h}\Psi, {\bm \pi}_{h}\mathbf{A})$ denote the interpolation functions of $(\Psi, \mathbf{A})$ in $\mathcal{V}_{h}^{r}\times\mathbf{V}_{h}^{r}$. Set $e_{\Psi}=I_{h}\Psi-\Psi$, $e_{\mathbf{A}}={\bm \pi}_h\mathbf{A}-\mathbf{A}$. By applying the interpolation error estimates (\[eq:2-3\]) and the regularity assumptions (\[eq:2-10\]), we have $$\label{eq:4-1}
\begin{array}{@{}l@{}}
{\displaystyle \|e_{\Psi}\|_{\mathcal{L}^2}+h\|e_{\Psi}\|_{\mathcal{H}^1}\leq Ch^{r+1},\quad \|e_{\mathbf{A}}\|_{\mathbf{L}^2}
+h\|e_{\mathbf{A}}\|_{\mathbf{H}^1}\leq C h^{r+1},}\\[2mm]
{\displaystyle
\|I_{h}\Psi\|_{\mathcal{L}^{\infty}}+\|{\bm \pi}_h\mathbf{A}\|_{\mathbf{H}^{1}}
+\|\nabla I_{h}\Psi \|_{\mathbf{L}^{3}}\leq C,}
\end{array}$$ where $ C $ is a constant independent of $ h $.
For convenience, we give the following identities, which will be used frequently in the sequel. $$\label{eq:4-2}
\begin{array}{@{}l@{}}
{\displaystyle \sum_{k=1}^{M}{(a_{k}-a_{k-1})b_{k}}=a_{M}b_{M}-a_{0}b_{1}-\sum_{k=1}^{M-1}{a_{k}(b_{k+1}-b_{k})},}\\[2mm]
{\displaystyle \sum_{k=1}^{M}{(a_{k}-a_{k-1})b_{k}}=a_{M}b_{M}-a_{0}b_{0}-\sum_{k=1}^{M}{a_{k-1}(b_{k}-b_{k-1})}.}
\end{array}$$
Let $\theta_{\Psi}^{k}=\Psi_{h}^{k}-I_{h}\Psi^{k} $, $\theta_{\mathbf{A}}^{k}=\mathbf{A}_{h}^{k}-{\bm \pi}_{h}\mathbf{A}^{k}$. By using the error estimates of the interpolation operators, we only need to estimate $\theta_{\Psi}^{k}$ and $\theta_{\mathbf{A}}^{k}$. Subtracting (\[eq:2-8\]) from (\[eq:2-9\]), we get the following equations for $\theta_{\Psi}^{k}$ and $\theta_{\mathbf{A}}^{k}$: $$\label{eq:4-3}
\begin{array}{@{}l@{}}
{\displaystyle -2\mathrm{i}(\partial \theta^{k}_{\Psi},\varphi)+B\left(\overline{\mathbf{A}}^{k}_{h};\overline{\theta}_{\Psi}^{k},\varphi\right)
= 2\mathrm{i}\left(\partial I_{h}\Psi^{k} - (\Psi_{t})^{k-\frac{1}{2}},\varphi\right)}\\[2mm]
{\displaystyle \quad+2V_0\left(\Psi^{k-\frac{1}{2}}-\overline{\Psi}_{h}^{k},\varphi\right)+ B(\mathbf{A}^{k-\frac{1}{2}};(\Psi^{k-\frac{1}{2}}-I_{h}\overline{\Psi}^{k}),\varphi)}\\[2mm]
{\displaystyle \quad+\left(B(\mathbf{A}^{k-\frac{1}{2}};I_{h}\overline{\Psi}^{k},\varphi)-B(\overline{\mathbf{A}}^{k}_{h}; I_{h}\overline{\Psi}^{k},\varphi)\right),\quad \forall \varphi\in\mathcal{V}_{h}^{r},}
\end{array}$$ and $$\label{eq:4-4}
\begin{array}{@{}l@{}}
{\displaystyle \left(\partial^{2}\theta^{k}_{\mathbf{A}},\mathbf{v}\right)+D(\widetilde{\theta^{k}_{\mathbf{A}}},\mathbf{v})
= \left((\mathbf{A}_{tt})^{k-1}-\partial^{2} {\bm \pi}_{h}\mathbf{A}^{k},\mathbf{v}\right)}\\[2mm]
{\displaystyle \quad +D(\mathbf{A}^{k-1}-\widetilde{{\bm \pi}_{h}\mathbf{A}^{k}},\mathbf{v})
+ \left(|\Psi^{k-1}|^{2}\mathbf{A}^{k-1}-|\Psi^{k-1}_{h}|^{2}\frac{\overline{\mathbf{A}}^{k}_{h}+\overline{\mathbf{A}}^{k-1}_{h}}{2},
\;\mathbf{v}\right)}\\[2mm]
{\displaystyle \quad+ \left(f(\Psi^{k-1},\Psi^{k-1})-f(\Psi^{k-1}_{h},\Psi^{k-1}_{h}),\;\mathbf{v}\right),\quad \forall\mathbf{v}\in\mathbf{V}^{r}_{h},}
\end{array}$$ where $ \overline{\mathbf{A}}_h^k $, $ \overline{\Psi}_h^k $, $ \widetilde{\theta^{k}_{\mathbf{A}}}$ and $ \widetilde{{\bm \pi}_{h}\mathbf{A}^{k}} $ are similarly given in (\[eq:2-2\]).
Now we briefly describe the outline of the proof of (\[eq:2-12\]). First, we take $\varphi=\overline{\theta}_{\Psi}^{k}$ in (\[eq:4-3\]) and obtain the estimate of $ \|\theta_\Psi^M\|_{\mathcal{L}^2} $. Second, we choose $\varphi=\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}$ in (\[eq:4-3\]) and derive the energy-norm estimate for $\theta^{M}_{\Psi}$. Finally, let $\mathbf{v}
=\frac{\displaystyle 1}{\displaystyle 2\Delta t}(\theta_{\mathbf{A}}^{k}-\theta_{\mathbf{A}}^{k-2})$ in (\[eq:4-4\]) and acquire the estimate involving $\theta_{\mathbf{A}}^k $. Combining the above three estimates, we will complete the proof of (\[eq:2-12\]).
Estimates for (\[eq:4-3\])
--------------------------
To begin with, choosing $\varphi=\overline{\theta}_{\Psi}^{k}$, $ k=1,2,\cdots, M $ as the test function in (\[eq:4-3\]), we get $$\label{eq:4-5}
{\displaystyle 2\mathrm{i}(\partial \theta^{k}_{\Psi},\overline{\theta}_{\Psi}^{k})-B\left(\overline{\mathbf{A}}^{k}_{h};\overline{\theta}_{\Psi}^{k},\overline{\theta}_{\Psi}^{k}\right)
=I_1^{(k)}+I_2^{(k)}+I_3^{(k)}+I_4^{(k)},}$$ where $$\begin{array}{@{}l@{}}
{\displaystyle I_1^{(k)}=2\mathrm{i}\left((\Psi_t)^{k-\frac{1}{2}}-\partial I_{h}\Psi^{k},\overline{\theta}_{\Psi}^{k}\right),\quad I_2^{(k)}=2V_0\left(\overline{\Psi}_{h}^{k}-\Psi^{k-\frac{1}{2}},\overline{\theta}_{\Psi}^{k}\right),}\\[2mm]
{\displaystyle I_3^{(k)}=B(\mathbf{A}^{k-\frac{1}{2}};(I_{h}\overline{\Psi}^{k}-\Psi^{k-\frac{1}{2}}),\overline{\theta}_{\Psi}^{k}),
\quad I_4^{(k)}=B(\overline{\mathbf{A}}^{k}_{h};I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})
-B(\mathbf{A}^{k-\frac{1}{2}};I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k}).}
\end{array}$$
Using the error estimates (\[eq:4-1\]) for the interpolation operator $I_{h}$ and the regularity of $\Psi$ in (\[eq:2-10\]), it is easy to see that $$\label{eq:4-6}
\begin{array}{@{}l@{}}
{\displaystyle |I_1^{(k)}|+ |I_2^{(k)}|\leq C\left((\Delta t)^{4}+h^{2r+2}\right)+C\left(\|\theta_{\Psi}^{k}\|^{2}_{\mathcal{L}^2}+\|\theta_{\Psi}^{k-1}\|^{2}_{\mathcal{L}^2}\right).}
\end{array}$$
We observe that $$\label{eq:4-7}
\begin{array}{@{}l@{}}
{\displaystyle B(\mathbf{A};\psi,\varphi)=(\nabla\psi,\nabla\varphi)+\left(|\mathbf{A}|^{2}\psi,\varphi\right)
+i\left(\varphi^{\ast}\nabla\psi-\psi\nabla\varphi^{\ast},\mathbf{A}\right)}\\[2mm]
{\displaystyle \quad \leq \|\nabla\psi\|_{\mathbf{L}^2}\|\nabla\varphi\|_{\mathbf{L}^2}+\|\mathbf{A}\|^{2}_{\mathbf{L}^6}\|\psi\|_{\mathcal{L}^6}\|\varphi\|_{\mathcal{L}^2}
+\|\mathbf{A}\|_{\mathbf{L}^6}\big(\|\psi\|_{\mathcal{L}^3}\|\nabla\varphi\|_{\mathbf{L}^2}}\\[2mm]
{\displaystyle \quad+\|\nabla\psi\|_{\mathbf{L}^2}\|\varphi\|_{\mathcal{L}^3}\big)\leq C\|\nabla\psi\|_{\mathbf{L}^2}
\|\nabla\varphi\|_{\mathbf{L}^2},\quad \forall \mathbf{A}\in \mathbf{L}^6(\Omega),
\,\,\,\psi,\varphi\in\mathcal{H}_{0}^{1}(\Omega),}
\end{array}$$ and $$\label{eq:4-8}
\begin{array}{@{}l@{}}
{\displaystyle I_3^{(k)}= B(\mathbf{A}^{k-\frac{1}{2}};(I_{h}\overline{\Psi}^{k}-\overline{\Psi}^{k}),\overline{\theta}_{\Psi}^{k})
+B(\mathbf{A}^{k-\frac{1}{2}};(\overline{\Psi}^{k}-\Psi^{k-\frac{1}{2}}),\overline{\theta}_{\Psi}^{k}).}
\end{array}$$
It follows from (\[eq:4-1\]), (\[eq:4-7\]) and (\[eq:4-8\]) that $$\label{eq:4-9}
{\displaystyle |I_3^{(k)}|\leq C\left(h^{2r}+(\Delta t)^{4}\right)+C\left(\|\nabla\theta_{\Psi}^{k}\|^{2}_{\mathbf{L}^2}
+\|\nabla\theta_{\Psi}^{k-1}\|^{2}_{\mathbf{L}^2}\right).}$$
Notice that $$\label{eq:4-10}
\begin{array}{@{}l@{}}
{\displaystyle I_4^{(k)}= \Big[B(\overline{\mathbf{A}}^{k}_{h}; I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})
-B({\bm\pi}_{h}\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})\Big]+\Big[B({\bm \pi}_{h}\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})}\\[2mm]
{\displaystyle \quad
-B(\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})\Big]+\Big[B(\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})
-B(\mathbf{A}^{k-\frac{1}{2}}; I_{h}\overline{\Psi}^{k},\overline{\theta}_{\Psi}^{k})\Big]}\\[2mm]
\end{array}$$
By using (\[eq:3-1\])-(\[eq:3-3\]) and (\[eq:3-10\]), we prove $$\label{eq:4-13}
\begin{array}{@{}l@{}}
{\displaystyle |I_4^{(k)}|\leq C\left(D({\theta}^{k}_{\mathbf{A}},{\theta}^{k}_{\mathbf{A}})
+D({\theta}^{k-1}_{\mathbf{A}},{\theta}^{k-1}_{\mathbf{A}})\right)}\\[2mm]
{\displaystyle\qquad\quad+C\big\{h^{2r}+(\Delta t)^{4}+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}+\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}^{2}\big\}.}
\end{array}$$
Taking the imaginary part of (\[eq:4-5\]), we have $$\label{eq:4-14}
\begin{array}{@{}l@{}}
{\displaystyle \frac{1}{\Delta t}\big(\|\theta^{k}_{\Psi}\|^{2}_{\mathcal{L}^2}
-\|\theta^{k-1}_{\Psi}\|^{2}_{\mathcal{L}^2}\big)
=\mathrm{Im}(I_1^{(k)})+\mathrm{Im}(I_2^{(k)})+\mathrm{Im}(I_3^{(k)})+\mathrm{Im}(I_4^{(k)})}\\[2mm]
{\displaystyle\qquad\quad\leq |I_1^{(k)}|+|I_2^{(k)}|+|I_3^{(k)}|+|I_4^{(k)}|
\leq C \left(D ({\theta}^{k}_{\mathbf{A}},{\theta}^{k}_{\mathbf{A}})
+D({\theta}^{k-1}_{\mathbf{A}},{\theta}^{k-1}_{\mathbf{A}})\right)}\\[2mm]
{\displaystyle\qquad\quad+C\big\{h^{2r}+(\Delta t)^{4}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}^{2}\big\},}
\end{array}$$ and therefore $$\label{eq:4-15}
\begin{array}{@{}l@{}}
{\displaystyle \|\theta^{M}_{\Psi}\|^{2}_{\mathcal{L}^2}\leq \|\theta^{0}_{\Psi}\|^{2}_{\mathcal{L}^2}
+C\big(h^{2r}+(\Delta t)^{4}\big)+C \Delta t
\sum_{k=1}^{M}{D({\theta}^{k}_{\mathbf{A}},{\theta}^{k}_{\mathbf{A}})}
+C\Delta t\sum_{k=1}^{M} {\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}}}\\[2mm]
{\displaystyle \quad \leq C\big(h^{2r}+(\Delta t)^{4}\big)
+C\Delta t\sum_{k=1}^{M}{D({\theta}^{k}_{\mathbf{A}},
{\theta}^{k}_{\mathbf{A}})}+C\Delta t\sum_{k=1}^{M} {\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}}.}
\end{array}$$ Here we have used the fact $\|\theta^{0}_{\Psi}\|^{2}_{\mathcal{L}^2} \leq C h^{2r+2} $.
To proceed further, we take $\varphi=\Delta t\partial{\theta_{\Psi}^{k}}=\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}$, $ k=1,2,\cdots, M $ in (\[eq:4-3\]), to find $$\label{eq:4-16}
{\displaystyle -2\mathrm{i}\Delta t(\partial \theta^{k}_{\Psi},\partial{\theta_{\Psi}^{k}})+\Delta t B\left(\overline{\mathbf{A}}^{k}_{h};\overline{\theta}_{\Psi}^{k},\partial{\theta_{\Psi}^{k}}\right)
=J_1^{(k)}+J_2^{(k)}+J_3^{(k)}+J_4^{(k)},}$$ where $$\begin{array}{@{}l@{}}
{\displaystyle J_1^{(k)}=2\mathrm{i}\left(\partial I_{h}\Psi^{k}-(\Psi_{t})^{k-\frac{1}{2}},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right),\quad J_2^{(k)}=2V_0\left(\Psi^{k-\frac{1}{2}}-\overline{\Psi}_{h}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right),}\\[2mm]
{\displaystyle J_3^{(k)}=B(\mathbf{A}^{k-\frac{1}{2}};(\Psi^{k-\frac{1}{2}}-I_{h}\overline{\Psi}^{k}),\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}),}\\[2mm]
{\displaystyle J_4^{(k)}=B(\mathbf{A}^{k-\frac{1}{2}}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})-B(\overline{\mathbf{A}}^{k}_{h}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}).}
\end{array}$$
By virtue of (\[eq:4-2\]), we get $$\label{eq:4-17}
\begin{array}{@{}l@{}}
{\displaystyle \sum_{k=1}^{M} J_1^{(k)}=2\mathrm{i}\sum_{k=1}^{M}\left(\partial I_{h}\Psi^{k}-(\Psi_{t})^{k-\frac{1}{2}},\theta_{\Psi}^{k}-{\theta_{\Psi}^{k-1}}\right)}\\[2mm]
{\displaystyle \quad= 2\mathrm{i}\left(\partial I_{h}\Psi^{M}-(\Psi_{t})^{M-\frac{1}{2}},\theta_{\Psi}^{M}\right)-2\mathrm{i}\left(\partial I_{h}\Psi^{1}
-(\Psi_{t})^{\frac{1}{2}},\theta_{\Psi}^{0}\right)}\\[2mm]
{\displaystyle \quad-2\mathrm{i}\sum_{k=1}^{M-1}\left(\partial I_{h}\Psi^{k+1}
-\partial I_{h}\Psi^{k}-(\Psi_{t})^{k+\frac{1}{2}}
+(\Psi_{t})^{k-\frac{1}{2}},\,\theta_{\Psi}^{k}\right).}
\end{array}$$
It follows from (\[eq:4-1\]) and (\[eq:4-17\]) that $$\label{eq:4-19}
\begin{array}{@{}l@{}}
{\displaystyle |\sum_{k=1}^{M} J_1^{(k)}|\leq C\big(h^{2r+2}+(\Delta t)^{4}\big)
+ C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+C\Delta t \sum_{k=1}^{M-1}{\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}}.}
\end{array}$$
To estimate the term $ J_2^{(k)}$, we rewrite it as $$\begin{array}{@{}l@{}}
{\displaystyle 2 V_0\left(\Psi^{k-\frac{1}{2}}-\overline{\Psi}_{h}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right) = 2V_0\left(\Psi^{k-\frac{1}{2}}
-I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)}\\[2mm]
{\displaystyle \quad-2V_0\left(\frac{1}{2}(\theta_{\Psi}^{k}+\theta_{\Psi}^{k-1}),\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)
\stackrel{\mathrm{def}}{=}J_2^{(k),1}+J_2^{(k),2}.}
\end{array}$$
By applying a standard argument, we find that $$\label{eq:4-22}
\begin{array}{@{}l@{}}
{\displaystyle |\mathrm{Re}\big(\sum_{k=1}^{M} J_2^{(k)}\big)|\leq |\sum_{k=1}^{M}J_2^{(k),1}| +|\mathrm{Re}\big(\sum_{k=1}^{M} J_2^{(k),2}\big)|}\\[2mm]
{\displaystyle \quad \leq C\big(h^{2r+2}+(\Delta t)^{4}\big)
+C \|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+ C\Delta t \sum_{k=1}^{M-1}{\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}}.}
\end{array}$$
We recall (\[eq:3-10\]) and rewrite $ J_3^{(k)} $ as follows: $$\label{eq:4-23}
\begin{array}{@{}l@{}}
{\displaystyle J_3^{(k)}= \left(\nabla( \Psi^{k-\frac{1}{2}}-I_{h}\overline{\Psi}^{k}),\;\nabla
(\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})\right)}\\[2mm]
{\displaystyle\quad \qquad +\left(|\mathbf{A}^{k-\frac{1}{2}}|^2(\Psi^{k-\frac{1}{2}}-I_{h}\overline{\Psi}^{k}),\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)}\\[2mm]
{\displaystyle \quad \qquad+ \mathrm{i}\left(\nabla(\Psi^{k-\frac{1}{2}}-I_{h}\overline{\Psi}^{k})\mathbf{A}^{k-\frac{1}{2}},\;\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)}\\[2mm]
{\displaystyle \quad \qquad-\mathrm{i}\left((\Psi^{k-\frac{1}{2}}-I_{h}\overline{\Psi}^{k})\mathbf{A}^{k-\frac{1}{2}},\;\nabla \theta_{\Psi}^{k}
-\nabla \theta_{\Psi}^{k-1}\right).}\\[2mm]
\end{array}$$
By employing (\[eq:4-1\]), (\[eq:4-2\]), the regularity assumption (\[eq:2-10\]) and Young’s inequality, we can prove the following estimate of $ \sum_{k=1}^{M}{J_3^{(k)}} $ . $$\label{eq:4-23-0}
{\displaystyle |\sum_{k=1}^{M}{J_3^{(k)}}| \leq C(h^{2r}+(\Delta t)^{4}) +C \Vert\theta_{\Psi}^{M}\Vert_{\mathcal{L}^2}^{2}
+ \frac{1}{16} \Vert\nabla\theta_{\Psi}^{M}\Vert_{\mathbf{L}^2}^{2}+C\Delta t \sum_{k=1}^{M-1}{\Vert\nabla\theta_{\Psi}^{k}\Vert_{\mathbf{L}^2}^{2}}.}$$
The proof is standard but tedious. Due to space limitations, we omit it here.
In order to estimate $\sum\limits_{k=1}^{M} J_4^{(k)}$, we rewrite $ J_4^{(k)}$ in the following form: $$\label{eq:4-30}
\begin{array}{@{}l@{}}
{\displaystyle J_4^{(k)}= \Big[B(\mathbf{A}^{k-\frac{1}{2}}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})
-B(\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})\Big]}\\[2mm]
{\displaystyle \quad\quad +\Big[B(\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})-B({\bm \pi}_{h}\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})\Big]}\\[2mm]
{\displaystyle \quad\quad +\Big[B({\bm \pi}_{h}\overline{\mathbf{A}}^{k}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})-B(\overline{\mathbf{A}}^{k}_{h}; I_{h}\overline{\Psi}^{k},\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1})\Big]}\\[2mm]
{\displaystyle\quad\quad \stackrel{\mathrm{def}}{=}J_4^{(k),1}+J_4^{(k),2}+J_4^{(k),3}.}
\end{array}$$
By applying (\[eq:3-15\]) and (\[eq:4-2\]), we deduce $$\label{eq:4-34}
\begin{array}{@{}l@{}}
{\displaystyle |\sum_{k=1}^{M} J_4^{(k),1}| + |\sum_{k=1}^{M} J_4^{(k),2}| \leq C\big(h^{2r}+(\Delta t)^{4}\big)
+C \|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}}\\[2mm]
{\displaystyle \quad+ \frac{1}{16} \|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}+C\Delta t \sum_{k=1}^{M-1}{(\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2})}.}
\end{array}$$
In order to estimate $|\sum\limits_{k=1}^{M} J_4^{(k),3}|$, we rewrite it as follows. $$\label{eq:4-35}
\begin{array}{@{}l@{}}
{\displaystyle \sum_{k=1}^{M}J_4^{(k),3}=\sum_{k=1}^{M}{ \left(I_{h}\overline{\Psi}^{k}({\bm \pi}_{h}\overline{\mathbf{A}}^{k}+\overline{\mathbf{A}}^{k}_{h})({\bm\pi}_{h}\overline{\mathbf{A}}^{k}
-\overline{\mathbf{A}}^{k}_{h}),\;\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)}}\\[2mm]
{\displaystyle\quad\quad\quad\quad-\sum_{k=1}^{M}{\mathrm{i}\left( I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}-\overline{\mathbf{A}}^{k}_{h}),\;
\nabla\theta_{\Psi}^{k}-\nabla\theta_{\Psi}^{k-1}\right)}}\\[2mm]
{\displaystyle \quad\quad\quad\quad+\sum_{k=1}^{M}{\mathrm{i}\left(\nabla I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}
-\overline{\mathbf{A}}^{k}_{h}),\;\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)}}\\[2mm]
{\displaystyle\quad\quad\quad \stackrel{\mathrm{def}}{=}Q_1+Q_2+Q_3.}
\end{array}$$
Note that $$\label{eq:4-36}
\begin{array}{@{}l@{}}
{\displaystyle Q_1= \sum_{k=1}^{M}\left(I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}+\overline{\mathbf{A}}^{k}_{h})
({\bm\pi}_{h}\overline{\mathbf{A}}^{k}-\overline{\mathbf{A}}^{k}_{h}),\;\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)}\\[2mm]
{\displaystyle \quad = -\left(I_{h}\overline{\Psi}^{M}({\bm\pi}_{h}\overline{\mathbf{A}}^{M}
+\overline{\mathbf{A}}^{M}_{h})
\overline{\theta}_{\mathbf{A}}^{M},\;\theta_{\Psi}^{M}\right) + \left(I_{h}\overline{\Psi}^{0}({\bm\pi}_{h}\overline{\mathbf{A}}^{0}+\overline{\mathbf{A}}^{0}_{h})
\overline{\theta}_{\mathbf{A}}^{0},\;\theta_{\Psi}^{0}\right)}\\[2mm]
{\displaystyle \quad + \sum_{k=1}^{M}\left(I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}
+\overline{\mathbf{A}}^{k}_{h})
\overline{\theta}_{\mathbf{A}}^{k}-I_{h}\overline{\Psi}^{k-1}({\bm\pi}_{h}\overline{\mathbf{A}}^{k-1}
+\overline{\mathbf{A}}^{k-1}_{h})
\overline{\theta}_{\mathbf{A}}^{k-1},\;\theta_{\Psi}^{k-1}\right).}
\end{array}$$
By applying the Young’s inequality and (\[eq:3-15\]), we can estimate the first two terms on the right side of (\[eq:4-36\]) as follows. $$\label{eq:4-37}
\begin{array}{@{}l@{}}
{\displaystyle |\left(I_{h}\overline{\Psi}^{M}({\bm\pi}_{h}\overline{\mathbf{A}}^{M}
+\overline{\mathbf{A}}^{M}_{h})\overline{\theta}_{\mathbf{A}}^{M},\;\theta_{\Psi}^{M}\right)|
+|\left(I_{h}\overline{\Psi}^{0}({\bm\pi}_{h}\overline{\mathbf{A}}^{0}
+\overline{\mathbf{A}}^{0}_{h})\overline{\theta}_{\mathbf{A}}^{0},\;\theta_{\Psi}^{0}\right)|}\\[2mm]
{\displaystyle \quad\leq \frac{1}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M})+ C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+ Ch^{2r}.}
\end{array}$$
From (\[eq:3-14\]) and (\[eq:3-15\]), we further deduce $$\label{eq:4-39}
\begin{array}{@{}l@{}}
{\displaystyle |\left(I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}
+\overline{\mathbf{A}}^{k}_{h})\overline{\theta}_{\mathbf{A}}^{k}-I_{h}\overline{\Psi}^{k-1}
({\bm\pi}_{h}\overline{\mathbf{A}}^{k-1}+\overline{\mathbf{A}}^{k-1}_{h})
\overline{\theta}_{\mathbf{A}}^{k-1},\;\theta_{\Psi}^{k-1}\right)|}\\[2mm]
{\displaystyle \quad\leq \Delta t \|I_{h}\overline{\Psi}^{k}\|_{\mathcal{L}^6}\| {\bm\pi}_{h}\overline{\mathbf{A}}^{k}+\overline{\mathbf{A}}^{k}_{h}\|_{\mathbf{L}^6}
\|\frac{1}{\Delta t} (\overline{\theta}_{\mathbf{A}}^{k}-\overline{\theta}_{\mathbf{A}}^{k-1})\|_{\mathbf{L}^2}\|\theta_{\Psi}^{k-1}\|_{\mathcal{L}^6}}\\[2mm]
{\displaystyle \quad \quad+ \Delta t \|\frac{I_{h}\overline{\Psi}^{k}-I_{h}\overline{\Psi}^{k-1}}{\Delta t}\|_{\mathcal{L}^2}\| {\bm\pi}_{h}\overline{\mathbf{A}}^{k}+\overline{\mathbf{A}}^{k}_{h}\|_{\mathbf{L}^6}
\|\overline{\theta}_{\mathbf{A}}^{k-1}\|_{\mathbf{L}^6}\|\theta_{\Psi}^{k-1}\|_{\mathcal{L}^6}}\\[2mm]
{\displaystyle \quad\quad +\Delta t \|I_{h}\overline{\Psi}^{k-1}\|_{\mathcal{L}^6}\| \overline{\theta}_{\mathbf{A}}^{k-1}\|_{\mathbf{L}^6}\|\frac{{\bm\pi}_{h}\overline{\mathbf{A}}^{k}
-{\bm\pi}_{h}\overline{\mathbf{A}}^{k-1}}{\Delta t}+\frac{\overline{\mathbf{A}}_{h}^{k}-\overline{\mathbf{A}}_{h}^{k-1}}{\Delta t}\|_{\mathbf{L}^2}\|\theta_{\Psi}^{k-1}\|_{\mathcal{L}^6}}\\[2mm]
{\displaystyle \quad\leq C\Delta t\Big\{\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}
+ \|\partial {\theta}_{\mathbf{A}}^{k-1}\|_{\mathbf{L}^2}^{2}
+ D({\theta}_{\mathbf{A}}^{k-1},{\theta}_{\mathbf{A}}^{k-1}) + +D({\theta}_{\mathbf{A}}^{k-2},{\theta}_{\mathbf{A}}^{k-2})
+\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}^{2}\Big\},}\\[2mm]
\end{array}$$ where we have used the fact: $$\Vert \partial I_{h}\overline{\Psi}^{k} \Vert_{\mathcal{L}^{2}} \leq \Vert \partial \overline{\Psi}^{k} \Vert_{\mathcal{L}^{2}} + \Vert I_{h}\partial \overline{\Psi}^{k} - \partial \overline{\Psi}^{k}\Vert_{\mathcal{L}^{2}} \leq C.$$
Hence we get the following estimate: $$\label{eq:4-40}
\begin{array}{@{}l@{}}
{\displaystyle |Q_1|= |\sum_{k=1}^{M}\left(I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}
+\overline{\mathbf{A}}^{k}_{h})({\bm\pi}_{h}\overline{\mathbf{A}}^{k}-\overline{\mathbf{A}}^{k}_{h}),\;\theta_{\Psi}^{k}
-\theta_{\Psi}^{k-1}\right)|}\\[2mm]
{\displaystyle \quad\quad\quad\leq \frac{1}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M})
+ C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+ C h^{2r} }\\[2mm]
{\displaystyle \quad\quad\quad + C\Delta t \sum_{k=1}^{M}\left(\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}
+ D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})+ \|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
Employing (\[eq:4-1\]) and integrating by parts, we discover $$\label{eq:4-41}
\begin{array}{@{}l@{}}
{\displaystyle \mathrm{i}Q_2=\left(\nabla I_{h}\overline{\Psi}^{M}\overline{\theta}_{\mathbf{A}}^{M},\;\theta_{\Psi}^{M}\right)
+ \left(I_{h}\overline{\Psi}^{M}\nabla\cdot\overline{\theta}_{\mathbf{A}}^{M},\;\theta_{\Psi}^{M}\right)
+\left(I_{h}\overline{\Psi}^{0}\overline{\theta}_{\mathbf{A}}^{0},\;\nabla\theta_{\Psi}^{0}\right)}\\[2mm]
{\displaystyle \quad\quad +\sum_{k=1}^{M}\left(I_{h}\overline{\Psi}^{k}\overline{\theta}_{\mathbf{A}}^{k}- I_{h}\overline{\Psi}^{k-1}\overline{\theta}_{\mathbf{A}}^{k-1},\;\nabla\theta_{\Psi}^{k-1}\right),}
\end{array}$$
By using the Young’s inequality, we can estimate the first three terms on the right side of (\[eq:4-41\]) as follows: $$\label{eq:4-42}
\begin{array}{@{}l@{}}
{\displaystyle |\left(\nabla I_{h}\overline{\Psi}^{M}\overline{\theta}_{\mathbf{A}}^{M},\;\theta_{\Psi}^{M}\right)|
+|\left(I_{h}\overline{\Psi}^{M}\nabla\cdot\overline{\theta}_{\mathbf{A}}^{M},\;\theta_{\Psi}^{M}\right)|
+|\left(I_{h}\overline{\Psi}^{0}\overline{\theta}_{\mathbf{A}}^{0},\;\nabla\theta_{\Psi}^{0}\right)|}\\[2mm]
{\displaystyle \quad\leq \frac{1}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M})+ C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+C h^{2r}.}
\end{array}$$
Using (\[eq:4-1\]), the last term on the right side of (\[eq:4-41\]) can be estimated by $$\label{eq:4-43}
\begin{array}{@{}l@{}}
{\displaystyle |\sum_{k=1}^{M}\left(I_{h}\overline{\Psi}^{k}\overline{\theta}_{\mathbf{A}}^{k}- I_{h}\overline{\Psi}^{k-1}\overline{\theta}_{\mathbf{A}}^{k-1},\;\nabla\theta_{\Psi}^{k-1}\right)|}\\[2mm]
{\displaystyle \quad\leq C\Delta t \sum_{k=1}^{M}\left(\|\overline{\theta}_{\mathbf{A}}^{k}\|_{\mathbf{H}^{1}}
\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}+\|\partial \overline{\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}\right)}\\[2mm]
{\displaystyle \quad\leq C\Delta t \sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
Hence we get $$\label{eq:4-44}
\begin{array}{@{}l@{}}
{\displaystyle |Q_2|=|-\mathrm{i}\sum_{k=1}^{M}\left(I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}-\overline{\mathbf{A}}^{k}_{h}),\;
\nabla\theta_{\Psi}^{k}-\nabla\theta_{\Psi}^{k-1}\right)|}\\[2mm]
{\displaystyle \quad\quad\leq \frac{1}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M})
+ C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+C h^{2r}}\\[2mm]
{\displaystyle \quad \quad\quad+ C\Delta t \sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
Reasoning as before, we can estimate $ Q_3 $ as follows: $$\label{eq:4-45}
\begin{array}{@{}l@{}}
{\displaystyle |Q_3|=|\mathrm{i}\sum_{k=1}^{M}\left(\nabla I_{h}\overline{\Psi}^{k}({\bm\pi}_{h}\overline{\mathbf{A}}^{k}
-\overline{\mathbf{A}}^{k}_{h}),\;\theta_{\Psi}^{k}-\theta_{\Psi}^{k-1}\right)|}\\[2mm]
{\displaystyle \quad\quad\leq \frac{1}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M}) + C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+C h^{2r}}\\[2mm]
{\displaystyle \quad\quad\quad + C\Delta t \sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
Combining (\[eq:4-40\]), (\[eq:4-44\]) and (\[eq:4-45\]) implies $$\label{eq:4-46}
\begin{array}{@{}l@{}}
{\displaystyle |\sum_{k=1}^{M} J_4^{(k),3}| \leq \frac{3}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M})
+ C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+C h^{2r}}\\[2mm]
{\displaystyle \quad\quad\quad + C\Delta t \sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
It follows from (\[eq:4-30\]), (\[eq:4-34\]) and (\[eq:4-46\]) that $$\label{eq:4-47}
\begin{array}{@{}l@{}}
{\displaystyle \displaystyle |\sum_{k=1}^{M} J_4^{(k)}|\leq C\left(h^{2r}+(\Delta t)^{4}\right)+\frac{3}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M}) +\frac{1}{16} \|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}
+C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2} }\\[2mm]
{\displaystyle \qquad+C\Delta t \sum_{k=0}^{M}\Big\{D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k}) +\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\Big\} }\\[2mm]
\end{array}$$
Now take the real part of (\[eq:4-16\]), and we get $$\label{eq:4-48}
\begin{array}{@{}l@{}}
{\displaystyle \Delta t \mathrm{Re}\left[B\left(\overline{\mathbf{A}}_{h}^{k};
\overline{\theta}_{\Psi}^{k},\partial \theta_{\Psi}^{k}\right)\right] =\mathrm{Re}
\big(J_1^{(k)}\big)+\mathrm{Re}\big(J_2^{(k)}\big)+\mathrm{Re}
\big(J_3^{(k)}\big)+\mathrm{Re}
\big(J_4^{(k)}\big).}
\end{array}$$
Similarly to (\[eq:3-11\]), we have $$\label{eq:4-49}
\begin{array}{@{}l@{}}
{\displaystyle \mathrm{Re}\left[B\left(\overline{\mathbf{A}}_{h}^{k};\overline{\theta}_{\Psi}^{k},\partial \theta_{\Psi}^{k}\right)\right]=
-\left(\frac{1}{2}(\overline{\mathbf{A}}_{h}^{k}+\overline{\mathbf{A}}_{h}^{k-1})|\theta_{\Psi}^{k-1}|^{2},\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\right)}\\[2mm]
{\displaystyle\quad +\frac{1}{2}\partial B(\overline{\mathbf{A}}_{h}^{k};\theta_{\Psi}^{k},\theta_{\Psi}^{k})-\left(f(\theta_{\Psi}^{k-1},\theta_{\Psi}^{k-1}),\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\right).}
\end{array}$$
Substituting (\[eq:4-49\]) into (\[eq:4-48\]) and summing over $k = 1,2,\cdots,M$, we get $$\label{eq:4-51}
\begin{array}{@{}l@{}}
{\displaystyle \frac{1}{2}B(\overline{\mathbf{A}}_{h}^{M};\theta_{\Psi}^{M},\theta_{\Psi}^{M})
=\sum_{k=1}^{M}\left[\mathrm{Re}
\big(J_1^{(k)}\big)+\mathrm{Re}\big(J_2^{(k)}\big)+\mathrm{Re}
\big(J_3^{(k)}\big)+\mathrm{Re}
\big(J_4^{(k)}\big)\right]}\\[2mm]
{\displaystyle\quad+ \frac{1}{2}B(\overline{\mathbf{A}}_{h}^{0};\theta_{\Psi}^{0},\theta_{\Psi}^{0})+\Delta t\sum_{k=1}^{M} \Big(\frac{1}{2}(\overline{\mathbf{A}}_{h}^{k}+\overline{\mathbf{A}}_{h}^{k-1})|\theta_{\Psi}^{k-1}|^{2},\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\Big)}\\[2mm]
{\displaystyle \quad+\Delta t\sum_{k=1}^{M}\Big(f(\theta_{\Psi}^{k-1},\theta_{\Psi}^{k-1}),\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\Big).}
\end{array}$$
Combining (\[eq:4-19\]), (\[eq:4-22\]), (\[eq:4-23-0\]) and (\[eq:4-47\]) implies $$\label{eq:4-53}
\begin{array}{@{}l@{}}
{\displaystyle
\sum_{k=1}^{M}\left[\mathrm{Re}
\big(J_1^{(k)}\big)+\mathrm{Re}\big(J_2^{(k)}\big)+\mathrm{Re}
\big(J_3^{(k)}\big)+\mathrm{Re}
\big(J_4^{(k)}\big)\right]}\\[2mm]
{\displaystyle\quad \leq |\sum_{k=1}^{M}J_1^{(k)}|+|\sum_{k=1}^{M}J_2^{(k)}|+|\sum_{k=1}^{M}J_3^{(k)}|+|\sum_{k=1}^{M}J_4^{(k)}|}\\[2mm]
{\displaystyle \leq C\left(h^{2r}+(\Delta t)^{4}\right)+\frac{3}{16}D(\overline{\theta}_{\mathbf{A}}^{M}, \overline{\theta}_{\mathbf{A}}^{M})
+\frac{1}{8} \|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2} +C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}}\\[2mm]
{\displaystyle \qquad+C\Delta t \sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
By employing Theorem \[thm3-1\], we discover $$\label{eq:4-53-0}
\Big(\frac{1}{2}(\overline{\mathbf{A}}_{h}^{k}+\overline{\mathbf{A}}_{h}^{k-1})|\theta_{\Psi}^{k-1}|^{2},\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\Big) \leq C \|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}^{2}.$$
Setting $$\label{eq:4-54}
J_5^{(k)} = \Big(f(\theta_{\Psi}^{k-1},\theta_{\Psi}^{k-1}),\frac{1}{2}(\partial \mathbf{A}_{h}^{k}+\partial \mathbf{A}_{h}^{k-1})\Big),$$ and applying (\[eq:4-53\]) and (\[eq:4-53-0\]), we obtain $$\label{eq:4-55}
\begin{array}{@{}l@{}}
{\displaystyle \frac{1}{2}B(\overline{\mathbf{A}}_{h}^{M};\theta_{\Psi}^{M},\theta_{\Psi}^{M}) \leq C\left(h^{2r}+(\Delta t)^{4}\right)+\frac{3}{16}D(\overline{\theta}_{\mathbf{A}}^{M},\overline{\theta}_{\mathbf{A}}^{M}) +\frac{1}{8} \|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}
+C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}}\\[2mm]
{\displaystyle \qquad +\Delta t \sum_{k=1}^{M} J_5^{(k)} + C\Delta t \sum_{k=0}^{M}\left\{D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right\}.}
\end{array}$$
Arguing as in the proof of Theorem \[thm3-1\], we discover $$\label{eq:4-56}
\|\left(\mathrm{i}\nabla
+\overline{\mathbf{A}}^{M}_{h}\right)\theta^{M}_{\Psi}\|_{\mathbf{L}^2}\geq \|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2} - \|\overline{\mathbf{A}}_{h}^{M}\theta_{\Psi}^{M}\|_{\mathbf{L}^2} \geq \frac{3}{4}\|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2} - C\|\theta_{\Psi}^{M}\|_{\mathbf{L}^2},$$ and thus $$\label{eq:4-57}
B(\overline{\mathbf{A}}_{h}^{M};\theta_{\Psi}^{M},\theta_{\Psi}^{M}) + C\|\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2} \geq \frac{9}{32}\|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}.$$
Substituting (\[eq:4-57\]) into (\[eq:4-55\]) , we obtain $$\label{eq:4-59}
\begin{array}{@{}l@{}}
{\displaystyle \frac{1}{64}\|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}
\leq C\left(h^{2r}+(\Delta t)^{4}\right)+\frac{3}{16}D(\overline{\theta}_{\mathbf{A}}^{M},\overline{\theta}_{\mathbf{A}}^{M}) +C\|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+\Delta t \sum_{k=1}^{M} J_5^{(k)}}\\[2mm]
{\displaystyle \quad\quad\quad+C\Delta t \sum_{k=0}^{M}\left\{D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right\}.}
\end{array}$$
Multiplying (\[eq:4-15\]) with $(C + 1)$ and adding to (\[eq:4-59\]), we end up with $$\label{eq:4-60}
\begin{array}{@{}l@{}}
{\displaystyle \|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+\frac{1}{64}\|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}
\leq C\left(h^{2r}+(\Delta t)^{4}\right) +\frac{3}{16}D(\overline{\theta}_{\mathbf{A}}^{M},\overline{\theta}_{\mathbf{A}}^{M})+\Delta t \sum_{k=1}^{M} J_5^{(k)}}\\[2mm]
{\displaystyle \quad\quad\quad+C\Delta t \sum_{k=0}^{M}\left\{D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right\}.}
\end{array}$$
Estimates for (\[eq:4-4\])
--------------------------
Setting $$\label{eq:4-61}
\begin{array}{@{}l@{}}
{\displaystyle K_1^{(k)}=\left((\mathbf{A}_{tt})^{k-1}-\partial^{2}{\bm\pi}_{h}\mathbf{A}^{k},\mathbf{v}\right),}\\[2mm]
{\displaystyle K_2^{(k)}= D(\mathbf{A}^{k-1}-\widetilde{{\bm\pi}_{h}\mathbf{A}^{k}},\mathbf{v}),}\\[2mm]
{\displaystyle K_3^{(k)} = \Big(|\Psi^{k-1}|^{2}\mathbf{A}^{k-1}-|\Psi^{k-1}_{h}|^{2}\frac{\overline{\mathbf{A}}^{k}_{h}
+\overline{\mathbf{A}}^{k-1}_{h}}{2},\;\mathbf{v}\Big),}\\[2mm]
{\displaystyle K_4^{(k)}= \left(f(\Psi^{k-1},\Psi^{k-1})-f(\Psi^{k-1}_{h},\Psi^{k-1}_{h}),\;\mathbf{v}\right),}
\end{array}$$ we rewrite (\[eq:4-4\]) as follows: $$\label{eq:4-62}
{\displaystyle \left(\partial^{2}\theta^{k}_{\mathbf{A}},\mathbf{v}\right)+D(\widetilde{\theta^{k}_{\mathbf{A}}},\mathbf{v})=
K_1^{(k)}+ K_2^{(k)}+ K_3^{(k)}+ K_4^{(k)}.}$$
We first estimate $\sum\limits_{k=1}^{M} K_1^{(k)}$, $\sum\limits_{k=1}^{M} K_3^{(k)}$ and $\sum\limits_{k=1}^{M} K_4^{(k)}$. Under the regularity assumption of $\mathbf{A}$ in (\[eq:2-10\]), we have $$\label{eq:4-63}
\sum_{k=1}^{M} |K_1^{(k)}| \leq \frac{C}{\Delta t}\left(h^{2r+2}+(\Delta t)^{4}\right)
+ C\sum_{k=1}^{M} \|\mathbf{v}\|_{\mathbf{L}^2}^{2}.$$
By applying the regularity assumption (\[eq:2-10\]), the interpolation error estimates (\[eq:4-1\]) and Theorem \[thm3-1\], it is easy to deduce $$\label{eq:4-68}
\sum_{k=1}^{M} K_3^{(k)} \leq \frac{C}{\Delta t}\left[h^{2r}+(\Delta t)^{4}\right] + C\sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k}, {\theta}_{\mathbf{A}}^{k})+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2} +\|\mathbf{v}\|_{\mathbf{L}^2}^{2}\right).$$
In order to estimate $\sum\limits_{k=1}^{M} K_4^{(k)}$, we rewrite $ K_4^{(k)}$ in the following form: $$\label{eq:4-69}
\begin{array}{@{}l@{}}
{\displaystyle K_4^{(k)}=\left(f(\Psi^{k-1},\Psi^{k-1})-f(I_{h}\Psi^{k-1}, I_{h}\Psi^{k-1}),\;\mathbf{v}\right) }\\[2mm]
{\displaystyle \quad +\left(f(I_{h}\Psi^{k-1}, I_{h}\Psi^{k-1})-f(\Psi^{k-1}_{h},\Psi^{k-1}_{h}),\;\mathbf{v}\right)\stackrel{\mathrm{def}}{=} K_4^{(k),1}+K_4^{(k),2}.}
\end{array}$$
We observe that $$\label{eq:4-70}
\begin{array}{@{}l@{}}
{\displaystyle f(\psi,\psi)-f(\varphi,\varphi)=\frac{\mathrm{i}}{2}(\psi^{\ast}\nabla \psi- \psi \nabla\psi^{\ast})
-\frac{\mathrm{i}}{2} (\varphi^{\ast} \nabla\varphi-\varphi \nabla\varphi^{\ast}) }\\[2mm]
{\displaystyle = -\frac{\mathrm{i}}{2}\left(\varphi^{\ast}\nabla(\varphi-\psi)-\varphi\nabla(\varphi-\psi)^{\ast}\right)
+\frac{\mathrm{i}}{2}\left((\varphi-\psi)\nabla\psi^{\ast}-(\varphi-\psi)^{\ast}\nabla\psi\right).}
\end{array}$$
We obtain from (\[eq:4-1\]) and (\[eq:4-70\]) that $$\label{eq:4-71}
\begin{array}{@{}l@{}}
{\displaystyle K_4^{(k),1} \leq C\left\{h^{2r}+\|\mathbf{v}\|_{\mathbf{L}^2}^{2}\right\}.}
\end{array}$$
Similarly, from (\[eq:4-1\]) and (\[eq:4-70\]), we deduce $$\label{eq:4-72}
\begin{array}{@{}l@{}}
{\displaystyle K_4^{(k),2}=\left(f(I_{h}\Psi^{k-1}, I_{h}\Psi^{k-1})-f(\Psi^{k-1}_{h},\Psi^{k-1}_{h}),\;\mathbf{v}\right)}\\[2mm]
{\displaystyle \quad = -\frac{\mathrm{i}}{2}\left((\theta_{\Psi}^{k-1})^{\ast}\nabla\theta_{\Psi}^{k-1}
-\theta_{\Psi}^{k-1}\nabla(\theta_{\Psi}^{k-1})^{\ast},\;\mathbf{v}\right) }\\[2mm]
{\displaystyle \quad\quad-\frac{\mathrm{i}}{2}\left((I_{h}\Psi^{k-1})^{\ast}\nabla\theta_{\Psi}^{k-1}
-I_{h}\Psi^{k-1}\nabla(\theta_{\Psi}^{k-1})^{\ast},\;\mathbf{v}\right) }\\[2mm]
{\displaystyle\quad \quad +\frac{\mathrm{i}}{2}\left(\theta_{\Psi}^{k-1}\nabla (I_{h}\Psi^{k-1})^{\ast}-(\theta_{\Psi}^{k-1})^{\ast}\nabla I_{h}\Psi^{k-1},\;\mathbf{v}\right)}\\[2mm]
{\displaystyle \quad \leq -\left(f(\theta_{\Psi}^{k-1},\theta_{\Psi}^{k-1}),\mathbf{v}\right)
+ C \| I_{h}\Psi^{k-1}\|_{\mathcal{L}^{\infty}}\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}\|\mathbf{v}\|_{\mathbf{L}^2}}\\[2mm]
{\displaystyle \quad\quad+C\|\nabla I_{h}\Psi^{k-1}\|_{\mathbf{L}^{3}}\|\theta_{\Psi}^{k-1}\|_{\mathcal{L}^6}
\|\mathbf{v}\|_{\mathbf{L}^2}}\\[2mm]
{\displaystyle \quad \leq -\left(f(\theta_{\Psi}^{k-1},\theta_{\Psi}^{k-1}),\mathbf{v}\right)
+C\left(\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}^{2}
+\|\mathbf{v}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
Combining (\[eq:4-69\]), (\[eq:4-71\]) and (\[eq:4-72\]) gives $$\label{eq:4-73}
\begin{array}{@{}l@{}}
{\displaystyle\sum_{k=1}^{M} K_4^{(k)}\leq \frac{C h^{2r}}{\Delta t} -\sum_{k=1}^{M}\left(f(\theta_{\Psi}^{k-1},\theta_{\Psi}^{k-1}),\mathbf{v}\right)
+C\sum_{k=1}^{M}\left(\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^2}^{2}
+\|\mathbf{v}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
From (\[eq:4-63\]), (\[eq:4-68\]) and (\[eq:4-73\]), we obtain $$\label{eq:4-74}
\begin{array}{@{}l@{}}
{\displaystyle \sum_{k=1}^{M} \Big(K_1^{k)}+K_3^{k)}+K_4^{k)}\Big)
\leq -\sum_{k=1}^{M}\left(f(\theta_{\Psi}^{k-1},
\theta_{\Psi}^{k-1}),\mathbf{v}\right)+\frac{C}{\Delta t}\left\{h^{2r}+(\Delta t)^{4}\right\}}\\[2mm]
{\displaystyle \qquad \quad+ C\sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k}, {\theta}_{\mathbf{A}}^{k})
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}+\|\mathbf{v}\|_{\mathbf{L}^2}^{2}\right).}
\end{array}$$
Now taking $\mathbf{v}= \frac{\displaystyle 1}{\displaystyle 2\Delta t}(\theta_{\mathbf{A}}^{k}
-\theta_{\mathbf{A}}^{k-2})=\frac{\displaystyle 1}{\displaystyle 2}(\partial \theta_{\mathbf{A}}^{k}
+\partial \theta_{\mathbf{A}}^{k-1})$ in (\[eq:4-62\]), we find $$\label{eq:4-75}
\begin{array}{@{}l@{}}
{\displaystyle \Big(\partial^{2}\theta^{k}_{\mathbf{A}},\frac{1}{2}(\partial \theta_{\mathbf{A}}^{k}+\partial \theta_{\mathbf{A}}^{k-1})\Big)
+D\Big(\widetilde{\theta^{k}_{\mathbf{A}}},\frac{1}{2}(\partial \theta_{\mathbf{A}}^{k}+\partial \theta_{\mathbf{A}}^{k-1})\Big)}\\[2mm]
{\displaystyle \quad = \frac{1}{2\varDelta t}\left(\|\partial \theta_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}
-\|\partial \theta_{\mathbf{A}}^{k-1}\|_{\mathbf{L}^2}^{2}\right)+\frac{1}{4\Delta t}\left(D({\theta}_{\mathbf{A}}^{k}, {\theta}_{\mathbf{A}}^{k})
-(D({\theta}_{\mathbf{A}}^{k-2}, {\theta}_{\mathbf{A}}^{k-2})\right)}\\[2mm]
{\displaystyle \quad = K_1^{(k)}+ K_2^{(k)}+ K_3^{(k)}+ K_4^{(k)}.}
\end{array}$$
Note that $$\mathbf{v} = \frac{1}{2}(\partial \theta_{\mathbf{A}}^{k}
+\partial \theta_{\mathbf{A}}^{k-1}) =\frac{1}{2}(\partial \mathbf{A}_{h}^{k}
+\partial \mathbf{A}_{h}^{k-1}) - \frac{1}{2}(\partial {\bm \pi}_{h}{\mathbf{A}}^{k}
+\partial {\bm \pi}_{h}{\mathbf{A}}^{k-1}).$$
Thus we have $$\label{eq:4-75-0}
\begin{array}{@{}l@{}}
{\displaystyle -\sum_{k=1}^{M}\left(f(\theta_{\Psi}^{k-1},
\theta_{\Psi}^{k-1}),\mathbf{v}\right) }\\[2mm]
{\displaystyle = -\sum_{k=1}^{M} J_5^{(k)} + \sum_{k=1}^{M}\left(f(\theta_{\Psi}^{k-1},
\theta_{\Psi}^{k-1}),\frac{\partial {\bm \pi}_{h}{\mathbf{A}}^{k}
+\partial {\bm \pi}_{h}{\mathbf{A}}^{k-1}}{2}\right) }\\[2mm]
{\displaystyle \leq -\sum_{k=1}^{M} J_5^{(k)} + C\sum_{k=1}^{M}\|\nabla\theta_{\Psi}^{k-1}\|_{\mathbf{L}^{2}}^{2}}.
\end{array}$$ Here we have used the definition of $ J_5^{(k)} $ in (\[eq:4-54\]).
Multiplying (\[eq:4-75\]) by $\Delta t $, and using (\[eq:4-74\]) and (\[eq:4-75-0\]), we obtain $$\label{eq:4-76}
\begin{array}{@{}l@{}}
{\displaystyle \frac{1}{2}\|\partial \theta_{\mathbf{A}}^{M}\|_{\mathbf{L}^2}^{2}
+\frac{1}{4}D({\theta}_{\mathbf{A}}^{M}, {\theta}_{\mathbf{A}}^{M})
+\frac{1}{4}D({\theta}_{\mathbf{A}}^{M-1}, {\theta}_{\mathbf{A}}^{M-1})}\\[2mm]
{\displaystyle \quad \leq C\left\{h^{2r}+(\Delta t)^{4} \right\}
+ C\Delta t\sum_{k=0}^{M}\left(D({\theta}_{\mathbf{A}}^{k}, {\theta}_{\mathbf{A}}^{k})+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}
+ \|\partial \theta_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2} \right) }\\[2mm]
{\displaystyle \quad+ \Delta t \sum_{k=1}^{M} K_2^{(k)}
-\Delta t\sum_{k=1}^{M} J_5^{(k)}.}
\end{array}$$
Since $\mathbf{v}=\frac{\displaystyle 1}{\displaystyle 2}(\partial \theta_{\mathbf{A}}^{k}+\partial \theta_{\mathbf{A}}^{k-1})$, by applying (\[eq:4-2\]) and the Young’s inequality, we get $$\label{eq:4-77}
\begin{array}{@{}l@{}}
{\displaystyle \Delta t \sum_{k=1}^{M} K_2^{(k)}=\Delta t \sum_{k=1}^{M} D(\mathbf{A}^{k-1}
-\widetilde{{\bm\pi}_{h}\mathbf{A}^{k}}, \frac{1}{2}(\partial \theta_{\mathbf{A}}^{k}
+\partial \theta_{\mathbf{A}}^{k-1}))}\\[2mm]
{\displaystyle \leq C\left\{h^{2r} + (\Delta t)^{4}\right\} + \frac{1}{32} D\left(\theta_{\mathbf{A}}^{M},\theta_{\mathbf{A}}^{M}\right)
+\frac{1}{32} D\left(\theta_{\mathbf{A}}^{M-1},\theta_{\mathbf{A}}^{M-1}\right)
+ C\Delta t \sum_{k=0}^{M}D\left(\theta_{\mathbf{A}}^{k},\theta_{\mathbf{A}}^{k}\right)}\\[2mm]
\end{array}$$
Combining (\[eq:4-60\]), (\[eq:4-76\]) and (\[eq:4-77\]) implies $$\label{eq:4-81}
\begin{array}{@{}l@{}}
{\displaystyle \|\theta_{\Psi}^{M}\|_{\mathcal{L}^2}^{2}+\frac{1}{64}\|\nabla\theta_{\Psi}^{M}\|_{\mathbf{L}^2}^{2}
+\frac{1}{2}\|\partial \theta_{\mathbf{A}}^{M}\|_{\mathbf{L}^2}^{2} + \frac{1}{8}D({\theta}_{\mathbf{A}}^{M}, {\theta}_{\mathbf{A}}^{M})
+\frac{1}{8}D({\theta}_{\mathbf{A}}^{M-1}, {\theta}_{\mathbf{A}}^{M-1})}\\[2mm]
{\displaystyle \quad\leq C\left\{h^{2r}+(\Delta t)^{4}\right\}
+ C\Delta t \sum_{k=0}^{M}\left\{D({\theta}_{\mathbf{A}}^{k},{\theta}_{\mathbf{A}}^{k})
+\|\partial {\theta}_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}+\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}\right\}.}
\end{array}$$
By applying the discrete Gronwall’s inequality, we have $$\label{eq:4-82}
\max_{1\leq k \leq M}\left\{\|\theta_{\Psi}^{k}\|_{\mathcal{L}^2}^{2}
+\|\nabla\theta_{\Psi}^{k}\|_{\mathbf{L}^2}^{2}
+\|\partial \theta_{\mathbf{A}}^{k}\|_{\mathbf{L}^2}^{2}
+D({\theta}_{\mathbf{A}}^{k}, {\theta}_{\mathbf{A}}^{k})\right\} \leq C\left\{h^{2r}+(\Delta t)^{4}\right\}.$$
Combine (\[eq:4-82\]) with the interpolation error estimates (\[eq:4-1\]) and we can complete the proof of Theorem \[thm2-1\].
Numerical tests {#sec-6}
===============
To validate the developed algorithm and to confirm the theoretical analysis reported in this paper, we present numerical simulations for the following case studies.
\[exam6-1\] We consider the Maxwell-Schrödinger system (\[eq:1-2\]), where the initial-boundary conditions are as follows: $$\begin{array}{@{}l@{}}
{\displaystyle \Psi(\mathbf{x},t)=0,\quad \mathbf{A}(\mathbf{x},t)\times\mathbf{n}=0, \quad (\mathbf{x},t)\in
\partial \Omega\times(0,T),} \\[2mm]
{\displaystyle \Psi(\mathbf{x},0) = 2\sqrt{2}\sin(\pi x_1)\sin(\pi x_2)\sin(\pi x_3) , \quad
\mathbf{A}_{t}(\mathbf{x},0)=\mathbf{A}_1(\mathbf{x})=0, }\\[2mm]
{\mathbf{A}(\mathbf{x},0)=\mathbf{A}_0(\mathbf{x})
=\big(10x_1 x_2 x_3(1-x_2)(1-x_3), 10x_1 x_2 x_3(1-x_1)(1-x_3), }\\[2mm]
{\displaystyle\qquad \qquad \qquad \qquad 10x_1 x_2 x_3(1-x_1)(1-x_2)\big). }
\end{array}$$ Here we take $\Omega=(0,1)^{3} $, $ V_0 = 0 $, $ T = 0.5$, and the time step $\Delta t = 0.0025 $. Note that the initial wave function $\Psi(\mathbf{x},0)$ is the eigenfunction of the stationary Schrödinger’s equation.
The numerical results are displayed in Fig. 5.1.
![Example \[exam6-1\] : (a) The evolution of the density function $ \rho(\mathbf{x},t) $ on the line $
x_1=x_2=x_3 $ at time $t =0.1, 0.2, 0.3 ,0.4, 0.5$; (b) the evolution of $ \rho(\mathbf{x}^{(1)},t) $, $ \rho(\mathbf{x}^{(2)},t) $ and $ \rho(\mathbf{x}^{(3)}, t)$, where $ \mathbf{x}^{(1)}=(0.25,0.5,0.75) $, $\mathbf{x}^{(2)}=(0.5,0.5,0.5) $ and $ \mathbf{x}^{(3)}=(0.4,0.5,0.6)$. ](./diagrho.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-1\] : (a) The evolution of the density function $ \rho(\mathbf{x},t) $ on the line $
x_1=x_2=x_3 $ at time $t =0.1, 0.2, 0.3 ,0.4, 0.5$; (b) the evolution of $ \rho(\mathbf{x}^{(1)},t) $, $ \rho(\mathbf{x}^{(2)},t) $ and $ \rho(\mathbf{x}^{(3)}, t)$, where $ \mathbf{x}^{(1)}=(0.25,0.5,0.75) $, $\mathbf{x}^{(2)}=(0.5,0.5,0.5) $ and $ \mathbf{x}^{(3)}=(0.4,0.5,0.6)$. ](./points.eps "fig:"){width="6cm" height="6cm"}
\[fig:6-1\]
\[rem6-1\] The numerical results illustrated in Fig. 5.1 clearly show that the change of $\rho(\mathbf{x}, t)$ is smooth with respect to $t$ and the assumption on which the modified Maxwell-Schrödinger equations are based is valid in this case.
\[exam6-2\] We consider the modified Maxwell–Schrödinger’s equations as follows: $$\label{eq:6-11}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}\frac{\partial \Psi}{\partial t}+
\frac{1}{2}\left(\mathrm{i}\nabla +\mathbf{A}\right)^{2}\Psi
+ V_{0}\Psi = f ,\,\, (\mathbf{x},t)\in
\Omega\times(0,T),}\\[2mm]
{\displaystyle \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}}+\nabla\times
(\nabla\times \mathbf{A}) - \gamma\nabla(\nabla \cdot \mathbf{A}) +\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big)}\\[2mm]
{\displaystyle \quad\quad+\,\,\vert\Psi\vert^{2}\mathbf{A} =\mathbf{g}(\mathbf{x},t),\quad \,\, (\mathbf{x},t)\in \Omega\times(0,T),}
\end{array}
\right.$$ where the initial-boundary conditions are given in (\[eq:1-7\])-(\[eq:1-8\]). We take $\Omega = (0,1)^3$, $T = 4$, $ \gamma = 1$ and $V_0 = 5$. The exact solution $(\Psi,\mathbf{A})$ of (\[eq:6-11\]) is given by $$\begin{array}{@{}l@{}}
{\displaystyle \Psi(\mathbf{x}, t) = 20 e^{\mathrm{i}t}(1+3t^2)\exp\big((x_1+x_2+x_3)/5\big)x_1 x_2 x_3(1-x_1)(1-x_2)(1-x_3) }\\[2mm]
{\displaystyle \qquad\qquad\qquad+ 5.0e^{\mathrm{i}\pi t} \sin(2\pi x_1)\sin(2\pi x_2)\sin(2\pi x_3), }
\end{array}$$ $$\begin{array}{@{}l@{}}
{\displaystyle \mathbf{A}(\mathbf{x},t)=\sin(\pi t)\Big(\cos(2\pi x_1)\sin(2\pi x_2)\sin(2\pi x_3),
\sin(2\pi x_1)\cos(2\pi x_2)\sin(2\pi x_3),}\\[2mm]
{\displaystyle \quad \sin(2\pi x_1)\sin(2\pi x_2)\cos(2\pi x_3)\Big)
+ \cos(\pi t)\Big(\cos(\pi x_1)\sin(\pi x_2)\sin(\pi x_3),}\\[2mm]
{\displaystyle \quad
\sin(\pi x_1)\cos(\pi x_2)\sin(\pi x_3), \sin(\pi x_1)\sin(\pi x_2)\cos(\pi x_3)\Big).}
\end{array}$$ The functions $f$ and $\mathbf{g}$ in (\[eq:6-11\]) are chosen correspondingly to the exact solution $(\Psi,\mathbf{A})$.
A uniform tetrahedral partition is generated with $M+1$ nodes in each direction and $6M^{3}$ elements in total. We solve the system(\[eq:6-11\]) by the proposed Crank-Nicolson Galerkin finite element scheme (\[eq:2-9\]) with linear elements and quadratic elements, respectively. To confirm our error analysis, we take $\Delta t = h^{\frac{1}{2}}$ for the linear element method and $\Delta t = h$ for the quadratic element method respectively. Numerical results for the linear element method and the quadratic element method at time $t=1.0, 2.0, 3.0, 4.0 $ are listed in Tables \[table6-1\] and \[table6-2\], respectively.
$\Vert \mathbf{A}^{k}_{h} - \mathbf{A}(\cdot, t_k) \Vert _{\mathbf{H}^{1}} $
----- ------------ ------------------------------------------------------------------------------ ------------ -------
t M=25 M=50 M=100 Order
1.0 9.7908e-01 4.8501e-01 2.1082e-01 1.11
2.0 7.6414e-01 3.7807e-01 1.7681e-01 1.06
3.0 6.3094e-01 3.1006e-01 1.5308e-01 1.02
4.0 7.2739e-01 3.5204e-01 1.7705e-01 1.02
$ \Vert \Psi_{h}^{k}-\Psi(\cdot,t_k)\Vert_{\mathcal{H}^1}$
t M=25 M=50 M=100 Order
1.0 6.8289e-01 3.3004e-01 1.5046e-01 1.09
2.0 8.0035e-01 3.6227e-01 1.6032e-01 1.16
3.0 4.1192e-01 1.9485e-01 1.0286e-01 1.00
4.0 2.3418e-01 1.1430e-01 5.6022e-02 1.03
: $H^1$ error of linear FEM with $h = \frac{1}{M}$ and $\Delta t = h^{\frac12}$.[]{data-label="table6-1"}
$\Vert \mathbf{A}^{k}_{h} - \mathbf{A}(\cdot, t_k) \Vert _{\mathbf{H}^{1}} $
----- ------------ ------------------------------------------------------------------------------ ------------ -------
t M=25 M=50 M=100 Order
1.0 3.3770e-02 8.4984e-03 2.2743e-03 1.95
2.0 2.2786e-02 5.7068e-03 1.4966e-03 1.96
3.0 3.4016e-02 8.9115e-03 2.4360e-03 1.90
4.0 3.9787e-02 9.0740e-03 2.3467e-03 2.04
$ \Vert \Psi_{h}^{k}-\Psi(\cdot,t_k)\Vert_{\mathcal{H}^1}$
t M=25 M=50 M=100 Order
1.0 2.8221e-02 6.8364e-03 1.8024e-03 1.98
2.0 4.5738e-02 1.1608e-02 2.6719e-03 2.05
3.0 3.2712e-02 8.5698e-03 2.2101e-03 1.94
4.0 2.1868e-02 5.1495e-03 1.3250e-03 2.02
: $H^1$ error of quadratic FEM with $h = \Delta t= \frac{1}{M}.$[]{data-label="table6-2"}
\[rem6-2\] Numerical results in Tables \[table6-1\] and \[table6-2\] are in good agreement with the theoretical analysis, see Theorem \[thm2-1\].
\[exam6-3\] We consider the following modified Maxwell–Schrödinger’s equations $$\label{eq:6-12}
\left\{
\begin{array}{@{}l@{}}
{\displaystyle -\mathrm{i}\frac{\partial \Psi}{\partial t}+
\frac{1}{2}\left(\mathrm{i}\nabla +\mathbf{A}\right)^{2}\Psi
+ V_{0}\Psi = 0 ,\,\, (\mathbf{x},t)\in
\Omega\times(0,T),}\\[2mm]
{\displaystyle \frac{\partial ^{2}\mathbf{A}}{\partial t^{2}}+\nabla\times
(\nabla\times \mathbf{A}) - \gamma\nabla(\nabla \cdot \mathbf{A}) +\frac{\mathrm{i}}{2}\big(\Psi^{*}\nabla{\Psi}-\Psi\nabla{\Psi}^{*}\big)}\\[2mm]
{\displaystyle \quad\quad+\,\,\vert\Psi\vert^{2}\mathbf{A} =\mathbf{g}(\mathbf{x},t),\quad \,\, (\mathbf{x},t)\in \Omega\times(0,T).}\\[2mm]
{\displaystyle \Psi(\mathbf{x},t)=0,\quad \mathbf{A}(\mathbf{x},t)\times\mathbf{n}=0, \quad \nabla \cdot \mathbf{A}(\mathbf{x},t) = 0, \quad (\mathbf{x},t)\in
\partial \Omega\times(0,T),}\\[2mm]
{\displaystyle \Psi(\mathbf{x},0) = \Psi_0(\mathbf{x}),\quad\mathbf{A}(\mathbf{x},0)=\mathbf{A}_{0}(\mathbf{x}),\quad
\mathbf{A}_{t}(\mathbf{x},0)=\mathbf{A}_{1}(\mathbf{x}),}
\end{array}
\right.$$ with $$\begin{array}{@{}l@{}}
{\displaystyle \Psi_{0}(\mathbf{x}) = 2\sqrt{2}\sin(\pi x_1)\sin(\pi x_2)\sin(\pi x_3) , \quad
\mathbf{A}_{0}(\mathbf{x})=\mathbf{A}_{1}(\mathbf{x})=0,} \\[2mm]
{\displaystyle \mathbf{g}(\mathbf{x}, t) = \left(10\sin(1.5{\pi}^{2} t), 10\sin(1.5{\pi}^{2} t), 10\cos(1.5{\pi}^{2} t)\right).}
\end{array}$$
In this example we take $\Omega = (0,1)^3$, $T = 10.0$ , $\gamma = 1$, $V_0 = 0$. Using the mesh in Example \[exam6-2\] with M = 50, we solve the system (\[eq:6-12\]) by the proposed Crank-Nicolson Galerkin finite element scheme (\[eq:2-9\]) with linear elements. The time step $\Delta t = 0.0025 $.
In Fig. 5.2 we display the numerical results of $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$.
In Fig. 5.3 we plot the numerical results of $\rho = \vert\Psi\vert^{2}$ on the intersections $x_1 = 0.4, \,x_2=0.4 \,$ and $ x_3 = 0.4$ at time $t =10.0$, respectively.
![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and the contour plots of $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$; (b) the contour plot at time $t =2.0$; (c) the contour plot at time $t =4.0$; (d) the contour plot at time $t =6.0$; (e) the contour plot at time $t =8.0$; (f) the contour plot at time $t =10.0$; []{data-label="fig6.2"}](./diagrho_2.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and the contour plots of $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$; (b) the contour plot at time $t =2.0$; (c) the contour plot at time $t =4.0$; (d) the contour plot at time $t =6.0$; (e) the contour plot at time $t =8.0$; (f) the contour plot at time $t =10.0$; []{data-label="fig6.2"}](./contour_2.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and the contour plots of $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$; (b) the contour plot at time $t =2.0$; (c) the contour plot at time $t =4.0$; (d) the contour plot at time $t =6.0$; (e) the contour plot at time $t =8.0$; (f) the contour plot at time $t =10.0$; []{data-label="fig6.2"}](./contour_4.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and the contour plots of $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$; (b) the contour plot at time $t =2.0$; (c) the contour plot at time $t =4.0$; (d) the contour plot at time $t =6.0$; (e) the contour plot at time $t =8.0$; (f) the contour plot at time $t =10.0$; []{data-label="fig6.2"}](./contour_6.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and the contour plots of $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$; (b) the contour plot at time $t =2.0$; (c) the contour plot at time $t =4.0$; (d) the contour plot at time $t =6.0$; (e) the contour plot at time $t =8.0$; (f) the contour plot at time $t =10.0$; []{data-label="fig6.2"}](./contour_8.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ and the contour plots of $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.5$ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the line $
x_1=x_2=x_3 $ at time $t =2.0, 4.0, 6.0 ,8.0, 10.0$; (b) the contour plot at time $t =2.0$; (c) the contour plot at time $t =4.0$; (d) the contour plot at time $t =6.0$; (e) the contour plot at time $t =8.0$; (f) the contour plot at time $t =10.0$; []{data-label="fig6.2"}](./contour_10.eps "fig:"){width="6cm" height="6cm"}
![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.4,\,x_2 = 0.4$ and $x_3 = 0.4$ at time $t = 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.4$ ; (b) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_2 = 0.4$; (c) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_3 = 0.4$; []{data-label="fig6.3"}](./slice_x1.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.4,\,x_2 = 0.4$ and $x_3 = 0.4$ at time $t = 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.4$ ; (b) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_2 = 0.4$; (c) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_3 = 0.4$; []{data-label="fig6.3"}](./slice_x2.eps "fig:"){width="6cm" height="6cm"} ![Example \[exam6-3\]: numerical results of the density function $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.4,\,x_2 = 0.4$ and $x_3 = 0.4$ at time $t = 10.0$. (a) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_1 = 0.4$ ; (b) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_2 = 0.4$; (c) $\rho = \vert\Psi\vert^{2}$ on the intersection $x_3 = 0.4$; []{data-label="fig6.3"}](./slice_x3.eps "fig:"){width="6cm" height="6cm"}
Althouth the Maxwell-Schrödinger system and the time-dependent Ginzburg-Landau equations are somehow formally similar, they describe the different physical phenomenons. The time-dependent Ginzburg-Landau equations describe the vortex dynamics of superconductor [@Gao-1; @Gao-2; @Gao-3] while the Maxwell-Schrödinger equations describe the wave packet dynamics of an electron. As can be seen in Fig. 5.2-5.3, the wave packet of the electron is located at the center of the computational domain at first and the external and its self-induced electromagnetic fields cause the motion of the wave packet. Unlike the time-dependent Ginzburg-Landau equations, no stable state is observed in our computation.
Conclusions
===========
We have presented the optimal $H^1$ error estimates of a Crank-Nicolson Galerkin finite element method for the modified Maxwell-Schrödinger equations , which are derived from the original equations under some assumptions. The techniques used in this paper may also be applied to other nonlinear PDEs, such as the Ginzburg-Landau equations. The original Maxwell-Schrödinger system is challenging and difficult to perform numerical computation and theoretical analysis. Our work can serve as an elementary attempt for the numerical analysis of this system. We will study the original system in a further work using the mixed finite element method.
[99]{}
, [*Simulation of plasmonics nanodevices with coupled Maxwell and schrödinger equations using the FDTD method*]{}, Advanced Electromagnetics (2012), 1(1): pp. 76–83.
,[*The Mathematical Theory of Finite Element Methods*]{}, Springer, New York, 2002.
, [*Unconditionally optimal error estimates of a Crank–Nicolson Galerkin method for the nonlinear thermistor equations*]{}, SIAM J. Numer. Anal.(2014), 52(2): pp. 933–954.
, [*Numerical studies of a non-stationary Ginzburg-Landau model for superconductivity*]{}, Adv. Math. Sci. Appl. (1995), 5: pp. 363–389.
,[*Vector Potential Electromagnetics with Generalized Gauge for Inhomogeneous Media: Formulation*]{}, Progress In Electromagnetics Research (2014), 149: pp. 69–84.
, [*Analysis and approxiamtion of the GL model of superconductivity*]{}, SIAM Rev.(1992), 34: pp. 54-81.
, [*Partial Differential Equations*]{}, American Mathematical Society, 1998.
, [*Optimal error estimates of linearized Crank-Nicolson Galerkin FEMs for the time-dependent Ginzburg-Landau equations in superconductivity*]{}, SIAM J. Numer. Anal. (2014), 52(3): pp. 1183–1202.
, [*An efficient fully linearized semi-implicit Galerkin-mixed FEM for the dynamical Ginzburg-Landau equations of superconductivity*]{}, J. Comput. Phys .(2015), 294: pp. 329–345.
, [*A new mixed formulation and efficient numerical solution of Ginzburg-Landau equations under the temporal gauge*]{}, SIAM J. Sci . Comput.(2016), 38(3):pp. 1339–1357.
, [*Long range scattering and modified wave operators for the Maxwell-Schrödinger system I. The case of vanishing asymptotic magnetic field*]{}, Commun. Math. Phys.(2003), 236: pp. 395–448.
,[*Finite Element Methods for Navier-Stokes Equations*]{}, Springer-Verlag, Berlin, 1986.
, [*Global finite-energy solutions of the Maxwell-Schrödinger system*]{}, Commun. Math. Phys. (1995), 170: pp. 181-196.
,[*Linear and Quasilinear Equations of Parabolic Type*]{}, American Mathematical Society, 1968
, [*A numerical Maxwell-Schrödinger model for intense laser-matter interaction and propagation*]{}, Comput. Phys. Comm. (2007), 177(12): pp. 908–932.
, [*Efficient and accurate numerical modeling of a micro-macro nonlinear optics model for intense and short laser pulses*]{}, J. Comput. Sci. (2012), 3(3): pp. 159–168.
, [*A linearized Crank-Nicolson-Galerkin method for the Ginzburg-Landau model*]{}, SIAM J. Sci. Comput. (1997), 18: pp. 1028–1039.
, [*An alternating Crank-Nicolson method for decoupling the Ginzburg- Landau equations*]{}, SIAM J. Numer. Anal. (1998), 35(5): pp. 1740–1761.
, [*The Cauchy problem for the coupled Maxwell-Schródinger equations*]{}, J. Math. Phys. (1986), 27(1): pp. 211–216.
, Mathematische Annalen (2005), 332(3): pp. 565–604.
, [*Global existence and uniqueness of solutions to the Maxwell-Schrödinger equations*]{}, Commun. Math. Phys. (2007), 276(2): pp. 315–339.
, [*Coupled analysis of Maxwell-schrödinger equations by using the length gauge: harmonic model of a nanoplate subjected to a 2D electromagnetic field*]{}, Intern. J. of Numer. Model.: Electronic Networks, Devices and Fields (2013), 26(6): pp. 533–544.
, [*A new 3-D transmission line matrix scheme for the combined Schrödinger-Maxwell problem in the electronic/electromagnetic characterization of nanodevices*]{}, IEEE Transactions on Microwave Theory and Techniques (2008), 56(3): pp. 654–.
, [*Finite-difference time-domain simulation of the Maxwell-Schrödinger system*]{}, M.S. Thesis, 2015.
, [*Maxwell+ TDDFT multi-scale simulation for laser-matter interactions*]{}, Adv. Simul. Sci. Engng. (2014), 1: pp. 98–110.
, [*Modified wave operator for Maxwell-Schrödinger equations in three space dimensions*]{}, Ann. Henri Poincaré (2003), 4: pp. 661–683.
, [*Compact sets in the space $L^p(0,T; B)$.*]{}, Ann. Mat. Pura. Appl.(1987), 146: pp. 65-96.
, [*Including quantum effects in electromagnetic system for FDTD solution to Maxwell-Schrödinger equations,*]{} Microwave Symposium, IEEE/MTT-S International, pp. 1979-1982, 2007.
, [*Global existence and asymptotic behavior of solutions for the Maxwell-Schrödinger equations in three space dimensions*]{}, Commun. Math. Phys. (1993), 151(3): pp. 543–576.
, [*A FDTD solution to the Maxwell-Schrödinger coupled model at the microwave range*]{}, Electromagnetics in Advanced Applications (ICEAA), 2012 International Conference on. IEEE, 2012: pp. 363–366.
, J. Functional Analysis (2012), 263: pp. 1-24.
[^1]: [Institute of Computational Mathematics and Scientific/Engineering Computing, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China; ([[email protected]]{}).]{}
[^2]: [Corresponding author. LSEC, NCMIS, Institute of Computational Mathematics and Scientific/Engineering Computing, Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China; ([[email protected]]{}).]{}
[^3]: [This work is supported by National Natural Science Foundation of China (grant 11571353, 91330202), and Project supported by the Funds for Creative Research Group of China (grant 11321061).]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We develop a microscopic theory for the relaxation dynamics of an optically pumped two-level system (TLS) coupled to a bath of weakly interacting Bose gas. Using Keldysh formalism and diagrammatic perturbation theory, expressions for the relaxation times of the TLS Rabi oscillations are derived when the boson bath is in the normal state and the Bose-Einstein condensate (BEC) state. We apply our general theory to consider an irradiated quantum dot coupled with a boson bath consisting of a two-dimensional dipolar exciton gas. When the bath is in the BEC regime, relaxation of the Rabi oscillations is due to both condensate and non-condensate fractions of the bath bosons for weak TLS-light coupling and dominantly due to the non-condensate fraction for strong TLS-light coupling. Our theory also shows that a phase transition of the bath from the normal to the BEC state strongly influences the relaxation rate of the TLS Rabi oscillations. The TLS relaxation rate is approximately independent of the pump field frequency and monotonically dependent on the field strength when the bath is in the low-temperature regime of the normal phase. Phase transition of the dipolar exciton gas leads to a non-monotonic dependence of the TLS relaxation rate on both the pump field frequency and field strength, providing a characteristic signature for the detection of BEC phase transition of the coupled dipolar exciton gas.'
author:
- 'Vadim M. Kovalev$^{1,2}$'
- 'Wang-Kong Tse$^{3,4}$'
title: |
Relaxation of Radiation-Driven Two-Level Systems Interacting\
with a Bose-Einstein Condensate Bath
---
\[sec:level1\]Introduction
==========================
The dynamics of a quantum two-level system (TLS) is a topic of fundamental importance. Its sustained influence is evident in the continual interest in the dynamics of spin or pseudospin systems ranging from quantum optics [@QO_ref_1; @QO_ref_2] to quantum information [@QI_ref_1; @QI_ref_2; @QI_ref_3]. The spin-boson model [@2_Weiss] captures the interaction between the TLS and its environment by a spin $1/2$ degree of freedom coupled linearly to an oscillator bath [@3_Makhlin]. Despite the simplicity of such a model, it exhibits a rich variety of behavior and describes a diverse array of physical systems and phenomena [@1_Leggett; @2_Weiss]. One of the quantum systems that is well described by a TLS is the quantum dot (QD). Fueled by interests in quantum information processing, coherent optical control of quantum dots has seen substantial development in the past decade [@QD_review1]. New functionalities or tuning capabilities can be achieved with hybrid systems by further coupling QDs to other materials, such as nano-sized cavity [@QD_cavity], graphene [@QD_graphene], and superconductor [@QD_SC].
Hybrid quantum systems comprising a fermion gas coupled to a boson gas constitute the condensed matter analogue of $^3$He-$^4$He mixtures. In systems where an electron gas is coupled with excitons or exciton-polaritons, it was recently predicted that the transition of the excitonic subsystem to the Bose Einstein condensate (BEC) phase strongly modifies the properties of the electronic subsystem, resulting in polariton-mediated superconductivity and supersolidity [@6_RefImamogluPRB2016; @7_RefShelykhPRL2010; @8_RefShelykhPRL1051404022010; @8_1Shelykh2012; @9_RefMatuszewskiPRL1080604012012]. The topic of phase transition of the exciton or exciton-polariton Bose system into the BEC state is itself an intriguing topic that has garnered much attention [@4_1Butov; @4_2Butov; @4_3Butov; @5_Timofeev; @5_1Timofeev]. For dipolar exciton systems realized in GaAs double quantum well (DQW) structures, the critical temperature to reach the condensate phase is about $3-5\,\mathrm{K}$. Recent works have demonstrated that double-layer structures based on transition metal dichalcogenides (TMD) monolayers [@10_3butovAPL2016; @10_4Rivera2015] can further push the transition temperature to $\sim
10-30\,\mathrm{K}$ [@10_2butovNature2014; @10_5Wu2015; @10_6Berman2016].
Motivated by recent interest in hybrid fermion-boson systems and exciton-polariton physics mentioned above, in this paper we consider a radiation-driven quantum dot coupled to a dipolar exciton gas and study the influence of the latter’s BEC phase transition on the dynamics and relaxation of the QD states. The problem of TLS relaxation coupled to a fermionic bath transitioning to a superconducting state was studied in the context of metallic glasses [@10_6.1BlackFulde1979]; however, the question of TLS relaxation coupled to a *bosonic* bath transitioning to a BEC state has not, up to the authors’ knowledge, been considered before. It is noteworthy to mention that our current work is closely connected to the problem of a mobile impurity moving in a BEC [@11_Mobile_impurity], since the renormalization of physical properties of a moving electron that strongly interacts with the surrounding medium (polaron problem) can be described by a quantum particle coupled with a bath [@2_Weiss]. Our theory consists of a TLS modeling the ground and lowest excited states of the QD, which is coupled [@10_9.1Melik1978; @10_9.2Weisskopf1930; @10_9.3Gordon1963] to a bath of weakly-interacting Bose gas modeling the dipolar exciton system. In contrast to the simple spin-boson model, the interaction between the QD and the 2D dipolar exciton gas in our system is described by a *nonlinear* coupling Hamiltonian. We take the Bose gas to be weakly interacting, exhibiting a normal phase as well as a BEC phase described by the Bogoliubov model [@10_10Beliaev1958; @10_11PitaevskiiStringari2003; @10_10.1Griffin1998]. Our results demonstrate that the damping of the Rabi oscillations of the TLS is highly sensitive to the phase transition of the bosonic bath. The rest of our paper is organized as follows. The second section is devoted to the development of general theory for the relaxation rate of an illuminated TLS coupled to a bosonic bath. We then apply our general results to the situation of a QD coupled with a dipolar exciton gas in the third section. Finally, in the fourth section we present numerical results of the relaxation rates and discuss their behavior as a function of the optical pump field’s parameters. In the Appendix we present details of our calculations.
General theory
==============
Driven TLS and Rabi oscillations
--------------------------------
First, we consider dynamics of isolated TLS system under strong external electromagnetic field and describe the system’s response using the non-equilibrium Keldysh Green function technique. The TLS Hamiltonian is given by $$\label{eq1}
%\begin{bmatrix}
\hat{\tilde{H}}_0(t)=\left(
\begin{array}{cc}
\Delta & \lambda e^{-i\omega t} \\
\lambda^\ast e^{i\omega t} & -\Delta \\
\end{array}
\right),$$ where $\pm\Delta$ are the energies of the upper and lower states of the TLS, and quantities with an overhead caret ($\,\hat{}\,$) symbol denotes a matrix quantity. The interaction Hamiltonian with the electromagnetic field is written here in the Rotating Wave Approximation (RWA). $\lambda$ is the interaction matrix element and $\omega$ the frequency of the electromagnetic field. It is also assumed in Eq. (\[eq1\]) that the wavelength of the electromagnetic field is much larger than the geometrical size of the TLS so that the field is uniform on our scale of interest. In this work, we denote quantities in the laboratory frame and the rotating frame, respectively, with and without an overhead tilde. The dynamics of the TLS is described by the time-ordered Green’s function $\hat{\tilde{G}}_0(t,t')$ satisfying the equation of motion $$\begin{gathered}
\label{eq2}
\left(
\begin{array}{cc}
i\partial_t-\Delta & -\lambda e^{-i\omega t} \\
-\lambda^\ast e^{i\omega t} & i\partial_t+\Delta \\
\end{array}
\right)\hat{\tilde{G}}_0(t,t')=\delta(t-t').\end{gathered}$$ To remove the explicit time dependence, it is convenient to transform this equation to the rotating frame using the unitary transformation $$\begin{gathered}
\label{eq3}
\hat{S}=\left(
\begin{array}{cc}
e^{i\omega t/2} & 0 \\
0 & e^{-i\omega t/2} \\
\end{array}
\right),\,\,\,\hat{S}^{-1}=\left(
\begin{array}{cc}
e^{-i\omega t/2} & 0 \\
0 & e^{i\omega t/2} \\
\end{array}
\right).\end{gathered}$$ As a result, the Green’s function in the rotating frame, $\hat{G}_0(t,t')=\hat{S}(t)\hat{\tilde{G}}_0(t,t')\hat{S}^{-1}(t')$, is described by the equation $$\begin{gathered}
\label{eq4}
\left(
\begin{array}{cc}
i\partial_t-\varepsilon_0 & -\lambda \\
-\lambda^\ast & i\partial_t+\varepsilon_0 \\
\end{array}
\right)\hat{G}_0(t,t')=\delta(t-t'),\end{gathered}$$ where $\varepsilon_0=\Delta-{\omega}/{2}$. To find the self-energies and lifetimes, we need the retarded and the lesser components of the non-equilbrium Green’s function. The retarded Green’s function is derived as (see Appendix A), $$%\bar{G}^{R}_{ij}(\varepsilon)
%=\frac{A_{ij}}{\varepsilon-\Omega+i\delta}
% +\frac{B_{ij}}{\varepsilon+\Omega+i\delta},\\\nonumber
\hat{G}_0^{R}(\varepsilon)
=\frac{\hat{A}}{\varepsilon-\Omega+i\delta}
+\frac{\hat{B}}{\varepsilon+\Omega+i\delta},\label{eq5}$$ where $\Omega=\sqrt{\varepsilon_0^2+|\lambda|^2}$ is the Rabi frequency, and $\hat{A}$ and $\hat{B}$ are matrices defined as $$\hat{A}=\left(
\begin{array}{cc}
u^2 & uv \\
u^\ast v^\ast & v^2 \\
\end{array}
\right),\,\,
\hat{B}=\left(
\begin{array}{cc}
v^2 & -uv \\
-u^\ast v^\ast & u^2 \\
\end{array}
\right),\label{eq5_1}$$ with $u^2=\left(1+{\varepsilon_0}/{\Omega}\right)/2$, $v^2=\left(1-{\varepsilon_0}/{\Omega}\right)/2$ and $uv={\lambda}/{2\Omega}$. Eqs. (\[eq5\])-(\[eq5\_1\]) imply that new quasiparticles emerge from the light-matter coupling that renormalizes the original TLS states into dressed states with energies $\pm \Omega$. $\hat{A}$ and $\hat{B}$ are the projection operators to these dressed states.
It is instructive to recover the result for Rabi oscillations using the above retarded Green’s function Eq. (\[eq5\]). The TLS wave function at a latter time $t$ is obtained by propagating the initial time ($t = 0$) wave function, $$\begin{aligned}
\psi_i(t) &=&[\hat{S}^{-1}(t)\hat{G}^R(t)\hat{S}(0)]_{ij}\psi_j(0)
\nonumber \\
&=& \left[\hat{S}^{-1}(t)\int\frac{d\varepsilon}{2\pi}\hat{G}^R(\varepsilon)e^{-i\varepsilon
t}\right]_{ij}\psi_j(0), \label{eq6}\end{aligned}$$ where $\psi_i$ are the wave functions of the TLS upper $(i=1)$ and lower $(i=2)$ levels. Here $\hat{S}^{-1}(t)\hat{G}^R(t)\hat{S}(0) \equiv
\hat{\tilde{G}}^R(t,0)$ is the retarded Green’s function in the laboratory frame. When only the lower level is initially occupied, $\psi_2(0)=1$ and $\psi_1(0)=0$. Using Eqs.(\[eq3\])-(\[eq5\_1\]), the transition probability to the upper level is then given by $$\begin{aligned}
|\langle\psi_2^+(0)\psi_1(t)\rangle|^2&=&\left|\frac{\lambda}{\Omega}\sin(\Omega
t)e^{-i\omega t/2}\right|^2 \nonumber \\
&=&\frac{|\lambda|^2}{2\Omega^2}\left(1-\cos\Omega t\right), \label{eq7}
%|\langle\psi_2^+(0)\psi_1(t)\rangle|^2=\left|\frac{\lambda}{\Omega}\sin(\Omega t)e^{-i\omega t/2}\right|^2=\frac{|\lambda|^2}{2\Omega^2}\left(1-\cos\Omega t\right),\end{aligned}$$ which is the Rabi oscillations [@12_LL]. The lesser Green’s function can be expressed in terms of the distribution functions $n_{\pm \Omega}$ of the upper and lower dressed states as $$\begin{aligned}
\hat{G}^{<}(\varepsilon) &=&
-n_{\varepsilon}\left[\hat{G}^{R}(\varepsilon)-\hat{G}^{A}(\varepsilon)\right]
\nonumber \\
&=& 2\pi i
n_{\varepsilon}\left[\hat{A}\delta(\varepsilon-\Omega)+\hat{B}\delta(\varepsilon+\Omega)\right]
\nonumber \\
&=& 2\pi i
\left[\hat{A}n_{\Omega}\delta(\varepsilon-\Omega)+\hat{B}n_{-\Omega}\delta(\varepsilon+\Omega)\right]. \label{Glesser1}\end{aligned}$$ Note that $n_{\Omega}+n_{-\Omega} = 1$. Assuming the radiation is turned on adiabatically, we can obtain $n_{\pm \Omega}$ in the following. The density matrix $\hat{f}$ in the original basis of TLS upper and lower levels satisfies the kinetic equation (see Appendix A): $$\begin{aligned}
\label{KE}
%\begin{equation}
&&\frac{\partial \hat{f}}{\partial t} +i[\hat{H}_0,\hat{f}] = 0, \label{KE} \\
&&\hat{f} = \left(\begin{array}{cc}
f_{11} & f_{12} \\
f_{21} & f_{22} \\
\end{array}
\right), \label{distfcn}
%\end{equation}\end{aligned}$$ where the subscripts $1,2$ respectively denotes the original (*i.e.*, unrenormalized by light) upper and lower levels of the TLS, and $\hat{H}_0 = \hat{S}(t)\hat{\tilde{H}}_0(t,0)\hat{S}^{-1}(0)$ is the Hamiltonian in the rotating frame. Writing the Hamiltonian as $\hat{H}_0 =
\hat{\bm{\sigma}}\cdot {\bm{B}}_0/2$, we can define an effective magnetic field ${\bm{B}}_0 =2\lambda_R{\bm{e}_x}-2\lambda_I{\bm{e}_y}+
(2\Delta-\omega){\bm{e}_z}$ that drives the TLS pseudospin degrees of freedom, where $\lambda_{R,I}$ are the real and imaginary parts of $\lambda$ and $\bm{e}_{x,y,z}$ the unit vectors along the $x,y,z$ directions. Then, decomposing the density matrix $\hat{f}$ as $\hat{f} =
C+\bm{S}\cdot\hat{\bm{\sigma}}/2$, the kinetic equation can be written as a Bloch equation: $$\frac{\partial \bm{S}}{\partial t}+\bm{S}\times {\bm{B}}_0=
0. \label{Bloch}$$ From the definition of $\hat{f}$ in Eq. (\[distfcn\]), we can relate the distribution functions in the two representations as $S_{z} = f_{11}-f_{22}$ and $S^{(+)} \equiv S_{x}+iS_{y} = 2f_{12}^* = 2f_{21}$. With the laser field switched on adiabatically, the optical response follows adiabatically the driving field and is therefore stationary in the rotating frame, *i.e.*, $\partial/\partial t =
0$. Before laser is turned on, the TLS initial state is in the lower level so that $\bm{S}(t =
0) = -{\bm{e}_z}$. Since $\vert \bm{S} \vert$ is a constant of motion, this implies that $\vert \bm{S}(t) \vert = 1$ for all times $t$. Here we focus on the regime without population inversion, so that $S_{z} = f_{11}-f_{22}$ is always less than zero. We obtain $\bm{S}$ as $$\begin{aligned}
S^{(+)} &=& -\frac{\mathrm{sgn}(2\Delta-\omega)2\lambda^*}{\sqrt{(2\Delta-\omega)^2+4\lambda^2}}, \\
S_{z} &=& -\frac{\vert 2\Delta-\omega \vert}{\sqrt{(2\Delta-\omega)^2+4\lambda^2}}.\end{aligned}$$ Using $f_{11}+f_{22} =1$, we also find the density matrix $\hat{f}$ in the original basis of the TLS upper and lower levels: $$\begin{aligned}
f_{11} &=& \frac{1}{2}\left[1-\frac{\vert 2\Delta-\omega
\vert}{\sqrt{(2\Delta-\omega)^2+4\lambda^2}}\right], \label{fpp} \\
f_{12} &=&
-\frac{\mathrm{sgn}(2\Delta-\omega)2\lambda}{\sqrt{(2\Delta-\omega)^2+4\lambda^2}}. \label{fpm}\end{aligned}$$ On the other hand, Eq. (\[Glesser1\]) gives the density matrix in the basis of the dressed quasiparticles as follows $$\begin{aligned}
\hat{f}(t) &=& -i\hat{G}^{<}(t,t) =
-i\int\frac{d\varepsilon}{2\pi}\hat{G}^{<}(\varepsilon), \nonumber \\
&=& \hat{A}n_{\Omega}+\hat{B}n_{-\Omega}.
\label{Glesser2}\end{aligned}$$ We can determine $n_{\Omega}$ by comparing the expressions of the density matrix in Eq. (\[Glesser2\]) and Eq. (\[fpp\]). The $11$ element, for instance, gives $f_{11} = u^2
n_{\Omega}+v^2(1-n_{\Omega})$ or $n_{\Omega} =
({f_{11}-v^2})/({u^2-v^2})$, from which we determine $$%&=& %\frac{f_{11}-v^2}{u^2-v^2} \nonumber \\
n_{\pm \Omega} = \frac{1}{2}\left[1\mp \mathrm{sgn}\left(\varepsilon_0\right)\right]. \label{nOmega}
% \frac{1}{2}\left[1-\mathrm{sgn}\left(\Delta-\omega/2\right)\right]. \label{nOmega}$$ It follows that $n_{\Omega}$ takes on the values $0$ or $1$ depending on whether the light frequency $\omega$ is smaller or larger, respectively, than the energy difference $2\Delta$ of the TLS.
Coupling to bosonic bath and TLS self-energies
----------------------------------------------
In the absence of bath coupling, the TLS is described by bare Green’s functions with a vanishing level broadening. To focus on the effects of bosonic bath coupling, we ignore the effects of spontaneous and stimulated emission due to electrons’ coupling to light. The only damping effects on the TLS dynamics, once coupled to the bosonic bath, will be due to interlevel transitions caused by absorption or emission of the bosons. To analyze the TLS-bath coupling, we add to the bare TLS Hamiltonian Eq. (\[eq1\]) the TLS interaction term with the bath and the bath Hamiltonian $$\begin{gathered}
\label{eq8}
\left(
\begin{array}{cc}
W_{11}[\varphi] & 0 \\
0 & W_{22}[\varphi] \\
\end{array}
\right)+\hat{H}_{bath}[\varphi].\end{gathered}$$ Here the first term is TLS-bath coupling Hamiltonian where the matrix elements describe the interaction of the upper and lower levels with the bath bosons. Both terms in Eq. (\[eq8\]) are functionals of the quantum field $\varphi$, which describes the dynamics of bath degrees of freedom. The structure of the bath Hamiltonian depends on whether it is in the normal or Bose-condensed phase and will be specified later on. We assume short-range interaction between the TLS and the bosonic bath $$\begin{gathered}
\label{eq9}
W_{ii}[\varphi]=g_i\int d\textbf{r}|\psi_{i}(\textbf{r})|^2|\varphi(\textbf{r},t)|^2,\end{gathered}$$ with the coupling constant $g_i$. The form of this interaction contains $\varphi^2$ and is markedly different from the conventional coupling to a phonon-like bath, which is linear in $\varphi$ [@2_Weiss]. Anticipating further application of our general theory to consider a QD interacting with a 2D exciton gas, we note that $W_{ii}$ in Eq. (\[eq9\]) takes into account the most important direct contribution to the Coulomb interaction energy between electrons in the QD and the 2D exciton gas.
To elucidate the influence of the bosonic bath on the TLS dynamics, we use the diagrammatic perturbation theory. Within this approach the retarded Green function can be found from the Dyson equation, $$\label{eq10}
\hat{G}^R=\hat{G}_0^{R}+\hat{G}_0^{R}\hat{\Sigma}^R\hat{G}^{R},
%\bar{G}^R_{ij}=\bar{G}^{0R}_{ij}+\bar{G}^{0R}_{im}\bar{\Sigma}^R_{mn}\bar{G}^{R}_{nj},$$ where the bare Green’s function $\hat{G}_0^{R}$ is given by Eq. (\[eq5\]). The solution of Eq. (\[eq10\]) gives the interacting Green’s function as follows $$\hat{G}^R =
\frac{1}{\Lambda}\left[{\hat{G}_{0}^{R}-\det(\hat{G}_0^{R})\sigma_y
(\hat{\Sigma}^R)^{\mathrm{T}}\sigma_y}\right], \label{eq11}$$ where $\Lambda =
1-\mathrm{Tr}(\hat{G}_0^{R}\hat{\Sigma}^R)+\det(\hat{G}_0^{R}\hat{\Sigma}^R)$, $(\hat{\Sigma}^R)^{\mathrm{T}}$ is the matrix transpose of $\hat{\Sigma}^R$, and $\sigma_y$ is the Pauli matrix. We arrive at the following form of the Green’s function (see Appendix B) $$\begin{aligned}
\hat{G}^R(\epsilon) &\approx&
%\frac{\hat{G}_0^{R}(\epsilon)}{1-\mathrm{Tr}\left[\hat{G}_0^{R}(\epsilon)\hat{\Sigma}^R(\epsilon)\right]}
% \nonumber \\ &=&
\frac{\hat{A}}{\epsilon-\Omega+i/2\tau_u}
+\frac{\hat{B}}{\epsilon+\Omega+i/2\tau_l}, \label{eq12}
%\bar{G}^R_{ij}(\epsilon)=\frac{\bar{G}^{0R}_{ij}(\epsilon)}{1-Tr\left[\bar{G}^{0R}(\epsilon)\bar{\Sigma}^R(\epsilon)\right]}\approx\\
%\approx\frac{A_{ij}}{\epsilon-\Omega+i/2\tau_u}
%+\frac{B_{ij}}{\epsilon+\Omega+i/2\tau_l},\end{aligned}$$ where we have introduced the relaxation times $\tau_{u,l}$ of the upper (subscript $u$) and lower (subscript $l$) dressed quasiparticles
$$\begin{aligned}
\frac{1}{2\tau_u} &=&-\textmd{Im} \,
\mathrm{Tr}\left[\hat{A}\hat{\Sigma}^R(\Omega)\right],\label{eq13_1} \\
\frac{1}{2\tau_l} &=&-\textmd{Im} \,
\mathrm{Tr}\left[\hat{B}\hat{\Sigma}^R(-\Omega)\right], \label{eq13_2}\end{aligned}$$
and neglected the shift of the levels due to the real part of self-energy. To obtain expressions of the relaxation times, we apply the diagrammatic perturbation theory in the leading order of the TLS-bath interaction potential and account for the lowest-order non-vanishing diagrams for the self-energy. The explicit form of such diagrams depends on whether the bath is in the normal or the condensate state. First we consider the normal state.
### Bath in the normal state
![Feynman diagram for TLS self-energy when the bath is in the normal phase. Double blue line represents the TLS Green function, dashed lines the TLS electron-bath interaction, and solid black lines the Green functions of the normal-state bath particles.[]{data-label="Figure1"}](figure1.pdf){width="4.5cm"}
In the normal state of the bath, we assume that the bosons are non-interacting with a kinetic energy $E_{\textbf{p}}=\vert\textbf{p}\vert^2/2m \equiv p^2/2m$ and chemical potential $\mu$. To lowest order in the TLS-bath interaction, the self-energy diagram is shown in Fig. \[Figure1\]. Detailed calculation of the self-energy is presented in the Appendix C. The result reads
$$\begin{aligned}
\label{eq14}
\hat{\Sigma}^R(\omega) &=& \sum_{\textbf{k},\textbf{p}}\left[
\hat{M}_{\textbf{k}}\hat{A}\hat{M}_{-\textbf{k}}\frac{(1-n_\Omega)[n_B(\xi_{\textbf{p}})-n_B(\xi_{\textbf{p}+\textbf{k}})]+n_B(\xi_{\textbf{p}+\textbf{k}})
[1+n_B(\xi_{\textbf{p}})]}{\omega-\Omega+E_{\textbf{p}}-E_{\textbf{p}+\textbf{k}}+i\delta}\right.
\nonumber \\
&&\left.+\hat{M}_{\textbf{k}}\hat{B}\hat{M}_{-\textbf{k}}\frac{(1-n_{-\Omega}) [n_B(\xi_{\textbf{p}})-n_B(\xi_{\textbf{p}+\textbf{k}})]+n_B(\xi_{\textbf{p}+\textbf{k}}) [1+n_B(\xi_{\textbf{p}})]}{\omega+\Omega+E_{\textbf{p}}-E_{\textbf{p}+\textbf{k}}+i\delta}
\right],\end{aligned}$$
where $\xi_{\textbf{p}} = E_{\textbf{p}}-\mu$ is the energy of the bath bosons rendered from the chemical potential, $n_B(\xi) =
[\exp(\xi/T)-1]^{-1}$ is the Bose distribution function, and $$\hat{M}_{\textbf{k}}=\left(
\begin{array}{cc}
g_1\int d\textbf{r}e^{i\textbf{kr}}|\psi_1(\textbf{r})|^2 & 0 \\
0 & g_2\int d\textbf{r}e^{i\textbf{kr}}|\psi_2(\textbf{r})|^2 \\
\end{array}
\right),\label{eq15}$$ Taking the imaginary part of Eq. (\[eq14\]), integrating over the angle between $\textbf{k}$ and $\textbf{p}$, and substituting the resulting expression into Eqs. (\[eq13\_1\])-(\[eq13\_2\]) we find the relaxation times $$\begin{aligned}
\frac{1}{2\tau_u} &=&\frac{m}{(2\pi)^2}\int_0^\infty
dk \left[\int_{k/2}^\infty
\frac{\alpha_kf_\Omega^u(p)pdp}{\sqrt{p^2-(k/2)^2}}\right.\label{eq15_1} \\
&&\left.+\int_{|{k}/2-{2m\Omega}/{k}|}^\infty
\frac{\beta_kf_{-\Omega}^u(p)pdp}{\sqrt{p^2-\left({k}/{2}-{2m\Omega}/{k}\right)^2}}\right],\nonumber\\
\frac{1}{2\tau_l} &=&\frac{m}{(2\pi)^2}\int_0^\infty
dk\left[\int_{k/2}^\infty
\frac{\gamma_kf_{-\Omega}^l(p)pdp}{\sqrt{p^2-(k/2)^2}}\right. \label{eq15_2} \\
&&\left.+\int_{|{k}/{2}+{2m\Omega}/{k}|}^\infty
\frac{\beta_kf_{\Omega}^l(p)pdp}{\sqrt{p^2-\left({k}/{2}+{2m\Omega}/{k}\right)^2}}\right]. \nonumber\end{aligned}$$ Here we have introduced the coefficients $$\begin{gathered}
\label{eq15_3}\nonumber
\alpha_k=\textmd{Tr}\left(\hat{A}\hat{M}_{\textbf{k}}\hat{A}\hat{M}^\ast_{\textbf{k}}\right)=\left|g_1(M_{\textbf{k}})_{11}u^2+g_2(M_{\textbf{k}})_{22}v^2\right|^2,\\\nonumber
\beta_k=\textmd{Tr}\left(\hat{B}\hat{M}_{\textbf{k}}\hat{A}\hat{M}^\ast_{\textbf{k}}\right)=|uv|^2\left|g_1(M_{\textbf{k}})_{11}-g_2(M_{\textbf{k}})_{22}\right|^2,\\
\gamma_k=\textmd{Tr}\left(\hat{B}\hat{M}_{\textbf{k}}\hat{B}\hat{M}^\ast_{\textbf{k}}\right)=\left|g_1(M_{\textbf{k}})_{11}v^2+g_2(M_{\textbf{k}})_{22}u^2\right|^2,\end{gathered}$$ and $$\begin{aligned}
%\label{eq15_4}
f^u_{\Omega}(p) &=& f^l_{-\Omega}(p)=n_B(\xi_{\textbf{p}})[1+n_B(\xi_{\textbf{p}})],\nonumber\\
f^u_{-\Omega}(p) &=& n_{\Omega}[n_B(\xi_{\textbf{p}})-n_B(\xi_{\textbf{p}}+2\Omega)]\nonumber \\
&&+n_B(\xi_{\textbf{p}}+2\Omega) [1+n_B(\xi_{\textbf{p}})],\nonumber\\
f^l_{\Omega}(p) &=& \theta(E_{\textbf{p}}-2\Omega)\left\{n_{-\Omega}[n_B(\xi_{\textbf{p}})-n_B(\xi_{\textbf{p}}-2\Omega)]\right. \nonumber \\
&&\left.+n_B(\xi_{\textbf{p}}-2\Omega) [1+n_B(\xi_{\textbf{p}})]\right\}, \label{eq15_4}
%f^u_{\Omega}(p)=f^l_{-\Omega}(p)=n_{\textbf{p}}(1+n_{\textbf{p}}),\\\nonumber
%f^u_{-\Omega}(p)=n_{\Omega}(n_{\textbf{p}}-n^+_{\textbf{p}})+n^+_{\textbf{p}}(1+n_{\textbf{p}}),\\\nonumber
%f^l_{\Omega}(p)=\theta(E_{\textbf{p}}-2\Omega)[n_{-\Omega}(n_{\textbf{p}}-n^-_{\textbf{p}})
%+n^-_{\textbf{p}}(1+n_{\textbf{p}})],\end{aligned}$$
### Bath in the condensed state
We now consider the bath to be in the Bose condensed phase and obtain the TLS self-energy. The elementary excitations of the Bose-condensed system are Bogoliubov quasi-particles. An explicit form of the dispersion law of Bogoliubov excitations depends on the model used to describe the system of interacting bosons. In the case of small particle density an appropriate theoretical model is the Bogoluibov model of weakly-interacting Bose gas. In the framework of this model, the dispersion law of elementary excitations has the form of $\epsilon_{\textbf{p}}=\sqrt{E_{\textbf{p}}\left(E_{\textbf{p}}+2g_0n_c\right)}$, where $n_c$ is particle density in the condensate and $g_0$ the strength of inter-particle interaction. In the low-energy, long-wavelength limit $E_{\textbf{p}} \ll 2g_0n_c$ the elementary excitations comprise sound quanta, with a dispersion $\epsilon_{\textbf{p}}\approx sp$ where $s=\sqrt{g_0n_c/m}$ is the sound velocity. In the Bose-condensed state, most of the particles are in the condensate but there are also noncondensate particles, due to both interaction and finite temperature effects (thermal-excited particles). All three fractions of particles contribute to relaxation times of TLS. We consider the quantum limit $T\ll sp$ when thermal excitations are not important and the theory can be developed for $T=0$. In the present dilute boson gas limit, the density of the noncondensate particles is small, and one can neglect the interaction term due to fluctuations of the condensate density and the noncondensate density. Thus, the contribution to relaxation times due to the condensate and noncondensate particles can be calculated independently.
\[h\] ![Feynman diagrams for TLS self-energy when the bath is in the BEC phase. Type (a) describes the condensate particles contribution, $\bar{\Sigma}^c$, and (b) is due to the non-condensate particles, $\bar{\Sigma}^n$. Double blue line represents the TLS Green function and dashed lines the TLS electron-bath interaction. Red lines denote the normal or anomalous Green functions of the non-condensate particles, and the zigzag lines stand for $\sqrt{n_c}$ corresponding to the condensate particles.[]{data-label="Figure2"}](figure2.pdf "fig:"){width="8.5cm"}
The self-energy diagrams in the lowest order with respect to the TLS-bath coupling are depicted in Fig. \[Figure2\]. Fig. \[Figure2\](a) corresponds to the contribution to the self-energy from the condensate particles and describes virtual processes in which a condensate particle is scattered by a TLS electron through an intermediate non-condensate state. Fig. \[Figure2\](b) corresponds to the contribution from the non-condensate particles and describes polarization of the non-condensate particles induced by the TLS electrons. These self-energy contributions due to condensate particles $\hat{\Sigma}^{c R}$ and non-condensate $\hat{\Sigma}^{nR}$ particles have the following form (see Appendix C)
$$\begin{aligned}
\hat{\Sigma}^{c R}(\omega) &=&\frac{n_c}{2ms}\sum_{\textbf{k}}k
\left[\hat{M}_{\textbf{k}}\hat{A}\hat{M}_{-\textbf{k}}\left(\frac{1-n_\Omega}{\omega-\Omega-\epsilon_{\textbf{k}}+i\delta}+\frac{n_\Omega}{\omega-\Omega+\epsilon_{\textbf{k}}+i\delta}\right)
\right.\nonumber \\
&&\left.+\hat{M}_{\textbf{k}}\hat{B}\hat{M}_{-\textbf{k}}\left(\frac{1-n_{-\Omega}}{\omega+\Omega-\epsilon_{\textbf{k}}+i\delta}+\frac{n_{-\Omega}}{\omega+\Omega+\epsilon_{\textbf{k}}+i\delta}\right)\right],\label{eq16_1}\end{aligned}$$
$$\begin{aligned}
\hat{\Sigma}^{nR}(\omega) &=&\frac{(ms^2)^2}{2}\sum_{\textbf{k},\textbf{p}}\left[\frac{\hat{M}_{\textbf{k}}\hat{A}\hat{M}_{-\textbf{k}}}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}
\left(\frac{1-n_\Omega}{\omega-\Omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}+i\delta}+\frac{n_\Omega}{\omega-\Omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right)\right. \nonumber\\
&&\left.+\frac{\hat{M}_{\textbf{k}}\hat{B}\hat{M}_{-\textbf{k}}}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\left(\frac{1-n_{-\Omega}}{\omega+\Omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}+i\delta}+\frac{n_{-\Omega}}{\omega+\Omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right)
\right]. \label{eq16_2}\end{aligned}$$
Similar steps leading to Eqs. (\[eq15\_1\])-(\[eq15\_2\]) yields $$\begin{aligned}
%\label{b10.1}
\frac{1}{2\tau^c_u}&=&\frac{1}{2\tau^c_l}
=n_{\Omega}\frac{\pi n_c}{2ms}\sum_kk\beta_k\delta(-2\Omega+\epsilon_{\textbf{k}}),\label{b10.1}\\
\frac{1}{2\tau^n_u}&=&\frac{1}{2\tau^n_l} \label{b10.2} \\
&=& n_{\Omega}\frac{\pi
(ms^2)^2}{2}\sum_{\textbf{k},\textbf{p}}\frac{\beta_k}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\delta(2\Omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}). \nonumber\end{aligned}$$ Explicit expressions for relaxation times depend on the shape of the wave functions of the TLS upper and lower states through the matrix elements Eq. (\[eq15\]). In the next section we propose and analyze an experimental setup in which explicit expressions of the relaxation time can be obtained.
Application to coupled QD-dipolar exciton bath {#Sec:QD}
==============================================
Applying our theory developed in the previous sections, we consider the nanostructure depicted in Fig. 3. A double quantum well (DQW) with closely separated electron-doped and hole doped wells realizes a 2D gas of indirect excitons (also called dipolar excitons). A self-assembled QD is positioned above the DQW and is irradiated with a frequency close to the lowest exciton energy of the QD. In the electron-hole representation of the QD, the lower quantum state describes the unexcited QD state (*i.e.* vaccum) $|vac\rangle$, while the lowest excited QD state describes the state $|eh\rangle$ of an excited electron-hole pair upon irradiation [@QD_TLS1; @QD_TLS2; @QD_TLS3]. With a shift in energy that does not affect the system’s dynamics, we can assume that these two QD states correspond exactly to the lower and upper TLS levels in Eq. (\[eq1\]) with $\Delta=(E_e+E_h)/2$, where $E_{e,h}$ are the lowest energies of electrons and holes in the QD (corresponding to the energies at the conduction and valence band edges), and $\lambda$ represents the dipole matrix element weighted by the electron-hole pair envelope wave function integrated over all space.
\[h\] ![Schematic of the coupled QD-dipolar excitons system. The quantum dot is positioned through a dielectric spacer on top of a double quantum well that hosts the dipolar exciton gas.[]{data-label="Figure3"}](figure3_2.pdf "fig:"){width="8.5cm"}
Let us discuss the exciton gas model we use for our calculations. The indirect excitons have a dipole moment $\textbf{p}$ in the out-of-plane direction of the DQW. The excitons are modeled as rigid dipole molecules that are free to move on the DQW plane as described by the center-of-mass motion of the dipoles, a valid assumption as long as the dipole’s internal degrees of freedom are not excited. The exciton density $n_{\mathrm{ex}}$ is assumed to be small so that $n_{\mathrm{ex}}a_B^2\ll 1$, where $a_B$ is an exciton Bohr radius. For simplicity, we assume that there is no particle tunneling between the QD and the DQW so that the exchange contribution to the interaction potential is negligible. This can be guaranteed using a large-band gap dielectric spacer as a substrate between the QD and the DQW. The direct contribution to the interaction potential between the electron-hole pair in the QD and the excitons in the DQW is given by $$\begin{gathered}
\label{eq17}
%\label{interactionpotential}\nonumber
U=\int d\textbf{r}\int d\textbf{r}'\left(|\chi_h(\textbf{r})|^2-|\chi_e(\textbf{r})|^2\right)|\varphi(\textbf{r}')|^2u(\textbf{r}-\textbf{r}'),\\
u(\textbf{r}-\textbf{r}')=\frac{e^2}{\varepsilon\sqrt{(\textbf{r}-\textbf{r}')^2+a^2}}-\frac{e^2}{\varepsilon\sqrt{(\textbf{r}-\textbf{r}')^2+(a+d)^2}},\end{gathered}$$ where $\chi_{e,h}(\textbf{r})$ are the electron (e) and hole (h) wave functions in the QD, $\varphi(\textbf{r})$ is the QW exciton center-of-mass wave function, $\varepsilon$ is an effective dielectric constant taking account of the dielectric environment of the spacer and the DQW, $a$ and $d$ are the distance of the QD to the DQW surface and the separation between the positive charges and negative charges of the dipole layer (see Fig. \[Figure3\]). The Fourier transform of $u(\textbf{r})$ yields $$\begin{gathered}
\label{eq18}
u(\textbf{k})=\frac{2\pi e^2}{\varepsilon k}\left(1-e^{-kd}\right)e^{-ka}.\end{gathered}$$ Typical values of the wave vector $k$ here are determined by the bath excitons of DQW. At temperatures above the condensation temperature $T>T_c$, the bath in the normal state, the DQW exciton energy is of the order of $T$ and thus the wave vector $k\sim p_T=\sqrt{2mT}$, where $m$ is the exciton mass. At $T=0$, the bath is in the condensed phase, the typical value of the bath excitation momentum is of the order of $p\leq ms$. Hereafter we assume the distance of the QD from the DQW as well as the inter-well distance in the DQW to be sufficiently small so that $p_Ta, p_Td \ll 1$ and $(ms)a, (ms)d \ll
1$. This allows us to simplify the expression Eq. (\[eq18\]) assuming that $kd,ka\ll 1$. Thus we have $$u(\textbf{k})\equiv U_0=\frac{2\pi e^2d}{\varepsilon},
\,\,\,u(\textbf{r})=U_0\delta(\textbf{r}). \label{eq19}$$ Under this approximation the interaction Eq. (\[eq17\]) becomes contact-like, $$U=U_0\int d\textbf{r}\left(|\chi_h(\textbf{r})|^2-|\chi_e(\textbf{r})|^2\right)|\varphi(\textbf{r})|^2. \label{eq20}$$ In our coupled QD-exciton gas system, since the lower level is the vacuum, we have in Eqs.(\[eq8\]) and (\[eq9\]) that the coupling constants $g_1=U_0,\,\,g_2=0$ and their respective matrix elements $W_{11}=U,\,\,W_{22}=0$. The QD is described by a two-dimensional system of electrons and holes confined by a parabolic potential. For strongly confined QDs where the QD size $L$ is small compared to the Bohr radius of the electron-hole pair, the lowest-energy electrons and holes are characterized by the wavefunctions [@QD_TLS2] $\chi_i(\textbf{r})=\exp(-\rho^2/2a^2_i)/({a_i\sqrt{\pi}})$, where $a_i$ is an electron $i=e$ or a hole $i=h$ characteristic length determined by the QD confinement potential. Using these wavefunctions, we find the matrix element $(\hat{M}_{\textbf{k}})_{11}=\mathrm{exp}[{-(ka_h/2)^2}]-\mathrm{exp}[{-(ka_e/2)^2}]$.
Relaxation times for normal-state bath
--------------------------------------
With $ka_i \ll 1$, we perform the integration over $k$ in Eqs. (\[eq15\_1\])-(\[eq15\_2\]) and obtain the following expressions for the relaxation times $$\begin{aligned}
\frac{1}{2\tau_u}&=&\frac{m\alpha_0}{(2\pi)^2}\int_0^\infty
F_+(p,0)n_B(\xi_{\textbf{p}})[1+n_B(\xi_{\textbf{p}})]pdp\nonumber
\\
&&+\frac{m\beta_0}{(2\pi)^2}\int_0^\infty
F_+(p,p_0)f^u_{-\Omega}(p)pdp, \label{eq21_1} \\
\frac{1}{2\tau_l}&=&\frac{m\gamma_0}{(2\pi)^2}\int_0^\infty F_-(p,0)n_B(\xi_{\textbf{p}})[1+n_B(\xi_{\textbf{p}})]pdp\nonumber\\
&&+\frac{m\beta_0}{(2\pi)^2}\int_0^\infty
F_-(p,p_0)f^l_{\Omega}(p)pdp, \label{eq21_2}
%\frac{1}{2\tau_u}&=&\frac{m\alpha_0}{(2\pi)^2}\int_0^\infty F_+(p,0)n_{\textbf{p}}(1+n_{\textbf{p}})pdp+\\\nonumber
%&&+\frac{m\beta_0}{(2\pi)^2}\int_0^\infty
% F_+(p,p_0)f^u_{-\Omega}(p)pdp, \label{eq21_1} \\
%\frac{1}{2\tau_l}&=&\frac{m\gamma_0}{(2\pi)^2}\int_0^\infty F_-(p,0)n_{\textbf{p}}(1+n_{\textbf{p}})pdp+\\\nonumber
%&&+\frac{m\beta_0}{(2\pi)^2}\int_0^\infty
% F_-(p,p_0)f^l_{\Omega}(p)pdp, \label{eq21_2}\end{aligned}$$ where $F_\pm(p,p_0)=6\pi(p^4\pm p^2p_0^2+p_0^4/6)\theta(p^2\pm
p_0^2)$ with $p_0^2=4m\Omega$, and $$\begin{aligned}
\alpha_0 &=& U_0^2(a_h^2-a_e^2)^2u^4/16,\nonumber \\
\beta_0 &=& U_0^2(a_h^2-a_e^2)^2|uv|^2/16,\nonumber \\
\gamma_0 &=& U_0^2(a_h^2-a_e^2)^2v^4/16. \label{eq21_3}\end{aligned}$$ To simplify Eqs. (\[eq21\_1\])-(\[eq21\_2\]) we consider two limiting cases. At large temperatures $T \gg\Omega$ \[while still small compared to $\hbar^2/(2m d^2)$\], we can keep up to the zeroth order in $\Omega/T$ with $F_{\pm}(p,p_0) \approx 6\pi p^4$ and $f^u_{-\Omega}(p)\approx f^l_{\Omega}(p)\approx
n_B(\xi_{\textbf{p}})[1+n_B(\xi_{\textbf{p}})]$ in Eqs. (\[eq21\_1\])-(\[eq21\_2\]). The relaxation rates for $\Omega\ll T$ then read $$\frac{1}{\tau_{u,l}}
=\frac{1}{\tau_T}\left(1\pm \frac{\varepsilon_0}{\Omega}\right)\mathrm{Li}_2\left(e^{\mu/T}\right),
%\int_0^\infty\frac{xdx}{e^{x-(\mu/T)}-1},
\label{eq22}$$ where the $+,-$ signs apply for $\tau_{u,l}$ respectively, and $$\frac{1}{\tau_T} =
\frac{3 U_0^2 m^4T^3(a_h^2-a_e^2)^2}{4\pi\hbar^9}, \label{eq22_1}$$ and $\mathrm{Li}_2(x)$ is the polylogarithm function of order $2$. In the opposite limit of low temperatures $T \ll \Omega$, we keep up to the zeroth order in $T/\Omega$. Then the respective first terms in Eqs. (\[eq21\_1\])-(\[eq21\_2\]) drop out, and we have $f^u_{-\Omega}(p)\approx n_\Omega n_B(\xi_{\textbf{p}})$ and $f^l_{\Omega}(p)\approx n_\Omega \theta(E_{\textbf{p}}-2\Omega)
n_B(\xi_{\textbf{p}}-2\Omega)$ in the second terms. The result is $$\frac{1}{\tau_{u}} = \frac{1}{\tau_{l}}
=\frac{1}{\tau_T}n_{\Omega}\frac{\pi^2}{3}\left(\frac{\lambda}{T}\right)^2\frac{\hbar^2
n_{\mathrm{ex}}}{mT}, \label{eq23}$$ where $n_{\mathrm{ex}}$ is the exciton density in the bath. We have restored $\hbar$ in the expressions for the relaxation rates here and in the following section.
Relaxation times for Bose-condensed bath
----------------------------------------
Straightforward but cumbersome integration in Eqs. (\[b10.1\])-(\[b10.2\]) results in the following expressions for relaxation times due to the condensate and non-condensate particles in the bath $$%\frac{1}{2\tau^c_u}=\frac{1}{2\tau^c_l}
%=n_{\Omega}\frac{|\lambda|^2n_cu_0^2}{4Ms^4}M^2_{2\Omega/s}, \label{eq24}
\frac{1}{2\tau^c_u}=\frac{1}{2\tau^c_l}
=n_{\Omega}\frac{|\lambda|^2n_cU_0^2}{4\hbar^3 ms^4}\left[e^{-(\Omega a_h/s)^2}-e^{-(\Omega
a_e/s)^2}\right]^2, \label{eq24}$$ and $$\begin{aligned}
&&\frac{1}{2\tau^n_u} =\frac{1}{2\tau^n_l}
=n_{\Omega}\frac{\vert\lambda\vert^2 m^2sU_0^2}{16\pi^2\hbar^4 \Omega^2}\label{eq25} \\
%=n_{\Omega}|uv|^2\frac{sm^2U_0^2}{4\pi^2}\label{eq25} \\
%=n_{\Omega}|uv|^2\frac{2(\pi Ms^2u_0)^2}{(2\pi s)^3}\times\\\nonumber
&&\times\left[\frac{F(\sqrt{2}\Omega
a_h/s)}{a_h\sqrt{2}}+\frac{F(\sqrt{2}\Omega
a_e/s)}{a_e\sqrt{2}}-2\frac{F(\sqrt{a_h^2+a_e^2}\Omega/s)}{\sqrt{a_h^2+a_e^2}}\right], \nonumber\end{aligned}$$ where $F(x)$ is the Dawson integral $$F(x)=e^{-x^2}\int_0^x e^{t^2}dt.$$
Discussion
==========
Simple analysis of Rabi oscillations given above accounting for the finite values of relaxation times yields $$\begin{aligned}
|\langle\psi_2^+(0)\psi_1(t)\rangle|^2 &=&\frac{|\lambda|^2}{2\Omega^2}
\left(1-\cos\Omega
t\right)e^{-{t}\left({1}/{2\tau_u}+{1}/{2\tau_l}\right)}\nonumber \\
&&+\frac{|\lambda|^2}{4\Omega^2}\left(e^{-{t}/{2\tau_u}}-e^{-{t}/{2\tau_l}}\right)^2. \label{eq26}\end{aligned}$$
We focus our discussion on the low temperature regime $T \ll \Omega$. First we note that the relaxation rates for the upper and lower levels coincide both in the normal phase \[Eq. (\[eq23\])\] and in the Bose-condensed phase \[Eqs. (\[eq24\])-(\[eq25\])\]. With relaxation times for the upper and lower levels being equal, Eq. (\[eq26\]) is simplified to $$\begin{gathered}
\label{eq28}\nonumber
|\langle\psi_2^+(0)\psi_1(t)\rangle|^2=\frac{|\lambda|^2}{2\Omega^2}\left(1-\cos\Omega t\right)e^{-t/\tau},\end{gathered}$$ where $1/\tau=1/\tau_u=1/\tau_l$, with $1/\tau$ given by Eq. (\[eq23\]) in the normal phase and by $1/\tau^c+1/\tau^n$ from Eqs. (\[eq24\])-(\[eq25\]) in the BEC phase. Secondly, the relaxation rates are all proportional to the distribution function $n_{\Omega}$ of the dressed quasiparticle states given in Eq. (\[nOmega\]). Since $n_{\Omega} =
1$ only if the upper dressed quasiparticle state $+\Omega$ is occupied and vanishes otherwise, finite relaxation of the TLS Rabi oscillations occurs only when the pump field frequency exceeds the TLS energy level difference $\omega \geq 2\Delta$ [@Remark1]. In our following discussion, therefore, we focus on the regime $\omega
\geq 2\Delta$. In the normal phase, we note that the low-temperature relaxation rates Eq. (\[eq23\]) are independent of the driving frequency and increases monotonically with the TLS-light coupling as $\lambda^2$. In the BEC phase, we find that the relaxation rates Eqs. (\[eq24\])-(\[eq25\]) exhibit strong non-monotonic dependence on the driving frequency through the Rabi frequency and the TLS-light coupling.
\[h\] ![Relaxation rate $\hbar/\tau^{c}$ due to condensate particles (normalized by $\Delta$) versus frequency detuning $y$. Red (dot-dashed), blue (dashed) and black (solid) lines correspond to $\lambda=0.1\,\mathrm{meV}$, $0.2\,\mathrm{meV}$ and $0.3\,\mathrm{meV}$ respectively.[]{data-label="Figure4"}](figure4_updated_3.pdf "fig:"){width="8.5cm"}
\[h\] ![Relaxation rate $\hbar/\tau^{n}$ due to non-condensate particles (normalized by $\Delta$) versus frequency detuning $y$. Red (dot-dashed), blue (dashed) and black (solid) lines correspond to $\lambda=0.1\,\mathrm{meV}$, $0.2\,\mathrm{meV}$ and $0.3\,\mathrm{meV}$ respectively.[]{data-label="Figure5"}](figure5_updated_3.pdf "fig:"){width="8.5cm"}
\[h\] ![Relaxation rates due to condensate and non-condensate particles versus TLS-light coupling $\lambda/\Delta$ at frequency detuning $y = 0.01$. Solid (black) line corresponds to $\hbar/\tau^{c}$ and dashed (red) line to $\hbar/\tau^{n}$, respectively.[]{data-label="Figure6"}](figure6_updated_3_1.pdf "fig:"){width="8.5cm"}
\[h\] ![Three-dimensional plot of relaxation rates $\hbar/\tau^{c}$ and $\hbar/\tau^{n}$ as a function of TLS-light coupling $\lambda/\Delta$ and frequency detuning $y$. Grey (red) surface corresponds to $\hbar/\tau^{c}$ and black (blue) surface to $\hbar/\tau^{n}$, respectively.[]{data-label="Figure7"}](figure7_updated_3_1.pdf "fig:"){width="8.5cm"}
Below, we proceed to analyze the numerical dependence of the relaxation rates on the driving frequency and TLS-light coupling in the BEC phase. For the TLS, we take the following parameters $\Delta = 500 \,\mathrm{meV}$, $m_e=0.067\,m_0$ and $m_h=0.45\,m_0$ ($m_0$ is the electron mass) typical for GaAs-based QDs [@GaAs_QD1; @GaAs_QD2; @GaAs_QD3]. The characteristic lengths of the hole and the electron wavefunctions are taken as $a_h = 2\,
\mathrm{nm}$ and $a_e \approx a_h \sqrt{m_h/m_e} = 2.23 a_h$. With the dipole matrix element of the QD $\sim
10\,-\,100\,\mathrm{Debye}$ ($1\,\mathrm{Debye} = 3.3\times
10^{-30}\,\mathrm{Cm}$) and optical field strength $0.1\,-\,10\,\mathrm{MVm^{-1}}$, the TLS-light coupling constant takes the range of values $\lambda \sim 0.1\,-\,10\,\mathrm{meV}$. For the dipolar exciton gas, we take $d=10\,\mathrm{nm}$, $n_c=10^{10}\,\mathrm{cm}^{-2}$, $\varepsilon = 12.5$ and $m = m_e+m_h = 0.517\,m_0$ typical for GaAs DQW structures. To provide an estimate for the inter-particle interaction $g_0$, we assume a simple point-charge treatment of the dipolar excitons, and the exciton-exciton interaction potential takes the form [@Zimmermann] $g(r) =
(2e^2/\varepsilon)(1/r-1/\sqrt{r^2+d^2})$. Fourier transform of $g(r)$ then gives $g(k) = (4\pi e^2/\varepsilon k)[1-\mathrm{exp}(-kd)]$ and the coupling constant $g_0 \equiv g(k = 0) = 4\pi e^2 d/\varepsilon$. The Bogoliubov speed of sound $s$ is thus also fixed from $s = \sqrt{g_0
n_c/m}$.
For convenience, we display the frequency $\omega$ in terms of the dimensionless frequency detuning $y=(\omega-2\Delta)/(2\Delta)$. Figs. \[Figure4\]-\[Figure5\] show the relaxation rates $\hbar/\tau^{c}$ and $\hbar/\tau^{n}$ as a function of $y$ for relatively small values of $\lambda \sim 0.1\,\mathrm{meV}$. We find that both relaxation rates behave non-monotonically as a function of detuning, reaching maximum values at $y \sim 0.01$ and then becoming exponentially suppressed at larger values of $y$. $\hbar/\tau^{c}$ is more strongly suppressed than $\hbar/\tau^{n}$. Secondly, we observe that, for the present values of $\lambda \sim 0.1\,\mathrm{meV}$, the condensate and non-condensate fractions contribute to the relaxation rate by the same order of magnitude, with $\hbar/\tau^{c}$ exceeding $\hbar/\tau^{n}$. This trend is maintained until $\lambda$ reaches $\sim 1\%$ of $\Delta$ (corresponding to $5\,\mathrm{meV}$), when $\hbar/\tau^{c}$ starts to drop signifcantly faster than $\hbar/\tau^{n}$. Fig. \[Figure6\] shows both quantities plotted versus $\lambda/\Delta$ at a frequency detuning $y = 0.01$, from which we observe that $\hbar/\tau^{c}$ decreases much more abruptly than $\hbar/\tau^{n}$. When $\lambda$ is increased beyond $\sim 0.02\Delta$, it is seen that $\hbar/\tau^{n}$ now overtakes $\hbar/\tau^{c}$. For $\lambda$ values beyond $0.03\Delta$, $\hbar/\tau^{c}$ has dropped essentially to zero and the non-condensate fraction constitutes the dominant contribution to relaxation.
Although the relaxation rates vanish expectedly when $\lambda = 0$, they do not vanish at zero frequency detuning $y = 0$, as one might conclude by inspecting Figs. \[Figure4\]-\[Figure5\]. To examine more fully the behavior of $\hbar/\tau^{c}$ and $\hbar/\tau^{n}$, we plot them in Fig. \[Figure7\] in the full range of $\lambda$ and $y$. At $y = 0$, both relaxation rates become small only when $\lambda \ll \Delta$; in addition $\hbar/\tau^{c}$ also become small when $\lambda \gtrsim 0.03\Delta$. Around $\lambda \sim
0.01\Delta$, both relaxation rates as a function of $y$ reach maximum at $y = 0$.
Conventionally, the BEC phase transition of a dipolar exciton gas is detected using optical spectroscopy. In the BEC phase, the excitons or exciton-polaritons are described by a single coherent wave function and emit light coherently. The resulting luminescence peak becomes much narrower in comparison with that in the normal phase, signaling formation of the condensate state. Another way to detect the BEC phase transition has been theoretically suggested recently [@10_RefKovalevChaplikJETP2016; @10_1RefKovalevChaplikJETPLetters2016]. BEC phase transition strongly influences the non-equilibrium properties of a dipolar exciton gas driven by an external surface acoustic waves (SAW). Under phase transition, the SAW attenuation effect and the SAW-exciton drag current become strongly modified, allowing one to detect the BEC phase transition using acoustic spectroscopy. On top of the foregoing, our findings in principle provide a new strategy to detect the BEC phase transition of the dipolar exciton gas. While the QD’s relaxation rate displays only a monotonic linear dependence on the light intensity when the exciton gas is in the normal phase, it becomes strongly non-monotonic as a function of both the pump field’s frequency and intensity once the exciton gas is in the BEC phase. Thus, by monitoring the Rabi oscillation dynamics of the QD, the normal and condensed phases of the exciton gas can be distinguished by the dependence of the relaxation rate on the frequency and intensity of the driving field.
Conclusion
==========
To conclude, we have developed a theory for the relaxation of optically pumped two-level systems coupled to a bosonic bath using the nonequilibrium Keldysh technique and the diagrammatic perturbation theory. To elucidate the effects of bath phase transition, we have considered the cases when the bosonic bath is in the normal state and in the Bose-condensed state. We then apply our theory to study the scenario of an illuminated quantum dot coupled to a dipolar exciton gas. The condensate and non-condensate fractions of the bath particles contribute to the relaxation rate by variable proportions depending on the value of pump field amplitude. When the pump field is weak, both fractions contribute by about the same order of magnitude; while for strong pump field, the non-condensate fraction becomes the dominant contribution. Our findings also show that the phase transition of the dipolar exciton gas to the BEC regime results in a strong dependence of the relaxation rate on the optical pump field. The relaxation rate then exhibits a strong non-monotonic behavior, reaching a maximum and then becoming exponentially suppressed as a function of both the pump field’s frequency and amplitude. Such a non-monotonic dependence could in principle serve as a smoking gun for detecting BEC phase transition of the coupled dipolar exciton gas. Finally, we point out that despite our focus on dipolar exciton gas in this work, the theory we have developed is also applicable to other types of Bose gas, such as 2D exciton-polaritons [@10_7Deng2010], magnons [@10_8Pokrovskii2013] and cold atoms [@10_9Pethick2002].
Acknowledgments
===============
V.M.K. acknowledges the support from RFBR grant $\#16-02-00565a$. W.K. acknowledges the support by a startup fund from the University of Alabama.
Appendix {#Appendix}
========
Non-Equilibrium Green’s Functions {#Sec:NEGF}
---------------------------------
Because of the time-dependent perturbation from light, we employ the Keldysh formalism to calculate the Green’s function and distribution function of the system. Following established routes in non-equilibrium Green’s function formalism, the left-multiplied and right-mulitplied Dyson equations for the contour-ordered Green’s function ${G}^c$ are $$\begin{aligned}
G_0^{-1} {G}^{c} &=& 1+{\Sigma}^{c} {G}^{c}, \\
{G}^{c} G_0^{-1} &=& 1+{G}^{c} {\Sigma}^{c}. \label{Dyson}\end{aligned}$$ We are interested in the Green’s function of the TLS under irradiation, therefore the self-energy ${\Sigma}$ due to interaction with the bath is set to zero. In the rotating frame, we already find $$\begin{gathered}
\label{5}
\hat{G}_0^{-1}(t,t') = \left(
\begin{array}{cc}
i\partial_t-\left(\Delta-\frac{\omega}{2}\right) & -\lambda \\
-\lambda^\ast & i\partial_t+\left(\Delta-\frac{\omega}{2}\right) \\
\end{array}
\right).\end{gathered}$$ First we derive the retarded Green’s function. Applying Langreth’s rules [@14_Jauhobook] to the two equations in Eq. (\[Dyson\]) and summing them together, we have $\hat{G}_0^{-1}\hat{G}^{R}+\hat{G}^{R}\hat{G}_0^{-1}
= 2\delta(t-t')$, $$\begin{aligned}
%\bar{G}_0^{-1}\bar{G}^{R}+\bar{G}^{R}\bar{G}_0^{-1} &=& 2\delta(t-t')
%\nonumber
%\\
\left(i\frac{\partial \hat{G}^R}{\partial t}-\hat{H}_0
\hat{G}^R\right)+\left(-i\frac{\partial \hat{G}^R}{\partial
t'}-\hat{G}^R \hat{H}_0\right) = 2\delta(t-t'). \nonumber \\
\label{Dyson2}\end{aligned}$$ We transform the time variables $t,t'$, into the Wigner coordinates with the average time $T = (t+t')/2$ and relative time $\tau =
t-t'$. Eq. (\[Dyson2\]) becomes $$\begin{aligned}
i2\frac{\partial \hat{G}^R}{\partial
\tau}-\left\{\hat{H}_0,\hat{G}^R\right\} = 2\delta(\tau).\end{aligned}$$ Performing Fourier transformation with respect to $\tau$ gives $$\begin{aligned}
2\varepsilon \hat{G}^R-\left\{\hat{H}_0,\hat{G}^R\right\} = 2.\end{aligned}$$ Solving this matrix equation yields the retarded Green’s function in Eq. (\[eq5\]): $$\begin{aligned}
\hat{G}^R(\varepsilon,T) &=&\frac{1}{2\Omega}\left(
\begin{array}{cc}
\varepsilon+\varepsilon_0 & \lambda \\
\lambda^\ast & \varepsilon-\varepsilon_0 \\
\end{array}
\right) \nonumber \\
&&\times
\left[\frac{1}{\varepsilon-\Omega+i\delta}-\frac{1}{\varepsilon+\Omega+i\delta}\right]
\nonumber \\
&=&\hat{A}\frac{1}{\varepsilon-\Omega+i\delta}+\hat{B}\frac{1}{\varepsilon+\Omega+i\delta}, \label{7}\end{aligned}$$ with $\hat{A},\hat{B}$ defined in Eq. (\[eq5\_1\]).
Now from the contour-ordered Dyson’s equations Eq. (\[Dyson\]) and applying Langreth’s rule, then subtracting the two equations, we get $G_0^{-1}G^{<}-G^{<} G_0^{-1} = 0$, $$\begin{aligned}
\left(i\frac{\partial \hat{G}^{<}}{\partial
t}-\hat{H}_0 \hat{G}^{<}\right)-\left(-i\frac{\partial \hat{G}^{<}}{\partial
t'}-\hat{G}^{<}\hat{H}_0\right) = 0,
%\bar{G}_0^{-1}\bar{G}^{<} = i\frac{\partial \bar{G}^{<}}{\partial
% t}-\bar{H}_0 \bar{G}^{<}, \\
%\bar{G}^{<} \bar{G}_0^{-1}= -i\frac{\partial \bar{G}^{<}}{\partial
% t'}-\bar{G}^{<}\bar{H}_0,\end{aligned}$$ Transforming into the Wigner coordinates, we obtain the kinetic equation $$i\frac{\partial \hat{G}^{<}}{\partial t}-\left[\hat{H}_0,\hat{G}^{<}\right]
= 0$$ The density matrix $f(t)$ is given by the equal-time Keldysh Green’s function $\hat{f}(t) = -i\hat{G}^{<}(t,t) = -i\hat{G}^{<}(T = t, \tau = 0)$, which satisfies $$\frac{\partial \hat{f}}{\partial T} +i[\hat{H}_0,\hat{f}] = 0.$$
Quasiparticle Lifetimes {#Sec:QPL}
-----------------------
We start from Eq. (\[eq11\]) in the main text $$\hat{G}^R =
\frac{1}{\Lambda}\left[{\hat{G}_{0}^{R}-\det(\hat{G}_0^{R})\sigma_y
(\hat{\Sigma}^R)^{\mathrm{T}}\sigma_y}\right]. \nonumber %\label{eq11}$$ From Eqs. (\[eq5\])-(\[eq5\_1\]) it can be easily evaluated that $\det(\hat{G}_0^{R}) = 1/(\varepsilon^2-\Omega^2+i\delta)$. Upon substitution of Eq. (\[eq5\]), the first term of Eq. (\[eq11\]) can be written as follows
$$\begin{aligned}
\frac{\hat{G}_{0}^{R}}{\Lambda}
&=& \frac{\hat{A}}{\varepsilon-\Omega-\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)-\mathrm{Tr}(\hat{B}\hat{\Sigma}^R)
(\varepsilon-\Omega)/(\varepsilon+\Omega)+\det(\hat{\Sigma}^R)/(\varepsilon+\Omega)}
\nonumber \\
&+&\frac{\hat{B}}{\varepsilon+\Omega-\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)(\varepsilon+\Omega)/(\varepsilon-\Omega)-\mathrm{Tr}(\hat{B}\hat{\Sigma}^R)+\det(\hat{\Sigma}^R)/(\varepsilon-\Omega)}. %\label{eq11}\end{aligned}$$
In the vicinity of the poles we have $$\begin{aligned}
\frac{\hat{G}_{0}^{R}}{\Lambda}
&\approx&
\frac{\hat{A}}{\varepsilon-\Omega-\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)\vert_{\varepsilon
= \Omega} +\det(\hat{\Sigma}^R) \vert_{\varepsilon
= \Omega}/(2\Omega)}
\label{eqSB1} \\
&+&\frac{\hat{B}}{\varepsilon+\Omega-\mathrm{Tr}(\hat{B}\hat{\Sigma}^R)\vert_{\varepsilon
= -\Omega}-\det(\hat{\Sigma}^R)\vert_{\varepsilon
= -\Omega}/(2\Omega)}. \nonumber \end{aligned}$$ Since $\hat{\Sigma}^R \sim 1/\tau$ where $\tau$ is the quasiparticle lifetime, $1/\tau \ll \Omega$ is satisfied for our perturbative calculations. We see that the last terms in the denominators above $\det(\hat{\Sigma}^R)\vert_{\varepsilon=\pm\Omega}/(2\Omega) \sim (1/\tau)\times 1/(\Omega\tau)$ are a factor of $1/(\Omega\tau)$ smaller than $1/\tau$ and thus can be neglected. Next we consider the second term of Eq. (\[eq11\]), whereupon substituting Eq. (\[eq5\]) becomes
$$\begin{aligned}
\frac{\det(\hat{G}_0^{R})\sigma_y
(\hat{\Sigma}^R)^{\mathrm{T}}\sigma_y}{\Lambda} &=&
\frac{\sigma_y(\hat{\Sigma}^R)^{\mathrm{T}}\sigma_y}{\varepsilon^2-\Omega^2-(\varepsilon+\Omega)\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)-(\varepsilon-\Omega)\mathrm{Tr}(\hat{B}\hat{\Sigma}^R)+\det(\hat{\Sigma}^R)}. \label{eqSB2} \end{aligned}$$
In the vicinity of the upper level where $\varepsilon \approx \Omega$, $$\begin{aligned}
&&\frac{\det(\hat{G}_0^{R})\sigma_y
(\hat{\Sigma}^R)^{\mathrm{T}}\sigma_y}{\Lambda} \label{eqSB3} \\
&\approx& \frac{1}{2\Omega}
\frac{\sigma_y(\hat{\Sigma}^R)^{\mathrm{T}}\sigma_y}{\varepsilon-\Omega-\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)\vert_{\varepsilon
= \Omega}+\det(\hat{\Sigma}^R)\vert_{\varepsilon
= \Omega}/(2\Omega)}. \nonumber
%\nonumber \\
%&\sim& \frac{1}{\Omega
% \tau}\frac{1}{\varepsilon-\Omega-\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)}. \label{eqSB3} \end{aligned}$$ We see that the expression in the last line above is a factor of $1/(\Omega \tau)$ smaller than the corresponding $\varepsilon \approx \Omega$ contribution (*i.e.*, the term $\propto \hat{A}$) in Eq. (\[eqSB1\]), and hence can be neglected. A similar analysis shows that the same is true for the $\varepsilon \approx -\Omega$ contribution in Eq. (\[eqSB2\]). Therefore we find $$\begin{aligned}
\hat{G}^R(\epsilon) &\approx& \frac{\hat{A}}{\varepsilon-\Omega-\mathrm{Tr}(\hat{A}\hat{\Sigma}^R)\vert_{\varepsilon
= \Omega}} \nonumber \\
&+&\frac{\hat{B}}{\varepsilon+\Omega-\mathrm{Tr}(\hat{B}\hat{\Sigma}^R)\vert_{\varepsilon
= -\Omega}}, \end{aligned}$$ which gives Eq. (\[eq12\]) with Eqs. (\[eq13\_1\])-(\[eq13\_2\]) in the main text.
TLS Self-Energy {#Sec:TLS_SE}
---------------
Here we derive the expression for the TLS self-energy. Analytic expression of the diagram depicted in Fig. \[Figure1\] reads $$\begin{gathered}
\label{A1}\nonumber
\hat{\Sigma}(t-t')=i\int \frac{d\textbf{k}}{(2\pi)^2}\hat{M}_{\textbf{k}}\hat{{G}}(t-t')\hat{M}_{-\textbf{k}}\Pi(\textbf{k},t-t'),\\
\Pi(\textbf{k},t-t')=i\sum_{\textbf{p}}\mathcal{G}(\textbf{p},t-t')\mathcal{G}(\textbf{p}+\textbf{k},t'-t),\end{gathered}$$ where times $t,t'$ are located on the Keldysh contour, $\Pi$ and $\mathcal{G}$ are the polarization operator and Green’s function of the bath’s particles.
To proceed further, let us first perform an analytic continuation to the real time domain. Using the Langreth’s rules [@14_Jauhobook] we find $$\begin{aligned}
\hat{\Sigma}^<(\omega) &=&
i\sum_{\textbf{k},\varepsilon}\hat{M}_{\textbf{k}}\hat{{G}}^<(\omega-\varepsilon)\hat{M}_{-\textbf{k}}\Pi^<(\textbf{k},\varepsilon),
\label{App2_1} \\
\hat{\Sigma}^R(\omega) &=&
i\sum_{\textbf{k},\varepsilon}\hat{M}_{\textbf{k}}\left[\hat{{G}}^<(\omega-\varepsilon)\Pi^R(\textbf{k},\varepsilon) \right.\nonumber \\
&&+\hat{{G}}^R(\omega-\varepsilon)\Pi^<(\textbf{k},\varepsilon) \nonumber\\
&&\left.+\hat{{G}}^R(\omega-\varepsilon)\Pi^R(\textbf{k},\varepsilon)\right]\hat{M}_{-\textbf{k}},
\label{App2_2}\end{aligned}$$ $$\begin{aligned}
\Pi^<(\textbf{k},\omega) &=& i\sum_{\textbf{p},\varepsilon}\mathcal{G}^<(\textbf{p}+\textbf{k},\varepsilon+\omega)\mathcal{G}^>(\textbf{p},\varepsilon),\label{App2_3} \\
\Pi^R(\textbf{k},\omega) &=&
i\sum_{\textbf{p},\varepsilon}[\mathcal{G}^<(\textbf{p}+\textbf{k},\varepsilon+\omega)\mathcal{G}^A(\textbf{p},\varepsilon)
\nonumber \\
&&+\mathcal{G}^R(\textbf{p}+\textbf{k},\varepsilon+\omega)\mathcal{G}^<(\textbf{p},\varepsilon)].\label{App2_4}\end{aligned}$$
### TLS self-energy in normal state bath
Lets consider the bath in normal phase state, $T>T_c$. In this case the bare Green’s functions of the bath’s particles are $$\begin{aligned}
%\label{A3}
\mathcal{G}^<(\textbf{k},\varepsilon) &=&-2\pi i
n_B(\xi_{\textbf{k}})\delta(\varepsilon-E_{\textbf{k}}),
\label{A3_1}\\
\mathcal{G}^>(\textbf{k},\varepsilon)&=&-2\pi i[1+n_B(\xi_{\textbf{k}})]\delta(\varepsilon-E_{\textbf{k}}),\label{A3_2}\\
\mathcal{G}^R(\textbf{k},\varepsilon)&=&\frac{1}{\varepsilon-E_{\textbf{k}}+i\delta},\label{A3_3}
%\mathcal{G}^<(\textbf{k},\epsilon)=-2\pi in_{\textbf{k}}\delta(\epsilon-E_{\textbf{k}})\\\nonumber
%\mathcal{G}^>(\textbf{k},\epsilon)=-2\pi i[n_{\textbf{k}}+1]\delta(\epsilon-E_{\textbf{k}})\\\nonumber
%\mathcal{G}^R(\textbf{k},\epsilon)=\frac{1}{\epsilon-E_{\textbf{k}}+i\delta}\end{aligned}$$ where $n_B(\xi_{\textbf{k}})$ is equilibrium Bose distribution and $E_{\textbf{k}}=k^2/2m$ is the energy of the bath’s particles (which corresponds to the kinetic energy of the exciton’s center-of-mass motion for the excitonic bath we consider in Section \[Sec:QD\]. Using these functions we find the following polarization operators $$\begin{aligned}
%\label{A4}\nonumber
\Pi^<(\textbf{k},\omega)&=&-2\pi
i\sum_{\textbf{p}}n_B(\xi_{\textbf{k}+\textbf{p}})[1+n_B(\xi_{\textbf{p}})]
\nonumber \\
&&\times\delta(\omega+E_{\textbf{p}}-E_{\textbf{p}+\textbf{k}}), \label{A4_1}\\
\Pi^R(\textbf{k},\omega)&=&\sum_{\textbf{p}}\frac{n_B(\xi_{\textbf{p}})-n_B(\xi_{\textbf{k}+\textbf{p}})}{\omega+E_{\textbf{p}}-E_{\textbf{p}+\textbf{k}}+i\delta}. \label{A4_2}
%\Pi^<(\textbf{k},\omega)=-2\pi i\sum_{\textbf{p}}n_{\textbf{p}+\textbf{k}}(n_{\textbf{p}}+1)\delta(\omega+E_{\textbf{p}}-E_{\textbf{p}+\textbf{k}})\\
%\Pi^R(\textbf{k},\omega)=\sum_{\textbf{p}}\frac{n_{\textbf{p}}-n_{\textbf{p}+\textbf{k}}}{\omega+E_{\textbf{p}}-E_{\textbf{p}+\textbf{k}}+i\delta}\end{aligned}$$ Substituting these expressions into Eq. (\[App2\_1\]), we obtain the self-energy expression Eq. (\[eq14\]).
### TLS self-energy in BEC bath
If the bath is in the Bose-condensed state, the elementary excitations in the Bogoliubov’s theory of weakly-interacting Bose gas have the energy dispersion $$\epsilon_{\textbf{k}}=\sqrt{\frac{k^2}{2m}\left(\frac{k^2}{2m}+2ms^2\right)}, \label{A7_0}$$ where $s^2=g_0n_c/m$ is the Bogolubov quasiparticle’s speed of sound, $g_0$ is the inter-particle interaction strength, and $n_c$ particles density in the condensate. The Green functions are given by $$\begin{aligned}
\label{A7}
\hat{\mathfrak{G}}^R(\textbf{k},\varepsilon) &=& \left(
\begin{array}{cc}
\mathfrak{G} & \mathfrak{F}^+ \\
\mathfrak{F} & \tilde{\mathfrak{G}} \\
\end{array}
\right)^R=\frac{1}{(\varepsilon+i\delta)^2-\epsilon_{\textbf{k}}^2}\label{A7_1}\\
&&\times\left(
\begin{array}{cc}
\varepsilon+k^2/2m+ms^2 & -ms^2 \\
-ms^2 & -\varepsilon+k^2/2m+ms^2 \\
\end{array}\right),
\nonumber\end{aligned}$$ and $$\begin{aligned}
%\label{A8}
&&\hat{\mathfrak{G}}^<(\textbf{k},\varepsilon) =
n_B(\varepsilon)[\hat{\mathfrak{G}}^R(\textbf{k},\varepsilon)-\hat{\mathfrak{G}}^A(\textbf{k},\varepsilon)]
\nonumber \\
&=& -\frac{2\pi i}{2\epsilon_{\textbf{k}}}\left(
\begin{array}{cc}
\varepsilon+k^2/2m+ms^2 & -ms^2 \\
-ms^2 & -\varepsilon+k^2/2m+ms^2 \\
\end{array}\right)\nonumber \\
&&\times\left\{n_B(\epsilon_{\textbf{k}})\delta(\varepsilon-\epsilon_{\textbf{k}})+[1+n_B(\epsilon_{\textbf{k}})]\delta(\varepsilon+\epsilon_{\textbf{k}})\right\},\label{A8_1}\end{aligned}$$ $$\begin{aligned}
&&\hat{\mathfrak{G}}^>(\textbf{k},\varepsilon) = [1+n_B(\varepsilon)][\hat{\mathfrak{G}}^R(\textbf{k},\varepsilon)-\hat{\mathfrak{G}}^A(\textbf{k},\varepsilon)]\nonumber
\\
&=&-\frac{2\pi i}{2\epsilon_{\textbf{k}}}\left(
\begin{array}{cc}
\varepsilon+k^2/2m+ms^2 & -ms^2 \\
-ms^2 & -\varepsilon+k^2/2m+ms^2 \\
\end{array}\right)\nonumber \\
&&\times\left\{[1+n_B(\epsilon_{\textbf{k}})]\delta(\varepsilon-\epsilon_{\textbf{k}})+n_B(\epsilon_{\textbf{k}})\delta(\varepsilon+\epsilon_{\textbf{k}})\right\}.\label{A8_2}
% \hat{\mathfrak{G}}^<(\textbf{k},\epsilon)=N_\epsilon[\hat{\mathfrak{G}}^R(\textbf{k},\epsilon)-\hat{\mathfrak{G}}^A(\textbf{k},\epsilon)]\\\nonumber=-\frac{2\pi i}{2\epsilon_k}\left(
% \begin{array}{cc}
% \epsilon+k^2/2m+ms^2 & -ms^2 \\
% -ms^2 & -\epsilon+k^2/2m+ms^2 \\
% \end{array}\right)\times\\\nonumber
% \times[N_k\delta(\epsilon-\epsilon_k)+(1+N_k)\delta(\epsilon+\epsilon_k)],\\\nonumber
% \hat{\mathfrak{G}}^>(\textbf{k},\epsilon)=(1+N_\epsilon)[\hat{\mathfrak{G}}^R(\textbf{k},\epsilon)-\hat{\mathfrak{G}}^A(\textbf{k},\epsilon)]\\\nonumber=-\frac{2\pi i}{2\epsilon_k}\left(
% \begin{array}{cc}
% \epsilon+k^2/2m+ms^2 & -ms^2 \\
% -ms^2 & -\epsilon+k^2/2m+ms^2 \\
% \end{array}\right)\times\\\nonumber
% \times[(1+N_k)\delta(\epsilon-\epsilon_k)+N_k\delta(\epsilon+\epsilon_k)],\\\nonumber
% N_k=\frac{1}{e^{\epsilon_k/T}-1},\end{aligned}$$ The polarization operator has two contributions. The first one comes from the condensate particles and the other one from non-condensate particles, Fig. \[Figure1\]. If the Bose bath is a two-dimensional system, the condensate occurs at zero temperature only. In this case one assumes that $n_B(\epsilon_{\textbf{k}})=0$. The contribution $P_c(\textbf{k},\omega)$ to the polarization operator from the condensate particles is $$\begin{aligned}
%\label{A9}\nonumber
P^R_c(\textbf{k},\omega) &=&
n_c[\mathfrak{G}^R+\mathfrak{F}^{+R}+\mathfrak{F}^R+\tilde{\mathfrak{G}}^R] \nonumber \\
&=& n_c\frac{k^2/m}{(\omega+i\delta)^2-\epsilon_{\textbf{k}}^2}, \label{A9_1}\\
P^<_c(\textbf{k},\omega) &=&
n_c[\mathfrak{G}^<+\mathfrak{F}^{+<}+\mathfrak{F}^<+\tilde{\mathfrak{G}}^<]
\nonumber \\
&=&-\pi in_c\frac{k^2/m}{\epsilon_{\textbf{k}}}\delta(\omega+\epsilon_{\textbf{k}}).\label{A9_2}
% P^R_c(\textbf{k},\omega) &=&
% n_c[\mathfrak{G}^R+\mathfrak{F}^{+R}+\mathfrak{F}^R+\tilde{\mathfrak{G}}^R] \nonumber \\
% &=& n_c\frac{k^2/m}{(\varepsilon+i\delta)^2-\epsilon_{\textbf{k}}^2}, \label{A9_1}\\
% P^<_c(\textbf{k},\omega) &=&
% n_c[\mathfrak{G}^<+\mathfrak{F}^{+<}+\mathfrak{F}^<+\tilde{\mathfrak{G}}^<]
% \nonumber \\
% &=&-\pi in_c\frac{k^2/m}{\epsilon_k}\delta(\varepsilon+\epsilon_{\textbf{k}}).\label{A9_2}
% %P^R_c(\textbf{k},\omega)=n_c[\mathfrak{G}^R+\mathfrak{F}^{+R}+\mathfrak{F}^R+\tilde{\mathfrak{G}}^R]=n_c\frac{k^2/m}{(\epsilon+i\delta)^2-\epsilon_k^2},\\\nonumber
% %P^<_c(\textbf{k},\omega)=n_c[\mathfrak{G}^<+\mathfrak{F}^{+<}+\mathfrak{F}^<+\tilde{\mathfrak{G}}^<]\\=-2\pi in_c\frac{k^2/m}{2\epsilon_k}\delta(\epsilon+\epsilon_k).\end{aligned}$$ Using this functions one finds the condensate contribution to the retarded self-energy, Eq. (\[eq16\_1\]).
Now let us find the contribution to the self-energy from the non-condensate particles. First, we need the retarded and lesser polarization operators. In the regime $ms^2 \gg k^2/2m$ where linear dispersion $\epsilon_{\textbf{k}}=sk$ of the Bogoliubov quasiparticles holds, using Eqs. (\[A7\])-(\[A8\_2\]) we find for the retarded and lesser polarization operators of non-condensate particles $$\begin{aligned}
%\label{A11}
P_n^R(\textbf{k},\omega) &=&
2i\sum_{\textbf{p},\varepsilon}\left[\mathfrak{F}^<(\textbf{p}+\textbf{k},\varepsilon+\omega)\mathfrak{F}^{+A}(\textbf{p},\varepsilon)\right.\label{A11_1}
\\
&&\left.+\mathfrak{F}^R(\textbf{p}+\textbf{k},\varepsilon+\omega)\mathfrak{F}^{+<}(\textbf{p},\varepsilon)\right]\nonumber \\
&=&\frac{(ms^2)^2}{2}\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\left(\frac{1}{\omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right. \nonumber\\
&&\left.-\frac{1}{\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right),\nonumber\end{aligned}$$ $$\begin{aligned}
P_n^<(\textbf{k},\omega) &=&
2i\sum_{\textbf{p},\varepsilon}\mathfrak{F}^<(\textbf{p}+\textbf{k},\varepsilon+\omega)\mathfrak{F}^{+>}(\textbf{p},\varepsilon)\label{A11_2} \\
&=&-\pi
i(ms^2)^2\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\delta(\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}). \nonumber
% P_n^R(\textbf{k},\omega)=2i\sum_{\textbf{p},\epsilon}[\mathfrak{F}^<(\textbf{p}+\textbf{k},\epsilon+\omega)\mathfrak{F}^{+A}(\textbf{p},\epsilon)+\\\nonumber
% +\mathfrak{F}^R(\textbf{p}+\textbf{k},\epsilon+\omega)\mathfrak{F}^{+<}(\textbf{p},\epsilon)]\\\nonumber
% =\frac{(Ms^2)^2}{2}\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\left[\frac{1}{\omega-\epsilon_{\textbf{p}}-\epsilon_{\textbf{p}+\textbf{k}}+i\delta}-\frac{1}{\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}+i\delta}\right],\\\nonumber
% P_n^<(\textbf{k},\omega)=2i\sum_{\textbf{p},\epsilon}\mathfrak{F}^<(\textbf{p}+\textbf{k},\epsilon+\omega)\mathfrak{F}^{+>}(\textbf{p},\epsilon)\\\nonumber
% =-2\pi i\frac{(Ms^2)^2}{2}\sum_{\textbf{p}}\frac{1}{\epsilon_{\textbf{p}}\epsilon_{\textbf{p}+\textbf{k}}}\delta(\omega+\epsilon_{\textbf{p}}+\epsilon_{\textbf{p}+\textbf{k}}).\end{aligned}$$ Now the calculation of the non-condensate particles’ contribution to the self-energy of TLS is simple, and we arrive at the expression Eq. (\[eq16\_2\]) of the main text.
[50]{}
L. Mandel and E. Wolf, *Optical Coherence and Quantum Optics* (Cambridge University Press, 1995).
L. Allen and J. H. Eberly, *Optical Resonance and Two Level Atoms* (Dover Publications, 1987).
L. M. K. Vandersypen and I. L. Chuang, Rev. Mod. Phys. **76**, 1037 (2005).
I. Chiorescu, P. Bertet, K. Semba, Y. Nakamura, C. J. P. M. Harmans, and J. E. Mooij, Nature **431**, 159 (2004).
A. Wallraff, D. I. Schuster, A. Blais, L. Frunzio, R. S. Huang, J. Majer, S. Kumar, S. M. Girvin, and R. J. Schoelkopf, Nature **431**, 162 (2004).
U. Weiss, *Quantum dissipative systems*, 2nd ed. (World Scientific, 1999).
A. Shnirman, Y. Makhlin and G. Sch$\ddot{o}$n, [ Physica Scripta]{} **102**, 147 (2002).
A. J. Leggett, S. Chakravarty, A. T. Dorsey, Matthew P. A. Fisher, Anupam Garg, and W. Zwerger, Rev. Mod. Phys. **59**, 1 (1987).
A. J. Ramsay, Semicond. Sci. Technol. **25**, 103001 (2010).
P. Lodahl, S. Mahmoodian, and S. Stobbe, Rev. Mod. Phys. **87**, 347 (2015).
G. Konstantatos, M. Badioli, L. Gaudreau, J. Osmond, M. Bernechea, F. P. Garcia de Arquer, F. Gatti, and F. H. L. Koppens, Nat. Nanotechnol. **7**, 363 (2012).
S. D. Franceschi, L. Kouwenhoven, C. Schönenberger, and W. Wernsdorfer, Nat. Nanotechnol. **5**, 703 (2010).
O. Cotlet, S. Zeytinoglu, M. Sigrist, E. Demler, and A. Imamoglu, [Phys. Rev. B]{} **93**, 054510 (2016).
F. P. Laussy, A. V. Kavokin, and I. A. Shelykh, [Phys. Rev. Lett.]{} **104**, 106402 (2010).
I. A. Shelykh, T. Taylor, and A. V. Kavokin, [Phys. Rev. Lett.]{} **105**, 140402 (2010).
F. P. Laussy, T. Taylor, I. A. Shelykh, and A. V. Kavokin, [J. Nanophotonics]{} **6**, 064502 (2012).
M. Matuszewski, T. Taylor, and A. V. Kavokin, [Phys. Rev. Lett.]{} **108**, 060401 (2012).
L. V. Butov, [Sol. State Comm.]{} 127, 89 (2003).
L. V. Butov, [J. Phys.: Cond. Matt]{} **16**, R1577 (2004).
L. V. Butov, [J. Phys.: Cond. Matt.]{} **19**, 295202 (2007).
V. B. Timofeev, A. V. Gorbunov, Phys. Status Solidi C **5**, 2379 (2008).
V. B. Timofeev, A. V. Gorbunov, [J. Exp. Theor. Phys.]{} **84** 390 (2006).
E. V. Calman, C. J. Dorow, M. M. Fogler, L. V. Butov, S. Hu, A. Mishchenko, and A. K. Geim, [Appl. Phys. Lett.]{} 108, 101901 (2016).
P. Rivera, J. R. Schaibley, A. M. Jones, J. S. Ross, S. Wu, G. Aivazian, P. Klement, K. Seyler, G. Clark, N. J. Ghimire, J. Yan, D. G. Mandrus, W. Yao, and X. Xu, *et al.*, [Nat. Commun.]{} **6**, 6242 (2015).
M. M. Fogler, L. V. Butov, and K. S. Novoselov, [Nat. Commun.]{} **5**, 4555 (2014).
F.-C. Wu, F. Xue, and A. H. MacDonald, [Phys. Rev. B]{} **92**, 165121 (2015).
O. L. Berman and R. Ya. Kezerashvili, [Phys. Rev. B]{} **93**, 245410 (2016).
J. L. Black and P. Fulde, [Phys. Rev. Lett.]{} **43**, 453 (1979).
R. S. Christensen, J. Levinsen and G. M. Bruun, [Phys. Rev. Lett.]{} **115**, 160401 (2015).
T. K. Melik-Barkhudarov, [Sov. Phys. JETP]{} **48**, 48 (1978).
V. Weisskopf and E. Wigner, [Z. Phys.]{}, **63** **54** (1930).
J. P. Gordon, L. R. Walker, and W. H. Louisell, [Phys. Rev.]{} **130**, 804 (1963).
S. T. Beliaev, [Sov. Phys. JETP]{} **7**, 289 (1958).
H. Shi and A. Griffin, [Phys. Rep.]{} **304**, 1 (1998).
L. P. Pitaevskii and S. Stringari, *Bose-Einstein Condensation*, International Series of Monographs on Physics (Clarendon Press, Oxford, 2003).
L. D. Landau and E. M. Lifshitz, *Quantum Mechanics*, (Pergamon Press, 1965).
W. Que, Phys. Rev. B **45**, 11036 (1992).
L. Jacak, P. Hawrylak, and A. Wojs, *Quantum Dots*, NanoScience and Technology (Springer, 2013).
H. Kamada, H. Gotoh, J. Temmyo, T. Takagahara, and H. Ando, Phys. Rev. Lett. **87** 246401 (2001).
This follows from our assumption of low-temperatures: the relaxation rate in the normal state obtained in the limit of low temperatures $T \ll \Omega$ is a zero-temperature result, while the relaxation rate in the Bose-condensed phase is obtained for temperatures below the BEC critical temperature (typically a few kelvins) where the Bogoliubov theory at $T = 0$ applies. At higher temperatures, the relaxation rates become non-zero when $\omega < 2\Delta$, as indicated by our result Eq. (\[eq22\]).
Y. Turki-Ben Ali, G. Bastard, R. Bennaceur, Physica E **27**, 67 (2005).
P. Lelong and G. Bastard, Solid State Commun. **98**, 819 (1996).
G. W. Bryant, Phys. Rev. B **37**, 8763 (1988).
C. Schindler and R. Zimmermann, Phys. Rev. B **78**, 045313 (2008).
V. M. Kovalev and A. V. Chaplik, [J. Exp. Theor. Phys.]{} **122** 499 (2016).
M. V. Boev, V. M. Kovalev and A. V. Chaplik, [J. Exp. Theor. Phys. Lett.]{} **104** 204 (2016).
H. Deng, H. Haug, and Y. Yamamoto, [Rev. Mod. Phys.]{} **82**, 1489 (2010).
F. Li, W. M. Saslow, V. L. Pokrovsky, [Scientific Reports]{}, **3**, 1372 (2013).
C. J. Pethick and H. Smith, *Bose-Einstein Condensation in Dilute Gases*, 2nd ed. (Cambridge University Press, 2002).
H. Haug and A. P. Jauho, *Quantum Kinetics in Transport and Optics of Semiconductors*, 2nd ed. (Springer-Verlag, 2008).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'There is a finite set of points on the boundary of the three-dimensional unit ball whose minimal tree is knotted. This example answers a problem posed by Michael Freedman.'
address: 'Auburn University, Auburn, AL 36830-5310, USA'
author:
- Krystyna Kuperberg
title: A knotted minimal tree
---
[^1]
Introduction
============
In \[1\], “Problems in Low-dimensional Topology” by Rob Kirby, one can find the following:
[**Problem 5.17 (Freedman)**]{} [*Given a finite set of points $X$ in $\partial B^3$, let $T$ be a tree in $B^3$ of minimal length with $\partial T=X$. Is $T$ unknotted, that is, is there a ${\rm PL}$ imbedded 2-ball in $B^3$ containing $T$?*]{}
It is shown here that there is a finite set $X$ on the boundary unit 3-ball in ${\mathbb R}^3$ whose minimal tree [*is*]{} knotted. The cardinality of $X$ is quite large, but the construction essentially depends on seven elements only: 6 points and an arc on the boundary of the ball. An outline of the example, provided by the author of this paper, is contained in \[1\] following the statement of Problem 5.17.
The arc lies close to the equator circumventing it one and a half times. In a plane perpendicular to the equator, consider a regular hexagon whose one pair of antipodal vertices is very close to the endpoints of the arc. The set $X$ contains one of the other pairs of antipodal vertices. The remaining two vertices of the hexagon are split into two points each so that the points are closer to the equator and the minimal tree connects these points to the endpoints of the arc. Finally, the arc is replaced by a sequence of points of small mesh.
Preliminaries and notation
==========================
Let $A$ be a locally connected compact set in ${\mathbb R}^3$ with finitely many components. A [*connecting graph*]{} for $A$ is a pair $({\mathcal
E},{\mathcal V})$, where ${\mathcal E}$ is a finite collection of straight line segments (edges) and ${\mathcal V}$ is a set of points (vertices) consisting of endpoints of edges of ${\mathcal E}$, such that the set $A\cup
\bigcup_{E\in {\mathcal E}} E$ is connected. We informally say that the graph consists of these edges. The [*length*]{} of a connecting graph is the sum of the lengths of its edges. A [*minimal graph*]{} for $A$ is a connecting graph for $A$ whose length is a minimum. A minimal graph for $A$ is denoted by $G(A)$, possibly with a subscript if more than one of such graphs is considered. The length of $G(A)$ is denoted by $|G(A)|$. No additional notation is used for the union of the elements of $G(A)$ – it is also denoted by $G(A)$. For simplicity, assume that every point of $A\cap G(A)$ is a vertex of $G(A)$ but otherwise $G(A)$ has the minimum number of vertices, i.e., if two edges meeting at a vertex are collinear, then the vertex belongs to $A$. The [*order*]{} of a vertex is the number of edges meeting at this vertex.
If $A$ is finite, then a minimal graph $G(A)$ is a tree, i.e., it is connected and acyclic. It is then called a [*minimal tree*]{} for $A$ and it is denoted by $T(A)$. Its length is denoted by $|T(A)|$. A [*simple triod*]{} is a tree consisting of three edges meeting at a vertex. For a three point set $A=\{a,b,c\}$, $T(A)$ is unique. If one of the angles of the triangle $\triangle
(a,b,c)$ is greater than or equal to $\frac{2\pi}{3}$, then the minimal tree consists of two edges. If all angles of $\triangle (a,b,c)$ are less than $\frac{2\pi}{3}$, then the minimal tree is a simple triod whose edges form $\frac{2\pi}{3}$ angles. In general, a minimal tree is not unique. For example, the set of vertices of a square has two minimal trees.
If four or more half-lines in ${\mathbb R}^3$ have a common endpoint $p$, then at least one of the angles between the half-lines is less than $\frac{2\pi}{3}$. If $q_1$ and $q_2$ are two distinct points equidistant to $p$ that are on two half-lines meeting at $p$ at an angle less than $\frac{2\pi}{3}$, then the minimal tree $T(\{p,q_1,q_2\})$ is a triod. Therefore, a vertex of a minimal graph $G(A)$ is either of order 3 or it belongs to $A$. The angles between the edges meeting at a vertex not in $A$ equal $\frac{2\pi}{3}$ and the edges are coplanar.
The segment joining the points $p$ and $q$ is denoted by $[p,q]$. The Euclidean distance is denoted by ${\rm d}(p,q)$. For distinct points $p$, $q$ and $r$, denote by ${\rm L}(q,r)$ the line passing through $q$ and $r$, and by ${\rm d}(p,{\rm L}(q,r))$ the perpendicular distance between $p$ and ${\rm L}(q,r)$. The Hausdorff distance between the sets $A$ and $B$ is denoted by ${\rm d_H}(A,B)$. We say that two sets $A$ and $B$ with the same finite number of components, $A_1,\ldots ,A_k$ and $B_1,\ldots ,B_k$, respectively, are Hausdorff $\epsilon$-close, if there is a permutation $\tau :\{1,\ldots ,k\}\to\{1,\ldots ,k\}$ such that for $i=1,\ldots k$, ${\rm d_H}(A_i,B_{\tau (i)})<\epsilon$. If $p$ and $q$ are non-antipodal points on a circle or a sphere $C$, then ${\rm arc}_C(pq)$ denotes the shortest arc in $C$ joining the points $p$ and $q$.
In here, the 3-ball $B^3$ is exactly the unit ball in $ {\mathbb R}^3$. A PL imbedded 2-ball $D$ in $B^3$ is properly imbedded, i.e., $\partial B^3\cap D= \partial D$. A tree for a finite set $A\subset\partial B^3$ is [*unknotted*]{} if there is a PL 2-ball $D$ containing the tree, or equivalently, if there is an isotopy of $B^3$ onto itself such that the image of the tree under the final stage of the isotopy is contained in the $xy$-plane.
Throughout the paper we use the following notation: $$\begin{array}{rcl}
S^2=\partial B^3&=&\{(x,y,z)\in {\mathbb R}^3\ |\ x^2+y^2+z^2=1\},\\
P&=&\{(x,y,z)\in S^2\ |\ y=0\},\\
Q&=&\{(x,y,z)\in S^2\ |\ z=0\},
\end{array}$$ where $(x,y,z)$ denotes the Cartesian coordinates of a point in ${\mathbb R}^3$.
Some special graphs
===================
Let $H\subset P$ be the regular hexagon with vertices:
$$\begin{array}{lll}
a_1=(-\frac{1}{2},0 ,\frac{\sqrt{3}}{2}),
&b_1=( -1,0,0),
&c_1=(-\frac{1}{2},0,-\frac{\sqrt{3}}{2}),\\
a_2=(\frac{1}{2},0,-\frac{\sqrt{3}}{2}),
&b_2=(1,0,0),
&c_2=(\frac{1}{2},0,\frac{\sqrt{3}}{2}).
\end{array}$$
An easy verification gives the following:
\[four\] Every minimal tree $T(A)$ for the set $A$ consisting of four consecutive vertices of $H$ consists of three edges of $H$.
\[five\] Every minimal tree $T(A)$ for the set $A$ consisting of five vertices of $H$ consists of four edges of $H$.
Let $A=\{a_1, b_1,c_1,a_2,c_2\}$.
Suppose that $T(A)$ contains at least one edge of $H$. If $b_1$ is an order 1 vertex of $T(A)$ and it is one of the endpoints of the only edge of $H$ belonging to $T(A)$, then $|T(A)|=1+2\sqrt{3}$. The cases when either $a_1$ or $c_1$ is of order 1 can also be easily eliminated. If either $a_2$ or $c_2$ is of order 1, then Lemma \[four\] can be used.
Now suppose that no edge of $H$ belongs to $T(A)$, i.e., $T(A)$ has three additional vertices. We may assume that every edge of $T(A)$ with an endpoint in $H$ has length less than one. The combinatorial scheme for the edges of $T(A)$ is as in Figure \[finterior\]. The tree to the left shows that starting with any edge joining two interior vertices and with the four adjacent edges, by attaching the sixth edge to one of the legs, we obtain the tree to the right. The edge $K_1$ is different from the other edges that have a vertex of $H$ as one of the endpoints: $K_1$ does not have a common endpoint with any of the other edges $K_i$. Consider three cases: $K_1=[p,b_1]$, $K_1=[p,c_1]$, and $K_1=[p,a_2]$ for some $p$ in the interior of the hexagon, see Figures \[ffivebc\] and \[ffivea\]. Assume that all angles at interior vertices equal $\frac{2\pi}{3}$.
Suppose that $K_1$ has $b_1$ as one of its endpoints. Then the pentagon whose sides are the segment $[a_2,c_2]$ and four edges of $T(A)$, as shown in Figure \[ffivebc\], would have three $\frac{2\pi}{3}$ angles and the remaining two angles each less than $\frac{\pi}{2}$.
Suppose that $K_1$ has $c_1$ as one of its endpoints. Then $T(A)$ contains two paths, one from $a_1$ to $c_1$ and the other from $a_2$ to $c_2$ not overlapping in a segment, see Figure \[ffivebc\]. Hence $|T(A)|> {\rm d}(a_1,a_2)+{\rm
d}(c_1,c_2)=4$.
Suppose that $K_1$ has $a_2$ as one of its endpoints. Let $K_2=[q,c_2]$. The angle $\angle(q,c_2,a_2)$ is greater than $\frac{\pi}{6}$, otherwise $\angle(c_2,a_2,p)\geq\frac{\pi}{2}$. The segment $[c_1,c_2]$ either intersects the interior of $[p,q]$ or it intersects the edge $K_1$. In either case denote the point of intersection by $r$.
If $r$ is in the interior of $[p,q]$, then by reflecting the part of $T(a)$ below $r$ in the line passing through $c_1$ and $c_2$ we obtain a graph of the same length as $T(a)$ and with an additional vertex at $r$, which is not possible.
Suppose now that $[c_1,c_2]$ intersects $K_1$. Let $x={\rm d}(p,{\rm L}(c_1,a_2))$ and $y={\rm d}(p_1,p_2)$, where the segment $[p_1,p_2]$ is parallel to ${\rm L}(c_1,a_2)$, $p\in [p_1,p_2]$, $p_1\in {\rm L}(c_1,a_1)$, and $p_2\in {\rm L}(c_1,c_2)$. Let $y_1={\rm d}(p_1,p)$ and $y_2={\rm d}(p_2,p)$, see Figure \[ffivexy\].
Note that
1. since $|K_1|<1$, $p$ is on the same side of ${\rm L}(c_1,a_1)$ as $a_2$,
2. $|K_1|= {\rm d}(a_2,r)+{\rm d}(r,p)\geq \frac{\sqrt 3}{2}+ {\rm
d}(p,{\rm L}(c_1,c_2))\geq \frac{\sqrt 3}{2}+\frac{\sqrt 3}{2}y_2 $,
3. ${\rm d}(p,{\rm L}(b_1,c_1))=\frac{x+\sqrt{3}y_1}{2}$,
4. $y=\frac{x}{\sqrt 3}$,
5. $|T(\{p,a_1,c_2\})|\geq |G(\{a_1,c_2\}\cup{\rm L}(p_1,p_2))|=
\frac{3}{2}\sqrt{3}-x$,
6. $|T(\{p,b_1,c_1\})|\geq {\rm d}(p,{\rm L}(b_1,c_1))+\frac{\sqrt 3}{2} =
\frac{x}{2}+\frac{\sqrt 3}{2}y_1+\frac{\sqrt 3}{2}$ (this holds true even if ${\rm d}(p,{\rm L}(b_1,c_1))<\frac{1}{2\sqrt 3}$ and $T(\{p,b_1,c_1\}$ is not a triod).
We have $ |T(A)|=|K_1|+ |T(\{p,a_1,c_2\})|+ |T(\{p,b_1,c_1\})|
\geq \frac{\sqrt 3}{2} +\frac{\sqrt 3}{2}y_2 +
\frac{3}{2}\sqrt 3-x+ \frac{x}{2}+\frac{\sqrt 3}{2}y_1+ \frac{\sqrt 3}{2} =
\frac{5\sqrt 3}{2}- \frac{x}{2}+ \frac{\sqrt 3}{2}y= \frac{5\sqrt
3}{2}>4$.
\[hexa\] Every minimal graph $G(A)$ for the set $A=Q\cup \{a_1,c_1,a_2,c_2\}$ consists of four edges of $H$.
Let $b_4=(0,1,0)$ and $b_5=(0,-1,0)$.
If $G(A)\cap Q\subset \{b_1, b_2\}$, then $G(A)$ is planar; otherwise the projection of $G(A)$ onto the $xz$-plane would be of shorter length. Then the vertices of a component of $G(A)$ that are in $H$ form a sequence of consecutive vertices of $H$. By the previous lemmas, $G(A)$ consists of 4 edges of $H$.
Suppose that one of the edges $[a_1,c_2]$ or $[c_1,a_2]$ belongs to $G(A)$. The only segments in the convex hull of $A$ forming with these edges an angle $\geq \frac{2\pi}{3}$ at one of the endpoints $a_1,a_2,c_1$, or $c_2$ are contained in another edge of $H$. So if $G(A)$ contains either of these two edges, then it contains an edge adjacent to $[a_1,c_2]$ or $[c_1,a_2]$. If $G(A)$ does not contain $[a_1,c_2]$ (resp. $[c_1,a_2]$) but it contains an edge of $H$ adjacent to it, then this edge can be replaced by $[a_1,c_2]$ (resp. $[c_1,a_2]$) to get a connecting graph of the same length. Hence we may assume that if $G(A)$ contains an edge of $H$, then it also contains another edge on the same side of the $xy$-plane. Our consideration may be reduced to graphs $G(A)$ whose components contain either two or four of the vertices $a_1,a_2,c_1,c_2$. The case when one component contains $a_1$ and $a_2$, and the other $c_1$ and $c_2$ can be easily eliminated. If one of the components contains $a_1$ and $c_1$, and the other contains $a_2$ and $c_2$, then $G(A)$ is planar. Suppose that a non-planar component of $G(A)$ contains exactly two vertices of $H$, both on the same side of the $xy$-plane, say $a_1$ and $c_2$. Then $G(A)$ has a component that is a simple triod with one additional endpoint $a_3\in Q$ different from $b_1$ and $b_2$, and a vertex $s$ of order 3. Since $a_3$ is the closest point to $s$ on $Q$, the line ${\rm L}(s,a_3)$ intersects the $z$-axis. The only possible choices for $a_3$ so that $\angle (a_1,s,a_3)=\angle (c_2,s,a_3)$ are $b_4$ and $b_5$. But for $i=4,5$, $|T(\{a_1,c_2,b_i\})|=\frac{\sqrt 7+\sqrt 3}{2}>2$.
Hence if $G(A)$ is non-planar, then $G(A)$ is connected and contains a point $b_3\in Q$ different from $b_1$ and $ b_2$. We may assume that $b_3\in {\rm arc}_Q
(b_2,b_4)$. $G(A)$ is combinatorially equivalent to the graph pictured in Figure \[finterior\]. Let $p$ be the endpoint of $K_1$ that does not belong to $H$. The possible types of configurations are:
1. $K_1=[b_3,p]$ and the endpoints of $K_2$ and $K_3$ that belong to $H$ are on the same side of the $xy$-plane (see Figure \[fiveb\]),
2. $K_1=[b_3,p]$ and the endpoints of $K_2$ and $K_3$ that belong to $H$ are on the opposite sides of the $xy$-plane,
3. $K_1=[c_2,p]$, $K_2$ connects to $a_1$ and $K_3$ connects to $b_3$ (see Figure \[fiveac\]),
4. $K_1=[a_1,p]$, $K_2$ connects to $c_2$ and $K_3$ connects to $b_3$ (see Figure \[fiveac\]),
5. $K_1=[c_2,p]$, $K_2$ connects to $b_3$ and $K_3$ connects to either $c_1$ or $a_2$,
6. $K_1=[a_1,p]$, $K_2$ connects to $b_3$ and $K_3$ connects to either $c_1$ or $a_2$.
The most interesting is Configuration 1. Let $q$ and $r$ be the remaining two interior vertices different from $p$, with $q$ above the $xy$-plane and $r$ below. Let $p',q',r',b'_3$ be the points obtained from $p,q,r,b_3$, respectively, by a rotation of the tree $T(\{p,r,b_3\})$ in the $z$-axis so that ${\rm d}(b'_3,b_2)<{\rm d} (b_3,b_2)$. Let $E$ be the ellipsoid given by the equation ${\rm d}(x,a_1)+{\rm d}(x,c_2)=|K_2|+|K_3|$. Since $|K_2|={\rm
d}(c_2,q)\leq |K_3|={\rm d}(a_1,q)$, the point $q'$ is inside the ellipsoid ${E}$. Therefore ${\rm d}(a_1,q')+{\rm d}(c_2,q')<|K_2|+ |K_3|$. Similarly ${\rm d}(c_1,r')+{\rm d}(a_2,r')<|K_4|+ |K_5|$ and we obtain a connecting graph for the set $A$ of shorter length. Hence $b_3=b_2$.
It is easy to eliminate Configurations 2, 5, and 6. In each of these cases $G(A)$ contains two non-overlapping paths from the set $\{a_1,c_2\}$ to the set $\{a_2,c_1\}$, and since $G(A)$ is connected, $|G(A)|>4$.
Now consider Configuration 3. We have, $$|G(A)|>|T(\{a_1,b_3,p\})| +|T(\{c_1,a_2,c_2\})|>$$ $${\rm d}(a_1,b_4)+|T(\{c_1,a_2,(0,0,\frac{\sqrt 3}{2})\})={\sqrt
2}+3\frac{\sqrt 3}{2}>4.$$
Finally consider Configuration 4. Let $q$ and $r$ be the two additional interior vertices with $K_2=[q,c_2]$ and $r$ being the common vertex of $K_4$ and $K_5$ as in the right-hand picture in Figure \[fiveac\]. Note that $b_3$ is the point in $Q$ closest to $q$, hence the line $L(b_3,q)$ intersects the $z$-axis. If the segment $[p,r]$ intersects the $yz$-plane at $\tilde r$ different from $r$, then by reflecting the tree $T(\{ {\tilde r},a_2,c_1\})$ in the $yz$-plane, we obtain a connecting graph for the set $A$ of the same length as $G(A)$ with an additional vertex $\tilde r$. Therefore, we may assume that the points $p$ and $r$ are not on the opposite sides of the $yz$-plane.
Suppose that $r$ is either on the $yz$-plane or on the same side of the $yz$-plane as $c_2$. Let $b'_3,p',q',r'$ be the new vertices corresponding to $b_3,p,q,r$ obtained by rotating the path $[b_3,q]\cup
[q,p]\cup [p,r]$ about the $z$-axis. The points $\overline a_1$, $\overline c_2$, $\overline p$, $\overline p'$, $\overline q$, $\overline q'$ and $\overline r$ in Figure \[frotate\] are the projections of $a_1$, $c_2$, $p$, $p'$, $q$, $q'$ and $r$ onto the $xy$-plane; $o$ is the origin. Since $p$ is in the convex hull of $\{a_1,c_1,a_2,c_2,b_3\}$ and $q$ is in the triangle $\triangle
(p,b_3,c_2)$, then $\overline p\in \triangle (\overline a_1,o,\overline q)$.
If $b'_3\in {\rm arc}_Q(b_2,b_3)$ and $b'_3\not= b_3$, then, arguing in a similar fashion as for Configuration 1, we get ${\rm d}(c_1,r')+{\rm d}(a_2,r')<
|K_4|+|K_5|$ and ${\rm d}(a_1,q')+{\rm d}(c_2,q') <{\rm d}(a_1,q)+{\rm d}(c_2,q)$. Also note that ${\rm d}(a_1,q')>{\rm d}(a_1,q)$ and ${\rm d}(c_2,q')<{\rm
d}(c_2,q)$. Suppose that $\overline q$, $\overline p$ and $o$ are not collinear, and $\overline q'$, $\overline p'$ and $\overline a_1$ are collinear as in the right-hand picture in Figure \[frotate\]. Then $\angle (q',a_1,p')<\angle (q,a_1,p)$ and $\angle
(a_1,q',p')<\angle (a_1,q,p)$, and since ${\rm d}(p',q')={\rm d}(p,q)$, then ${\rm
d}(a_1,q')-{\rm d}(a_1,p')> {\rm d}(a_1,q)-{\rm d}(a_1,p)$, see Figure \[fangles\]. Hence ${\rm d}(a_1,q')-{\rm d}(a_1,q)> {\rm d}(a_1,p')-{\rm d}(a_1,p)$, which combined with ${\rm d}(a_1,q')+{\rm d}(c_2,q') <{\rm d}(a_1,q)+{\rm d}(c_2,q)$ gives ${\rm d}(a_1,p')+{\rm
d}(c_2,q') <{\rm d}(a_1,p)+{\rm d}(c_2,q) = |K_1|+|K_2|$. We obtain a new connecting graph for the set $A$ with interior vertices $p',q',r'$ whose length is less than $|G(A)|$.
If the points $\overline a_1$, $\overline p$ and $\overline q$ are collinear, then $a_1$, $p$ and $q$ are in a plane perpendicular to the $xy$-plane and so is the tree $T(\{a_1,q,r\})$, in particular, so is the segment $[p,r]$. Then $G(A)$ is planar.
If the points $\overline o$, $\overline p$ and $\overline q$ are collinear, then $b_3$, $q$, $p$ and $a_1$ are in a plane that is perpendicular to the $xy$-plane and passes through the $z$-axis. This plane also contains $c_2$ and $G(A)$ is planar.
If $p$ and $r$ are on the same side or on the $yz$-plane as $a_1$ (including the case when one or both of these points are on the $yz$-plane), then $$|G(A)|>|T(\{a_1,c_1,a_2\})|+ {\rm min}_{v\in yz{\rm -plane}}|T(\{v,c_2,b_3\})|.$$ If ${\rm min}_{v\in yz{\rm -plane}}|T(\{v,c_2,b_3\})|$ is attained at $v_0\in
yz$-plane, then either
1. $v_0=b_3=b_4$, $|T(\{v_0,c_2,b_3\})|=\sqrt 2$ and $|G(A)|= \sqrt
2+\frac{3\sqrt 3}{2}>4$, or
2. $v_0=(0,0,\frac{\sqrt 3}{2})$, $|T(\{v_0,c_2,b_3\})|=\frac{3}{2}$ and $|G(A)|= \frac{3}{2}+\frac{3\sqrt 3}{2}>4$ , or
3. $|T(\{v_0,c_2,b_3\})|$ is a simple triod.
In the last case, let $t$ be the order 3 vertex of $T(\{v_0,c_2,b_3\})$. Note that the line ${\rm L}(v_0,t)$ is perpendicular to the $yz$-plane and the line ${\rm
L}(b_3,t)$ intersects the $z$-axis. The plane $K$ containing ${\rm L}(v_0,t)$ and ${\rm L}(b_3,t)$ also contains the tree $T(\{v_0,c_2,b_3\})$, see Figure \[ftrapezoid\]. If $b_3=(x,y,0)$, then $\sqrt 3x=\sqrt{(\frac{\sqrt3}{2})^2+y^2}=\sqrt{\frac{7}{4}-x^2}$. Then $x=\frac{\sqrt 7}{4}$, $|T(\{v_0,c_2,b_3\})|=\frac{1}{4} +\frac{\sqrt 7}{2}$ and $|G(A)|> 2(x- \frac{1}{2})+ \frac{1}{2}+ \frac{\sqrt
3}{2}(\sqrt{\frac{7}{4} -x^2}-\sqrt 3(x-\frac{1}{2})) =\frac{3\sqrt 3}{2}+
\frac{1}{4}+\frac{\sqrt 7}{2}>4$.
\[Q+graph\] For every $\epsilon >0$ there is a $\delta >0$ such that if a set $A$ is Hausdorff $\delta$-close to $Q\cup \{a_1,c_1,a_2,c_2\}$, then every minimal graph $G(A)$ is in an $\epsilon$-neighborhood of four edges of $H$.
In the following lemmas, the notation $Q_\delta$ is used for a circle in $B^3$ that is in a plane parallel to the $xy$-plane, with center on the $z$-axis, and such that ${\rm d_H}(Q,Q_\delta )<\delta $.
\[x\] There are an $\epsilon >0$, a $\delta >0$, and an $\eta >0$ such that if
1. $a'_1$ and $c'_2$ are points in the $xz$-plane such that ${\rm
d}(a_1,a'_1)<\eta$, ${\rm d}(c_2,c'_2)<\eta$, and
2. $G(A)$ is a minimal graph for $A=Q_\delta \cup
\{a'_1,c'_2\}$ contained in an $\epsilon$-neighborhood of the set $[a_1, b_1]\cup [a_1,c_2]$,
then $G(A)$ is contained in the $xz$-plane.
Let $q$ be the point of $G(A)$ that belongs to $Q_\delta$. Thus $G(A)$ is the tree $T(\{q,a'_1,c'_2\})$. For small $\epsilon$, $\delta $, and $\eta$, the tree $T(\{q,a'_1,c'_2)$ consists of 2 segments in the $xz$-plane or it is a simple triod with an additional vertex $p$ close to $a'_1$. Then the point of intersection $r$ of the line ${\rm L}(p,q)$ and the edge $[a'_1,c'_1]$ is also close to $a'_1$. Since $q$ is the point on $Q_\delta$ that is closest to $p$, the line ${\rm L}( p,q)$ intersects the $z$-axis at some point $s$. Hence ${\rm L}(p,q)$ has two distinct points $r$ and $s$ in the $xz$-plane and $T(\{q,a'_1,c'_2\})$ is in the $xz$-plane.
For $0<\gamma <1$, let $c_1(\gamma)=(-\frac{1}{2},0,-\frac{\sqrt{3}}{2}(1-\gamma ))$ and $c_2(\gamma)=(\frac{1}{2},0,\frac{\sqrt{3}}{2}(1-\gamma ))$.
\[1-x\] There is a $\gamma_0 >0$, such that the set $A=Q\cup \{a_1,c_1(\gamma),a_2,c_2(\gamma)\}$, ${0<\gamma<\gamma_0}$, has a unique minimal graph $G(A)$ consisting of two simple triods $$T(\{a_1,b_1,c_1(\gamma )\})\ and\ T(\{a_2,b_2,c_2(\gamma )\}).$$
By Corollary \[Q+graph\], there are six cases of a minimal graph $G(A)$ to consider. $G(A)$ may be close to one of the sets consisting of the following fours edges of $H$:
$$\begin{array}{lll}
\vspace{.1in}
1.\ [a_1,b_1]\cup [b_1,c_1]\cup [a_2,b_2]\cup [b_2,c_2],
&2.\ [a_1,b_1]\cup [a_1,c_2]\cup [c_1,a_2]\cup [a_2,b_2],\\
\vspace{.1in}
3.\ [a_1,c_2]\cup [c_2,b_2]\cup [a_2,c_1]\cup [c_1,b_1],
&4.\ [a_1,b_1]\cup [b_1,c_1]\cup [a_1,c_2]\cup [a_2,b_2],\\
\vspace{.1in}
5.\ [a_1,b_1]\cup [b_1,c_1]\cup [c_1,a_2]\cup [b_2,c_2],
&6.\ [a_1,b_1]\cup [b_1,c_1]\cup [a_1,c_2]\cup [c_1,a_2].
\end{array}$$
The remaining cases of subsets of $H$ consisting of four edges are three cases symmetric with respect to the origin to Cases 4, 5, or 6, and six possibilities of graphs that do not connect to at least one of the vertices of $H$.
For $i=1,\ldots ,6$, let $G_i(A)$ be a minimal graph for $A$ corresponding to Case $i$, see Figure \[QH\]. Note that $G_1(A)$ and $G_3(A)$ are clearly contained in the $xz$-plane. The see that the remaining graphs are also in the $xz$-plane, denote by $q$ the additional vertex of the graph $G_i(A)$ so that there is an edge $[q,b_1]$ for $i=4,5,6$, and $q=b_1$ for $i=2$. The vertex $q$ separates $G_i(A)$ into subgraphs. The two subgraphs different from the edge $[q,b_1]$ connect one or two of the points $a_1,c_1(\gamma),
a_2,c_2(\gamma)$ to a circle $Q_\delta$. By Lemma \[x\], we may assume that the subgraphs are subsets of the $xz$-plane. Since the distance between a point in $Q_\delta$ and the circle $Q$ is constant, it easily follows that each $G_i(A)$ is in the $xz$-plane. We analyze the six cases as follows:
1. $|G_1(A)|<2({\rm d}(a_1,b_1) + {\rm d}(b_1,c_1(\gamma)))< 2(1+1-\gamma +\frac{\gamma
}{2})=4-\gamma$.
2. $|G_2(A)| = 2({\rm d}(a_1,p)+{\rm d}(p,b_1)+{\rm d}(c_2(\gamma),p))>2(1+{\rm
d}(c_2(\gamma),p))$. Since $p$ is on the same side of the line ${\rm L}(a_1,a_2)$ as $q$, and ${\rm d}(p,{\rm L}(a_1,c_2))< {\rm d}(c_2(\gamma ),{\rm
L}(a_1,c_2))$, then $|G_2(A)|>2(1+1-\frac{\gamma }{2})>|G_1(A)|$.
3. $|G_3(A)|> 2({\rm d}(a_1,c_2(\gamma))+{\rm d}(c_2(\gamma),b_2))>2(1+{\rm
d}(c_2(\gamma),b_2))>|G_1(A)|$.
4. Let $c'_1=(-\frac{1}{2}+\frac{3\gamma}{4},0,- \frac{\sqrt 3}{2}+\frac{\sqrt
3\gamma}{4})$, i.e., $c'_1$ is the point symmetric to $c_1(\gamma)$ with respect to the line ${\rm L}(c_1,c_2)$. Let $C$ be the circle of radius $\frac{\sqrt 3}{2}\gamma $ centered at $c'_1$. Since $c_1(\gamma)\in C$, $|G_4(A)|> |G(\{c_2(\gamma),a_1,b_1\}\cup C)|-{\rm d}(c_1,c'_1)
+1 = 1-\frac{\gamma }{2}+1-\gamma +1-\frac{\gamma }{2}+ 2\gamma -\frac{\sqrt 3}{2}\gamma
+1=4-\frac{\sqrt 3}{2}\gamma >|G_1(A)|$.
5. Suppose that $G_5(A)$ has two vertices $q$ and $u$ of order 3 and an edge $[c_2(\gamma),u]$. Then the sum of the angles of the trapezoid with vertices $a_1,q,u,a_2$ is less than $2\pi$. Hence $c_1(\gamma)$ is of order 2. We have $|G_5(A)|=\frac{1}{2}|G_1(A)| +{\rm d}(c_1(\gamma),a_2) +{\rm
d}(c_2(\gamma),b_2)=\frac{1}{2}|G_1(A)|+\frac{1}{2}|G_3(A)|>|G_1(A)|$.
6. Suppose that $G_6(A)$ has three vertices $p$, $q$ and $u$ of order 3 and edges $[a_1,p]$, $[b_1,q]$ and $[c_1(\gamma),u]$. The pentagon $U$ with vertices $a_2,u,q,p,c_2(\gamma)$ is inside the pentagon with vertices $a_2,c_1(\gamma),b_1,a_1,c_2(\gamma)$. Since the edge $[a_2, c_1(\gamma)]$ is parallel to the edge $[a_1, c_2(\gamma)]$, we have $\angle (a_1, c_2(\gamma),a_2)+ \angle
(c_2(\gamma),a_2, c_1(\gamma))=\pi$ and $\angle (p, c_2(\gamma),a_2)+ \angle
(c_2(\gamma),a_2, u)<\pi$. This is a contradiction since each of the remaining angles of $U$ equals $\frac{2\pi}{3}$. Hence $c_1(\gamma)$ is of order 2. We have $|G_6(A)|=|G_4(A)|-1 + {\rm d}(c_1(\gamma), a_2)>|G_4(A)|$.
Hence $G_1(A)$ is the unique minimal graph.
For a given $\gamma$ and $i=1,2$, let $L_i(\gamma )$ be the line passing through $c_i(\gamma )$ and perpendicular to the $xz$-plane. Denote by $e_i(\gamma )$ and $f_i(\gamma )$ the two points in which $L_i(\gamma )$ intersects the sphere $S^2$.
\[split\] There are a $\gamma_0 >0$ and a $\delta >0$, such that for $0<\gamma <\gamma_0$ and $i=1,2$,
1. the minimal graph for the set $Q_\delta\cup \{e_i(\gamma ),f_i(\gamma )\}$ is the triod $$T(\{v_i,e_i(\gamma),f_i(\gamma ) \}),$$
2. the minimal graph for the set $Q_\delta\cup \{a_i, e_i(\gamma ),f_i(\gamma )\}$ is the minimal tree $$T(\{a_i, v_i, e_i(\gamma),f_i(\gamma )\}),$$
where $v_i\in Q_{\delta}$ is a point in the $xz$-plane close to $b_i$.
For a given $\delta$, let $e'_i$ and $f'_i$ be points on a $Q_\delta$ closest to $e_i(\gamma )$ and $f_i(\gamma )$ respectively. Let $p\in {\rm arc}_{Q_\delta}
(e'_i,f'_i)$. The minimal tree $T(\{q,e_i(\gamma),f_i(\gamma ) \})$ has an edge containing $q$, whose extension intersects the $z$-axis, see Figure \[fsplit\]. For small $\gamma$ and $\delta$ this is possible only for $q\in xz$-plane. From this, both Conclusions 1 and 2 of Lemma \[split\] follow.
Similarly, one can prove the following:
\[splitmore\] There are a $\gamma_0 >0$ and an $\epsilon >0$, such that for $0<\gamma <\gamma_0$ and $i=1,2$, if the minimal graph $G(A)$ for the set $$A=Q\cup\{a_1,e_1(\gamma ),f_1(\gamma ),a_2,e_2(\gamma ),f_2(\gamma )\}$$ is in the $\epsilon$-neighborhood of the hexagon $H$, then $G(A)$ is symmetric with respect to the $xz$-plane.
Finally, in this sequence of lemmas we have:
\[splitfinal\] There is a $\gamma_0 >0$, such that for $0<\gamma <\gamma_0$ and $i=1,2$, the minimal graph $G(A)$ for the set $$A=Q\cup
\{a_1,e_1(\gamma ),f_1(\gamma ),a_2,e_2(\gamma ),f_2(\gamma )\}$$ is unique and consists of the two minimal trees $T(\{a_i, b_i, e_i(\gamma),f_i(\gamma )\})$, $i=1,2$.
\[circles\] There is a $\delta >0$ such that if $b'_2\in P$ is a point below the $x$-axis and ${\rm d}(b_2,b'_2)<\delta$, then the length of the minimal tree $|T(\{a_2,v,c_2(\gamma)\})|$ is a strictly monotone function of $v\in {\rm
arc}_P(b_2,b'_2)$ and attains its maximum at $b_2$.
$P$ is the circle circumscribed around the hexagon $H$. Let $a$ and $b$ be points in ${\rm arc}_P(b_2,b'_2)$ with $b\in
{\rm arc}_P(b_2,a)$. Note that $T(\{a_2,b,c_2(\gamma)\})$ is a simple triod; denote the vertex of order 3 by $q$. Let $E$ be the ellipse with foci $a_2$ and $c_2(\gamma)$, and passing through the point $q$. Let $p$ be the point on $E$ closest to $a$, see Figure \[fcircles\]. We have
$$|T(\{a_2,a,c_2(\gamma)\})|<{\rm d}(a,p) + {\rm d}(p,a_2) + {\rm d}(p,c_2(\gamma))=$$ $${\rm d}(a,p) + {\rm d}(q,a_2) + {\rm d}(q,c_2(\gamma))<|T(\{a_2,b,c_2(\gamma)\})|.$$
\[splitcircle\] There are a $\gamma >0$ and a $\delta >0$ such that if $b'_2\in P$ is a point below the $x$-axis and ${\rm d}(b_2,b'_2)<\delta$, then the length of the minimal tree $|T(\{a_2,v,e_2(\gamma),f_2(\gamma)\})|$ is a strictly monotone function of $v\in {\rm arc}_P(b_2,b'_2)$ and attains its maximum at $b_2$.
A knotted minimal tree
======================
In this section, we use the spherical coordinates, denoted by $(r,\theta , \phi)_s$, where $x=r\sin\phi \cos \theta$, $y=r\sin \phi \sin\theta$, and $z=r\cos \phi $.
Let $d_1(\delta)=(1,\pi,\frac{\pi}{2}-\delta)_s$, $d_2(\delta)=(1,0,\frac{\pi}{2}+\delta)_s$, and
$$\begin{array}{rcl}
M_1(\delta)&=&{\rm arc}_{S^2}(d_1,(1,\frac{5\pi}{4},\frac{\pi}{2}-\delta)_s),\\
M_2(\delta)&=&{\rm
arc}_{S^2}((1,\frac{5\pi}{4},\frac{\pi}{2}-\delta)_s,(1,\frac{7\pi}{4},\frac{\pi}{2})_s),\\
M_3&=&Q- {\rm
arc}_{S^2}((1,\frac{5\pi}{4},\frac{\pi}{2})_s,(1,\frac{7\pi}{4},\frac{\pi}{2})_s),\\
M_4(\delta)&=&{\rm
arc}_{S^2}((1,\frac{5\pi}{4},\frac{\pi}{2})_s,(1,\frac{7\pi}{4},\frac{\pi}{2}+\delta)_s),\\
M_5(\delta)&=&{\rm arc}_{S^2}((1,\frac{7\pi}{4},\frac{\pi}{2}+\delta)_s,d_2),\\
M(\delta)&=&\cup_{i=1}^5M_i(\delta).
\end{array}$$
Thus $M(\delta)$ is a path from $d_1(\delta)$ to $d_2(\delta)$, close to $Q$, and of length approximately $\frac{3}{2}$ times the length of $Q$.
\[model\] There are a $\gamma $ and a $\delta$ such that the minimal graph $G(B)$ for the set $$B=M(\delta)\cup \{a_1,e_1(\gamma),f_1(\gamma),a_2,e_2(\gamma),f_2(\gamma)\}$$ is unique and consists of the two minimal trees $T(\{a_i,d_i(\delta),e_i(\gamma),f_i(\gamma)\})$, $i=1,2$.
By Lemma \[splitfinal\], for if $\gamma $ is sufficiently small, then the minimal graph $G(A)$ for the set $$A=Q\cup\{a_1,e_1(\gamma ),f_1(\gamma ),a_2,e_2(\gamma ),f_2(\gamma )\}$$ is unique and $$G(A)=T(\{a_1, b_1, e_1(\gamma),f_1(\gamma
)\})\cup T(\{a_2, b_2, e_2(\gamma),f_2(\gamma )\}).$$ Since the sets $A$ and $B$ are Hausdorff $\delta$-close, there is a $\delta$ such that $$G(B)=T(\{a_1, m_1,
e_1(\gamma),f_1(\gamma )\})\cup T(\{a_2, m_2, e_2(\gamma),f_2(\gamma )\})$$ for some $m_1$ and $m_2$ on $M$. By Lemma \[split\], the points $m_i$ are on the $xz$-plane. For sufficiently small $\delta$, by Corollary \[splitcircle\], $m_i=d_i(\delta )$.
For some positive constants $\gamma $ and $\delta$ chosen so that Lemma \[model\] is satisfied, let $M=M(\delta )$, $d_i=d_i(\gamma )$, $e_i=e_i(\gamma )$, and $f_i=f_i(\gamma )$. For $\epsilon >0$, let $t_2, \dots ,t_{n(\epsilon )-1}$ be points on $M$ such that ${\rm d}(t_i,t_{i+1}) <\epsilon$ for $i=1,\ldots , n(\epsilon )-2$, $t_2=(1,\pi+\epsilon ,\frac{\pi}{2}-\delta)_s$, and $t_{{n(\epsilon )}-1}=(1,-\epsilon
,\frac{\pi}{2}+\delta)_s$. In addition let $t_1=(1,\pi-\epsilon ,\frac{\pi}{2}-\delta)_s$ and $t_{n(\epsilon )} =(1,\epsilon ,\frac{\pi}{2}+\delta)_s$.
\[example\] There is an $\epsilon >0$ such that the minimal tree for the set $$X=\{a_1,e_1,f_1,a_2,e_2,f_2, t_1,\ldots ,t_{n(\epsilon )}\}$$ is unique and knotted.
Let $A_1=\{a_1, t_1,t_2, e_1,f_1\}$ and $A_2=\{a_2, t_{n(\epsilon
)-1},t_{n(\epsilon )}, e_2,f_2\}$. For small $\epsilon$, $T(X)$ contains two subgraphs close to the minimal trees $T(\{a_1, d_1, e_1,f_1\})$ and $T(\{a_2, d_2, e_2,f_2\})$. Since ${\rm d}(t_1,t_2)={\rm d}(t_{n(\epsilon)-1},t_{n(\epsilon)})$ is approximately $2\epsilon$ and ${\rm d}(t_i,t_{i+1})<\epsilon$ for the remaining points $t_i$, for sufficiently small $\epsilon$, the two subgraphs are $T(A_1)$ and $T(A_2)$. Thus there is an $\epsilon >0$ such that $T(X)$ is unique and $$T(X) =T(A_1)
\cup [t_2,t_3]\cup \ldots \cup [t_{n(\epsilon )-2},t_{n(\epsilon )-1}]
\cup T(A_2).$$ Such $T(X)$ is knotted, see Figure \[fknotted\].
[**Remark.**]{} A slight change of the arc $M$ and an appropriate choice of the sequence of the points $t_i$ can yield a finite set in $S^2$ with two minimal trees, one knotted and the other unknotted.
The example of the knotted minimal tree raises the following questions:
1. (M. Freedman) What does the set of $k$-tuples in $S^2$ whose minimal tree is unknotted look like in the $k$-fold product $S^2\times\cdots\times S^2$? In particular, what is the measure of this set?
2. What is the minimum number $k$ for which there is a set of points in $S^2$ whose minimal tree is knotted?
3. (W. Kuperberg) What is the minimum number of vertices of order 3 in a knotted minimal tree for a finite subset of $S^2$? The described knotted tree has 6 vertices of order 3.
4. (G. Kuperberg) There is a finite set whose minimal tree is knotted on the surface of an ellipsoid with one of the axes much shorter than the other two axes. What are the strictly convex closed surfaces in ${\mathbb R}^3$ containing a finite set whose minimal tree is knotted? Can any knot be realized in a minimal tree of a finite set on some convex surface?
The author would like to thank Greg Kuperberg for helpful discussions and William Chen for bringing the problem to her attention.
[99]{} Rob Kirby, [*Problems in low-dimensional topology*]{}, 35–473, in: Geometric Topology [*by*]{} William H. Kazez, Studies in Advanced Mathematics, AMS/IP; Proceedings of the 1993 Georgia International Topology Conference, Athens, Georgia.
[^1]: This research was supported in part by NSF grant \# DMS-9401408.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The application of deep learning (DL) models to the decoding of cognitive states from whole-brain functional Magnetic Resonance Imaging (fMRI) data is often hindered by the small sample size and high dimensionality of these datasets. Especially, in clinical settings, where patient data are scarce. In this work, we demonstrate that transfer learning represents a solution to this problem. Particularly, we show that a DL model, which has been previously trained on a large openly available fMRI dataset of the Human Connectome Project, outperforms a model variant with the same architecture, but which is trained from scratch, when both are applied to the data of a new, unrelated fMRI task. Even further, the pre-trained DL model variant is already able to correctly decode 67.51% of the cognitive states from a test dataset with 100 individuals, when fine-tuned on a dataset of the size of only three subjects.'
author:
- 'Armin W. Thomas'
- |
\
Klaus-Robert M[ü]{}ller
- Wojciech Samek
bibliography:
- 'main.bib'
title: |
Deep Transfer Learning For\
Whole-Brain fMRI Analyses
---
Introduction
============
Over the recent years, deep learning (DL) methods have been shown to outperform more conventional machine learning techniques in a variety of decoding tasks (for a review, see [@lecun2015deep]). The success of DL methods is often attributed to their ability to autonomously learn highly abstracted representations of the raw input data, through a hierarchical sequence of non-linear transforms.
While researchers have started exploring the application of DL methods to the analysis of functional Magnetic Resonance Imaging (fMRI) data (e.g., [@plis2014deep]), their application to whole-brain fMRI data is still limited (e.g., [@huang2017modeling; @jang2017task]). Mainly, due to the small sample sizes and high dimensionality of fMRI datasets (and a lack of interpretability of DL models [@LapNCOMM19]). Particularly, in clinical settings, where fMRI datasets often only contain 10 - 20 patients and several hundred fMRI samples (i.e., volumes) per patient. Yet, each fMRI volume can easily contain several hundred thousand dimensions (i.e., voxels). In such classification settings, in which the number of data dimensions far exceeds the number of data samples, DL methods, as well as traditional machine learning approaches, are prone to overfitting (for a review, see [@lemm2011introduction]).
This problem has been similarly encountered in other research domains (e.g., [@oquab2014learning]). Here, researchers have discovered that the successful application of DL models to small datasets can strongly benefit from [*transfer learning*]{}. Transfer learning describes a process in which a model is trained on one dataset and subsequently applied to another dataset [@oquab2014learning]. Thereby, the knowledge about the first dataset, contained in the parameter estimates of the trained model, is utilized to benefit the application of the model to the second dataset. This procedure often drastically improves the classification performance of the model, while also reducing the amount of time and data required to train it.
In this work, we explore whether transfer learning is similarly beneficial for the application of DL models to the decoding of cognitive states (e.g., viewing the image of a face vs the image of a house) from fMRI data. In particular, we show that a DL model that has been trained on the data of six out of seven task-fMRI datasets of the Human Connectome Project database [@barch2013function] performs better in decoding the cognitive states underlying a seventh, unrelated task, when compared to a model variant that is trained entirely from scratch on the data of this task. For this comparison, we utilize the DeepLight framework [@thomas2018interpretable], which decodes a cognitive state from whole-brain fMRI data, by combining convolutional and recurrent DL elements (see Fig. \[fig1\] and Section \[sec\_deeplight\]).
Methods
=======
Data
----
### Experiment tasks {#sec_tasks}
We analyzed the fMRI data of 400 unrelated participants in the following seven experiment tasks (for further details, see Table 1 and [@barch2013function]):
- **Working Memory (WM):** Participants are asked to decide whether a currently presented image (of body parts, faces, places or tools) is the same as a previously presented target image.
- **Gambling:** Participants are asked to guess whether the value of a card (with values between 1-9) is below or above 5. Participants win or loose if they guess correctly/incorrectly. Trials are neutral if the value of the card is 5.
- **Motor:** Participants are presented with visual cues asking them to tap their left or right fingers, squeeze their left or right toes, or move their tongue.
- **Language:** Participants either hear a brief fable (story trials) or an arithmetic problem (math trials) and are subsequently given a two-alternative forced choice question about the story / arithmetic problem.
- **Social:** Participants are presented with short video clips of objects that either interact in some way or move randomly. Subsequently, participants are asked to decide whether the objects interacted with one another, did not have an interaction, or if they are not sure.
- **Relational:** Participants are presented with different shapes, filled with different textures. In relational trials, participants see a pair of objects at the top of the screen and a pair at the bottom. They are then asked to decide whether the bottom pair differs along the same dimension (shape or texture) as the top pair. In match trials, participants see one object at the top and bottom and are asked to decide whether the objects match on a specified dimension.
- **Emotion:** Participants are asked to decide which of two faces presented on the bottom of the screen matches the face at the top of the screen. The faces have an either angry or fearful expression.
Task Decoding targets Target count Duration (s)
------------ ------------------------------------------- -------------- --------------
WM body, face, place, tool 4 400
Gambling win, loss, neutral 3 224
Motor left/right finger, left/right toe, tongue 5 312
Language story, math 2 480
Social interaction, no interaction 2 200
Relational relational, matching 2 216
Emotion fear, neutral 2 252
**Total** **20** **2,084**
: Overview of the fMRI Data. For each experiment task, the decoding targets (i.e., cognitive states), the number of decoding targets, as well as the duration of the fMRI data that are included in the analysis are presented.[]{data-label="tab1"}
### FMRI data
All analyzed fMRI data were provided in a preprocessed format by the Human Connectome Project (HCP), WU Minn Consortium (Principal Investigators: David VanEssen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Center for Systems Neuroscience at Washington University. Whole-brain EPI acquisitions were acquired with a 32 channel head coil on a modified 3T Siemens Skyra with TR=720 ms and TE=33.1 ms (for further details on fMRI acquisition, see [@uugurbil2013pushing]).
### FMRI data preprocessing
The HCP preprocessing pipeline for fMRI data [@glasser2013minimal] includes the following steps: gradient unwarping, motion correction, fieldmap-based EPI distortion correction, brain-boundary based registration of EPI to structural T1-weighted scan, non-linear registration into MNI152 space, and grand-mean intensity normalization (for further details, see [@uugurbil2013pushing; @glasser2013minimal]). In addition, we applied the following preprocessing: volume-based smoothing with a 3mm Gaussian kernel, linear detrending and standardization of the single voxel signal time-series (resulting in a zero-centered voxel time-series with unit variance) and temporal filtering of the single voxel time-series with a butterworth highpass filter and a cutoff of 128s. We further excluded the first two TRs of every fMRI experiment block (for experiment details, see [@barch2013function]) from all analyses, as we did not expect any task-related hemodynamic response within this time period. Each fMRI volume contained 91 x 109 x 91 voxels (X x Y x Z).
### Data splitting {#sec_data_split}
We further divided the fMRI data into a distinct pre-training and test dataset, by assigning the data of the working memory task to the test data and all other experiment tasks to the pre-training data.
DeepLight {#sec_deeplight}
---------
![Illustration of the DeepLight framework [@thomas2018interpretable]. DeepLight first separates a whole-brain fMRI volume into a sequence of axial slices. Each axial slice is then processed by a convolutional feature extractor. The resulting sequence of higher-level axial slice representations is processed by a bi-directional LSTM unit, before a decoding prediction is made through a fully connected softmax output layer.[]{data-label="fig1"}](Figure_1_DeepLight_MLCN.png){width="\textwidth"}
DeepLight [@thomas2018interpretable] consists of three distinct computational modules (see Fig. \[fig1\]). Namely, a feature extractor, an LSTM unit and output layer. To decode a cognitive state, DeepLight first separates a whole-brain fMRI volume into a sequence of axial slices. These slices are then sequentially processed by a convoltional feature extractor. The feature extractor used here consists of the following 12 convolution layers [@lecun1995convolutional]: conv3-16(1), conv3-16(1), conv3-16(2), conv3-16(1), conv3-32(2), conv3-32(1), conv3-32(2). conv3-32(1), conv3-64(2), conv3-64(1), conv3-64(2), conv3-64(1) (notation: conv(kernel size) - (number of kernels)(stride size). All convolution kernels were activated through a rectified linear unit function. This sequence of convolution layers resulted in a 768-dimensional representation of each axial volume slice. To integrate the information provided by the resulting sequence of higher-level slice representations into a higher-level representation of the observed whole-brain activity, DeepLight applies a bi-directional LSTM [@hochreiter1997long], containing two independent LSTM units. Each of the two LSTM units contains 64 neurons and iterates through the entire sequence of input slices, but in reverse order (one from bottom-to-top and the other from top-to-bottom). Lastly, to make a decoding decision, DeepLight applies a fully-connected softmax output layer, containing one output neuron per decoding target in the data.
### Training {#sec_train}
All DeepLight variants that were used in this study were trained as follows (if not reported otherwise): We iteratively trained DeepLight through backpropagation, by the use of the ADAM optimization algorithm, as implemented in tensorflow 1.13. During parameter estimation, we applied dropout regularization to all network layers as follows: We set the dropout probability to 50 % for the LSTM unit and softmax output layer, For the convolution layers, however, we set the dropout probability to 0% for the first four convolution layers, 20% for the next four convolution layers, and 40% for the last four convolution layers (in line with [@thomas2018interpretable]). We further used a learning rate of $1e^{-4}$ and a batch size of 24 fMRI volumes. DeepLight’s weights were initialized by the use of a normal-distributed random initialization scheme [@glorot2010understanding].
Results
=======
Pre-training data {#sec_pretrain}
-----------------
![DeepLight pre-training statistics. A-B: Mean decoding accuracy in the training (A) and validation (B) data, as a function of training epochs. C: Mean decoding accuracy in the validation data after 40 training epochs. Lines represent grand means, surrounded by standard error bands. Bar heights indicate grand means, while scatter points indicate subject means. Colors indicate tasks. Dashed lines indicate chance level.[]{data-label="fig2"}](MICCAI_MLCN_figure_2.png){width="\textwidth"}
The goal of the first analysis was to pre-train DeepLight on the data of the six tasks contained in the pre-training dataset (see Section \[sec\_data\_split\]). To this end, we divided the data within each task into a distinct training and validation dataset, by assigning the data of 300 randomly selected subjects to the training data and the data of the remaining 100 subjects to the validation data. During pre-training, DeepLight’s output layer contained 16 neurons, one for each cognitive state of each task in the pre-training dataset (for an overview, see Table 1). Thereby, DeepLight has no knowledge of the underlying tasks and is able to identify an individual’s cognitive state without knowing which task the individual performed. Overall, we trained DeepLight for a period of 40 epochs (Fig. \[fig2\]). Each epoch was defined as an iteration over the entire training data.
After 40 training epochs, DeepLight achieved an average decoding accuracy of 76.04% in the training dataset (Fig. \[fig2\] A) and a decoding accuracy of 70.55% in the left-out validation data (Fig. \[fig2\] B, C). Interestingly, DeepLight’s average decoding accuracy was between 70.00-86.12% for five out of the six tasks in the validation data, while the decoding accuracy for the sixth task (the gambling task) was only 28.18%. When excluding the data of the gambling task from the decoding analysis, DeepLight’s average decoding accuracy increased to 84.56% in the training dataset and 79.02% in the validation data.
Test data
---------
![Comparison of a “pre-trained” DeepLight variant with a “not pre-trained” variant that is trained entirely from scratch, when both are applied to subsets of 1%, 5%, 10%, 20%, 40%, 60%, and 100% of the full training dataset ($N=300$) of the test task (the working memory task). A, B, D, E: Decoding accuracy as a function of training epochs in the training (A-B) and validation data (D-E). C, F: Difference in decoding accuracy between the pre-trained and not pre-trained DeepLight variant after 50 training epochs. Stars indicate a statistically meaningful difference in a t-test using Bonferroni adjusted alpha levels of 0.05/7. Colors indicate the fraction of training data that is used. Lines show grand means with standard error bands surrounding them. Bar heights indicate grand means. Scatter points indicate subject means.[]{data-label="fig3"}](MICCAI_MLCN_figure_3.png){width="\textwidth"}
The goal of the second analysis was to explore the benefits of transfer learning for the application of DL models to fMRI data. To this end, we compared the performance of the pre-trained DeepLight variant (see Section \[sec\_pretrain\]) to that of a variant that was trained entirely from scratch, when both are applied to the data of the left-out test task (the working memory task, see Section \[sec\_data\_split\]). We again divided the data of the test task into a separate training and validation dataset, by randomly assigning 300 distinct subjects to the training data and the remaining 100 to the validation data. We then trained a DeepLight variant with transfer learning, and one without, on the training data of the test task. The output layer of both variants was set to contain four neurons (one per decoding target in the working memory task, see Table 1). Otherwise, the architecture and training procedures (see Section \[sec\_train\]) of both variants were identical.
The first variant (“not pre-trained”) does not apply transfer learning and was trained entirely from scratch, with weights initialized according to the normal-distributed random initialization scheme [@glorot2010understanding]. After 50 training epochs, this variant achieved an average decoding accuracy of 88.57% in the training data of the test task (Fig. \[fig3\] A) and 81.91% in the validation data (Fig. \[fig3\] D). The second variant (“pre-trained”) applies transfer learning and is based on the DeepLight variant that we previously trained on the pre-training dataset (see Section \[sec\_data\_split\] and \[sec\_pretrain\]). Particularly, we initialized the parameters of all network layers, except for the output layer (Fig. \[fig1\]), to those weights obtained by the pre-trained DeepLight variant and only initialized the weights of the output layer according to the normal-distributed random initialization scheme [@glorot2010understanding]. After 50 training epochs, the pre-trained variant achieved an average decoding accuracy of 92.43% in the training data of the test task (Fig. \[fig3\] B) and 83.83% in the validation data (Fig. \[fig3\] E) and thereby performed meaningfully better in decoding the cognitive states from the validation data than the not pre-trained variant ($t(99)=8.42, p < 0.0001$), Fig. \[fig3\] F).
We were further interested in exploring how both DeepLight variants performed, when trained on smaller fractions of the original training dataset of the test task. Therefore, we repeatedly trained both variants on 1%, 5%, 10%, 20%, 40% and 60% of the full training dataset of the test task ($N=300$), while evaluating their performance on the full validation data of the test task ($N=100$). Overall, the pre-trained variant consistently achieved higher decoding accuracies in the training (Fig. \[fig3\] C) and validation (Fig. \[fig3\] F) data, and required less training time, when compared to the not pre-trained variant. Importantly, the pre-trained DeepLight variant already achieved an average decoding accuracy of 67.51% (Fig. \[fig3\] E) in the validation data, when being trained on only 1% of the training dataset (equal to the data of three subjects). The not pre-trained variant, on the other hand, achieved a decoding accuracy of only 32.49% (Fig. \[fig3\] D), when being trained on 1% of the training data and thereby performed meaningfully worse (the pre-trained DeepLight variant outperformed the not pre-trained variant by 35.02% ($t(99)=49.68, p<0.0001$)). Lastly, we also tested how much of the training data the pre-trained DeepLight variant requires to performs as well as (or better than) the not pre-trained variant that has been trained on the full training data. Interestingly, the pre-trained variant already achieved a meaningfully better decoding accuracy than the not pre-trained variant (which was trained on the full training dataset), when the pre-trained variant was trained on only 40% of the training data ($t(99)=2.82, p=0.0057$).
Conclusion
==========
The broad application of DL models to fMRI data has long been hindered by the small sample size, and high dimensionality, of typical fMRI datasets. Here, we have demonstrated that transfer learning is strongly beneficial for the application of DL models to small fMRI datasets. A DL model that has been pre-trained on a large, openly available fMRI dataset, generally requires less training data and time, and achieves higher decoding accuracies, when compared to a model variant with the same architecture that is trained entirely from scratch. Even further, the pre-trained model variant already performs well in decoding the cognitive states of 100 individuals in an unrelated fMRI task, when fine-tuned on a dataset of the size of only three subjects.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We extend a few recent results of Górnicki (2011) asserting that the set of fixed points of an asymptotically regular mapping is a retract of its domain. In particular, we prove that in some cases the resulting retraction is Hölder continuous. We also characterise Bynum’s coefficients and the Opial modulus in terms of nets.'
address: 'Andrzej Wiśnicki, Institute of Mathematics, Maria Curie-Skłodowska University, 20-031 Lublin, Poland'
author:
- Andrzej Wiśnicki
title: 'On the structure of fixed-point sets of asymptotically regular semigroups'
---
Introduction.
=============
The notion of asymptotic regularity, introduced by Browder and Petryshyn in [@BrPe], has become a standing assumption in many results concerning fixed points of nonexpansive and more general mappings. Recall that a mapping $T:M\rightarrow M$ acting on a metric space $(M,d)$ is said to be asymptotically regular if $$\lim_{n\rightarrow \infty }d(T^{n}x,T^{n+1}x)=0$$for all $x\in M.$ Ishikawa [@Is] proved that if $C$ is a bounded closed convex subset of a Banach space $X$ and $T:C\rightarrow C$ is nonexpansive, then the mapping $T_{\lambda }=(1-\lambda )I+\lambda T$ is asymptotically regular for each $\lambda \in (0,1).$ Edelstein and O’Brien [@EdOb] showed independently that $T_{\lambda }$ is uniformly asymptotically regular over $x\in C,$ and Goebel and Kirk [@GoKi3] proved that the convergence is even uniform with respect to all nonexpansive mappings from $C$ into $C$. Other examples of asymptotically regular mappings are given by the result of Anzai and Ishikawa [@AnIs] (see also [@XuYa]): if $T$ is an affine mapping acting on a bounded closed convex subset of a locally convex space $X $, then $T_{\lambda }=(1-\lambda )I+\lambda T$ is uniformly asymptotically regular.
In 1987, Lin [@Li] constructed a uniformly asymptotically regular Lipschitz mapping in $\ell _{2}$ without fixed points which extended an earlier construction of Tingley [@Ti]. Subsequently, Maluta, Prus and Wośko [@MaPrWo] proved that there exists a continuous fixed-point free asymptotically regular mapping defined on any bounded convex subset of a normed space which is not totally bounded (see also [@Er]). For the fixed-point existence theorems for asymptotically regular mappings we refer the reader to the papers by T. Domínguez Benavides, J. Górnicki, M. A. Japón Pineda and H. K. Xu (see [@DoJa; @DoXu; @Go1]).
It was shown in [@SeWi] that the set of fixed points of a k-uniformly Lipschitzian mapping in a uniformly convex space is a retract of its domain if $k$ is close to $1$. In recent papers [@GoT; @GoTai; @GoN], J. Górnicki proved several results concerning the structure of fixed-point sets of asymptotically regular mappings in uniformly convex spaces. In this paper we continue this work and extend a few of Górnicki’s results in two aspects: we consider a more general class of spaces and prove that in some cases, the fixed-point set $\mathrm{Fix\,}T$ is not only a (continuous) retract but even a Hölder continuous retract of the domain. We present our results in a more general case of a one-parameter nonlinear semigroup. We also characterise Bynum’s coefficients and the Opial modulus in terms of nets.
Preliminaries
=============
Let $G$ be an unbounded subset of $[0,\infty )$ such that $t+s,t-s\in G$ for all $t,s\in G$ with $t>s$ (e.g., $G=[0,\infty )$ or $G=\mathbb{N}$). By a nonlinear semigroup on $C$ we shall mean a one-parameter family of mappings $\mathcal{T}=\{T_{t}:t\in G\}$ from $C$ into $C$ such that $T_{t+s}x=T_{t}\,T_{s}x$ for all $t,s\in G$ and $x\in C$. In particular, we do not assume in this paper that $\{T_{t}:t\in G\}$ is strongly continuous. We use a symbol $|T|$ to denote the exact Lipschitz constant of a mapping $T:C\rightarrow C$, i.e., $$|T|=\inf \{k:\Vert Tx-Ty\Vert \leq k\Vert x-y\Vert \ \text{for\ all}\ x,y\in
C\}.$$If $T$ is not Lipschitzian we define $|T|=\infty $.
A semigroup $\mathcal{T}=\{T_{t}:t\in G\}$ from $C$ into $C$ is said to be asymptotically regular if $\lim_{t}\left\Vert T_{t+h}x-T_{t}x\right\Vert =0$ for every $x\in C$ and $h\in G.$
Assume now that $C$ is convex and weakly compact and $\mathcal{T}=\{T_{t}:t\in G\}$ is a nonlinear semigroup on $C$ such that $s(\mathcal{T})=\liminf_{t}|T_{t}|<\infty .$ Choose a sequence $(t_{n})$ of elements in $G$ such that $\lim_{n\rightarrow \infty }t_{n}=\infty $ and $s(\mathcal{T})=\lim_{n\rightarrow \infty }\left\vert T_{t_{n}}\right\vert .$ By Tikhonov’s theorem, there exists a pointwise weakly convergent subnet $(T_{t_{n_{\alpha }}})_{\alpha \in \emph{A}}$ of $(T_{t_{n}}).$ We denote it briefly by $(T_{t_{\alpha }})_{\alpha \in \emph{A}}.$ For every $x\in C$, define $$Lx=w\text{-}\lim_{\alpha }T_{t_{\alpha }}x, \label{Lx}$$i.e., $Lx$ is the weak limit of the net $(T_{t_{\alpha }}x)_{\alpha \in
\emph{A}}$. Notice that $Lx$ belongs to $C$ since $C$ is convex and weakly compact. The weak lower semicontinuity of the norm implies $$\Vert Lx-Ly\Vert \leq \liminf_{\alpha }\Vert T_{t_{\alpha }}x-T_{t_{\alpha
}}y\Vert \leq \limsup_{n\rightarrow \infty }\Vert T_{t_{n}}x-T_{t_{n}}y\Vert
\leq s(\mathcal{T})\Vert x-y\Vert .$$We formulate the above observation as a separate lemma.
\[nonexp\]Let $C$ be a convex weakly compact subset of a Banach space $X$ and let $\mathcal{T}=\{T_{t}:t\in G\}$ be a semigroup on $C$ such that $s(\mathcal{T})=\liminf_{t}|T_{t}|<\infty .$ Then the mapping $L:C\rightarrow C$ defined by (\[Lx\]) is $s(\mathcal{T})$-Lipschitz.
We end this section with the following variant of a well known result which is crucial for our work (see, e.g., [@BeLi Prop. 1.10]).
\[holder\]Let $(X,d)$ be a complete bounded metric space and let $L:X\rightarrow X$ be a k-Lipschitz mapping. Suppose there exists $0<\gamma
<1 $ and $c>0$ such that $\Vert L^{n+1}x-L^{n}x\Vert \leq c\gamma ^{n}$ for every $x\in X$. Then $Rx=\lim_{n\rightarrow \infty }L^{n}x$ is a Hölder continuous mapping.
We may assume that $\operatorname{diam}X<1$. Fix $x\neq y$ in $X$ and notice that for any $n\in \mathbb{N}$, $$d(Rx,Ry)\leq d(Rx,L^{n}x)+d(L^{n}x,L^{n}y)+d(L^{n}y,Ry)\leq 2c\frac{\gamma
^{n}}{1-\gamma }+k^{n}d(x,y).$$Take $\alpha <1$ such that $k\leq \gamma ^{1-\alpha ^{-1}}$ and put $\gamma
^{n-r}=d(x,y)^{\alpha }$ for some $n\in \mathbb{N}$ and $0<r\leq 1$. Then $k^{n-1}\leq (\gamma ^{1-\alpha ^{-1}})^{n-r}$ and hence $$d(Rx,Ry)\leq 2c\frac{\gamma ^{n-r}}{1-\gamma }+k(\gamma ^{n-r})^{1-\alpha
^{-1}}d(x,y)=(\frac{2c}{1-\gamma }+k)d(x,y)^{\alpha }.$$
Bynum’s coefficients and Opial’s modulus in terms of nets
=========================================================
From now on, $C$ denotes a nonempty convex weakly compact subset of a Banach space $X$. Let $\mathcal{A}$ be a directed set, $(x_{\alpha })_{\alpha \in
\mathcal{A}}$ a bounded net in $X$, $y\in X$ and write $$\begin{aligned}
r(y,(x_{\alpha }))& =\limsup_{\alpha }\Vert x_{\alpha }-y\Vert , \\
r(C,(x_{\alpha }))& =\inf \{r(y,(x_{\alpha })):y\in C\}, \\
A(C,(x_{\alpha }))& =\{y\in C:r(y,(x_{\alpha }))=r(C,(x_{\alpha }))\}.\end{aligned}$$
The number $r(C,(x_{\alpha }))$ and the set $A(C,(x_{\alpha }))$ are called, respectively, the asymptotic radius and the asymptotic center of $(x_{\alpha
})_{\alpha \in \mathcal{A}}$ relative to $C$. Notice that $A(C,(x_{\alpha
})) $ is nonempty convex and weakly compact. Write $$r_{a}(x_{\alpha })=\inf \{\limsup_{\alpha }\Vert x_{\alpha }-y\Vert :y\in
\overline{\operatorname{conv}}(\{x_{\alpha }:\alpha \in \mathcal{A}\})\}$$and let $$\operatorname{diam}_{a}(x_{\alpha })=\inf_{\alpha }\sup_{\beta ,\gamma \geq \alpha }\Vert
x_{\beta }-x_{\gamma }\Vert$$denote the asymptotic diameter of $(x_{\alpha })$.
The normal structure coefficient $\operatorname{N}(X)$ of a Banach space $X$ is defined by $$\operatorname{N}(X)=\sup \left\{ k:k\,r(K)\leq \operatorname{diam}K\ \ \text{for\ each\ bounded\
convex\ set}\ K\subset X\right\} ,$$where $r(K)=\inf_{y\in K}\sup_{x\in K}\Vert x-y\Vert $ is the Chebyshev radius of $K$ relative to itself. Assuming that $X$ does not have the Schur property, the weakly convergent sequence coefficient (or Bynum’s coefficient) is given by $$\operatorname{WCS}(X)=\sup \left\{ k:k\,r_{a}(x_{n})\leq \operatorname{diam}_{a}(x_{n})\ \ \text{for\ each\
sequence}\ x_{n}\overset{w}{\longrightarrow }0\right\} ,$$where $x_{n}\overset{w}{\longrightarrow }0$ means that $(x_{n})$ is weakly null in $X$ (see [@By]). For Schur spaces, we define $WCS(X)=2$.
It was proved independently in [@DoLop; @Pr; @Zh] that
$$WCS(X)=\sup \left\{ k:k\,\limsup_{n}\Vert x_{n}\Vert \leq \operatorname{diam}_{a}(x_{n})\
\text{for each\ sequence}\ x_{n}\overset{w}{\longrightarrow }0\right\}
\label{wcs1}$$
and, in [@DoLoXu], that $$WCS(X)=\sup \left\{ k:k\,\limsup_{n}\Vert x_{n}\Vert \leq D[(x_{n})]\ \text{for\ each\ sequence}\ x_{n}\overset{w}{\longrightarrow }0\right\} ,$$where $D[(x_{n})]=\limsup_{m}\limsup_{n}\left\Vert x_{n}-x_{m}\right\Vert
.\smallskip $
Kaczor and Prus [@KaPr] initiated a systematic study of assumptions under which one can replace sequences by nets in a given condition. We follow the arguments from that paper and use the well known method of constructing basic sequences attributed to S. Mazur (see [@Pe]). Let us first recall a variant of a classical lemma which can be proved in the same way as for sequences (see, e.g., [@Pe Lemma]).
\[Ma\] Let $\{x_{\alpha }\}_{\alpha \in \mathcal{A}}$ be a bounded net in $X$ weakly converging to $0$ such that $\inf_{\alpha }\Vert x_{\alpha
}\Vert >0$. Then for every $\varepsilon >0$, $\alpha ^{\prime }\in \mathcal{A}$ and for every finite dimensional subspace $E$ of $X$, there is $\alpha
>\alpha ^{\prime }$ such that $$\Vert e+tx_{\alpha }\Vert \geq (1-\varepsilon )\Vert e\Vert$$for any $e\in E$ and every scalar $t.$
Recall that a sequence $(x_{n})$ is basic if and only if there exists a number $c>0$ such that $\Vert \sum_{i=1}^{q}t_{i}x_{i}\Vert \leq c\Vert
\sum_{i=1}^{p}t_{i}x_{i}\Vert $ for any integers $p>q\geq 1$ and any sequence of scalars $(t_{i})$. In the proof of the next lemma, based on Mazur’s technique, we follow in part the reasoning given in [KaPr]{}. Set $D[(x_{\alpha })]=\limsup_{\alpha }\limsup_{\beta }\left\Vert
x_{\alpha }-x_{\beta }\right\Vert .$
\[KaPr\] Let $(x_{\alpha })_{\alpha \in \mathcal{A}}$ be a bounded net in $X$ which converges to $0$ weakly but not in norm. Then there exists an increasing sequence $(\alpha _{n})$ of elements of $\mathcal{A}$ such that $\lim_{n}\Vert x_{\alpha _{n}}\Vert =\limsup_{\alpha }\Vert x_{\alpha }\Vert $, $\operatorname{diam}_{a}(x_{\alpha _{n}})\leq D[(x_{\alpha })]$ and $(x_{\alpha _{n}})$ is a basic sequence.
Since $(x_{\alpha })_{\alpha \in \mathcal{A}}$ does not converge strongly to $0$ and $D[(x_{\alpha _{s}})]\leq D[(x_{\alpha })]$ for any subnet $(x_{\alpha _{s}})_{s\in \mathcal{B}}$ of $(x_{\alpha })_{\alpha \in \mathcal{A}}$, we can assume, passing to a subnet, that $\inf_{\alpha }\Vert
x_{\alpha }\Vert >0$ and the limit $c=\lim_{\alpha }\Vert x_{\alpha }\Vert $ exists. Write $d=D[(x_{\alpha })]$. Let $(\varepsilon _{n})$ be a sequence of reals from the interval $(0,1)$ such that $\Pi _{n=1}^{\infty
}(1-\varepsilon _{n})>0$. We shall define the following sequences $(\alpha
_{n})$ and $(\beta _{n})$ by induction.
Let us put $\alpha _{1}<\beta _{1}\in \mathcal{A}$ such that $\left\vert
\Vert x_{\alpha _{1}}\Vert -c\right\vert <1$ and $\sup_{\beta \geq \beta
_{1}}\Vert x_{\alpha _{1}}-x_{\beta }\Vert <d+1$. By the definitions of $c$ and $d$, there exists $\alpha ^{\prime }>\beta _{1}$ such that $\left\vert
\Vert x_{\alpha }\Vert -c\right\vert <\frac{1}{2}$ and $\inf_{\beta ^{\prime
}}\sup_{\beta \geq \beta ^{\prime }}\Vert x_{\alpha }-x_{\beta }\Vert <d+\frac{1}{2}$ for every $\alpha \geq \alpha ^{\prime }.$ It follows from Lemma \[Ma\] that there exists $\alpha _{2}>\alpha ^{\prime }$ such that$$\Vert t_{1}x_{\alpha _{1}}+t_{2}x_{\alpha _{2}}\Vert \geq (1-\varepsilon
_{2})\Vert t_{1}x_{\alpha _{1}}\Vert$$for any scalars $t_{1},t_{2}.$ Furthermore, $\left\vert \Vert x_{\alpha
_{2}}\Vert -c\right\vert <\frac{1}{2},$ and we can find $\beta _{2}>\alpha
_{2}$ such that $\sup_{\beta \geq \beta _{2}}\Vert x_{\alpha _{2}}-x_{\beta
}\Vert <d+\frac{1}{2}.$
Suppose now that we have chosen $\alpha _{1}<\beta _{1}<...<\alpha
_{n}<\beta _{n}$ $(n>1)$ in such a way that $\left\vert \Vert x_{\alpha
_{k}}\Vert -c\right\vert <\frac{1}{k}$, $\sup_{\beta \geq \beta _{k}}\Vert
x_{\alpha _{k}}-x_{\beta }\Vert <d+\frac{1}{k}$ and $$(1-\varepsilon _{k})\Vert t_{1}x_{\alpha _{1}}+...+t_{k-1}x_{\alpha
_{k-1}}\Vert \leq \Vert t_{1}x_{\alpha _{1}}+...+t_{k}x_{\alpha _{k}}\Vert$$for any scalars $t_{1},...,t_{k}$, $k=2,...,n.$ From the definitions of $c$ and $d$, and by Lemma \[Ma\], we can find $\beta _{n+1}>\alpha
_{n+1}>\beta _{n}$ such that $\left\vert \Vert x_{\alpha _{n+1}}\Vert
-c\right\vert <\frac{1}{n+1}$, $\sup_{\beta \geq \beta _{n+1}}\Vert
x_{\alpha _{n+1}}-x_{\beta }\Vert \leq d+\frac{1}{n+1}$ and (considering a subspace $E$ spanned by the elements $x_{\alpha _{1}},...,x_{\alpha _{n}}$ and putting $e=t_{1}x_{\alpha _{1}}+...+t_{n}x_{\alpha _{n}}$), $$(1-\varepsilon _{n+1})\Vert t_{1}x_{\alpha _{1}}+...+t_{n}x_{\alpha
_{n}}\Vert \leq \Vert t_{1}x_{\alpha _{1}}+...+t_{n+1}x_{\alpha _{n+1}}\Vert$$for any scalars $t_{1},...,t_{n+1}$.
Notice that the sequence $(x_{\alpha _{n}})$ defined in this way satisfies $\lim_{n\rightarrow \infty }\Vert x_{\alpha _{n}}\Vert =c$ and $\operatorname{diam}_{a}(x_{\alpha _{n}})\leq d$. Furthermore, $$\Vert t_{1}x_{\alpha _{1}}+...+t_{p}x_{\alpha _{p}}\Vert \geq \Pi
_{n=q+1}^{p}(1-\varepsilon _{n})\Vert t_{1}x_{\alpha
_{1}}+...+t_{q}x_{\alpha _{q}}\Vert$$for any integers $p>q\geq 1$ and any sequence of scalars $(t_{i})$. Hence $(x_{\alpha _{n}})$ is a basic sequence.
We are now in a position to give a characterization of the coefficient WCS(X) in terms of nets. The abbreviation $\left\{
x_{\alpha }\right\} $ is r.w.c. means that the set $\left\{ x_{\alpha }:\alpha \in \mathcal{A}\right\} $ is relatively weakly compact.
\[Wi1\] Let $X$ be a Banach space without the Schur property and write$$\begin{aligned}
w_{1}& =\sup \left\{ k:k\,r_{a}(x_{\alpha })\leq \operatorname{diam}_{a}(x_{\alpha })\
\text{ for\ each\ net}\ x_{\alpha }\overset{w}{\longrightarrow }0,\text{ }\left\{ x_{\alpha }\right\} \text{ is r.w.c.}\right\} , \\
w_{2}& =\sup \left\{ k:k\,\limsup_{\alpha }\Vert x_{\alpha }\Vert \leq \operatorname{diam}_{a}(x_{\alpha })\ \text{for\ each\ net}\ x_{\alpha }\overset{w}{\longrightarrow }0,\text{ }\left\{ x_{\alpha }\right\} \text{ is r.w.c.}\right\} , \\
w_{3}& =\sup \left\{ k:k\,\limsup_{\alpha }\Vert x_{\alpha }\Vert \leq
D[(x_{\alpha })]\ \text{for\ each\ net}\ x_{\alpha }\overset{w}{\longrightarrow }0,\text{ }\left\{ x_{\alpha }\right\} \text{ is r.w.c.}\right\} .\end{aligned}$$Then $$\operatorname{WCS}(X)=w_{1}=w_{2}=w_{3}.$$
Fix $k>w_{3}$ and choose a weakly null net $(x_{\alpha })$ such that the set $\left\{ x_{\alpha }:\alpha \in \mathcal{A}\right\} $ is relatively weakly compact and $k\,\limsup_{\alpha }\Vert x_{\alpha }\Vert >D[(x_{\alpha })].$ Then, by Lemma \[KaPr\], there exists an increasing sequence $(\alpha
_{n}) $ such that $$k\,\lim_{n}\Vert x_{\alpha _{n}}\Vert >D[(x_{\alpha })]\geq \operatorname{diam}_{a}(x_{\alpha _{n}})$$ and $(x_{\alpha _{n}})$ is a basic sequence. Since the set $\left\{
x_{\alpha }:\alpha \in \mathcal{A}\right\} $ is relatively weakly compact, we can assume (passing to a subsequence) that $(x_{\alpha _{n}})$ is weakly convergent. Since it is a basic sequence, its weak limit equals zero. It follows from (\[wcs1\]) that $\operatorname{WCS}(X)\leq k$ and letting $k$ go to $w_{3}$ we have $$\operatorname{WCS}(X)\leq w_{3}\leq w_{2}\leq w_{1}\leq
\operatorname{WCS}(X).$$
Notice that a similar characterisation holds for the normal structure coefficient.
For a Banach space $X$,$$\operatorname{N}(X)=\sup \left\{ k:k\,r_{a}(x_{\alpha })\leq \operatorname{diam}_{a}(x_{\alpha })\
\text{for\ each bounded net}\ (x_{\alpha })\text{ in }X\right\} .$$
Let $$N_{1}=\sup \left\{ k:k\,r_{a}(x_{\alpha })\leq \operatorname{diam}_{a}(x_{\alpha })\ \text{for\ each bounded net}\ (x_{\alpha })\text{ in }X\right\} .$$Set $k>N_{1}$ and choose a bounded net $(x_{\alpha })$ such that $k\,r_{a}(x_{\alpha })>\operatorname{diam}_{a}(x_{\alpha }).$ Fix $y\in \overline{\operatorname{conv}}(\{x_{\alpha }:\alpha \in \mathcal{A}\})$ and notice that $k\,\limsup_{\alpha }\Vert x_{\alpha }-y\Vert >\operatorname{diam}_{a}(x_{\alpha }).$ In a straightforward way, we can choose a sequence $(\alpha _{n})$ such that $$k\,\lim_{n}\Vert x_{\alpha _{n}}-y\Vert =k\,\limsup_{\alpha }\Vert x_{\alpha
}-y\Vert >\operatorname{diam}_{a}(x_{\alpha })\geq \operatorname{diam}_{a}(x_{\alpha _{n}}).$$It follows from [@By Th. 1] that $\operatorname{N}(X)\leq k$ and letting $k$ go to $N_{1}$ we have $\operatorname{N}(X)\leq N_{1}.$ By [@Lim Th. 1], $\operatorname{N}(X)\geq N_{1}$ and the proof is complete.
In the next section we shall need a similar characterisation for the Opial modulus of a Banach space $X,$ defined for each $c\geq 0$ by $$r_{X}(c)=\inf \left\{ \liminf_{n\rightarrow \infty }\left\Vert
x_{n}+x\right\Vert -1\right\} ,$$where the infimum is taken over all $x\in X$ with $\left\Vert x\right\Vert
\geq c$ and all weakly null sequences $(x_{n})$ in $X$ such that $\liminf_{n\rightarrow \infty }\left\Vert x_{n}\right\Vert \geq 1$ (see [LiTaXu]{}). We first prove the following counterpart of Lemma \[KaPr\].
\[KaPr2\] Let $(x_{\alpha })_{\alpha \in \mathcal{A}}$ be a bounded net in $X$ which converges to $0$ weakly but not in norm and $x\in X.$ Then there exists an increasing sequence $(\alpha _{n})$ of elements of $\mathcal{A}$ such that $\lim_{n}\Vert x_{\alpha _{n}}+x\Vert =\liminf_{\alpha }\Vert
x_{\alpha }+x\Vert ,$ $\lim_{n}\Vert x_{\alpha _{n}}\Vert \geq
\liminf_{\alpha }\Vert x_{\alpha }\Vert $ and $(x_{\alpha _{n}})$ is a basic sequence.
Since $(x_{\alpha })_{\alpha \in \mathcal{A}}$ does not converge strongly to $0$ and $$\liminf_{s}\Vert x_{\alpha _{s}}\Vert \geq \liminf_{\alpha }\Vert x_{\alpha
}\Vert$$for any subnet $(x_{\alpha _{s}})_{s\in \mathcal{B}}$ of $(x_{\alpha
})_{\alpha \in \mathcal{A}}$, it is sufficient (passing to a subnet) to consider only the case that $\inf_{\alpha }\Vert x_{\alpha }\Vert >0$ and the limits $c_{1}=\liminf_{\alpha }\Vert x_{\alpha }+x\Vert $, $c_{2}=\liminf_{\alpha }\Vert x_{\alpha }\Vert $ exist. Let $(\varepsilon
_{n})$ be a sequence of reals from the interval $(0,1)$ such that $\Pi
_{n=1}^{\infty }(1-\varepsilon _{n})>0$. We shall define the sequence $(\alpha _{n})$ by induction.
Let us put $\alpha _{1}\in \mathcal{A}$ such that $\left\vert \Vert
x_{\alpha _{1}}+x\Vert -c_{1}\right\vert <1$ and $\left\vert \Vert x_{\alpha
_{1}}\Vert -c_{2}\right\vert <1$. By the definitions of $c_{1}$ and $c_{2}$, there exists $\alpha ^{\prime }>\alpha _{1}$ such that $\left\vert \Vert
x_{\alpha }+x\Vert -c_{1}\right\vert <\frac{1}{2}$ and $\left\vert \Vert
x_{\alpha }\Vert -c_{2}\right\vert <\frac{1}{2}$ for every $\alpha \geq
\alpha ^{\prime }.$ It follows from Lemma \[Ma\] that there exists $\alpha
_{2}>\alpha ^{\prime }$ such that$$\Vert t_{1}x_{\alpha _{1}}+t_{2}x_{\alpha _{2}}\Vert \geq (1-\varepsilon
_{2})\Vert t_{1}x_{\alpha _{1}}\Vert$$for any scalars $t_{1},t_{2}.$ We can now proceed analogously to the proof of Lemma \[KaPr\] to obtain a basic sequence $(x_{\alpha _{n}})$ with the desired properties.
\[Wi2\]For a Banach space $X$ without the Schur property and for $c\geq
0,$$$r_{X}(c)=\inf \left\{ \liminf_{\alpha }\left\Vert x_{\alpha }+x\right\Vert
-1\right\} ,$$where the infimum is taken over all $x\in X$ with $\left\Vert x\right\Vert
\geq c$ and all weakly null nets $(x_{\alpha })$ in $X$ such that $\liminf_{\alpha }\Vert x_{\alpha }\Vert \geq 1$ and the set $\left\{
x_{\alpha }:\alpha \in \mathcal{A}\right\} $ is relatively weakly compact.
Let $r_{1}(c)=\inf \left\{ \liminf_{\alpha }\left\Vert x_{\alpha
}+x\right\Vert -1\right\} ,$ where the infimum is taken as above. Fix $c\geq
0$ and take $k>r_{1}(c).$ Then there exist $x\in X$ with $\left\Vert
x\right\Vert \geq c$ and a weakly null net $(x_{\alpha })_{\alpha \in
\mathcal{A}}$ such that $\liminf_{\alpha }\Vert x_{\alpha }\Vert \geq 1,$ $\left\{ x_{\alpha }:\alpha \in \mathcal{A}\right\} $ is relatively weakly compact and $$\liminf_{\alpha }\left\Vert x_{\alpha }+x\right\Vert -1<k.$$By Lemma \[KaPr2\], there exists an increasing sequence $(\alpha _{n})$ of elements of $\mathcal{A}$ such that $\lim_{n}\Vert x_{\alpha _{n}}\Vert \geq
1,\lim_{n}\Vert x_{\alpha _{n}}+x\Vert -1<k$ and $(x_{\alpha _{n}})$ is a basic sequence. Since $\left\{ x_{\alpha }:\alpha \in \mathcal{A}\right\} $ is relatively weakly compact, we can assume (passing to a subsequence) that $(x_{\alpha _{n}})$ is weakly null. Hence $r_{X}(c)<k$ and since $k$ is an arbitrary number greater than $r_{1}(c)$, it follows that $r_{X}(c)\leq
r_{1}(c).$ The reverse inequality is obvious.
Fixed-point sets as Hölder continuous retracts
==============================================
The following lemma may be proved in a similar way to [@DoJaLo Th. 7.2 ].
\[main\]Let $C$ be a nonempty convex weakly compact subset of a Banach space $X$ and $\mathcal{T}=\{T_{t}:t\in G\}$ an asymptotically regular semigroup on $C$ such that $s(\mathcal{T})=\lim_{\alpha }\left\vert
T_{t_{\alpha }}\right\vert $ for a pointwise weakly convergent subnet $(T_{t_{\alpha }})_{\alpha \in \emph{A}}$ of $(T_{t})_{t\in G}.$ Let $x_{0}\in C$, $x_{m+1}=w$-$\lim_{\alpha }T_{t_{\alpha }}x_{m},m=0,1,...,$ and $$B_{m}=\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m+1}\right\Vert .$$Assume that
1. $s(\mathcal{T})<\sqrt{\operatorname{WCS}(X)}$ or,
2. $s(\mathcal{T})<1+r_{X}(1).$
Then, there exists $\gamma <1$ such that $B_{m}\leq \gamma
B_{m-1} $ for any $m=1,2,...$.
It follows from the asymptotic regularity of $\{T_{t}:t\in G\}$ that $$\limsup_{\alpha }\left\Vert T_{t_{\alpha }-l}\,x-y\right\Vert
=\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x-y\right\Vert$$for any $l\in G$ and $x,y\in C$. Thus$$\begin{aligned}
& D[(T_{t_{\alpha }}x_{m})]=\limsup_{\beta }\limsup_{\alpha }\left\Vert
T_{t_{\alpha }}x_{m}-T_{t_{\beta }}x_{m}\right\Vert \\
& \ \leq \limsup_{\beta }\left\vert T_{t_{\beta }}\right\vert
\limsup_{\alpha }\left\Vert T_{t_{\alpha }-t_{\beta }}x_{m}-x_{m}\right\Vert
=s(\mathcal{T})\limsup_{\alpha }\left\Vert T_{t_{\alpha
}}x_{m}-x_{m}\right\Vert .\end{aligned}$$Hence, from Theorem \[Wi1\] and from the weak lower semicontinuity of the norm,$$\begin{aligned}
B_{m}& \leq \frac{D[(T_{t_{\alpha }}x_{m})]}{\operatorname{WCS}(X)}\leq \frac{s(\mathcal{T})}{\operatorname{WCS}(X)}\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m}\right\Vert
\\
& \leq \frac{s(\mathcal{T})}{\operatorname{WCS}(X)}\limsup_{\alpha }\liminf_{\beta
}\left\Vert T_{t_{\alpha }}x_{m}-T_{t_{\beta }}x_{m-1}\right\Vert \\
& \leq \frac{s(\mathcal{T})}{\operatorname{WCS}(X)}\limsup_{\alpha }\left\vert
T_{t_{\alpha }}\right\vert \limsup_{\beta }\left\Vert x_{m}-T_{t_{\beta
}-t_{\alpha }}x_{m-1}\right\Vert =\frac{(s(\mathcal{T}))^{2}}{\operatorname{WCS}(X)}B_{m-1}.\end{aligned}$$This gives (a). For (b), we can use Theorem \[Wi2\] and proceed analogously to the proof of [@DoJaLo Th. 7.2 ] (see also [GoN]{}).
We are now in a position to prove a qualitative semigroup version of [@DoJaLo Th. 7.2 (a) (b)] which is in turn based on the results given in [DoJa, DoXu]{} (see also [@Ku]). It also extends, in a few directions, [@GoN Th. 5].
\[Thwcs\]Let $C$ be a nonempty convex weakly compact subset of a Banach space $X$ and $\mathcal{T}=\{T_{t}:t\in G\}$ an asymptotically regular semigroup on $C.$ Assume that
1. $s(\mathcal{T})<\sqrt{\operatorname{WCS}(X)}$ or,
2. $s(\mathcal{T})<1+r_{X}(1).$
Then $\mathcal{T}$ has a fixed point in $C$ and $\operatorname{Fix}\mathcal{T}=\{x\in C:T_{t}x=x,\,t\in G\}$ is a Hölder continuous retract of $C.$
Choose a sequence $(t_{n})$ of elements in $G$ such that $\lim_{n\rightarrow
\infty }t_{n}=\infty $ and $s(\mathcal{T})=\lim_{n\rightarrow \infty
}\left\vert T_{t_{n}}\right\vert .$ Let $(T_{t_{n_{\alpha }}})_{\alpha \in
\emph{A}}$ (denoted briefly by $(T_{t_{\alpha }})_{\alpha \in \emph{A}}$) be a pointwise weakly convergent subnet of $(T_{t_{n}}).$ Define, for every $x\in C$, $$Lx=w-\lim_{\alpha }T_{t_{\alpha }}x.$$Fix $x_{0}\in C$ and put $x_{m+1}=Lx_{m},m=0,1,....$ Let $B_{m}=\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m+1}\right\Vert .$ By Lemma \[main\], there exists $\gamma <1$ such that $B_{m}\leq \gamma
B_{m-1}$ for any $m\geq 1.$ Since the norm is weak lower semicontinuous and the semigroup is asymptotically regular,$$\begin{aligned}
& \Vert L^{m+1}x_{0}-L^{m}x_{0}\Vert =\left\Vert x_{m+1}-x_{m}\right\Vert
\leq \liminf_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m}\right\Vert \\
& \ \leq \liminf_{\alpha }\liminf_{\beta }\left\Vert T_{t_{\alpha
}}x_{m}-T_{t_{\beta }}x_{m-1}\right\Vert \leq \limsup_{\alpha }\left\vert
T_{t_{\alpha }}\right\vert \limsup_{\beta }\left\Vert x_{m}-T_{t_{\beta
}-t_{\alpha }}x_{m-1}\right\Vert \\
& \ =s(\mathcal{T})B_{m-1}\leq s(\mathcal{T})\gamma ^{m-1}\operatorname{diam}C\end{aligned}$$for every $x_{0}\in C$ and $m\geq 1.$ Furthermore, by Lemma \[nonexp\], the mapping $L:C\rightarrow C$ is $s(\mathcal{T})$-Lipschitz. It follows from Lemma \[holder\] that $Rx=\lim_{n\rightarrow \infty }L^{n}x$ is a Hölder continuous mapping on $C$. We show that $R$ is a retraction onto $\operatorname{Fix}\mathcal{T}.$ It is clear that if $x\in \operatorname{Fix}\mathcal{T},$ then $Rx=x.$ Furthermore, for every $x\in C,m\geq 1$ and $\alpha \in \emph{A},$$$\Vert T_{t_{\alpha }}Rx-Rx\Vert \leq \left\Vert T_{t_{\alpha
}}Rx-T_{t_{\alpha }}L^{m}x\right\Vert +\left\Vert T_{t_{\alpha
}}L^{m}x-L^{m+1}x\right\Vert +\left\Vert L^{m+1}x-Rx\right\Vert$$and hence$$\lim_{\alpha }\Vert T_{t_{\alpha }}Rx-Rx\Vert \leq s(\mathcal{T})\left\Vert
Rx-L^{m}x\right\Vert +B_{m}+\left\Vert L^{m+1}x-Rx\right\Vert .$$Letting $m$ go to infinity, $\limsup_{\alpha }\Vert T_{t_{\alpha
}}Rx-Rx\Vert =0.$ Since $s(\mathcal{T})=\lim_{\beta }\left\vert T_{t_{\beta
}}\right\vert <\infty ,$ there exists $\beta _{0}\in \emph{A}$ such that $\left\vert T_{t_{\beta }}\right\vert <\infty $ for every $\beta \geq \beta
_{0}.$ Then, the asymptotic regularity of $\mathcal{T}$ implies$$\begin{aligned}
\Vert T_{t_{\beta }}Rx-Rx\Vert & \leq \left\vert T_{t_{\beta }}\right\vert
\limsup_{\alpha }\Vert Rx-T_{t_{\alpha }}Rx\Vert +\lim_{\alpha }\Vert
T_{t_{\beta }+t_{\alpha }}Rx-T_{t_{\alpha }}Rx\Vert \\
& +\limsup_{\alpha }\Vert T_{t_{\alpha }}Rx-Rx\Vert =0.\end{aligned}$$Hence $T_{t_{\beta }}Rx=Rx$ for every $\beta \geq \beta _{0}$ and, from the asymptotic regularity again, $$\Vert T_{t}Rx-Rx\Vert =\lim_{\beta }\left\Vert T_{t+t_{\beta
}}Rx-T_{t_{\beta }}Rx\right\Vert =0$$for each $t\in G.$ Thus $Rx\in \operatorname{Fix}\mathcal{T}$ for every $x\in C$ and the proof is complete.
It is well known that the Opial modulus of a Hilbert space $H,$$$r_{H}(c)=\sqrt{1+c^{2}}-1,$$and the Opial modulus of $\ell _{p},p>1,$ $$r_{\ell _{p}}(c)=(1+c^{p})^{1/p}-1$$for all $c\geq 0$ (see [@LiTaXu]). The following corollaries are sharpened versions of [@GoT Th. 2.2] and [@GoN Cor. 8].
Let $C$ be a nonempty bounded closed convex subset of a Hilbert space $H.$ If $\mathcal{T}=\{T_{t}:t\in G\}$ is an asymptotically regular semigroup on $C$ such that $$\liminf_{t}|T_{t}|<\sqrt{2},$$then $\operatorname{Fix}\mathcal{T}$ is a Hölder continuous retract of $C.$
Let $C$ be a nonempty bounded closed convex subset of $\ell _{p},1<p<\infty
. $ If $\mathcal{T}=\{T_{t}:t\in G\}$ is an asymptotically regular semigroup on $C$ such that $$\liminf_{t}|T_{t}|<2^{1/p},$$then $\operatorname{Fix}\mathcal{T}$ is a Hölder continuous retract of $C.$
Let $1\leq p,q<\infty .$ Recall that the Bynum space $\ell _{p,q}$ is the space $\ell _{p}$ endowed with the equivalent norm $\Vert x\Vert
_{p,q}=(\Vert x^{+}\Vert _{p}^{q}+\Vert x^{-}\Vert _{p}^{q})^{1/q},$ where $x^{+},x^{-}$ denote, respectively, the positive and the negative part of $x.$ If $p>1,$ then $$r_{\ell _{p,q}}(c)=\min \{(1+c^{p})^{1/p}-1,(1+c^{q})^{1/q}-1\}$$for all $c\geq 0$ (see, e.g., [@AyDoLo]). The following corollary extends [@GoN Cor. 10].
Let $C$ be a nonempty convex weakly compact subset of $\ell
_{p,q},1<p<\infty ,1\leq q<\infty .$ If $\mathcal{T}=\{T_{t}:t\in G\}$ is an asymptotically regular semigroup on $C$ such that $$\liminf_{t}|T_{t}|<\min \{2^{1/p},2^{1/q}\},$$then $\operatorname{Fix}\mathcal{T}$ is a Hölder continuous retract of $C.$
Let us now examine the case of $p$-uniformly convex spaces. Recall that a Banach space $X$ is $p$-uniformly convex if $\inf_{\varepsilon >0}\delta
(\varepsilon )\varepsilon ^{-p}>0,$ where $\delta $ denotes the modulus of uniform convexity of $X.$ If $X$ is $p$-uniformly convex, then (see [Xu91]{})$$\left\Vert \lambda x+(1-\lambda )y\right\Vert ^{p}\leq \lambda \left\Vert
x\right\Vert ^{p}+(1-\lambda )\left\Vert y\right\Vert
^{p}-c_{p}W_{p}(\lambda )\left\Vert x-y\right\Vert ^{p} \label{ineq_Xu}$$for some $c_{p}>0$ and every $x,y\in X,0\leq \lambda \leq 1,$ where $W_{p}(\lambda )=\lambda (1-\lambda )^{p}+\lambda ^{p}(1-\lambda ).$ A Banach space $X$ satisfies the Opial property if$$\liminf_{n\rightarrow \infty }\left\Vert x_{n}-x\right\Vert
<\liminf_{n\rightarrow \infty }\left\Vert x_{n}-y\right\Vert$$for every sequence $x_{n}\overset{w}{\longrightarrow }x$ and $y\neq x.$
The following theorem is an extension of [@GoN Th. 7], and a partial extension of [@GoTai Th. 9].
\[Pconvex\]Let $C$ be a nonempty bounded closed convex subset of a $p$-uniformly convex Banach space $X$ with the Opial property and $\mathcal{T}=\{T_{t}:t\in G\}$ an asymptotically regular semigroup on $C$ such that$$\liminf_{t}|T_{t}|<\max \left\{ (1+c_{p})^{1/p},\left( \frac{1}{2}\left(
1+(1+4c_{p}WCS(X)^{p})^{1/2}\right) \right) ^{1/p}\right\} .$$Then $\mathcal{T}$ has a fixed point in $C$ and $\operatorname{Fix}\mathcal{T}$ is a Hölder continuous retract of $C.$
Choose a sequence $(t_{n})$ of elements in $G,$ $\lim_{n\rightarrow \infty
}t_{n}=\infty ,$ such that $s(\mathcal{T})=\lim_{n\rightarrow \infty
}\left\vert T_{t_{n}}\right\vert $ and let $(t_{\alpha })_{\alpha \in \emph{A}}$ denotes a pointwise weakly convergent subnet of $(t_{n}).$ Define, for every $x\in C$,$$Lx=w\text{-}\lim_{\alpha }T_{t_{\alpha }}x.$$Fix $x_{0}\in C$ and put $x_{m+1}=Lx_{m},m\geq 0.$ Let $B_{m}=\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m+1}\right\Vert .$ Since $X$ satisfies the Opial property, it follows from [@KaPr Prop. 2.9] that $$\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m+1}\right\Vert
<\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-y\right\Vert$$for every $y\neq x_{m+1},$ i.e., $x_{m+1}$ is the unique point in the asymptotic center $A(C,(T_{t_{\alpha }}x_{m})),m\geq 0.$ Applying ([ineq\_Xu]{}) yields$$\begin{aligned}
& c_{p}W_{p}(\lambda )\left\Vert x_{m}-T_{t_{\alpha }}x_{m}\right\Vert
^{p}+\left\Vert \lambda x_{m}+(1-\lambda )T_{t_{\alpha }}x_{m}-T_{t_{\beta
}}x_{m-1}\right\Vert ^{p} \\
& \leq \lambda \left\Vert x_{m}-T_{t_{\beta }}x_{m-1}\right\Vert
^{p}+(1-\lambda )\left\Vert T_{t_{\alpha }}x_{m}-T_{t_{\beta
}}x_{m-1}\right\Vert ^{p}.\end{aligned}$$for every $\alpha ,\beta \in \emph{A},0<\lambda <1,m>0.$ Following [@GoTai Th. 9] (see also [@Xu90]) and using the asymptotic regularity of $\mathcal{T},$ we obtain$$\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m}\right\Vert ^{p}\leq
\frac{s(\mathcal{T})^{p}-1}{c_{p}}(B_{m-1})^{p}. \label{in1}$$for any $m>0.$ By Theorem \[Wi1\] and the weak lower semicontinuity of the norm, we have $$B_{m}\leq \frac{D[(T_{t_{\alpha }}x_{m})]}{\operatorname{WCS}(X)}\leq \frac{s(\mathcal{T})}{\operatorname{WCS}(X)}\limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m}\right\Vert .
\label{in2}$$Furthermore, by the Opial property, $$B_{m}\leq \limsup_{\alpha }\left\Vert T_{t_{\alpha }}x_{m}-x_{m}\right\Vert .
\label{in3}$$Combining (\[in1\]) with (\[in2\]) and (\[in3\]) we see that$$(B_{m})^{p}=\limsup_{\alpha }\left\Vert T_{t_{\alpha
}}x_{m}-x_{m+1}\right\Vert ^{p}\leq \gamma ^{p}(B_{m-1})^{p},$$where $$\gamma ^{p}=\max \left\{ \frac{s(\mathcal{T})^{p}-1}{c_{p}},\frac{s(\mathcal{T})^{p}-1}{c_{p}}\left( \frac{s(\mathcal{T})}{\operatorname{WCS}(X)}\right) ^{p}\right\}
<1,$$by assumption. Hence $B_{m}\leq \gamma B_{m-1}$ for every $m\geq 1$ and, proceeding in the same way as in the proof of Theorem \[Thwcs\], we conclude that $\operatorname{Fix}\mathcal{T}$ is a nonempty Hölder continuous retract of $C.$
[99]{} K. Anzai, S. Ishikawa, On common fixed points for several continuous affine mappings, Pacific J. Math. 72 (1977), 1–4.
J. M. Ayerbe Toledano, T. Domínguez Benavides, G. López Acedo, Measures of Noncompactness in Metric Fixed Point Theory, Birkhäuser Verlag, Basel, 1997.
Y. Benyamini, J. Lindenstrauss, Geometric Nonlinear Functional Analysis, Vol. 1, American Mathematical Society, Providence, RI, 2000.
F. E. Browder, W. V. Petryshyn, The solution by iteration of nonlinear functional equations in Banach spaces, Bull. Amer. Math. Soc. 72 (1966), 571–575.
W. L. Bynum, Normal structure coefficients for Banach spaces, Pacific J. Math. 86 (1980), 427–436.
T. Domínguez Benavides, M. A. Japón Pineda, Opial modulus, moduli of noncompact convexity and fixed points for asymptotically regular mappings, Nonlinear Anal. 41 (2000), 617–630.
T. Domínguez, M. A. Japón, G. López, Metric fixed point results concerning measures of noncompactness, in: Handbook of Metric Fixed Point Theory, W. A. Kirk, B. Sims (eds.), Kluwer Acad. Publ., Dordrecht, 2001, 239–268.
T. Domínguez Benavides, G. López Acedo, Lower bounds for normal structure coefficients, Proc. Roy. Soc. Edinburgh Sect. A 121 (1992), 245–252.
T. Domínguez Benavides, G. López Acedo, H. K. Xu, Weak uniform normal structure and iterative fixed points of nonexpansive mappings, Colloq. Math. 68 (1995), 17–23.
T. Domínguez Benavides, H. K. Xu, A new geometrical coefficient for Banach spaces and its applications in fixed point theory, Nonlinear Anal. 25 (1995), no. 3, 311–325.
M. Edelstein, R. C. O’Brien, Nonexpansive mappings, asymptotic regularity and successive approximations, J. London Math. Soc. (2) 17 (1978), no. 3, 547–554.
N. A. Erzakova, Asymptotically regular mappings, Izv. Vyssh. Uchebn. Zaved. Mat. (2006), 17–21 (in Russian), translation in Russian Math. (Iz. VUZ) 50 (2006), 15–19 (2007).
K. Goebel, W. A. Kirk, Iteration processes for nonexpansive mappings, in: Topological Methods in Nonlinear Functional Analysis, S. P. Singh, S. Thomeier, B. Watson (eds.), AMS, Providence, R.I., 1983, 115–123.
J. Górnicki, Fixed point theorems for asymptotically regular mappings in $L^{p}$ spaces, Nonlinear Anal. 17 (1991), no. 2, 153–159.
J. Górnicki, On the structure of fixed point sets of asymptotically regular mappings in Hilbert spaces, Topol. Methods Nonlinear Anal. 34 (2009), 383–389.
J. Górnicki, Structure of the fixed-point set of asymptotically regular mappings in uniformly convex Banach spaces, Taiwanese J. Math. 15 (2011), 1007–1020.
J. Górnicki, Geometrical coefficients and the structure of the fixed-point set of asymptotically regular mappings in Banach spaces, Nonlinear Anal. 74 (2011), 1190–1199.
S. Ishikawa, Fixed points and iteration of a nonexpansive mapping in a Banach space. Proc. Amer. Math. Soc. 59 (1976), no. 1, 65–71.
W. Kaczor, S. Prus, Asymptotical smoothness and its applications, Bull. Austral. Math. Soc. 66 (2002), no. 3, 405–418.
T. Kuczumow, Opial’s modulus and fixed points of semigroups of mappings, Proc. Amer. Math. Soc. 127 (1999), 2671–2678.
T. C. Lim, On the normal structure coefficient and the bounded sequence coefficient, Proc. Amer. Math. Soc. 88 (1983), 262–264.
P. K. Lin, A uniformly asymptotically regular mapping without fixed points, Canad. Math. Bull. 30 (1987), 481–483.
P. K. Lin, K. K. Tan, H. K. Xu, Demiclosedness principle and asymptotic behavior for asymptotically nonexpansive mappings, Nonlinear Anal. 24 (1995), 929–946.
E. Maluta, S. Prus, J. Wośko, Fixed point free mappings which satisfy a Darbo type condition, in: Fixed Point Theory and its Applications, H. Fetter Nathansky at al. (eds.), Yokohama Publ., Yokohama, 2006, 171–184.
A. Pe[ł]{}czyński, A note on the paper of I. Singer ”Basic sequences and reflexivity of Banach spaces”, Studia Math. 21 (1961/1962), 371–374.
S. Prus, On Bynum’s fixed point theorem, Atti Sem. Mat. Fis. Univ. Modena 38 (1990), 535–545.
E. Sȩd[ł]{}ak, A. Wiśnicki, On the structure of fixed-point sets of uniformly Lipschitzian mappings, Topol. Methods Nonlinear Anal. 30 (2007), 345–350.
D. Tingley, An asymptotically nonexpansive commutative semigroup with no fixed points, Proc. Amer. Math. Soc. 97 (1986), 107–113.
H. K. Xu, Fixed point theorems for uniformly Lipschitzian semigroups in uniformly convex spaces, J. Math. Anal. Appl. 152 (1990), 391–398.
H. K. Xu, Inequalities in Banach spaces with applications, Nonlinear Anal. 16 (1991), 1127–1138.
H. K. Xu, I. Yamada, Asymptotic regularity of linear power bounded operators, Taiwanese J. Math. 10 (2006), 417–429.
G. L. Zhang, Weakly convergent sequence coefficient of product space, Proc. Amer. Math. Soc. 117 (1993), 637–643
| {
"pile_set_name": "ArXiv"
} |
-.1 cm April 2012\
.4in
[**Anomalies, instantons and chiral symmetry breaking**]{} .1cm [**at a Lifshitz point[^1]**]{}
0.8in
[**Ioannis Bakas**]{} 0.2in [*Department of Physics, School of Applied Mathematics and Physical Sciences\
National Technical University, 15780 Athens, Greece\
0.2in* ]{}\
0.6in
**Abstract**
0.2in We give a new twist to an old-fashioned topic in quantum field theory describing violations of the chiral charge conservation of massless fermions through Adler-Bell-Jackiw anomalies in the background of instanton fields in the context of non-relativistic Lifshitz theories. The results we report here summarize in a nut-shell our earlier work on the subject found in arXiv:1103.5693 and arXiv:1110.1332. We present simple examples where index computations can be carried out explicitly focusing, in particular, to gravitational models of Lifshitz type and highlight their differences from ordinary gravity in four space-time dimensions.
16 pt
Introduction
============
The axial anomalies arising upon quantization of massless fermions in a given gauge and/or metric field background are deeply connected to the analytic index of the fermion operator. The Atiyah-Singer index theorem asserts that the difference of positive and negative chirality normalizable zero modes of the Dirac operator in a given background is provided by (one-half) the integrated form of the anomalous axial current conservation law. This profound relation constitutes the back-bone of our study and it can be easily established without knowing the local form of the axial anomaly. It will be discussed first for relativistic fermion theories in four space-time dimensions and then it will be generalized to non-relativistic models of Lifshitz type. Of course, the computation of the index relies on the form of the axial anomaly, which is a topological density given by the Chern-Pontryagin class of the gauge and/or metric field background. The coefficient of the anomaly turns out to be universal (it is the same for both relativistic and Lifshitz theories) in accordance with the general expectation that the axial anomaly is an infra-red phenomenon in disguise. Furthermore, if the index of the Dirac operator is non-zero, the corresponding axial charge will not be conserved in time, leading to violation of baryon and lepton number in physical processes within a given theory. These issues will be addressed in detail focusing on the similarities and the differences exhibited by relativistic and Lifshitz quantum field theories.
The results we report in the following are restricted to four-dimensional theories and they are based on our previous work on the subject [@dieter; @bakas] to which we refer the interested reader for further details. However, the presentation we adopt here is somewhat different emphasizing more the general ideas rather than the technical details. Also, the instanton backgrounds on which the index computations will be made are described in all generality in our work on gravitational Hořava-Lifshitz models [@BBLP; @bakas2]. References to other original papers can also be found there.
Relativistic field theories
===========================
First, to set up the stage, we consider the Dirac operator in a four-dimensional space-time $M_4$ with Euclidean signature which may also be coupled to a (generally non-Abelian) background gauge field $A_{\mu}$ via the rule of minimal substitution D\_ = \_ + [1 8]{} \[\_a , \_b\] [\_]{}\^[ab]{} - i A\_ . \[raloua\] Here, ${\omega_{\mu}}^{ab}$ are the components of the spin connection with tangent space-time indices $a$, $b$ and $\gamma_a$ are the corresponding Dirac matrices satisfying the anti-commutation relations \_+ = 2 \_[ab]{} . In most applications, the four-dimensional metric will be taken to be of the special form ds\^2 = dt\^2 + g\_[ij]{} (t, x) dx\^i dx\^j . \[loukos\] This is also appropriate for the description of Lifshitz theories satisfying the so called projectable condition, which will occupy most of our attention later. In the relativistic case, however, the space-time metric can be of general form.
Then, the Euclidean Dirac operator and the associated $\gamma_5$-matrix $\gamma_5 = - \gamma_0 \gamma_1 \gamma_2 \gamma_3$ that anti-commutes with it assume the following form in the chiral representation, respectively, i \^ D\_ = i (
[ccc]{} 0 & & [Q]{}\_-\
& &\
[Q]{}\_+ & & 0
), \_5 = (
[ccc]{} & & 0\
& &\
0 & & -
) , setting for notational convenience \_ = [t]{} i \_I [E\_I]{}\^i D\_i . \[loura\] In writing [(\[loura\])]{} we use the temporal choice $A_0 = 0$ in the presence of gauge fields. We also use the Pauli matrices $\sigma_I$ as well as the inverse dreibeins ${E_I}^i$ associated to the metric $g_{ij}$ in [(\[loukos\])]{} (summation over the space indices $i$ and the tangent space indices $I$ are implicitly assumed). The $2 \times 2$ blocks ${\cal Q}_{\pm}$ are first-order operators mapping the two-component Weyl spinors $\Psi_{\pm}$ to $\Psi_{\mp}$ and they are mutually related by conjugation as $({\cal Q}_{\pm})^{\dagger} = - {\cal Q}_{\mp}$. The massless Dirac equation $i\gamma^{\mu} D_{\mu} \Psi (t, x) = 0$ acting on four-component spinors reduces to the following system of Weyl equations on $M_4$, \_ \_ (t, x) = 0 , \[soutz\] whose number of normalizable solutions will be denoted by $n_{\pm}$, respectively.
The index of the Dirac operator is defined as the difference between the number of positive and negative chirality zero modes, i.e., (D) = n\_+ - n\_- = [dim Ker]{} [Q]{}\_+ - [dim Ker]{} [Q]{}\_- = [Tr]{} \_5 . The last equality is actually a tautology following from the chirality condition $\gamma_5 \Psi_{\pm} = \pm \Psi_{\pm}$ by taking the trace over the zero energy states in the fermion Hilbert space. One can extend the trace to the entire Hilbert space without affecting the index because the non-zero energy states are always paired and for them the difference between the two chiralities cancel. Thus, only zero energy states contribute to the index. Since the zero modes of ${\cal Q}_{\pm}$ are also zero modes of ${\cal Q}_{\mp} {\cal Q}_{\pm}$, the index takes the equivalent form (D) = [dim Ker]{} ([Q]{}\_- [Q]{}\_+) - [dim Ker]{} ([Q]{}\_+ [Q]{}\_-) = [Tr]{} (\_5 e\^[- (i \^ D\_)\^2]{} ) . \[mourta\] The operators ${\cal Q}_{\mp} {\cal Q}_{\pm}$ are elliptic and they are better behaved than ${\cal Q}_{\pm}$, since they are related to the square of the Dirac operator as - (i \^ D\_)\^2 = (
[ccc]{} [Q]{}\_- [Q]{}\_+ & & 0\
& &\
0 & & [Q]{}\_+ [Q]{}\_-
) . The equality [(\[mourta\])]{} is very useful for comparison with the axial anomaly computations. It holds for any $\tau > 0$ and it provides a regulated version of ${\rm Tr} \gamma_5$. The trace is taken oven the entire fermionic Hilbert space, since the non-zero energy states cancel each other.
The massless fermion theory with Lagrangian density ${\cal L} = \bar{\Psi} i \gamma^{\mu}
D_{\mu} \Psi$ exhibits an axial current conservation law $\nabla_{\mu} J_5^{\mu} (t, x) = 0$, where $J_5^{\mu} (t, x) = \bar{\Psi} \gamma^{\mu} \gamma_5 \Psi$. There is an invariance of the classical theory that can be easily found by applying Noether’s procedure with respect to the chiral rotations $\delta_{\epsilon} \Psi = i \epsilon \gamma_5 \Psi$. There is also an associated chiral charge Q\_5 = d\^3 x J\_5\^0 (t, x) which is conserved in time under the appropriate boundary conditions at spatial infinity (typically the spatial slices are taken to be compact without boundary). Quantum mechanically, however, the situation changes drastically as there is an obstruction to the axial current conservation law, called axial anomaly. The anomaly is most conveniently described in the Euclidean domain after Wick rotation of the time coordinate and it originates from the non-invariance of the fermionic path integral measure $({\cal D} \bar{\Psi})
({\cal D} \Psi)$ under chiral rotations. Careful investigation of the partition function shows that the variation of the action combines with the variation of the measure to produce the anomalous conservation law \_ J\_5\^ (t, x) = 2 \_ \_n \_n\^ (t, x) \_5 e\^[- (i\^ D\_)\^2 / \^2]{} \_n (t, x) \[tsoutsia\] after introducing a cut-off $\Lambda$ to regulate the infinite sum that otherwise is ill-defined. Here, $\varphi_n (t, x)$ are the eigen-states of the interacting fermion operator $i \gamma^{\mu} D_{\mu}$.
Integration over space-time is combined with the sum over $n$ in [(\[tsoutsia\])]{} to yield the trace over the entire fermionic Hilbert space of the theory. Thus, comparison with the index formula [(\[mourta\])]{} leads to the profound relation (D) = [1 2]{} \_[M\_4]{} dt d\^3 x \_ J\_5\^ (t, x) , \[raka\] which connects the physics of axial anomalies with the mathematical theory of Atiyah-Singer index theorem for the Dirac operator. Clearly, if the index is non-zero, the axial charge $Q_5$ will not be conserved, leading to baryon and lepton number violation in the theory, as Q\_5 = 2 [Ind]{} (D) .
So far there has been no explicit reference to the form of the axial anomaly in the background of gauge and/or metric fields. Likewise, there has been no reference to the Atiyah-Singer formula for computing the index of the Dirac operator analytically. This step is crucial for telling the rest of the story in physics and mathematics. Manipulating the regulated sum [(\[tsoutsia\])]{}, it turns out that \_ J\_5\^ = [1 4 \^2]{} [Tr]{}(F F) , \_ J\_5\^ = [1 96 \^2]{} [Tr]{}(R R) \[expras\] for the gauge and the gravitational field contribution to the anomalous divergence of the axial current, respectively. The obstruction to the axial current conservation law can be easily guessed in both cases. The anomaly should be a topological density built out of the field strength of the background fields, i.e., the curvature two-forms $F$ and $R_{ab}$ of the gauge and metric fields, respectively, so that the divergence of the axial current is a total derivative gauge invariant term. Furthermore, it should be odd under parity since $J_5^{\mu}$ is a pseudo-vector current. The only available such quantities in four space-time dimensions are the characteristic classes ${\rm Tr}(F \wedge F)$ and ${\rm Tr}(R \wedge R)$, where the trace is taken over the color indices of the non-Abelian gauge field (per flavor of Dirac fermions) and the tangent space-time indices of $M_4$, respectively. Then, it only remains to fix the coefficient of the anomaly to complete the derivation of [(\[expras\])]{}. The outcome is the same using either physical or mathematical techniques to make sense of the regulated sum for ${\rm Tr} \gamma_5$.
Soon after the discovery of instantons in four-dimensional gauge theories and gravity, the question arose whether the index of the Dirac operator is non-zero on such topologically non-trivial backgrounds. For non-Abelian gauge fields (e.g., $SU(2)$) on $M_4 \simeq S^4$, obtained as one-point compactification of $\mathbb{R}^4$ by imposing appropriate boundary conditions on $A_{\mu}(x)$, the instanton number $k \in \mathbb{Z}$ is provided by the second Chern number k = [1 8 \^2]{} \_[M\_4]{} [Tr]{}(F F) and, therefore, ${\rm Ind} (D) = k$ on such instanton backgrounds. Then, it follows that the axial charge conservation law is violated by $\Delta Q_5 = 2k$. Likewise, on compact Riemannian manifolds without boundaries, but with non-vanishing Hirzebruch signature, (M\_4) = [1 24 \^2]{} \_[M\_4]{} [Tr]{}(R R) , the index of the Dirac operator is non-zero, ${\rm Ind} (D) = \tau/8$ (recall that the signature of all compact four dimensional spin manifolds without boundaries is integer multiple of $8$). The only gravitational instanton of this kind is $K3$, which is a hyper-Kähler manifold with self-dual Riemann tensor whose signature is $16$ and ${\rm Ind} (D) = 2$.
In the presence of boundaries one considers bound states of the Dirac equation by imposing appropriate boundary conditions on the spinors near $\partial M_4$ and uses the Atiyah-Patodi-Singer (APS) index theorem to count the difference between positive and negative chirality zero modes ($L^2$-index of the Dirac operator). The APS theorem is based on the integrated form of the axial anomaly with the same bulk contribution as before, but it also contains suitable boundary terms given by (D) = [1 192 \^2]{} \_[M\_4]{} [Tr]{} (R R) - [1 192 \^2]{} \_[M\_[4]{}]{} [Tr]{} (R) - [1 2]{} \_[D]{} (M\_4) . \[aps\] The first boundary term involves the Chern-Simons secondary characteristic class written in terms of the second fundamental form $\theta$, which accounts for the possible deviation of the space-time metric $ds^2 = dt^2 + g_{ij} (t, x) dx^i dx^j$ from cross-product form at the boundary; as such it is a higher derivative analogue of the Gibbons-Hawking-York boundary term encountered in general relativity. The second boundary term is non-local and it is provided by the $\eta$-invariant of the tangential part of the Dirac operator restricted to boundary; as such it counts the spectral asymmetry between positive and negative chirality modes of the corresponding three-dimensional Dirac operator on $\partial M_4$ and it is made rigorous using zeta-function regularization.
When $M_4 = I \times \Sigma_3$ with a compact three-manifold $\Sigma_3$ without boundaries (e.g., $\Sigma_3 \simeq S^3$), the index of the Dirac operator on $M_4$ can be calculated in practice by spectral flow methods. ${\rm Ind}(D)$ is provided by the net number of level crossings that occur in the spectrum of the three-dimensional Dirac operator on $\Sigma_3$, (D) = S(\_3) , \[specri\] as the metric $g_{ij} (t, x)$ deforms from one end of the time interval $I$ to the other; recall that the $\eta$-invariant jumps by $\pm 2$ units when an eigen-value crosses from negative to positive values or conversely, and, therefore, each level crossing contributes $\pm 1$ units to the index. Formula [(\[specri\])]{} and its generalization to Lifshitz theories is very useful for the applications. In those cases that the complete spectrum of the Dirac operator on $\Sigma_3$ can be found and the spectral flow can be studied explicitly, the index of the four-dimensional fermion operator on $I \times \Sigma_3$ can be computed in closed form.
It is well known that all non-compact instanton solutions of Einstein gravity can not support extreme geometric deformations on $\Sigma_3$ that are capable to induce level crossing, and, hence, ${\rm Ind}(D) = 0$. Prime examples of this kind are provided by the Taub-NUT and Eguchi-Hanson instantons which admit complete metrics [(\[loukos\])]{} with self-dual Riemann curvature tensor on $I \times \Sigma_3$ with $I$ being the semi-infinite real line of proper time $t$ and $\Sigma_3$ is $S^3$ and $S^3 / \mathbb{Z}_2$, respectively, endowed with homogeneous and partially isotropic geometries with $SU(2) \times U(1)$ isometry group. It can be explicitly seen in those cases that the individual terms contributing to the index [(\[aps\])]{} cancel against each other, as required on general grounds based on Lichnerowicz’s theorem: non-compact four-metrics with non-negative Ricci scalar curvature admit no bound state solutions of the Dirac equation (if such states existed, they would be covariantly constant, and, hence, non-normalizable leading to contradiction). Thus, it is not possible to have chiral symmetry breaking induced by gravitational instantons of topology $I \times \Sigma_3$ in ordinary Einstein-Dirac theory.
Lifshitz field theories
=======================
Next, we extend the scope of our discussion to Lifshitz fermion theories with anisotropy scaling exponent $z = 2 \alpha + 1$ by considering the non-relativistic analogue of the Dirac operator i \^ [D]{}\_ = i \^0 D\_0 + [1 2]{} i \^i \[D\_i (- D\_k D\^k + M\^2)\^ + (- D\_k D\^k + M\^2)\^ D\_i\] \[arou\] in the presence of background gauge and/or metric fields. Here, $M$ is an arbitrary mass scale that is introduced for convenience to extrapolate between the Dirac and Lifshitz fermion models. The Lifshitz operator acts on four-component spinors $\Psi (t, x)$ and $D_0$ and $D_i$ are the time and space components of the ordinary covariant derivative [(\[raloua\])]{} that provides the minimal coupling to the background fields. Then, the massless Lifshitz fermion theory in $3+1$ space-time dimensions is defined by the Lagrangian density $\bar{\Psi} i \gamma^{\mu} {\cal D}_{\mu} \Psi$ and gives rise to the axial current conservation law $\nabla_{\mu} J_5^{\mu} (t, x) = 0$, as in the relativistic case. The spatial components of the axial current are different from the relativistic case, since they involve a number of derivatives that depend upon $z$, whereas the time component $J_5^0 (t, x)$, and, hence, the axial charge $Q_5$ is the same.
The axial symmetry is broken quantum mechanically and explicit computation shows that the anomalous term in the divergence of the axial current is identical to the relativistic case for the gauge and/or the metric field couplings, as in equation [(\[expras\])]{}. More precisely, the anomalous divergence of the axial current in the presence of gauge fields is \_ J\_5\^ = [1 4 \^2]{} \^[0ijk]{} [Tr]{}(F\_[0i]{} F\_[jk]{}) = [1 4 \^2]{} [Tr]{}(F F) , whereas the metric field contribution to the axial anomaly turns out to be \_ J\_5\^ = - [1 96 \^2]{} \^[0ijk]{} [R\^[ab]{}]{}\_[0i]{} R\_[ab jk]{} = [1 96 \^2]{} [Tr]{}(R R) . Here, we write the result in $3+1$ terms using the electric and magnetic components of the corresponding curvature 2-forms and then recast it in the form [(\[expras\])]{}. The calculation is performed in the Euclidean domain by analytic continuation of the space-time foliation used for the formulation of Lifshitz models. It is also implicitly assumed that the space-time metric is restricted to the projectable case, meaning that the lapse and shift functions in the ADM decomposition of the metric are taken to be $1$ and $0$, respectively, so that its analytic continuation is given by [(\[loukos\])]{} in proper time $x^0 = t$.
The arguments leading to the general relation [(\[raka\])]{} between the index of the fermion operator and the axial anomaly generalize easily to the Lifshitz operator [(\[arou\])]{}. One simply has to consider the higher derivative analogue of the operators ${\cal Q}_{\pm}$ shown in [(\[loura\])]{} by replacing $D_i$ with $[D_i (- D_k D^k + M^2)^{\alpha} + (- D_k D^k + M^2)^{\alpha} D_i]/2$ and also replace the regulator ${\rm exp}[- (i\gamma^{\mu} D_{\mu})^2 / \Lambda^2]$ by ${\rm exp}[- (i\gamma^{\mu} {\cal D}_{\mu})^2 / \Lambda^{2z}]$ in the sum [(\[tsoutsia\])]{} which is now taken oven the eigen-states $\varphi_n (t, x)$ of the interacting Lifshitz fermion operator $i\gamma^{\mu} {\cal D}_{\mu}$. As a result, the index of the Lifshitz fermion operator [(\[arou\])]{} is equal to the index of the Dirac operator for all values of the anisotropy exponent $z$, ([D]{}) = [Ind]{}(D) . \[univad\]
In space-times with boundaries the Atiyah-Patodi-Singer index theorem for the Lifshitz fermion operator assumes the same form as in [(\[aps\])]{}. The $\eta$-invariant is now referring to the tangential part of the Lifshitz operator restricted to the boundary $\partial M_4$, which turns out to be equal to the $\eta$-invariant of the corresponding three-dimensional Dirac operator. Thus, relation [(\[univad\])]{} is universal, as it extends to all space-times with or without boundaries. If the index does not vanish, the axial charge will not be conserved in time, i.e., $\Delta Q_5 = 2 ~ {\rm Ind}({\cal D})$, leading to baryon and lepton number violation as before. It will be seen shortly that chiral symmetry breaking effects in non-relativistic theories of Lifshitz type are common practice.
Anomalies, instantons and chiral symmetry breaking at a Lifshitz point are closely interrelated as in the relativistic case. The task is to find solutions of Lifshitz theories that allow for violation of chiral charge conservation and then compare the results to relativistic field theories. Here, we provide a brief account of bosonic Lifshitz theories and review their instanton solutions, following our earlier work on the subject [@BBLP; @bakas2]. In general, they provide classical backgrounds for Lifshitz fermion propagation in $3+1$ space-time dimensions. Special emphasis is placed on gravitational theories of Lifshitz type (in the so called Hořava-Lifshitz gravity) coupled to Lifshitz fermion models. Then, ignoring the backreaction of fermions to the gravitational instanton backgrounds, we find that – unlike ordinary gravity – non-conservation of $Q_5$ becomes possible for a certain range of the gravitational coupling parameters. We provide some simple solutions that realize this novel possibility and give a qualitative interpretation of its origin. The case of Lifshitz gauge field theories appears to be conceptually simpler and looks easier for comparison with the relativistic case, but its instanton solutions are not explicitly known to this day; it is an open problem for future study which we hope to address elsewhere in detail.
The bosonic Lifshitz field theories in $3+1$ space-time dimensions resemble point particle systems with configuration space ${\cal C}$ and local coordinates $q_I$ that correspond to Euclidean relativistic fields in three spatial dimensions. Thus, ${\cal C}$ is the infinite dimensional space of all scalar, vector or more generally tensor fields on a Riemannian manifold $\Sigma_3$, i.e., $q_I = \varphi (x)$, $A_i (x)$ or $g_{ij} (x)$ etc, which will be called superspace in all cases. These field theories are non-relativistic models with anisotropic scaling in space and time $x \rightarrow ax$ and $t \rightarrow a^z t$ with exponent $z$ which is provided by the order of the classical equations of motion of the relativistic fields $q_I$ defined on $\Sigma_3$. Their action is often taken to be of the form S = [1 2]{} dt \_[I, J]{} ([d q\_I dt]{} [O]{}\^[IJ]{} [d q\_J dt]{} - [W q\_I]{} [O]{}\_[IJ]{} [W q\_J]{} ) \[parta\] assuming that the potential term is derivable from a superpotential $W[q]$ using the metric ${\cal O}^{IJ}$ and its inverse ${\cal O}_{IJ}$ on the superspace ${\cal C}$. This class of Lifshitz models are said to satisfy the detailed balance condition with local superpotential functional $W$ being the action of a suitably chosen relativistic field theory on $\Sigma_3$. Of course, one may deviate from detailed balance by having additional terms in the potential that cannot be casted in the form [(\[parta\])]{} using a local superpotential functional, but such generalizations will not be in focus here. In all cases, the Lifshitz theories admit an effective point particle description in superspace that proves useful in many respects.
Next, we consider instanton solutions of Lifshitz theories on $\mathbb{R} \times \Sigma_3$, assuming for simplicity that the metric in superspace is positive definite (it can also become degenerate in some important cases that will be discussed later). We first note that the minima of the potential provide static solutions of the equations of motion following from [(\[parta\])]{}; they are configurations that satisfy the classical equations of motion $\partial W / \partial q_I = 0$ of the underlying relativistic field theory defined by the action $W$ on $\Sigma_3$ and they are all degenerate with zero energy. Then, the instantons are defined as extrema of the Euclidean action derived from [(\[parta\])]{} by Wick rotation $t \rightarrow it$ (equivalently by inverting the potential) that interpolate smoothly between different degenerate vacua of the effective point particle system as $t$ extends all the way from $-\infty$ to $+\infty$. By completing the square, as usual, it can be easily seen that the instantons satisfy the system of first-order equations in time = \_[IJ]{} [W q\_J]{} \[eterna\] and their action equals $S_{\rm instanton} = |\Delta W | \equiv |W(t = + \infty) -
W(t = - \infty)|$. The two choices of sign in [(\[eterna\])]{} correspond to instanton and anti-instanton configurations.
Thus, the instantons of Lifshitz theories (with detailed balance) are eternal solutions of the gradient flow equations [(\[eterna\])]{} derived from the superpotential functional $W$ and their action is finite, as required on general grounds. Yet explicit solutions of the gradient flow equations can not be easily obtained unless additional symmetries are imposed on the fields, leading to consistent mini-superspace truncations of the configuration space ${\cal C}$. Otherwise, only qualitative features of the solutions can be studied, in general, at least in those cases that the mathematical tools of geometric analysis are powerful enough to explore the problem of long-time existence against the possible formation of singularities along the flow lines. In gravitational theories of Lifshitz type the defining relations [(\[eterna\])]{} are nothing else but geometric flows for the metrics on $\Sigma_3$. Specific examples and explicit solutions will be described later and compared to the instantons of Einstein gravity. Note for completeness that if we were considering Lifshitz theories without detailed balance, instanton solutions would be much more difficult to find, if they existed at all as finite Euclidean action configurations.
Specializing to Hořava-Lifshitz gravity, we consider $(3+1)$-dimensional space-times $M_4 \simeq \mathbb{R} \times \Sigma_3$ endowed with Lorentzian metrics $ds^2 = -dt^2 + g_{ij}(t, x) dx^i dx^j$ (the so called projectable case) and write down the following action in canonical form, S = [1 2]{} \_[M\_4]{} dt d\^3 x , \[horav\] where \^[ijkl]{} = [1 2]{} (g\^[ik]{} g\^[jl]{} + g\^[il]{} g\^[jk]{}) - g\^[ij]{} g\^[kl]{} is the ($\lambda$-deformed) DeWitt metric in superspace ${\cal C}$ consisting of all Riemannian metrics on $\Sigma_3$ and ${\cal G}_{ijkl}$ is its inverse. We also choose as superpotential functional $W[g]$ the action of topologically massive gravity on the three-manifold $\Sigma_3$, W\_[TMG]{}\[g\] = [2 \_[w]{}\^2]{} \_[\_3]{} d\^3x (R-2\_[w]{}) + [1 ]{} W\_[CS]{} \[g\] , where W\_[CS]{} \[g\] = \_[\_3]{} d\^3x \^[ijk]{} \_[im]{}\^l (\_j \_[lk]{}\^m + [2 3]{} \_[jn]{}\^m \_[kl]{}\^n ) is the gravitational Chern-Simons action, which is conveniently written here in terms of the Christoffel symbols of the metric $g$ on $\Sigma_3$. The resulting non-relativistic gravitational theory [(\[horav\])]{} in $3+1$ dimensions exhibits anisotropic scaling $z=3$ which reduces to $z=2$ only when the gravitational Chern-Simons term is absent.
The essential feature of topologically massive gravity is the presence of an adjustable scale that supplies the mass to one of the two helicity gravitons in the weak field approximation of the theory around flat space; the other helicity graviton remains massless and the two are interrelated by orientation reversing transformations on $\Sigma_3$ that flip the sign of $\omega$. The associated range of gravitational interactions in three dimensions is given by the relative ratio of the three-dimensional gravitational constant to the Chern-Simons coupling, $\kappa_{\rm w}^2 / |\omega|$, which will play role later in the formulation of a geometric criterion for having chiral symmetry breaking effects by gravitational instantons in the associated $(3+1)$-dimensional Hořava-Lifshitz gravity. Further generalizations arise by adding quadratic (or higher) curvature terms to $W$, as in three-dimensional new massive gravity (and generalizations thereof), which lead to $(3+1)$-dimensional Hořava-Lifshitz models with higher anisotropy scaling exponent $z$. In those cases, the three-dimensional gravitons remain massive (although the masses of the two helicity states need not be the same) leading to similar phenomena as with $W_{\rm TMG}$; such generalizations will not be addressed here at all to simplify the presentation.
The instanton equation [(\[eterna\])]{} specializes to the Ricci-Cotton flow. This is a third order equation derived from $W_{\rm TMG}[g]$ as gradient flow for the metric $g_{ij}$ on $\Sigma_3$. The flow lines depend on the superspace parameter $\lambda$, but the fixed points, which are classical solutions of three-dimensional topological massive gravity, do not depend on it for general values of $\lambda$. Here, we only consider the spacial case $\lambda = 1/3$, so that the defining equation of gravitational instantons takes the form \_t g\_[ij]{} = (R\_[ij]{} - [1 3]{} R g\_[ij]{} ) C\_[ij]{} , \[ricot\] where $R_{ij}$ is the Ricci curvature tensor of $g$ and $C_{ij}$ its Cotton tensor. The driving curvature terms that arise in this case are traceless. This, in turn, implies that the fixed points have undetermined Ricci scalar curvature $R$, which can vary from one fixed point to the other. Said differently, we are considering a unimodular version of three-dimensional topological massive gravity in which the cosmological constant $\Lambda_{\rm w}$ has the interpretation of an integration constant that can assume arbitrary values. This restriction is necessary for having level crossing, and, hence, $\Delta Q_5 \neq 0$ as $t$ varies from $-\infty$ to $+\infty$ in the corresponding gravitational instanton background. Otherwise, if $\lambda \neq 1/3$, the two end-points of the instanton will have the same curvature $R (= 6 \Lambda_{\rm w})$ forbidding any net level crossing to occur in the theory.
More technically speaking, the metric in superspace takes the following form at $\lambda = 1/3$, \^[ijkl]{} = [1 2]{} (g\^[ik]{} g\^[jl]{} + g\^[il]{} g\^[jk]{}) - [1 3]{} g\^[ij]{} g\^[kl]{} , and as such it projects any symmetric two-tensor to its traceless part. Then, the inverse metric in superspace becomes ill-defined. Instead, one defines \_[ijkl]{} = [1 2]{} (g\_[ik]{} g\_[jl]{} + g\_[il]{} g\_[jk]{}) - [1 3]{} g\_[ij]{} g\_[kl]{} , which follows formally from the inverse metric as $\lambda \rightarrow \pm \infty$ and it also projects any symmetric two-tensor to its traceless part. The two quantities are simply related to each other by the generalized orthonormality condition \^[ijkl]{} [G]{}\_[klmn]{} = [1 2]{} (\_m\^i \_n\^j + \_n\^i \_m\^j ) - [1 3]{} g\^[ij]{} g\_[mn]{} that follows from the standard one by subtracting the trace part for consistency of the projection. As a result, the conformal factor of the metric decouples from the dynamics and the Ricci-Cotton flow [(\[ricot\])]{} preserves the volume of space ${\rm Vol} (\Sigma^3)$. This is precisely the class of models that allow for violations of chiral charge conservation by gravitational instanton effects in Hořava-Lifshitz gravity.
Our next task is to consider simple examples that illustrate the situation and derive the necessary and sufficient conditions on the parameters of the theory for having chiral symmetry breaking. We consider the simple case of Bianchi IX homogeneous geometries on $\Sigma_3 = S^3$ with isometry group $SU(2) \times U(1)$, known as Berger spheres, ds\^2 = (t) , which provide consistent mini-superspace reduction of the Ricci-Cotton flow to an ordinary differential equation. Here, $\sigma^I$ are the left-invariant 1-forms of $SU(2)$ satisfying d\^I + [1 2]{} [\^I]{}\_[JK]{} \^J \^K = 0 and $\delta \in [0, ~ \infty)$ is a parameter measuring the anisotropy of the model. Instanton solutions of Hořava-Lifshitz gravity can be explicitly constructed in this case and then compared to the analogous solutions (Eguchi-Hanson and Taub-NUT) of Euclidean Einstein gravity. Homogeneous solutions with higher degree of anisotropy can also be studied by relaxing the additional $U(1)$ isometry (axial symmetry) of the 3-sphere, but they will not be discussed here as they introduce unnecessary technical complications.
The normalized Ricci-Cotton flow [(\[ricot\])]{} admits the round metric with $\delta = 1$ as fixed point, but it also admits a second fixed point within the class of Berger sphere metrics provided that $\omega < 0$ (for a given choice of orientation of $S^3$) with + [\_[w]{}\^2 2 ]{} = 0 . \[othfix\] The presence of two fixed points is prerequisite for the existence of instantons. There is a smooth flow line interpolating between these two fixed points, but the precise form of the solution will not be important in the following. All it matters here is the variation of the 3-curvature R = [1 2 ]{} (4- \^2) \[kourva\] as one moves from one fixed point to the other and the ability to induce level crossing by changing the shape of $S^3$, whereas the volume ${\rm Vol} (S^3) = 16 \pi^2 \delta \gamma^{3/2}$ remains fixed throughout the evolution. Fortunately, the computation of the index of the Dirac-Lifshitz operator is a tractable problem in the background of gravitational instanton solutions with $SU(2) \times U(1)$ isometry.
Let $\zeta$ denote the eigen-values of the three-dimensional Dirac operator $i \gamma^i D_i$ on a Berger sphere, which can be determined in closed form together with their multiplicities. They split into positive and negative eigen-values that depend upon $\delta$. Zero modes (often called harmonic spinors) also arise for special values of $\delta \geq 4$, \^2 = 2 , (p, q) \^2 \[moutsia\] with multiplicities $p+q$. Setting $p=q=1$ it follows that the first zero modes of $i \gamma^i D_i$ arise when $\delta = 4$, which is the critical value of the anisotropy parameter for inducing level crossing by varying $\delta$. Comparison with equation [(\[kourva\])]{} shows that the curvature of the Berger sphere should be sufficiently negative to allow for the occurrence of zero modes and subsequently for level crossing as $\delta > 4$. This is also consistent with Lichnerowicz’s theorem for the three-dimensional Dirac operator that requires negative curvature for the existence of harmonic spinors.
Extending the discussion to the three-dimensional Dirac-Lifshitz operator $i \gamma^i {\cal D}_i$, we note that its eigen-values $Z$ on Berger spheres are simply expressed in terms of the eigen-values $\zeta$ of the corresponding Dirac operator as Z = (\^2 + [1 8 ]{} (\^2 - 4) + M\^2 )\^ . It turns out that $Z$ have the same sign as $\zeta$ for all values of $\delta$, and, hence, level crossing occurs at the values of $\delta$ given by [(\[moutsia\])]{} as before. Also, since the multiplicity of the eigen-values of $Z$ is the same as $\zeta$, the number of modes that undergoes level crossing, $\Delta S(S^3)$, is the same for both operators. This is consistent with the fact that the index of the four-dimensional Dirac-Lifshitz operator is the same as the index of the Dirac operator on all such geometrical backgrounds $\mathbb{R} \times S^3$, i.e., ${\rm Ind}({\cal D})={\rm Ind}(D)$, and which can be computed by spectral flow methods via equation [(\[specri\])]{}.
The instanton of $z=3$ Hořava-Lifshitz gravity with $SU(2) \times U(1)$ isometry that interpolates smoothly between the round sphere and the Berger sphere with $\delta$ given by [(\[othfix\])]{} leads to violation of chiral charge conservation provided that the volume of space (which remains fixed for all $t$) is bounded from below as (S\^3) = 16 \^2 \^[3/2]{} = 54 \^2 \^4 (- [\_[w]{}\^2 ]{} )\^3 > 13824 \^2 (- [\_[w]{}\^2 ]{} )\^3 letting $\delta > 4$ in the final step. Thus, chiral symmetry breaking becomes possible when the mean radius of space is sufficiently larger than the range of interaction $\sim \kappa_{\rm w}^2 / |\omega|$ in topologically massive gravity, which is associated to the superpotential functional $W[g]$. This provides a qualitative criterion for chiral symmetry breaking in Hořava-Lifshitz gravity that satisfies the detailed balance condition and which generalizes beyond the simple example we have considered here. $\Delta Q_5$ is given by the number of modes (including their multiplicities) that have undergone level crossing.
Comparison to Einstein gravity reveals that novel phenomena become possible in non-relativistic gravitational theories leading to fermion number violation. According to a scenario, such theories are thought to provide the ultra-violet completion of gravity sacrificing relativistic invariance for power counting renormalizability at very high energies. The phase of Hořava-Lifshitz gravity arising at $\lambda = 1/3$ seems to be more appropriate in this context although the flow to ordinary gravity in the infra-red regime and the emergence of relativistic invariance as low energy phenomenon has not been made quantitative to this day. The asymptotic safety programme for gravity seems to provide a promising framework to address this fundamental issue in future studies.
Conclusions
===========
We outlined the general relation between anomalies, instantons and chiral symmetry breaking in relativistic and Lifshitz field theories. The quantum anomaly of the axial current conservation law of massless fermions is independent of the anisotropy scaling parameter and coincides with the result obtained for the relativistic case in the background of gauge and metric fields. Likewise, the index of the Dirac-Lifshitz operator is universal given by the integrated form of the axial anomaly. This is in agreement with the infra-red nature of the axial anomaly, which is inert to higher derivative terms that become relevant in the ultra-violet regime.
The difference between relativistic and Lifshitz field theories lies in the ability of their instantons to affect the conservation law of chiral charge. The main result in this context is the construction of simple instanton solutions of gravitational Lifshitz theories and their use to induce chiral symmetry breaking for certain range of the couplings, leading to baryon and lepton number violation triggered by gravity. This novel possibility does not arise in general relativity. It remains to construct instanton solutions of Lifshitz gauge theories and examine their effect on chiral symmetry breaking in comparison to ordinary gauge theories.
**Acknowledgements**
I thank the conference organizers for their kind invitation to give an account of this work in an exciting scientific environment as well as for the partial financial support.
[99]{} I. Bakas and D. Lüst, “Axial anomalies of Lifshitz fermions”, Fortschr. Phys. (2011) 937 \[arXiv:1103.5693 \[hep-th\]\]. I. Bakas, “More on axial anomalies of Lifshitz fermions”, Fortschr. Phys. (2012) 224 \[arXiv:1110.1332 \[hep-th\]\]. I. Bakas, F. Bourliot, D. Lüst and M. Petropoulos, “Geometric flows in Hořava-Lifshitz gravity”, JHEP (2010) 131 \[arXiv:1002.0062 \[hep-th\]\]. I. Bakas, “Gradient flows and instantons at a Lifshitz point”, J. Phys. Conf. Ser. (2011) 012004 \[arXiv:1009.6173 \[hep-th\]\].
[^1]: Based on talks delivered at the XLI-ème Institut d’été de LPTENS [*Cordes, Particules, et l’Univers*]{}, 17 August – 2 September 2011, Paris, France and at the EISA Summer Institute [*Workshop on Fields and Strings: Theory – Cosmology – Phenomenology*]{}, 14 – 18 September 2011, Corfu, Greece; contribution to appear in the Proceedings of Science.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Luis V. Dieulefait, Jorge Jiménez-Urroz, Kenneth A. Ribet[^1]\'
title: '[**Modular forms with large coefficient fields via congruences** ]{}'
---
-20mm
Introduction
============
In this paper we will exploit the theory of congruences between modular forms to deduce the existence of newforms (in particular, cuspidal Hecke eigenforms) with levels of certain specific types having arbitrarily large coefficient fields. We will only consider newforms of weight $2$ and trivial nebentypus.\
If the level is allowed to be divisible by a large power $n$ of a fixed prime, or by the cube of a large prime $p$, then the coefficient fields of [**all**]{} newforms of this level will grow with $n$ (with $p$, respectively) due to results of Hiroshi Saito (cf. [@Sa], Corollary 3.4; see also [@Br]) showing that the maximal real subfield of certain cyclotomic field whose degree grows with $n$ (with $p$, respectively) will be contained in these fields of coefficients. Thus, it is natural to deal with the question when the levels are square-free or almost-square-free, i.e., square-free except for the fact that they are divisible by a fixed power of a small prime.\
In the square-free case, for any given number $t$, we will prove that in levels which are the product of exactly $t$ primes there are newforms with arbitrarily large coefficient fields. We will recall results of Mazur on reducible primes for newforms of prime level that give the case of $t=1$. Then, a generalization of these results will allow us to deduce the case $t=2$. For $t \geq 3$ we follow a completely different approach, namely, we exploit congruences involving certain elliptic curves whose construction is on the one hand related to Chen’s celebrated results on (a partial answer to) Goldbach’s conjecture (cf. [@Ch]) and on the other hand inspired by Frey curves as in the proof of Fermat’s Last Theorem (the diophantine problem that we will consider will be a sort of Fermat-Goldbach mixed problem). It is via the level lowering results in [@Ri] that the desired congruence will be guaranteed. The precise statement of our first main result is the following:\
\[squarefree\] Let $B$ and $t$ be two given positive integers. Then, there exist $t$ different primes $p_1, p_2, ...., p_t$ such that if we call $N$ their product, in the space of cuspforms of weight $2$, level $N$ and trivial nebentypus there exists a newform $f$ whose field of coefficients $\Q_f$ satisfies: $$[\Q_f : \Q] > B.$$
In the almost-square-free case, we will consider levels $N=2^kp_1\dots p_t$ which are square-free except that they are divisible by a small power of $2$. We will prove that for any fixed $t$, among newforms with such levels the fields of coefficients have unbounded degree. We will use congruences with certain $Q$-curves constructed from solutions to the problem of finding prime values attained by the expression $(x^4+y^2)/c$. Again, these $Q$-curves will also have some features inspired by Frey curves, and the existence of the desired congruences will be a consequence of level lowering. For $c=1$ it is a celebrated result of Friedlander and Iwaniec that infinitely many primes are of the form $x^4 + y^2$ (cf. [@fi2]). Here we give the following generalization
\[mainintro\] Let $B>0$ fixed, and $\Lambda$ the usual Von Mangold function. Then, we have uniformly in $c\le (\log x)^B$ $$\sum\sum_{\kern-10pt c|a^2+b^4\le cx}\Lambda((a^2+b^4)/c)=K(c)x^{3/4}+o\left(x^{3/4}\right)$$ where $a,b$ run over positive integers, and $K$ is completely explicit in terms of $c$.
In Theorem \[main\] in section 6 we give the precise value of $K(c)$. For our application to congruences between modular forms we only need a mild version of the particular case with fixed $c$ of the form $c =5^\ell$. The following is a direct consequence of the previous theorem
\[primevalues\] Let $c$ be a positive odd integer. Then there are infinitely many primes of the form $(x^4+y^2)/c$ if and only if $c$ can be written as the sum of two squares.\
The precise statement of our second main result, the one covering the almost-square-free level case, is the following:
\[almostsquarefree\] Let $B$ and $t$ be two given positive integers. Then, there exist $\alpha \in \{ 5, 8\}$ and $t$ different odd primes $p_1, p_2, ...., p_t$ such that if we call $N$ the product of these $t$ primes, in the space of cuspforms of weight $2$, level $2^\alpha N$ and trivial nebentypus there exists a newform $f$ with field of coefficients $\Q_f$ satisfying: $$[\Q_f : \Q] > B.$$
Let us stress that the results on prime values of $(x^4+y^2)/c$, Theorem \[mainintro\] and its corollary, are interesting in its own right, independently of the application to finding newforms with large coefficient fields.
Theorem \[squarefree\] for the case of prime level: Mazur’s argument
====================================================================
Suppose that the level $N$ is prime and that $\ell >3$ is a prime that divides $N - 1$. Then it is proved in [@Ma] that the prime is Eisenstein, meaning that there is a newform $f$ of weight $2$ and level $N$ such that if we call $K_f$ its field of coefficients there is a prime $\lambda$ dividing $\ell$ in the ring of integers of $K_f$ for which we have $a_p \equiv 1 + p \; \mod{\lambda}$ for all primes $p$. The residual mod $\lambda$ Galois representation attached to $f$ is reducible. In particular, we have $$a_2 \equiv 3
\; \mod{\lambda} \qquad \quad (1)$$ The coefficients of the modular form $f$ and those of any Galois conjugate $f^\sigma$ all satisfy the bound $| a_p | \leq 2 \sqrt{p}$, in particular $a_2$ and all its Galois conjugates have absolute value bounded above by $2 \sqrt{2}
< 3$. Then, $a_2 -3$ is a non-zero algebraic integer, whose norm is divisible by $\ell$ because of congruence (1) and with absolute value at most $(3 + 2 \sqrt{2})^{\deg k_f}$. Hence $$\ell \leq (3 + 2 \sqrt{2})^{\deg k_f}$$ Thus, $\deg K_f$ is bigger than a fixed constant times $\log \ell$. Taking $\ell$ big and using Dirichlet’s theorem to find an $ N
\equiv 1 \; \mod{\ell}$, we can make $\deg K_f$ as big as we like. This proves the case of prime level ($t=1$) of Theorem \[squarefree\].\
A Frey curve adapted to Chen results, and the case $t
\geq 3$ of Theorem \[squarefree\]
=====================================================
Let $\ell$ be a (large) prime number, and assuming for the moment the truth of Goldbach’s conjecture let us write the even number $2^{\ell + 4}$ as the sum of two prime numbers: $2^{\ell + 4} = p +
q$. Since $p$ and $q$ are clearly non-congruent modulo $4$, we assume without loss of generality that $p \equiv 3 \; \mod{4}$. Let $F$ be the semistable Frey curve associated to the triple $p , q,
2^{\ell + 4}$: $$y^2 = x (x- p) (x+ 2^{\ell + 4})$$
Its conductor is $2pq$, while its minimal discriminant is $\Delta =
(2^{\ell
+ 4} pq)^2/ 2^8 = (2^\ell pq)^2$.\
The modularity of all semistable elliptic curves, proved by Wiles in [@Wi], implies that there is a newform $f$ of weight $2$ and level $2pq$ corresponding to $F$.\
The mod $\ell$ Galois representation $F[\ell]$ of ${{\rm Gal}}(\bar{\Q}/\Q)$ is irreducible by results of Mazur, and unramified at $2$ because the $2$-adic valuation of the discriminant is divisible by $\ell$ (as in the original Frey curves related to solutions of Fermat’s Last Theorem). Although it comes initially from a newform $f$ of level $2pq$, by level-lowering (cf. [@Ri]) it arises also from a newform $f'$ of level $pq$.\
The trace of the action of ${{\rm Frob }}\; 2$ on $F[\ell]$ is $\pm (1 +
2)$, because this is the well-known necessary condition for level-raising, i.e., for the existence of an $\ell$-adic Galois representation with semistable ramification at $2$ providing a lift of $F[\ell]$, and we have such a lift by construction: it is given by the Galois action on the full $\ell$-adic Tate module $T_\ell(F)$ of the curve $F$. So if we call $\{ a_p\}$ the coefficients of $f'$ we get $$a_2 \equiv \pm 3 \; \mod{\lambda} \qquad \; (2)$$ for a prime $\lambda$ dividing $\ell$ in the field of coefficients of $f'$. From this congruence we can argue as we did in the previous section using congruence (1) an conclude easily that just by choosing the prime $\ell$ sufficiently large we can construct newforms of weight $2$ and level $N= pq$ a product of two different primes with arbitrarily large field of coefficients.\
Since Goldbach’s conjecture remains open, in order to get an unconditional result we need to move to the case of three primes in the level. Using the results of Chen on Goldbach’s problem (cf. [@Ch]) we know that for $\ell$ sufficiently large $2^{\ell + 4}$ can be written as the sum of a prime and a “pseudo-prime", i.e., a number that is either a prime or the product of two different primes. Then, in particular, one of the following is true for infinitely many $\ell$: $2^{\ell + 4}$ can be written as the sum of two primes $p$ and $q$, or $2^{\ell + 4}$ can be written as the sum of a prime $p$ and the product of two primes $qr$. If the first is true, as we have just seen, this will prove Theorem \[squarefree\] for levels which are the product of two primes, and if the second is true a similar argument with the triple $p, qr, 2^{\ell + 4}$ shows that the theorem is true for levels which are the product of three primes.\
Thus, to finish the proof of the case $t=3$ of Theorem \[squarefree\], it remains to show that if the result holds for $t=2$ then it also holds for $t=3$. But this is just an application of raising the level (cf. [@Rilr]), because whenever we have a modular form of level $pq$ and an irreducible mod $\ell$ Galois representation attached to it, the same residual representation is also realized in some newform of level $pqr$, as long as the prime $r$ satisfies the required condition for level raising (and it is well known that there are infinitely many primes $r$ that satisfy this condition, cf. [@Rilr]). Thus, whenever we have found a newform $f'$ as in the previous argument, of level $pq$ and satisfying (2), there are also newforms with levels of the form $pqr$ also satisfying (2) and from this the proof of the theorem for the case $t=3$ follows exactly as explained above.\
To treat the case of more than three primes, we modify the argument above by further raising the level. Starting with the irreducible mod $\ell$ representation afforded by $F[\ell]$ of conductor either $pq$ or $pqr$, and for any given $t \geq 4$, we can find forms giving the same residual representations in levels equal to the product of $t$ primes by just raising the level $t-2$ ($t-3$, respectively) times. For this, we have to take care not to lose spurious primes as we add on new ones. The required analysis is carried out in [@DT].\
We conclude that Theorem \[squarefree\] is true for any $t\geq 3$. Since the case of prime level was dealt with in the previous section, at this point only the case of $t=2$ remains unsolved (and a proof of Goldbach’s conjecture would be enough to handle it).
The case $t=2$ of Theorem \[squarefree\] via a result of Ogg
============================================================
One way to treat Theorem \[squarefree\] for the case where $N =
pq$ (without proving Goldbach’s conjecture) is to appeal to the results of Ogg in [@O]. If $p$ and $q$ are distinct primes, Ogg finds a degree $0$ cuspidal divisor on $X_0(pq)$ whose image on $J_0(pq)$ has order equal to the numerator of the fraction $(p -
1)(q + 1)/24$. Take $\ell > 3$. If $\ell$ divides $(p - 1)(q + 1)$, using exactly the same arguments applied in (cf. [@Ma]) in the case of prime level we find an eigenform $f$ at level $pq$ that is Eisenstein mod $\ell$, therefore giving a reducible residual mod $\ell$ representation. In particular, this means that the coefficient $a_2$ satisfies again congruence (1) as in section 2, for some prime $\lambda$ dividing $\ell$ in its field of coefficients $K_f$, and we deduce as before that the degree of $K_f$ is large (it grows with $\ell$).\
We need to ensure that $f$ is genuinely a newform, i.e., that its eigenvalues do not arise at level $p$ or at level $q$. We begin as before by taking $\ell$ large. Then we find $q \equiv -1 \;
\mod{\ell}$ and pick $p$ to be a random prime that is not congruent to $1$ mod $\ell$. Since the Eisenstein primes at prime level $N$ are divisors of $N -1$, we see that $\ell$ is not an Eisenstein prime at either level $p$ or level $q$ while it is an Eisenstein prime at level $pq$, thus the form $f$ must be a newform of level $pq$. This completes the proof of the case $t=2$ of Theorem \[squarefree\]. Thus, putting together the results of the last three sections, we conclude that Theorem \[squarefree\] for any positive value of $t$.
The proof of Theorem \[almostsquarefree\]
==========================================
To prove Theorem \[almostsquarefree\] we follow a strategy similar to the one explained in section 3, except that now we will start from a diophantine equation such that the elliptic curve corresponding to any solution is a $Q$-curve defined over $\Q(i)$. For the Fermat-type equation $x^4+ y^2 = z^p$ an attached $Q$-curve was proposed by Darmon in [@Da] and in the work of Ellenberg [@El] it was shown using the modularity of this curve that the diophantine equation does not have non-trivial solutions for large $p$. We will consider instead the diophantine problem:
$$x^4 + y^2 = 5^{\ell} p$$ The result that we will prove in the next sections (see Theorem \[main\] in section 6, specialized to the case $c= 5^\ell$), which is a generalization of the case $\ell = 0$ solved by Friedlander-Iwaniec in [@fi2], implies that for any prime exponent $\ell$ there exist infinitely many primes $p$ such that there are integer solutions $A,B$ to this equation.\
Thus, if $\ell$ is a given prime and $A, B, p$ satisfy $$A^4 + B^2 = 5^\ell p$$ with $p$ prime, we consider, as in the work of Darmon and Ellenberg, the elliptic curve $E$: $$y^2 = x^3 + 4 A x^2 + 2 (A^2 + i B) x$$ For simplicity, and since we have infinitely many primes $p$ satisfying the equation, we assume $p \neq \ell$.\
The following properties of this curve are known (cf. [@El]):\
$E$ is $2$-isogenous to its Galois conjugate, in particular it is a $Q$-curve. If we call $W$ its Weil restriction defined over $\Q$, it is a ${{\rm GL}}_2$-type abelian surface and thus it has a compatible family of $2$-dimensional Galois representations of $G_\Q$ attached. These representations have coefficients in $\Q(\sqrt{2})$. This abelian surface is semistable outside $2$ and it is modular. Computing the conductor of $W$ it follows that the modular form $f$ attached to $W$ has level $2^\alpha 5 p$, with $\alpha = 5$ or $8$. It has weight $2$ and trivial nebentypus. This newform has coefficients in $\Q(\sqrt{2})$ and it has an inner twist.\
We now consider for the prime $\ell$ we started with and $\lambda \mid \ell$ in $\Q(\sqrt{2})$ the residual mod $\lambda$ representation $\bar{\rho}_{W,\lambda}$ attached to $W$. Assuming that $\ell > 13$ it is known that this representation is irreducible (cf. [@El]). Since the discriminant of $E$ is $512 (A^2 + i B) 5^\ell p $ we can, as in [@El], apply the Frey trick at the semistable prime $5$ (observe that $5$ is unramified in $\Q(i) / \Q$): locally at $5$ the valuation of the discriminant is divisible by $\ell$ (on the other hand, this does not happen locally at the prime $p$). Thus, we conclude that $\bar{\rho}_{W,\lambda}$ is unramified at $5$: more precisely it has conductor $2^\alpha p$ with $\alpha \in \{ 5 , 8\}$. If we apply lowering the level (cf. [@Ri]) we see that there is a newform $f'$ of level $2^\alpha p$, weight $2$ and trivial nebentypus such that some of its corresponding residual Galois representations in characteristic $\ell$ is $\bar{\rho}_{W,\lambda}$.\
Now we can conclude as in section 3: If we look at the coefficient $a_5$ of $f'$ since we know that there is a mod $\lambda$ congruence with a newform $f$ corresponding to an abelian variety which is semistable at $5$ (namely, the abelian surface $W$) then by the necessary condition for level raising we know that it must hold: $$a_ 5 \equiv \pm 6 \qquad \mod{\lambda}$$ From this congruence we see, as in sections 2 and 3, that the minimal field of definition of $a_5$, and a fortiori the field of coefficients of $f'$, has a degree that grows with $\ell$. This solves the case of almost-square-free level $2^\alpha p $ with $\alpha = 5$ or $8$, i.e., the case $t=1$, of Theorem \[almostsquarefree\]. The case of Theorem \[almostsquarefree\] for levels of the form $2^\alpha p_1 \cdot ....\cdot p_t$, with $t > 1$ fixed and the $p_i$ odd, different primes, can be deduced from this by $t-1$ applications of raising the level as explained in section 3.
Prime values of $(x^4 + y^2)/c$
================================
We now introduce some notation which will be used from now on. For any given prime $p$, and any integer $d$ we denote $v_p(d)$ the highest power of $p$ dividing $d$. Moreover, we will write $d$ as $d=d_1d_2^2=d_1d_3^2d_4^4$, where $d_1,d_3$ are squarefree. We will consider $\Lambda(r)$ the usual Von Mangoldt function, extended as zero over non integer numbers. Then, the main result of this section is the following theorem.
\[main\] Let $B>0$ fixed. We have uniformly in $c\le (\log x)^B$ $$\sum\sum_{\kern-10pt(a^2+b^4)/c\le x}\Lambda((a^2+b^4)/c)=4\pi^{-1}\kappa G(c) (cx)^{3/4}+o\left(x^{3/4}\right)$$ where $a,b$ run over positive integers, $G$ is a multiplicative function, and $$\label{kappa}
\kappa=\int_0^1(1-t^4)^{1/2}dt=\Gamma(\frac{_{_1}}{^{^4}})^2/6\sqrt{2\pi}.$$
[**Remark:**]{} It is important to emphasize here that the constant $G(c)$, explicitly described in Lemma \[multgeneral\], can take the zero value, and it does precisely in those $c$ which are non representable as the sum of two squares, or such that $v_2(c)\equiv
3\tm 4$. For trivial reasons there is at most one prime in the sequence $(a^2+b^4)/c$ in these cases, since none of the elements is in fact coprime with $c$ if $c$ is non representable as the sum of two squares, and if $v_2(c)\equiv 3\tm 4$, every integer of the sequence has to be even. Hence, the proof that follows restricts to those values of $c$ such that this constant is non zero since, in any other case, the result is trivial.
Theorem \[main\] is an easy generalization of Theorem 1.1 in [@fi2]. However, some of the computations done in [@fi2] do not apply to this case in a straightforward manner and, hence, they must be done now with the required level of generality in the variable $c$. In particular, the proof of Theorem \[main\] relies in the verification of the hypothesis needed to apply the Asymptotic Sieve due to Friedlander and Iwaniec in [@fi1], but now for the sequence $a(c)_n=0$ for any $(n,c)>1$, and for $n$ coprime to $c$ given by $$\label{sequence}
\ac{c}_n=\sum\sum_{\kern-10pt(a^2+b^2)/c=n}\mathfrak Z(b),$$ where $a,b$ are integers non necessarily positive, and $\mathfrak Z$ is the function with value $\mathfrak Z(m^2)=2$, for any integer $m\ne 0$, $\mathfrak Z(0)=1$, and $\mathfrak Z(b)=0$ in any other case. From now on we will only consider integers $n$ coprime to $c$ and, then, we have $\ac{c}_n= a^{old}_{cn}$ where $a^{old}_n$ is the sequence related with Theorem 1 of [@fi2]. We now include for reading convenience the hypotheses and main result of the Asymptotic Sieve. The following, with the exception of (\[divisor\]), is basically a copy of Section 2 in [@fi2]. We explain the difference between (\[divisor\]) and (2.8) in [@fi2] at the end of this section.
Consider a sequence of real, nonnegative, numbers ${\cal A}=(a_n)_{n\ge 1}$, and $x$ a positive number. We want to obtain an asymptotic formula for $$S(x)=\sum_{p\le x}a_p\log p,$$ where the sum runs over prime numbers, in terms of $\A(x)=\sum_{n\le x} a_n$. We suppose $$\begin{aligned}
\A(x)&\gg&\A(\sqrt{x})(\log x)^2\label{a:uno},\\
\A(x)&\gg&x^{1/3}\left(\sum_{n\le x}a_n^2\right)^{1/2}.\label{a:dos}\end{aligned}$$ As usual in sieve theory, we will assume that for any integer $d>1$ $$\A_d(x)=\sum_{{n\le x}\atop{d|n}}a_n=g(d)\A(x)+r_d(x),$$ where $g$ is a multiplicative function, and $r_d(x)$ is regarded as an error term. For the function $g$ we assume the following hypotheses $$\begin{aligned}
&&0\le g(p^2)\le g(p)<1,\label{g:uno}\\
&&g(p)\ll p^{-1},\label{g:dos}\\
&&g(p^2)\ll p^{-2},\quad\text{ and }\label{g:tres}\\
&&\sum_{p\le y}g(p)=\log\log y+ e+O((\log y)^{-10}),\label{g:cuatro}\end{aligned}$$ for every $y$ and some $e$ depending only on $g$. For the error term we will assume $$\label{super}
\sum_{d\le DL^2}^{\kern15pt 3}|r_d(t)|\le A(x)L^{-2},$$ uniformly in $t\le x$, for some $D$ in the range $$\label{range:D}
x^{2/3}<D<x.$$ The superscript $3$ in (\[super\]) restrict the sum to cube free moduli and $L=(\log x)^{2^{24}}$. We also require $$\label{divisor}
\A_d(x)\ll d^{-1}\tau(d)^8\A(x)\log x\text{ uniformly in }d\le x^{1/3},$$ and finally an estimate in bilinear forms like $$\label{bilinear}
\sum_{m}|\sum_{{N<n\le 2N}\atop{{mn\le x}\atop{(n,m\Pi)=1}}}\beta(n)\mu(mn)a_{mn}|\le \A(x)L^{-4},$$ where the coefficients are given by $$\label{coeff:bilinear}
\beta(n)=\beta(n,K)=\sum_{k|n,k\le K}\mu(k),$$ for any $K$ in the range $$\label{range:K}
1\le K\le xD^{-1},$$ $N$ verifing $$\label{range:N}
\Delta^{-1}\sqrt{D}< N<\delta^{-1}\sqrt{x}$$ for some $\Delta\ge\delta\ge 2$, and $\Pi$ is the product of all primes $p<P$ for some $P$ which can be chosen conveniently in the range $$\label{range:P}
2\le P\le\Delta^{1/2^{35}\log\log x}.$$ In this conditions we have
\[prop1\] Let $\cal A$ be a sequence verifying the above hypotheses. Then, $$S(x)=H\A(x)\left\{1+O\left(\frac{\log \delta}{\log \Delta}\right)\right\}$$ where $H$ is the positive constant given by the convergent product $$\label{hache}
H=\prod_p(1-g(p))(1-\frac1p)^{-1},$$ and the implied constant depends only on the function $g$.
Normally $\delta$ is a large power of $\log x$ and $\Delta$ a small power of $x$.
[**Remark:**]{} It is important to note that (\[divisor\]) is not the original assumption (1.6) in [@fi1], but a slightly weaker. However, as the authors mention in that paper, (1.6) is only required to reduce the hypotheses $$\label{bilinear:dos}
\sum_{m}\tau_5(m)|\sum_{{N<n\le 2N}\atop{{mn\le x}\atop{(n,m\Pi)=1}}}\beta(n)\mu(mn)a_{mn}|\le \A(x)(\log x)^{-3},$$ and $$\label{super:dos}
\sum_{d\le D}\mu^2(d)\tau_5(d)|r_d(t)|\le A(x)(\log x)^{-3},$$ to (\[super\]) and (\[bilinear\]). We just have to follow the reasoning in Section 2, p. 1047 of [@fi1] to see that this reduction is also possible with our hypothesis (\[divisor\]).
Proof of Theorem \[main\]
=========================
To prove Theorem \[main\] we will use Proposition \[prop1\] for the sequence given in (\[sequence\]). Hence, we have to check that the sequence verifies hypotheses (\[a:uno\]) trhough (\[bilinear\]). Given an integer $d\ge 1$, we denote $\A_d(x;c)=\sum_{n\le x,n\equiv 0\tm d}\ac{c}_n$. The first thing that needs to be done is to find a good aproximation of $\A_d(x;c)$ in terms of a multiplicative function. Now, $\A_d(x;c)=0$ for $(d,c)>1$ and for $(d,c)=1$ we have $$\label{moebius}
\A_d(x;c)=\sum_{k|c}\mu(k)\A^{old}_{ckd}(cx),$$ and we know by [@fi2] that $$\A^{old}_d(x)=g(d)\A^{old}(x)+r^{old}_d(x),$$ where the functions $g$, $r^{old}$ satisfy conditions (\[a:uno\]) trhough (\[bilinear\]). Note that for any integer $d$ the definition of $\A^{old}_{d}(x)$ is implicit in (\[moebius\]) for $c=1$, and observe that $g$, $r^{old}$ are precisely the functions $g$, $r$ appearing in [@fi2]. Hence, to approximate $\A_d(x;c)$ we are tempted to use the approximation of $\A^{old}(x)$ given in Lemma 3.4 of [@fi2]. However, this lemma only works for cubefree integers $d$ which do not cover completely our case, since $c$ will be any number $c\le (\log x)^B$. Hence, our next objective is to generalize Lemma 3.4 of [@fi2] to any integer $d$. As in [@fi2], we start approximating $\A_d(x;c)$ by $$M_d(x;c)=\sum_{k|c}\mu(k)\frac1{ckd}\sum\sum_{\kern-10pt0<(a^2+b^2)\le cx}\mathfrak Z(b)\rho(b;ckd),$$ for any $d$ coprime to $c$, where $\rho(b,d)$ denotes the number of solutions $\ap\tm d$ to the congruence $\ap^2+b^2\equiv 0\tm{d}$, and $M_d(x;c)=0$ otherwise. The following is a trivial consequence of Lemma 3.1 of [@fi2].
\[error\] Let $B>0$. For any $c\le (\log x)^B$ we have $$\sum_{d\le D}|\A_d(x;c)-M_d(x;c)|\ll D^{1/4}x^{9/16+\ep}$$ for any $D\ge 1$ and $\ep>0$ and the implied constant depending only on $\ep$.
Now, we need to find out the main term of $M_d(x;c)$, as we mentioned, by generalizing Lemma 3.4 of [@fi2].
\[multgeneral\] Let $B>0$. We have uniformly for any $c\le (\log x)^B$ $$M_d(x;c)=g_c(d)\left(4\kappa c^{3/4}G(c)\right) x^{3/4}+O\left(h(d)H(c)x^{1/2}\right),$$ where $\kappa$ is given in (\[kappa\]), $g_c(d)=0$ for any $(c,d)>1$, and $g_c(d)=g(d)$ otherwise where $g$ and $h$ are the multiplicative functions given by $$\label{defg}
g(d)=\frac1d\sum_{\nu_4|d_4}\nu_4^2\sum_{{\nu_3|d_3}\atop{(\nu_3,\frac{d_4}{\nu_4})=1}}\nu_3(\nu_3,d_1^*)\sum_{{\nu_1|\frac{d_1^*}{(\nu_3,d_1^*)}}\atop{(\nu_1,\frac{d_3d_4}{\nu_3\nu_4})=1}}
\rho\left(\delta(\nu_1,\nu_3,\nu_4)^2\right)\frac{\varphi\left(\delta(\nu_1,\nu_3,\nu_4)\right)}{d_1^*d_3d_4},$$ and
$$\label{defh}
\kern-2cm h(d)=\frac 1d\sum_{\nu_4|d_4}\nu_4^2\sum_{{\nu_3|d_3}\atop{(\nu_3,\frac{d_4}{\nu_4})=1}}\nu_3\sum_{{\nu_1|\frac{d_1^*}{(\nu_3,d_1^*)}}\atop{(\nu_1,\frac{d_3d_4}{\nu_3\nu_4})=1}}
\rho\left(\delta(\nu_1,\nu_3,\nu_4)^2\right)\tau(\delta(\nu_1,\nu_3,\nu_4))).$$
Here $d_1^*=d_1/(d_1,2)$ and $\delta(\nu_1,\nu_3,\nu_4)=\frac{d_1^*d_3d_4}{(d_1^*,\nu_3)\nu_1\nu_3\nu_4}$. Finally $G(c)=\sum_{k|c}\mu(k)g(ck)$ and $H(c)=c^{1/2}\sum_{k|c}h(ck)$.
[*Proof:*]{} We restrict only to integers $d$ coprime to $c$ since the result is trivial otherwise. Given $d=d_1d_2^2=d_1d_3^2d_4^4$, with $d_1, d_3$ squarefree, and an integer $b$, let us call $b_2=(b,d_2)$, $b_1=(b/b_2,d_1^*)$. Then, it is fairly straightforward to prove that $$\rho(b;d)=b_2\rho\left((d_1^*d_2/b_1b_2)^2\right)$$ where $\rho$ is the multiplicative function given by $$\rho(p^\ap)=1+\chi_4(p),$$ whit $\chi_4$ the character of conductor $4$, except $\rho(d)=0$ if $4|d$. Now, by definition, we have $$\label{summob}
M_d(x;c)=\sum_{k|c}\mu(k)M_{ckd}^{old}(cx)$$ where $$\begin{aligned}
M_d^{old}(x)&=&\frac1{d}\sum\sum_{\kern-10pt0<(a^2+b^2)\le x}\mathfrak Z(b)\rho(b;d)=\frac2{d}\sum_{|r|\le x^{1/4}}\rho(r^2,d)\left\{ (x-r^4)^{1/2}+O(1)\right\}\\
&=&T_d+E_d,\end{aligned}$$ $T_d$ being the main term of $M_d^{old}(x)$, and $$E_d\ll\frac1{d}\sum_{|r|\le x^{1/4}}\rho(r^2,d)\ll x^{1/4}.$$ Now suppose $(r,d_4)=\nu_4$. Then $(r^2,d_3d_4^2)=\nu_4^2(r/\nu_4,d_3)$, and splitting the sum over the divisors of $d_4$ we get, after some calculations, $$\begin{aligned}
T_d(x)&=&\frac{2}{d}\sum_{\nu_4|d_4}\nu_4^4\sum_{{|r|\le z^{1/4}}\atop{(r,\frac{d_4}{\nu_4})=1}}(r,d_3)\rho\left(\left(\frac{d_1^*}{(r,d_1^*)}\frac{d_3}{(r,d_3)}\frac{d_4}{\nu_4}\right)^2\right)(z-r^4)^{1/2},\end{aligned}$$ where $z=z(\nu_4)=x/\nu_4^4$. Similarly, splitting the inner sum over the divisors of $d_3$ and $d_1^*$, it is easy to get $$T_d(x)=\frac{2}{d}\sum_{\nu_4|d_4}\nu_4^4\sum_{{\nu_3|d_3}\atop{(\nu_3,\frac{d_4}{\nu_4})=1}}
\nu_3^3\sum_{{\nu_1|\frac{d_1^*}{(\nu_3,d_1^*)}}\atop{(\nu_1,\frac{d_3d_4}{\nu_3\nu_4})=1}}
\nu_1^2
\rho\left(\delta(\nu_1,\nu_3,\nu_4)^2\right)
\sum_{{|r|\le u^{1/4}}\atop{{(r,\delta(\nu_1,\nu_3,\nu_4)=1}}}
(u-r^4)^{1/2},$$ where $\delta(\nu_1,\nu_3,\nu_4)=\frac{d_1^*d_3d_4}{(\nu_3,d_1^*)\nu_1\nu_3\nu_4}$, and $u=u(\nu_1,\nu_3,\nu_4)=z/(\nu_3\nu_1)^4$. Estimating the inner sum, (also done in Lemma 3.4 of [@fi2]), we get $$\begin{gathered}
T_d(x)=\frac{2}{d}\sum_{\nu_4|d_4}\nu_4^4\sum_{{\nu_3|d_3}\atop{(\nu_3,\frac{d_4}{\nu_4})=1}}\nu_3^3\sum_{{\nu_1|\frac{d_1^*}{(\nu_3,d_1^*)}}\atop{(\nu_1,\frac{d_3d_4}{\nu_3\nu_4})=1}}\nu_1^2
\rho\left(\delta(\nu_1,\nu_3,\nu_4)^2\right)\left\{\frac{\varphi\left(\delta(\nu_1,\nu_3,\nu_4)\right)}{\delta(\nu_1,\nu_3,\nu_4)}{2\kappa u^{3/4}}+\right.\\
\left. \frac{}{}O(\tau(\delta(\nu_1,\nu_3,\nu_4))u^{1/2})\right\}.\end{gathered}$$ Substituting the value of $u$ in the previous formula we get, for any integer $d\ge 1$ not necessarily coprime with $c$, $$M_d^{old}(x)=g(d)4\kappa x^{3/4}+O(h(d)x^{1/2}),$$ where $g$ and $h$ are the multiplicative functions given by (\[defg\]) and (\[defh\]) respectively. We just have to plug this into (\[summob\]) to get the result with $g_c=g$ if $(d,c)=1$, and $0$ otherwise. Note that $$G(c)=\sum_{k|c}\mu(k)g(ck)=\prod_{p|c}\left(g(p^{v_p(c)})-g(p^{v_p(c)+1})\right),$$ defines a multiplicative function. Also, observe that the multiplicativity of both $g$ and $h$ is a direct consequence of the definition in each case. It is straightforward to see that the value at prime powers is given by $g(2^{4\ap+r})=1/2^{3\ap+r}$, and for $p$ odd, $$\label{firstg}
g(p^{4\ap+r})=
\frac1{p^{4\ap+r}}(1+\chi_4(p))\left(1-\frac1{p}\right)\frac{p^{\ap}-1}{p-1}+\frac1{p^{3\ap+r}}g_r(p)$$ where $$\label{secondg}
g_{r}(p)=\begin{cases}1
&\text{ if } r=0\\
1+\chi_4(p)\left(1-\frac1p\right)&\text{ if } r=1\\
1+(1+\chi_4(p))\left(1-\frac1p\right)&\text{ if } r=2,3.\\
\end{cases}$$ For $h$ we will only need its value for cubefree integers which comes from the following $$\label{h:primes}
h(p)p=1+2\rho(p),\qquad h(p^2)p^2=p+2\rho(p),$$ already gotten in Lemma 3.4 of [@fi2].
We are now in position to verify hypotheses (\[a:uno\]) trhough (\[bilinear\]) for the approximation $$\A_d(x;c)=g_c(d)\A(x;c)+r_d(x).$$ First of all we note that, by Lemmas \[error\] and \[multgeneral\], we have $$\label{newa}
\A_d(x;c)=g(d)4\kappa G(c) (cx)^{3/4}+O(d^{1/4}x^{9/16+\ep})+O(H(c)h(d)x^{1/2}),$$ meanwhile, trivially, we have $a_c(n)\ll \tau(n)$. From here, (\[a:uno\]) and (\[a:dos\]) follow immediately by noting that $H(c)\ll c^\ep\ll \log x$, which is an easy consequence of the definition of $h$. Also (\[g:uno\]),(\[g:dos\]),(\[g:tres\]) and (\[g:cuatro\]) are easy consequences of (\[firstg\]), (\[secondg\]) and the Prime Number Theorem in the arithmetic progression modulo $4$. Note also that these conditions were already verified in [@fi2] since, for cubefree integers, $g(d)$ is the same function as the one appearing in that reference. In order to get (\[super\]) we note that, by (\[newa\]) used for any given $d$ and for $d=1$, we have $$r_d(x)=O(g(d)d^{1/4}x^{9/16+\ep})+O(H(c)h(d)x^{1/2}).$$ Moreover $$\sum_{d\le x}^{\kern15pt 3}h(d)\ll \prod_{p\le x}(1+h(p))(1+h(p)^2)\ll (\log x)^6,$$ and $$\sum_{d\le x}^{\kern15pt 3}g(d)\ll \prod_{p\le x}(1+g(p))(1+g(p)^2)\ll (\log x)^2,$$ which gives $$\sum_{d\le D}^{\kern15pt 3}\left|r_d(t)\right|\ll D^{1/4}x^{9/16+\ep},$$ and, in particular, implies (\[super\]). For the remainder conditions, (\[divisor\]) and (\[bilinear\]), we want to use the analogous results obtained in [@fi2] for the sequence $a^{old}_n$. It is then mandatory to obtain the relation between the size of $\A_d(x;c)$ and $\A^{old}_d(x)$. Now, (3.18) of [@fi2] is, with our notation $$\label{olda}
\A^{old}(x)=4\kappa x^{3/4}+O(x^{1/2}).$$ Hence, in view of (\[newa\]), with $d=1$, to compare $\A^{old}(x)$ with $\A(x;c)$ we need to control $G(c)$, and $G(c)^{-1}$ for any $c\le (\log x)^B$. This is the content of the next lemma.
\[Gc\] Let $c$ be an integer. If $c$ is not representable as the sum of two squares, or $v_2(c)\equiv 3\tm 4$, then $G(c)=0$. Otherwise we have $$\frac1{c}\le G(c)\le \frac1{c^{3/4}}.$$
[**Remark:**]{} The lemma is intended to show bounds which are enough for our purpose and by no means need to be optimal.
From (\[firstg\]) and (\[secondg\]) we see that $$G(p^{\ap})\ge \frac 1{p^\ap},$$ which in particular implies the lower bound. The upper bound is a direct consequence of (\[olda\]) and (\[newa\]) since, by definition, $\A(x;c)\le\A(x;1)=\A^{old}(x)$.
We now deal with (\[divisor\]), for $(d,c)=1$. It is clear that in this case $\A_d(x;c)\le \A_{cd}^{old}(cx)$ and it is trivial to get $$\A_d^{old}(x)\ll \frac1d\tau(d)\A^{old}(x),$$ uniformly in $d\le x^{1/2-\ep}$. Hence, we have $$\A_d(x;c)\ll\frac1{cd}\tau(cd)\A^{old}(cx)\le\frac1{d}\tau(d)\tau(c)\A(x;c)\ll\frac1{d}\tau(d)\A(x;c)\log x,$$ where we have used the lower bound in Lemma \[Gc\], together with (\[newa\]) with $d=1$, (\[olda\]) and the bound $c\le(\log x)^B$. Hence, we are left with the bilinear condition (\[bilinear\]). We will get this bound from the similar one achieved in Proposition 4.1 of [@fi2].
\[errorbilinear\] Let $\eta>0$, $A>0$ and $B>0$. Then for any $c\le (\log x)^B$ we have $$\label{bilinear:concreta}
\sum_{m}|\sum_{{N<n\le 2N}\atop{{mn\le x}\atop{(n,m\Pi)=1}}}\mu(mn)\beta(n)\,\ac{c}_{mn}|\le \, \A(x;c)L^{4-A},$$ for every $N$ with $$x^{1/4+\eta}<N<x^{1/2}(\log x)^{-U},$$ where the coefficients $\beta(n)$ are given by (\[coeff:bilinear\]) for any $1\le C\le N^{1-\eta}$, $$(\log\log x)^2\le \log P\le (\log x)(\log\log x)^{-2}.$$ where $\Pi=\prod_{p<P}p$. Here $U$ and the implied constant in [(\[bilinear:concreta\])]{} need to be taken sufficiently large in terms of $\eta$ and $A$.
[*Proof of Proposition \[errorbilinear\]:*]{}
$$\begin{aligned}
&&\sum_{m}|\sum_{{N<n\le 2N}\atop{{mn\le x}\atop{(n,m\Pi)=1}}}\mu(mn)\beta(n)\,\ac{c}_{mn}|=
\sum_{(m,c)=1}|\sum_{{N<n\le 2N}\atop{{mn\le x}\atop{(n,m\Pi)=1}}}\mu(mn)\beta(n)\,a^{old}_{cmn}|\\
&\le&
\sum_{c|m}|\sum_{{N<n\le 2N}\atop{{mn\le cx}\atop{(n,m\Pi)=1}}}\mu(mn)\beta(n)\,a^{old}_{mn}|\le
\sum_{m}|\sum_{{N<n\le 2N}\atop{{mn\le cx}\atop{(n,m\Pi)=1}}}\mu(mn)\beta(n)\,a^{old}_{mn}|\\
&\le&
\A^{old}(x)L^{4-A},\end{aligned}$$
where we have used Proposition 4.1 of [@fi2], and $c\le (\log x)^B$ together with the lower bound on $P$. We just have to use (\[newa\]) with $d=1$, (\[olda\]), and the lower bound in Lemma \[Gc\] to get the result.
Theorem \[main\] is now a direct consequence of Proposition \[prop1\] and the upper bound in Lemma \[Gc\]. Recall that $H$ is given by (\[hache\]), which in this case is $$H=\prod_p(1-\chi_4(p)p^{-1})=L(1,\chi_4)^{-1}=\frac 4\pi,$$ and we are counting only positive integers $a,b$ in Theorem \[main\].
[999]{} A. Brumer, The rank of $J_0(N)$. Columbia University Number Theory Seminar (New York, 1992). Astérisque No. 228 (1995), 3, 41-68.
J. R. Chen, On the representation of a larger even integer as the sum of a prime and the product of at most two primes. Sci. Sinica 16 (1973), 157-176.
H. Darmon. Serre’s conjectures. In V. Kumar Murty, editor, Seminar on Fermat’s Last Theorem, number 17 in CMS Conference Proceedings, 135-153, 1995.
F. Diamond, R. Taylor, Nonoptimal levels of mod $l$ modular representations. Invent. Math. 115 (1994), no. 3, 435-462.
J. S. Ellenberg, Galois representations attached to $\mathbb Q$-curves and the generalized Fermat equation $A^4+B^2=C^p$. Amer. J. Math. 126 (2004), no. 4, 763-787.
J.B. Friedlander, H. Iwaniec, Asymptotic sieve for primes, Ann. of Math. 148 (1998), 1041-1065.
J.B. Friedlander, H. Iwaniec, The polynomial $X^2 +Y^4$ captures its primes. Ann. Math. 148 (1998), no. 3, 965-1040.
B. Mazur, Modular curves and the Eisenstein ideal. Inst. Hautes ' Etudes Sci. Publ. Math. No. 47 (1977), 33-186
A. P. Ogg, Hyperelliptic modular curves. Bull. Soc. Math. France 102 (1974), 449-462.
K. A. Ribet, Raising the levels of modular representations. Séminaire de Théorie des Nombres, Paris 1987-88, 259-271, Progr. Math., 81, Birkhauser Boston, Boston, MA, 1990.
K. A. Ribet, On modular representations of ${\rm Gal}(\overline{\mathbb Q}/{\mathbb Q})$ arising from modular forms. Invent. Math. 100 (1990), no. 2, 431-476.
H. Saito, On a decomposition of spaces of cusp forms and trace formula of Hecke operators. Nagoya Math. J. 80 (1980), 129-165.
A. Wiles, Modular elliptic curves and Fermat’s last theorem. Ann. of Math. 141 (1995), no. 3, 443-551.
Luis Víctor Dieulefait Departament d’Algebra i Geometria Universitat de Barcelona Gran Via de les Corts Catalanes 585 08007 Barcelona Spain email: [email protected]\
Jorge Jiménez Urroz Departament de Matemàtica Aplicada IV Universitat Politècnica de Catalunya (UPC) Edifici C3-Campus Nord Jordi Girona, 1-3. E-08034 Barcelona Spain email: [email protected]\
Kenneth Alan Ribet Department of Mathematics 3840 University of California Berkeley, CA 94720-3840 USA email: [email protected]
[^1]: first named author partially supported by grant MTM2009-07024 and by an ICREA Academia prize, second by DGICYT Grant MTM2009-11068
| {
"pile_set_name": "ArXiv"
} |
The DNA molecule is central to every living organism. It carries the genetic code of the organism, and continuously controls the synthesis of proteins, which are vital to its functioning and existence. Since the pioneering discoveries of Watson and Crick, a considerable effort was devoted to the study of the DNA, and much has been understood about its structure and activity [@Biochemistry]. Yet there are some key features which pose open questions. One such feature is the long range correlation and control between segments of the DNA. Synthesis of proteins is done via local ”reading” of specific sectors in the DNA. The initiation of the reading is done by another segment, which can be located far away along the DNA sequence (thousands of bases away). Consider this and other examples [@Genome_Wisdom], it seems as if the DNA molecule has the capability of transmitting information over long distances and in a specific manner (the information is transmitted to/from specific targets). Possible efficient candidates for such a transmission are solitons and solitary waves (rather than point particles/wave packets and linear waves). The existence and propagation of conformational solitons related to the DNA replication process have been studied in the past [@Replic_sol]. Non-topological charge solitons in proteins have also been studied [@Davidov]. Motivated by our studies of topological charge solitons in one dimensional (1D) arrays of mesoscopic tunnel junctions [@Normal_Solitons], [@Charge_Soliton], we have investigated the possibility of the existence of topological charge solitons in DNA.
We have developed a novel electromagnetical model of the DNA molecule, which is based on the properties of charge dynamics in the DNA (see Fig. \[DNA\_fig\]). We view each of the DNA strands as a 1D array. A unit (or ’grain’) of the array is a sugar and a base attached to it. The grains are connected longitudinally by Phosphate groups (P-bonds), and parallely, i.e., to the other strand, by the Hydrogen bonds (H-bonds) between bases. An additional electron residing on an atom belonging to the sugar-base grain can hop from atom to atom, thus obtaining a kinetic energy. The H- and P-bonds form barriers to the charge propagation. The proton in the H-bond can effectively screen a net charge density on either side of the bond by shifting its position towards this side. (By ’net’ charge we mean the change from the charge distribution of the unperturbed DNA.) As a result, electrons do not cross the H-bond. Hence the bond can be viewed as a capacitor. The P-bond barrier stems from the two oxygens which are transversely connected to the phosphorus. These oxygens share three electrons with the phosphorus, giving rise to two $\sigma$ bonds and one $\pi$ bond. As the $\pi$ electron can be shared with both oxygens, it behaves as an electron in a double well, and occupies the lowest level. When another electron approaches the well it encounters a barrier due to the energy difference to the next level of the well. However, since this barrier is narrow, the approaching electron can tunnel through the well. Thus from the charge dynamics point of view, the P-bond behaves as a tunnel junction.
We start with the model for a single strand (see Fig. \[DNA\_fig\]). For simplicity we assume that all the bases are of the same type. Each grain $i$ is composed of four sub-grains ($j=0,1,2,3$), assigned with phase variables, $\phi_{i,j}$. These phases are related to the electric potential of the sub-grains through $\phi_{i,j}(t)\equiv{1\over\hbar}\int_{-\infty}^t\,V_{i,j}(t')dt'$. The conjugate variable to each $\phi_{i,j}$ is the charge on the sub-grain, $Q_{i,j}$. The Lagrangian of this model is $$\begin{aligned}
\label{Full_Lagrangian_Single}
L&=&\sum_i \left[{C\over 2}(\dot\phi_{i+1,1}-\dot\phi_{i,3})^2+
{C_S+C_0\over 2}\dot\phi_{i,2}^2+
\right.
\nonumber \\
& & {1\over 2L/2}(\phi_{i,1}-\phi_{i,2})^2-
{1\over 2L/2}(\phi_{i,2}-\phi_{i,3})^2+
\nonumber \\
& & \left. E_J\cos(\phi_{i+1,1}-\phi_{i,3})-
{1\over L_0/2}\left(\phi_{i,0}-\phi_{i,2}\right)^2 \right]\ .\end{aligned}$$ It includes three types of energies: inductive energies which represent the hopping of electrons, capacitive energies which represent the capacitive properties of the H- and P-bonds, and a tunneling energy, which represents the tunneling process in the P-bond. This latter energy is proportional to the cosine of the phase difference across the bond, according to the tight binding picture [@Tight_Binding]. The values of the parameters can be obtained, in principle, from experiments, but for now we employ only a qualitative view. The $L_0/2$ inductance denotes the hopping from the open side of a base to its sugar, while the $L/2$ inductance denotes the hopping between P-bonds. Since the former involves more hopping around carbon rings, we assume that $L_0>L$. Both $C_S$ and $C_0$ denote in this single-strand model capacitances to the outside world. $C_S$ is the capacitance seen from the sugar, while $C_0$ is the capacitance seen from the open H-bond. We assume that both are of the same order of magnitude and much smaller than $C$, the capacitance of the P-bond tunnel junction. $E_J$ is the tunneling strength of the P-bond. The length scale in our model, $a$, is the distance between grains, which is $3.4\AA$. (In the model we use units in which $a=1$.)
From the kinetic part of (\[Full\_Lagrangian\_Single\]) we see that out of the four variables describing the $i$’th grain only three are independent. According to our experience in the study of 1D arrays of tunnel junctions [@Charge_Soliton], we introduce non-local (or integral) charge variables. These variables are very useful in trying to probe the collective, non-local dynamics of the DNA. We envision a propagation of charge from a certain point in the far left of the chain (’$-\infty$’) to a certain point in the far right (’$\infty$’), passing on its way the $i$’th grain. Referring to the $i$’th junction as the junction between the $i$’th and the $(i+1)$’th grains, we define $R_i$ as the charge that has [*passed*]{} through this junction: $$\begin{aligned}
\label{R_Variable_Single}
R_i & \equiv & \sum_{i+1,1}^{\infty}Q_{i',j} = \sum_{i+1}
(Q_{i',1}+Q_{i',2}+Q_{i',0}+Q_{i'})-Q_{\infty}
\nonumber \\
& & = Q_{i,1}+\sum_{i+1}(Q_{i',2}+Q_{i',0}) \ .\end{aligned}$$ $\sum_{i+1,1}^{\infty}Q_{i',j}$ means summation over the sub-grain charges, starting from $(i'=i+1,j=1)$ and going right, with the internal order $(i+1,1)
\rightarrow(i+1,2)\rightarrow(i+1,0)\rightarrow(i+1,3)\rightarrow(i+2,1)$, and so on. Next we define $q_i$ as the charge that has [*reached*]{} junction $i$: $$\label{q_Variable_Single}
q_i \equiv \sum_{i,3}^{\infty}Q_{i',j} = Q_{i,3}+R_i=
\sum_{i+1}(Q_{i',2}+Q_{i',0}) \ .$$ After the canonical transformation to the variables: $\pi_i\equiv\phi_{i+1,2}-\phi_{i,2}-\theta_i$, $\theta_i\equiv
\phi_{i+1,1}-\phi_{i,3}$ and $\tilde\phi_{i,0}\equiv\phi_{i,0}-\phi_{i,2}$, we obtain the following Euclidean Lagrangian (in the continuum limit which we justify later): $$\begin{aligned}
\label{Euclidean_Lagrangian_Single}
L_E&=&\int\,dx\left\{{1\over 2}L\dot q^2(x)-E_J\cos\theta(x)+
{1\over 2}{L_0\over 2}\dot q_0^2(x)+
\right. \nonumber \\
& & {1\over 2C}\left[q(x)-R(x)\right]^2+
{1\over 2C_S}q_x^2(x)+
\nonumber \\
& & \left. {1\over C_S}q_0(x)q_x(x)+{1\over 2C_{S0}}q_0^2(x)+
i\dot\theta(x)R(x) \right\} \ ,\end{aligned}$$ where $C_{S0}\equiv C_SC_0/(C_0+C_S)$ is the effective capacitance across the chain (parallel coupling of $C_S$ and $C_0$). As $C_S$ is much smaller than $C_0$, $C_{S0}\approx C_S$. We define a tunneling inductance $L_J\equiv \Phi_0^2/(2\pi E_J)^2$ ($\Phi_0\equiv h/e$ is the flux quantum), which we take to be much smaller than $L$. From (\[Euclidean\_Lagrangian\_Single\]) one can identify three characteristic frequencies in the system: $\Omega^2\equiv 1/(LC)$ of the $q$ mode, $\omega_J^2\equiv 1/(L_JC)$ of the $R$ mode, and $\omega^2\equiv 2/(L_0C_{S0})$ of the $q_0$ mode. In the limits we are working, we have $\Omega^2\ll\omega_J^2, \omega^2$. Thus we can integrate out the $R$ and $q_0$ modes, and obtain an effective theory for the $q$ mode. The effective real time Lagrangian is $$\begin{aligned}
\label{Full_eff_Lagrangian_Single}
L^{eff}&=&\int\,dx\,\left\{{1\over 2}L\dot q^2-{2\over(2\pi)^2}E_C
\left[1-\cos\left({2\pi\over e}q\right)\right]-
\right. \nonumber \\
& & \left. {1\over 2(C_0+C_S)}q_x^2+
{1\over 2}{C_0^2L_0\over 2(C_0+C_S)^2}\dot q_x^2\right\}\ ,\end{aligned}$$ where $E_C\equiv e^2/(2C)$. The equation of motion of (\[Full\_eff\_Lagrangian\_Single\]) is the following modified sine-Gordon equation: $$\begin{aligned}
\label{Single_Eq}
L\ddot q+V_D\sin\left({2\pi\over e}q\right)-{1\over C_0+C_S}q_{xx}-
{C_0^2L_0\over 2(C_0+C_S)^2}\ddot q_{xx}
\nonumber \\ =0 \ ,\end{aligned}$$ where $V_D\equiv e/(2\pi C)$. This is a Kirchoff law for the equivalent electrical circuit. Since the characteristic length scale in a sine-Gordon model is $C/(C_0+C_S)$, the continuum limit is justified when $C>C_0+C_S$, which indeed corresponds to our assumption. Transforming into dimensionless space, time and charge variables: $x\rightarrow x'\equiv\sqrt{C/C_0}x, \ \ t\rightarrow t'\equiv\sqrt{LC}t, \ \
q\rightarrow q'\equiv {e\over 2\pi}q$, we obtain: $$\label{Scaled_Single_Eq}
\ddot q+\sin q-{1\over 1+\mu_S}q_{xx}-{\zeta\over 2(1+\mu_S)^2}
\ddot q_{xx}=0\ ,$$ where $\zeta\equiv{L_0C_0\over LC}$, and $\mu_S\equiv C_S/C_0$.
Next we introduce the double-strand model (see again Fig. \[DNA\_fig\]). We distinguish between the two strands by the superscripts $\alpha, \beta$. The capacitance $C_0$ describes now the capacitive coupling between the two strands, i.e., over the H-bonds. We assume that it is much larger than the capacitance to the outside world, $C_S$, but still smaller than $C$. We take the parameters of the two strands to be equal. Following the same steps of derivation as in the single-strand model (details will be given elsewhere [@Next_paper]), we obtain two (dimensionless) coupled modified sine-Gordon equations of motion: $$\label{AVE2_Eq}
\ddot{\bar q}+\sin\bar q\cos q-{1\over 2\mu}\bar q_{xx}=0 \ ,$$ $$\begin{aligned}
\label{REL2_Eq}
\ddot q+\sin q\cos\bar q-{1\over 2(1+\mu)}q_{xx}-
{\zeta\over 2(1+\mu)^2}\ddot q_{xx}=0 \ ,\end{aligned}$$ where $\mu \equiv C_S/2C_0$, and we have used the average and the relative charge variables: $$\label{New_Charges}
\bar q\equiv{1\over 2}\left(q^\alpha+q^\beta\right) \hskip 2cm
q\equiv{1\over 2}\left(q^\alpha-q^\beta\right) \ ,$$ Similar equations, without the $\zeta$ term, where studied in the past in connection with the stacked Josephson junctions model [@Coupled_Josephson1].
We turn now to study the modified sine-Gordon equations we have derived, starting from the single-strand model. The pure Sine-Gordon equation has exact topological solitons solutions. We have checked numerically and found that the extra term in equation (\[Scaled\_Single\_Eq\]) does not effect the stability of the topological soliton solution, though it does induce interaction with the plasmons. This interaction causes the soliton to slowly radiate away its kinetic energy. Using the collective coordinate: $$\label{Collective_coord_eq}
X = -{1 \over 2\pi}\int q_x x\,dx,$$ we find that the soliton velocity actually oscillates with relatively high frequency, but its average is almost constant, decreasing only slightly over large period of time. Therefore, though the soliton’s motion is not persistent in the exact sense, charge can propagate relatively large distances along the strand without the need to apply a driving force in the form of a potential difference.
The extra term in equation (\[Scaled\_Single\_Eq\]) destroys the Lorenz invariance of this equation. The dispersion relation for small amplitude linear waves is $$\omega^2 = {k^2/(1+\mu_S) + 1\over {\zeta\over 2}k^2/(1+\mu_S) + 1}\ ,$$ and the group velocity tend to zero for both small and large k values. When the soliton velocity exceed the maximal group velocity, it leaves in its wake all the small amplitude waves.
In the double-strand model, the dimensionless parameter $\mu$ serves as a measure of the strength of the interaction between the strands (it is small for strong interaction). Since $C_S \ll C_0$, $\mu$ is very small. In the extreme $\mu = 0$ limit, $\bar q_x$ should be zero as well, in order that the Lagrangian for this model would be finite. As both $q^\alpha$ and $q^\beta$ have integer values at the two edges of the chain, $\bar q\approx 2\pi n$ everywhere. We are thus left with a single equation for q: $$\label{Double_q_Eq}
\ddot q+\sin q-{1\over 2}q_{xx}-{\zeta\over 2}\ddot q_{xx}=0 \ .$$ This equation is equivalent to the one of the single-strand model, but now $q$ represents a soliton anti-soliton pair (with zero total charge) rather than a single charged soliton.
Equations (\[AVE2\_Eq\],\[REL2\_Eq\]) have two simple topological solutions. One is the “symmetrical” solution, namely, $q=0$ and $\bar q$ is a usual Sine-Gordon kink. This solution is unstable ,at least at low energy, since it corresponds to two charges with equal sign which tend to separate. It is, however, possible that it becomes stable at high energy, as was demonstrated for the stacked Josephson junctions model [@Coupled_Josephson2]. The other “anti-symmetrical” solution has $\bar q=0$ and q in the form of a kink. It is a stable solution which correspond to an electron-hole pair (each on a different strand). It is plausible that this excitation is used in the DNA to transmit information over long distances. It can be created at a specific segment, which is responsible for the initiation of a certain protein synthesis, by tunneling of an electron through the H-bond (as a result of external agent). The pair can, then, propagate almost freely along the DNA, and be annihilated at the protein synthesis segment by a reverse tunneling, thus transmitting the order for this synthesis. The targeting of a specific annihilation segment is probably done through the specific base sequence of both segments, which was neglected in our model.
There is also a way to approximate a solution which corresponds to an electron (hole) moving with velocity $v$ in the, say, $\alpha$ strand: $$\begin{aligned}
\label{aprox_sol_eq}
q & = & 2 \arctan\,\exp
\left[ -\gamma(x - vt)/c \right] \nonumber \\
\bar q & = & 2 \arctan\,\exp
\left[ -\bar\gamma (x - vt)/\bar c \right],\end{aligned}$$ where $\gamma$ is the relativistic factor, and $c=1/\sqrt{2(\mu + 1)}$, $\bar c=1/\sqrt{2\mu}$, are the linear wave velocities in equations (\[AVE2\_Eq\],\[REL2\_Eq\]) (when the last term in equation (\[REL2\_Eq\]) is neglected). They correspond to the different linear waves velocity in the two equations. The exact solution can be obtained numerically and is shown in Fig. \[soliton\_fig\].
An important way to test our model is via measurements of the current-voltage characteristics. An electron can be injected to one of the strands to form a topological charge soliton, and may be subjected to an applied external voltage $V$ and an Ohmic dissipation $R$ (both per unit length). We implement this in our model by adding the terms $F + \alpha \dot{\bar q}$ to equation (\[AVE2\_Eq\]) and $\alpha \dot{\bar q}$ to equation (\[REL2\_Eq\]), where $F \equiv {V/ V_D}$ and $\alpha \equiv {RC/\sqrt{LC}}$, are the external force and the dissipation rate respectively. For given values of force and dissipation rate a soliton reaches a limiting velocity which corresponds to the measured current. The I-V curves which we obtain for different choices of the $\mu$ and $\zeta$ parameters are shown in Fig. \[IV\_fig\]. It should be noted that above some threshold voltage close to the maximum voltage indicated in each curve of Fig. \[IV\_fig\], the system becomes unstable, as more and more soliton anti-soliton pairs are created.
We proceed to estimate the limiting velocity. To do so, we equate the energy gain by the external force and the dissipative energy [@Power_balance]. We find that $\dot{X} (\gamma/c + \bar\gamma/\bar c) = {\pi F/2\alpha}$. This reduces in the small $\mu$ limit to $$\label{Theoretic_IV_eq}
\dot{X}={1\over\sqrt{2\left[\left({2\alpha\over\pi F}\right)^2 + 1\right]}}.$$ This theoretical curve is also shown in Fig. \[IV\_fig\]. The ascending of the numerical curve in the “relativistic” regime above the theoretical curve is obviously due to the $\zeta$ term.
We turn now to discuss some possible experimental tests of our predictions. The most straightforward test is to measure the I-V characteristics of a DNA that was mentioned earlier. Such experiments are now being developed [@Experiment]. We propose to use first a single strand vs a double strand. It is easier to inject solitons into the former. In the case of a double strand, it would be useful if each end is made of a short single strand (a different one on each end). In both cases it would be easier to compare the experimental results with the theoretical predictions if artificial strands composed of repetitia of one base are used. Another approach would be to study the magnetic response of circular strands. The idea is to place many circular strands on a surface, apply a time-dependent magnetic flux (say, of a saw tooth form), and measure the response of the system. For short circular strands at low temperature we expect to observe persistent current carried by the charge solitons. Finally, a less trivial experiment, is to inject a charge soliton at the end of the DNA via Scanning Tunneling Microscope (STM), and look for a response at the other end (for example, via an attached molecule that has a fluorescent response to an incoming charge).
These are just three examples of many more possible tests of our predictions about the existence of topological charge solitons in DNA molecules. If turned to be verified, we expect charge solitons to have crucial role in the DNA activities as means of transfer of information and energy over long distances, and to specific locations. To study this role, one can use our model as a starting point, incorporating into it the inhomogeneity of the chain due to the different bases, as well as interactions with external molecules at specific sites.
We are most thankful to E. Braun for sharing with us his ideas about experimental measurements of DNA electrical transport. This research is supported in part by a GIF grant.
[10]{}
L. Stryer, [*Biochemistry*]{}, W. H. Freeman and Company, New York (1995).
F. Joset, J. Guespin-Michel, [*Prokaryotic Genetics: Genome organization, transfer and plasticity*]{}, Blackwell Scientific Publications (1993).
J. J.-L. Ting, M. Peyrard, Phys. Rev. E [**53**]{}, 1011 (1996).
A. S. Davydov, [*Biology & Quantum Mechanics*]{}, Pergamon Press (1982).
E. Ben-Jacob, K. Mullen and M. Amman, Phys. Lett. A [**135**]{}, 390 (1989); M. Amman, E. Ben-Jacob and Z. Hermon, in [*Single Electron Tunneling and Mesoscopic Devices*]{}, edited by H. Koch and H. Lübbig (Springer-Verlag, Berlin, 1992).
Z. Hermon, E. Ben-Jacob and G. Schön, Phys. Rev. B [**54**]{}, 1234 (1996).
N. W. Ashcroft, N. D. Mermin, [*Solid State Physics*]{}, Saunders College Publishing (1976).
Z. Hermon, S. Caspi and E. Ben-Jacob (to be published).
M. B. Mineev, G. S. Mkrtchyan and V. V. Schmidt, J. Low Temp. Phys. [**45**]{}, 497 (1981); A. Petraglia, A. V. Ustinov, N. F. Pedersen and A. Sakai, J. Appl. Phys [**77**]{}, 1171 (1995).
N. Gronbech-Jensen, D. Cai and M. R Samuelsen, Phys. Rev. B [**48**]{}, 16160 (1993).
D. W. McLaughlin, A. C. Scott, Phys. Rev. A [**18**]{}, 1652 (1978).
E. Braun and U. Sivan, private Communication. They have communicated with us their approach to measure the DNA electrical transport prior to our theoretical studies.
| {
"pile_set_name": "ArXiv"
} |
=cmbx10 at 15pt =cmbx10 at 12pt
[Centre de Physique Théorique[^1], CNRS Luminy, Case 907]{}
[F-13288 Marseille – Cedex 9]{}
[Light-cone wave functions of mesons]{}
[**Gautier STOLL**]{}
[**Abstract**]{}
A review of light-cone, covariant and gauge-invariant wave-functions of mesons is presented. They are basic non-perturbative objects needed for hard exclusive processes and for the method of light-cone QCD sum rules. The emphasis is on the vector mesons and a new (model-independent) way of computing the mass corrections of vector-meson wave functions is given.
Key-Words: Non-perturbative QCD, Conformal expansion.
December 1998
CPT-98/P.3735
anonymous ftp : ftp.cpt.univ-mrs.fr
web : www.cpt.univ-mrs.fr
Introduction
============
One of the main goal in QCD is to understand the non-perturbative aspects of this theory. In this report, I describe a way of parameterizing these non-perturbative effects in a covariant and gauge-invariant way : the “meson wave functions” or “light-cone distributions amplitudes of meson”, with the emphasis on vector meson (the pseudoscalar case, like the $\pi$-meson, is simpler and the results can be found in [@BF]). By definition, a “wave function” is the matrix element of (gluon and) quark operators on the light-cone between the vacuum and a light meson. It is conceptually different to the OPE applied to QCD (like in the method of QCD sum rules [@SVZ]) where the non-perturbative inputs are vacuum expectation values of [*local*]{} operators.
The standard approach to wave function (due to Brodsky and Lepage[@BrLe; @BrLe2]) considers the parton decomposition in the infinite momentum frame. A mathematically equivalent formalism is the light-cone quantization[@BrPa]. Here, the concept of wave function is defined in a different way. Although it is less intuitive, it keeps the Lorentz and gauge invariance, in order to use the equation of motion of the QCD.With this approach, one gets directly the non-perturbative objects needed for the method of light-cone QCD sum rules[@B], which allows to compute exclusive decays of heavy hadron beyond perturbative QCD.
To compute a whole non-perturbative function seems an impossible task. Therefore the “conformal expansion” is used (it is like a partial-wave expansion, but referred to the conformal invariance of QCD), and the unknown input come from local matrix elements, which can be computed with the help of QCD sum rules[@SVZ]. Within this expansion the renormalization-scale dependence is also under control.
Like in DIS, the different wave functions are classified by their increasing twist. In order to be more precise when using these wave functions, the non-leading twist ones are needed. A part of these non-leading twist contributions comes from the mass correction, which can be important if $K^*$ is considered for example.
In this work, after two sections of preliminaries and definitions, I present the conformal expansion of these wave functions : I introduce the conformal group, the notion of conformal operators on the light-cone, the link between conformal invariance and renormalization, the matrix elements of conformal operators. Then I give an example of computation of non-leading twist three-point wave function. Finally, I give methods of computing mass corrections, with some new formulas which gives the mass corrections directly from the leading twist wave function (without making a conformal expansion). I add some appendices which contain lists of wave functions and technical results.
General framework
=================
The two basic objects are (for the $\rho$-meson) :
1\) The two-point wave functions, extracted from this kind of matrix element : 0 |(x)\[x,-x\] d(-x)|\[2matel\]
2\) The three-point wave functions, extracted from this kind of matrix element : 0 | (x)\[x,vx\] g G\_(vx)\[vx,-x\] d(-x)|\[3matel\] Where $x$ is almost on the light cone, $v \in [0,1]$, $\Gamma$ any kind of product of $\gamma_{\mu}$ matrices and $[x,y]$ a path-ordered gauge factor along the straight line connecting $x$ and $y$ : =[P]{}
This gauge factor is a way to introduce interaction (see [@Bali]), it ensure the gauge invariance of these non-local matrix elements (I will omit to write it sometimes).
These matrix elements depend on three vectors :
- $P_{\mu}$ : momentum of the $\rho$-meson
- $\epl_{\mu}$ : polarization vector of the $\rho$-meson
- $x_{\mu}$
With the relations : P\^2&=&m\^2\_\
&=&-1\
P&=&0
The parameterization of these different matrix elements is based on the Operator Product Expansion on the light-cone[@BP]. So one needs light-like vectors $p$ and $z$, built from $P$, $\epl$ and $x$ :
p\^2&=&0\
z\^2&=&0 such that $p\rightarrow P$ in the limit $m^2_{\rho} \rightarrow 0$ and $z\rightarrow x$ for $x^2\rightarrow 0$ (see [@BBKT]) : z\_&=& x\_- P\_\[defz\]\
p\_&=& P\_- z\_. \[defp\]
Useful scalar products are zP=zp&=&\
p&=&-z.
The polarization vector $\epl$ can be decomposed in projections onto the two light-like vectors and the orthogonal plane : \_=(p\_-z\_)+\_. \[defeperp\]
Note that (z)=(x).
For a practical use, one defines the projector onto the direction orthogonal to $p,z$ : g\^\_=g\_-(p\_z\_+ p\_z\_), \[gperp\] and these notations : a.&& a\_z\^\
a\_\* & & a\_p\^/ (pz) \[dotstar\]
Definitions of twist \[deftw\]
==============================
The different contributions to the matrix elements \[2matel\], \[3matel\] are classified by their “twist”, but there are two different definitions of this concept :
For local operators, the “twist” means dimension minus Lorentz-spin. To extract the contribution of a definite twist for non-local matrix elements like \[2matel\], \[3matel\], an expansion in local operator (OPE) has to be done and all local operators which have a definite twist in the sense given above must be re-summed. This will give the non-local term which has a definite power in $x^2$. To understand how this definition of “twist” means “power in $x^2$”, see [@BP]. From now on, this definition will be called “theoretical twist”.
The second definition is built in analogy to partons distributions. In the latter case, the different structures of non-local matrix elements are separated with their different powers of $Q$ (the hard momentum transfer) in deep inelastic scattering : a term of twist $t$ contributes to the inclusive cross section with coefficients which contain $t-2$ or more power of $1/Q$. A good description of this classification can be found in [@JJ]. For the matrix elements \[2matel\], \[3matel\], the same type of definition can be used, like in [@BBKT] : structures with $p\cdot z$, with $p_\mu$ or with $\epl\cdot z$ gives one power of $Q$, but $\epl_\perp$ behaves as $Q^0$. From now on, this definition will be called “physical twist”.
As an example, one gives the parameterization of a two-point matrix element :
\
&=& f\_m\_where =2u-1, $\mu$ is the renormalization scale and $f_\rho$ is defined by the following local matrix element : 0| (0)\_d(0)|\^-(P,)= f\_m\_\_.
All the three functions $\phi_\|,g_\perp^{(v)},g_3$ are normalized like this : \^1\_0 du \_ (u)=1
Following the classification of [@BBKT], $\phi_\|$ is a (“physical”-)twist 2 contribution, $g_\perp^{(v)}$ is (“physical”-)twist 3 and $g_3$ is (“physical”-)twist 4. But when one looks at the definition given above, one would argue that these contributions should be twist 1, 2, 3. In fact, the authors of [@BBKT] used another definition of twist, which just shifts it to one unity, based on the light-cone quantization formalism. The parameterization of all matrix elements \[2matel\] and \[2matel\] up to twist four can be found in the first appendix.
A physical meaning for the different functions $\phi_\|, g_\perp^{(v)}, g_3$ can be given : they describe the probability amplitude to find the $\rho$ in a state with a quark and an antiquark which carry momentum fractions $u$ for the quark and $1-u$ for the antiquark respectively and a small transverse separation of order $1/\mu$.
Roughly speaking, one uses the “physical twist” to define the different wave functions. But in order to compute them, an expansion in “theoretical twist” is made.
Conformal expansion
===================
When the different wave functions which can be extracted from non-local matrix elements like \[2matel\] and \[3matel\] has been classified, one needs to compute them. The first idea is to expand them in the different moments, like in [@CZ]. But it is like doing an expansion in local operators (OPE). This must be avoided if these wave functions are needed for a light-cone QCD sum rule (see [@B]).
The method used to compute wave functions is the “conformal expansion” : the idea is to expand the wave function in a series of polynomials, whose coefficients are renormalized (almost) multiplicatively. Here is an example for $\phi_\|$ (with $\xi=2u-1$) : \_ (u)=6u\^\_[n=0]{}a\^\_nC\_n\^[3/2]{} (), =2u-1 \[confex\] where $C_n^{3/2}$ is the Gegenbauer polynomial (see appendix \[app:a\]) and $a_n^\|$ are coefficients which are renormalized multiplicatively (to one loop). These $a_n^\|$ can be computed using the usual QCD sum rules and exact relations (see section \[PsiPhi\]).
In this section, the method of building such an expansion will be described. For it, one needs to know what is the structure of the conformal group, in what sense is QCD conformal invariant, how to build local and polylocal conformal operators and finally what happens for the matrix element of polylocal conformal operators.
Structure of the conformal group
--------------------------------
The conformal group is a Lie group which contains the Poincaré group and the dilatation. To describe its structure, it is simpler to work directly with the algebra. The generators of the conformal group are the generators of the Poincaré algebra $P_\mu$ and $M_{\mu\nu}$, the dilatation $D$ and of the special conformal transformations $K_\mu$.
They satisfy the following commutation relations[@BF; @Oh; @MS] : & = & iK\_\
\[ K\_,P\_\] & = & -2i (g\_D+M\_)\
\[D,M\_\]&=&0\
\[D,P\_\]&=&-iP\_\
\[K\_,M\_\]&=&i(g\_K\_-g\_K\_)\
\[K\_,K\_\]&=&0 plus those of the Poincaré algebra. The action of these generators on a field $\Phi(x)$ with an arbitrary spin and canonical dimension $l$ is the following : &=&-i\_(x) \[conf1\]\
\[M\_,(x)\]&=&\[i(x\_\_-x\_\_)-\_\] (x)\
\[D,(x)&=&-i(x\^\_+l)(x)\
\[K\_,(x)\]&=&-i(2x\_x\^\_-x\^2\_+2x\_l-2ix\_\_)(x) \[conf4\] where $\Sigma_{\mu\nu}$ is the generator of spin. It acts on a fermion field $\psi$ and a gauge field $G_{\alpha\beta}$ in the following way : \_&=&\_\[fspin\]\
\_G\_&=&i(g\_G\_-g\_ G\_)-() \[gspin\]
Conformal invariance and QCD \[sec:confQCD\]
--------------------------------------------
Anyone can convince himself that the unrenormalized QCD-action (with massless quark) is conformal invariant. But this symmetry is broken by two effects :
1\) Renormalization
2\) Quark-mass terms
But a careful analysis of these effects shows that the conformal expansion of wave functions like in \[confex\] is still relevant.
### Renormalization effect on the conformal invariance
Suppose one has a conformal operator in non-interacting and massless QCD : it is an operator which does not mix with other ones under conformal transformation (such local and polylocal operators are constructed in the next subsection). Following [@Ma], one can prove that such operators do not mix under renormalization at the leading log approximation (if simple derivatives are transformed into covariant ones). It means that conformal symmetry (and conformal expansion for wave function) is still a “good” symmetry of massless QCD.
Consider an operator which renormalizes multiplicatively. Its matrix $\Gamma$ satisfies the Callan-Symanzik equation : (p,;g)=\_(g)(p,;g). \[CaSy\] where $p\equiv (p^1 _\mu, \ldots ,p^N_\mu)$ is a set of momenta and $\gamma_\Gamma$ is the anomalous dimension of the operator considered. If this matrix element comes from a massless theory, it is an homogeneous function of $p$ and $\mu$. The Euler’s theorem yields : (p,;g)=d\_(p,;g), where $d_\Gamma$ is the canonical dimension of the operator considered. So equation \[CaSy\] can be transformed in (p,;g)=(p,;g).
If one considers the theory at one loop level, $\beta(g)$ can be neglected because it starts with $g^3$; one gets : (p,;g)=(p,;g).
This last equation shows that $\Gamma$ satisfies the dilatational Ward identity, so if the interaction is switched off, one gets a matrix element which transforms in itself under a dilatation : it is the matrix element of a conformal operator.
So if an operator is renormalized multiplicatively at one loop order, it is conformal if the interaction is switched off. Is the converse true ?
Suppose one has a set of operators. A basis which diagonalizes the anomalous dimension matrix at one loop order is built. From the preceding arguments, one knows that this basis is made of conformal operators. If one directly builds a basis of conformal operators (like in the next section), almost the same basis of multiplicatively renormalized operator is obtained, except if the eigenvalues of the anomalous matrix (or in the representation of the conformal group) are degenerate.
So conformal operators renormalize almost multiplicatively at one loop order, they can only mix with the ones which have the same conformal representation. The coefficient of the dilatational Ward identity transforms like this d\_d\_-g\^2\_2 when the interaction is switched on at one loop order. $g^2\gamma_2$ is the value of the anomalous dimension $\gamma_\Gamma(g)$ at the order $g^2$ : \_(g)=g\^2\_2+g\^4\_4+…
This property of conformal operators works also in the leading logarithmic approximation, otherwise it would contradict this property at one loop order (see [@Ma]).
In [@Oh], the anomalous dimension of some conformal operators is computed and for these operators, it is checked that they renormalized multiplicatively to one loop.
Some complications arise when one deals with gauge theories. In order to have gauge-invariant operators, usual derivatives have to be replaced by covariant ones. But in that case, it is not sure that the diagonalization of the anomalous dimensions matrix is really possible. But one hopes that for color-singlet operators, it works (see [@Oh]). The other problem is that an operator which transforms in itself under a dilatation may not be a conformal one (the action of special conformal transformations $K_\mu$ may introduce some gauge-variant terms). But it seems very hard to construct such “almost”-conformal operators, so it is assumed that they do not exist (see [@Sa]).
The coefficients of the conformal expansion of a wave function (like the $a_n^\|$ in equation \[confex\]) are matrix elements of local conformal operators (see sections \[sec:locconf\] and \[PsiPhi\]). So they are renormalized multiplicatively, that is why conformal expansion like in \[confex\] make sense.
### Quark-mass effect on the conformal invariance
Masses of light-quarks break explicitly the conformal invariance of the QCD-Lagrangian. So they can break the multiplicative renormalizability of conformal operators. But if the renormalization scale is high enough compare to the light quark masses, this effect is small and the conformal expansion of wave functions is still a “good” expansion.
Quark masses can also appear explicitly in the twist-expansion of matrix elements of non-local operators. It that case, it introduce some new wave functions, but the latter can also be expanded in a conformal way.
So the effect of quark masses can be controlled and the conformal expansion is still a good way to compute the wave functions.
Construction of local and polylocal conformal operators
-------------------------------------------------------
### Local operators on the light-cone \[sec:locconf\]
For the use of QCD light-cone sum rules, one deals with fields $\Phi(z)$ varying on the light cone ($z^2=0$) or almost on the light-cone. But the representation of the conformal group is much simpler on the light cone, that is why a light-like basis $z_\mu$, $p_\mu$ was introduced at the beginning of this chapter (equations \[defz\] and \[defp\]).
Following [@BF], one can see that only the components $P.$, $D$, $M_*.$ and $K_*$ of the algebra of the conformal group only act non-trivially on the field $\Phi(z)$ (the meaning of $.$ and ${}_*$ is given at equation \[dotstar\]). These generators form the algebra of a subgroup called collinear conformal subgroup $SO(2,1)\cong SU(1,1)\cong SL_2(R)$ which generates projective transformations on the line (see e. g. [@Oh]).
This algebra can be brought into a more standard form, with the following linear combinations :
J\_+=J\_1+iJ\_2&=&P.\
J\_-=J\_1-iJ\_2&=&K\_\*\
J\_3&=&(D+M\_\*.)\
E&=&(D-M\_\*.)\
J\^2&=&J\_3\^2-J\_1\^2-J\_2\^2=J\_3\^2-J\_3-J\_+J\_-.
Then the commutation relations are &=&-2J\_3\
\[J\_3,J\_\]&=&J\_\
\[E,J\_i\]&=&0\
\[J\^2,J\_i\]&=&0
In order to build a representation of this subgroup, all the generators $P.$, $D$, $M_*.$ and $K_*$ must reduce to differential operators. Equations \[conf1\] to \[conf4\] imply that fields must be eigenvectors of the operator $\Sigma_*.$, that is fields having fixed projections ($s$) of the Lorentz spin on to the line $z_\mu$ : \_\*.(z)=is(z)
A spinor field $\psi$ has two components with the projection $s=\pm \frac{1}{2}$ : $\gamma .\psi$ and $\gamma_*\psi$. One can see it using equation \[fspin\] :
\_\*..=\_\*..=-. \[confop1\]\
\_\*.\_\*=\_\*.\_\*=+\_\*
For the gauge field $G_{\mu\nu}$ there are three possible value of the projection : $s=-1,0,1$. The different components are (see equation \[gspin\]) : \_\*.G.\_&=&+iG.\_\[confop3\]\
\_\*.G\_[\*]{}&=&-iG\_[\*]{}\
\_\*.G\_\*.=\_\*.G\_&=& 0 \[confop5\] where $\perp$ is a component in the plane orthogonal to the plane of the light-like vectors $z_\mu$ and $p_\mu$ (the projection operator associated to this plane is $g^\perp_{\mu\nu}$, defined by the equation \[gperp\]).
If one has a field $\Phi(uz)$ ($u$ is a real number) which has a fixed projection $s$ of the Lorentz spin on to the line $z_\mu$, the generators of the collinear conformal subgroup act in the following way : &=&(uz) \[diffconf1\]\
\[J\_3,(uz)\]&=&(l+s+2u)(uz)\
\[J\_-,(uz)\]&=&(u(l+s)+u\^2)(uz)\
\[E,(uz)\]&=&(l-s)(uz). \[diffconf4\]
From this relation the irreducible representations of the collinear conformal subgroup $SO(2,1)$ which contains the non-trivial conformal transformation on operators on the light-cone can be constructed. These representations are classified by the eigenvalues of the Casimir operator $J^2$. Equations \[confop1\] to \[confop5\] describe the construction of operators $\Phi(uz)$ which are eigenvectors of $J^2$ :
=j(j-1)(uz) where j=(l+s) \[confspin\] $j$ is called “conformal spin”. For $\Phi(uz)$, the algebra is reduced to differential operators, so one can work with functions of one real variable instead of quantum fields. Having in mind the application to the computation of wave functions, one can make the Fourier Transform of the quantum field $\Phi$ :
(uz)=de\^[-iupz]{}().
In this formalism, the irreducible representations of the collinear conformal subgroup can be built, classified by the eigenvalues of $J^2$ and $-J_3$. They are the set of the following functions : |j,n=()\^[j+n-1]{} \[irrep\]
Equations \[diffconf1\] to \[diffconf4\] give the action of the generators of the collinear conformal subgroup on the states $|j,n\rangle$ :
J\_+|j,n&=&(j+n)|j,n+1\
J\_-|j,n&=&(n-j)|j,n-1\
J\_3|j,n&=&-n|j,n.
One can see that $J_+$ transforms a state one step upper and $J_-$ one step lower. The lowest value of $n$ is $j$.
To build a local conformal operator on the light-cone, one has to have fixed projection of the Lorentz spin on to the light-cone (equations \[confop1\] to \[confop5\]). If the operator is not taken at $z=0$, one has to expand it in the states \[irrep\]. But in order to compute wave-functions, the $z$-dependence of polylocal operators must be controlled, like those in \[2matel\] and \[3matel\].
### Polylocal operators on the light-cone
Suppose that $\Phi_1(u_1z),\ldots,\Phi_n(u_nz)$ are $n$ local operators which have a fixed conformal spin. What happens for the product $\Phi_1(u_1z)\ldots\Phi_n(u_nz)$ ? When one thinks of irreducible representation of the collinear conformal subgroup, the corresponding “Clebsch-Gordon coefficients” of a tensor product of different irreducible representations are needed : |j,n=\_[n\_1+…+n\_k=n]{}C\^[j,n]{}\_[j\_1,n\_1,…,j\_k,n\_k]{} |j\_1,n\_1…|j\_k,n\_k\[CGcoeff\]
This seems a very complicated task, but one has to recall what is the conformal collinear subgroup : it is the Lorentz group in $2+1$ dimensions ($SO(2,1)$). So the irreducible representations $|j,n\rangle$ can be interpreted as relativistic particles in an abstract or internal space of $2+1$ dimensions. With this point of view, the state $|j,n\rangle$ is a particle with a mass $j$ and an energy $n$ (that is why $n\ge j$).
Consider a system of $k$ particles. The lowest invariant mass of this system is the sum of the masses of all particles, $j=j_1+\ldots+j_k$. If the state of the lowest invariant mass has also the lowest energy, all the particles must be a their lowest energy level, $n_i=j_i$. This state is non-degenerate. In the language of irreducible representations of $SO(2,1)$, it gives : |j\_=\_[i=1]{}\^[k]{}j\_i,n\_=j\_& \~& |j\_1,j\_1…|j\_k,j\_k\
&\~&\_1\^[2j\_1-1]{}…\_k\^[2j\_k-1]{}
If $n>j$, the operator $J_+=\sum_{i=1}^{k}J_+^{(i)}$ can be used to “raise” the “energy” of the state. Since $J_+\phi(\alpha)=\frac{1}{\sqrt{2i}}\alpha\phi(\alpha)$, one gets : |j\_=\_[i=1]{}\^[k]{}j\_i,n&\~& (\_1+…+\_k)\^[n-j]{}|j\_1,j\_1…|j\_k,j\_k\
&\~&(\_1+…+\_k)\^[n-j]{}\_1\^[2j\_1-1]{}…\_k\^[2j\_k-1]{} \[jminCG\]
To have the whole sum for the state $|j,n\rangle$ (equation \[CGcoeff\]), not only the states at rests $|j_i,j_i\rangle$ are needed but also the higher ones ($|j_i,n_i\rangle$ with $n_i>j_i$). In that case, there are different possibilities and the Clebsch-Gordan coefficients contains binomial coefficients (see [@Oh]). For a bilocal operator, one gets : |j,n= (\_1+\_2)\^[n-j]{}\_[n\_1+n\_2=j-j\_1-j\_2]{}(
[c]{} n\_1 +n\_2\
n\_1
)|j\_1,j\_1+n\_1>|j\_2,j\_2+n\_2>\
\[bilocCG\]
The summation gives :
|j,n>\~(\_1+\_2)\^[n+j+1]{}(1+)\^[2j\_1-1]{}(1-)\^[2j\_2-1]{} P\_[j-j\_1-j\_2]{}\^[(2j\_1-1,2j\_2-1)]{}() \[repJa\] with = where $P_n^{(\nu_1,\nu_2)}(\xi)$ are the Jacobi polynomials[@Er] (see appendix \[app:a\]). Analogously, for product of three local operators, one gets Appell polynomials[@Er] (see appendix \[app:a\]). The important thing to remark about equation \[repJa\] is that the only dependence in $n$ is in the factor $(\alpha_1+\alpha_2)^{n+j+1}$, the other parts of this equation depend only of the conformal spins $j_1$, $j_2$ and $j$.
An important property will be needed for the next subsection. For any kind of irreducible representation of the collinear conformal subgroup $|j, n\rangle$ which is constructed from the tensor product of $k$-irreducible representations, there is the following homogeneity property : |j,n\_[j,n]{}(\_1,…,\_k)=(\_1+…+\_k )\^[n+j-1]{}\_[j,n]{}(\_1,…,\_k) \[hom\] where \_i=
This property comes almost directly from the constructions of the Clebsch-Gordon coefficients (equations \[jminCG\] and \[bilocCG\]). Hence irreducible representations $|j,n\rangle$ of the collinear conformal subgroup on the space of $k$ variables induce the irreducible representations on the functions defined on the simplex $\alpha_1+\ldots+\alpha_k=1$. This property will be used for the next subsection.
Matrix element of polylocal conformal operators
-----------------------------------------------
Suppose that one has the matrix element of a product of local conformal operators $\Phi_1(u_1z)\ldots\Phi_k(u_k)$ on the light-cone between the vacuum and a massless meson state $h$ of momentum $p_\mu$. This matrix element can be parameterized like this : 0|\_1(u\_1z)…\_k(u\_kz)|h(p)=\_i e\^[-ipz(u\_1\_1+…+u\_k\_k)]{}(\_1,…,\_k) where \_i=d\_1…d\_k(\_[i=1]{}\^[k]{} \_i-1)
According to the property of equation \[hom\], the wave function $\phi(\alpha_1,\ldots,\alpha_k)$ can be expanded in different parts which have a fixed conformal spin $j$. The minimum value of the conformal spin is the sum of the conformal spins of every conformal operators $\Phi_i(u_iz)$. The different parts of this conformal expansion are mutually orthogonal polynomials (Jacobi polynomials if there are only two operators) and form the complete set of functions on the simplex $\alpha_1+\ldots+\alpha_k=1$.
In front of each of these polynomials there is a coefficient which is renormalized multiplicatively. This conformal expansion is justified by the construction of irreducible representations of the collinear conformal subgroup.
The first term of such an expansion is called “asymptotic” wave function and can be easily computed (equation \[jminCG\]) : \_(\_1,…,\_k)= \_1\^[2j\_1-1]{} …\_k\^[2j\_k-1]{} \[as\] with \_(\_i)=1
Now one can understand better the meaning of the example given at the beginning of this section : \_ (u)=6u\^\_[n=0]{}a\^\_nC\_n\^[3/2]{} (). \[confex2\]
$\phi_\| (u)$ is the contribution of the matrix element $\langle 0|\ov{u} (z) \gamma_\mu [z,-z] d ( -z)|\rho^- (P, \lambda) \rangle$ where each quark field has a positive spin projection $s=+\frac{1}{2}$ (see equation \[confop1\]). Knowing that a quark field has the canonical dimension $l=\frac{3}{2}$, each quark field has a conformal spin $j_q=1$ (see equation \[confspin\]). So the asymptotic distribution amplitude \[as\] equals $\phi_{\text{as}}(\alpha_q, \alpha_{\ov{q}})=6\alpha_q\alpha_{\ov{q}}$ and has the conformal spin $j=2$. The higher terms in \[confex2\] corresponds of higher values of $j$ and $n$ in equation \[repJa\]. Denoting $u=\alpha_q$ with $\alpha_q+\alpha_{\ov{q}}=1$, one gets the expansion \[confex2\] (the Gegenbauer polynomials $C_n^{3/2}$ are proportional to the Jacobi ones $P^{(1,1)}_n$ (see Appendix C) which appear when $j_1=j_2=1$, see \[repJa\]).
Example of computation \[PsiPhi\]
=================================
In this section, the method of computing wave functions (defined in the appendix A) is illustrated with non-leading twist three-point distributions for the $\rho$-meson (following [@BBS]).
One starts with the definition of the chiral-even three-points distributions :
\
&=& f\_m\_\
& &+\[exechie3\]\
&=&if\_m\_\
& &+ \[exechie3b\]
One wants to compute the non-leading twist distributions $\Phi(v,pz)$, $\Psi(v,pz)$, $\tilde{\Phi}(v,pz)$ and $\tilde{\Psi}(v,pz)$. The leading twist distributions have the (following) conformal expansion (see [@BBKT]) : A(v,pz)=e\^[ipz(\_u-\_d+v\_g)]{}A() A()=360 \_3\_d\_u\_g\^2 +…V(v,pz)=e\^[ipz(\_u-\_d+v\_g)]{}V() V()=\_3\_3\^V\[(\_d-\_u)\_d\_u\_g\^2 +…\]
To compute $\Phi(v,pz)$ and $\tilde \Phi(v,pz)$, the wave function of conformal operators is needed : 0||u(z)gG\_.\_\_5\_\*d(-z) |(P,)=f\_m\_\^3\^(v,pz) \[t3conf1\] 0||u(z)gG\_\_\*\_\_5.d(-z) |(P,)=f\_m\_\^3\^(v,pz) \[t3conf2\] (this parameterization corresponds to the highest conformal spin for $\tilde G_{\mu\nu}$, although this gluon field does not look explicit like a conformal operator, see equations \[confop3\] to \[confop5\]).
These two matrix elements have fixed conformal spin, so a conformal expansion (with the help of the Appell polynomials) can be done : \^()=K\^ 60\_u\_g\^2\^()=K\^ 60\_d\_g\^2
Then it is easy to get $\Phi$ and $\tilde \Phi$ from $\Phi^{\ud}$ and $\Phi^{\du}$ : ()&=&\
()&=&
The G-parity (charge conjugation $\times$ isospin) applied to the matrix elements \[t3conf1\] and \[t3conf2\] implies the following relations : K\^&=&K\^\
\_[1,0]{}\^&=&\_[0,1]{}\^\
\_[1,0]{}\^&=&\_[0,1]{}\^
Finally, the expression for $\Phi$ and $\tilde \Phi$ (knowing that $1-\alpha_g-\alpha_u-\alpha_d=0$) is the following : ()=K\^30\_g\^2{(1-\_g)&+& \_[1,0]{}\^\
&+&.\_[0,1]{}\^} ()=K\^30\^2\_g(\_u-\_d)
Now, the different constants $K^{\ud}$, $\omega_{1,0}^{\ud}$ and $\omega_{0,1}^{\ud}$ has to be extracted from matrix elements of local operators.
To find $K^{\ud}$, the definition of $\tilde \Phi(v,pz)$ in \[exechie3b\] is taken. It implies : f\_m\_\^3(z)g\_ ()=0||uG.\_\_\_5 d|(P,)
Using the equation \[defxi\] which defines $\zeta_3$ and $\zeta_4$, $$\begin{aligned}
\langle0|\bar u g\tilde G_{\mu\nu}\gamma_\alpha
\gamma_5 d|\rho^-(P,\lambda)\rangle &=&
f_\rho m_\rho \zeta_{3}
\Bigg[
e^{(\lambda)}_\mu\Big(P_\alpha P_\nu-\frac{1}{3}m^2_\rho \,g_{\alpha\nu}\Big)
-e^{(\lambda)}_\nu\Big(P_\alpha P_\mu-\frac{1}{3}m^2_\rho \,g_{\alpha\mu}\Big)
\Bigg]
\nonumber\\
&&{}+\frac{1}{3}f_\rho m_\rho^3 \zeta_{4}
\Bigg[e^{(\lambda)}_\mu g_{\alpha\nu}- e^{(\lambda)}_\nu g_{\alpha\mu}\Bigg]
\label{exedefxi}\end{aligned}$$ one obtains for $K^{\ud}$ : K\^=\[Kud\]
The two other constants $\omega_{1,0}^{\ud}$ and $\omega_{0,1}^{\ud}$ have to be computed. The idea is to relate these coefficients to local matrix elements. To do it, the matrix elements \[t3conf1\] are expanded to the order $(z_\mu)$. It gives\
&=&i[f\_]{}[m\_]{}\^3(z)(p z)g\^\_(\_u-\_d)\^() \[derconft3a\] (\_u-\_d)\^()=K\^(+ \_[0,1]{}\^-\_[1,0]{}\^) 0||ug(.gG\_) .\_\_5\_\*d|0&=&-i[f\_]{}[m\_]{}\^3(z)(p z)g\^\_\_g\^()\
\
\_g\^()&=&K\^(+ \_[0,1]{}\^+\_[1,0]{}\^). \[derconft3d\]
These equations can be also rewritten like this (replacing $K^{\ud}$ by its value in \[Kud\]) :\
&=&-[f\_]{}[m\_]{}\^3.P.g\_+…\[Ra\]\
\
&=&[f\_]{}[m\_]{}\^3.P.g\_+…\[Rb\]\
where R\_1K\^\_[0,1]{}\^ \[defR1\]\
R\_2K\^\_[1,0]{}\^ \[defR2\]
Because of the G-parity symmetry, one can suspect that only two local operators are involved (apart from operators with total derivative). One defines : O\^1\_&=&|u(i\_gG\_+gG\_ i\_)\_\_5d\
O\^2\_&=&|u(-\_g G\_+g G\_ \_)\_d
The different Lorentz-structures for $O^1_{\alpha\beta\mu\nu}$ are the following :\
&=&{\_.\
& &-\_\
& &.-[m\_]{}\^2}A\_1[f\_]{}[m\_]{}\
& &+P\_(\_g\_-\_g\_)B\_1 +P\_(\_g\_-\_g\_)C\_1\
& &+\_(P\_g\_-P\_g\_)D\_1 + \_(P\_g\_-P\_g\_)E\_1\
& &-g\_(P\_\_-P\_\_) \[defO1\] And the constants $A_2$ to $F_2$ can be similarly defined, replacing $O^1$ by $O^2$ in this last equation.
Traces can be taken in order to reduce the number of independent constants (it works for both $O^1$ and $O^2$) : g\_ : B+C-D-E+4F=0 g\_ : -\_P\_(B+3C+F)-\_P\_(D+3E-F)= 0|O\_|(P,)g\_ : -\_P\_(3B+C+F)-\_P\_(3D+E-F)= 0|O\_|(P,)
One defines $a^1$, $a^2$, $b^1$ and $b^2$ in the following way : 0|O\^[1,2]{}\_|(P,)=(\_P\_+\_P\_)a\_[1,2]{} [f\_]{}[m\_]{}\^3 + \[defa1\] 0|O\^[1,2]{}\_|(P,)=(\_P\_+\_P\_)b\_[1,2]{} [f\_]{}[m\_]{}\^3 +
These higher (theoretical-)twist terms are not taken into account, because they are bigger than twist 4. This approximation gives : B&=&D=(a-3b)[m\_]{}\^3[f\_]{}\
E&=&C=(b-3a)[m\_]{}\^3[f\_]{}\[CE\]\
F&=&0
The goal is to express the unknown coefficients $\omega^{\ud}_{0,1}$ and $\omega_{1,0}^{\ud}$ in function of $A\cdots F$ and $a$, $b$.
For it, the definition of $\zeta_3$ and $\zeta_4$ in [@BBKT] (equation \[defxi\]) is used. It can be written like this : 0||ugG.\_\_\_5 d|(P,)=.g\_[f\_]{}[m\_]{}\^3(-\_3 +\_4) + …\[xi34\] Equation \[defO1\] implies : 0|O\^[1,2]{}\_|(P,)= e\^\_ P.g\_(-[f\_]{}[m\_]{}\^3A\_[1,2]{}+C\_[1,2]{}+E\_[1,2]{}) +…\[O..\]
Taking \[Ra\] and \[Rb\] with the total derivative of \[xi34\] (with $i\partial .\rightarrow P.$), combining \[CE\] with \[O..\], one gets : (-\_3+\_4)+R\_1 -R\_2=-\[A\_2-b\_2+3a\_2\] (-\_3+\_4)+R\_1 +R\_2=(-\_3 +\_4) +\[A\_1-b\_1+3a\_1\]
The constants $A_{1,2}$ and $B_{1,2}$ can be extracted from the coefficients of the twist 3 wave functions $A(\un \alpha)$ and $V(\un \alpha)$. For that, the equations \[exechie3\] and \[exechie3b\] are expanded at the order $(z_\mu)$ : 0 ||u(-\_.gG.\_+gG.\_.\_) d|(P,)= [f\_]{}[m\_]{}\_p\_\^3\_3\_3\^V 0 ||u.\_5(i\_gG.\_) d|(P,)= -[f\_]{}[m\_]{}\_p\_\^3\_3( +\_3\^A)
The definition of $A_{1,2}$ applied to these equations gives : A\_1&=&\_3( +\_3\^A)\
A\_2&=&-\_3\_3\^V.
Now, the three coefficients ($K^{\ud}$, $R_1$ and $R_2$ for $\Phi(\un \alpha)$ and $\tilde \Phi(\un \alpha)$ are expressed in functions of constants coming for the twist 3 wave functions and some unknown coefficients $a_{1,2}$, $b_{1,2}$. Their number can be reduced using exact operator relations (with the equation of motion) :
The equation D\_G\_=0 implies (after taking the matrix element) b\_1=-\_3-\_4
The equation O\^2\_-O\^2\_=O\^1\_ -O\^1\_+g\_implies a\_2-b\_2=a\_1-b\_1
And the equation\
&=&-12i|u\_(G\_\_-\_G\_ +()) d\
& &+\^2(\_|u\_d+ \_|u\_d)-4\_|u (\_G\_+\_G\_) \_5 d\
& &-\_|u\_G\_ \_5 d-\_|u\_G\_ \_5 d\
& &+g\_\_|u\_G\_ d implies 24a\_2=-O\_2+ O\_0+\_3 -\_4 with $O_{\mu\alpha\beta}$ the following leading twist operator (symmetric and traceless) : O\_=|u(\_D\_D\_)\_ d-
and $\langle\langle O_n\rangle\rangle$ the different moments of the leading twist part of the two-point wave function $\phi_\|$ : O\_n=M\_n\^[\_]{} with M\_n\^[\_]{}=du \^n\_(u) ($\langle\langle O_n\rangle\rangle$ is also defined by equation \[defOn\]).
Taking everything together, one gets : R\_2&=&(A\_1+A\_2)+a\_1- O\_2+ O\_0+\_4\
\
R\_1&=&(A\_1-A\_2)+a\_1+ O\_2- O\_0-\_3+\_4\
So the unknown coefficients $R_{1,2}$ (equation \[defR1\] and \[defR2\]) which are in the wave functions $\Phi$ and $\tilde \Phi$ are expressed in function of known coefficients, expect for $a_1$ which has to be computed by the method of QCD sum rules.
With the same method, $\Psi$ and $\tilde \Psi$ can be computed : ()&=&N\_1260(\_d\_u)\_d\_u\_g\
()&=&120\_u\_d\_gwith N\_&=&-O\_2+O\_0-a\_1-A\_2-\_3 -\_4\
\
\_0&=&\_3+\_4\
\_1&=&a\_1-A\_1 +\_3+\_4
So, in order to have the non-leading twist wave functions $\Phi$, $\tilde \Phi$, $\Psi$ and $\tilde \Psi$, the coefficient $a_1$ has to be computed. But it is defined from the matrix element of a local operator in equation \[defa1\], hence the method of two-point QCD sum rule described in [@SVZ] can be used.
In [@BBKT], there is a good presentation of the renormalization dependence of the coefficients appearing in the wave functions. The basic idea is to defined them from local conformal operators, where the renormalization dependence is better known (see section \[sec:confQCD\]).
Meson mass correction
=====================
In this section, some new techniques (partially reproduced in [@BBS]) to compute mass corrections of wave functions are described.
The mass of the $\rho$-meson induce some non-leading (physical-)twist structure for the non-local matrix element.
For example, for the chiral-even two-point distributions $$\begin{aligned}
\lefteqn{\langle 0|\bar u(x) \gamma_{\mu} [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle = }\makebox[2cm]{\ } \nonumber \\
&=& f_{\rho} m_{\rho} \left[ P_{\mu}
\frac{e^{(\lambda)}\cdot x}{P \cdot x}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(A^{(e)}(u)+
\frac{x^2 m_\rho^2}{4}\hat{A}^{(e)}(u)\right) \right.
\nonumber \\
& &{}+ \left(e^{(\lambda)}_{\mu}-P_\mu\frac{\epl\cdot x}{P\cdot x}\right)
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(B^{(e)}(u)
+\frac{x^2 m_\rho^2}{4}\hat{B}^{(e)}(u)\right)\nonumber \\
& & \left.{}- \frac{1}{2}x_{\mu}
\frac{e^{(\lambda)}\cdot x }{(P \cdot x)^{2}} m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} C(u)^{(e)} \right]\end{aligned}$$ some wave functions would disappear if the mass of the $\rho$ were $0$. Hence mass corrections are higher (physical-)twist contributions to non-local matrix elements (in the previous example, $\hat{A}^{(e)}(u)$ and $C^{(e)}(u)$ are twist 4, $\hat{B}^{(e)}$ is twist 5). In fact, the $m^2_\rho$ contribution can first be seen in the difference between the light-like basis ($p_\mu$, $z_\mu$ and $\epl_{\perp\mu}$ defined in \[defp\], \[defz\] and \[defeperp\]) needed for the conformal expansion and the vectors which characterize the $\rho$-meson ($P_\mu$ and $\epl_\mu$). These different basis introduce $m^2_\rho$ terms which are sort of “explicit” and “direct” corrections; but there are also the other corrections like $\hat{A}^{(e)}(u)$, $\hat{B}^{(e)}$ and $C^{(e)}(u)$.
Example of mass correction for a scalar theory \[sbscalar\]
-----------------------------------------------------------
Suppose that $\phi(x)$ is a scalar field with $M$ a scalar particle of momentum $P_\mu$ and mass $m$. If $x^2$ is almost on the light-cone, we have the following expansion : 0|(x)(-x)|M(P)=\_0\^1du e\^[iPx]{}\[scanloc\] with =2u-1
It is quite hard to compute directly $\hat{g}(u)$. In fact, the following approximation is done : in the Operator Product Expansion of $\phi(x)\phi(-x)$, only the leading-twist local operators (it is the “theoretical twist”, see section \[deftw\]) is taken. The leading (theoretical-)twist expansion of \[scanloc\] is defined in the following way : 0|\_|M(P)=\_0\^1du e\^[iPx]{}\[ltsca\]
In that case, $\hat{f}(u)$ can be computed in term of $f(u)$. The first way to do it is the “brute force” method, i. e. doing a local expansion of \[ltsca\] : 0|\_ |m=(iPx)\^nM\_n+m\^2x\^2n(n-1)\_[n-1]{}(iPx)\^[n-2]{} \[scmom\] where M\_n&=&\_0\^1du\^nf(u)\
\_n&=&\_0\^1du\^n(u) \[defmom\]
The leading twist operator is always symmetrical in all his indices and traceless.
It implies :\
&&{P\_[\_1]{}…P\_[\_n]{}- \_[i<j]{}\^ng\_[\_i\_j]{}P\_[\_1]{}…P\_[\_[i-1]{}]{}P\_[\_[i+1]{}]{}… P\_[\_[j-1]{}]{} P\_[\_[j+1]{}]{} …P\_[\_n]{}+O(m\^4)}\
Hence\
&=&{(Px)\^n-(Px)\^[n-2]{}+O(m\^4)} O\_n
The comparison of this last equation with \[scmom\] gives : M\_n&=&O\_n\
n\_[n-2]{}&=&O\_n=M\_n
The integrated version of these equations is (u)=2(2u-1)(u-1/2)\_u\^1dtf(t)+2(1-2u)(1/2-u)\_0\^udtf(t). \[fchap\]
The same answer can be obtained using the techniques developed in [@BaBr] directly in an integrated form : 0|\_|M(P)=\_0\^1du \_f(u) where \_f(u)=e\^[iPx]{}+\^2 \_0\^1dvve\^[ivPx]{}+O(m\^4)
It can be checked that these last equations reproduce \[fchap\] :\
&=&2\_0\^1 dt e\^[iPx(2t-1)]{}\_0\^1duf(u)\^2\_0\^1dv v (2t-1-v) and \_0\^1dv v(2t-1-v)&=&\
\
&=&(2t-1)\
The goal is to reproduce this formalism in the physical case : a vector $\rho$-meson in QCD.
Mass correction for two-point wave function : theoretical twist method \[thtmasscorr\]
--------------------------------------------------------------------------------------
### Leading theoretical twist
The idea is to compute the mass corrections in the leading theoretical twist approximation. In this part, the case with the two-point chiral even distributions is computed.
The different wave functions are : $$\begin{aligned}
\lefteqn{\langle 0|\bar u(x) \gamma_{\mu} [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle = }\makebox[2cm]{\ } \nonumber \\
&=& f_{\rho} m_{\rho} \left[ P_{\mu}
\frac{e^{(\lambda)}\cdot x}{P \cdot x}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(A^{(e)}(u)+
\frac{x^2 m_\rho^2}{4}\hat{A}^{(e)}(u)\right) \right.
\nonumber \\
& &{}+ \left(e^{(\lambda)}_{\mu}-P_\mu\frac{\epl\cdot x}{P\cdot x}\right)
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(B^{(e)}(u)
+\frac{x^2 m_\rho^2}{4}\hat{B}^{(e)}(u)\right)\nonumber \\
& & \left.{}- \frac{1}{2}x_{\mu}
\frac{e^{(\lambda)}\cdot x }{(P \cdot x)^{2}} m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} C^{(e)}(u) \right]
\label{tpmasscor}\end{aligned}$$
The “brute force” method, i. e. the expansion of the distributions in moments ($x\rightarrow 0$) can be used. It gives :\
&=&P\_\
& &+ (e\^[()]{}\_-P\_)\
& &-x\_m\_\^2(Px)\^nM\_n\^c where the moments $M_n^A$, $M_n^B$, $M_n^C$ $\hat{M}_n^A$ and $\hat{M}_n^B$ are defined like in \[defmom\].
The leading (theoretical-)twist of the operators in the left hand side can be written. The condition of symmetry and zero trace gives :\
&=&{ P\_(Px)\^n+ (e\^[()]{}\_-P\_) .\
& &-. P\_(Px)\^[n-2]{} n-x\_ (Px)\^nm\_\^2}O\_n\[defOn\]\
The comparison of these two equations implies : M\_n\^A&=&O\_n\[ltmom1\]\
M\_n\^B&=&O\_n\
\^A\_[n-2]{}&=&O\_n\
\^B\_[n-2]{}&=& O\_n\
M\_n\^C&=&O\_n\[ltmom5\]
There is another method to get the leading (theoretical-)twist part of the distributions \[tpmasscor\], giving directly integrated equations between the different wave functions. The different relations for non-local operators in appendix \[app:oprel\] are used : they give integrated forms for the leading twist part of two-point operators.
One takes equation \[BBformula\] :
\_=\_0\^1 du (ux)(-ux) \[vopsym\] where $\hat{x}=x^\nu\gamma_\nu$. The subscript “sym.” means that in the Operator Product Expansion (local expansion around $x=0$), only the operators which have symmetrical Lorentz-indices are taken. But for the leading theoretical twist, the traces has also to be removed, so the leading twist part of the right hand side of \[vopsym\] is also taken : \_=\_0\^1 du \_\[voplt\]
The condition for the leading twist part of $\ov{\psi}(ux)\hat{x}\psi(-ux)$ is the following (equation \[ltxhatapp\]) : \_=0 \[ltxhat\]
Now the matrix element of $\ov{\psi}(ux)\hat{x}\psi(-ux)$ can be taken. The parameterization of the leading twist part up to $m_\rho^2$ corrections is the following :
\
&=&(x)\_0\^1du g(u)\[xhatpar\] where $g(u)$ is the leading twist wave functions in the limit $m_\rho\rightarrow
0$.
The condition \[ltxhat\] can be applied; it gives (after some calculation) : \_0\^1 du g(u)=0 \[condfxi\]
The following ansatz for $f_\xi(Px)$ can be done (having in mind the case of scalar field in the previous subsection \[sbscalar\]) : f\_=a()\_0\^1b(v)e\^[ivPx]{}
Then equation \[condfxi\] gives a()&=&\^2\
b(v) &=&v\^2
This can be written in this form :\
&=&\_0\^1du g(u)\_ with\
&=&(x)
The equation \[voplt\] can be applied. After some algebra, one gets the distributions $A^{(e)}(u)$, $B^{(e)}(u)$, $C^{(e)}(u)$, $\hat{A}^{(e)}(u)$ and $\hat{B}^{(e)}(u)$ defined in \[tpmasscor\] in the leading (theoretical-) twist approximation : A\^[(e)]{}(u)&=&g(u) \[Aelt\]\
.\_0\^1duB\^[(e)]{}(u)e\^[iPx]{})\_&=&\_0\^1du g(u)\_0\^1dte\^[itPx]{}\
.\_0\^1duC\^[(e)]{}(u)e\^[iPx]{})\_&=&\_0\^1dug(u)\
\
.\_0\^1du\^[(e)]{}(u)e\^[iPx]{})\_&=& \_0\^1dug(u)\^2\_0\^1dtt\^2\
.\_0\^1du\^[(e)]{}(u)e\^[iPx]{})\_&=& \_0\^1dug(u)\_0\^1dt\^2\_0\^1v\^2dve\^[ivtPx]{} \[Bhatelt\]
There is no index “l. t.” for $A^{(e)}(u)$ because this distribution contains only the leading twist part of the matrix element $\langle 0|\bar u(x) \gamma_\mu [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle$. So these last equations show that all the distributions can be computed when one has $A^{(e)}(u)$, in the leading (theoretical-)twist approximation. They reproduced exactly the relations between moments (equations \[ltmom1\] to \[ltmom5\]). The advantage is that they are directly in an integrated form.
### Adding twist 3 and twist 4 terms
The idea is to add contributions of (theoretical-)twist 3 and 4 without the gluonic contribution. In the appendix \[app:oprel\], there is the following operator relation (\[dertot\]) : (x)(-x)&=&\_ -x\^2\_0\^1dv v\^2\^2 (vx)(-vx)\
& &+ +…\[txtdertot\]
The matrix element of the term with a total derivative, gives\
&=&m\^2\_x\^2\_0\^1dv v\^20|u(vx)\[vx,-vx\]d(-vx)| (P,)
The $O(m_\rho^0)$ term (equation \[xhatpar\]) in the parameterization of the matrix element in the right hand side implies :\
&=&(x)\_0\^1 du g(u)\_0\^1dv v\^2e\^[ivPx]{}
So the leading twist part with the total derivative term is :\
&=&\_0\^1du g(u)\_ with\
&=&(x)\[totdercor\]
This last form can be used to give some twist 4 corrections to equation \[Aelt\] to \[Bhatelt\].
Now the twist 3 part without gluon can be added. The matrix element of the equations \[twist3\] is taken, suppressing the gluonic terms :\
&=&-i\_\_0\^1dtt x\^(-iP\^)0||u(tx) \_\_5 \[tx,-tx\] d(-tx)|\^-(P,)\
\
\
&=&-i\_\_0\^1dtt x\^(-iP\^)0||u(tx) \_ d(-tx)|\^-(P,)\
Combining these two equations, replacing the right hand side of the second one with the parameterization of \[tpmasscor\], one gets :\
&=&(\_(Px)\^2+P\_(x)(Px)\_0\^1du \_0\^1 dtt \_0\^1 dv v\^2 B\^[(e)]{}(u)e\^[ivtPx]{}\
& &+m\^2\_\_0\^1du\_0\^1 dt t\_0\^1 dv v\^2 e\^[ivtPx]{}\
& &\
Including the leading twist term, the total derivative term (\[totdercor\]) and this last equation, equations for $A^{(e)}(u)$, $B^{(e)}(u)$, $C^{(e)}(u)$, $\hat{A}^{(e)}(u)$ and $\hat{B}^{(e)}(u)$ can be obtained : A\^[(e)]{}(u)&=&g(u) \[Aet234\]\
\_0\^1duB\^[(e)]{}(u)e\^[iPx]{}&=&\_0\^1du g(u)\_0\^1dte\^[itPx]{}\
& &-(Px)\^2\_0\^1du\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}B\^[(e)]{}(u)\
\
-\_0\^1duC\^[(e)]{}(u)e\^[iPx]{}&=&\_0\^1dug(u)\_0\^1dtt\^2\_0\^1dvv\^2e\^[ivtPx]{}\
& &-\_0\^1du B\^[(e)]{}(u)\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}\
\
\_0\^1du\^[(e)]{}(u)e\^[iPx]{}&=& \_0\^1dug(u)(\^2+1)\_0\^1dtt\^2\_0\^1v\^2dve\^[ivtPx]{}\
& &+4\_0\^1duB\^[(e)]{}(u)\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}\
& &-(Px)\^2\_0\^1du\^[(e)]{}(u)\_0\^1dtt\^3 \_0\^1dvv\^3e\^[itvPx]{}\
\[Bhatet234\]\
& & \_0\^1dug(u)(\^2+1)\_0\^1dtt\^2\_0\^1dvv\^2(i(Px)+1) t\^2e\^[ivtPx]{}\
& &+4\_0\^1duB\^[(e)]{}(u)\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}\
Theoretically, all the chiral-even distributions $B^{(e)}(u)$, $C^{(e)}(u)$, $\hat{A}^{(e)}(u)$ can be computed once the leading (physical-)twist contribution $A^{(e)}(u)$ is known. But, practically, it quite long and hard if one has a conformal expansion for $A^{(e)}(u)$ (except if only the asymptotic part of the wave function is taken, see [@BBS]). The advantage of these equations is that they are independent of the model used to computed $A^{(e)}(u)$ (another expansion can be used as the conformal one). They can be the basis of a numerical computation if one has discrete values for $A^{(e)}(u)$.
In the appendix \[app:masscorr\], similar equations for the chiral-odd distributions are given.
Mass correction using the equation of motion
--------------------------------------------
Equation of motion is used to relate the two-point wave functions to the three points ones (these calculations come from [@BBS]).
From [@BF], page 245, one has : i\_u(x)\^d(-x)=\_[-1]{}\^1dvu(x)gG\_ (vx)\^d(-x) The symbol $\partial_\mu$ is the derivative over the total translation define in \[totder\].
When the matrix element $\langle 0|\ldots|\rho(P,\lambda)\rangle$ is taken (replacing $x$ by $z$), on gets (using the definitions \[eq:vda\] and \[chie3\] for $\phi_\|(u)$, $g_3(u)$, $\Phi(u)$ and $\Psi(u)$) :\
&=&-if\_m\_\^3(z)\_[-1]{}\^1dv e\^[-ipz(\_u-\_d+v\_g)]{}\
The integration of the right hand side can be transformed : \_[-1]{}\^1 dvF()e\^[-ipz(\_u-\_d+ v\_g)]{}=2\_0\^1dwe\^[ipz(2w-1)]{}\_0\^wd\_d\_0\^[1-x]{}d\_u F()
And a relation between the two-points and three-points distributions is obtained : =\_0\^ud\_d \_0\^[1-u]{}d\_u
In a similar way, the operator relation (in [@BF]) u(x)\_d(-x) = -i\_[-1]{}\^1dvv u(x)x\^g G\_(vx)\^d(-x) gives \^[(e)]{}(u)+2C\_2\^[(e)]{}(u)=\_0\^ud\_d\_0\^[1-u]{}d\_u (\_uu-\_d(1-u))with C\_2\^[(e)]{}(u)=\_0\^u\_0\^vdtC\^[(e)]{}(t)
Hence from the wave functions computed in section \[PsiPhi\], the mass correction of the two-points functions can be computed. The advantage of this method is that it is quite simple, but it depends on the model for $\Phi$ and $\Psi$, and there is no general methods to find equations that relate the two-point to the three points functions.
Conclusion
==========
In this work, I describe the different wave functions of meson, the method conformal expansion and its application to the non-leading twist wave functions. I also give methods of computing the mass correction, which helps to apply these results to heavier vector mesons as the $\rho$.
The use of conformal expansion and exact operators relations coming from the equation of motion is still long and heavy, a more general method would be certainly useful, and there is no work of higher gluonic corrections, like four-points wave functions for example.
These non-local, non-perturbative objects are needed for the computation of exclusive decays, but they also describe the nature of the quark confinement in meson in a Lorentz-invariant and gauge invariant way. A better knowledge of the non-leading twist wave functions would probably help to understand how QCD describes the physics of hadron.
Another question is the possible link to effective low energy theories; for example, the quark condensate $\langle 0|\ov{q}q|0\rangle$ is an order parameter of the breakdown of the chiral symmetry at low energy; can one found such thing with non-local wave functions ?
The basis of the conformal expansion is the conformal invariance of QCD to one loop. But a analysis of the two-loop effects on this expansion, a better renormalization group dependence of these wave functions would be interesting.
Other ways of determining these wave functions (by experiments or using another expansion as the conformal one) is certainly desirable, and the “theoretical twist method” of determining the mass corrections that I give in this work will still be applicable.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work is supported by Schweizerischer Nationalfond. I am grateful to V. M. Braun and P. Ball for introducing me to the subject and to D. Wyler for encouragement and for proofreading this work.
List of vector-meson wave functions up to twist $4$
===================================================
In this appendix, I give a list of vector-meson wave functions (the chosen meson is the $\rho$), following the notation of [@BBKT] and [@BBS].
Two-points distributions
------------------------
Chiral-odd $\rho$ two-point distributions : $$\begin{aligned}
\lefteqn{\hspace*{-1.5cm}\langle 0|\bar u(z) \sigma_{\mu \nu} [z,-z]
d(-z)|\rho^-(P,\lambda)\rangle =} \nonumber \\
&=& i f_{\rho}^{T} \left[ ( e^{(\lambda)}_{\perp \mu}p_\nu -
e^{(\lambda)}_{\perp \nu}p_\mu )
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} \phi_{\perp}(u, \mu^{2}) \right.
\nonumber \\
& &{}+ (p_\mu z_\nu - p_\nu z_\mu )
\frac{e^{(\lambda)} \cdot z}{(p \cdot z)^{2}}
m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} {h_\parallel^{(t)}}(u, \mu^{2})
\nonumber \\
& & \left.{}+ \frac{1}{2}
(e^{(\lambda)}_{\perp \mu} z_\nu -e^{(\lambda)}_{\perp \nu} z_\mu)
\frac{m_{\rho}^{2}}{p \cdot z}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} h_{3}(u, \mu^{2}) \right],
\label{eq:tda}\end{aligned}$$ $$\langle 0|\bar u(z) [z,-z]
d(-z)|\rho^-(P,\lambda)\rangle
= -i \left(f_{\rho}^{T} - f_{\rho}\frac{m_{u} + m_{d}}{m_{\rho}}
\right)(e^{(\lambda)}\cdot z) m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} {h_\parallel^{(s)}}(u, \mu^{2}),
\label{eq:sda}$$
Chiral-even two-points distributions : $$\begin{aligned}
\lefteqn{\langle 0|\bar u(z) \gamma_{\mu} [z,-z]
d(-z)|\rho^-(P,\lambda)\rangle = }\makebox[2cm]{\ } \nonumber \\
&=& f_{\rho} m_{\rho} \left[ p_{\mu}
\frac{e^{(\lambda)}\cdot z}{p \cdot z}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} \phi_{\parallel}(u, \mu^{2}) \right.
+ e^{(\lambda)}_{\perp \mu}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} g_{\perp}^{(v)}(u, \mu^{2})
\nonumber \\
& & \left.{}- \frac{1}{2}z_{\mu}
\frac{e^{(\lambda)}\cdot z }{(p \cdot z)^{2}} m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} g_{3}(u, \mu^{2}) \right]
\label{eq:vda}\end{aligned}$$ and $$\begin{aligned}
\lefteqn{\langle 0|\bar u(z) \gamma_{\mu} \gamma_{5}[z,-z]
d(-z)|\rho^-(P,\lambda)\rangle = }\makebox[2cm]{\ } \nonumber \\
&=& \frac{1}{2}\left(f_{\rho} - f_{\rho}^{T}
\frac{m_{u} + m_{d}}{m_{\rho}}\right)
m_{\rho} \epsilon_{\mu}^{\phantom{\mu}\nu \alpha \beta}
e^{(\lambda)}_{\perp \nu} p_{\alpha} z_{\beta}
\int_{0}^{1} \!du\, e^{i \xi p \cdot z} g^{(a)}_{\perp}(u, \mu^{2}).
\label{eq:avda}\end{aligned}$$ with the notation $$\xi = u - (1-u) = 2u-1.$$
The vector and tensor decay constants $f_\rho$ and $f_\rho^T$ are defined as usually as $$\begin{aligned}
\langle 0|\bar u(0) \gamma_{\mu}
d(0)|\rho^-(P,\lambda)\rangle & = & f_{\rho}m_{\rho}
e^{(\lambda)}_{\mu},
\label{eq:fr}\\
\langle 0|\bar u(0) \sigma_{\mu \nu}
d(0)|\rho^-(P,\lambda)\rangle &=& i f_{\rho}^{T}
(e_{\mu}^{(\lambda)}P_{\nu} - e_{\nu}^{(\lambda)}P_{\mu}),
\label{eq:frp}\end{aligned}$$
--------- -------------------- ---------------------------------------------- --------------
Twist 2 3 4
$O(1)$ $O(1/Q)$ $O(1/Q^{2})$
$ $ $\phi_{\parallel}$ ${h_\parallel^{(t)}}$, ${h_\parallel^{(s)}}$ $g_{3}$
$ $ $\phi_{\perp}$ $g_{\perp}^{(v)}$, $g_{\perp}^{(a)}$ $h_{3}$
\[2pt\]
--------- -------------------- ---------------------------------------------- --------------
: “Physical twist” classification of two-points distributions[]{data-label="tab:2"}
If every $m^2_\rho$ terms are needed, $x^2$ corrections has to be made. In that case, the notation is a little different :
Chiral-odd two-points distributions : $$\begin{aligned}
\lefteqn{\hspace*{-1.5cm}\langle 0|\bar u(x) \sigma_{\mu \nu} [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle =} \nonumber \\
&=& i f_{\rho}^{T} \left[ ( e^{(\lambda)}_{\mu}P_\nu -
e^{(\lambda)}_{\nu}P_\mu )
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(A^{(o)}(u)
+\frac{m^2_\rho x^2}{4}\hat{A}^{(o)}(u)\right)\right.
\nonumber \\
& &{}+ (P_\mu x_\nu - P_\nu x_\mu )
\frac{e^{(\lambda)} \cdot x}{(P \cdot x)^{2}}m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x}\left(B^{(o)}(u)
+\frac{m^2_\rho x^2}{4}\hat{B}^{(o)}(u)\right)
\nonumber \\
& & \left.{}+ \frac{1}{2}
(e^{(\lambda)}_{\mu} x_\nu -e^{(\lambda)}_{\nu} x_\mu)
\frac{m_{\rho}^{2}}{P \cdot x}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} C^{(o)}(u) \right],
\label{eq:frpm}\end{aligned}$$ with A\^[(o)]{}(u)&=&\_(u)\
B\^[(o)]{}(u)&=&h\_\^[(t)]{}(u)-\_(u)-h\_3(u)\
C\^[(o)]{}(u)&=&h\_3(u)-\_(u)
Chiral-even two-points distributions : $$\begin{aligned}
\lefteqn{\langle 0|\bar u(x) \gamma_{\mu} [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle = }\makebox[2cm]{\ } \nonumber \\
&=& f_{\rho} m_{\rho} \left[ P_{\mu}
\frac{e^{(\lambda)}\cdot x}{P \cdot x}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(A^{(e)}(u)+
\frac{x^2 m_\rho^2}{4}\hat{A}^{(e)}(u)\right) \right.
\nonumber \\
& &{}+ \left(e^{(\lambda)}_{\mu}-P_\mu\frac{\epl\cdot x}{P\cdot x}\right)
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(B^{(e)}(u)
+\frac{x^2 m_\rho^2}{4}\hat{B}^{(e)}(u)\right)\nonumber \\
& & \left.{}- \frac{1}{2}x_{\mu}
\frac{e^{(\lambda)}\cdot x }{(P \cdot x)^{2}} m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} C^{(e)}(u) \right]
\label{eq:vdam}\end{aligned}$$ with A\^[(e)]{}(u)&=&\^(u)\
B\^[(e)]{}(u)&=&g\^[(v)]{}\_(u)\
C\^[(u)]{}(u)&=&g\_3(u)+\_(u)-2g\^[(v)]{}\_(u)
Three-points distributions
--------------------------
Chiral-odd three-points distributions :
\
&=& f\_\^T m\_\^2\
where ${\cal T}$ is a twist 3 distribution (“physical twist”), the other are twist 4. There are other Lorentz-structures which are twist 5.
\
&=& i f\^T\_m\_\^2\[\_p\_-e\_\]S(v,pz)
\
&=& i f\^T\_m\_\^2\[\_p\_-e\_\](v,pz)
These $S(v,pz)$ and $\tilde{S}(v,pz)$ are twist 4 distributions.
Chiral-even three-points distributions :\
&=& f\_m\_\[chie3\]\
&=&if\_m\_where ${\cal A}(v,pz)$ and ${\cal V}(v,pz)$ are twist 3 distributions, the other twist 4.
Every functions ${\cal F}(v,pz)$ have such form : $${\cal F}(v,pz) =\int {\cal D}\underline{\alpha}
e^{-ipz(\alpha_u-\alpha_d+v\alpha_g)}{\cal F}(\underline{\alpha}),$$ and $\underline{\alpha}$ is the set of three momentum fractions $\underline{\alpha}=\{\alpha_d,\alpha_u,\alpha_g\}$. The integration measure is defined as $$\int {\cal D}\underline{\alpha} \equiv \int_0^1 d\alpha_d
\int_0^1 d\alpha_u\int_0^1 d\alpha_g \,\delta(1-\sum \alpha_i).
\label{eq:measure}$$
Models for $\rho$ wave functions up to twist 4
==============================================
I reproduce the results of [@BB] here, which helps to understand section \[PsiPhi\]. Their computations can be found in [@BBKT] and [@BBS], where a more complete list can be found.
Chiral-even distributions
-------------------------
The model for the leading twist 2 distribution amplitude $\phi_\parallel$ is $$\label{eq:phipar}
\phi_\parallel(u) = 6 u\bar u \left[ 1 + 3 a_1^\parallel\, \xi +
a_2^\parallel\, \frac{3}{2} ( 5\xi^2 - 1 ) \right]$$ with parameter values as specified in Tab. \[tab:para\]. The expressions for higher-twist distributions given below correspond to the simplest self-consistent approximation which satisfies the QCD equations of motion [@BBKT; @BBS] :
- Three-particle distributions of twist 3 : $$\begin{aligned}
{\cal V} (\underline{\alpha}) &=&
540\, \zeta_3 \omega^V_3 (\alpha_d-\alpha_u)\alpha_d \alpha_u\alpha_g^2,
\label{modelV}\\
{\cal A} (\underline{\alpha}) &=&
360\,\zeta_3 \alpha_d \alpha_u \alpha_g^2
\Big[ 1+ \omega^A_{3}\frac{1}{2}(7\alpha_g-3)].
\label{modelA}
\end{aligned}$$
- Two-particle distributions of twist 3 : $$\begin{aligned}
g_\perp^{(a)}(u) & = & 6 u \bar u \left[ 1 + a_1^\parallel \xi +
\left\{\frac{1}{4}a_2^\parallel +
\frac{5}{3}\, \zeta_{3} \left(1-\frac{3}{16}\,
\omega^A_{3}+\frac{9}{16}\omega^V_3\right)\right\}
(5\xi^2-1)\right]\nonumber\\
& & {} + 6\, \widetilde{\delta}_+ \, (3u \bar u + \bar u \ln \bar u +
u \ln u ) +
6\, \widetilde{\delta}_- \, (\bar u \ln \bar u - u \ln u),\\
g_\perp^{(v)}(u) & = & \frac{3}{4}(1+\xi^2)
+ a_1^\parallel\,\frac{3}{2}\, \xi^3
+ \left(\frac{3}{7} \,
a_2^\parallel + 5 \zeta_{3} \right) \left(3\xi^2-1\right)
\nonumber\\
& & {}+ \left[ \frac{9}{112}\, a_2^\parallel
+ \frac{15}{64}\, \zeta_{3}\Big(3\,\omega_{3}^V-\omega_{3}^A\Big)
\right] \left( 3 - 30 \xi^2 + 35\xi^4\right)\nonumber\\
& & {}+\frac{3}{2}\,\widetilde{\delta}_+\,(2+\ln u + \ln\bar u) +
\frac{3}{2}\,\widetilde{\delta}_-\, ( 2 \xi + \ln\bar u - \ln u),\label{eq:gv}\end{aligned}$$
- Three-particle distributions of twist 4 : $$\begin{aligned}
\widetilde \Phi (\underline{\alpha}) &=&
\Big[-\frac{1}{3}\zeta_{3}+\frac{1}{3}\zeta_{4}\Big]
30(1-\alpha_g)\alpha_g^2,
\nonumber\\
\Phi (\underline{\alpha}) &=&
\Big[-\frac{1}{3}\zeta_{3}+\frac{1}{3}\zeta_{4}\Big]
30(\alpha_u-\alpha_d)\alpha_g^2,
\nonumber\\
\widetilde\Psi (\underline{\alpha}) &=&
\Big[\frac{2}{3}\zeta_{3}+\frac{1}{3}\zeta_{4}\Big]
120 \alpha_u\alpha_d\alpha_g,
\nonumber\\
\Psi (\underline{\alpha}) &=& 0.\end{aligned}$$
- Two-particle distributions of twist 4 : $$\begin{aligned}
A^{(e)}(u) &=& \Bigg[\frac{4}{5}+\frac{20}{9} \zeta_{4}
+\frac{8}{9} \zeta_{3}\Bigg]30 u^2(1-u)^2,
\nonumber\\
g_3(u) &=& 6u(1-u) + \Bigg[\frac{10}{3} \zeta_{4}
-\frac{20}{3} \zeta_{3}\Bigg](1-3 \xi^2),
\nonumber\\
C^{(e)}(u) &=& \Bigg[\frac{3}{2}+\frac{10}{3} \zeta_{4}
+\frac{10}{3} \zeta_{3}\Bigg](1-3 \xi^2),\end{aligned}$$
where the dimensionless couplings $\zeta_3$ and $\zeta_4$ are defined as local matrix elements $$\begin{aligned}
\langle0|\bar u g\tilde G_{\mu\nu}\gamma_\alpha
\gamma_5 d|\rho^-(P,\lambda)\rangle &=&
f_\rho m_\rho \zeta_{3}
\Bigg[
e^{(\lambda)}_\mu\Big(P_\alpha P_\nu-\frac{1}{3}m^2_\rho \,g_{\alpha\nu}\Big)
-e^{(\lambda)}_\nu\Big(P_\alpha P_\mu-\frac{1}{3}m^2_\rho \,g_{\alpha\mu}\Big)
\Bigg]
\nonumber\\
&&{}+\frac{1}{3}f_\rho m_\rho^3 \zeta_{4}
\Bigg[e^{(\lambda)}_\mu g_{\alpha\nu}- e^{(\lambda)}_\nu g_{\alpha\mu}\Bigg]
\label{defxi}\end{aligned}$$ and have been estimated from QCD sum rules [@ZZC85; @BK86]. The terms in $\delta_\pm$ and $\widetilde\delta_\pm$ specify quark-mass corrections in twist 3 distributions induced by the equations of motion. The numerical values of these and other coefficients are listed in Tabs. \[tab:para\] and \[tab:para2\][^2]. Note that SU(3) breaking effects are neglected in twist 4 distributions and in gluonic parts of twist 3 distributions.
The numbers $a_1^\|$, $a_2^\|$, $\omega_3^A$, $\omega_3^V$ are computed in the appendix C of [@BBKT], using usual two-points QCD sum rule. $\tilde\delta_+$ and $\tilde\delta_-$ have following definitions : \_=
Chiral-odd distributions
------------------------
$$\begin{array}{|c|cccc|}
\hline
V & \rho^\pm & K^*_{u,d} & \bar{K}^*_{u,d}& \phi\\ \hline
f_V [{\rm MeV}] & 198\pm 7 & 226 \pm 28& 226 \pm 28 & 254 \pm 3\\
f^T_V [{\rm MeV}] &
\begin{array}{c} 160\pm 10 \\ 152\pm 9 \end{array}&
\begin{array}{c} 185\pm 10 \\ 175\pm 9 \end{array}&
\begin{array}{c} 185\pm 10 \\ 175\pm 9 \end{array}&
\begin{array}{c} 215\pm 15 \\ 204\pm 14 \end{array}
\\ \hline
a_1^\parallel & 0 &
\begin{array}{c} 0.19 \pm 0.05 \\ 0.17\pm 0.04 \end{array}&
\begin{array}{c} -0.19 \pm 0.05 \\ -0.17\pm 0.04 \end{array}&
\phantom{-}0\\
a_2^\parallel &
\begin{array}{c} 0.18 \pm 0.10 \\ 0.16\pm 0.09 \end{array}&
\begin{array}{c} 0.06 \pm 0.06 \\ 0.05\pm 0.05 \end{array}&
\begin{array}{c} \phantom{-}0.06 \pm 0.06 \\
\phantom{-}0.05\pm 0.05 \end{array}&
0\pm0.1\\
a_1^\perp & 0 &
\begin{array}{c} 0.20 \pm 0.05 \\ 0.18\pm 0.05 \end{array}&
\begin{array}{c} -0.20 \pm 0.05 \\ -0.18\pm 0.05 \end{array}&
\phantom{-}0\\
a_2^\perp &
\begin{array}{c} 0.20 \pm 0.10 \\ 0.17\pm 0.09 \end{array}&
\begin{array}{c} 0.04 \pm 0.04 \\ 0.03\pm 0.03 \end{array}&
\begin{array}{c} \phantom{-}0.04 \pm 0.04 \\
\phantom{-}0.03\pm 0.03 \end{array}&
0\pm0.1\\ \hline
\delta_+ & 0 &
\begin{array}{c} \phantom{-}0.24 \\ \phantom{-}0.22 \end{array}&
\begin{array}{c} 0.24 \\ 0.22 \end{array}&
\begin{array}{c} 0.46 \\ 0.41 \end{array}
\\
\delta_- & 0 &
\begin{array}{c} -0.24 \\ -0.22 \end{array}&
\begin{array}{c} 0.24 \\ 0.22 \end{array}&
0 \\
\widetilde{\delta}_+ & 0 &
\begin{array}{c} \phantom{-}0.16 \\ \phantom{-}0.13 \end{array}&
\begin{array}{c} 0.16 \\ 0.13 \end{array}&
\begin{array}{c} 0.33 \\ 0.27 \end{array}
\\
\widetilde{\delta}_- & 0 &
\begin{array}{c} -0.16 \\ -0.13 \end{array}&
\begin{array}{c} 0.16 \\ 0.13 \end{array}&
0\\ \hline
\end{array}$$
$$\begin{array}{|c|ccccccc|}\hline
& \zeta_3 & \omega_3^A & \omega_3^V & \omega_3^T & \zeta_4 & \zeta_4^T
& \tilde{\zeta_4^T}\\ \hline
V &
\begin{array}{c} 0.032\\ 0.023 \end{array}&
\begin{array}{c} -2.1 \\ -1.8 \end{array}&
\begin{array}{c} 3.8 \\ 3.7 \end{array}&
\begin{array}{c} 7.0 \\ 7.5 \end{array}&
\begin{array}{c} 0.15 \\ 0.13 \end{array}&
\begin{array}{c} 0.10 \\ 0.07 \end{array}&
\begin{array}{c} -0.10\\ -0.07 \end{array}
\\ \hline
\end{array}$$
The model for the leading twist 2 distribution amplitude $\phi_\perp$ is $$\label{eq:phiperp}
\phi_\perp(u) = 6 u\bar u \left[ 1 + 3 a_1^\perp\, \xi +
a_2^\perp\, \frac{3}{2} ( 5\xi^2 - 1 ) \right]$$ with parameter values as specified in Tab. \[tab:para\]. The expressions for higher-twist distributions given below correspond to the simplest self-consistent approximation which satisfies all QCD equations of motion [@BBKT; @BBS] :
- Three-particle distribution of twist 3 : $${\cal T} (\underline{\alpha}) =
540\, \zeta_3 \omega^T_3 (\alpha_d-\alpha_u)\alpha_d \alpha_u\alpha_g^2.$$
- Two-particle distributions of twist 3 : $$\begin{aligned}
h_\parallel^{(s)}(u) & = & 6u\bar u \left[ 1 + a_1^\perp \xi + \left( \frac{1}{4}a_2^\perp +
\frac{5}{8}\,\zeta_{3}\omega_3^T \right) (5\xi^2-1)\right]\nonumber\\
& & {}+ 3\, \delta_+\, (3 u \bar u + \bar u \ln \bar u + u \ln u) +
3\,\delta_-\, (\bar u
\ln \bar u - u \ln u),\label{eq:e}\\
h_\parallel^{(t)}(u) &= & 3\xi^2+ \frac{3}{2}\,a_1^\perp \,\xi (3 \xi^2-1)
+ \frac{3}{2} a_2^\perp\, \xi^2 \,(5\xi^2-3)
+\frac{15}{16}\zeta_{3}\omega_3^T(3-30\xi^2+35\xi^4)\nonumber\\
& & {} + \frac{3}{2}\,\delta_+
\, (1 + \xi \, \ln \bar u/u) + \frac{3}{2}\,\delta_- \, \xi\, ( 2
+ \ln u + \ln\bar u )
\label{eq:hL}\end{aligned}$$
- Three-particle distributions of twist 4 : $$\begin{aligned}
T^{(4)}_1(\underline{\alpha}) &=& T^{(4)}_3(\underline{\alpha}) ~~=~~0,
\nonumber\\
T^{(4)}_2(\underline{\alpha}) &=&
30 \widetilde \zeta^T_{4}(\alpha_d-\alpha_u)\alpha_g^2,
\nonumber\\
T^{(4)}_4(\underline{\alpha}) &=&
- 30 \zeta^T_{4}(\alpha_d-\alpha_u)\alpha_g^2,
\nonumber\\
S(\underline{\alpha}) &=& 30 \zeta^T_{4}(1-\alpha_g)\alpha_g^2,
\nonumber\\
\widetilde S(\underline{\alpha}) &=&
30 \widetilde \zeta^T_{4}(1-\alpha_g)\alpha_g^2.\end{aligned}$$
- Two-particle distributions of twist 4 : $$\begin{aligned}
h_3(u) &=& 6u(1-u)+5[\zeta^T_4+\widetilde \zeta^T_4](1-3\xi^2),
\nonumber\\
A^{(o)}(u) &=& 30 u^2(1-u)^2
\Bigg[\frac{2}{5}+\frac{4}{3}\zeta^T_4-\frac{8}{3}
\widetilde \zeta^T_4\Bigg].\end{aligned}$$
The constants $\zeta^T_4$ and $\widetilde \zeta^T_4$ are defined as $$\begin{aligned}
\langle 0|\bar u gG_{\mu \nu}d|\rho^-(P,\lambda)\rangle &=&
if_\rho^T m_\rho^3 \zeta^T_4(
e^{(\lambda)}_{\mu}P_\nu - e^{(\lambda)}_{\nu}P_\mu),
\nonumber\\
\langle 0|\bar u g\widetilde G_{\mu \nu}i\gamma_5
d|\rho^-(P,\lambda)\rangle &=&
if_\rho^T m_\rho^3 \widetilde \zeta^T_4(
e^{(\lambda)}_{\mu}P_\nu - e^{(\lambda)}_{\nu}P_\mu)\end{aligned}$$ and have been estimated in [@BBK] from QCD sum rules: $$\zeta^T_4 \simeq - \widetilde \zeta^T_4 \simeq 0.10.$$ Other parameters are given in Tab. \[tab:para\][^3]. Like in the chiral-even case, SU(3) breaking corrections are neglected in twist 4 distributions.
The number $\omega_3^T$, $a_1^\perp$, $a_2^\perp$ are computed in the appendix C of [@BBKT], using usual two-points QCD sum rule. $\delta_+$ and $\delta_-$ have the following definition : \_=
Formulas for Orthogonal Polynomials\[app:a\]
============================================
This appendix reproduces useful formulas for the Appell, Jacobi and Gegenbauer polynomials.
Differentiation formula for Gegenbauer polynomials : $$\frac{d}{d\xi} (1-\xi^{2})C_{n}^{3/2}(\xi) =
-(n+1)(n+2) C_{n}^{1/2}(\xi).
\label{eq:jg1}$$ Differentiation formulae for Jacobi polynomials : $$\begin{aligned}
\frac{d}{d\xi} P_{n}^{(\nu_{1}, \nu_{2})}(\xi)
&=& \frac{n + \nu_{1} + \nu_{2} + 1}{2}
P_{n-1}^{(\nu_{1}+1,\nu_{2}+1)}(\xi),
\label{eq:jd2}\\
(1 + \xi) P_{n}^{(1,1)}(\xi)
&=& 2\frac{d}{d\xi}\left[
\frac{(n+1)}{(n+2)(2n+3)} P_{n+2}^{(0,0)}(\xi)\right.
\nonumber \\
& &\left. {}+\frac{1}{n+2} P_{n+1}^{(0,0)}(\xi)
+ \frac{1}{2n+3}P_{n}^{(0,0)}(\xi)\right].
\label{eq:jd1}\end{aligned}$$ The equation (\[eq:jd1\]) is obtained from (\[eq:jd2\]) combined with (\[eq:jrec1\]) below.
Recurrence formulae for Jacobi polynomials : $$\begin{aligned}
(1+\xi) P_{n}^{(1,1)}(\xi) &=& \frac{(n+1)(n+3)}{(n+2)(2n+3)}
P_{n+1}^{(1,1)}(\xi) + P_{n}^{(1,1)}(\xi) + \frac{n+1}{2n+3}
P_{n-1}^{(1,1)}(\xi)
\label{eq:jrec1}\nonumber \\ \\
&=&\frac{2(n+1)}{2n+3} \left(P_{n}^{(1,0)}(\xi) + P_{n+1}^{(1,0)}(\xi)
\right),
\label{eq:jrec12}\\
(1-\xi) P_{n}^{(1,1)}(\xi) &=&
\frac{2(n+1)}{2n+3} \left(P_{n}^{(0,1)}(\xi) - P_{n+1}^{(0,1)}(\xi)
\right),
\label{eq:jrec13}\\
P_{n}^{(0,0)}(\xi) &=& \frac{n+1}{2n+1}P_{n}^{(1,0)}(\xi)
- \frac{n}{2n+1}P_{n-1}^{(1,0)}(\xi)
\nonumber \\
&=&
\frac{n+1}{2n+1}P_{n}^{(0,1)}(\xi)
+ \frac{n}{2n+1}P_{n-1}^{(0,1)}(\xi),
\label{eq:jrec2}\end{aligned}$$ P\_n\^[(0,0)]{}() + P\_[n+1]{}\^[(0,0)]{}() &=& (1+)P\^[(0,1)]{}\_n(), \[A1\]\
P\_n\^[(0,0)]{}() - P\_[n+1]{}\^[(0,0)]{}() &=& (1-)P\^[(1,0)]{}\_n(). \[A2\] Relations between Jacobi and Gegenbauer polynomials : $$\begin{aligned}
(1+\xi) P_{n}^{(0,1)}(\xi) + (1-\xi)P_{n}^{(1,0)}(\xi) &=&
2 C_{n}^{1/2}(\xi),
\label{eq:jg2}\\
(1+\xi) P_{n}^{(0,1)}(\xi) - (1-\xi)P_{n}^{(1,0)}(\xi) &=&
2 C_{n+1}^{1/2}(\xi),
\label{eq:jg3}\end{aligned}$$ $$(n+2)P_{n}^{(1,1)}(\xi) = 2 C_{n}^{3/2}(\xi).
\label{eq:jg4}$$ Orthogonality relations for Appell polynomials [@Er] : $$\int{\cal D}\underline{\alpha} \: \alpha_{d}\alpha_{u}\alpha_{g}^{2}
J_{k,l}(\alpha_{d}, \alpha_{u})
J_{m,n}(\alpha_{d}, \alpha_{u})
= \delta_{k+l, m+n}
\frac{(-1)^{k+l}}{2^{k+l+3}(k+l+3)(2k+2l+5)!!}
W_{k,m}^{(k+l+1)},
\label{eq:appello0}$$ where $W_{k,m}^{(k+l+1)} \equiv
\partial^{m+n}J_{k,l}(\alpha_{d}, \alpha_{u})/
\partial \alpha_{d}^{m} \partial \alpha_{u}^{n}$ is a $(k+l+1)\times (k+l+1)$ symmetric matrix. This result can be obtained from the following relations: $$\begin{aligned}
\lefteqn{\int {\cal D}
\underline{\alpha}\: \alpha_{d}^{m+1}\alpha_{u}^{n+1}
\alpha_{g}^{2} J_{k,l}(\alpha_{d}, \alpha_{u}) =}\makebox[2cm]{\ }
\nonumber \\
&=&\left\{
\begin{array}{@{\,}ll}
0& (m+n < k+l)\\
\delta_{m,k}
\displaystyle{\frac{(-1)^{k+l}k!l!}{2^{k+l+3}(k+l+3)(2k+2l+5)!!}}&
(m+n=k+l),
\end{array}
\right.
\label{eq:appello}\end{aligned}$$ while the integral is in general nonzero for $m+n>k+l$.
Integral formulae for Appell polynomials : $$\begin{aligned}
&\mbox{$ $}&\frac{d}{du}\int_{0}^{u}\!d\alpha_{d} \int_{0}^{\overline{u}}
\!d\alpha_{u}\,
\frac{1}{1 -\alpha_{d}-\alpha_{u}}\left(
\alpha_{d}\frac{\partial}{\partial \alpha_{d}}
+\alpha_{u}\frac{\partial}{\partial \alpha_{u}}
- 1\right) \alpha_{d} \alpha_{u} (1 - \alpha_{d}-\alpha_{u})^{2}
J_{k,l}(\alpha_{d}, \alpha_{u}) =
\nonumber\\
&\mbox{$ $}&\;\;\;\;\;\;\;\;\;\;\;\;
= \frac{u\overline{u}}{2}\frac{k!l!(-1)^{k}}{(k+l+2)!}
\left(\frac{k-l}{k+l+3}P_{k+l+2}^{(1,1)}(\xi)
+ P_{k+l+1}^{(1,1)}(\xi)\right),
\label{eq:apint}\end{aligned}$$ $$\begin{aligned}
\lefteqn{\frac{d}{du}\int_{0}^{u}d\alpha_{d} \int_{0}^{\overline{u}}
d\alpha_{u}
\alpha_{d} \alpha_{u} (1 - \alpha_{d}-\alpha_{u})
J_{k,l}(\alpha_{d}, \alpha_{u}) =}\makebox[1cm]{\ }
\nonumber\\
&=& \frac{u\overline{u}}{2}\frac{k!l!(-1)^{k}}{(k+l+3)!}
\left(\frac{k-l}{k+l+3}P_{k+l+2}^{(1,1)}(\xi)
- P_{k+l+1}^{(1,1)}(\xi)\right),
\label{eq:apint2}\end{aligned}$$ & &[ddu]{} \_0\^ud\_d \_0\^[|[u]{}]{}d\_u [11- \_d -\_u]{} ( \_d [\_d ]{} - \_u[\_u]{}) \_d\_u (1-\_d-\_u)\^2 J\_[k,l]{}(\_d,\_u) =\
& &=[u|[u]{}2]{}[k!l!(-1)\^k(k+l+3)!]{} . \[A3\] & &\_0\^ud\_d \_0\^[|[u]{}]{}d\_u [11- \_d -\_u]{} ( [\_d ]{} + [\_u]{}) \_d\_u (1-\_d-\_u)\^2 J\_[k,l]{}(\_d,\_u) =\
& &=[k!l!(-1)\^k(k+l+3)!4]{} . \[A4\] The results (\[eq:apint\])$-$(\[A3\]) can be obtained by differentiating and/or integrating the Appell polynomials $J_{k,l}(\alpha_{d}, \alpha_{u})$ term by term. To obtain (\[A4\]), it is convenient to calculate its derivative first, which can be done similarly to (\[eq:apint\]), (\[eq:apint2\]) and (\[A3\]), and then integrate the result with the condition that it vanishes at $u=0$.
Operators relations for different theoretical twist \[app:oprel\]
=================================================================
This appendix contains some relations taken from [@BaBr], [@BB96] and [@BBK], which help to isolate the different twist parts of two-points non-local operators.
Chiral-even operator
--------------------
Consider the non-local operators $\ov{\psi}(x)\gamma_\alpha\psi(-x)$. The local expansion around $x=0$ can be written in the following way : (x)\_(-x)=\_[n=0]{}\^x\_[\_1]{}…x\_[\_n]{} \_[\_1]{}…\_[\_n]{}\_
In order to get the leading twist contribution, the symmetrization has to be done over all indices and subtract the traces : \_\_[n=0]{}\^x\_[\_1]{}…x\_[\_n]{} (\_[\_1]{}…\_[\_n]{}\_) \_-
The symmetrization has an integrated solution : $$\begin{aligned}
\lefteqn{\Big[\bar \psi(x)\gamma_\alpha \psi(-x)\Big]_{\text{sym}} \equiv}
\nonumber\\
&\equiv &
\sum_{n=0}^\infty \frac{x^{\mu_1}\ldots x^{\mu_n}}{n!}
\bar \psi(0)\Bigg\{\frac{1}{n+1}\Drl_{\mu_1}\ldots \Drl_{\mu_n}\gamma_\alpha
+\frac{n}{n+1}\Drl_{\alpha}\Drl_{\mu_1}\ldots \Drl_{\mu_{n-1}}\gamma_{\mu_n}
\Bigg\} \psi(0)\,.\makebox[1cm]{\ } \nonumber \\
\label{defsym} \\
& =&
\int_0^1 dv\,\frac{\partial}{\partial x_\alpha}\bar \psi(vx)\hat{x}
\psi(-v x)
\label{BBformula}
\eq
with $\hat{x}\equiv\gamma_\mu x^\mu$.
In order to have really the leading twist part of $\ov{\psi}(x)\gamma_\alpha\psi(-x)$, the leading twist part of $\bar \psi(x)\hat{x} \psi(-x)$ is needed. If one does a local expansion of this latter non-local operator, one will already have symetrized local operators. But traces has alos to be removed; this can be expressed as a differential equation :
\be
\frac{\partial}{\partial x_\alpha\partial x^\alpha}\left[\ov{\psi}(x)\hat{x}
\psi(-x)\right]_{\text{l. t.}}=0 \label{ltxhatapp}
\ee
which has the formal solution :
\bq
\left[\ov{\psi}(x)\hat{x}\psi(-x)\right]_{\text{l. t.}}&=&\ov{\psi}(x)\hat{x}\psi(-x)
\nonumber \\
& &{}+\sum_{n=1}^\infty\int_0^1\left(-\frac{1}{4}x^2\right)^n
\frac{\left(\frac{1-t}{t}\right)^{n-1}}{(n-1)!n!}
\left(\frac{\partial^2}{\partial x^\alpha\partial x_\alpha}\right)^n
\ov{\psi}(tx)\hat{x}\psi(-tx) \nonumber \\ \label{solltxhat}
\eq
Equations for the twist 3 part of $\ov \psi(x)\gamma_\alpha\psi(-x)$ are the following :
\begin{eqnarray}
\Big[\bar \psi(x)\gamma_\mu \psi(-x)\Big]_{\text{twist 3}} &=&
{}-g_s\int_0^1 \!du\int_{-u}^u \!dv\,\bar \psi(ux)
\left[
u\tilde G_{\mu\nu}(vx)x^\nu \hat{x}\gamma_5\right. \nonumber \\
& &{} -ivG_{\mu\nu}(vx)x^\nu \hat{x}
\Big]\psi(-ux)
\nonumber\\
&&{}+i\epsilon_{\mu}^{\phantom{\mu}\nu\alpha\beta}\int_0^1 udu\,
x_\nu\partial_\alpha
\Big[\bar \psi(ux)\gamma_\beta\gamma_5 \psi(-ux)\Big]\,,
\nonumber\\
\Big[\bar \psi(x)\gamma_\mu\gamma_5 \psi(-x)\Big]_{\text{twist 3}} &=&
-g_s\int_0^1 \!du\int_{-u}^u \!dv\,\bar \psi(ux)
\left[
u\tilde G_{\mu\nu}(vx)x^\nu\hat{x}\right. \nonumber \\ & &{}-ivG_{\mu\nu}(vx)x^\nu\hat{x}\gamma_5
\Big]\psi(-ux)
\nonumber\\
&&{}+i\epsilon_{\mu}^{\phantom{\mu}\nu\alpha\beta}\int_0^1 udu\,
x_\nu\partial_\alpha
\Big[\bar \psi(ux)\gamma_\beta \psi(-ux)\Big]\,,
\label{twist3}\end{aligned}$$ where $G_{\mu\nu}$ is the gluon field strength, $\tilde G_{\mu\nu}
=(1/2)\epsilon_{\mu\nu\alpha\beta}G^{\alpha\beta}$, and $\partial_\alpha$ is the derivative over the total translation : $$\partial_\alpha\Big[\bar \psi(ux)\gamma_\beta \psi(-ux)\Big] \equiv
\left.\frac{\partial}{\partial y_\alpha}
\Big[\bar \psi(ux+y)\gamma_\beta \psi(-ux+y)\Big]\right|_{y\to 0}. \label{totder}$$
These equations come from [@BB96] and can be obtained with the equation of motion, neglecting quark masses.
Another useful formula is the equation (5.13) in [@BaBr] which can be written like this : (x)(-x)= -\^2(x)(-x) +
The expansion of \[solltxhat\] at the order $x^2$, gives (x)(-x)&=&\_ +x\^2\_0\^1dv(vx)(-vx) + …\
&=&\_ -x\^2\_0\^1dv v\^2\^2 (vx)(-vx)\
& &+ ++…\[dertot\]
This last equation gives a twist 4 contribution with a total derivative (defined in \[totder\]).
Chiral-odd operator
-------------------
Consider the non-local operators $\ov{\psi}(x)\sigma_{\alpha\beta}\psi(-x)$. To get the most symmetrical part in the local expansion, one can compare it with the solution in \[BBformula\] and get directly : $$\begin{aligned}
\lefteqn{\Big[\bar \psi(x)\sigma_{\alpha\beta} \psi(-x)\Big]_{\text{sym}} =} \nonumber
\\
& =&
\int_0^1 dv\,\frac{\partial}{\partial x_\beta}\bar \psi(vx)
\sigma_{\alpha\nu}x^\nu
\psi(-v x)-(\alpha \leftrightarrow \beta)
\label{BBformulasig}
\eq
In \cite{BBK}, the equations for the leading twist part of $\bar \psi(x)\sigma_{\alpha\nu}x^\nu \psi(-x)$ can be founds~:
\bq
\frac{\partial}{\partial x_\alpha}\left[\bar \psi(x)\sigma_{\alpha\nu}x^\nu \psi(-x)\right]_{\text{l. t.}}&=&0 \\
\frac{\partial^2}{\partial x_\rho \partial x^\rho}\left[\bar \psi(x)\sigma_{\alpha\nu}x^\nu \psi(-x)\right]_{\text{l. t.}}&=&0
\eq
Like in equation \ref{dertot}, the contribution of total derivative can be added :
\bq
\bar \psi(x)\sigma_{\alpha\nu}x^\nu \psi(-x)&=&
\left[\bar \psi(x)\sigma_{\alpha\nu}x^\nu \psi(-x)\right]_{\text{l. t.}} -\frac{1}{4} x^2x^\nu\int_0^1 dt\,t^2 \partial^2\left(\bar \psi(tx)\sigma_{\alpha\nu}\psi(-tx)\right) \nonumber \\
& &{}+\text{operators with gluons}+\ldots
\eq
{\setcounter{equation}{0}\section{Mass corrections to two-points distributions \label{app:masscorr}}}
In this appendix, the mass corrections for the distributions are given, using the method described in section \ref{thtmasscorr}.
\subsection{Chiral even distributions}
For the different wave functions
\begin{eqnarray}
\lefteqn{\langle 0|\bar u(x) \gamma_{\mu} [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle = }\makebox[2cm]{\ } \nonumber \\
&=& f_{\rho} m_{\rho} \left[ P_{\mu}
\frac{e^{(\lambda)}\cdot x}{P \cdot x}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(A^{(e)}(u)+
\frac{x^2 m_\rho^2}{4}\hat{A}^{(e)}(u)\right) \right.
\nonumber \\
& &{}+ \left(e^{(\lambda)}_{\mu}-P_\mu\frac{\epl\cdot x}{P\cdot x}\right)
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(B^{(e)}(u)
+\frac{x^2 m_\rho^2}{4}\hat{B}^{(e)}(u)\right)\nonumber \\
& & \left.{}- \frac{1}{2}x_{\mu}
\frac{e^{(\lambda)}\cdot x }{(P \cdot x)^{2}} m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} C^{(e)}(u) \right]\end{aligned}$$ one has \_0\^1duB\^[(e)]{}(u)e\^[iPx]{}&=&\_0\^1du A[(e)]{}(u)\_0\^1dte\^[itPx]{}\
& &-(Px)\^2\_0\^1du\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}B\^[(e)]{}(u)\
\
-\_0\^1duC\^[(e)]{}(u)e\^[iPx]{}&=&\_0\^1duA\^[(e)]{}(u)\_0\^1dtt\^2\_0\^1dvv\^2e\^[ivtPx]{}\
& &-\_0\^1du B\^[(e)]{}(u)\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}\
\
\_0\^1du\^[(e)]{}(u)e\^[iPx]{}&=& \_0\^1duA\^[(e)]{}(u)(\^2+1)\_0\^1dtt\^2\_0\^1v\^2dve\^[ivtPx]{}\
& &+4\_0\^1duB\^[(e)]{}(u)\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}\
& &-(Px)\^2\_0\^1du\^[(e)]{}(u)\_0\^1dtt\^3 \_0\^1dvv\^3e\^[itvPx]{}\
\
& & \_0\^1duA\^[(e)]{}(u)(\^2+1)\_0\^1dtt\^2\_0\^1dvv\^2(i(Px)+1) t\^2e\^[ivtPx]{}\
& &+4\_0\^1duB\^[(e)]{}(u)\_0\^1dtt\_0\^1dvv\^2e\^[ivtPx]{}\
Chiral odd distributions
------------------------
For the different wave functions $$\begin{aligned}
\lefteqn{\hspace*{-1.5cm}\langle 0|\bar u(x) \sigma_{\mu \nu} [x,-x]
d(-x)|\rho^-(P,\lambda)\rangle =} \nonumber \\
&=& i f_{\rho}^{T} \left[ ( e^{(\lambda)}_{\mu}P_\nu -
e^{(\lambda)}_{\nu}P_\mu )
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} \left(A^{(o)}(u)
+\frac{m^2_\rho x^2}{4}\hat{A}^{(o)}(u)\right)\right.
\nonumber \\
& &{}+ (P_\mu x_\nu - P_\nu x_\mu )
\frac{e^{(\lambda)} \cdot x}{(P \cdot x)^{2}}m_{\rho}^{2}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x}\left(B^{(o)}(u)
+\frac{m^2_\rho x^2}{4}\hat{B}^{(o)}(u)\right)
\nonumber \\
& & \left.{}+ \frac{1}{2}
(e^{(\lambda)}_{\mu} x_\nu -e^{(\lambda)}_{\nu} x_\mu)
\frac{m_{\rho}^{2}}{P \cdot x}
\int_{0}^{1} \!du\, e^{i \xi P \cdot x} C^{(o)}(u) \right],\end{aligned}$$ one has\
&=&\_u\^1duE(u)\_0\^1dt\_0\^1dvv\^3e\^[ivtPx]{}\
\
&=&(Px)\_0\^1duE(u)\_0\^1dt\_0\^1dvv\^2e\^[itvPx]{}\
\
&=&(Px )\_0\^1duE(u)\_0\^1dt\_0\^1dvv\^2e\^[itvPx]{} where E(u)iA\^[(o)]{}(u)
There is no equation for $\hat{B}^{(e)}(u)$ because it is already a $O(m^4_\rho)$ contribution.
[999]{}
S. J. Brodsky and G. P. Lepage, Phys. Lett. [**B87**]{} (1979) 359.
S. J. Brodsky and G. P. Lepage, in: [*Perturbative Quantum Chromodynamics*]{}, ed. by A. H. Mueller, p. 93, World Scientific (Singapore) 1989.
S. J. Brodsky, H.-C. Pauli and S. S. Pinsky, Phys. Rept. [**301**]{} (1998) 299.
M. A. Shifman, A. I. Vainshtein and Zakharov, Nucl. Phys. [**B147**]{} (1979) 385.
V. L. Chernyak and A. R. Zhitnitsky, Phys. Rep. [**112**]{} (1984) 173.
V. M. Braun and I. E. Filyanov, Z. Phys. [**C48**]{} (1990) 239.
I. I. Balitskii, V. M. Braun and A. V. Kolesnichenko, Nucl. Phys. [**B312**]{} (1989) 509.
P. Ball, V. M. Braun, Y. Koike and K. Tanaka, hep-ph/9802299
R. A. Brandt and G. Preparata, Nucl. Phys. [**27**]{} (1971) 541-567.
R. L. Jaffe and X. Ji, Nucl. Phys. [**B375**]{} (1992) 527.
P. Ball and V. M. Braun, hep-ph/9810475.
Th. Ohrndorf, Nucl. Phys. [**B198**]{} (1982) 26.
S. Sakar, Nucl. Phys. [**B83**]{} (1974) 108.
G. Mack and A. Salam, Ann. of Phys. [**53**]{} (1969) 174.
Yu. M. Makeenko, Yad. Fiz. [**33**]{} (1981) 842.
A. Erdélyi et al., [sl Higher Transcendental Functions]{}, vol. II (McGraw-Hill, New York, 1953).
V.M. Braun and A.V. Kolesnichenko, Phys. Lett. B [**175**]{} (1986) 485; Sov. J. Nucl. Phys. [**44**]{} (1986) 489.
P. Ball, V. M. Braun, hep-ph/9805422
I. I. Balitsky and V. M. Braun, Nucl. Phys. [**B311**]{} (1989) 541.
P. Ball and V. M. Braun, phys. Rev. [**D54**]{} (1996) 2182.
V. M. Braun, hep-ph/9801222
I. I. Balitsky, Phys. Lett. [**124**]{} (1983) 230.
A. Ali and V. M. Braun, Phys. Lett. [**B359**]{} (1995) 223.
A.R. Zhitnitsky, I.R. Zhitnitsky and V.L. Chernyak, Sov. J. Nucl. Phys. [**41**]{} (1985) 284.
[^1]: Unité Propre de Recherche 7061
[^2]: In the notations of Ref. [@BBKT], $\omega_{1,0}^A\equiv \omega_3^A$, $ \zeta_3^A\equiv \zeta_3$ and $\zeta_3^V \equiv (3/28)\zeta_3\omega_3^V$.
[^3]: In notations of Ref. [@BBKT] $\zeta_3^T \equiv (3/28)\zeta_3\omega_3^T$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Social dynamic opinion models have been widely studied to understand how interactions among individuals cause opinions to evolve. Most opinion models that utilize spin interaction models usually produce a consensus steady state in which only one opinion exists. Because in reality different opinions usually coexist, we focus on non-consensus opinion models in which above a certain threshold two opinions coexist in a stable relationship. We revisit and extend the non-consensus opinion (NCO) model introduced by Shao [*et al.*]{}[@NCO]. The NCO model in random networks displays a second order phase transition that belongs to regular mean field percolation and is characterized by the appearance (above a certain threshold) of a large spanning cluster of the minority opinion. We generalize the NCO model by adding a weight factor $W$ to individual’s own opinion when determining its future opinion (NCO$W$ model). We find that as $W$ increases the minority opinion holders tend to form stable clusters with a smaller initial minority fraction compared to the NCO model. We also revisit another non-consensus opinion model based on the NCO model, the inflexible contrarian opinion (ICO) model [@ICO], which introduces inflexible contrarians to model a competition between two opinions in the steady state. Inflexible contrarians are individuals that never change their own opinion but may influence opinions of others. To place the inflexible contarians in the ICO model we use two different strategies, random placement and one in which high-degree nodes are targeted. In both strategies, the inflexible contrarians effectively decrease the size of the largest cluster of the rival opinion but the effect is more pronounced under the targeted method. All of the above models have previously been explored in terms of a single network. However human communities are rarely isolated, instead are usually interconnected. Because opinions propagate not only within single networks but also between networks, and because the rules of opinion formation within a network may differ from those between networks, we study here the opinion dynamics in coupled networks. Each network represents a social group or community and the interdependent links joining individuals from different networks may be social ties that are unusually strong, e.g., married couples. We apply the non-consensus opinion (NCO) rule on each individual network and the global majority rule on interdependent pairs such that two interdependent agents with different opinions will, due to the influence of mass media, follow the majority opinion of the entire population. The opinion interactions within each network and the interdependent links across networks interlace periodically until a steady state is reached. We find that the interdependent links effectively force the system from a second order phase transition, which is characteristic of the NCO model on a single network, to a hybrid phase transition, i.e., a mix of second-order and abrupt jump-like transitions that ultimately becomes, as we increase the percentage of interdependent agents, a pure abrupt transition. We conclude that for the NCO model on coupled networks, interactions through interdependent links could push the non-consensus opinion type model to a consensus opinion type model, which mimics the reality that increased mass communication causes people to hold opinions that are increasingly similar. We also find that the effect of interdependent links is more pronounced in interdependent scale free networks than in interdependent Erdös Rényi networks.'
author:
- Qian Li$^1$
- 'Lidia A. Braunstein$^{2,1}$'
- 'Huijuan Wang$^{3,1}$'
- Jia Shao$^1$
- 'H. Eugene Stanley$^1$'
- Shlomo Havlin$^4$
date: '10 September 2012 — lbwssh10sep.tex'
title: 'Non-consensus opinion models on complex networks'
---
Contribution to the special issue: Statistical Mechanics and Social Sciences edited by Sidney Redner.
Introduction
============
Statistical physics methods have been successfully applied to understand the cooperative behavior of complex interactions between microscopic entities at a macroscopic level. In recent decades many research fields, such as biology, ecology, economics, and sociology, have used concepts and tools from statistical mechanics to better understand the collective behavior of different systems either in individual scientific fields or in some combination of interdisciplinary fields. Recently the application of statistical physics to social phenomena, and opinion dynamics in particular, has attracted the attention of an increasing number of physicists. Statistical physics can be used to explore an important question in opinion dynamics: how can interactions between individuals create order in a situation that is initially disordered? Order in this social science context means agreement, and disorder means disagreement. The transition from a disordered state to a macroscopic ordered state is a familiar territory in traditional statistical physics, and tools such as Ising spin models are often used to explore this kind of transition. Another significant aspect present in social dynamics is the topology of the substrate in which a process evolves. This topology describes the relationships between individuals by identifying, e.g., friendship pairs and interaction frequencies. Researchers have mapped the topology of social connections onto complex networks in which the nodes represent agents and the links represent the interactions between agents [@socialCastellano; @socialGalam; @Boc_01; @Dor_02; @Pas_01; @ER1; @ER1959; @Bollobas; @SF1; @Cohen; @SF; @Complexnetwork; @Newmanbook]. Various versions of opinion models based on spin models have been proposed and studied, such as the Sznajd model [@Sznajd], the voter model [@voter1; @voter2], the majority rule model [@majority1; @majority2], and the social impact model [@social1; @social2].
Almost all spin-like opinion models mentioned above are based on short-range interactions that reach an ordered steady state, with a consensus opinion that can be described as a consensus opinion model. However, in real life different opinions are mostly present and coexist. In a presidential election in a country with two political parties in which each party has its own candidate, for example, a majority opinion and a minority opinion coexist. The opinions among the voters differ, with one fraction of the voters supporting one candidate and the rest supporting the other, and rarely will the two opinions reach consensus. This reality has motivated scientists to explore opinion models that are more realistic, ones in which two opinions can stably coexist. Shao [*et al.*]{} [@NCO] proposed a nonconsensus opinion (NCO) model that achieves a steady state with two opinions coexisting. Unlike the majority rule model and the voter model in which the dynamic of an agent’s opinion is not influenced by the agent’s own current opinion but only by its neighbors, the NCO model assumes that during the opinion formation process an agent’s opinion is influenced by [*both*]{} its own current opinion and the opinions of friends, modeled as nearest-neighbors in a network. This NCO model begins with a disordered state with a fraction $f$ of $\sigma_+$ opinion and a fraction $1-f$ of $\sigma_-$ opinion distributed randomly on the nodes of a network. Through interactions the two opinions compete and reach a non-consensus stable state with clusters of $\sigma_+$ and $\sigma_-$ opinions. In the NCO model, at each time step each node adopts the majority opinion of its “neighborhood”, which consists of the node’s nearest neighbors and itself. When there is a tie, the node does not change its opinion. The NCO model takes each node’s own current opinion into consideration, and this is a critical condition for reaching a nonconsensus steady state. Beginning with a random initial condition, this novel nontrivial stable state in which both majority and minority opinions coexist is achieved after a relatively short sequence of time steps in the dynamic process. The NCO model has a smooth phase transition with the control parameter $f$. Below a critical threshold $f_c$, only the majority opinion exists. Above $f_c$, minorities can form large spanning clusters across the total population of size $N$. Using simulations, Shao [*et al.*]{} [@NCO] suggested that the smooth phase transition in the NCO model in random networks is of the same universality class as regular mean field (MF) percolation. But simulations of the NCO model in Euclidean lattices suggest that the process might belong to the universality class of invasion percolation with trapping (TIP) [@NCO; @NCOPRESS]. Apparently this is the first time, to the best of our knowledge, that a social dynamic model has been mapped to percolation, an important tool in statistical physics. However, the nature of this percolation on $2D$ lattice is still under debate [@NCOPRESS; @Grassberger]. Exact solutions of the NCO model in one dimension and in a Cayley tree have been developed by Ben-Avraham [@NCO_solution].
Here we present simulations suggesting that the behavior of the NCO model, in which two opinions coexist, disappears when the average network degree increases. When the average degree of a network is high, the agent’s own opinion becomes less effective and the NCO model converges to the majority voter model. This was argued analytically by Roca [@Percolate_or_die] and claimed also by Sattari [*et al*]{} [@Grassberger]. In the present paper, we also generalize the NCO model and create a nonconsensus opinion model by adding a weight (NCO$W$ model) to an agent’s own opinion. The weight $W \ge 1$ represents the strength of an individual’s own opinion. Note that in the NCO model $W$ is assumed to be $1$ like the weight of its neighbors opinion. We find that the NCO$W$ model inherits all the features of the NCO model, except that the critical threshold $f_c$ of the NCO$W$ model with $W>1$ decreases when $W$ increases. This means that strengthening one’s own opinion helps smaller minority opinion groups to survive.
The NCO model reaches a steady state in which the two opinions coexist. This is only partially realistic. In real life, two opinions do not simply coexist—they continue to compete. Real-world examples include the decades-long competition between the Windows and Macintosh operating systems and between Republicans and Democrats in US presidential politics. All the participants in these competitions have the same goal: winning. In order to increase their prospects of winning, they need as many supporters (or customers) as possible. Thus, it is interesting to study how two opinions continue to compete after they have reached a steady state. In order to consider both aspects, the nonconsensus steady state and the competition, Li [*et al.*]{} [@ICO] proposed an inflexible contrarian opinion (ICO) model in which a fraction $\phi$ of inflexible contrarians are introduced into the final steady state of the NCO model and two different competition strategies are then applied. The concepts of inflexible agents and contrarian agents were introduced by Galam [@contrarian; @galam1] in his work on opinion models. In the ICO model, an inflexible contrarian is an agent that holds an opinion contrary to that held by the majority of its surrounding group and its opinion is not influenced by its surrounding group—it never changes. Inflexible contrarians have one goal: to change the opinion of the current supporters in the rival group. We see this strategy when, for example, companies send a free product to potential customers in order to convince them to adopt the product and influence their friends to do the same. We study the ICO model in order to determine, for example, whether these free products actually do help to win the competition, how many free products are needed to be sent, and who are the best candidates to receive the free product. Reference [@ICO] presents two strategies for introducing inflexible contrarians into the steady state of the $\sigma_+$ opinion groups: (i) the random strategy and (ii) the targeted (high degree) strategy. Using these strategies, we find that the relative size of the largest cluster in state $\sigma_+$ undergoes a second-order phase transition at a critical fraction of inflexible contrarians $\phi_c$ below which the two opinions can coexist and above which only $\sigma_-$ exists. Thus the ICO also belongs to the type of nonconsensus opinion models. The results also indicate that the largest cluster in state $\sigma_+$ undergoes a second order phase transition that can be mapped into MF percolation similar to the NCO model.
All opinion models described above have been studied on a single network. However, in real social opinion dynamics, individuals belonging to different social communities can as well communicate. In a traditional agrarian village, for example, two separate working relationship networks often form. Men work in the fields with other men and women work in their homes with other women. Marriages between men and women in this setting create interdependencies between the two working relationship networks. As far as we know, there has been no model study of how this kind of strong social connection between two such different groups influences the exchange of opinions. In studying the opinion dynamics across different groups we utilize a concept that has recently gained wide attention: the resilience of interdependent networks to cascading failures [@couple; @parshani; @gaonature; @gaoprl; @weiprl; @Bashan; @Leicht; @Brummitt]. Connecting two networks together with interdependent links allows individuals to exchange opinions between networks. In our model, two nodes from different networks that are connected by interdependent links represent a pair of nodes that have strong social relations. In interdependent networks we usually distinguish between the connectivity links between agents within each network or community and interdependent links between agents from different networks.
To study the effect of interdependent links on opinion dynamics, we propose a non-consensus opinion (NCO) model on coupled networks in which we assume different opinion formation rules for internal connectivity and interdependent links. We assume that during the dynamic process of opinion formation the agents that are connected with interdependent links will have the same opinion, this being the case because their social relationship is strong. In our model, the NCO rules are applied in each individual network. For the coupled pairs the following rule is applied: if two interdependent nodes have the same opinion, they will keep this opinion, but if they have different opinions, they will follow the majority opinion of the interdependent network system (global majority rule). Many other possible rules could be tested for the interdependent pairs, but we adopt here, for simplicity, the majority rule. When an opinion is shared by two interdependent individuals, such as a married couple, because their social relationship is strong and close, they will tend to maintain their opinion against outside influence. If their opinions differ initially, they tend to eventually resolve their differences and share the same opinion. In the process of resolving their differences, however, they can be significantly influenced by outside forces, e.g., mass media, and thus we assume that they often end up sharing the majority opinion. When we increase the number of interdependent links between the coupled networks, the transition changes from a pure second order phase transition to a hybrid phase transition and finally to a seemingly abrupt transition. The hybrid transition contains both a second order and an abrupt transition. The model type of the NCO model on coupled networks also changes as the number of interdependent links increases, and thus the system goes from being a kind of nonconsensus opinion model to being a kind of consensus opinion model. This suggests that strong interactions between different social groups is pushing our world in the direction of becoming more uniform in their opinions.
The paper is organized as follows, in Sec. \[percolation\] we revisit some important concepts on the topology of opinion clusters and percolation. We then present the results and discussions on NCO and NCO$W$ model in Sec. \[NCO\], on ICO model in Sec. \[ICO\] and on NCO model on coupled networks in Sec. \[NCOCOUPLE\]. Finally, we present our summary in Sec. \[conclusion\].
Topology of Opinion Clusters and Percolation {#percolation}
============================================
In recent decades, many researchers have studied how network topology affects the processes that evolve in them. Examples of such processes are the spreading of rumors, opinions, diseases, and percolation [@Complexnetwork; @Boc_01; @Dor_02; @Pas_01; @Vespignaniprl2001; @LewiStone; @KitsakNatureP; @Cohenprl2003; @Braunsteinprl2003; @quenched_shlomo; @percolation]. Classical percolation processes deal with the random failure of nodes (or links) and present a geometrical second order phase transition with a control parameter $p$ that represents the fraction of nodes (or links) remaining after a random failure of a fraction $1-p$ of nodes (or links).
There exists a critical probability $p_c$ above which a “giant component” (GC) appears. The number of nodes in the GC, $S_1$, is called the order parameter of the phase transition. Below criticality there is no GC and only finite clusters exist. For $p<p_c$ the size distribution of the clusters is $n_s \sim s^{- \tau}$ with a cutoff that diverges when approaching $p_c$. At criticality, in the thermodynamic limit, the size of the second largest component $S_2$ diverges at $p_c$ as $S_2 \sim |p-p_c|^{- \gamma}$ just as the susceptibility with the distance to the critical temperature. For large networks ($N \to \infty$), $p_c = 1/(\kappa-1)$, where $\kappa$ is the branching factor given by $\kappa= \langle k^2 \rangle /\langle
k \rangle$, where $\langle k \rangle$ and $\langle k^2 \rangle$ are the first and second moments of the degree distribution $P(k)$ of the network respectively [@Cohen]. We perform all our simulations on both Erdös-Rényi (ER) networks [@ER1; @ER1959; @Bollobas] and scale-free (SF) networks [@SF1]. ER networks are characterized by a Poisson degree distribution, $P(k)=e^{- \langle k
\rangle} \langle k \rangle^k /k!$. In SF networks the degree distribution is given by a power law, $P(k)\sim k^{-\lambda}$, for $k_{\rm min}\leq k \leq k_{\rm max}$, where $k_{\rm min}$ is the lowest degree of the network and $k_{\rm max}$ is the highest degree of the network. For random SF networks $k_{\rm max} \sim
N^{1/(\lambda-1)}$ is the degree cutoff [@Cohen], where $N$ is the system size and $\lambda$ is the broadness of the distribution.
We begin by examining percolation in ER networks. At criticality, percolation in ER networks is equivalent to percolation on a Cayley tree or percolation at the upper critical dimension $d_c=6$ where all the exponents have mean field (MF) values with $\tau=5/2$ and $\gamma=1$. Note that in the ER case, the mass of the incipient infinite cluster $S_1$ scales as $N^{2/3}$ at criticality. We can understand this result by using the framework of percolation theory for the upper critical dimension $d_c=6$. Since $S_1\sim R^{d_f}$ and $N\sim
R^d$ (where $d$ is the dimension of the initial lattice, $d_f$ the fractal dimension, and $R$ the spatial diameter of the cluster), it follows that $S_1\sim N^{d_f/d_c}$ and since $d_c=6$ and $d_f=4$ we obtain $S_1\sim N^{2/3}$ [@Braunsteinprl2003].
For SF networks, the GC at criticality is $S_1\sim N^{2/3}$ for $\lambda>4$, and $S_1\sim N^{(\lambda-2)/(\lambda-1)}$ for $3<\lambda\leq 4$ [@cohenpre2002]. For SF networks, with $\lambda <
3$, $\langle k^2 \rangle \to \infty$ when $N \to \infty$ because $k_{\rm
max} \to \infty$ and thus $p_c=0$, making these networks extremely robust against random failures [@Cohen].
However if we decrease $k_{\rm max}$ by targeting and removing the highest degree nodes (hubs), $p_c$ is finite [@cohenprl2001] and we recover a second order phase transition with MF exponents as for ER networks. We will show below that this is also true for our model here. A similar MF behavior in SF networks with $\lambda < 3$ was found also by Valdez [*et al.*]{} [@L.Valdez] for the percolation of susceptible clusters during the spread of an epidemic. In our simulations we always choose $k_{\rm min} = 2$ for SF networks in order to ensure that they are almost fully connected [@Cohen].
The NCO Model {#NCO}
=============
In the NCO model [@NCO] on a single network with $N$ nodes, opinion $\sigma_+$ and $\sigma_-$ are initially randomly assigned to each node with a fraction of $f$ and $1-f$ respectively. The basic assumption of the NCO model is that the opinion of an agent is influenced by both its own opinion and the opinions of its nearest neighbors (the agent’s friends). The opinion formation rule states that at each time step, each node adopts the majority opinion, which includes [*both*]{} the opinions of its neighbors and itself. If there is a tie, the node’s opinion will remain unchanged. Using this rule, each node is tested at each simulation step to see whether its opinion has changed. All these updates are performed simultaneously and in parallel until no more changes occur and a steady state is reached.
Figure \[fig:NCOf1\] demonstrates the dynamic behavior of the NCO model on a small network with nine nodes. At time $t=0$, five nodes are randomly assigned opinion $\sigma_+$ (empty circle), and the remaining four, opinion $\sigma_-$ (solid circle). After checking the status of each node, we find that only node A belongs to a local minority with opinion $\sigma_+$, so at the end of this time step, node A changes its opinion to $\sigma_-$. At time t=1 only node B belongs to a local minority, so at the end of this time step, the opinion of node B will be updated to $\sigma_-$. At time t=3, every node has the same opinion as its local majority, where the final nonconsensus steady state is reached.
Simulation Results
------------------
In the steady state $s_1=S_1/N$ is the normalized size of the largest opinion $\sigma_+$ cluster, $s_2=S_2/N$ is the normalized size of the second largest opinion $\sigma_+$ cluster, and $F$ is the normalized fraction of opinion $\sigma_+$ nodes. Figure \[fig:NCOf2\] shows plots of $s_1$, $s_2$, and $F$ as a function of the initial fraction $f$ of the opinion $\sigma_+$ nodes for both ER and SF networks. We find that due to the symmetrical status of both opinions, $F$ is a monotonically increasing function of $f$ with symmetry around $f=0.5$. Figure \[fig:NCOf2\] also shows the emergence of a second order phase transition. Note that there is a critical threshold $f_c$, which is characterized by the sharp peak of $s_2$. Below $f_c$, $s_1$ approaches zero, where only the majority opinion can form steady clusters, and above $f_c$, $s_1$ increases as $f$ increases and a state with stable coexistence of both majority and minority opinion clusters appears. Although for both ER and SF networks $f_c<0.5$ as expected, ER networks have smaller values of $f_c$ than SF networks for the same average degree $\langle k\rangle$. For example for SF networks with $k_{\rm min}=2$ and $\lambda=2.5$ where $\langle
k\rangle \approx 5.5$, $f_c \approx 0.45$, while for the same average degree $\langle k\rangle = 5.5$, for ER networks, $f_c \approx
0.4$. These differences in $f_c$ indicate that the minorities in SF networks need more initial supports to form final steady state clusters, than minorities in ER networks. This might be understood due to the high degree nodes (hubs) of a SF network. In the NCO model a hub, because of its large number of connections, is strongly influenced by its neighbors which are with high probability of the majority opinion, and puts the minority opinion at a disadvantage. Reference [@NCO] presents also studies of the NCO model in a two-dimensional Euclidean lattice and of the NCO model on real-world networks.
We next present numerical simulations indicating that the phase transition observed in the NCO model is in the same universality class as regular MF percolation. Percolation in random networks (e.g., ER and SF networks with $\lambda > 4$ for random failures or all $\lambda$ for targeted attacks) [@percolation; @quenched_shlomo; @Complexnetwork] is obtained by MF theory, which predicts that at criticality the cluster size distribution is $n_s \sim s^{-\tau}$ with $\tau=2.5$ and $S_1 \sim N^{\theta}$, where $\theta=d_f/d_c$ with $d_f=4$ and $d_c=6$ represent the fractal and the upper critical dimension of percolation respectively and thus $\theta=2/3$ (See Sec. \[percolation\]). Figure \[fig:NCOf3\](a) shows the finite cluster size distribution $n_s$ of the $\sigma_+$ opinion cluster as a function of $s$ at criticality ($f=f_c$). Figure \[fig:NCOf3\](b) show $S_1$ at criticality $f_c$ as a function of $N$ for ER and SF networks with $\langle k \rangle=4$ and $\lambda=2.5$, respectively. Note that in both networks $\tau \approx 2.5$ and $\theta \approx 2/3$. These two exponents strongly indicate that the NCO model in random networks behaves like a second order phase transition that belongs to the same universality class as regular MF percolation.
In our above results we focus on networks that have a relatively low average degree $\langle k \rangle$. We test the model for networks with higher average degrees. In SF networks we increase $\langle k
\rangle$ by increasing the value of $k_{\rm
min}$. Figure \[fig:NCOf4\] shows $s_1$ and $s_2$ as a function of $f$ for different values of $\langle k \rangle$ for ER networks and SF networks. As the values of $\langle k \rangle$ increase, i.e., as the network becomes increasingly condensed and the number of interactions between agents increases, a sharper change of $s_1$ at a critical threshold is observed. This may suggest (but can not be proved by simulations) the existence of a critical value $\langle k \rangle=k_c$ that is strongly affected by the topology of the network. Below $k_c$, as $\langle k \rangle$ increases, $f_c$ shifts to the right, as can be seen from the shift of the peak of $s_2$. Above $f_c$ two opinions can continue to coexist and remain stable. Above $k_c$ the smooth second order phase transition is replaced by a sharp jump of $s_1$ at approximately $f=0.5$ that is accompanied by the disappearance of the peaks of $s_2$. Note also that as the values of $\langle k \rangle$ increase, the region in which two opinions coexist becomes increasingly smaller and approach zero for very large values of $\langle k \rangle$ possibly above $k_c$. In terms of the NCO model, as the number of connections between individuals increase, the opinion of each individual becomes less important and each individual becomes increasingly susceptible to the influence of the majority opinion across the entire system. Thus the majority opinion can easily overwhelm the minority opinion, causing the critical behavior of the NCO model, the second-order phase transition, to disappear at large $\langle k \rangle$ and the NCO model to converge to the majority voter model yielding a possible global consensus throughout the system. Note that analytical arguments for an abrupt transition of the NCO model at large $\langle k \rangle$ are given in Ref. [@Percolate_or_die].
How can one help the minority opinions to survive? As we have seen, as the number of friends of an agent increases, the importance of the agent’s own opinion decreases. In this way the majority opinion gradually eliminates the minority opinion. If we generalize the NCO model by adding a weight value $W$ to each agent’s own opinion, as $W$ of an agent increases, the influence of the opinion of the agent’s neighbors decreases. We call this generalization of the NCO model the NCO$W$ model. As in the NCO model, in the NCO$W$ model we change an agent’s opinion if he is in a local minority but we also weight the agent’s own opinion $W$ times more than its nearest neighbors. The NCO model is actually a special case of the NCO$W$ model in which $W=1$. Figure \[fig:NCOf5\] shows plots of $s_1$ and $s_2$ as a function of $f$ for both $W=1$ and $W=4$. Note that as $W$ increases, the second-order phase transition becomes flatter and the peak of the $s_2$ shifts to the left, which indicates a smaller critical threshold $f_c$. The smaller value of $f_c$ for larger values of $W$ means that from the minority point of view, it needs fewer initial supporters to form and maintain stable finite clusters. When weight is added to the agents own opinion (indicating stubbornness) they become less susceptible to outside influence. Thus in the NCO$W$ model the majority is aided when the agents make more friends, but the minority in turn is aided when the agents treat their own opinion as more important than their friends’ opinions.
The ICO Model {#ICO}
=============
The initial configuration of the inflexible contrarian opinion (ICO) model corresponds to the final steady state of the NCO model in which two opinions $\sigma_+$ and $\sigma_-$ coexist. At $t=0$ a fraction $\phi$ of inflexible contrarians of opinion $\sigma_-$ are introduced into clusters of $\sigma_+$ by replacing nodes of $\sigma_+$. The inflexible contrarians are agents that hold a strong and unchangeable $\sigma_-$ opinion, that theoretically could influence the $\sigma_+$ opinion of their neighbors as the system evolves with NCO dynamics. Because the opinion held by the inflexible contrarians is unchanging, they function as a quenched noise in the network. The system evolves according to NCO dynamics until a new steady state is reached. In this steady state the agents form clusters of two different opinions above a new threshold $f_c \equiv
f_c(\phi)$. Because the contrarians hold the $\sigma_-$ opinion, the size of the $\sigma_+$ clusters decreases as $\phi$ increases. Figure \[fig:ICOf1\] demonstrates the dynamic of the ICO model. We use two different strategies to introduce a fraction $\phi$ of inflexible contrarians. In strategy I we chose the fraction $\phi$ of nodes with $\sigma_+$ opinion at random. In strategy II the inflexible contrarians are chosen from the agents with $\sigma_+$ opinion in decreasing order of their connectivity. Strategy II is thus a targeted strategy.
Simulation Results
------------------
We present our simulation results for ER networks with $\langle k
\rangle=4$ and $N=10^5$. For simulation results of ICO model for SF networks see Ref. [@ICO]. Figure \[fig:ICOf2\] shows plots of $s_1$ and $s_2$ as a function of $f$ for different values of $\phi$ for strategies I and II, respectively. Note that the ICO model inherits some of the properties of the NCO model. This is the case because there is a smooth phase transition with a critical threshold $f_c$, where $f_c$ is characterized by the sharp peak of $s_2$. However, for the ICO model, $f_c$ is also a function of $\phi$. Thus we denote the new $f_c$ in the ICO model by $f_c(\phi)$. We find that, as $\phi$ increases, the critical value $f_c(\phi)$ increases, which means that the largest cluster composed of $\sigma_+$ agents becomes less robust due to the increase in the number of inflexible contrarians of opinion $\sigma_-$. Note also that for $f>f_c(\phi)$, $s_1$ decreases as $\phi$ increases. Thus, we conclude that inflexible contrarians with opinion $\sigma_-$ have two effects: (i) they increase the value of $f_c(\phi)$ and thus the $\sigma_+$ opinion needs more initial support in order to survive, and (ii) they decrease the size of the largest $\sigma_+$ opinion cluster at $f>f_c(\phi)$. Note also that in the ICO model when $\phi$ is large the largest $\sigma_+$ cluster is fully destroyed and the second-order phase transition is lost. This is probably due to the fact that when $\phi$ is large, minority groups do not have high degree nodes and thus their average connectivity becomes smaller than $1$ and, as a consequence, will no longer be able to form stable clusters [@ER1]. As expected (see Fig. \[fig:ICOf2\]) strategy II is more efficient in destroying the largest minority component. This is plausible because, when selecting the initial fraction $\phi$ of inflexible contrarians using a targeted strategy, almost all the inflexible contrarians will be in the largest initial $\sigma_+$ cluster since this cluster includes most of the high degree nodes. Figure \[fig:ICOf3\] test this hypothesis and shows at the final stage of the NCO the ratio $F(k)$, which is the number of nodes within the GC of $\sigma_+$ opinion with degree $k$ divided by the total number of nodes of opinion $\sigma_+$ with degree $k$ in the entire network system, for different values of $f$. We find that for large values of $k$, $F(k) \to 1$. These results support our previous hypothesis that almost all the high degree nodes belong to the largest cluster, and this explains why strategy II is more efficient than strategy I.
We next test whether the ICO model undergoes a phase transition as a function of $\phi$ and what it its type. Figures \[fig:ICOf4\](a) and \[fig:ICOf4\](c) show plots of $s_1$ as a function of $\phi$ for different values of $f$ for strategy I and strategy II, respectively. Figures \[fig:ICOf4\](b)(top) and \[fig:ICOf4\](d)(top) show plots of $s_2$ as a function of $\phi$ for different values of $f$ for strategy I and strategy II, respectively. We can see that in both strategies $s_2$ has a peak at $\phi=\phi_c(f)$, which is a characteristic of a second-order phase transition. Figures \[fig:ICOf4\](b)(bottom) and \[fig:ICOf4\](d)(bottom) further support the presence of a second order phase transition by showing plots of the derivative of $s_1$ with respect to $\phi$ for different values of $f$. Note that there is an abrupt change with $\phi$ in $\Delta s_1/\Delta \phi$ at the same position of the peak of $s_2$, suggesting that the transition is of second order. We next show that the second order phase transition has the same exponents as MF percolation. Figure \[fig:ICOns\] plots the finite cluster size distribution of $\sigma_+$ agents, $n_s$ as a function of $s$ at $f=f_c(\phi)$, from where we obtain $\tau=5/2$. From $s_2$ we also compute the exponent $\gamma$ and obtain $\gamma
\approx 1$ (not shown). These two exponents indicate that the ICO model on random graphs belongs to the same universality class as MF percolation.
The NCO on Coupled Networks Model {#NCOCOUPLE}
=================================
Figure \[fig:f0\] demonstrates the dynamics of the NCO model on coupled networks. In coupled networks, the two networks represent two groups of people. The links within each network denote the relationships between nodes. For simplicity, we assume that the two networks have the same number of nodes $N$ and the same degree distribution. We also assign these two networks the same initial opinion condition, i.e., in both networks there is initially a fraction $f$ of nodes holding the $\sigma_+$ opinion, and a fraction $1-f$ holding the $\sigma_-$ opinion. To represent the strong social coupling between the two groups, we randomly choose a fraction $q$ of the nodes from both networks to form $qN$ pairs of one-to-one interdependent pairs regardless of their original opinions. At time t=0, in both networks we apply the same opinion formation rule, the NCO model, to decide whether an agent will change its opinion regardless of the interdependent links. This means that at this stage opinions propagate in each single network independently—as though the other network does not exist. All opinion updates are made simultaneously and in parallel. At t=1, if two nodes with an interdependent link have the same opinion they keep that opinion. If they do not, they follow the majority opinion of the coupled networks (global majority rule). All interdependent agents update their opinions simultaneously at the end of this time step. We repeat these two steps until the system reaches a steady state.
Simulation Results
------------------
We perform simulations of the NCO on coupled networks where both of the interdependent networks are either ER networks with $\langle k
\rangle=4$ or SF networks with $k_{\rm min}=2$ and $\lambda=2.5$. For an initial fraction $f$ of opinion $\sigma_+$ and a fraction $q$ of interdependent links, the NCO on coupled networks is simulated on $10^4$ network realizations to explore how interdependent links affect opinion dynamics.
### NCO on Coupled ER Networks
We first investigate $s_1$ as a function of $f$ for different values of $q$. Figure \[fig:f1\](a) shows that when $q=0$, which corresponds to the NCO model on a single network, the system undergoes a second order phase transition with a critical threshold $f_c$ [@NCO]. When $q>0$, there are two regions $0 < q \le 0.5$ and $q >
0.5$. For the region $0 < q \le 0.5$, as in the NCO model on a single network, the second order phase transition still exists, but the critical value $f \equiv f_c(q)$ is increasing with $q$. The value of $f_c(q)$ can be determined by the location of the peak of $s_2$, which is shown in Fig. \[fig:f1\](b), where we plot $s_2$ as a function of $f$ for different values of $q$. The inset of Fig. \[fig:f1\](a) shows a plot of $f_c(q)$ as a function of $q$. We find that the peak of $s_2$ shifts to the right for $q \le 0.5$ as $q$ increases, which means that $f_c(q)$ increases as $q$ increases. This suggests that if we add more interdependent links between the two networks, the minority opinion will need a larger initial fraction in order to exist. In the region $0 < q \le 0.5$ we also find that, unlike the NCO model on a single ER network, there is an abrupt change of $s_1$ at $f=0.5$, indicating that in addition to the smooth second order phase transition at $f_c(q)$, there may also be a discontinuous transition at $f=0.5$. Our results suggest that when $0 < q \le 0.5$ the system may undergo a hybrid phase transition [@hybrid], which is a mixture of both an abrupt and a second order phase transition. We also find that as $q$ increases the discontinuity around $f=0.5$ becomes more pronounced. Although the system possesses a seemingly discontinuous phase transition for $0 < q \le 0.5$ , the model itself is still a non-consensus model, i.e., when $f$ is above the critical value $f_c(q)$ the two opinions coexist in a steady state. When $q >
0.5$ the smooth second order phase transition of $s_1$ disappears and is replaced by an abrupt transition at $f=0.5$. When $q > 0.5$ the peak of $s_2$ disappears, supporting the loss of the second order phase transition \[see Fig. \[fig:f1\](b)\], and the system undergoes a pure abrupt transition. This suggests that when interactions between networks are sufficiently strong the hybrid phase transition is replaced by a pure abrupt transition. For all values of $q$, the region where two opinions can coexist decreases as $q$ increases, and the NCO coupled networks model moves at large $q$ toward the consensus type opinion model. To further support our finding of the existence of a discontinuous transition when $q>0$, in Fig. \[fig:f1\](c) and its inset we plot respectively the jump of $S_1$, $\Delta
S_1=S_1(0.51)-S_1(0.49)$, and $\Delta S_1/N$, around $f=0.5$ as a function of the system size $N$ for different values of $q$. The linear relationship between $\Delta S_1$ and $N$ supports our assumption of the existence of a discontinuous transition around $f=0.5$ for all values of $q>0$. Note that, as the value of $q$ increases, the value of $\Delta S_1$ increases, which means that as we increase the value of $q$ the abrupt transition becomes more pronounced.
To further support our conclusions, we investigate the number of iterations (NOI), which is the number of time steps needed to reach the steady state, as a function of $f$ for different values of $q$. Figure \[fig:f1\](d) shows a plot of the NOI as a function of $f$ for different values of $q$. As described in Ref. [@ParshaniPNAS], in a pure first order phase transition due to cascading failures the location of the peak of the NOI determines the critical threshold of the transition, which is the case for $q>0.5$ in our model. Figure \[fig:f1\](d) shows that there is only one peak for the NOI curves for $q>0.5$ at $f=0.5$, which is the position of the critical threshold of the abrupt transition. In the hybrid phase transition for $q \le 0.5$, the relation between the peak of the NOI and the critical threshold is unclear because there are two critical thresholds, one for the discontinuous transition at $f=0.5$ and the other for the second order phase transition at $f_c(q)$. Figure \[fig:f1\](d) shows that when $q<0.5$ the NOI exhibits two symmetric peaks. The inset of Fig. \[fig:f1\](d) shows a plot of the location of the left peaks of the NOI as a function of $q$. Comparing the insets in Fig. \[fig:f1\](a) and Fig. \[fig:f1\](d), we find that the curve of the peak locations of the NOI is always above the $f_c(q)$ curve, which suggests that for a hybrid phase transition the peak of the NOI is located between the critical thresholds of the second order phase transition and the abrupt transition.
Figure \[fig:f2\] shows a log-log plot of the NOI at $f=0.5$ as a function of the system size $N$ for different values of $q$, and the inset of Fig. \[fig:f2\] shows the same in a log-linear plot. The accuracy of the simulations is such that we cannot distinguish the relationship between exponential and logarithmic. However, the increase of NOI with system size indicates that there is a real jump at approximately $f=0.5$ rather than a finite size effect. This supports our previous conjecture that for all values of $q>0$, the NCO on coupled networks exhibits an abrupt transition at $f=0.5$.
We next present results indicating that, when $q \le 0.5$ and when $f$ is close to $f_c(q)$, our model is in the same universality class as regular MF percolation, even though a discontinuity appears at larger $f$. For regular percolation on random graphs at criticality, the cluster sizes follow a power law distribution, $n_s \sim s^{-\tau}$ with $\tau=2.5$ [@percolation; @quenched_shlomo; @Complexnetwork]. Figure \[fig:f3\] shows a plot of $n_s$ as a function of $s$ for finite $\sigma_+$ clusters at criticality, $f_c(q)$. We see that for $q \le 0.5$, $\tau \approx 2.5$, and for $q>0.5$, the power law no longer holds. The exponent values we obtain strongly indicate that, for small value of $q$, the NCO model on coupled ER networks close to $f_c$ is in the same universality class as mean field percolation in random networks. The power law for the cluster size distribution at $q>0.5$ disappears, so we conclude that the NCO coupled networks model changes the phase transition type as $q$ increases from $q \le 0.5$ to $q>0.5$.
### NCO on Coupled SF Networks
Empirical studies show that many real-world social networks are not ER. They instead exhibit a SF degree distribution [@SF1] in which $P(k)\sim k^{-\lambda}$ and $\lambda$ characterize the broadness of the distribution. A feature of SF is the existence of hubs, i.e., very high degree nodes. These large hubs make the opinion dynamic processes in SF networks much more efficient than in ER networks [@SFbetter1; @SFbetter2; @SFbetter3; @SFbetter4; @SFbetter5; @SFbetter6].
Because of its large number of connections, a hub in the NCO model tends to follow the opinion of the majority and effectively influence the opinions of its neighbors. In a SF network the hubs help the majority dominate the minority, and thus the NCO model on a single SF network has a larger $f_c$ and exhibits a much sharper jump around $f_c$ than in ER networks with the same average degree [@NCO]. This is also the case in interdependent SF networks. Figures \[fig:f4\](a), \[fig:f4\](b), and \[fig:f4\](c) depict $s_1$, $s_2$, and NOI as a functions of $f$ for different values of $q$, respectively. The results of the NCO model on coupled SF networks are similar to those of coupled ER networks, except that the region of the hybrid phase transition is much smaller in coupled SF networks. This is confirmed by the fact that the peak of $s_2$ drops much faster for small $q$ values and that the single peak of NOI shows up at smaller $q$ values for coupled SF networks in contrast to the case of coupled ER networks. This indicates that the pure abrupt phase transition occurs at smaller $q$ values in coupled SF networks compared to coupled ER networks, which suggests that in coupled SF networks a smaller number of interdependent agents are needed to achieve a consensus state compared to coupled ER networks. Figure \[fig:f5\] shows a plot of $n_s$ as a function of $s$ for finite $\sigma_+$ clusters at criticality. Note that in SF networks when $q \leq 0.1$ the $n_s$ decays as a power law with $\tau=2.5$, and when $q>0.1$ the power law decay of $n_s$ no longer holds. This suggests that only for small values of $q$ our NCO model on coupled SF networks is in the same universality class as regular MF percolation. Comparing Fig. \[fig:f5\] with Fig. \[fig:f3\], we find that the power law decay disappears at smaller $q$ values in coupled SF networks compared to coupled ER networks. This supports our hypothesis that interdependent links push the entire system to an abrupt phase transition more effectively in coupled SF networks than in coupled ER networks.
In both coupled ER and SF networks, our non-consensus opinion second order phase transition model is transformed into a consensus opinion type abrupt transition model when the number of interdependent links is increased. This suggests that increasing the interactions between different groups in our world will push humanity to become increasingly homogeneous, i.e., interdependent pairs in the NCO coupled networks model helps the majority opinions supporters to eliminate the minority opinion, making uniformity (consensus) a possible final result.
Summary {#conclusion}
=======
In this paper we revisit and extended the non-consensus opinion (NCO) model, introduced by Shao et al. [@NCO]. We introduce the NCO$W$ model in which each node’s opinion is given a weight $W$ to represent the nodes’ resistance to opinion changes. We find that in both the NCO and the NCO$W$ models the size of the largest minority cluster with $\sigma_+$ opinion undergoes a second order MF percolation transition in which the control parameter is $f$. The NCO$W$ model is more robust than the NCO model because the weighted nodes reinforce the largest $\sigma_+$ minority cluster, and shift the critical value $f_c$ to values lower than those found in the NCO model. We also show that when the average network degree $\langle k\rangle$ in the NCO model is increased, the second order phase transition is replaced by an abrupt transition, making the NCO model converge to a consensus type opinion model. We also review another non-consensus opinion model, the ICO model [@ICO], which introduces into the system, using both random and targeted strategies, a fraction $\phi$ of inflexible contrarians (which act as quenched noise). As $\phi$ increases, both random and targeted strategies reduce the size of the largest $\sigma_+$ cluster and, above a critical threshold $\phi=\phi_c$, the largest $\sigma_+$ cluster disappears and the second order phase transition is also lost. The targeted strategy is more efficient in eliminating the largest $\sigma_+$ cluster or decreasing its size. This is due to the fact that the contarians are introduced (targeted) mainly into the largest cluster, which contains most of the high degree nodes. Thus a smaller $\phi_c$ value is needed to eliminate the largest cluster of minority in the targeted strategy compared to the random strategy. We also study an opinion model in which two interdependent networks are coupled by a fraction $q$ of interdependent links. The internal dynamics within each network obey the NCO rules, but the cross-network interdependent nodes, when their opinions differ, obey the global majority rule. These interdependent links force the system from a second order phase transition, characteristic of the NCO model on a single network, to a hybrid phase transition, i.e., a mix of a second order transition and an abrupt transition. As the fraction of interdependent links increases, the system evolves to a pure abrupt phase transition. Above a certain value of $q$, which is strongly dependent on network topology, the interdependent link interactions push the non-consensus opinion model to a consensus opinion model. Because scale free networks have large hubs, the effect of interdependent links is more pronounced in interdependent scale free networks than in interdependent Erdös Rényi networks. We are investigating whether the same effect appears in other opinion models of interdependent networks. The results will be presented in a future paper.
We acknowledge support from the DTRA, ONR, NSF CDI program, the European EPIWORK, LINC and MULTIPLEX projects, the Israel Science Foundation, the PICT 0293/00 and UNMdP, NGI and CONGAS.
[99]{}
J. Shao [*et al.*]{}, Phys. Rev. Lett. [**103**]{}, 01870 (2009).
Q. Li [*el al.*]{}, Phys. Rev. E. [**84**]{}, 066101 (2011).
C. Castellano, S. Fortunato, and V. Loreto, Rev. Mod. Phys. [**81**]{}, 591 (2009).
S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D. Hwang, Physics Reports [**424**]{}, 175 (2006).
S. N. Dorogovtsev and J. F. F. Mendes, [*Evolution of Networks*]{} (Oxford University Press, Oxford, 2003).
R. Pastor-Satorras and A. Vespignani, [*Evolution and Structure of the Internet: A Statistical Approach*]{} (Cambridge University Press, Cambridge, 2004).
P. Erdös and A. Rényi, Publications of the Mathematical Institute of the Hungarian Academy of Science [**5**]{}, 17 (1960).
P. Erdös and A. Rényi, A. I. Publ. Math. [**6**]{}, 290 (1959).
B. Bollobás, [*Random Graphs*]{} (Academic Press, London, 1985).
R. Albert and A.-L. Barabási, Reviews of Modern Physics [**74**]{}, 47–97 (2002).
R. Cohen [*et al.*]{}, Phys. Rev. Lett. [**85**]{}, 4626 (2000).
A.-L. Barabási and R. Albert, Science [**286**]{}, 509 (1999).
R. Cohen and S. Havlin, [*Complex Networks: Structure, Robustness and Function*]{} (Cambridge University Press, Cambridge, 2010).
M. Newman, [*Networks: An Introduction*]{} (Oxford University Press, Oxford, 2010).
S. Galam, Europhys. Lett. [**70**]{}, 705 (2005).
K. Sznajd-Weron and J. Sznajd, Int. J. Mod. Phys. C [**11**]{}, 1157 (2000).
T. M. Liggett, [*Stochastic Interacting Systems: Contact Voter, and Exclusion Processes*]{} (Springer, Berlin, 1999).
R. Lambiotte and S. Redner, Europhys. Lett. [**82**]{}, 18007 (2008).
S. Galam, Eur. Phys. J. B [**25**]{}, 403 (2002).
P. L. Krapivsky and S. Redner, Phys. Rev. Lett. [ **90**]{}, 238701 (2003).
B. Latané, Am. Psychol [**36**]{}, 343 (1981).
A. Nowak [*et al.*]{}, Psychol. Rev. [**97**]{}, 362 (1990).
J. Shao, S. Havlin, and H. E. Stanley, Phys. Rev. Lett. [**109**]{}, 079802 (2012).
A. Sattari, M. Paczuski, and P. Grassberger, Phys. Rev. Lett. [**109**]{}, 079801 (2012).
D. ben-Avraham, Phys. Rev. E. [**83**]{}, 050101 (2011).
C. P. Roca, M. Draief, and D. Helbing, arXiv:1101.0775v1 (2011).
S. Galam, arXiv:0803.2453v1 (2008).
S. Galam, Physica A [**333**]{} 453–460 (2004).
S. V. Buldyrev, R. Parshani, G. Paul, H. E. Stanley, and S. Havlin, Nature [**464**]{}, 1025 (2010).
R. Parshani, S. V. Buldyrev, and S. Havlin, Phys. Rev. Lett. [**105**]{}, 048701 (2010).
J. Gao, S. V. Buldyrev, S. Havlin, and H. E. Stanley, Phys. Rev. Lett. [**107**]{}, 195701 (2011).
J. Gao, S. V. Buldyrev, H. E. Stanley, and S. Havlin, Nature Physics [**8**]{}, 40–48 (2012).
W. Li, A. Bashan, S. V. Buldyrev, H. E. Stanley, and S. Havlin, Phys. Rev. Lett. [**108**]{}, 228702 (2012).
A. Bashan [*et al.*]{}, arXiv:1206.2062v1 (2012).
E. A. Leicht and R. M. D’Souza, arXiv:0907.0894v1 (2009).
C. D. Brummitt, R. M. D’Souza and E. A. Leicht, Proc. Natl. Acad. Sci. USA [**109**]{}, 680 (2012).
R. Paster-Satorras and A. Vespignani, Phys. Rev. Lett. [**86**]{}, 3200–3203 (2001).
O. Givan, N. Schwartz, A. Cygelberg and L. Stone, Journal of Theoretical Biology [**288**]{}, 21–28 (2011).
M. Kitsak, L. K. Gallos, S. Havlin, F. Liljeros, L. Muchnik, H. E. Stanley, and H. A. Makse, Nature Physics [**6**]{}, 888–893 (2010).
R. Cohen, S. Havlin and D. ben-Avraham, Phys. Rev. Lett. [**91**]{}, 247901 (2003).
L. A. Braunstein, S. V. Buldyrev, R. Cohen, S. Havlin, and H. E. Stanley, Phys. Rev. Lett. [**91**]{}, 168701 (2003).
A. Bunde and S. Havlin, [*Fractals and Disordered Systems*]{} (Springer-Verlag, Berlin, 1994).
D. Stauffer and A. Aharony, [*Introduction to Percolation Theory*]{} (Taylor & Francis, New York, 2003).
R. Cohen, D. Ben-Avraham, and S. Havlin, Phys. Rev. E. [**66**]{}, 036113 (2002). R. Cohen, K. Erez, D. ben-Avraham, and S. Havlin, Phys. Rev. Lett. [**86**]{}, 3682 (2001). L. D. Valdez, P. A. Macri, and L. A. Braunstein, Phys. Rev. E [**85**]{}, 036108 (2012).
G. J. Baxter, S. N. Dorogovtsev, A. V. Goltsev, and J. F. F. Mendes, Phys. Rev. E [**82**]{}, 011103 (2010).
R. Parshani, S. V. Buldyrev, and S. Havlin, Proc. Natl. Acad. Sci. USA [**108**]{} 1007–1010 (2011).
R. Pastor-Satorras and A. Vespignani, Phys. Rev. Lett. [**86**]{}, 3200 (2001).
E. López [*et al.*]{}, Phys. Rev. Lett. [**94**]{}, 248701 (2005).
Z. Wu [*et al.*]{}, Phys. Rev. E [**71**]{}, 045101(R) (2005).
A. E. Motter, Phys. Rev. Lett. [**93**]{}, 098701 (2004).
G. Korniss, Phys. Rev. E [**75**]{}, 051121 (2007).
M. Boguñá and D. Krioukov, Phys. Rev. Lett. [ **102**]{}, 058701 (2009).
![Dynamics of the NCO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At t=0, five nodes are randomly assigned to be $\sigma_+$ (empty circle), and the remaining four nodes are assigned with $\sigma_-$ (solid circle). In the set comprising of node $A$ and its 4 neighbors (dashed box), node $A$ is in a local minority opinion, while the remaining nodes are not. Thus at the end of this simulation step, node $A$ is converted into $\sigma_-$ opinion. (b) At t=1, in the set of nodes comprising node $B$ and its 6 neighbors (dashed box), node $B$ becomes in a local minority opinion, while the remaining nodes are not. Thus, node $B$ is converted into $\sigma_-$ at the end of simulation step $t=1$. (c) At, $t=2$, the nine nodes system reaches a stable state. \[fig:NCOf1\]](NCOPLOT1.eps "fig:"){width="8cm" height="8cm"} ![Dynamics of the NCO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At t=0, five nodes are randomly assigned to be $\sigma_+$ (empty circle), and the remaining four nodes are assigned with $\sigma_-$ (solid circle). In the set comprising of node $A$ and its 4 neighbors (dashed box), node $A$ is in a local minority opinion, while the remaining nodes are not. Thus at the end of this simulation step, node $A$ is converted into $\sigma_-$ opinion. (b) At t=1, in the set of nodes comprising node $B$ and its 6 neighbors (dashed box), node $B$ becomes in a local minority opinion, while the remaining nodes are not. Thus, node $B$ is converted into $\sigma_-$ at the end of simulation step $t=1$. (c) At, $t=2$, the nine nodes system reaches a stable state. \[fig:NCOf1\]](NCOPLOT2.eps "fig:"){width="8cm" height="8cm"} ![Dynamics of the NCO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At t=0, five nodes are randomly assigned to be $\sigma_+$ (empty circle), and the remaining four nodes are assigned with $\sigma_-$ (solid circle). In the set comprising of node $A$ and its 4 neighbors (dashed box), node $A$ is in a local minority opinion, while the remaining nodes are not. Thus at the end of this simulation step, node $A$ is converted into $\sigma_-$ opinion. (b) At t=1, in the set of nodes comprising node $B$ and its 6 neighbors (dashed box), node $B$ becomes in a local minority opinion, while the remaining nodes are not. Thus, node $B$ is converted into $\sigma_-$ at the end of simulation step $t=1$. (c) At, $t=2$, the nine nodes system reaches a stable state. \[fig:NCOf1\]](NCOPLOT3.eps "fig:"){width="8cm" height="8cm"}
![Plots of $s_1$, $s_2$ and $F$ of opinion $\sigma_+$ as a function of $f$ with network size $N=10000$, for (a) ER networks with $\langle k\rangle=4$ and (b) SF networks with $\lambda=2.5$ and $k_{\rm min}=2$. All simulations were done for $10^4$ networks realizations.\[fig:NCOf2\]](NCO_fig1a.eps "fig:"){width="7cm" height="7cm"} ![Plots of $s_1$, $s_2$ and $F$ of opinion $\sigma_+$ as a function of $f$ with network size $N=10000$, for (a) ER networks with $\langle k\rangle=4$ and (b) SF networks with $\lambda=2.5$ and $k_{\rm min}=2$. All simulations were done for $10^4$ networks realizations.\[fig:NCOf2\]](NCO_fig1b.eps "fig:"){width="7cm" height="7cm"}\
![ (a) Plots of $n_s$ as a function of $s$ at criticality for ER networks with $\langle k\rangle=4$ and SF networks with $\lambda=2.5$ and $k_{\rm min}=2$. The dashed line is a guide to show that the slope obtained is $\tau=2.5$. (b) Plots of $S_1$ as a function of $N$ at criticality for ER networks with $\langle
k\rangle=4$ and SF networks with $\lambda=2.5$ and $k_{\rm
min}=2$. The dashed lines are guides to show that the slope obtained is $\theta \approx 2/3$. All simulations were done for $N=10000$ and over $10^4$ networks realizations.\[fig:NCOf3\]](NCO_fig2.eps "fig:"){width="7cm" height="7cm"} ![ (a) Plots of $n_s$ as a function of $s$ at criticality for ER networks with $\langle k\rangle=4$ and SF networks with $\lambda=2.5$ and $k_{\rm min}=2$. The dashed line is a guide to show that the slope obtained is $\tau=2.5$. (b) Plots of $S_1$ as a function of $N$ at criticality for ER networks with $\langle
k\rangle=4$ and SF networks with $\lambda=2.5$ and $k_{\rm
min}=2$. The dashed lines are guides to show that the slope obtained is $\theta \approx 2/3$. All simulations were done for $N=10000$ and over $10^4$ networks realizations.\[fig:NCOf3\]](S1_N.eps "fig:"){width="7cm" height="7cm"}\
![Plots of (a) $s_1$ and (b) $s_2$ of opinion $\sigma_+$ as a function of $f$ for ER networks with different values of $\langle
k\rangle$ for $N=10000$. (c) $s_1$ and (d) $s_2$ of opinion $\sigma_+$ as a function of $f$ for SF networks with different values of $\langle k\rangle$ for $N=10000$ and $\lambda=2.5$. The solid lines are guides for the eyes. All simulations were done for $10^4$ networks realizations. \[fig:NCOf4\]](NCO_fig3a.eps "fig:"){width="7cm" height="7cm"} ![Plots of (a) $s_1$ and (b) $s_2$ of opinion $\sigma_+$ as a function of $f$ for ER networks with different values of $\langle
k\rangle$ for $N=10000$. (c) $s_1$ and (d) $s_2$ of opinion $\sigma_+$ as a function of $f$ for SF networks with different values of $\langle k\rangle$ for $N=10000$ and $\lambda=2.5$. The solid lines are guides for the eyes. All simulations were done for $10^4$ networks realizations. \[fig:NCOf4\]](NCO_fig3b.eps "fig:"){width="7cm" height="7cm"}\
![Plots of (a) $s_1$ and (b) $s_2$ of opinion $\sigma_+$ as a function of $f$ for ER networks with different values of $\langle
k\rangle$ for $N=10000$. (c) $s_1$ and (d) $s_2$ of opinion $\sigma_+$ as a function of $f$ for SF networks with different values of $\langle k\rangle$ for $N=10000$ and $\lambda=2.5$. The solid lines are guides for the eyes. All simulations were done for $10^4$ networks realizations. \[fig:NCOf4\]](NCO_fig3c.eps "fig:"){width="7cm" height="7cm"} ![Plots of (a) $s_1$ and (b) $s_2$ of opinion $\sigma_+$ as a function of $f$ for ER networks with different values of $\langle
k\rangle$ for $N=10000$. (c) $s_1$ and (d) $s_2$ of opinion $\sigma_+$ as a function of $f$ for SF networks with different values of $\langle k\rangle$ for $N=10000$ and $\lambda=2.5$. The solid lines are guides for the eyes. All simulations were done for $10^4$ networks realizations. \[fig:NCOf4\]](NCO_fig3d.eps "fig:"){width="7cm" height="7cm"}
![Plots of (a) $s_1$ and (b) $s_2$ of opinion $\sigma_+$ as a function of $f$, for ER networks with different values of $W$ for $\langle k\rangle=4$. The solid lines are guides for the eyes. All simulations were done with $N=10000$ and for $10^4$ networks realizations.\[fig:NCOf5\]](NCOW_S1.eps "fig:"){width="7cm" height="7cm"} ![Plots of (a) $s_1$ and (b) $s_2$ of opinion $\sigma_+$ as a function of $f$, for ER networks with different values of $W$ for $\langle k\rangle=4$. The solid lines are guides for the eyes. All simulations were done with $N=10000$ and for $10^4$ networks realizations.\[fig:NCOf5\]](NCOW_S2.eps "fig:"){width="7cm" height="7cm"}\
![Schematic plot of the dynamics of the ICO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At $t=0$, we have a stable state where opinion $\sigma_+$ (open circle) and opinion $\sigma_-$ (filled circle) coexist. (b) At $t=1$, we change node $1$ into a inflexible contrarian (filled square), which will hold $\sigma_-$ opinion. Node $2$ is now in a local minority opinion while the remaining nodes are not. Notice that node $1$ is an inflexible contrarian and even if he is in the local minority he will not change his opinion. At the end of this simulation step, node $2$ is converted into $\sigma_-$ opinion. (c) At $t=2$, node $3$ is in a local minority opinion and therefore will be converted into $\sigma_-$ opinion. (d) At $t=3$, the system reaches a stable state where the system breaks into four disconnected clusters, one of them composed of six $\sigma_-$ nodes and the other three with one $\sigma_+$ node. \[fig:ICOf1\]](ICO_net_plot_1.eps "fig:"){width="6cm" height="6cm"} ![Schematic plot of the dynamics of the ICO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At $t=0$, we have a stable state where opinion $\sigma_+$ (open circle) and opinion $\sigma_-$ (filled circle) coexist. (b) At $t=1$, we change node $1$ into a inflexible contrarian (filled square), which will hold $\sigma_-$ opinion. Node $2$ is now in a local minority opinion while the remaining nodes are not. Notice that node $1$ is an inflexible contrarian and even if he is in the local minority he will not change his opinion. At the end of this simulation step, node $2$ is converted into $\sigma_-$ opinion. (c) At $t=2$, node $3$ is in a local minority opinion and therefore will be converted into $\sigma_-$ opinion. (d) At $t=3$, the system reaches a stable state where the system breaks into four disconnected clusters, one of them composed of six $\sigma_-$ nodes and the other three with one $\sigma_+$ node. \[fig:ICOf1\]](ICO_net_plot_2.eps "fig:"){width="6cm" height="6cm"}\
![Schematic plot of the dynamics of the ICO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At $t=0$, we have a stable state where opinion $\sigma_+$ (open circle) and opinion $\sigma_-$ (filled circle) coexist. (b) At $t=1$, we change node $1$ into a inflexible contrarian (filled square), which will hold $\sigma_-$ opinion. Node $2$ is now in a local minority opinion while the remaining nodes are not. Notice that node $1$ is an inflexible contrarian and even if he is in the local minority he will not change his opinion. At the end of this simulation step, node $2$ is converted into $\sigma_-$ opinion. (c) At $t=2$, node $3$ is in a local minority opinion and therefore will be converted into $\sigma_-$ opinion. (d) At $t=3$, the system reaches a stable state where the system breaks into four disconnected clusters, one of them composed of six $\sigma_-$ nodes and the other three with one $\sigma_+$ node. \[fig:ICOf1\]](ICO_net_plot_3.eps "fig:"){width="6cm" height="6cm"} ![Schematic plot of the dynamics of the ICO model showing the approach to a stable state on a network with $N=9$ nodes. (a) At $t=0$, we have a stable state where opinion $\sigma_+$ (open circle) and opinion $\sigma_-$ (filled circle) coexist. (b) At $t=1$, we change node $1$ into a inflexible contrarian (filled square), which will hold $\sigma_-$ opinion. Node $2$ is now in a local minority opinion while the remaining nodes are not. Notice that node $1$ is an inflexible contrarian and even if he is in the local minority he will not change his opinion. At the end of this simulation step, node $2$ is converted into $\sigma_-$ opinion. (c) At $t=2$, node $3$ is in a local minority opinion and therefore will be converted into $\sigma_-$ opinion. (d) At $t=3$, the system reaches a stable state where the system breaks into four disconnected clusters, one of them composed of six $\sigma_-$ nodes and the other three with one $\sigma_+$ node. \[fig:ICOf1\]](ICO_net_plot_4.eps "fig:"){width="6cm" height="6cm"}\
![For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ and (b) $s_2$ as a function of $f$ for different values of $\phi$ under strategy I. Plots of (c) $s_1$ and (d) $s_2$ as a function for $f$ for different values of $\phi$ under strategy II. The solid lines in $s_2$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations. \[fig:ICOf2\]](ICO_fig1_a.eps "fig:"){width="7cm" height="7cm"} ![For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ and (b) $s_2$ as a function of $f$ for different values of $\phi$ under strategy I. Plots of (c) $s_1$ and (d) $s_2$ as a function for $f$ for different values of $\phi$ under strategy II. The solid lines in $s_2$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations. \[fig:ICOf2\]](ICO_fig1_b.eps "fig:"){width="7cm" height="7cm"}\
![For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ and (b) $s_2$ as a function of $f$ for different values of $\phi$ under strategy I. Plots of (c) $s_1$ and (d) $s_2$ as a function for $f$ for different values of $\phi$ under strategy II. The solid lines in $s_2$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations. \[fig:ICOf2\]](ICO_fig1_c.eps "fig:"){width="7cm" height="7cm"} ![For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ and (b) $s_2$ as a function of $f$ for different values of $\phi$ under strategy I. Plots of (c) $s_1$ and (d) $s_2$ as a function for $f$ for different values of $\phi$ under strategy II. The solid lines in $s_2$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations. \[fig:ICOf2\]](ICO_fig1_d.eps "fig:"){width="7cm" height="7cm"}
![$F(k)$ as a function of $k$ for different values of $f$ for ER networks with $\langle k\rangle=4$ . We can see that as $f$ increases $F(k) \to 1$ for smaller values of $k$. All simulations were done with $N=10000$ and $10^4$ network realizations. \[fig:ICOf3\]](ICO_Fk_ER.eps){width="8cm" height="8cm"}
![ For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ (b)(top) $s_2$ and (b)(bottom) $\Delta S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy I. Plot of (c) $s_1$, (d)(top) $s_2$ and (d)(bottom) $\Delta
S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy II. The solid lines in $s_2$ and $\Delta S_1/\Delta
\phi$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations.\[fig:ICOf4\]](ICO_fig2_a.eps "fig:"){width="8cm" height="8cm"} ![ For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ (b)(top) $s_2$ and (b)(bottom) $\Delta S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy I. Plot of (c) $s_1$, (d)(top) $s_2$ and (d)(bottom) $\Delta
S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy II. The solid lines in $s_2$ and $\Delta S_1/\Delta
\phi$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations.\[fig:ICOf4\]](ICO_fig2_b.eps "fig:"){width="8cm" height="8cm"}\
![ For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ (b)(top) $s_2$ and (b)(bottom) $\Delta S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy I. Plot of (c) $s_1$, (d)(top) $s_2$ and (d)(bottom) $\Delta
S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy II. The solid lines in $s_2$ and $\Delta S_1/\Delta
\phi$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations.\[fig:ICOf4\]](ICO_fig2_c.eps "fig:"){width="8cm" height="8cm"} ![ For ER networks with $\langle k\rangle=4$, plots of (a) $s_1$ (b)(top) $s_2$ and (b)(bottom) $\Delta S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy I. Plot of (c) $s_1$, (d)(top) $s_2$ and (d)(bottom) $\Delta
S_1/\Delta \phi$ as a function of $\phi$ for different values of $f$ under strategy II. The solid lines in $s_2$ and $\Delta S_1/\Delta
\phi$ are guides for the eyes. All simulations were done with $N=10000$ and $10^4$ network realizations.\[fig:ICOf4\]](ICO_fig2_d.eps "fig:"){width="8cm" height="8cm"}
![Plots of $n_s$ as a function of $s$ for both strategies at $f_c(\phi)$. For the random strategy, $f_c(\phi)=0.36$ and for the targeted strategy, $f_c(\phi)=0.45$ when $\phi=0.2$. The dashed line is a guide to show that the slope obtained is $\tau \approx
5/2$. All simulations were done with $N=10000$ and $10^4$ network realizations. \[fig:ICOns\]](ICO_ns.eps){width="14cm" height="14cm"}
![(Color online) Schematic plot of the dynamics of the NCO on coupled networks showing the approach to a stable state on a system of interdependent networks $A$ and $B$ with $N=6$ nodes in each network. Open circles represent opinion $\sigma_+$ and solid circles represent opinion $\sigma_-$. The solid lines connecting nodes in each network are connectivity links within the networks, and the dashed lines connecting nodes from two networks are interdependent links. In the initial state, each node is randomly assigned with opinion $\sigma_+$ or $\sigma_-$. (a) At t=0, opinion dynamics evolve within each single network. In networks $A$ and $B$, nodes $A_1$, $A_2$, $A_3$, $A_4$ and $B_1$ are in a local minority opinion within their network, so at the end of this time step these nodes will change their opinions. The remaining nodes will keep their opinions. (b) At t=1, the two networks interact through the interdependent links. Notice that the global majority opinion is $\sigma_+$ now. Thus, the pairs $A_1 -B_1$ and $A_5 -B_5$ where two nodes have different opinions, will follow the global majority opinion. So at the end of this time step nodes $A_1$ and $A_5$ will change their opinions. The pair $A_3 -B_3$, remains as $\sigma_-$ since both nodes share the same opinion. (c) At t=2, in network $A$, node $A_3$ is in a local minority, so at the end of the time step it will change opinion. (d) At t=3, the two networks interact through the interdependent links. Notice that the global majority opinion is still $\sigma_+$, the only pair with different opinion is $A_3
-B_3$, so at the end of this time step, $B_3$ will change its opinion. (d) At t=4, the interdependent networks reach a stable state and no more changes will happen. \[fig:f0\]](image1.eps "fig:"){width="5cm" height="6cm"} ![(Color online) Schematic plot of the dynamics of the NCO on coupled networks showing the approach to a stable state on a system of interdependent networks $A$ and $B$ with $N=6$ nodes in each network. Open circles represent opinion $\sigma_+$ and solid circles represent opinion $\sigma_-$. The solid lines connecting nodes in each network are connectivity links within the networks, and the dashed lines connecting nodes from two networks are interdependent links. In the initial state, each node is randomly assigned with opinion $\sigma_+$ or $\sigma_-$. (a) At t=0, opinion dynamics evolve within each single network. In networks $A$ and $B$, nodes $A_1$, $A_2$, $A_3$, $A_4$ and $B_1$ are in a local minority opinion within their network, so at the end of this time step these nodes will change their opinions. The remaining nodes will keep their opinions. (b) At t=1, the two networks interact through the interdependent links. Notice that the global majority opinion is $\sigma_+$ now. Thus, the pairs $A_1 -B_1$ and $A_5 -B_5$ where two nodes have different opinions, will follow the global majority opinion. So at the end of this time step nodes $A_1$ and $A_5$ will change their opinions. The pair $A_3 -B_3$, remains as $\sigma_-$ since both nodes share the same opinion. (c) At t=2, in network $A$, node $A_3$ is in a local minority, so at the end of the time step it will change opinion. (d) At t=3, the two networks interact through the interdependent links. Notice that the global majority opinion is still $\sigma_+$, the only pair with different opinion is $A_3
-B_3$, so at the end of this time step, $B_3$ will change its opinion. (d) At t=4, the interdependent networks reach a stable state and no more changes will happen. \[fig:f0\]](image2.eps "fig:"){width="5cm" height="6cm"}\
![(Color online) Schematic plot of the dynamics of the NCO on coupled networks showing the approach to a stable state on a system of interdependent networks $A$ and $B$ with $N=6$ nodes in each network. Open circles represent opinion $\sigma_+$ and solid circles represent opinion $\sigma_-$. The solid lines connecting nodes in each network are connectivity links within the networks, and the dashed lines connecting nodes from two networks are interdependent links. In the initial state, each node is randomly assigned with opinion $\sigma_+$ or $\sigma_-$. (a) At t=0, opinion dynamics evolve within each single network. In networks $A$ and $B$, nodes $A_1$, $A_2$, $A_3$, $A_4$ and $B_1$ are in a local minority opinion within their network, so at the end of this time step these nodes will change their opinions. The remaining nodes will keep their opinions. (b) At t=1, the two networks interact through the interdependent links. Notice that the global majority opinion is $\sigma_+$ now. Thus, the pairs $A_1 -B_1$ and $A_5 -B_5$ where two nodes have different opinions, will follow the global majority opinion. So at the end of this time step nodes $A_1$ and $A_5$ will change their opinions. The pair $A_3 -B_3$, remains as $\sigma_-$ since both nodes share the same opinion. (c) At t=2, in network $A$, node $A_3$ is in a local minority, so at the end of the time step it will change opinion. (d) At t=3, the two networks interact through the interdependent links. Notice that the global majority opinion is still $\sigma_+$, the only pair with different opinion is $A_3
-B_3$, so at the end of this time step, $B_3$ will change its opinion. (d) At t=4, the interdependent networks reach a stable state and no more changes will happen. \[fig:f0\]](image3.eps "fig:"){width="5cm" height="6cm"} ![(Color online) Schematic plot of the dynamics of the NCO on coupled networks showing the approach to a stable state on a system of interdependent networks $A$ and $B$ with $N=6$ nodes in each network. Open circles represent opinion $\sigma_+$ and solid circles represent opinion $\sigma_-$. The solid lines connecting nodes in each network are connectivity links within the networks, and the dashed lines connecting nodes from two networks are interdependent links. In the initial state, each node is randomly assigned with opinion $\sigma_+$ or $\sigma_-$. (a) At t=0, opinion dynamics evolve within each single network. In networks $A$ and $B$, nodes $A_1$, $A_2$, $A_3$, $A_4$ and $B_1$ are in a local minority opinion within their network, so at the end of this time step these nodes will change their opinions. The remaining nodes will keep their opinions. (b) At t=1, the two networks interact through the interdependent links. Notice that the global majority opinion is $\sigma_+$ now. Thus, the pairs $A_1 -B_1$ and $A_5 -B_5$ where two nodes have different opinions, will follow the global majority opinion. So at the end of this time step nodes $A_1$ and $A_5$ will change their opinions. The pair $A_3 -B_3$, remains as $\sigma_-$ since both nodes share the same opinion. (c) At t=2, in network $A$, node $A_3$ is in a local minority, so at the end of the time step it will change opinion. (d) At t=3, the two networks interact through the interdependent links. Notice that the global majority opinion is still $\sigma_+$, the only pair with different opinion is $A_3
-B_3$, so at the end of this time step, $B_3$ will change its opinion. (d) At t=4, the interdependent networks reach a stable state and no more changes will happen. \[fig:f0\]](image4.eps "fig:"){width="5cm" height="6cm"} ![(Color online) Schematic plot of the dynamics of the NCO on coupled networks showing the approach to a stable state on a system of interdependent networks $A$ and $B$ with $N=6$ nodes in each network. Open circles represent opinion $\sigma_+$ and solid circles represent opinion $\sigma_-$. The solid lines connecting nodes in each network are connectivity links within the networks, and the dashed lines connecting nodes from two networks are interdependent links. In the initial state, each node is randomly assigned with opinion $\sigma_+$ or $\sigma_-$. (a) At t=0, opinion dynamics evolve within each single network. In networks $A$ and $B$, nodes $A_1$, $A_2$, $A_3$, $A_4$ and $B_1$ are in a local minority opinion within their network, so at the end of this time step these nodes will change their opinions. The remaining nodes will keep their opinions. (b) At t=1, the two networks interact through the interdependent links. Notice that the global majority opinion is $\sigma_+$ now. Thus, the pairs $A_1 -B_1$ and $A_5 -B_5$ where two nodes have different opinions, will follow the global majority opinion. So at the end of this time step nodes $A_1$ and $A_5$ will change their opinions. The pair $A_3 -B_3$, remains as $\sigma_-$ since both nodes share the same opinion. (c) At t=2, in network $A$, node $A_3$ is in a local minority, so at the end of the time step it will change opinion. (d) At t=3, the two networks interact through the interdependent links. Notice that the global majority opinion is still $\sigma_+$, the only pair with different opinion is $A_3
-B_3$, so at the end of this time step, $B_3$ will change its opinion. (d) At t=4, the interdependent networks reach a stable state and no more changes will happen. \[fig:f0\]](image5.eps "fig:"){width="5cm" height="6cm"}
![Plots of NCO on coupled ER networks, with $\langle
k_A\rangle=\langle k_B\rangle=4$ and for each network $N=10000$. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function of $f$ for several values of $q$. As seen, when $q=0$, the system undergoes a smooth second order phase transition (regular NCO model). As $q$ increases until $q=0.5$, it becomes a hybrid phase transition, which contains both a smooth second order type and a seemingly abrupt jump, [ *i.e.*]{}, $s_1$ changes smoothly close to $f_c(q)$ and followed by a sharp jump at $f=0.5$. When $q$ is further increased ($q>0.5$), the smooth phase transition disappears, the system undergoes a pure abrupt phase transition. In the inset of (a) we plot $f_c$ as a function of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. As seen, when $q$ increases, the peaks of the $s_2$, which characterize the critical threshold value of the second order phase transition, shift to the right. We can see that beyond $q=0.5$, the peak of $s_2$ disappears, which indicates that there is no second order phase transition. (c) Plot of the change of $S_1$ around $f=0.5$, $\Delta S_1$, as a function of system size $N$ for different values of $q>0$. In the inset of (c), we plot $\Delta
S_1/N$ as a function of $N$ for different values of $q$. The linear relationship between $\Delta S_1$ and $N$ suggests that for $q>0$, around $f=0.5$, there exists a discontinuous transition. (d) Plot of the number of cascading steps (NOI) of the networks as a function of $f$ for different values of $q$. As seen there are two peaks for each $q$ value for $q<0.5$, and for $q \ge 0.5$ there is only one peak at $f=0.5$. In the inset of (d), plot of the location of NOI peak as a function of $q$. The solid lines are guides for the eyes. \[fig:f1\]](fig1_a1.eps "fig:"){width="6cm" height="5cm"} ![Plots of NCO on coupled ER networks, with $\langle
k_A\rangle=\langle k_B\rangle=4$ and for each network $N=10000$. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function of $f$ for several values of $q$. As seen, when $q=0$, the system undergoes a smooth second order phase transition (regular NCO model). As $q$ increases until $q=0.5$, it becomes a hybrid phase transition, which contains both a smooth second order type and a seemingly abrupt jump, [ *i.e.*]{}, $s_1$ changes smoothly close to $f_c(q)$ and followed by a sharp jump at $f=0.5$. When $q$ is further increased ($q>0.5$), the smooth phase transition disappears, the system undergoes a pure abrupt phase transition. In the inset of (a) we plot $f_c$ as a function of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. As seen, when $q$ increases, the peaks of the $s_2$, which characterize the critical threshold value of the second order phase transition, shift to the right. We can see that beyond $q=0.5$, the peak of $s_2$ disappears, which indicates that there is no second order phase transition. (c) Plot of the change of $S_1$ around $f=0.5$, $\Delta S_1$, as a function of system size $N$ for different values of $q>0$. In the inset of (c), we plot $\Delta
S_1/N$ as a function of $N$ for different values of $q$. The linear relationship between $\Delta S_1$ and $N$ suggests that for $q>0$, around $f=0.5$, there exists a discontinuous transition. (d) Plot of the number of cascading steps (NOI) of the networks as a function of $f$ for different values of $q$. As seen there are two peaks for each $q$ value for $q<0.5$, and for $q \ge 0.5$ there is only one peak at $f=0.5$. In the inset of (d), plot of the location of NOI peak as a function of $q$. The solid lines are guides for the eyes. \[fig:f1\]](fig1_b1.eps "fig:"){width="6cm" height="5cm"}\
![Plots of NCO on coupled ER networks, with $\langle
k_A\rangle=\langle k_B\rangle=4$ and for each network $N=10000$. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function of $f$ for several values of $q$. As seen, when $q=0$, the system undergoes a smooth second order phase transition (regular NCO model). As $q$ increases until $q=0.5$, it becomes a hybrid phase transition, which contains both a smooth second order type and a seemingly abrupt jump, [ *i.e.*]{}, $s_1$ changes smoothly close to $f_c(q)$ and followed by a sharp jump at $f=0.5$. When $q$ is further increased ($q>0.5$), the smooth phase transition disappears, the system undergoes a pure abrupt phase transition. In the inset of (a) we plot $f_c$ as a function of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. As seen, when $q$ increases, the peaks of the $s_2$, which characterize the critical threshold value of the second order phase transition, shift to the right. We can see that beyond $q=0.5$, the peak of $s_2$ disappears, which indicates that there is no second order phase transition. (c) Plot of the change of $S_1$ around $f=0.5$, $\Delta S_1$, as a function of system size $N$ for different values of $q>0$. In the inset of (c), we plot $\Delta
S_1/N$ as a function of $N$ for different values of $q$. The linear relationship between $\Delta S_1$ and $N$ suggests that for $q>0$, around $f=0.5$, there exists a discontinuous transition. (d) Plot of the number of cascading steps (NOI) of the networks as a function of $f$ for different values of $q$. As seen there are two peaks for each $q$ value for $q<0.5$, and for $q \ge 0.5$ there is only one peak at $f=0.5$. In the inset of (d), plot of the location of NOI peak as a function of $q$. The solid lines are guides for the eyes. \[fig:f1\]](DeltaS1_N_1.eps "fig:"){width="6cm" height="5cm"} ![Plots of NCO on coupled ER networks, with $\langle
k_A\rangle=\langle k_B\rangle=4$ and for each network $N=10000$. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function of $f$ for several values of $q$. As seen, when $q=0$, the system undergoes a smooth second order phase transition (regular NCO model). As $q$ increases until $q=0.5$, it becomes a hybrid phase transition, which contains both a smooth second order type and a seemingly abrupt jump, [ *i.e.*]{}, $s_1$ changes smoothly close to $f_c(q)$ and followed by a sharp jump at $f=0.5$. When $q$ is further increased ($q>0.5$), the smooth phase transition disappears, the system undergoes a pure abrupt phase transition. In the inset of (a) we plot $f_c$ as a function of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. As seen, when $q$ increases, the peaks of the $s_2$, which characterize the critical threshold value of the second order phase transition, shift to the right. We can see that beyond $q=0.5$, the peak of $s_2$ disappears, which indicates that there is no second order phase transition. (c) Plot of the change of $S_1$ around $f=0.5$, $\Delta S_1$, as a function of system size $N$ for different values of $q>0$. In the inset of (c), we plot $\Delta
S_1/N$ as a function of $N$ for different values of $q$. The linear relationship between $\Delta S_1$ and $N$ suggests that for $q>0$, around $f=0.5$, there exists a discontinuous transition. (d) Plot of the number of cascading steps (NOI) of the networks as a function of $f$ for different values of $q$. As seen there are two peaks for each $q$ value for $q<0.5$, and for $q \ge 0.5$ there is only one peak at $f=0.5$. In the inset of (d), plot of the location of NOI peak as a function of $q$. The solid lines are guides for the eyes. \[fig:f1\]](fig1_c1.eps "fig:"){width="6cm" height="5cm"}
![ Plot of the number of cascading steps between the networks (ER with $\langle k_A\rangle=\langle k_B\rangle=\langle k\rangle=4$) at $f=0.5$ as a function of system size for different values of $q$ in log-log scale and in log-linear scale in the inset respectively. The dashed lines are the power law and logarithmic fittings respectively. \[fig:f2\]](Cascading_steps_middle.eps "fig:"){width="14cm" height="14cm"}\
![ Plot of $n_s$ as a function of $s$ at criticality, $f_c$, for different values of $q$ for the NCO model on coupled ER networks, with $\langle k_A\rangle=\langle k_B\rangle=\langle
k\rangle=4$ and $N=10000$. For each network the results are averaged over $10^4$ realizations. As $q$ increases, $n_s$ losses the power law shape indicating that the second order phase transition is lost. The dashed line is a guide to show the slope $\tau=2.5$.\[fig:f3\]](fig3_a1.eps "fig:"){width="14cm" height="14cm"}\
![Study of NCO model on coupled SF networks, with $k_{\rm
min}=2$, $\lambda=2.5$ and $N=10000$ for each network. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function $f$ for different values of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. (c) Plot of the number of cascading steps, NOI, of the coupled networks as a function of $f$ for different values of $q$. The solid lines are guides for the eyes.\[fig:f4\]](fig4_a.eps "fig:"){width="7cm" height="7cm"} ![Study of NCO model on coupled SF networks, with $k_{\rm
min}=2$, $\lambda=2.5$ and $N=10000$ for each network. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function $f$ for different values of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. (c) Plot of the number of cascading steps, NOI, of the coupled networks as a function of $f$ for different values of $q$. The solid lines are guides for the eyes.\[fig:f4\]](fig4_b.eps "fig:"){width="7cm" height="7cm"}\
![Study of NCO model on coupled SF networks, with $k_{\rm
min}=2$, $\lambda=2.5$ and $N=10000$ for each network. (a) Plot of $s_1$ of opinion $\sigma_+$ as a function $f$ for different values of $q$. (b) Plot of $s_2$ as a function of $f$ for different values of $q$. (c) Plot of the number of cascading steps, NOI, of the coupled networks as a function of $f$ for different values of $q$. The solid lines are guides for the eyes.\[fig:f4\]](fig4_c.eps "fig:"){width="7cm" height="7cm"}
![ Plot of $n_s$ as a function of $s$ at criticality for different value of $q$ for the NCO model on coupled SF networks, with $k_{\rm min}=2$, $\lambda=2.5$ and $N=10000$. For each case the results are averaged over $10^4$ realizations. As $q$ increases, $n_s$ losses the power law shape indicating that the second order phase transition is lost. The dashed line is a guide to show the slope $\tau=2.5$.\[fig:f5\]](fig5_a1.eps "fig:"){width="14cm" height="14cm"}\
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The general sum-connectivity index of a graph $G$ is defined as $\chi_{\alpha}(G)= \sum_{uv\in E(G)} (d_u + d_{v})^{\alpha}$ where $d_{u}$ is degree of the vertex $u\in V(G)$, $\alpha$ is a real number different from $0$ and $uv$ is the edge connecting the vertices $u,v$. In this note, the problem of characterizing the graphs having extremum $\chi_{\alpha}$ values from a certain collection of polyomino chain graphs is solved for $\alpha<0$. The obtained results together with already known results (concerning extremum values of polyomino chain graphs) give the complete solution of the aforementioned problem.'
author:
- '**Akbar Ali[^1] , Tahir Idrees**'
date: 'Submitted on January 2, 2018'
title: ' **A note on polyomino chains with extremum general sum-connectivity index**'
---
Introduction
============
All graphs considered in this note are simple, finite and connected. Those notations and terminologies from graph theory which are not defined here can be found in the books [@Harary69; @Bondy08].
The connectivity index (also known as Randić index and branching index) is one of the most studied graph invariants, which was introduced in 1975 within the study of molecular branching [@r3]. The connectivity index for a graph $G$ is defined as $$R(G)=\displaystyle\sum_{uv\in E(G)}(d_{u}d_{v})^{-\frac{1}{2}},$$ where $d_{u}$ represents the degree of the vertex $u\in V(G)$ and $uv$ is the edge connecting the vertices $u,v$ of $G$. Detail about the mathematical properties of this index can be found in the survey [@Li08], recent papers [@Ali17R; @Cui17; @Li16; @Mansour17; @Das-17; @Gutman-18; @An2] and related references contained therein.
Several modified versions of the connectivity index were appeared in literature. One of such versions is the sum-connectivity index [@Zhou09], which is defined as $$\chi(G)=\sum_{uv\in E(G)}(d_{u}+d_{v})^{-\frac{1}{2}}.$$ Soon after the appearance of sum-connectivity index, its generalized version was proposed [@Zhou10], whose definition is given as $$\chi_{\alpha}(G)=\sum_{uv\in E(G)}(d_{u}+d_{v})^{\alpha},$$ where $\alpha$ is a non-zero real number. In this note, we are concerned with the general sum-connectivity index $\chi_{\alpha}$. Details about $\chi_{\alpha}$ can be found in the recent papers [@Zhu16; @Cui-17; @Tomescu16; @Ali17; @Ali-18; @Wang17; @Ramane17; @Akhter17; @Arshad17; @Jamil17] and related references listed therein. We recall that $2\chi_{-1}(G)=H(G)$, where $H$ is the harmonic index [@Fajtlowicz87], and $\chi_{1}$ coincides with the first Zagreb index [@Gutman72], whose mathematical properties can be found in the recent surveys [@Borovicanin-17; @Borovicanin-MCM19; @Ali18] and related references cited therein.
A polyomino system is a connected geometric figure obtained by concatenating congruent squares side to side in a plane in such a way that the figure divides the plane into one infinite (external) region and a number of finite (internal) regions, and all internal regions must be congruent squares. For possible applications of polyomino systems, see, for example, [@Liu-Ou-13; @Zhang-13; @Gutman-99; @Harary-97] and related references mentioned therein. Two squares in a polyomino system are adjacent if they share a side. A polyomino chain is a polyomino system in which every square is adjacent to at most two other squares. Every polyomino chain can be represented by a graph known as polyomino chain graph. For the sake of simplicity, in the rest of this note, by the term *polyomino chain* we always mean *polyomino chain graph*.
The problem of characterizing graphs having extremum $\chi_{{\alpha}}$ values over the collection of certain polyomino chains, with fixed number of squares, was solved in [@z; @AA2; @y2] for ${\alpha}=1$. The results established in [@deng] give a solution of the aforementioned problem for ${\alpha}=-1$. An and Xiong [@An18] solved this problem for ${\alpha}>1$. While, the same problem was also addressed in [@Ali-2016] and its solution for the case $0<{\alpha}<1$ was reported there. The main purpose of the present note is to give the solution of the problem under consideration for all remaining values of ${\alpha}$, that is, for ${\alpha}<-1$ and $-1<{\alpha}<0$.
Main Results
============
Before proving the main results, we recall some definitions concerning polyomino chains. In a polyomino chain, a square adjacent with only one (respectively two) other square(s) is called terminal (respectively non-terminal) square. A kink is a non-terminal square having a vertex of degree 2. A polyomino chain without kinks is called *linear chain*. A polyomino chain consisting of only kinks and terminal squares is known as zigzag chain. A segment is a maximal linear chain in a polyomino chain, including the kinks and/or terminal squares at its ends. The number of squares in a segment $S_r$ is called its length and is denoted by $l(S_r)$ (or simply by $l_r$). If a polyomino chain $B_{n}$ has segments $S_{1}, S_{2},...,S_{s}$ then the vector $(l_{1},l_{2},...,l_{s})$ is called length vector of $B_n$. A segment $S_r$ is said to be external (internal, respectively) segment if $S_r$ contains (does not contain, respectively) terminal square.
[@z] For $2\leq i\leq s-1$ and $1\leq j\leq s$, $$\alpha_{i} =
\begin{cases}
1 & \text{if } l_i=2\\
0 & \text{if } l_i\geq3
\end{cases}$$ $$\beta_{j}=
\begin{cases}
1 & \text{if } l_j=2\\
0 & \text{if } l_j\geq3
\end{cases}$$ and $\alpha_{1}=\alpha_{s}=0.$
Let $\Omega_{n}$ be the collection of all those polyomino chains, having $n$ squares, in which no internal segment of length 3 has edge connecting the vertices of degree 3.
\[t1\] [@Ali-2016] Let $B_{n}\in\Omega_{n}$ be a polyomino chain having $s$ segment(s) $S_{1}, S_{2},S_{3},...,S_{s}$ with the length vector $(l_{1},l_{2},...,l_{s})$. Then, $$\begin{aligned}
\chi_\alpha(B_{n})&=&3\cdot 6^\alpha n+(2\cdot5^\alpha-6^{\alpha+1}+4\cdot 7^\alpha)s
+(2\cdot 4^\alpha+2\cdot 5^\alpha+6^\alpha-4\cdot 7^\alpha)\\
&&+ \ (2\cdot6^\alpha-5^\alpha- 7^\alpha)[\beta_{1}+\beta_{s}]+ \ (5\cdot 6^\alpha-2\cdot 5^\alpha-4\cdot 7^\alpha+ 8^\alpha)\sum_{i=1}^{s}\alpha_{i}.\end{aligned}$$
Let $$f({\alpha})=2\cdot5^\alpha-6^{\alpha+1}+4\cdot 7^\alpha, \ g({\alpha})= 2\cdot6^\alpha-5^\alpha- 7^\alpha,$$ $$h({\alpha})=5\cdot 6^\alpha-2\cdot 5^\alpha-4\cdot 7^\alpha+ 8^\alpha.$$ Furthermore, let $\Psi_{\chi_\alpha}(S_{1})=f({\alpha})+g({\alpha})\beta_{1}, \ \Psi_{\chi_\alpha}(S_{s})=f({\alpha})+g({\alpha})\beta_{s}$ and for $s\geq3$, assume that $\Psi_{\chi_\alpha}(S_{i})=f({\alpha})+h({\alpha})\alpha_{i}$ where $2\leq i\leq s-1$. Then $$\label{Eq.11}
\Psi_{\chi_\alpha}(B_{n})=\sum_{i=1}^{s}\Psi_{\chi_\alpha}(S_{i})=f({\alpha})s+g({\alpha})(\beta_{1}+\beta_{s})
+h({\alpha})\sum_{i=1}^{s}\alpha_{i} \ .$$ Hence, the formula given in Theorem \[t1\] can be rewritten as $$\label{Eq.12}
\chi_\alpha(B_{n})=3\cdot 6^\alpha n +(2\cdot 4^\alpha+2\cdot 5^\alpha+6^\alpha-4\cdot 7^\alpha)+\Psi_{\chi_\alpha}(B_{n}).$$ The next lemma is a direct consequence of the relation (\[Eq.12\]).
\[L1\] [@Ali-2016] For any polyomino chain $B_{n}$ having $n\geq3$ squares, $\chi_\alpha(B_{n})$ is maximum (respectively minimum) if and only if $\Psi_{\chi_\alpha}(B_{n})$ is maximum (respectively minimum).
Lemma \[L1\] will play a vital role in proving the main results of the present note.
\[t22\] [@Ali-2016] Let $B_{n}\in\Omega_{n}$ be a polyomino with $n\geq3$ squares. If $f({\alpha})$, $f({\alpha})+2g({\alpha})$ and $f({\alpha})+2h({\alpha})$ are all negative, then $$\chi_\alpha(Z_{n})\leq \chi_\alpha(B_{n})\leq \chi_\alpha(L_{n}).$$ Right (respectively left) equality holds if and only if $B_{n}\cong L_{n}$ (respectively $B_{n}\cong Z_{n}$).
\[p-1\] Let $B_{n}\in\Omega_{n}$ be a polyomino chain having $n\geq3$ squares. Let $x_0\approx -3.09997$ be a root of the equation $f({\alpha})=0$. Then, for $x_0 < \alpha <0 $, it holds that $$\chi_\alpha(Z_{n}) \leq \chi_\alpha(B_{n})\leq \chi_\alpha(L_{n}),$$ with right (respectively left) equality if and only if $B_{n}\cong L_{n}$ (respectively $B_{n}\cong Z_{n}$).
It can be easily checked that $f({\alpha})$, $f({\alpha})+2g({\alpha})$ and $f({\alpha})+2h({\alpha})$ are negative for $x_0<\alpha<0$, and hence, from Lemma \[t22\], the required result follows.
\[p-2\] Let $B_{n}\in\Omega_{n}$ be a polyomino with $n\geq3$ squares. Let $x_0\approx -3.09997$ be a root of the equation $f({\alpha})=0$. Then, for $\alpha\le x_0$, the following inequality holds $$\chi_\alpha(B_{n}) \ge \chi_\alpha(Z_{n}),$$ with equality if and only if $B_{n}\cong Z_{n}$.
We note that $f({\alpha})$ is non-negative and both $g({\alpha})$, $h({\alpha})$ are negative for $\alpha\le x_0 \approx -3.09997$. Suppose that the polyomino chain $B_{n}^{*}\in\Omega_{n}$ has the minimum $\Psi_{\chi_\alpha}$ value for $\alpha\le x_0$. Further suppose that $S_{1}, S_{2},...,S_{s}$ be the segments of $B_{n}^{*}$ with the length vector $(l_{1},l_{2},...,l_{s})$. It holds that $$\Psi_{\chi_\alpha}(Z_{n}) = 2f({\alpha}) + 2g({\alpha}) + (n-3)(f({\alpha}) + h({\alpha})) \le 2f({\alpha}) + 2g({\alpha}) < f({\alpha}) = \Psi_{\chi_\alpha}(L_{n}),$$ which implies that $s\ge2$.
If at least one of external segments of $B_{n}^{*}$ has length greater than 2. Without loss of generality, assume that $l_{1}\geq3$. Then, there exist a polyomino chain $B_{n}^{(1)}\in\Omega_{n}$ having length vector $(\underbrace{2,2,\cdots,2}_{(l_1-1)-times},l_{2},...,l_{s})$ and $$\Psi_{\chi_\alpha}(B_{n}^{(1)})-\Psi_{\chi_\alpha}(B_{n}^{*})= g({\alpha}) + \left(l_1 -2\right)(f({\alpha}) + h({\alpha}))\le f({\alpha}) + g({\alpha}) + h({\alpha})<0,$$ for $\alpha\le x_0 \approx -3.09997$, which is a contradiction to the definition of $B_{n}^{*}$. Hence both external segments of $B_{n}^{*}$ must have length 2.
If some internal segment of $B_{n}^{*}$ has length greater than 2, say $l_{j}\geq3$ where $2\leq j\leq s-1$ and $s\geq3$. Then, there exists a polyomino chain $B_{n}^{(2)}\in\Omega_{n}$ having length vector $(l_{1},l_{2},...,l_{j-1},2,l_{j}-1,...,l_{s})$ and $$\Psi_{\chi_\alpha}(B_{n}^{(2)})-\Psi_{\chi_\alpha}(B_{n}^{*})= f({\alpha}) + (1+y)h({\alpha})<0, \ \ \ (\text{where $y=0$ or 1})$$ for $\alpha\le x_0 \approx -3.09997$, which is again a contradiction. Hence, every internal segment of $B_{n}^{*}$ has length 2.
Therefore, $B_{n}^{*}\cong Z_{n}$ and from Lemma \[L1\], the desired result follows.
\[p-3\] Let $B_{n}\in\Omega_{n}$ be a polyomino with $n\geq3$ squares. Let $\alpha\approx -3.09997$ be a root of the equation $f({\alpha})=0$. Then, the following inequality holds $$\chi_\alpha(B_{n}) \le 3\cdot 6^\alpha n +(2\cdot 4^\alpha+2\cdot 5^\alpha+6^\alpha-4\cdot 7^\alpha),$$ with equality if and only if $B_{n}$ does not contain any segment of length 2.
From Equation (\[Eq.11\]), it follows that $$\Psi_{\chi_\alpha}(B_{n})=g({\alpha})(\beta_{1}+\beta_{s})+h({\alpha})\sum_{i=1}^{s}\alpha_{i} \le 0 .$$ Clearly, the equality $\Psi_{\chi_\alpha}(B_{n})=0$ holds if and only if $B_{n}$ does not contain any segment of length 2. Hence, by using Lemma \[L1\], we have the required result.
Let $\mathcal{Z}_{n}^*$ be a subclass of $\Omega_{n}$ consisting of those polyomino chains which do not contain any segment of length equal to 2 or greater than 4, and contain at most one segment of length 4. Let $\mathcal{Z}_{n}$ be a subclass of $\Omega_{n}$ consisting of those polyomino chains in which every internal segment (if exists) has length 3 or 4, every external segment has length at most 4, at most one external segment has length 2, at most one segment has length 4 and if some internal segment has length 4 then both the external segments have length 3. Let $Z_{n}^\dag\in\Omega_{n}$ be the polyomino chain in which every internal segment (if exists) has length 3, every external segment has length at most 3 and at most one external segment has length 2.
\[p-4\] Let $B_{n}\in\Omega_{n}$ be a polyomino with $n\geq3$ squares. Let $x_0 \approx -3.09997$ and $x_1 \approx -5.46343$ be the roots of the equations $f({\alpha})=0$ and $f({\alpha})+g({\alpha})=0$, respectively. Then, for $x_1 < \alpha < x_0$, the following inequality holds $$\label{Eq-Z1}
\chi_\alpha(B_{n}) \le \chi_\alpha\left( Z_{n}^* \right),$$ with equality if and only if $B_n \cong Z_{n}^*\in \mathcal{Z}_{n}^*$. Also, for $\alpha = x_1$, the following inequality holds $$\label{Eq-Z3}
\chi_\alpha(B_{n}) \le \chi_\alpha\left( Z_{n}^\maltese \right),$$ with equality if and only if $B_n \cong Z_{n}^\maltese \in \mathcal{Z}_{n}$. Furthermore, for $\alpha < x_1$, the following inequality holds $$\label{Eq-Z2}
\chi_\alpha(B_{n}) \le \chi_\alpha\left( Z_{n}^\dag \right),$$ with equality if and only if $B_n \cong Z_{n}^\dag$.
For $n=3$, the result is obvious. We assume that $n\ge4$. It can be easily checked that $f({\alpha})$ is positive and both $g({\alpha})$, $h({\alpha})$ are negative for $x_1 < \alpha < x_0$. Suppose that for the polyomino chain $B_{n}^{*}\in\Omega_{n}$, $\Psi_{\chi_\alpha}(B_{n}^{*})$ is maximum for $\alpha < x_0$. Let $B_{n}^{*}$ has $s$ segments $S_{1}, S_{2},...,S_{s}$ with the length vector $(l_{1},l_{2},...,l_{s})$.
If $s\ge 3$ and at least one of internal segments of $B_{n}^{*}$ has length 2, say $l_{i}=2$ for $2\le i \le s-1$, then there exists a polyomino chain $B_{n}^{(1)}\in\Omega_{n}$ having length vector $$\begin{cases}
(l_1,l_{2},..., l_{s-1} + l_{s}-1) & \text{ if } i=s-1, \\
(l_1,l_{2},..., l_{i-1}, l_i + l_{i+1}-1, l_{i+2},...,l_{s}) & \text{ otherwise},
\end{cases}$$ and $$\Psi_{\chi_\alpha}(B_{n}^{*})-\Psi_{\chi_\alpha}(B_{n}^{(1)})=
\begin{cases}
f({\alpha}) + x\cdot g({\alpha}) + h({\alpha})<0 & \text{ if } i=s-1, \\
f({\alpha}) + (1+ y)h({\alpha})<0 & \text{ otherwise},
\end{cases}$$ for $\alpha < x_0$, where $x,y\in \{0,1 \}$. This is a contradiction. Hence, every internal segment (if exists) of $B_{n}^{*}$ has length greater than 2.
If at least one of segments of $B_{n}^{*}$ has length greater than 4, say $l_{i}\ge5$ for $1\le i \le s$, then there exists a polyomino chain $B_{n}^{(2)}\in\Omega_{n}$ having length vector $$\begin{cases}
(3,l_1-2,l_{2},l_3,...,l_{s}) & \text{ if } i=1, \\
(l_1,l_{2},..., l_{i-1}, 3, l_i - 2, l_{i+1}, l_{i+2},...,l_{s}) & \text{ if } 2\le i \le s-1,\\
(l_1,l_{2},..., l_{s-1}, 3, l_s - 2) & \text{ if } i = s,
\end{cases}$$ and $$\Psi_{\chi_\alpha}(B_{n}^{*})-\Psi_{\chi_\alpha}(B_{n}^{(2)})= -f({\alpha}) <0,$$ a contradiction. Hence, every segment of $B_{n}^{*}$ has length less than than 5.
If at least two segments of $B_{n}^{*}$ have length 4, say $l_i=l_j=4$ for $1\le i,j \le s$, then there exists a polyomino chain $B_{n}^{(3)}\in\Omega_{n}$ having length vector $(3,l_1,l_{2},..., l_{i-1}, l_{i}-1, l_{i+1},...,l_{j-1}, l_{j}-1, l_{j+1},...,l_{s})$ and $$\Psi_{\chi_\alpha}(B_{n}^{*})-\Psi_{\chi_\alpha}(B_{n}^{(3)})= -f({\alpha}) <0,$$ a contradiction. Hence, $B_{n}^{*}$ contains at most one segment of length 4.
If both the external segments of $B_{n}^{*}$ have length 2, then ($s\ge3$ because $n\ge4$) there exists a polyomino chain $B_{n}^{(4)}\in\Omega_{n}$ having length vector $(l_1+1,l_{2},l_3,...,l_{s-1})$ and $$\Psi_{\chi_\alpha}(B_{n}^{*})-\Psi_{\chi_\alpha}(B_{n}^{(4)})= f({\alpha}) + 2g({\alpha})<0,$$ which is again a contradiction. Hence, at most one external segment has length 2. In what follows, without loss of generality, we assume that $l_s=2$ whenever some external segment has length 2.
If some external segment of $B_{n}^{*}$ has length greater 2, say $l_{1}=2$, then there exists a polyomino chain $B_{n}^{(5)}\in\Omega_{n}$ having length vector $(l_{2}+1, l_{3}, l_{4},...,l_{s})$ and $$\Psi_{\chi_\alpha}(B_{n}^{*})-\Psi_{\chi_\alpha}(B_{n}^{(5)})= f({\alpha}) + g({\alpha})<0, \ \text{ (because $l_2\ge3$)}$$ for $x_1 < \alpha < x_0$, which is again a contradiction. Hence, if $\Psi_{\chi_\alpha}(B_{n}^{*})$ is maximum for $x_1<\alpha < x_0$ then every external segment of $B_{n}^{*}$ has length greater than 2. Therefore, if $\Psi_{\chi_\alpha}(B_{n}^{*})$ is maximum for $x_1<\alpha < x_0$ then $B_{n}^{*}\cong Z_{n}^*$ and thence from Lemma \[L1\], inequality (\[Eq-Z1\]) follows.
In the remaining proof, we assume $\alpha \le x_1$.
If $B_{n}^{*}$ contains a segment of length 4, say $l_{i}=4$ for $1\le i \le s$, then there exists a polyomino chain $B_{n}^{(6)}\in\Omega_{n}$ having length vector $$\begin{cases}
(2,l_1-1,l_{2},l_3,...,l_{s}) & \text{ if } i=1, \\
(l_1,l_{2},..., l_{i-1}, l_i -1 , l_{i+1}, l_{i+2},...,l_{s}+1) & \text{ if } 2\le i \le s-1 \text{ and } l_s=2,\\
(l_1,l_{2},..., l_{i-1}, l_i -1 , l_{i+1}, l_{i+2},...,l_{s},2) & \text{ if } 2\le i \le s-1 \text{ and } l_s=3,\\
(l_1,l_{2},..., l_{s-1}, l_s - 1,2) & \text{ if } i = s,
\end{cases}$$ and $$\Psi_{\chi_\alpha}(B_{n}^{*})-\Psi_{\chi_\alpha}(B_{n}^{(6)})=
\begin{cases}
g({\alpha}) & \text{ if } 2\le i \le s-1 \text{ and } l_s=2,\\
-f({\alpha}) - g({\alpha}) & \text{ otherwise}.
\end{cases}$$ This last equation together with the fact that for $\alpha < x_1$, both $g({\alpha})$ and $-f({\alpha}) - g({\alpha})$ are negative, gives a contradiction. The same equation together with the fact that for $\alpha = x_1$, only $g({\alpha})$ is negative, arises also a contradiction if $2\le i \le s-1$ and $l_s=2$. Therefore, if $\Psi_{\chi_\alpha}(B_{n}^{*})$ is maximum for $\alpha < x_1$ then $B_{n}^{*}\cong Z_{n}^\dag$ and if $\Psi_{\chi_\alpha}(B_{n}^{*})$ is maximum for $\alpha = x_1$ then $B_{n}^{*} \in \mathcal{Z}_{n}$, and thence from Lemma \[L1\], inequalities (\[Eq-Z3\]) and (\[Eq-Z2\]) follow.
Propositions \[p-1\], \[p-2\], \[p-3\] and \[p-4\], together with the already reported results in [@z; @AA2; @y2; @deng; @An18; @Ali-2016], yield Table \[table-1\] which gives information about the polyomino chains having extremum $\chi_\alpha$ values in the collection $\Omega_{n}$ for $n\ge3$.
Polyomino Chain(s) with Maximal $\chi_\alpha$ Value Polyomino Chain(s) with Minimal $\chi_\alpha$ Value
------------------- ----------------------------------------------------- -----------------------------------------------------
$\alpha >0 $ $Z_n$ $L_n$
$x_0 <\alpha <0$ $L_n$ $Z_n$
$\alpha = x_0$ chains having no segment of length 2 $Z_n$
$x_1 <\alpha<x_0$ members of $\mathcal{Z}_{n}^*$ $Z_n$
$\alpha= x_1$ members of $\mathcal{Z}_{n}$ $Z_n$
$\alpha < x_1$ $Z_{n}^\dag$ $Z_n$
: Polyomino chains having extremum $\chi_\alpha$ values in the collection $\Omega_{n}$ for $n\ge3$.[]{data-label="table-1"}
[999]{}
M. An, L. Xiong, Extremal polyomino chains with respect to general Randić index, *J. Comb. Optim.* **31** (2016) 635-647.
M. An, L. Xiong, Extremal polyomino chains with respect to general sum-connectivity index, *Ars Combinatoria* [**131**]{} (2017) 255–271.
S. Akhter, M. Imran, Z. Raza, Bounds for the general sum-connectivity index of composite graphs, *J. Inequal. Appl.* **2017** (2017) 76.
A. Ali, An alternative but short proof of a result of Zhu and Lu concerning general sum-connectivity index, *Asian-European J. Math.* **11** (2018) DOI: 10.1142/S1793557118500304, in press.
A. Ali, A. A. Bhatti, Z. Raza, A note on the zeroth-order general Randić index of cacti and polyomino chains, *Iranian J. Math. Chem.* **5** (2014) 143-152.
A. Ali, A. A. Bhatti, Z. Raza, Some vertex-degree-based topological indices of polyomino chains, *J. Comput. Theor. Nanosci.* **12** (2015) 2101-2107.
A. Ali, D. Dimitrov, On the extremal graphs with respect to bond incident degree indices, *Discrete Appl. Math.* **238** (2018) 32–40.
A. Ali, Z. Du, On the difference between atom-bond connectivity index and Randić index of binary and chemical trees, *Int. J. Quantum Chem.* **117** (2017) e25446.
A. Ali, I. Gutman, E. Milovanović, I. Milovanović, Sum of powers of the degrees of graphs: extremal results and bounds, *MATCH Commun. Math. Comput. Chem.* **80** (2018) 5–84.
A. Ali, Z. Raza, A. A. Bhatti, Bond incident degree (BID) indices of polyomino chains: A unified approach, *Appl. Math. Comput.* **287-288** (2016) 28–37.
M. Arshad, I. Tomescu, Maximum general sum-connectivity index with $-1\le {\alpha}<0$ for bicyclic graphs, *Math. Reports* **19** (2017) 93–96.
J. A. Bondy, U. S. R. Murty, [*Graph Theory*]{}, Springer, London, 2008.
B. Borovićanin, K. C. Das, B. Furtula, I. Gutman, Bounds for Zagreb indices, [*MATCH Commun. Math. Comput. Chem.*]{} [**78**]{} (2017) 17–100. B. Borovićanin, K. C. Das, B. Furtula, I. Gutman, Zagreb indices: Bounds and extremal graphs, in: I. Gutman, B. Furtula, K. C. Das, E. Milovanović, I. Milovanović (Eds.), [*Bounds in Chemical Graph Theory – Basics*]{}, Univ. Kragujevac, Kragujevac, 2017, pp. 67–153.
Q. Cui, L. Zhong, On the general sum-connectivity index of trees with given number of pendent vertices, *Discrete Appl. Math.* **222** (2017) 213–221.
Q. Cui, L. Zhong, The general Randić index of trees with given number of pendent vertices, *Appl. Math. Comput.* **302** (2017) 111–121.
K. C. Das, S. Balachandran, I. Gutman, Inverse degree, Randić index and harmonic index of graphs, *Appl. Anal. Discr. Math.* **11** (2017) 304–313.
H. Deng, S. Balachandran, S. K. Ayyaswamy, Y. B. Venkatakrishnan, The harmonic indices of polyomino chains, *Natl. Acad. Sci. Lett.* **37** (2014) 451-455.
S. Fajtlowicz, On conjectures of Graffiti-II, *Congr. Numer.* **60** (1987) 187–197.
I. Gutman, B. Furtula, V. Katanić, Randić index and information, *AKCE Int. J. Graphs Comb.* DOI:10.1016/j.akcej.2017.09.006, in press.
I. Gutman, N. Trinajstić, Graph theory and molecular orbitals. Total $\pi$-electron energy of alternant hydrocarbons, *Chem. Phys. Lett.* **17** (1972) 535–538.
I. Gutman, S. Klavzar, A. Rajapakse, Average distances in square–cell configurations. *Int. J. Quant. Chem.* **76** (1999) 611–617.
F. Harary, [*Graph Theory*]{}, Addison–Wesley, Reading, 1969.
F. Harary, P.G. Mezey, Cell–shedding transformations, equivalence relations, and similarity measures for square–cell configurations. *Int. J. Quant. Chem.* **62** (1997) 353–361.
M. K. Jamil, I. Tomescu, Minimum general sum-connectivity index of trees and unicyclic graphs having a given matching number, *Discrete Appl. Math.* **222** (2017) 143–150.
X. Li, Y. Shi, A survey on the Randić index, *MATCH Commun. Math. Comput. Chem.* **59** (2008) 127–156.
F. Li, Q. Ye, The general connectivity indices of fluoranthene-type benzenoid systems, *Appl. Math. Comput.* **273** (2016) 897–911.
S. Liu, J. Ou, On maximal resonance of polyomino graphs, *J. Math. Chem.* **51** (2013) 603–619.
T. Mansour, M. A. Rostami, S. Elumalai, B. A. Xavier, Correcting a paper on the Randić and geometric-arithmetic indices, *Turk. J. Math.* **41** (2017) 27–32.
H. S. Ramane, V. V. Manjalapur, I. Gutman, General sum-connectivity index, general product-connectivity index, general Zagreb index and coindices of line graph of subdivision graphs, *AKCE Int. J. Graphs Comb.* **14** (2017) 92–100.
M. Randić, On characterization of molecular branching, *J. Am. Chem. Soc.* **97** (1975) 6609–6615.
I. Tomescu, On the general sum-connectivity index of connected graphs with given order and girth, *Electron. J. Graph Theory Appl.* **4** (2016) 1–7.
H. Wang, J.-B. Liu, S. Wang, W. Gao, S. Akhter, M. Imran, M. R. Farahani, Sharp bounds for the general sum-connectivity indices of transformation graphs, *Discrete Dyn. Nat. Soc.* **2017** (2017) 2941615.
Z. Yarahmadi, A. R. Ashrafi, S. Moradi, Extremal polyomino chains with respect to Zagreb indices, *Appl. Math. Lett.* **25** (2012) 166-171.
L. Zhang, S. Wei, F. Lu, The number of Kekulé structures of polyominos on the torus, *J. Math. Chem.* **51** (2013) 354–368.
B. Zhou, N. Trinajstić, On a novel connectivity index, *J. Math. Chem.* **46** (2009) 1252–1270.
B. Zhou, N. Trinajstić, On general sum-connectivity index, *J. Math. Chem.* **47** (2010) 210–218.
Z. Zhu, H. Lu, On the general sum-connectivity index of tricyclic graphs, *J. Appl. Math. Comput.* **51** (2016) 177–188.
[^1]: Corresponding author.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'X-ray observations of pre-main sequence (pre-MS) stars of M-type probe coronal emission and offer a means to investigate magnetic activity at the stellar-substellar boundary. Recent observations of main sequence (MS) stars at this boundary display a decrease in fractional X-ray luminosity ($L_{X}$/$L_{bol}$) by almost two orders of magnitude for spectral types M7 and later. We investigate magnetic activity and search for a decrease in X-ray emission in the pre-MS progenitors of these MS stars. We present XMM-Newton X-ray observations and preliminary results for $\sim$10 nearby (30-70 pc), very low mass pre-MS stars in the relatively unexplored age range of 10-30 Myr. We compare the fractional X-ray luminosities of these 10-30 Myr old stars to younger (1-3 Myr) pre-MS brown dwarfs and find no dependence on spectral type or age suggesting that X-ray activity declines at an age later than $\sim$30 Myr in these very low-mass stars.'
---
Introduction
============
The early evolution of magnetic activity in very low mass pre-MS stars –stars of mid-M-type, which lie near the H-burning limit of 0.08 $M_{\odot}$– is very poorly understood. Yet understanding their pre-MS evolution is crucial for determining the emerging differences between very low-mass MS stars and brown dwarfs. X-ray emission offers a means to indirectly probe the effects of internal and surface magnetic activity in both pre-MS and MS stars alike ([@Vidotto2014 Vidotto et al. 2014]). Pre-MS and MS M-type stars are magnetically active and thus can be bright X-ray sources, as indicated by their high values of ($L_{X}$/$L_{bol} \sim 10^{-3}$). However, observations of nearby late M-type MS stars suggest that stars of $\sim$M7 and later appear to be under luminous in X-rays (e.g., $L_{X}$/$L_{bol}$ $\sim$ $10^{-5}$; [@Berger2010 Berger et al. 2010]). The narrow range of spectral types where these M stars become X-ray under luminous is roughly the same spectral type where a transition to predominantly neutral atmospheres occurs ([@Mohanty2002 Mohanty et al. 2002]). [@Berger2006 Berger (2006)] concluded that the decrease in X-ray activity (as well as H$\alpha$) toward late M-types is related to changes in magnetic field configuration or the decreasing ionization fractions in the atmospheres of these stars.
By determining the age at which this dramatic decrease in X-ray activity occurs for M-type stars, we can gain insight into the early pre-MS stellar evolution of such stars which lie at the low-mass-star/brown dwarf (H-burning) boundary. A recent survey combining GALEX, 2MASS, WISE and catalog proper motions have revealed a population of nearby late-M-type stars in the 10-30 Myr age range ([@Rodriguez2013 Rodriguez et al. 2013]) where X-ray activity of such stars has remained, until now, essentially unexplored.
Data and Preliminary Results
============================
We performed XMM-Newton EPIC X-ray observations of 8 nearby ($<$70 pc) $\sim$M5 members of the Tuc-Hor and $\beta$ Pic moving groups (ages $\sim$ 30 Myr and 12 Myr, respectively; Rodriguez et al. 2013 and ref. therein). Stellar evolution models ([@Dantona1997 D’Antona & Mazzitelli 1997]) suggest pre-MS stars of this age and spectral type will evolve to become MS $\sim$M7 (i.e., they may be progenitors of the under luminous MS M7 stars). Standard one and two temperature thermal plasma models were fit to the data to determine spectral parameters such as plasma temperature and $L_{X}$. Bolometric luminosities for each of our sources was estimated using their J band flux and the intrinsic colors of 5-30 Myr stars from [@Pecaut2013 Pecaut & Mamajek (2013)]. The fractional X-ray luminosity for each source is shown in Fig. 1 and compared to pre-MS stars of similar spectral type in younger (e.g 1-3 Myr) star-forming regions.
Conclusions
===========
We find no trend of decreasing fractional X-ray luminosity with age in these 10-30 Myr $\sim$M5 stars (Fig. 1). If MS stars of $\sim$M7 and later are under luminous in X-rays, these preliminary results suggest that either X-ray activity decreases at ages later than $\sim$30 Myr or that our sample of pre-MS stars is unusually magnetically active. The latter scenario may be more likely due to the fact that the sample from which we chose our sources required a GALEX UV detection (i.e., only M stars that were UV bright). UV emission is also an indicator of magnetic activity and thus our sample might be biased towards magnetically active (i.e., X-ray bright) pre-MS stars. More X-ray observations of both UV bright and UV faint mid-to-late M-type pre-MS stars are required to explore the age at which X-ray activity may diminish. A more detailed analysis of these data will be presented in Principe et al. (2015), in prep.
![Fractional X-ray luminosity as a function of M spectral subclass for $\sim$1-3 Myr old pre-MS stars in the Orion Nebula Cluster (asterisk), Taurus (cross), Cha I (triangle) and preliminary results of $\sim$10-30 Myr old Tuc-Hor and $\beta$ Pic moving group members (diamonds). Figure originally from [@Stelzer2007 Stelzer & Micela (2007)] and modified to include preliminary results. []{data-label="fig:grains"}](mstar_plot_lxlbol.eps){width="2.87in"}
2006, *ApJ*, 648, 629
2010, *ApJ*, 709, 332
1997, *Memorie della Societ[á]{} Astronomia Italiana*, 68, 807
2002, *ApJ*, 571, 469
2013, *ApJS*, 208, 9
2013, *ApJ*, 774, 101
2007, *A$\&$A*, 474, 129
2014, *MNRAS*, 441, 2361
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Understanding the cosmic ray (CR) ionization rate is crucial in order to simulate the dynamics of, and interpret the chemical species observed in molecular clouds. Calculating the CR ionization rate requires both accurate knowledge of the spectrum of MeV to GeV protons at the edge of the cloud as well as a model for the propagation of CRs into molecular clouds. Some models for the propagation of CRs in molecular clouds assume the CRs to stream freely along magnetic field lines, while in others they propagate diffusively due to resonant scattering off of magnetic disturbances excited by MHD turbulence present in the medium. We discuss the conditions under which CR diffusion can operate in a molecular cloud, calculate the local CR spectrum and ionization rate in both a free-streaming and diffusive propagation model, and highlight the different results from the two models. We also apply these two models to the propagation through the ISM to obtain the spectrum seen by Voyager 1, and show that such a spectrum favors a diffusive propagation model.'
author:
- 'Kedron Silsbee$^1$, Alexei V. Ivlev$^1$'
bibliography:
- 'diffusionNote.bib'
title: 'Diffusive Versus Free-Streaming Cosmic Ray Transport in Molecular Clouds'
---
Introduction
============
Cosmic rays (CRs) provide the dominant source of ionization in molecular clouds at visual extinctions greater than 1 to a few, depending on conditions, which corresponds to H$_2$ column depths greater than 1 to a few $\times 10^{21}$ cm$^{-2}$ [@McKee89; @Keto08]. They affect the gas-phase chemistry [@Dalgarno06], chemistry that occurs in dust grains [@Shingledecker18], and contribute heating to cold cores of molecular clouds [@Glassgold12; @Galli15].
Many widely adopted models for the propagation of CRs in molecular clouds [@Padovani09; @Padovani18] assume that CRs stream freely along magnetic field lines. Under this assumption, @Padovani18 showed that the CR ionization rate at a point in the cloud is a function of the [*effective*]{} column density $N$ to that point, integrated along the magnetic field lines.
There is also discussion in the literature of diffusive CR propagation. Turbulence can excite MHD waves that scatter the CRs’ pitch angles [@Kulsrud69], leading to spatial diffusion. This turbulence can arise from the anisotropy in the CR distribution function which arises near the cloud in response to CR absorption in the cloud center. The role of such turbulence has been discussed in several works [e.g., @Skilling76; @Morlino15; @Ivlev18]. However, it was found that for clouds with $N \lesssim 10^{23}$ cm$^{-2}$, the effect of such self-generated turbulence on CR penetration is only marginally significant [@Dogiel18].
Diffusive CR transport can also occur due to pre-existing turbulence. @Schlickeiser16 calculated the CR spectrum and resulting ionization rate, assuming an energy-dependent diffusion coefficient derived in @Schlickeiser10 and using the CR spectrum derived from the Voyager 1 results.
There is increasing evidence for both substantial variability from object to object, as well as a steep dependence of the ionization rate on the column density of the cloud. Estimates of the primary CR ionization rate per hydrogen atom $\zeta_p$ from observations of OH$^+$ in low-density clouds give values ranging from $3.9 \times 10^{16}$ s$^{-1}$ up to $1.6 \times 10^{-15}$ s$^{-1}$ [@Bacalla18]. A paper by @Neufeld17 uses H$_3^+$ observations in different clouds with known column densities to determine $\zeta_p$ as a function of column density $N$. While there is significant uncertainty in the results, they suggest a quite steep dependence of $\zeta_p$ on $N$. Finally, @Galli15 argue that the temperature and molecular abundance profile in the center of the starless core L1544 is best fit by $\zeta_p \sim 10^{-17}$ s$^{-1}$ or even lower.
In this paper we discuss the possible role of pre-existing MHD turbulence. Envelopes of molecular clouds are thought to be turbulent environments. There is some uncertainty, however, as to whether the MHD waves associated with the turbulence would have sufficient energy at small enough scales to be resonant with the CRs responsible for the majority of the ionization. Radio scintillation observations [e.g., @Armstrong95] suggest that turbulence in the ISM extends to scales at least as small as $10^{10}$ cm, comparable to the gyroradius of a sub-relativistic proton, but it is not clear that this result is relevant to molecular clouds. As we show in Section \[molCloudIon\], assuming diffusive propagation of CRs into molecular clouds to take place, it would create a steep dependence of the ionization rate on column density although this is only likely up to column densities of $\sim 10^{21}$ cm$^{-2}$ under conditions appropriate for local molecular clouds.
The CR ionization rate depends on both the propagation model for CRs, and on their spectrum at the edge of the cloud. Ionization at column densities in the range of $10^{20} - 10^{23}$ cm$^{-2}$ is dominated by CR protons with energies from 1 MeV to 1 GeV. Unfortunately, the spectrum of such particles cannot be measured accurately from near Earth because they are largely excluded by the solar wind [@Potgieter13]. The Voyager 1 probe has measured the spectrum of Galactic CRs down to 3 MeV [@Cummings16]. However, the magnetic field direction measured by the probe has not changed, as it would be expected to if Voyager were really in a region of space beyond the influence of solar modulation [@Gloeckler15]. Furthermore, @Padovani18 and @Phan18 noted that the proton and electron spectra from Voyager were too low by about a factor of 10 to explain the values of $\zeta_p$ observed in nearby molecular clouds. For these reasons, there is still considerable uncertainty about the density of low-energy CRs impinging on molecular clouds.
Models for the acceleration of CRs in shocks suggest that they should act as power-law source functions for CRs \[see e.g. @Drury83\]. This is very different from the Voyager spectrum, which shows a broad turnover around 30 MeV. Recently, there has been work to reproduce the spectrum seen by Voyager with the code GALPROP, using a complicated model including diffusion, advection, reacceleration, adiabatic momentum gain and loss and several energy loss processes [@Bischoff19]. They are able to well reproduce the spectrum seen by Voyager. In this paper we consider a simpler model, which includes the effect of a shell of high-density material surrounding the local bubble with magnetic field nearly parallel to the shell [@Alves18]. In section \[voyager\], we look at the spectrum of CRs that would be seen by Voyager after propagation through this shell. We find that diffusive propagation within this thin dense region could attenuate the power-law source spectrum of low-energy CRs, to produce something qualitatively resembling the Voyager spectrum. The value of the column density required for such attenuation is much more reasonable than that predicted by a model of free-streaming propagation.
Diffusive propagation Model
===========================
In this section we calculate the CR spectrum as a function of $N$, assuming CRs propagate diffusively through an attenuating column. The propagation is modelled as occurring due to a combination of diffusion along the magnetic field and energy losses due to ionization.
Following @Skilling75, we use a simplified expression for the CR diffusion coefficient $D$ due to the presence of weak MHD turbulence: $$D(E) = \frac{vB^2}{6\pi^2\mu_*k^2W},
\label{Dexp}$$ where $v$ is the speed of the CR particle, $B$ the magnetic field strength, and $\mu_*$ is the “effective" cosine of the resonant pitch angle. Furthermore, $k$ is the wavenumber of the resonant MHD wave, and $W(k)$ is the spectral energy density of MHD waves. $k$ in Equation is related to $E$ via the cyclotron resonance condition, which is expressed in terms of $\mu_*$ as $$k_{\rm res}(E) = \frac{m \Omega}{\mu_*p(E)}.
\label{kres}$$ Here, $m$ is the mass of the CR particle, $\Omega$ is the CR gyrofrequency given by $$\Omega = \frac{eB}{mc},$$ $c$ is the speed of light and $e$ the electron charge. We assume that $W$ is equal to the ion kinetic energy density of the weak turbulence: $$W(k) = \frac{\rho_{\rm ion} v^2_{\rm turb}(k)}{2k},$$ where $\rho_{\rm ion}$ is the mass density of ions in the medium. We assume a power law spectrum of the turbulence, such that $$v_{\rm turb} = v_* \left(\frac{k_*}{k}\right)^\lambda,
\label{vturb}$$ where $\lambda$ is 1/3 for a Kolmogorov turbulent spectrum, and 1/4 for a Kraichnan spectrum. Putting all of this together, we find $D \propto E^{1-\lambda} \rho_{\rm ion}^{-1}$.
Let us consider the flux of particles entering the column from the source along a field line. Let $s$ be the coordinate of distance along this line. The source is located at $s = 0$, and $s$ increases with increasing column away from the source. The transport equation for $n(E, s)$, the differential density of particles with energy $E$ at position $s$, considering spatial diffusion and energy losses is given by $$\frac{\partial n}{\partial t} = \frac{\partial}{\partial s}\left(D \frac{dn}{ds}\right) - \frac{\partial}{\partial E}\left(\frac{dE}{dt}n\right),
\label{transport}$$ where $dE/dt$ is given in terms of the loss function $L$ as $$\frac{dE}{dt} = -L(E)n_g v(E).
\label{dedt}$$ Here $n_g$ is the density of hydrogen atoms $n_g = n(H) + 2n(H_2)$. We are searching for a steady-state distribution of $n$, and set $\partial n/ \partial t = 0$. Noting that $dN = n_g ds$, and using Equation , Equation can be written as $$X(E) \frac{\partial F}{\partial E} + \frac{\partial ^2F}{\partial N^2}= 0,
\label{middleStep}$$ where $$F(E, N) = n(E, N)v(E)L(E),$$ and $X(E) =L(E)v(E)/\Delta(E)$, with the rescaled diffusion coefficient $\Delta$ given by $$\Delta(E) = n_g(N) D(E, N).$$ Note that we are able to write Equation in this form because we have assumed that both the turbulent velocity spectrum and the magnetic field strength are independent of $N$, and that the ionization fraction is constant, so $\rho_{\rm ion} \propto n_g$. Under these assumptions, $\Delta$ is a function just of $E$ (see discussion in Section \[rangeOfApp\]).
We then let $$T= -\int_0^E\frac{dE'}{X(E')},
\label{Meq}$$ where $-T$ corresponds to the amount of diffusion undergone by a particle of energy $E$ during the time it loses all of its energy. Then, Equation becomes a linear diffusion equation $$\frac{\partial F}{\partial T} = \frac{\partial ^2F}{\partial N^2},
\label{diffEq}$$ with $T(E)$ being the pseudo-time. This allows the solution in a general form of the problem where $F(T, N)$ at the boundary $N = 0$ is a given function of pseudo-time [@Landau59]: $$F(T, N) = \int_{-\infty}^T \frac{N\exp{(-\frac{N^2}{4(T-T_i)})}}{\sqrt{4\pi (T-T_i)^3}} F_i(T_i)dT_i,
\label{Fsol}$$ where $F_i(T_i)$ is determined by the spectrum $j_i$ of CRs on the outside of the cloud as a function of their initial energy $E_i$, with $T_i = T(E_i)$.
ionization rate in the envelopes of molecular clouds
====================================================
The primary CR ionization rate of H$_2$, $\zeta_{\rm H_2}$, can be calculated using the relation $$\zeta_{\rm H_2}(N) = \int_0^\infty j(E, N) \sigma_{\rm H_2}(E) dE,
\label{genericionization}$$ where $\sigma_{\rm H_2}$ is the ionization cross section for molecular hydrogen and $j(E, N) = n(E, N)v(E)$.
The loss function for protons is well approximated by a power law over the range of energies from $10^5$ to $10^9$ eV (relevant for ionization in molecular clouds) as: $$L(E) = L_0\left(\frac{E}{E_0}\right)^{-d},
\label{Leq}$$ with $L_0 = 1.27\times 10^{-15}$ eV cm$^2$, $E_0 = 1$ MeV, and $d = 0.82$ [@Padovani18]. The use of approximation facilitates the analysis below by simplifying the calculations significantly. In @Padovani18, they assumed all the hydrogen to be in molecular form and used a column density which was the number of [*particles*]{} per unit area. In this paper, since we are dealing with lower column densities where not all the hydrogen need be molecular, we define the column density $N$ as the number of [*hydrogen atoms*]{} per unit area. This means that at a given mass surface density, our column density is higher than that in @Padovani18 by a factor of 1.67, which we have taken into account in the value of $L_0$ in Equation . Using Equations through , we obtain $$\Delta(E) = \Delta_0 \left(\frac{E}{E_0}\right)^{1-\lambda},
\label{simpleD}$$ where the value of $\Delta_0$ is discussed in Section \[rangeOfApp\]. Using Equations and , we can write $T$ as $$T = -\frac{1}{4}N_{0d}^2\left(\frac{E}{E_0}\right)^\alpha
\label{analT}$$ where $$N_{0d} = \sqrt{\frac{4\Delta_0E_0}{\alpha v_0L_0}}$$ is the characteristic column density necessary to attenuate a particle with energy $E_0$ for diffusive transport, $\alpha = 3/2 + d - \lambda$, and $v_0 = \sqrt{2E_0/m}$. For protons in the range from $10^5$ eV to $5 \cdot 10^8$ eV, the ratio between the loss function and the H$_2$ ionization cross section is nearly constant [@Padovani18]. Using the expression for the ionization cross section given in @Rudd85, and the loss function in Equation , we determine this ratio to be approximately $\epsilon = 37$ eV. Note that this corresponds to an energy lost per H$_2$ ionization event of approximately 62 eV, which is reduced by the ratio of the hydrogen number density to the particle number density. Thus, we can write $\sigma_{\rm H_2} = L/\epsilon$, and Equation becomes $$\zeta_{\rm H_2}(N) = \int_0^\infty \frac{F(E, N)}{\epsilon} dE.
\label{zetaR}$$ If we assume the following initial CR spectrum, on the outside of the cloud: $$j_i(E) = j_0\left(\frac{E}{E_0}\right)^{-a},
\label{initJ}$$ then, using Equation , one can express Equation in terms of $E$ as $$j(E, N) = j_i(E) \int_0^1 \operatorname{erfc}{\left(\frac{N/N_{0d}}{\sqrt{\left(E/E_0\right)^\alpha (x^{-\frac{\alpha}{a+d}}-1)}}\right)} dx.
\label{Jeq}$$ For the local spectrum , we can, after some manipulation, write the ionization rate as $$\zeta_{\rm H_2}(N) = \frac{j_0L_0E_0 I_d}{\sqrt{\pi} \epsilon } \left(\frac{N}{N_{0d}}\right)^{-\gamma_d},
\label{analDiffusiveZeta}$$ where $$I_d = \int_0^\infty x^{\frac{\gamma_d-1}{2}} dx \int_0^1 e^{-\frac{x}{1-y^\alpha}} \left(1-y^\alpha \right)^{-3/2} dy,
\label{Ieq}$$ and $$\gamma_d = \frac{4(a+d-1)}{3+2d-2\lambda}.$$ For $\lambda = 1/3$ or $1/4$, and $a$ in the range \[0.5, 2\], the formula $I_d = (1.73-\lambda/3)/(a+d-1)$ is accurate to within 8%; $I_d$ is convergent as long as $\gamma_d > 0$.
ionization Rate for Free-Streaming CRs
--------------------------------------
In the free-streaming approximation, we replace the diffusive flux $-D dn/ds$ in Equation with the free-streaming flux $\mu j$, where $\mu$ is the cosine of the pitch angle. In this case, we can directly relate the initial energy $E_i$ to $E$, via $N$ and $\mu$, using the loss function: $$N = \mu \int_E^{E_i} \frac{dE}{L(E)}.
\label{genericN}$$ Then, in the continuously slowing-down approximation [@Padovani09], we find that $$F(E, N, \mu) = F_i(E_i, \mu),
\label{CDSAj}$$ where $F_i$ is determined by the initial spectrum $j_i$. Note that at the low densities relevant for our problem the magnetic field strength can be assumed to be constant [@Crutcher12]. In our analytic model, using Equation , and we can write $$E_i = \left(E^{1+d} + \frac{N}{\mu N_{0f}}E_0^{1+d}\right)^{1/(1+d)},
\label{Ei}$$ where $$N_{0f} = \frac{E_0}{(1+d)L_0}$$ is the characteristic column density necessary to attenuate a particle of energy $E_0$ for free-streaming transport. Then the ionization rate is $$\zeta_{\rm H_2}(N) = \int_0^1 d\mu \int_0^\infty \frac{F(E, N, \mu)}{\epsilon} dE.
\label{genericStreamingZeta}$$ Assuming an initial spectrum $j_i(E)$ given by Equation , the local spectrum is $$j(E, N, \mu) = j_i(E) \left[1 + \frac{N}{\mu N_{0f}} \left(\frac{E_0}{E}\right)^{1+d}\right]^{-\frac{a+d}{1+d}}.
\label{CDSAj}$$ Using Equations and we can then write $$\zeta_{\rm H_2}(N) = \frac{(1+d)}{(a+2d)} \frac{j_0L_0E_0 I_f}{\epsilon} \left(\frac{N}{N_{0f}}\right)^{-\gamma_f},
\label{analStreamingZeta}$$ where $$I_f = \int_0^\infty \left(x^{1+d} + 1\right)^{-\frac{a+d}{1+d}}dx$$ and $$\gamma_f = \frac{a+d-1}{1+d} \equiv \frac{3 + 2d - 2\lambda}{4(1+d)} \gamma_d.$$ In the range from $a = [0.5, 2]$, the approximation $I_f = 1.07/(a+d-1) + 0.42$ is accurate to within 2%.
Diffusion Constant and Range of Applicability {#rangeOfApp}
---------------------------------------------
The diffusion approximation is only appropriate at column densities such that the particle has lost the memory of the pitch angle with which it entered the cloud. Diffusion is approximately equivalent to a random walk at velocity $v$ with step length (in column density) $\delta N(E) = 3\Delta(E)/v$. Thus, for a particle with energy $E$, the transition from free-streaming to diffusive propagation should occur roughly at the column $N$ such that $N \sim \delta N(E)$. Using Equation , and keeping in mind that the particles responsible for the bulk of the ionization are sub-relativistic, we find $$\delta N(E) = \frac{3\Delta_0}{v_0} \left(\frac{E}{E_0}\right)^{1/2 - \lambda}.
\label{stepLength}$$ The particles dominating the ionization at column $N$ are those whose stopping range $N_{\rm st}(E)$ is comparable to $N$. The stopping range is calculated using Equation as $$N_{\rm st}(E) = N_{0f} \left(\frac{E}{E_0}\right)^{1+d}.
\label{range}$$ Solving Equation for $E$, and plugging the result into Equation , we find that the condition $N \gtrsim \delta N$ is appropriate for column densities $$N \gtrsim \frac{3\Delta_0}{v_0} \left(\frac{3\Delta_0}{v_0N_{0f}}\right)^\frac{1 - 2\lambda}{1 + 2d + 2\lambda}.
\label{appropriateColumn}$$ We estimate the diffusion constant $\Delta_0$, entering Equation , based on an assumed slope of the turbulent power spectrum. We normalise the power spectrum based on observations of turbulent velocities at large scales. To estimate $v_*$ and $k_*$ in Equation , we assume a turbulent velocity of 1 km s$^{-1}$ at a scale of 1 parsec. This requires a major extrapolation, and it is possible that the turbulence is damped by ion neutral friction at intermediate scales (see @Soler13 for a thorough discussion of which MHD modes propagate at the scales of ion-neutral decoupling). Despite these uncertainties, we point to @Armstrong95 as evidence that a Kolmogorov power spectrum over a very wide range of $k$ is possible in the ISM. We further assume that the ionization is dominated by singly-ionised carbon, with an abundance relative to hydrogen of $1.5 \times 10^{-4}$ [@Gerin15]. We take $B = 3\, \mu G$, consistent with the results from @Crutcher12, and set $\mu_* = 2/3$.
Assuming a Kolmogorov turbulent spectrum between 1 parsec and the range of interest ($\lambda = 1/3$), Equations through give $\Delta_0 = 4.2 \times 10^{28}$ cm$^{-1}$s$^{-1}$; for a Kraichnan spectrum ($\lambda = 1/4$), we find $\Delta_0 = 2.6 \times 10^{27}$ cm$^{-1}$ s$^{-1}$. Plugging these values of $\Delta_0$ into Equation , we find that the diffusion approximation is appropriate for $N \gtrsim 8 \times 10^{19}$ cm$^{-2}$ for the Kolmogorov spectrum, and $\gtrsim 3 \times 10^{18}$ cm$^{-2}$ for the Kraichnan spectrum. We note that there is some observational evidence [@Heyer15] for a steeper turbulent spectrum of $\lambda = 0.5$ at spatial scales much larger than those resonant with sub-relativistic CRs. However at these scales, the turbulence is supersonic, so the spectrum is not governed by the same physics.
At a certain higher column density, the ion density is expected to drop dramatically when there are no longer sufficient UV photons to keep carbon ionised. This depends on the strength of the UV field near the cloud, as well as on the assumed properties of the medium [@Hollenbach99]. Based on the work of @Keto08, we assume the transition to take place at $N_{\rm tran} \approx 2 \times 10^{21}$ cm$^{-2}$, though we note that @Neufeld17 find a very sharp drop in C$^{+}$ abundance near a column density of $6 \times 10^{20}$ cm$^{-2}$. For column densities greater than $N_{\rm tran}$, the ion density is expected to drop by a factor of $\sim$ 100, depending on $\zeta_{\rm H_2}$ [@Neufeld17], leading to the proportional increase in $\Delta_0$. Then it appears unlikely that the turbulence would be strong enough to greatly influence the CR propagation. In environments with higher CR fluxes or more incident UV radiation than assumed by @Neufeld17, this boundary may be moved to higher column density. Specifically, @Neufeld17 find that if $\zeta_{\rm H_2}/n_H > 1.2 \times 10^{-17} {\rm cm}^{3} {\rm s}^{-1}$, then the ionization fraction remains greater than $10^{-4}$ to a column density of $\sim 10^{22}$ cm$^{-2}$. Such conditions may be found near the Galactic center [@Petit16].
results for a model interstellar spectrum
=========================================
@Padovani18 propose an interstellar CR spectrum of the following model form $$j(E) = C \frac{E^\delta}{(E + E_t)^\beta}\, {\rm eV}^{-1} {\rm cm}^{-2} \rm{s}^{-1}.
\label{genericSpectrum}$$ The high-energy slope of this function, $\delta - \beta$, is well determined [e.g., @Aguilar14; @Aguilar15], while the low-end slope $\delta$ is uncertain. @Ivlev15 argue that the spectrum determined by Voyager [@Cummings16], represents a lower bound on the interstellar proton spectrum, and they estimate an upper bound based on observed ionization rates in nearby clouds, in which $C = 3.0 \times 10^{16}$, $E_t = 650$ MeV, $\delta = -0.8$, and $\beta = 1.9$.
Ionization in molecular cloud envelopes {#molCloudIon}
---------------------------------------
We use the spectrum described by Equation , truncated at 3 GeV, to calculate $\zeta_{\rm H_2}(N)$ for three different propagation models described below. The results are plotted in Figure \[zetaFig\]. In all cases, when calculating the ionization rate, we integrated Equation or as appropriate from 10 KeV to 1 GeV. We vary $\delta$ as labelled in the panels, using $C = 3.0 \times 10^{16}$, $E_t = 650$ MeV, and $\delta - \beta = -2.7$.
The blue curve assumes pure free-streaming propagation, in which $\zeta_{\rm H_2}(N)$ was determined using Equation .
The purple curve represents the hybrid model, which assumes that CRs propagate diffusively until the column depth $N_{\rm tran} = 2 \times 10^{21}$ cm$^{-2}$ (such that carbon is no longer ionised) and then stream freely. The left-hand part of the curve is described by Equation where $F(E, N)$ is given by Equation , with $\Delta_0$ evaluated in Section \[rangeOfApp\] for Kolmogorov turbulence. The hybrid model results in a region of nearly flat $\zeta_{\rm H_2}(N)$ at $N \gtrsim N_{\rm tran}$, where the spectrum is dominated by particles with $N_{\rm st} \gg N_{\rm tran}$. Therefore, further attenuation has little effect until the column penetrated in the free-streaming region is comparable to the actual column passed through by the particles as they propagated diffusively.
The red curve assumes pure diffusive propagation for the entire column, ignoring the expected sharp decrease in $\rho_{\rm ion}$ that occurs around $N_{\rm tran}$. This represents a lower bound on $\zeta_{\rm H_2}$, but such a curve is probably unrealistic unless there is some process (anomalously high UV field, or anomalously high $\zeta_{\rm H_2}$) that keeps a higher ionization fraction deeper within the cloud.
Finally, the dashed red and dashed blue lines are the corresponding analytic approximations (given by Equations and respectively), assuming a spectrum given by Equation , with $a = -\delta$ and $j_0 = C E_0^\delta/E_t^\beta$. This spectrum coincides with that in Equation at lower energies.
The data points and error bars in Figure \[zetaFig\] are taken from Figure 6 of @Neufeld17[^1], assuming one magnitude of visual extinction to be equivalent to a column density of $1.9 \times 10^{21}$ cm$^{-3}$. The H$_2$ column density for the black points was measured directly, whereas for the green points it was inferred from meausrements of CH or the reddening. The spectrum in the top panel, corresponding to $\delta = -0.8$ in Equation , was constructed by @Padovani18 so that the free-streaming model passes through the points. For the other models, this spectrum yields curves which are too low. In the middle panel we plot the resulting ionization rate if $\delta$ is changed to 1.0. The low-energy slope of the resulting spectrum corresponds to the spectrum of particles produced in strong shocks [@Drury83]. Finally, in the bottom panel, we consider a steeper low-energy slope of $\delta = 1.2$. In this case, the diffusive model provides the best fit. We also point out that the slope for $\zeta_{\rm H_2}(N)$ obtained from Neufeld ($1.05 \pm 0.36$) is fit better by the diffusive model with $\delta = 1.2$, (which has a slope of 1.1 at $N = 10^{21}$ cm$^{-2}$), compared with the free-streaming model with $\delta = 0.8$, (which has a slope of $-0.4$ at at $N = 10^{21}$ cm$^{-2}$). This argument would seem to favor the diffusive model. However, as is clear from the magnitude of the error bars, the slope suggested by @Neufeld17 is rather uncertain.
![Primary ionization rate of H$_2$, $\zeta_{\rm H_2}$, as a function of column density of hydrogen atoms $N$ for our three different propagation models, as indicated in the legend. Different panels correspond to different values of $\delta$ in the assumed initial spectrum \[see Equation \]. The points and error bars are estimated from Figure 6 of @Neufeld17. Black points are those for which the the H$_2$ column density has been measured directly [@Neufeld17]. The solid lines represent the results for CR spectrum , whereas the dashed lines are for the power-law spectrum in Equation . Details of the different propagation models are discussed in Section \[molCloudIon\]. Note that the column density displayed here is a factor 1.67 larger than that in @Padovani18[]{data-label="zetaFig"}](zetaFig.pdf){width="1.03\columnwidth"}
Voyager Spectrum {#voyager}
----------------
![The red and blue curves show the best-fit attenuated spectra from Equations and respectively. The black points with error bars represent the Voyager data [@Cummings16]. []{data-label="voyagerFig"}](specPlot.pdf){width="1.03\columnwidth"}
As mentioned in the previous section, one source of low-energy CRs are shocks in the ISM which are expected to produce a power-law spectrum of accelerated particles. In particular, in the non-relativistic regime, a strong shock will produce a spectrum of particles with $j(E) \propto E^{-1}$ [@Drury83]. On the other hand, there is evidence [@Alves18], that the local bubble is surrounded by a thin shell of dense material, with the magnetic field nearly in the plane of the shell. In this picture, CRs penetrating into the local bubble must pass through a significant column density in the shell.
Let us assume the source produces a spectrum of particles outside the shell given by Equation with $a = 1.0$, and $j_0$ a free parameter. Given a propagation model, then we can find the column density and value of $j_0$ which best fit the Voyager data. Figure \[voyagerFig\] shows the best-fit spectra where the column density and the strength of the source spectra are free parameters. The points are the data from the Voyager probe [@Cummings16]. The blue curve shows the best fit curve assuming free-streaming propagation \[Equation , integrated over $\mu$\], and the green curve shows the best fit assuming diffusive propagation \[Equation \].
One can see that the best-fit spectra have very similar shapes, although the diffusive propagation model fits the data marginally better. There is, however, an important difference: in the free-streaming model, the best fit is obtained with a column density of $1.4 \times 10^{23}$ cm$^{-2}$, or $4.7 \times 10^4$ pc/cm$^3$. Unless the shell is very dense ($>100$ cm$^{-3}$), or the magnetic field extremely close to parallel to the shell (i.e. field lines wrap around the shell multiple times before entering the bubble), it seems difficult to understand from where such a large column could arise. In the diffusive model, on the contrary, the results depend both on $N$ and $\Delta_0$. Assuming $\Delta_0$ to be the same as that used in Section \[molCloudIon\], then we find a best fit value of $N$ of $5 \times 10^{21}$ cm$^{-2} = 1.8 \times 10^{3}$ pc/cm$^3$. This shows that using this model, the required attenuation can occur with a reasonable physical column density. That said, it is clear that our best fits do deviate significantly from the data, so these simplified models must of course not be the whole story.
Conclusions and Outlook
=======================
We proposed a model for the change in the low-energy CR spectrum (and corresponding ionization rate) as CRs propagate diffusively through a medium (where a certain degree of pre-existing turbulence is present) losing their energy to ionization. This predicts a substantially steeper slope of the ionization rate $\zeta_{\rm H_2}$ as a function of column density compared with the free-streaming model. Under conditions appropriate for local molecular clouds, this mechanism would likely only operate up to column densities of $\sim 10^{21}$ cm$^{-2}$. However, we showed that the assumption of diffusive propagation makes a significant difference to the behavior of $\zeta_{\rm H_2}$, and there are reasonable sets of physical parameters under which it could operate. We have provided analytic solutions for $\zeta_{\rm H_2}(N)$, Equations and , that can be applied to a variety of environments.
We also considered the question of how the spectrum of CRs seen by Voyager can be produced. We note that, to produce such a spectrum from a power-law source spectrum (predicted from the theory of diffusive shock acceleration), would require a large attenuating column of $\sim 10^{23}$ cm$^{-2}$. It is difficult to understand from where this column could arise. However, if one uses a diffusive propagation model, a marginally better fit to the Voyager spectrum can be obtained while keeping the required column density well under $10^{22}$ cm$^{-2}$.
The principal aim of the present paper is to highlight the stark differences in the behavior of $\zeta_{\rm H_2}(N)$ depending on the mode of CR transport. More detailed observations and analyses must be performed to distinguish between the two modes. In particular, it would be desirable to perform a dedicated analysis of $\zeta_{\rm H_2}(N)$ measured in molecular clouds, to determine the most probable slope more reliably. Furthermore, the analysis of @Neufeld17 should be done that $\zeta_{\rm H_2}$ varies within the cloud, rather than assuming a constant value within each cloud. Also, it would be good to have a more detailed model of CR transport in the shell surrounding the local bubble, based on the current model of the $B$ field, and taking into account transverse diffusion.
We would like to thank Marco Padovani and Daniele Galli for useful discussions and suggestions.
[^1]: @Neufeld17 plot $\zeta_p$, the primary ionization rate per hydrogen, which they assume to be 1/2.3 times the total ionization rate $\zeta_t$ (including secondary ionizations) per H$_2$. Taking a ratio $\zeta_t/\zeta_{\rm H_2} = 1.7$ [@Glassgold12], we find that we must shift the points from @Neufeld17 upwards by a factor of 1.4.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We extend the results of Denisov–Rakhmanov, Szegő–Shohat–Nevai, and Killip–Simon from asymptotically constant orthogonal polynomials on the real line (OPRL) and unit circle (OPUC) to asymptotically periodic OPRL and OPUC. The key tool is a characterization of the isospectral torus that is well adapted to the study of perturbations.'
author:
- 'David Damanik$^{1}$, Rowan Killip$^{2}$, and Barry Simon$^3$'
date: 'December 5, 2008'
title: Perturbations of Orthogonal Polynomials With Periodic Recursion Coefficients
---
[^1]
[^2]
[^3]
Introduction
============
[s1]{} This is a paper about the spectral theory of orthogonal polynomials on the real line (OPRL) and orthogonal polynomials on the unit circle (OPUC), that is, the connection of the underlying (spectral) measure and the recursion coefficients.
Specifically, given a probability measure, $d{\eta}$, on ${{\mathbb{R}}}$ with bounded but infinite support, the orthonormal polynomials, $p_n(x)$, obey a recursion relation $${\label}{1.1}
xp_n(x) =a_{n+1} p_{n+1}(x) + b_{n+1} p_n(x) + a_n p_{n-1}(x)$$ where the Jacobi parameters $\{a_n,b_n\}_{n=1}^\infty$ obey $b_j\in{{\mathbb{R}}}$, $a_j\in (0,\infty)$. As is well known (see, e.g., [@OPUC1 Section 1.3]), sets up a one-one correspondence between uniformly bounded $\{a_n,b_n\}_{n=1}^\infty$ and such measures, $d{\eta}$ (this is sometimes called Favard’s theorem).
Similarly, probability measures, $d\mu$, on $\partial{{\mathbb{D}}}$ which are nontrivial (i.e., their support is not a finite set of points) are in one-one correspondence with sequences $\{\alpha_n\}_{n=0}^\infty$ of Verblunsky coefficients in ${{\mathbb{D}}}\equiv \{z : {\lvertz\rvert}<1\}$ via the recursion relation of the orthonormal polynomials $\varphi_n(z)$, namely, $${\label}{1.2}
z\varphi_n(z) = \rho_n \varphi_{n+1}(z) + \bar\alpha_n \varphi_n^*(z)$$ where $${\label}{1.3}
\varphi_n^*(z) = z^n \, {\overline}{\varphi_n (1/\bar z)} \qquad
\rho_n = (1-{\lvert\alpha_n\rvert}^2)^{1/2}$$
Underlying the association of measures and recursion coefficients are matrix representations. For OPRL, we take the matrix for multiplication by $x$ in the $\{p_n\}_{n=0}^\infty$ basis of $L^2 ({{\mathbb{R}}},d{\eta})$, which is the tridiagonal Jacobi matrix $${\label}{1.3a}
J = \begin{pmatrix}
b_1 & a_1 & 0 & \vphantom{\ddots} \\
a_1 & b_2 & a_2 & \ddots \\
0 & a_2 & \ddots & \ddots \\
{} & \ddots & \ddots & \ddots
\end{pmatrix}$$ For OPUC, one takes the matrix, ${{\mathcal{C}}}$, for multiplication by $z$ in the basis obtained by orthonormalizing $\{1,z,z^{-1},z^2, z^{-2}, \dots\}$ in $L^2 (\partial{{\mathbb{D}}}, d\mu)$. This CMV matrix (see [@OPUC1 Section 4.2]) has the form $$\begin{aligned}
{{\mathcal{C}}}&={{\mathcal L}}{{\mathcal M}}{\label}{1.3b} \\
{{\mathcal L}}&= \Theta(\alpha_0) \oplus\Theta(\alpha_2)\oplus\cdots {\label}{1.3c} \\
{{\mathcal M}}&= {{\boldsymbol{1}}}_{1\times 1} \oplus\Theta(\alpha_1)\oplus\Theta(\alpha_3)\oplus \cdots {\label}{1.3d} \\
\Theta(\alpha) &= \begin{pmatrix}
\bar\alpha & \rho \\
\rho & -\alpha \end{pmatrix} {\label}{1.3e}\end{aligned}$$ where $\rho\equiv(1-|\alpha|^2)^{1/2}$. Note that ${{\mathcal{C}}}$ is unitary, while $J$ is self-adjoint.
As a model for what we wish to prove, let us briefly survey some of the main results relating to (slowly decaying) perturbations of the free case, that is, $a_n\equiv 1$, $b_n\equiv 0$ for OPRL and $\alpha_n\equiv 0$ for OPUC.
\(1) [*Weyl’s Theorem*]{} [@Weyl; @Blu; @AK; @GBk1; @GSprep]. If $a_n\to 1$, $b_n\to 0$, then $\sigma_{\text{\rm{ess}}}(d{\eta})=[-2,2]$ and if $\alpha_n\to 0$, then $\sigma_{\text{\rm{ess}}}(d\mu) = \partial{{\mathbb{D}}}$. Here $\sigma_{\text{\rm{ess}}}(d{\eta})$ is the (topological) support of the measure, $d{\eta}$, with all isolated points removed.
\(2) [*Denisov–Rakhmanov Theorem*]{} [@Rakh77; @Rakh83; @MNT85a; @DenPAMS; @NTppt]. If $\sigma_{\text{\rm{ess}}}(d{\eta})=\Sigma_{\text{\rm{ac}}}(d{\eta}) =[-2,2]$, then $a_n\to 1$ and $b_n\to 0$. If $\sigma_{\text{\rm{ess}}}(d\mu)= \Sigma_{\text{\rm{ac}}}(d\mu) =\partial{{\mathbb{D}}}$, then $\alpha_n\to 0$. Here $\Sigma_{\text{\rm{ac}}}(d{\eta})$ is defined as follows: let $d{\eta}=d{\eta}_{\text{\rm{ac}}}+ d{\eta}_{\text{\rm{s}}}$ with $d{\eta}_{\text{\rm{s}}}$ singular and $d{\eta}_{\text{\rm{ac}}}=f(x)dx$, then $\Sigma_{\text{\rm{ac}}}(d{\eta})=\{x : f(x)\neq 0\}$ as a measure class, that is, modulo sets of Lebesgue measure zero.
\(3) [*Szegő’s Theorem*]{} [@Sz15; @Sz20; @Sh; @Nev79; @KS]. In the OPUC case, define $${\label}{1.4}
Z(d\mu) \equiv -\int \log \biggl( {\frac}{d\mu_{\text{\rm{ac}}}}{d\theta}\biggr)\,
{\frac}{d\theta}{2\pi}$$ Then $Z(d\mu)<\infty$ if and only if $${\label}{1.5}
\sum_{j=0}^\infty \, {\lvert\alpha_j(d\mu)\rvert}^2 <\infty$$ In the OPRL case, define $${\label}{1.6}
Z(d{\eta}) \equiv -\int_{-2}^2 \log \biggl( 2\pi(4-E^2)^{1/2}\,
{\frac}{d{\eta}_{\text{\rm{ac}}}}{dE}\biggr) \, {\frac}{dE}{2\pi (4-E^2)^{1/2}}$$ Then, if we assume ${\text{\rm{supp}}}(d{\eta})\subset [-2,2]$, we have $${\label}{1.7}
Z(d\rho) <\infty \iff \limsup \sum_{j=1}^N \log (a_j) >-\infty$$ and if that holds, then $${\label}{1.8}
\sum_{n=1}^\infty (a_n-1)^2 + b_n^2 <\infty$$ and $${\label}{1.9}
\sum (a_n-1) \quad\text{and}\quad \sum b_n$$ are conditionally convergent to finite numbers.
\(4) [*Killip–Simon Theorem*]{} [@KS]. For OPRL, define $${\label}{1.10}
Q(d{\eta}) =-\int_{-2}^2 \log\biggl( \pi (4-E^2)^{-1/2}\,
{\frac}{d{\eta}_{\text{\rm{ac}}}}{dE}\biggr) \, {\frac}{(4-E^2)^{1/2}\, dE}{\pi}$$ and let $\{E_j\}$ be the point masses of $d{\eta}$ (eigenvalues of $J$) outside $[-2,2]$. Then holds if and only if $\sigma_{\text{\rm{ess}}}(d{\eta}) =[-2,2]$, $Q(d{\eta})<\infty$, and $\sum_j ({\lvertE_j\rvert}-2)^{3/2} <\infty$.
\(5) [*Nevai’s Conjecture*]{} [@Nev92; @KS]. For OPRL, if $\sum_{n=1}^\infty
{\lverta_n-1\rvert} +{\lvertb_n\rvert}<\infty$, then $Z(d\rho)<\infty$ ($Z$ given by ).
The five results listed above capture different aspects of the philosophy that the measure is close to the free case if and only if the coefficients are asymptotic to the free ones. In this paper, we study extensions of all these results to perturbations of a periodic sequence of Jacobi or Verblunsky coefficients, that is, $${\label}{1.11}
a_{n+p}^{(0)} =a_n^{(0)} \qquad b_{n+p}^{(0)} = b_n^{(0)} \qquad n\geq 1$$ or $${\label}{1.12}
\alpha_{n+p}^{(0)} =\alpha_n^{(0)} \qquad n\geq 0$$ and some fixed $p\geq 1$. Note that $p=1$ is the perturbation of the free case considered above. For simplicity in the OPUC case, we will normally suppose $p$ is even—indeed, the shape of a CMV matrix repeats itself only after shifting by two rows/columns. As explained in Section \[s13\], the situation when $p$ is odd can be reduced to this using sieving. For OPRL, $p$ is arbitrary.
The philosophy described above becomes more subtle when we move to the periodic setting; rather than having a single ‘free operator’ we have a manifold of them (the isospectral torus). Nevertheless—and this is the main thrust of the paper—spectral measures that are close to those of the isospectral torus correspond to coefficients that approach the isospectral torus. One of the key obstructions here is that a sequence of coefficients may approach the isospectral torus without converging to any particular point therein.
In order to make these heuristics precise, we need to make a few definitions. To keep the presentation as coherent as possible, we will focus our attention on the OPRL/Jacobi case for the remainder of the introduction.
To any pair of $p$-periodic sequences, $\{a_n^{(0)},b_n^{(0)}\}_{n\in{{\mathbb{Z}}}}$, we can associate a two-sided Jacobi matrix $J_0$. Two such pairs of sequences are termed *isospectral* if the corresponding Jacobi matrices have the same spectrum. We write ${{\mathcal T}}_{J_0}$ for the set of $p$-periodic sequences that are isospectral to $J_0$. Topologically, this is a torus as explained in Subsection \[ssIT\] below.
Given two bounded sequences $\{a_n,b_n\}_{n=1}^\infty$ and $\{a'_n,b'_n\}_{n=1}^\infty$, we define $${\label}{1.43}
d_m ((a,b), (a', b')) = \sum_{k=0}^\infty e^{-k} [{\lverta_{m+k} -a'_{m+k}\rvert} + {\lvertb_{m+k} - b'_{m+k}\rvert}]$$ which is a metric for the product topology on ${\mathop{\mathchoice{\smash{\vcenter{\hbox{\LARGE$\times$}}}\vphantom{\prod}}{\smash{\vcenter{\hbox{\Large$\times$}}}\vphantom{\prod}}{\times}{\times}}\displaylimits}_m^\infty \bigl((0,R]\times [-R,R]\bigr)$. The OPUC analog is $${\label}{1.44}
d_m ((\alpha), (\alpha')) = \sum_{k=0}^\infty e^{-k} {\lvert\alpha_{m+k} -\alpha'_{m+k}\rvert}$$ a metric for ${\mathop{\mathchoice{\smash{\vcenter{\hbox{\LARGE$\times$}}}\vphantom{\prod}}{\smash{\vcenter{\hbox{\Large$\times$}}}\vphantom{\prod}}{\times}{\times}}\displaylimits}_m^\infty {{\mathbb{D}}}$. The distance from a point to a set is defined in the usual way: $${\label}{1.45}
d_m ((a,b),{{\mathcal T}}) = \inf \{d_m ((a,b), (a',b')) : (a',b')\in{{\mathcal T}}\}$$ and similarly in the OPUC case.
We begin with the periodic analog of Weyl’s Theorem.
\[T1.1\] Let $J_0$ be a two-sided periodic Jacobi matrix and $J$ a one-sided Jacobi matrix with Jacobi parameters $\{a_m, b_m\}_{m=1}^\infty$. If $$\label{1.46}
d_m ((a,b),{{\mathcal T}}_{J_0})\to 0$$ then[^4] $${\label}{1.47}
\sigma_{\text{\rm{ess}}}(J) = \sigma(J_0)$$
As indicated, this result first appeared in [@LS_jdam]. It is derived from a theorem that had earlier been proven with different methods by others [@GI1; @M1; @Rab5]. The inclusion $\sigma_{\text{\rm{ess}}}(J) \supset \sigma(J_0)$ follows easily using trial vectors; the reverse seems to be more sophisticated. In Section \[s7\], we prove this using the methods of this paper. The OPUC version appears here as Theorem \[T13.1\]; it was also proved in [@LS_jdam].
Note that does not imply that there is a sequence $\{(a'_n,b'_n)\}\in {{\mathcal T}}_{J_0}$ such that $$d_m ((a,b),(a',b'))\to 0$$ It is much weaker. Equality of essential spectra under this stronger hypothesis follows immediately from Weyl’s original theorem on compact perturbations.
Our first major new result is an analog of the Denisov–Rakhmanov Theorem.
[T1.2]{} Let $J_0$ be a two-sided periodic Jacobi matrix and $J$ a one-sided Jacobi matrix with Jacobi parameters $\{a_m,b_m\}_{m=1}^\infty$. If $\sigma_{\text{\rm{ess}}}(J) = \sigma(J_0)$ and $${\label}{1.48}
\Sigma_{\text{\rm{ac}}}(J) = \sigma (J_0)$$ then $d_m ((a,b),{{\mathcal T}}_{J_0})\to 0$.
Using Theorem \[T1.4\] below, we will show that the hypotheses of this theorem can hold while $(a,b)$ only approaches ${{\mathcal T}}_{J_0}$ without actually having a limit.
A two-sided $p$-periodic Jacobi matrix is said to have all gaps open if the spectrum has exactly $p$ connected components—the largest number possible. As explained in Section \[s1a\], this holds generically (indeed, on a dense open set).
Our next new result is
[T1.3]{} Let $J_0$ be a two-sided periodic Jacobi matrix with all gaps open and parameters $\{a_n^{(0)}, b_n^{(0)}\}_{n=-\infty}^\infty$. Also, let $J$ be a one-sided Jacobi matrix with parameters $\{a_n,
b_n\}_{n=1}^\infty$ and spectral measure $d{\eta}$. We assume that $\sigma_{\text{\rm{ess}}}(J) = \sigma(J_0)$ and $${\label}{1.49}
\sum_j {\text{\rm{dist}}}(E_j,\sigma_{\text{\rm{ess}}}(J))^{1/2} <\infty$$ where $\{E_j\}$ enumerates the eigenvalues of $J$ outside $\sigma(J_0)$. Then $${\label}{1.50}
-\int_{\sigma(J_0)} \log \biggl( {\frac}{d{\eta}_{\text{\rm{ac}}}}{dx}\biggr) {\text{\rm{dist}}}(x, {{\mathbb{R}}}\setminus
\sigma(J_0))^{-1/2}\, dx <\infty$$ implies $${\label}{1.58a}
\lim_{N\to\infty}\, \sum_{j=1}^{pN} \log \biggl( {\frac}{a_j}{a_j^{(0)}}\biggr)$$ exists and lies in $(-\infty,\infty)$. Conversely, holds so long as $${\label}{1.51}
\limsup_{N\to\infty} \, \sum_{j=1}^N \log \biggl( {\frac}{a_j}{a_j^{(0)}}\biggr) >-\infty$$ and in this case, the limit in exists and lies in $(-\infty,\infty)$.
Lastly, if or holds, then $${\label}{1.52}
\sum_{m=0}^\infty d_m ((a,b),{{\mathcal T}}_{J_0})^2 <\infty$$ and there exists $J_1\in{{\mathcal T}}_{J_0}$, so that $${\label}{1.60a}
d_m (J,J_1)\to 0$$
1\. Thus, when – hold, $J$ has a limit $J_1$ in ${{\mathcal T}}_{J_0}$. In the normal direction to ${{\mathcal T}}_{J_0}$, the convergence is $\ell^2$ (in the sense of ). But in the tangential direction, we only prove it has a limit. It would be interesting to know what can be said about how slowly can occur and to know if there are examples where – hold but $${\label}{1.60b}
\sum_{m=0}^\infty d_m (J,J_1)^2 =\infty$$
2\. Notice that will only fail if the partial sums converge to $-\infty$.
3\. The final statement that there exists $J_1\in{{\mathcal T}}_{J_0}$ with is not our result but one of Peherstorfer–Yuditskii [@PY]. With our methods, we can prove that the $a$’s and $b$’s approach a periodic limit only if we replace with the stronger assumption that the discrete spectrum is finite.
4\. By , all $a_j^{(0)}$ in and can be replaced by $\operatorname*{Cap}(\sigma(J_0))$, the logarithmic capacity of the spectrum of $J_0$.
Our third new result is
[T1.4]{} Let $J_0$ be a two-sided periodic Jacobi matrix with all gaps open and parameters $\{a_n^{(0)}, b_n^{(0)}\}_{n\in{{\mathbb{Z}}}}$. Let $J$ be a Jacobi matrix with parameters $\{a_n, b_n\}_{n=1}^\infty$ and spectral measure $d{\eta}$. Then $$\label{1.52b}
\sum_{m=0}^\infty d_m ((a,b),{{\mathcal T}}_{J_0})^2 <\infty$$ if and only if
$\sigma_{\text{\rm{ess}}}(J)=\sigma(J_0)$,
$\displaystyle \sum_j {\text{\rm{dist}}}(E_j,\sigma_{\text{\rm{ess}}}(J))^{3/2} <\infty$, and
$\displaystyle -\int_{\sigma(J_0)} \log \biggl( {\frac}{d{\eta}_{\text{\rm{ac}}}}{dx}\biggr) {\text{\rm{dist}}}(x, {{\mathbb{R}}}\setminus
\sigma(J_0))^{1/2}\, dx <\infty$.
Here $\{E_j\}$ enumerates the [[(]{}]{}discrete[[)]{}]{} spectrum of $J$ outside $\sigma(J_0)$.
1\. Since (i)–(iii) are equivalent to , one may easily construct examples where (i)–(iii) hold, but there is no $J_1$ with . This provides the examples promised in the remark after Theorem \[T1.2\]. It also shows a stark difference between – and (ii)–(iii). In terms of the spectral measure, this difference is reflected only in the behavior near the band edges.
2\. As we will see (Section \[s10\]), there are results even if all gaps are not open, but for Theorem \[T1.4\] they are not so easy to express directly in terms of the $a$’s and $b$’s.
3\. A special case of part of Theorem \[T1.4\] is known, namely, Killip [@Kil02] proved that $\Sigma_{\text{\rm{ac}}}(J) = \sigma (J_0)$ for $\{a_n, b_n\}_{n=0}^\infty$ obeying $\sum_{n=1}^\infty {\lverta_n - a_n^{(0)}\rvert}^2 + {\lvertb_n -b_n^{(0)}\rvert}^2 <\infty$ (which is a strictly stronger hypothesis than ).
[T1.5]{} Let $J_0$ be a two-sided periodic Jacobi matrix and $J$ a Jacobi matrix with Jacobi parameters $\{a_n, b_n\}_{n=1}^\infty$ and spectral measure $d{\eta}$. Suppose $${\label}{1.53}
\sum_{n=1}^\infty\, {\lverta_n-a_n^{(0)}\rvert} + {\lvertb_n - b_n^{(0)}\rvert} <\infty$$ Then holds.
1\. Condition can be replaced by $${\label}{1.55}
\sum_{n=1}^\infty d_n ((a,b), {{\mathcal T}}_{J_0}) <\infty$$ Indeed, if holds, then holds with $\{a_n^{(0)}, b_n^{(0)}\}$ replaced by some fixed sequence in ${{\mathcal T}}_{J_0}$.
2\. As we will show (see Proposition \[P2.5\]), the theorems above continue to hold if $d_m$ is replaced by $${\label}{1.54}
{\tilde }d_m ((a,b), (a',b'))=\sum_{k=0}^{p-1} ({\lverta_{m+k}-a'_{m+k}\rvert} +
{\lvertb_{m+k} - b'_{m+k}\rvert})$$ For OPUC, we need to sum $k$ from $0$ to $p$ in order to get an equivalence; see the discussion at the end of Section \[s3\].
In Section \[s13\], we prove an OPUC analog of each of these theorems. We need to replace the all-gaps-open hypothesis with a stronger one (that holds generically). The deficiency is not so much with our method, but rather that an independent question, which is known in the Jacobi case (independence of the Toda Hamiltonians), is currently unresolved in the CMV case. Our results confirm Conjectures 12.2.3 and 12.2.4 of [@OPUC2] as well as Conjectures 12.2.5 and 12.2.6 in the (generic) special case that all gaps are open.
For the case of OPUC with a single gap, the analog of Theorem \[T1.2\] is known and motivated Simon’s conjectures in [@OPUC2]. In that case, the isospectral tori are labelled by $a\in (0,1)$ and consist of $\{\alpha^{(\lambda)} :
\lambda\in\partial{{\mathbb{D}}}\}$ where $\alpha_n^{(\lambda)}=\lambda a$. Then $d_m (\alpha,
{{\mathcal T}}) \to 0$ is equivalent to ${\lvert\alpha_n\rvert}\to a$ and $\alpha_{n+1}/\alpha_n\to 1$. This is often called the López condition. Bello–López [@BHLL] proved the OPUC analog of Theorem \[T1.2\] for this case if $\sigma_{\text{\rm{ess}}}(J)=\sigma(J_0)$ is strengthened to $\sigma (J)=\sigma (J_0)$ (the analog of Rakhmanov’s result). The full analog for this special case appears in Simon [@OPUC2], Alfaro et al. [@ABMV], and Barrios et al. [@BCL].
Associated to each two-sided $p$-periodic Jacobi matrix, $J_0$, is a polynomial, $\Delta_{J_0}$, of degree $p$, known as the discriminant. This is a classical object described in detail in the next section. It is usually defined as the trace of the one-period transfer matrix. It is also the unique polynomial (with positive leading coefficient) such that $$\sigma(J_0) = \{ x : \Delta_{J_0}(x) \in [-2,2] \}$$ In particular, two sequences of coefficients are isospectral if and only if they give rise to the same discriminant.
The key to the proofs of our results is what we call the magic formula. Let $J$ be a [*two-sided*]{} Jacobi matrix, then $${\label}{1.56}
\Delta_{J_0} (J) = S^p + S^{-p}$$ if and only if $J\in{{\mathcal T}}_{J_0}$. Here $S$ is the right shift (cf. ). In particular, already implies that $J$ is periodic! In the OPUC case, $\Delta$ is a polynomial in $z$ and $z^{-1}$. It turns out that $\Delta({{\mathcal{C}}})$ is always self-adjoint; moreover, $${\label}{1.57}
\Delta_{{{\mathcal{C}}}_0}({{\mathcal{C}}}) =S^p + S^{-p}$$ if and only if ${{\mathcal{C}}}\in{{\mathcal T}}_{{{\mathcal{C}}}_0}$.
It has been previously noted that for periodic $J_0$, one has $$\Delta_{J_0} (J_0) = S^p + S^{-p}$$ That this holds for some polynomial in $J_0$ is in Naĭman [@Nai; @Nai2]. That the polynomial is the discriminant was found by Sebbar–Falliero [@SF]. After learning of our results, L. Golinskii has kindly pointed out to us that Naĭman [@Nai2] also has a theorem which implies any $J$ obeying is periodic, the core of proving the converse. We will discuss this further in Section \[s2\].
Nonetheless, the two facts that make magical to us—that it characterizes the isospectral torus and that it is ideal for the study of perturbations—seem to have escaped prior notice.
While $J$ may be tridiagonal and ${{\mathcal{C}}}$ five-diagonal, both $\Delta(J)$ and $\Delta({{\mathcal{C}}})$ are $2p+1$-diagonal, that is, vanishing except for the main diagonal and $p$ diagonals above and below. Thus, both will be tridiagonal if written as $p\times p$ blocks. The key to our proofs will be to extend results from the $a_n\equiv 1$, $b_n\equiv 0$ case to block tridiagonal matrices, and then use or to study perturbations of the periodic case.
The magic formula is very powerful and opens up many other avenues for study:
Szegő and Jost asymptotics for periodic perturbations and, in particular, the analogs of Damanik–Simon [@Jost1].
Periodic analogs of the results of Nevai–Totik [@NT89] and its various recent extensions [@Jost2; @Sim303; @Jost3].
Analogs of the Strong Szegő Theorem for the periodic case following Ryckman’s paper [@Ryck] for the Jacobi case.
We should point out a major limitation of our results. If $B$ is a disjoint finite union of closed intervals in ${{\mathbb{R}}}$ (or $\partial{{\mathbb{D}}}$), one can construct an isospectral torus of Jacobi (or CMV) matrices whose recursion coefficients are almost periodic. As discussed in Section \[s1a\], these are strictly periodic if and only if the harmonic measure of each interval is rational. There are obvious potential extensions of Theorems \[T1.1\]–\[T1.5\] to this setting, but except for Theorem \[T1.1\] (where the method of [@LS_jdam] applies) and Theorem \[T1.2\] (where Section \[new-s9\] has some extensions), we do not know how to prove them (or even if they are true). There is no analog of $\Delta$ in the almost periodic case, so our method does not work directly.
Here is the plan of this paper. Section \[s1a\] reviews the theory of the (unperturbed) periodic problem. In Section \[s2\], we prove the magic formula for OPRL, and in Section \[s3\], the magic formula for OPUC. While we will not discuss Schrödinger operators in detail here, we discuss the magic formula for such operators in Section \[s4\]. As we have mentioned, the magic formula brings block Jacobi matrices into play, so Section \[s5\] discusses matrix-valued OPRL and OPUC—mainly setting up notation. Section \[s6\] uses known results on Rakhmanov’s theorem for matrix-valued orthogonal polynomials to prove a Denisov-type extension which we use in Section \[s7\] to prove Theorem \[T1.2\]; the section also proves half of Theorem \[T1.1\]. Section \[new-s9\] provides two results that go beyond the periodic case to prove Denisov–Rakhmanov-type theorems for special almost periodic situations. Section \[s8\], following [@Sim288], proves the $P_2$ sum rule of Killip–Simon [@KS] and the $C_1$ sum rule for matrix-valued measures, and Section \[s9\] uses these results to prove Theorems \[T1.3\] and \[T1.4\]. Section \[s10\] explores what we can say if gaps are closed. Section \[s11\] proves analogs to the Lieb–Thirring bounds of Hundertmark–Simon [@HunS] as preparation for proving Theorem \[T1.5\] in Section \[s12\]. Finally, Section \[s13\] discusses the OPUC results.
Acknowledgements {#acknowledgements .unnumbered}
----------------
It is a pleasure to thank Leonid Golinskii, Irina Nenciu, Leonid Pastur, and Peter Yuditskii for useful discussions.
Note Added August, 2008 {#note-added-august-2008 .unnumbered}
-----------------------
During the refereeing of this paper, Remling (in [@R07]), motivated in part by this paper, found a positive resolution of the conjecture that, in the language of our Theorem \[Tn9.5\], every set in ${{\mathcal G}}$ is a Denisov-Rahkmanov set. His analysis depends on a very interesting theorem on right limits of Jacobi matrices with absolutely continuous spectrum – it provides a new approach to Denisov-Rahkmanov theorems.
Review of the Periodic Problem {#s1a}
==============================
In this section, we collect some of the major elements in the strictly periodic case. As this is textbook material, we forgo proofs and historical discussion. Full details can be found, for example, in [@CodLev; @East; @MagWin; @OPUC2; @SimonNew; @Teschl; @Toda] and the references therein.
To discuss the strictly periodic case, we need to extend our operators to be two-sided, that is, to act on $\ell^2({{\mathbb{Z}}})$. In the Jacobi/OPRL case, we simply continue the tridiagonal pattern with parameters $\{a_n,b_n\}_{n\in{{\mathbb{Z}}}}$. Two-sided (or extended) CMV matrices are formed as ${{\mathcal{C}}}={{\mathcal L}}{{\mathcal M}}$, where ${{\mathcal L}}$ and ${{\mathcal M}}$ are doubly infinite direct sums $$\begin{aligned}
{\label}{3.12new}
{{\mathcal L}}&= \cdots \oplus \Theta_{-2} (\alpha_{-2}) \oplus \Theta_0 (\alpha_0) \oplus \Theta_2 (\alpha_2) \oplus \cdots \\
{\label}{3.13new}
{{\mathcal M}}&= \hphantom{\oplus\Theta_0(}\cdots\oplus \Theta_{-1} (\alpha_{-1}) \oplus\Theta_1 (\alpha_1) \oplus\cdots\end{aligned}$$ that are misaligned by one row/column, just as in –.
We adopt the convention of indexing the elements of matrices so that $$\label{zeroentry}
J_{11} = b_1 \qquad {{\mathcal L}}_{00}=\bar\alpha_0 \qquad {{\mathcal M}}_{00}=-\alpha_{-1}$$ except ${{\mathcal M}}_{00} = 1$ in the one-sided case.
Transfer Matrices.
------------------
Let $J$ be a two-sided Jacobi matrix. A sequence $\{u_n\}$ obeys $(J-x)u\equiv 0$ if and only if $${\label}{1.27}
a_{n} u_{n+1} + (b_{n}-x) u_{n} + a_{n-1} u_{n-1} =0$$ or, what is equivalent, $${\label}{1.28}
\begin{pmatrix} u_{n+1} \\ a_{n} {u_n} \end{pmatrix} = \Lambda_n \begin{pmatrix} u_{n} \\ a_{n-1} u_{n-1} \end{pmatrix}$$ with $${\label}{1.14}
\Lambda_n(x) = \frac{1}{a_{n}} \begin{pmatrix} x-b_{n} & -1 \\ a_{n}^2 & 0 \end{pmatrix}$$ Note that the desire to have $\Lambda_n$ depend only on one pair $(a_{n},b_{n})$ and have determinant equal to one resulted in the factors $a_{n}$ and $a_{n-1}$ appearing in . (The same price is usually paid when writing Sturm–Liouville equations as first-order systems.) The choice is not the most common one (although it is used in Pastur–Figotin [@PF]), but we feel it is the ‘right’ one since, in particular, $\det (\Lambda_n(x))=1$.
In the OPUC case we define $${\label}{1.15}
M_n(z) = \rho_n^{-1}
\begin{pmatrix}
z & -\bar\alpha_n \\
-\alpha_n z & 1
\end{pmatrix}$$ which encodes the recurrence relation : $${\label}{1.28C}
\begin{pmatrix} \varphi_{n+1}(z) \\ \varphi^*_{n+1}(z) \end{pmatrix}
= M_n(z) \begin{pmatrix} \varphi_n(z) \\ \varphi^*_n(z) \end{pmatrix}$$ We will now explain the link to formal (i.e., not necessarily $\ell^2$) eigenvectors of a two-sided CMV matrix, ${{\mathcal{C}}}$. It is not as simple as .
Suppose $({{\mathcal{C}}}-z)u=0$ with $z\neq 0$ and let $v={{\mathcal Z}}^{-1}{{\mathcal M}}u$ where ${{\mathcal Z}}$ denotes the diagonal matrix with entries $${{\mathcal Z}}_{jj} = \begin{cases} z &: \text{ $j$ odd} \\ 1 & : \text{ $j$ even }\end{cases}$$ and ${{\mathcal M}}$ is as in . Then $$\label{CMVtransfer}
z \begin{pmatrix} u_{2n+2} \\ v_{2n+2} \end{pmatrix} = z
M_{2n+1}(z) \begin{pmatrix} v_{2n+1} \\ u_{2n+1} \end{pmatrix} =
M_{2n+1}(z) M_{2n}(z) \begin{pmatrix} u_{2n} \\ v_{2n}
\end{pmatrix}$$
The key observation used to verify is $$\label{ThetaA}
\begin{pmatrix} zy \\ y' \end{pmatrix} = \Theta(\alpha_n) \begin{pmatrix} x \\ x' \end{pmatrix}
\iff
\begin{pmatrix} x' \\ y' \end{pmatrix} = M_n(z) \begin{pmatrix} y \\ x \end{pmatrix}$$ This follows by simple algebraic manipulations: $$\begin{aligned}
&\!\!\!\begin{pmatrix} zy \\ y' \end{pmatrix} = \Theta(\alpha_n) \begin{pmatrix} x \\ x' \end{pmatrix} \\
\iff\ & \bar\alpha_n x + \rho_n x' = zy \quad\text{and}\quad \rho_n x - \alpha_n x' = y' \\
\iff\ & x' = \rho_n^{-1}(zy - \bar\alpha_n x) \quad\text{and}\quad y' = \rho_n x - \alpha_n x' \\
\iff\ & x' = \rho_n^{-1}(zy - \bar\alpha_n x) \quad\text{and}\quad y' = \rho_n^{-1} (-\alpha_n zy + x) \\
\iff\ &\!\!\!\begin{pmatrix} x' \\ y' \end{pmatrix} = M_{n}(z) \begin{pmatrix} y \\ x \end{pmatrix}\end{aligned}$$
With now in hand, we may argue as follows: $$\begin{aligned}
& ({{\mathcal{C}}}-z)u=0 \\
\iff\ & zu = {{\mathcal L}}{{\mathcal M}}u \\
\iff\ & v:={{\mathcal Z}}^{-1} {{\mathcal M}}u \quad\text{obeys}\quad z u = {{\mathcal L}}{{\mathcal Z}}v \\
\iff\ & {{\mathcal Z}}v = {{\mathcal M}}u \quad\text{and}\quad z u = {{\mathcal L}}{{\mathcal Z}}v \\
\iff\ &\!\!\! \begin{pmatrix} z v_{2n-1} \\ v_{2n} \end{pmatrix} = \Theta(\alpha_{2n-1})
\begin{pmatrix} u_{2n-1} \\ u_{2n} \end{pmatrix} \quad\text{and}\quad \begin{pmatrix} z u_{2n} \\ z u_{2n+1} \end{pmatrix}
= \Theta(\alpha_{2n}) \begin{pmatrix} v_{2n} \\ z v_{2n+1} \end{pmatrix} \\
\iff\ &\!\!\! \begin{pmatrix} u_{2n} \\ v_{2n} \end{pmatrix} = M_{2n-1}(z) \begin{pmatrix} v_{2n-1} \\ u_{2n-1} \end{pmatrix}
\quad\text{ and }\quad \begin{pmatrix} z v_{2n+1} \\ z u_{2n+1} \end{pmatrix} = M_{2n}(z) \begin{pmatrix} u_{2n} \\ v_{2n} \end{pmatrix} \\
\iff\ &\!\!\! \begin{pmatrix} u_{2n+2} \\ v_{2n+2} \end{pmatrix} = M_{2n+1}(z) \begin{pmatrix} v_{2n+1} \\ u_{2n+1} \end{pmatrix}
\quad\text{and}\quad z \begin{pmatrix} v_{2n+1} \\ u_{2n+1} \end{pmatrix} = M_{2n}(z) \begin{pmatrix} u_{2n} \\ v_{2n}
\end{pmatrix}\end{aligned}$$ which are the two parts of .
The Discriminant.
-----------------
As in Sturm–Liouville theory, the discriminant is defined as the trace of the one-period transfer matrix: $${\label}{1.21new}
\Delta(z) = {\text{\rm{Tr}}}( T(z) )$$ where $${\label}{1.21Anew}
T(z) = \begin{cases}
\ \Lambda_{p}(z) \cdots \Lambda_2(z) \Lambda_1(z) &\text{ (OPRL)} \\[0.25ex]
\ z^{-p/2} M_{p-1}(z) \cdots M_1(z) M_0(z) & \text{ (OPUC)}
\end{cases}$$ In the OPUC case, $p$ is even. Also, the factor $z^{-p/2}$ is there to cancel the extra factor of $z$ on the left-hand side of . From a strictly OPUC point of view, it is more natural to omit this factor (as in [@OPUC1; @OPUC2]); however, as the magic formula is an operator identity, we have elected to use the definition best adapted to this perspective. The only negative side effect of this choice is that our Lyapunov exponent (defined below) differs by $-\tfrac12\log|z|$ from that in [@OPUC1; @OPUC2].
For OPRL, the discriminant is a real polynomial of degree $p$ with leading behavior $$\begin{aligned}
{\label}{2.5xnew}
\Delta(x) &= (a_1 \dots a_p)^{-1} \biggl[\, \prod_{j=1}^p (x-b_j) + O(x^{p-2})\biggr] \\
{\label}{2.4new}
&= (a_1 \cdots a_p )^{-1} \biggl[ x^p - \biggl(\, \sum_{j=1}^p b_j\biggr) x^{p-1} + O(x^{p-2})\biggr]\end{aligned}$$
For OPUC, it is a Laurent polynomial of total degree $p$ with $${\label}{1.13}
\Delta(\bar z) = {\overline}{\Delta(1/z)}
$$ so $\Delta$ is real-valued on $\partial{{\mathbb{D}}}$. Moreover, $${\label}{1.13a}
\Delta(z) =(\rho_0 \rho_1 \cdots \rho_{p-1})^{-1} (z^{p/2} + \cdots + z^{-p/2})$$
The Lyapunov Exponent
---------------------
On an exponential scale, the behavior of formal eigenfunctions is determined by the Lyapunov exponent $$\begin{aligned}
{\label}{1.16}
\gamma(z) &= \lim_{n\to\infty}\, {\frac}{1}{np} \, \log \| T^n(z) \| \\
&= {\frac}{1}{p} \, \log \text{(spectral radius of $T(z)$)} \notag \\
&= {\frac}{1}{p}\, \log |\lambda_+(z)| {\label}{1.18x}\end{aligned}$$ where $\lambda_\pm$ are the eigenvalues of $T(z)$ with the convention ${\lvert\lambda_+\rvert} \geq {\lvert\lambda_-\rvert}$.
As $\det(T(z))\equiv 1$, these eigenvalues are the roots of $${\label}{1.22}
\lambda^2 -\Delta(z)\lambda + 1=0$$ which implies $${\label}{1.23}
\lambda_\pm(z) = {\frac}{\Delta(z)}{2} \pm {\frac}{\sqrt{\Delta^2(z) -4}}{2}$$ and so $${\label}{1.24}
\gamma(z) = \tfrac{1}{p} \log \left| \tfrac12 \Delta(z) + \tfrac12 \sqrt{\Delta(z)^2-4}\right|$$
Gaps and Bands.
---------------
Our recurrence relations admit bounded solutions for a given $z$ if and only if $\Delta(z)\in[-2,2]$. In the Jacobi/OPRL case, this is a collection of intervals in ${{\mathbb{R}}}$. For CMV/OPUC, it is a collection of arcs in $\partial {{\mathbb{D}}}$. In either case, one may partition this set into $p$ bands. These are the closures of the (disjoint) regions where $\Delta(z)\in (-2,2)$. These can only intersect at the ‘band edges’, $\Delta^{-1} (\{-2,2\})$.
The open gaps are the intervals/arcs that are complementary to the bands—excluding the two semi-infinite intervals in the OPRL case. When two bands touch, we refer to the common band edge as a closed gap.
$\Delta^2-4$ has simple zeros at the edges of the open gaps and double zeros at the closed gaps; indeed, this is a complete list of its zeros. It is possible to distinguish whether these zeros correspond to $\Delta(z)=\pm 2$ from the fact that there must be two zeros of $\Delta\pm2$ between consecutive zeros of $\Delta\mp2$ and the fact that $\Delta$ has positive leading coefficient.
Spectrum.
---------
In both cases, the spectrum of the two-sided operator (acting on $\ell^2({{\mathbb{Z}}})$) is the union of the bands: $\sigma = \Delta^{-1} ([-2,2])$. It is purely absolutely continuous and of multiplicity two.
The spectrum of a two-sided $p$-periodic operator uniquely determines its discriminant; see Lemma \[L2.4\]. One consequence of this was noted already in the introduction: isospectral tori are the classes of $p$-periodic recurrence coefficients that lead to the same discriminant.
In the case of a one-sided operator, the essential spectrum remains $\Delta^{-1} ([-2,2])$; it is absolutely continuous with multiplicity one. In addition, up to one eigenvalue may appear in each open gap.
Potential Theory.
-----------------
From the way they are defined, one can see that $\gamma(z)$ vanishes on the bands and is both positive and harmonic in the complement (in the OPUC case one must also exclude $z=0$). This leads to the solution of the Dirichlet problem for a charge at infinity, $$\label{1.24new}
g_{{{\mathbb{C}}}\setminus \sigma} (z;\infty) = \begin{cases}
\ {\frac}{1}{p} \log \left| {\frac}{\Delta}{2} + \sqrt{{\frac}{\Delta^2}{4}-1}\right| & \text{(OPRL)} \\[0.25ex]
\ {\frac}{1}{2} \log {\lvertz\rvert} + {\frac}{1}{p} \log \left| {\frac}{\Delta}{2} + \sqrt{{\frac}{\Delta^2}{4}-1}\right| & \text{(OPUC)}
\end{cases}$$ and so to the logarithmic capacity of the spectrum, $$\label{1.24newA}
\operatorname*{Cap}(\sigma) = \begin{cases}
\bigl(\prod_{j=1}^p a_j\bigr)^{1/p} & \text{(OPRL)} \\[1.25ex]
\bigl(\prod_{j=0}^{p-1} \rho_j\bigr)^{1/p} & \text{(OPUC)}
\end{cases}$$
Harmonic Measure.
-----------------
Taking normal derivatives in leads to a formula for harmonic measure on $\sigma$ (aka equilibrium measure for the logarithmic potential), $${\label}{1.25}
d\nu = \begin{cases}
{\frac}{2}{p} \, {\frac}{{\lvert\Delta'(x)\rvert}}{\sqrt{4-\Delta^2(x)}}\, {\frac}{dx}{2\pi} & \text{(OPRL)} \\[1.25ex]
{\frac}{2}{p} \, {\frac}{{\lvert\Delta'(e^{i\theta})\rvert}}{\sqrt{4-\Delta^2(e^{i\theta})}}\, {\frac}{d\theta}{2\pi} & \text{(OPUC)}
\end{cases}$$ where ${\text{\rm{supp}}}(d\nu)= \sigma=\{ z :{\lvert\Delta(z)\rvert}\leq 2\}$. ([*Note*]{}: [@OPUC2] has $1/p$ rather than $2/p$, but that is an error.)
Recognizing $${\label}{1.26}
{\frac}{\Delta'(x)}{\sqrt{4-\Delta^2 (x)}} = {\frac}{d}{dx}\, \arccos \biggl( {\frac}{\Delta(x)}{2}\biggr)$$ we see that the harmonic measure of each band is exactly $1/p$. In particular, the connected components of the union of the bands all have rational harmonic measure. This gives strong restrictions on sets that can be bands. In the OPRL case, rational harmonic measure of connected components is also sufficient for a set to be the spectrum of a periodic Jacobi matrix. In the OPUC case, there is an additional condition needed: after breaking the bands into arcs of harmonic measure $1/p$, the harmonic midpoints $\{\zeta_j\}_{j=1}^p$ of these intervals must obey $\prod_{j=1}^p \zeta_j =1$. Clearly, the condition on the harmonic midpoints can be achieved by simply rotating $\sigma$. Discarding this condition gives rise to Verblunsky coefficients that are $p$-automorphic, $\alpha_{n+p} = e^{i\phi} \alpha_n$, rather than $p$-periodic.
Thouless Formula {#ssTF}
----------------
Harmonic measure appears naturally in the theory in several other ways. It is the density of states measure: $$\label{DOS}
\lim_{N\to\infty} \frac{1}{2N} \sum_{n=-N}^N f(J)_{nn} = \int f(x) \,d\nu(x)$$ for every polynomial (or continuous function) $f$. The same formula holds with ${{\mathcal{C}}}$ replacing $J$. This connection, or more precisely the resulting expression for $\gamma(z)$ in terms of $\operatorname*{Cap}(\sigma)$ and the logarithmic potential of $d\nu$, is known as the Thouless formula.
Two further characterizations of $d\nu$ involve the orthogonal polynomials. $d\nu$ is the weak limit of ${\frac}{1}{N} \sum_{n=0}^{N-1} p_n^2 (x)\, d{\eta}(x)$ (resp. $ {\frac}{1}{N} \sum_{n=0}^{N-1}
{\lvert\varphi_n (e^{i\theta})\rvert}^2 d\mu(\theta)$). It is also the limiting density of zeros, that is, the weak limit of the probability measures, $d\nu_n$, which give weight $1/n$ to each of the zeros of $p_n$ (resp.$\varphi_n$). These two characterizations are closely linked to ; however, in the OPUC case one should keep in mind that for each $n$, the zeros of $\varphi_n$ lie strictly inside ${{\mathbb{D}}}$.
Floquet Theory.
---------------
Looking at the eigenvalues of $T$, one sees that when $\lambda_+\neq \lambda_-$, there is a basis of formal (i.e., non-$\ell^2$) eigenfunctions obeying $${\label}{1.29}
u_{m+kp} = \lambda_\pm^k u_m$$ If $\lambda_+ =\lambda_-$, which happens precisely at the band edges, then both are $\pm1$. If the edge abuts an open gap, there is only one eigenfunction obeying since $T$ has a Jordan block structure. At closed gaps, $T=\pm {{\boldsymbol{1}}}$ and so all solutions obey .
Solutions obeying $${\label}{1.30}
u_{m+kp} = e^{ik\theta} u_m$$ are called Floquet solutions and $e^{i\theta}$ is called the Floquet index; they have much the same role as plane waves in Fourier analysis. Since $\lambda_-=\lambda_+^{-1}$, if has a solution, then $e^{-i\theta}$ is also a Floquet index.
In the OPRL case, $${\label}{1.31}
\text{\eqref{1.30} holds} \iff \Delta(x) =2\cos\theta$$ Thus, by the discussion above, for each $\theta\in (0,\pi)$, or holds for exactly $p$ values of $x$: $x_1(\theta) <x_2(\theta) <\cdots < x_p(\theta)$. These $x_j(\theta)$ are known as the band functions.
The changes in the OPUC case are purely notational: $${\label}{1.34}
\text{\eqref{1.30} holds} \iff \Delta(z) = 2\cos\theta$$ For $\theta\in (0,\pi)$, this has $p$ solutions all of which lie in $\partial{{\mathbb{D}}}$.
Direct Integrals.
-----------------
Let $S$ denote the right shift, $${\label}{1.35}
(Su)_n = u_{n-1}$$ If the sequences of coefficients are $p$-periodic, then $J$ (or ${{\mathcal{C}}}$) commutes with $S^p$, which means that the two operators can be ‘simultaneously diagonalized’. We elaborate this point in the OPRL case; the OPUC is almost identical.
Let us write $${{\mathcal H}}_p := \int^\oplus \ell^2_\theta \, {\frac}{d\theta}{2\pi} = L^2\bigl([0,2\pi), \tfrac{d\theta}{2\pi}; {{\mathbb{C}}}^p\bigr)$$ where $\ell^2_\theta$ is the $p$-dimensional Hilbert space $$\ell^2_\theta = \{ u | u_{n+p}=e^{i\theta} u_n\}
\qquad
\langle u | v \rangle_\theta = \sum_{n=1}^p \bar{u}_nv_n$$
From Fourier analysis, there is a unitary operator ${{\mathcal F}}\colon \ell^2 ({{\mathbb{Z}}})\to {{\mathcal H}}_p$ so that ${{\mathcal F}}S^p{{\mathcal F}}^{-1}$ is multiplication by $e^{i\theta}{{\boldsymbol{1}}}$ and ${{\mathcal F}}J{{\mathcal F}}^{-1}$ acts fiber-wise (i.e., on each $\ell^2_\theta$) as a $p\times p$ matrix, $J(\theta)$. In particular, the eigenvalues of $J(\theta)$ are the solutions of , that is, they are the band functions $x_j(\theta)$.
Hyperelliptic Riemann Surfaces.
-------------------------------
As $\sqrt{\Delta^2 -4}$ appears repeatedly in the theory, it is natural that the associated Riemann surface should enter the analysis. $\Delta^2 -4$ has simple zeros at the edges of open gaps and at $\inf \sigma(J)$ and $\sup \sigma(J)$. It has double zeros at the closed gaps. Let $\ell$ denote the number of open gaps, then $\sqrt{\Delta^2 -4}$ has square root singularities at $2(\ell+1)$ points, and so its natural analyticity domain is the genus $\ell$ Riemann surface, ${{\mathcal S}}$, obtained by taking two copies of ${{\mathbb{C}}}\setminus\sigma(J)$, gluing at the bands and adding points at $\infty$. There is a natural projection $\pi \colon
{{\mathcal S}}\to {{\mathbb{C}}}\cup\{\infty\}$ which is $2$ to $1$ except at the branch points of $\sqrt{\Delta^2 -4}$. A similar analysis works for OPUC, but now there are $\ell$ gaps and the genus is $\ell-1$.
Minimal Herglotz and Carathéodory Functions.
--------------------------------------------
For a half-line periodic Jacobi matrix, the $m$ function is defined by $${\label}{1.40}
m(z) = \langle\delta_0, (J-z)^{-1}\delta_0\rangle\qquad\forall z\in{{\mathbb{C}}}\setminus\sigma(J)$$ This can be shown to obey a quadratic equation with polynomial coefficients $${\label}{1.41}
A(z) m(z)^2 + B(z) m(z) + C(z) =0$$ Moreover, these coefficients can be chosen to obey $${\label}{1.42}
B^2 -4AC=\Delta^2 - 4$$ This implies that $m(z)$ has meromorphic continuation to ${{\mathcal S}}$. Indeed, $m$ has minimal degree (i.e., degree $\ell+1$ in the OPRL case and $\ell$ in the OPUC case) among all meromorphic functions on ${{\mathcal S}}$ that are not of the form $g\circ\pi$ with $g$ meromorphic on the Riemann sphere. It can be shown that there is a one-one correspondence between minimal meromorphic functions obeying $\operatorname{Im}m(z) >0$ if $\operatorname{Im}z >0$ and $m(z) =-z^{-1} + O(z^{-2})$ on the top sheet of ${{\mathcal S}}$ and all periodic Jacobi parameters with the same $\Delta$. (We call these minimal Herglotz functions.)
There is a similar description for OPUC, but now one uses $${\label}{1.43x}
F(z) =\langle\delta_0, ({{\mathcal{C}}}+z) ({{\mathcal{C}}}-z)^{-1}\delta_0\rangle$$ which obeys $\operatorname{Re}F(z) >0$ if ${\lvertz\rvert}<1$ and $F(0)=1$. Again $F$ obeys a quadratic equation, showing that $F$ has a meromorphic continuation to ${{\mathcal S}}$ of minimal degree, and again there is a one-one correspondence between all $\{\alpha_n\}_{n=0}^{p-1}$ with the same $\Delta$ and all minimal Carathéodory functions.
Dirichlet Data.
---------------
One can describe the set of minimal Herglotz functions in terms of their poles. For each open gap, $\{G_j\}_{j=1}^\ell$, $\pi^{-1} ({\overline}{G}_j)\equiv T_j$ is a circle since $\pi$ is $2$ to $1$ on $G_j$ and one-one on ${\overline}{G}_j\setminus G_j$. A meromorphic Herglotz function has $\ell+1$ simple poles, one at $\infty$ on the second sheet and the other $\ell$, one in each $T_j$. Thus, the set of meromorphic Herglotz functions is homeomorphic to ${\mathop{\mathchoice{\smash{\vcenter{\hbox{\LARGE$\times$}}}\vphantom{\prod}}{\smash{\vcenter{\hbox{\Large$\times$}}}\vphantom{\prod}}{\times}{\times}}\displaylimits}_{j=1}^\ell T_j$ under the bijective map from such functions to its poles. A similar analysis holds for OPUC but now there is no pole at infinity, there are $\ell$ gaps, and ${\mathop{\mathchoice{\smash{\vcenter{\hbox{\LARGE$\times$}}}\vphantom{\prod}}{\smash{\vcenter{\hbox{\Large$\times$}}}\vphantom{\prod}}{\times}{\times}}\displaylimits}_{j=1}^\ell T_j$ describes the possible poles. The difference is that for OPRL, the dimension of the torus is $\ell$, and for OPUC it is $\ell-1$.
Isospectral Tori. {#ssIT}
-----------------
By combining the bijective maps from periodic OPRL to minimal Herglotz functions and of such functions to Dirichlet data, we see for a $\Delta$ of period $p$ with $\ell$ gaps, $$\{(a_n, b_n)_{n=1}^p : \text{the discriminant is }\Delta\}$$ is an $\ell$-dimensional torus in ${{\mathbb{R}}}^{2p}$. Generically, $\ell=p-1$. In the OPUC case, generically $\ell=p$ and the torus is naturally embedded in ${{\mathbb{C}}}^p$. This torus is the isospectral torus which we will denote by ${{\mathcal T}}$ or ${{\mathcal T}}_{J_0}$ if a given periodic $J_0$ underlies our construction. For clarity of exposition, we will typically blur the distinction between $p$-tuples $(a_n, b_n)_{n=1}^p$ and the corresponding infinite sequences $\{a_n,b_n\}_{n\in{{\mathbb{Z}}}}$ of period $p$. Because of our perturbation theory viewpoint, we use $J_0$ to label the torus, but we emphasize that from another point of view, the torus is associated to the set $\sigma_{\text{\rm{ess}}}(J_0)$ and not to $J_0$.
Isospectral Flows.
------------------
The fact that spaces of $p$-periodic coefficients foliate into tori suggests that there is some kind of completely integrable system in the background. That is true: it is the Toda flow in the OPRL case and the defocusing Ablowitz–Ladik flow in the OPUC case. Since we will not need these below, we say no more about them, but see Chapter 6 of [@SimonNew] for the OPRL case and Section 11.11 of [@OPUC2] for OPUC.
The Magic Formula for Jacobi Matrices
=====================================
[s2]{} Our goal in this section is to prove
[T2.1]{} Let $J_0$ be a two-sided $p$-periodic Jacobi matrix with discriminant $\Delta_{J_0}(x)$ and isospectral torus ${{\mathcal T}}_{J_0}$. Let $J$ be a two-sided [[(]{}]{}not a priori periodic[[)]{}]{} Jacobi matrix. Then $${\label}{2.1}
\Delta_{J_0}(J) = S^p + S^{-p} \iff J\in{{\mathcal T}}_{J_0}$$ where $S$ is the right shift, , on $\ell^2({{\mathbb{Z}}})$.
We provide two proofs of the ‘harder’ direction $\Rightarrow$ or rather of $$\label{2.11a}
\Delta_{J_0}(J) =S^p + S^{-p}\Rightarrow J\text{ is periodic}$$ which is the key step. Our first proof is immediately below; the second, suggested to us by L. Golinskii, appears after Lemma \[L2.4A\].
[L2.2]{} Let $\ell=1,2,\dots$. Then $${\label}{2.2}
(J^\ell)_{m,m+k} = \begin{cases}
0 & k>\ell \\
a_m a_{m+1} \cdots a_{m+k-1} & k=\ell \\
a_m a_{m+1} \cdots a_{m+\ell-2} (b_m + b_{m+1} + \cdots + b_{m+\ell -1}) & k=\ell-1
\end{cases}$$
Writing $${\label}{2.3}
(J^\ell)_{m,m+k} = \sum_{i_1, \dots, i_{\ell-1}} J_{m, i_1} J_{i_1, i_2} \cdots
J_{i_{\ell-1}, m+k}$$ We see that since $J$ is tridiagonal, all terms are zero if $k>\ell$, that we must have (with $i_0\equiv m$, $i_\ell = m+k$) that $i_q - i_{q-1}=1$ for $q=1, \dots,
\ell$ if $k=\ell$, and that if $k=\ell-1$, $i_q -i_{q-1}=1$ for all but one $q\in
\{1,\dots,\ell\}$ and it is zero for that $q$.
[L2.4]{} If $J$ and $J_0$ are periodic, then $\sigma(J)=\sigma(J_0)$ if and only if $\Delta_J=\Delta_{J_0}$.
This lemma says that the spectrum determines the discriminant and vice versa. That the discriminant determines the spectrum is elementary: $\sigma = \{ x : \Delta(x)\in[-2,2]\}$. Therefore we only prove the other direction—indeed, we give two proofs.
Harmonic measure $d\nu$ is intrinsic to the set $\sigma$; it is the solution of an electrostatic problem there. But then $d\nu$ determines $\Delta$ via .
$\sigma$ determines the gaps—even closed gaps—via harmonic measure. The gap edges determine the zeros of $\Delta -2$ and so $\Delta -2$ up to a constant. The zeros of $\Delta +2$ then determine the constant.
For all $\theta\in[0,2\pi)$, $J(\theta)$ is self-adjoint and so diagonalizable. Moreover, the eigenvalues of $J_0(\theta)$ are precisely the roots of $\Delta(x)=2\cos(\theta)$. Thus$${\label}{2.5}
\Delta_{J_0}(J_0 (\theta)) = (2\cos\theta){{\boldsymbol{1}}}$$ But then $\Delta_{J_0}(J_0)$ and $S^p + S^{-p}$ both have direct integral decomposition with fibers $(2\cos\theta){{\boldsymbol{1}}}$, so $${\label}{2.6}
\Delta_{J_0}(J_0) = S^p + S^{-p}$$ Since $J\in {{\mathcal T}}_{J_0}\Rightarrow \Delta_J=\Delta_{J_0}$, this proves $\Leftarrow$ in .
Now suppose LHS of holds. By , , and $${\label}{2.7}
(S^p + S^{-p})_{m, m+p}=1 \qquad
(S^p + S^{-p})_{m, m+p-1} =0$$ this implies $${\label}{2.8}
a_m \cdots a_{m+p-1} = a_1^{(0)} \cdots a_p^{(0)}$$ and $${\label}{2.9}
\sum_{j=0}^{p-1} b_{m+j} =\sum_{j=0}^{p-1} b_{j+1}^{(0)}$$ In particular, $$a_m\cdots a_{m+p-1} = a_{m+1} \cdots a_{m+p} \qquad
\sum_{j=0}^{p-1} (b_{m+j+1}-b_{m+j}) =0$$ which lead to $${\label}{2.10}
a_m = a_{m+p} \qquad b_m = b_{m+p}$$ so $J$ is periodic.
Since $J$ is periodic, $\Delta_J(J)=S^p + S^{-p}$; moreover, $\Delta_{J_0}(J)=S^p + S^{-p}$ by hypothesis. Thus we learn that applying the polynomial $\Delta_{J}-\Delta_{J_0}$ to $J$ gives zero. By the $k=\ell$ case of , it must therefore be the zero polynomial, that is, $\Delta_{J}=\Delta_{J_0}$. Lemma \[L2.4\] now completes the proof.
1\. Showing that $J$ was periodic only required equality in $\Delta_{J_0}({{\mathcal T}})=S^p + S^{-p}$, for the two most extreme upper (or lower) diagonals. Nevertheless, $J\in{{\mathcal T}}_{J_0}$ requires equality everywhere.
2\. We need not suppose a priori that each $a_n >0$ and can allow some $a_n =0$ ($J$ can still be defined on $\ell^2 ({{\mathbb{Z}}})$), for implies that if LHS of holds, then each $a_n >0$.
We now turn to our second proof of .
[L2.4A]{} Let $A$ be a two-sided [[(]{}]{}bounded[[)]{}]{} infinite matrix of finite width [[(]{}]{}i.e., for some $w$, we have that ${\lvertk-\ell\rvert} > w
\Rightarrow A_{k\ell} =0$[[)]{}]{}. Suppose $${\label}{2.11b}
[A, S^p + S^{-p}] =0$$ for some $p$, then $${\label}{2.11c}
[A,S^p] =0$$
1\. This is Lemma 2 in [@Nai2]; no proof is given.
2\. $[A,B]\equiv AB-BA$
3\. has an equivalent form: $${\label}{2.11d}
[A,S^p]=0 \iff A_{k+p,\ell+p}=A_{k,\ell} \qquad \text{for all $k,\ell$}$$
4\. As shows, $[J,S^p] =0$ for a Jacobi matrix if and only if $a_k$ and $b_k$ are $p$-periodic.
Since $A$ has finite width, we can find diagonal matrices $D_{k_1},D_{k_1+1}, \ldots,
D_{k_2}$ with $D_{k_1}\neq 0\neq D_{k_2}$, so that $${\label}{2.11e}
A=\sum_{j=k_1}^{k_2} D_j S^j$$ Since $D_j$ is diagonal, so is $S^p D_j S^{-p}$. Thus $$\begin{gathered}
(S^p + S^{-p})A = \sum_{j=k_1}^{k_2} (S^p D_j S^{-p}) S^{j+p} + \sum_{j=k_1}^{k_2}
(S^{-p} D_j S^p) S^{j-p} \\
A (S^p + S^{-p}) =\sum_{j=k_1}^{k_2} D_j S^{j+p} + \sum_{j=k_1}^{k_2} D_j S^{j-p}\end{gathered}$$ Since the composition uniquely determines each $D_j$, implies $$S^p D_{k_2} S^{-p} = D_{k_2}$$ that is, $D_{k_2}$ is periodic. Thus, $D_{k_2} S^p$ commutes with $S^p + S^{-p}$, so we can remove it from without losing . This shows inductively that each $D_j$ is periodic.
$J$ commutes with $\Delta (J)$, so holds.
Our next goal is to compare ${\tilde }d_m ((a,b),{{\mathcal T}})$ given by and $d_m
((a,b),{{\mathcal T}})$ given by . As well as satisfying natural curiosity, this relation also plays an important role (via Theorem \[T9.11\]) in the proofs of Theorems \[T1.3\] and \[T1.4\].
To capture the essence of what follows, let us pause to ponder the following: suppose $\tilde d_m((a,b),{{\mathcal T}})=0$ for all $m$, does this mean that $(a,b)\in{{\mathcal T}}$? The hypothesis tells us that each length-$p$ block belongs to the isospectral torus; it does not a priori even guarantee that the coefficients are periodic. Example \[E4.5\] shows that periodicity can fail in the OPUC case. However, such problems do not arise for OPRL. The reason is simple: within the isospectral torus, $a_1,\ldots,a_{p-1}$ determines $a_p$ and $b_1,\ldots,b_{p-1}$ determines $b_p$.
[P2.5]{} Given a $p$-periodic Jacobi matrix $J_0$, $1\leq q \leq \infty$, and ${\varepsilon}>0$, there is a constant $C$ so that $${\label}{2.13}
e^{1-p} \bigl\| \tilde d_m \bigl((a,b),{{\mathcal T}}_{J_0}\bigr) \bigl\|_{\ell^q}
\leq \bigl\| d_m \bigl((a,b),{{\mathcal T}}_{J_0}\bigr) \bigl\|_{\ell^q}
\leq C \bigl\| \tilde d_m \bigl((a,b),{{\mathcal T}}_{J_0}\bigr) \bigl\|_{\ell^q}$$ for all sequences $\{(a_n,b_n)\}$ obeying ${\varepsilon}^{-1} > a_n > {\varepsilon}>0$. All $\ell^q$ norms are over $m\in\{1,2,3,\ldots\}$.
The key input is
[L2.6]{} Given $\{(a_n,b_n)\}$ obeying ${\varepsilon}^{-1} > a_n > {\varepsilon}>0$, $$\bigl| a_n^{ } - a^{(0)}_n \bigr| + \bigl|b_n^{ } - b^{(0)}_n \bigr|
\leq \tilde d_{m}\bigl( (a,b), (a^{(0)},b^{(0)}) \bigr) +
C \sum_{r=m}^{n-p+1} \tilde d_{r}\bigl( (a,b), {{\mathcal T}}_{J_0} \bigr)$$ for all $n\geq m$. The constant $C$ depends only on $\epsilon$.
The proof is by induction on $n$. For $m\leq n\leq m+p-1$, the result is immediate from the definition of $\tilde d_m$.
For $n>m+p-1$, we consider the functions $$\begin{aligned}
f(a_1,\ldots,a_p) := \sum_{j=1}^p \bigl[ \log(a_j) - \log( a_j^{(0)} ) \bigr]
\qquad
g(b_1,\ldots,b_p) := \sum_{j=1}^p \bigl[ b_j - b_j^{(0)} \bigr]\end{aligned}$$ These vanish on ${{\mathcal T}}_{J_0}$, as explained in the proof of Theorem \[T2.1\].
As $g$ is Lipschitz (with constant $1$), $$\begin{aligned}
\bigl|b_{n} - b_{n-p} \bigr|
&= \bigl| g(b_{n},\ldots,b_{n-p+1}) - g(b_{n-1},\ldots,b_{n-p}) \bigr| \\
&\leq \bigl| g(b_{n},\ldots,b_{n-p+1}) \bigr| + \bigl| g(b_{n-1},\ldots,b_{n-p}) \bigr| \\
&\leq \tilde d_{n-p+1}\bigl( (a,b), {{\mathcal T}}_{J_0} \bigr)
+ \tilde d_{n-p}\bigl( (a,b), {{\mathcal T}}_{J_0} \bigr)\end{aligned}$$ In a similar way, $$\begin{aligned}
\bigl|\log[a_n] - \log[a_{n-p}] \bigr|
&= \bigl| f(a_{n},\ldots,a_{n-p+1}) - f(a_{n-1},\ldots,a_{n-p}) \bigr|\end{aligned}$$ leads to $$\begin{aligned}
\bigl| a_n - a_{n-p} \bigr|
&\leq C_\epsilon \bigl[ \tilde d_{n-p+1}\bigl( (a,b), {{\mathcal T}}_{J_0} \bigr)
+ \tilde d_{n-p}\bigl( (a,b), {{\mathcal T}}_{J_0} \bigr) \bigr]\end{aligned}$$
Combining these two inequalities gives $$\begin{aligned}
\bigl| a_n^{ } - a^{(0)}_n \bigr| + \bigl|b_n^{ } - b^{(0)}_n \bigr|
&\leq \bigl| a_{n-p}^{ } - a^{(0)}_{n-p} \bigr| + \bigl|b_{n-p}^{ } - b_{n-p}^{(0)} \bigr| \\
&\qquad+ (1+C_{\varepsilon}) \bigl[ \tilde d_{n-p+1}\bigl( (a,b),
{{\mathcal T}}_{J_0} \bigr)
+ \tilde d_{n-p}\bigl( (a,b), {{\mathcal T}}_{J_0} \bigr) \bigr]\end{aligned}$$ which completes the proof of the inductive step.
The left-hand inequality in follows immediately from the definitions of $d_m$ and $\tilde d_m$; we focus on the second inequality.
Choose $(a^{(0)},b^{(0)})$ minimizing $d_m((a,b),{{\mathcal T}}_{J_0})$; strictly, this amounts to a (inconsequential) change in $J_0$. Applying Lemma \[L2.6\] in the definition of $d_m$ gives $$\begin{aligned}
d_m\bigl((a,b),{{\mathcal T}}_{J_0}\bigr) &\leq \tfrac{e}{e-1}\, \tilde d_m\bigl((a,b),{{\mathcal T}}_{J_0}\bigr)
+ C \sum_{k=0}^\infty \sum_{r=m}^{m+k} e^{-k} \, \tilde d_{r} \bigl((a,b),{{\mathcal T}}_{J_0}\bigr) \\
&\leq C' \sum_{j=0}^\infty e^{-j} \, \tilde d_{m+j} \bigl((a,b),{{\mathcal T}}_{J_0}\bigr)\end{aligned}$$ The proposition follows because convolution with $e^{-j}\chi_{[0,\infty)}(j)$ is a bounded operator on all $\ell^q$ spaces.
The Magic Formula for CMV Matrices
==================================
[s3]{} Our goal in this section is to prove
[T3.1]{} Let $p$ be even and let ${{\mathcal{C}}}_0$ be a two-sided $p$-periodic CMV matrix with discriminant $\Delta_{C_0}(z)$ and isospectral torus ${{\mathcal T}}_{{{\mathcal{C}}}_0}$. Given a two-sided [[(]{}]{}not a priori periodic[[)]{}]{} CMV matrix, ${{\mathcal{C}}}$, $${\label}{3.1}
\Delta_{{{\mathcal{C}}}_0}({{\mathcal{C}}}) =S^p + S^{-p} \iff {{\mathcal{C}}}\in{{\mathcal T}}_{{{\mathcal{C}}}_0}$$
1\. Notice that since ${{\mathcal{C}}}$ is unitary and $\Delta (e^{i\theta})$ is real, $\Delta_{{{\mathcal{C}}}_0}({{\mathcal{C}}})$ is self-adjoint.
2\. By and the fact that ${{\mathcal{C}}}$ is five-diagonal, $\Delta_{{{\mathcal{C}}}_0}
({{\mathcal{C}}})$ has $2(p/2)$ diagonals above/below the main diagonal.
3\. As in Section \[s2\], we will first present our initial proof that $${\label}{3.1a}
\Delta_{{{\mathcal{C}}}_0}({{\mathcal{C}}})=S^p + S^{-p} \Rightarrow \{\alpha_n\} \text{ is periodic}$$ and then a proof based on Golinskii’s suggestion.
[L3.2]{} We have: $$\begin{aligned}
&({{\mathcal{C}}}^\ell)_{m,m+k}= ({{\mathcal{C}}}^{-\ell})_{m,m+k} = 0 \qquad \text{ if }k>2\ell {\label}{3.2} \\
&({{\mathcal{C}}}^\ell)_{2m,2m+2\ell} = \rho_{2m} \rho_{2m+1} \dots \rho_{2m+2\ell-1} {\label}{3.3} \\
&({{\mathcal{C}}}^\ell)_{2m+1, 2m+2\ell+1} = 0 {\label}{3.4} \\
&({{\mathcal{C}}}^{-\ell})_{2m, 2m+2\ell} = 0 {\label}{3.5} \\
&({{\mathcal{C}}}^{-\ell})_{2m+1, 2m+\ell +1} = \rho_{2m+1} \rho_{2m+2} \dots \rho_{2m+2\ell} {\label}{3.6} \\
&({{\mathcal{C}}}^\ell)_{2m, 2m+2\ell -1} = \rho_{2m} \rho_{2m+1} \dots \rho_{2m+2\ell-2}
\bar\alpha_{2m+2\ell-1} {\label}{3.7} \\
&({{\mathcal{C}}}^\ell)_{2m+1, 2m+2\ell} = -\alpha_{2m} \rho_{2m+1} \dots \rho_{2m+2\ell-1} {\label}{3.8} \\
&({{\mathcal{C}}}^{-\ell})_{2m, 2m+2\ell-1} = -\bar\alpha_{2m-1} \rho_{2m} \rho_{2m+1} \dots
\rho_{2m+2\ell-2} {\label}{3.9} \\
&({{\mathcal{C}}}^{-\ell})_{2m+1, 2m+2\ell} = \rho_{2m+1} \dots \rho_{2m+2\ell-1}
\alpha_{2m+2\ell} {\label}{3.10}\end{aligned}$$
As ${{\mathcal L}}$ and ${{\mathcal M}}$ are tridagonal, ${{\mathcal{C}}}^\ell$ is a product of $2\ell$ tridiagonal matrices, so is immediate.
We will prove the results for ${{\mathcal{C}}}^\ell$. The results for ${{\mathcal{C}}}^{-\ell}$ are similar if we note $${\label}{3.15}
\Theta (\alpha)^{-1} = \Theta (\bar\alpha)$$ since $\Theta$ is unitary and symmetric.
Equation follows from $${\label}{3.16}
{{\mathcal L}}_{2m, 2m+1} = \rho_{2m} \qquad
{{\mathcal M}}_{2m+1, 2m+2} = \rho_{2m+1}$$ and from $${\label}{3.17}
{{\mathcal L}}_{2m+1, 2m+2} =0$$ Because of , the only way for ${{\mathcal{C}}}^\ell$ to get from $2m$ to $2m+2\ell-1$ is to increase index in the first $2\ell-1$ factors, which leads to . For the same reason, to get from $2m+1$ to $2m+2\ell$, the last $2\ell -1$ factor must increase index, leading to .
[L3.3]{} If ${{\mathcal{C}}}$ and ${{\mathcal{C}}}_0$ are $p$-periodic, then $\sigma ({{\mathcal{C}}})=\sigma ({{\mathcal{C}}}_0)$ if and only if $\Delta_{{\mathcal{C}}}=\Delta_{{{\mathcal{C}}}_0}$.
Either proof of Lemma \[L2.4\] carries over with no change.
The proof that $${\label}{3.18}
\Delta_{{{\mathcal{C}}}_0} ({{\mathcal{C}}}_0) =S^p + S^{-p}$$ is identical to the proof of .
For the converse, suppose $${\label}{3.19}
\Delta_{{{\mathcal{C}}}_0}({{\mathcal{C}}}) = S^p + S^{-p}$$ In particular, $${\label}{3.20}
\Delta_{{{\mathcal{C}}}_0} ({{\mathcal{C}}})_{2m,2m+p-1} =0$$
By and Lemma \[L3.2\], this implies (recall $p$ is even) $$(\rho_0^{(0)}\dots\rho_{p-1}^{(0)})^{-1} (\rho_{2m} \rho_{m+1} \dots
\rho_{2m + p-2}) (\bar\alpha_{2m+p-1} - \bar\alpha_{2m-1}) = 0$$ so $${\label}{3.21}
\alpha_{2m+p-1} = \alpha_{2m-1}$$
Similarly, since $$\Delta_{{{\mathcal{C}}}_0} ({{\mathcal{C}}})_{2m+1, 2m+p} =0$$ we get $$(\rho_0^{(0)}\dots\rho_{p-1}^{(0)})^{-1} (\rho_{2m+1} \dots \rho_{2m+p-1})
(\alpha_{2m+2}-\alpha_{2m+p-1}) =0$$ which leads to $${\label}{3.22}
\alpha_{2m+p} = \alpha_{2m}$$
Thus, $\alpha$ has period $p$. That ${{\mathcal{C}}}\in{{\mathcal T}}_{{{\mathcal{C}}}_0}$ follows from Lemma \[L3.3\] and the same argument used in the OPRL case.
Next, we give a proof using Naĭman’s lemma. We will need
[L3.3A]{} Let ${{\mathcal{C}}}$ be the extended CMV matrix associated to $\{\alpha_n\}_{n=-\infty}^\infty$. Let $p$ be even. If $S^p{{\mathcal{C}}}={{\mathcal{C}}}S^p$, then $${\label}{3.22a}
\alpha_{n+p} =\alpha_n$$ for all $n$.
We have that ${{\mathcal{C}}}_{2j\,\, 2j+1}^2 + {{\mathcal{C}}}_{2j\,\, 2j+2}^2 =
\rho_{2j}^2$ (see (4.2.14) of [@OPUC1]), so $\rho_{2j}$ is periodic. Thus, ${{\mathcal{C}}}_{2j\,\, 2j+2}/\rho_{2j}=\rho_{2j+1}$ is also periodic. So $\bar\alpha_{2j+1}
={{\mathcal{C}}}_{2j\,\, 2j+1}/\rho_{2j}$ is periodic as is $\alpha_{2j} ={{\mathcal{C}}}_{2j+1\,\,2j+2}/
(-\rho_{2j+1})$.
${{\mathcal{C}}}$ commutes with $S^p + S^{-p}$, so by Naĭman’s lemma (Lemma \[L2.4A\]), which did not require that $A$ be self-adjoint, $S^p{{\mathcal{C}}}={{\mathcal{C}}}S^p$, which implies $\alpha$ is periodic by Lemma \[L3.3A\].
We now turn to the OPUC version of Proposition \[P2.5\]. As noted in the introduction, it is not sufficient to sum over exactly one period:
\[E4.5\] $(0,{\frac}12, 0, {\frac}12, 0, \dots)$ and $(0,-{\frac}12, 0, -{\frac}12, 0, \dots)$ are in the same isospectral torus, namely, the one with $p=2$ and $$\Delta(z) = \sqrt{\tfrac43}\, (z+z^{-1})$$ Now consider $\alpha =(0,{\frac}12, 0, -{\frac}12, 0, {\frac}12, 0, -{\frac}12, \dots)$. If ${\tilde }d_m (\alpha,
{{\mathcal T}}_{{{\mathcal{C}}}_0})$ were defined as sum from $k=0$ to $p-1$, it would be zero for all $m$, but $d_m (\alpha, {{\mathcal T}}_{{{\mathcal{C}}}_0})$ is not small.
The problem, as this example shows, is that for sequences in ${{\mathcal T}}_{{{\mathcal{C}}}_0}$, $(\alpha_0,\dots, \alpha_{p-2})$ does not determine $\alpha_{p-1}$. But by periodicity, $\alpha_0, \dots, \alpha_{p-1}$ determines $\alpha_p$. Thus, if we define $${\label}{3.23}
{\tilde }d_m (\alpha,\alpha'):=\sum_{k=0}^p\, {\lvert\alpha_{m+k}-\alpha'_{m+k}\rvert}$$ then $$| \alpha_{m+p} - \alpha_m | \leq \tilde d_m(\alpha,{{\mathcal T}}_{{{\mathcal{C}}}_0})$$ Plugging this into the proofs of Lemma \[L2.6\] and Proposition \[P2.5\] leads quickly to
[P3.5]{} Let ${{\mathcal{C}}}_0$ be a fixed periodic CMV matrix, then $${\label}{3.24}
e^{-p} \bigl\| \tilde d_m (\alpha,{{\mathcal T}}_{{{\mathcal{C}}}_0}) \bigl\|_{\ell^q}
\leq \bigl\| d_m (\alpha,{{\mathcal T}}_{{{\mathcal{C}}}_0}) \bigl\|_{\ell^q}
\leq C \bigl\| \tilde d_m (\alpha,{{\mathcal T}}_{{{\mathcal{C}}}_0}) \bigl\|_{\ell^q}$$ for any sequence of Verblunsky coefficients $\{\alpha_n\}$.
The Magic Formula for Schrödinger Operators
===========================================
[s4]{} In this section, we want to illustrate the potential applicability of our central idea to the spectral theory of one-dimensional Schrödinger operators, $$\label{SO}
H = -{\frac}{d^2}{dx} + V(x)$$ However, we will not pursue the applications in this paper.
We will suppose $V\in L_{\text{\rm{unif}}}^1$, that is, $\sup_x \int_{x-1}^{x+1}
{\lvertV(y)\rvert}\, dy <\infty$. In that case, $V$ is a form bounded perturbation of $-{\frac}{d^2}{dx^2}$ on $L^2 ({{\mathbb{R}}}, dx)$ with relative bound zero, so $H$ is a self-adjoint operator. Its form domain is the Sobolev space $H^1({{\mathbb{R}}})$.
We need to say something about periodic Schrödinger operators. Suppose $V_0$ has period $L$, that is, $${\label}{4.1}
V_0 (x+L) = V_0(x)$$
For arbitrary $V$ in $L_{\text{\rm{loc}}}^1$ and $E\in{{\mathbb{C}}}$, let $u_D (x,E;V)$ and $u_N
(x,E;V)$ (we will often drop the $V$ if it is clear which $V$ is intended) be the solutions of $${\label}{4.2}
-u'' + Vu =Eu$$ obeying the boundary conditions $${\label}{4.3}
u_D(0)=0 \qquad u'_D(0)=1 \qquad u_N(0)=1 \qquad u'_N(0)=0$$ There is a unique solution of in distributional sense which is absolutely continuous.
The transfer matrix that updates solutions of (with data written as $\binom{u}{u'}$) is $${\label}{4.4}
T(x,E;V) = \begin{pmatrix}
u_N(x,E) & u_D(x,E) \\
u'_N(x,E) & u'_D(x,E)
\end{pmatrix}$$ $\det(T)=1$ by constancy of the Wronskian. For periodic $V_0$, we define the discriminant by $$\begin{aligned}
\Delta_{V_0}(E) &= {\text{\rm{Tr}}}(T(L,E; V_0)) \notag \\
&= u_N (L,E) + u'_D(L,E) {\label}{4.5}\end{aligned}$$
As in the OPRL and OPUC cases, it is easy to see for the whole-line operator that $${\label}{4.6}
\sigma \bigl( -\tfrac{d^2}{dx^2} + V_0 \bigr) = \Delta_{V_0}^{-1} ([-2,2])$$ and is purely absolutely continuous. Moreover (see, e.g., [@RS4]), if $${\label}{4.7}
(S_y u)(x) = u(x-y)$$ then $H=-{\frac}{d^2}{dx^2}+V_0$ commutes with $S_L$ and so has a direct integral decomposition, $${\label}{4.8}
H=\int^\oplus H(\theta)\, {\frac}{d\theta}{2\pi}$$ whose fibers, $H(\theta)$, are the operator on $[0,L]$ with $${\label}{4.9}
u(L)=e^{i\theta} u(0) \qquad u'(L) =e^{i\theta} u'(0)$$ boundary conditions. $H(\theta)$ has purely discrete spectrum (i.e., $(H(\theta)
+i)^{-1}$ is compact); the eigenvalues are precisely the solutions of $${\label}{4.10}
\Delta (E) =2\cos (\theta)$$
Two periodic potentials of period $L$ are called isospectral if and only if they have the same $\Delta$. As in the Jacobi and CMV cases, the spectrum determines $\Delta$, but this is more difficult to prove in the Schrödinger case. It is also known ([@McKT] for nice $V_0$’s; [@BuFi; @FIT; @Iwa; @RT]) that the set of $V$’s isospectral to $V_0$ is a torus of dimension equal to the number of gaps which is typically infinite, so we will refer to an isospectral torus, ${{\mathcal T}}_{V_0}$. We can now state the main result in this section:
[T4.1]{} Let $V_0$ be periodic obeying and let $\Delta_{V_0}$ be its discriminant and ${{\mathcal T}}_{V_0}$ its isospectral torus. Let $V$ be in $L_{{\text{\rm{loc}}},{\text{\rm{unif}}}}^1$ on ${{\mathbb{R}}}$ and $H=-{\frac}{d^2}{dx^2} +V$. Then $${\label}{4.11}
\Delta_{V_0}(H) = S_L + S_{-L}\iff V\in{{\mathcal T}}_{V_0}$$ Here $S_{\pm L}$ denotes the shift operator, as in .
1\. $\Delta_{V_0}(H)$ is defined by the functional calculus.
2\. As in the last two sections, we will provide our initial proof that $${\label}{4.11a}
\Delta_{V_0}(H) = S_L + S_{-L}\Rightarrow V \text{ periodic}$$ and then a simpler proof using an analog of Naĭman’s lemma. This argument does not require Theorems \[T4.2\] and \[T4.3\] and the considerable machinery their proofs entail. That said, to show $\Delta_{V_0}(H)=S_L + S_{-L}$ plus $V$ periodic implies $V\in{{\mathcal T}}_{V_0}$ [*does*]{} require Theorem \[T4.3\], but it should be noted that one can prove Theorem \[T4.3\] fairly easily without needing transformation formulae of Delsarte, Levitan, Gel’fand, Marchenko type.
We need two preliminaries whose proofs we defer to later in the section. We first make a definition:
For any $y>0$, ${{\mathcal R}}_y$ consists of operators on $L^2 ({{\mathbb{R}}})$ of the form $${\label}{4.12}
(Af)(x) = \tfrac12\, f(x+y) + \tfrac12\, f(x-y) + \int_{x-y}^{x+y}
K(x,z)f(z)\, dz$$ where $K$ is continuous and uniformly bounded on $\{(x,z) : {\lvertx-z\rvert} \leq y\}$.
[*Note.*]{} It can happen that $K(x,x\pm y)\neq 0$, so if we think of $K$ as an integral kernel on ${{\mathbb{R}}}\times{{\mathbb{R}}}$, it can be discontinuous at ${\lvertx-z\rvert}=y$.
[T4.2]{} If $V_0$ is $L$-periodic and $V$ in $L_{{\text{\rm{loc}}},{\text{\rm{unif}}}}^1$, then ${\frac}12 \Delta(H)\in{{\mathcal R}}_L$ and $${\label}{4.13}
K(x,x+L)=-\tfrac14 \int_x^{x+L} (V(z) - V_0(z))\, dz$$
Note that describes the ‘matrix elements’ of $\Delta_{V_0} (H)-(S_L + S_{-L})$ that are farthest from the diagonal. Indeed, just as in the other cases, one does not need the full statement $\Delta_{V_0} (H)=S_L + S_{-L}$ to see that $V$ is periodic, only that $\langle f, (\Delta (H)-S_L - S_{-L})g\rangle
=0$ for $f$ supported near $x_0$ and $g$ near $x_0 +L$ (for all $x_0$).
[T4.3]{} $\Delta(E)$ is an entire function which obeys $$\begin{aligned}
{2}
& \text{\rm{(i)}} \qquad && {\lvert\Delta(E)\rvert} \leq C\exp \bigl( L\sqrt{{\lvertE\rvert}}\bigr) {\label}{4.14} \\
& \text{\rm{(ii)}} \qquad && \lim_{\substack{E\to -\infty \\ E\text{ real}}}
\, {\frac}{\Delta(E)}{\exp (L\sqrt{{\lvertE\rvert}})}=1 {\label}{4.15}\end{aligned}$$
If $V\in{{\mathcal T}}_{V_0}$, then $\Delta_V =
\Delta_{V_0}$, so for the $\Leftarrow$ direction we need only prove $$\Delta_{V_0}\bigl( \tfrac{d^2}{dx^2} + V_0) = S_L + S_{-L}$$ As before, this is equivalent to $\Delta_{V_0}(H(\theta))=2\cos\theta$ which follows from .
Conversely, if $\Delta_{V_0}(H)=S_L + S_{-L}$, then from Theorem \[T4.2\] and the periodicity of $V_0$, we see $${\label}{4.16}
\int_x^{x+L} V(z)\, dz = \text{constant}$$ This implies that $V(x+L)-V(x)=0$ for a.e. $x$, that is, $V$ is periodic.
If $H(\theta)$ are the fibers of $H$ in the direct integral decomposition, $\Delta_{V_0}(H)=S_L + S_{-L}$ implies $${\label}{4.17}
\Delta_{V_0}(H(\theta))=2\cos\theta$$ so, if $\Delta$ is the discriminant for $V$, we have $\Delta
(z)=\pm 2\Rightarrow \Delta_{V_0}(z) =\pm 2$. Moreover, implies $\sigma(H)
\subseteq \sigma (-{\frac}{d^2}{dx^2} + V_0)$, so any double zero of $\Delta\pm 2$ is a double zero of $\Delta_{V_0}\pm 2$. It follows that $${\label}{4.18}
g(z) = {\frac}{\Delta_{V_0}^2 (z) -4}{\Delta^2(z) -4}$$ is analytic.
Since $\Delta_{V_0}$ and $\Delta$ are entire functions of order ${\frac}12$ (by Theorem \[T4.3\]), $g(z)$ is of the form $${\label}{4.19}
g(z) = C\prod_{j=1}^J \biggl( 1-{\frac}{z}{z_j}\biggr)$$ where $z_1 < z_2 < \cdots$ are bounded from below. By , $\lim_{E\to -\infty}
g(E)=1$, which implies $g\equiv 1$, that is, $\Delta = \Delta_{V_0}$.
The argument used at the end of the proof to conclude that missing zeros cannot occur is reminiscent of ideas connected with the Hochstadt–Lieberman [@HL] and related theorems [@GShpa; @GStams].
We now turn to the proofs of Theorems \[T4.2\] and \[T4.3\]. A critical role is played by the wave equation and the transformation operator formalism of Gel’fand–Levitan, further important work is due to Delsarte, Levin, and Marchenko; see the book of Marchenko [@Marc] for references and history.
Define for $s>0$, $${\label}{4.20}
C_s(z) =\cos \bigl(s\sqrt{z}\bigr) \qquad
S_s(z) = z^{-1/2} \sin \bigl(s\sqrt{z}\bigr)$$ which are entire functions of $z$ bounded on $(a,\infty)$ for any $a\in{{\mathbb{R}}}$. Thus $C_s(H)$ and $S_s(H)$ are bounded operators for any $H$ that is bounded from below. We will need to study the form of $C_s (-{\frac}{d^2}{dx^2}+V)$. For bounded continuous $V$, this is discussed in Marchenko [@Marc]. While his proofs extend to the $L_{\text{\rm{loc}}}^1$ case, it seems simpler to sketch the ideas:
[P4.4]{} $C_s^0:=C_s (-{\frac}{d^2}{dx^2})\in{{\mathcal R}}_s$; indeed, $${\label}{4.21}
(C_s^0 f)(x) =\tfrac12\, [f(x+s) + f(x-s)]$$ If $S_s^0:=S_s (-{\frac}{d^2}{dx^2})$, then $${\label}{4.22}
(S_s^0 f)(x) = \tfrac12 \int_{x-s}^{x+s} f(y)\, dy$$
If $w(x,s):=(C_s^0 f)(x)+(S_sg)(x)$, then $w$ obeys the wave equation $({\frac}{\partial^2}{\partial s^2} -{\frac}{\partial^2}{\partial x^2}) w=0$ with initial data $w(x,0)=f$ and $\partial_s w(x,0)=g(x)$. Thus the proposition basically encodes d’Alembert’s solution of the wave equation. From this point of view, Theorem \[T4.2\] is connected to finite propagation speed for the wave equation.
Since $\cos$ is even, $${\label}{4.23}
\cos (s{\lvertk\rvert})=\cos (sk) = \tfrac12\, (e^{iks} + e^{-iks})$$ is just the Fourier transform of this. follows from $${\label}{4.24}
S_s(z) = \int_0^s C_t(z)\, dt$$ and .
We are heading towards
[T4.5]{} Let $V\in L_{{\text{\rm{loc}}},{\text{\rm{unif}}}}^1 ({{\mathbb{R}}})$ and let $H=-{\frac}{d^2}{dx^2}+V$. Then $C_s(H)\in{{\mathcal R}}_s$ and the associated kernel $K_s$ of obeys $${\label}{4.25}
K_s (x,x+s) = -\tfrac14\int_x^{x+s} V(u)\, du$$ and for each $t\in (0,\infty)$, $${\label}{4.26}
\sup_{x,y, {\lverts\rvert}\leq t}\, {\lvertK_s (x,y)\rvert} <\infty$$
In addition, $${\label}{4.27}
(S_s(H)f)(x) = \int_{x-s}^{x+s} L_s (x,y) f(y)\, dy$$ where $${\label}{4.28}
L_s (x,x+s) =\tfrac12$$
[L4.6]{} It suffices to prove Theorem \[T4.5\] for $s$ small.
Since $\cos (2u)=2\cos^2 (u)-1$, one sees $${\label}{4.29}
C_{2s}(A) = 2C_s(A)^2 -{{\boldsymbol{1}}}$$ Thus, if $C_s\in{{\mathcal R}}_s$, one sees $C_{2s}\in{{\mathcal R}}_{2s}$ and $${\label}{4.30}
K_{2s}(x,y) = K_s (x,y+s) + K_s (x,y-s) + K_s (x+s,y) +K_s (x-s,y)$$ where $K(x,y)=0$ if ${\lvertx-y\rvert}>s$. Thus $${\label}{4.31}
K_{2s} (x,x+2s) = K_s (x,y+s) + K_s (x+s, y+2s)$$
This shows that if the formula is known for ${\lverts\rvert}\leq T$, one gets it successively for $2T, 4T, 8T, \dots$.
Using , one sees that the result for $C_s(H)$ implies and .
If $A$ is a bounded self-adjoint operator on ${{\mathcal H}}$ which is bounded from below, and $B$ is the operator on ${{\mathcal H}}\oplus{{\mathcal H}}$ given by $${\label}{4.32}
B = \left(
\begin{array}{rc}
{{\boldsymbol{0}}}& {{\boldsymbol{1}}}\\
-A & {{\boldsymbol{0}}}\end{array}\right)$$ then $${\label}{4.33}
e^{sB} = \left(
\begin{array}{rc}
C_t(A) & S_t(A) \\
-AS_t(A) & C_t(A)
\end{array}\right)$$
This formula can be checked by showing that the right side of is a bounded semigroup whose derivative at $t=0$ is $B$. DuHamel’s formula for $A,{\tilde }A$ bounded says that $$\begin{aligned}
e^{t {\tilde }B} &= e^{tB} + \int_0^t e^{sB} ({\tilde }B-B) e^{(t-s){\tilde }B}\, ds {\label}{4.34} \\
&= e^{tB} + \int_0^t e^{s{\tilde }B} ({\tilde }B-B) e^{(t-s)B}\, ds {\label}{4.35}\end{aligned}$$ Using , we obtain $$\begin{aligned}
C_t({\tilde }A) &= C_t(A) - \int_0^t S_s ({\tilde }A) ({\tilde }A-A) C_{t-s} (A)\, ds {\label}{4.36} \\
&= C_t(A) - \int_0^t S_s (A) ({\tilde }A-A) C_{t-s} ({\tilde }A)\, ds {\label}{4.37}\end{aligned}$$
By taking limits, it is easy to obtain these formulae for $A=-{\frac}{d^2}{dx^2}$, ${\tilde }A=-{\frac}{d^2}{dx^2} +V$ with $V$ bounded. By obtaining a priori bounds below depending only on certain $L^1$ norms of $V$, we get estimates for $V$ in $L^1$ and so, using the lemma, prove the theorem.
By iterating , one gets an expansion (which converges if $V$ is bounded and whose estimates then extend), $$\begin{gathered}
C_t \bigl( -\tfrac{d^2}{dx^2} + V(x) \bigr) = C_t^{(0)} +
\sum_{n=1}^\infty C_t^{(n)} {\label}{4.38} \\
C_t^{(n)} = (-1)^n \int_{0\leq s_1 + \cdots + s_n\leq t} S_{s_1}^{(0)} V
S_{s_2}^{(0)} \dots V S_{s_n}^{(0)} V C_{t-s_1- \cdots -s_n}^{(0)}\, ds_1, \dots ds_n {\label}{4.39}\end{gathered}$$
Apply the integrand in $C_t^{(n)}$ to a function $f$ and evaluate at $x$ for fixed $s_1, \dots, s_n$. Each $S_{s_j}^{(0)}V$ evaluates $V$ at points and integrals using . The integrands in $V$ are in the interval $(x-t, x+t)$, so if we take absolute values, we see this integrand is bounded by $$\biggl(\tfrac12 \int_{x-t}^{x+t} {\lvertV(y)\rvert}\, dy \biggr)^n
[\tfrac12\, f(x+t-s_1 - \cdots -s_n) + \tfrac12\, f(x-t+s_1 + \cdots + s_n)]$$
Now we can do the integral over $s_1, \dots, s_n$. For $t-s_1 -\cdots - s_n$ fixed, the new integrand is independent of $s_1, \dots, s_{n-1}$ and is bounded by $t^{n-1}$. We find $${\label}{4.40}
{\lvert(C_t^{(n)}f)(x)\rvert}\leq t^{n-1} \biggl( \tfrac12\int_{x-t}^{x+t} {\lvertV(y)\rvert}\, dy
\biggr)^n \int_{x-t}^{x+t} {\lvertf(y)\rvert}\, dy$$
Moreover, $C_t^{(n)}$ has a continuous integral kernel $K_t^{(n)}(x,y)$ supported in ${\lvertx-y\rvert}\leq t$. Since $V$ is uniformly locally $L^1$, by taking $t$ small, we can be sure $\sup_x {\frac}12 \int_{x-t}^{x+t} {\lvertV(y)\rvert}\, dy <1$, which yields uniform convergence of $K_t^{(n)}$ to a uniformly bounded kernel.
By , we get from $C_s(H)\in{{\mathcal R}}_s$, and comes from noting that $${\label}{4.41}
{\lvertL_s (x,y)-\tfrac12\rvert}\leq (s-{\lvertx-y\rvert})\, \sup_{x,y,u\leq s}\,
{\lvertK_u (x,y)\rvert}$$
Finally, using , we see that $$K_t (x,x+t) = -\tfrac12 \int_0^t L_s (x ,x+s) V(x+s)\, ds$$ proving .
To complete the proofs of Theorems \[T4.2\] and \[T4.3\] (and so Theorem \[T4.1\]), we need the transformation formulae of Delsarte, Levitan, Gel’fand, and Marchenko [@Marc]:
[T4.7]{} If $V\in L^1 ([0,R])$ for $R<\infty$, then there exist functions $K_N,K_D$ $C^1$ in $\{(y,x) : 0\leq y\leq x\leq R\}$ so that for $0\leq x\leq R$, $$\begin{aligned}
u_N(x,E) &=C_x(E) + \int_0^x K_N (y,x) C_y (E)\, dy {\label}{4.42} \\
u_D (x,E) &= S_x (E) + \int_0^x K_D (y,x) S_y (E)\, dy {\label}{4.43}\end{aligned}$$ Moreover, $${\label}{4.44}
K_D(x,x) = K_N(x,x) = \tfrac12 \int_0^x V(t)\, dt$$
1\. These formulae are in Marchenko [@Marc p. 9 and (1.2.28)]. He supposes $V$ is continuous, but his proof works if $V$ is $L^1$; indeed, see Remark 2.
2\. Defining ${\tilde }u_X (x,k) = u_X (x,k^2)$ for $X=D,N$ and $Q_X (x,y)$ as the Fourier transform of ${\tilde }u_X$ in $k$, we see becomes $${\label}{4.45}
{\frac}{\partial^2 Q}{\partial x^2} - {\frac}{\partial^2 Q}{\partial y^2} = VQ(x,y)$$ with initial conditions $$\begin{aligned}
{2}
Q_N(x=0,y)&=\delta (y) \qquad & Q_n' (x=0,y) &=0 \\
Q_D(x=0,y) &= 0 \qquad & Q'_D (x=0,y) &= \delta(y)\end{aligned}$$ Thus, Theorem \[T4.7\] is essentially Theorem \[T4.5\] with a time-dependent $V$ used.
3\. By , , and , we obtain a critical representation for $\Delta$: $${\label}{4.46}
\begin{split}
\Delta(E) = 2C_x (E) &+ \int_0^L L_1 (t) C_t (E)\, dt \\
& \qquad + \int_0^L L_2 (t) S_t(E)\, dt + K_D (L,L) S_L(E)
\end{split}$$ where $L_1,L_2$ are continuous in $[0,L]$. Indeed, $$L_1(t) = K_N (t,L) \qquad L_2 (t) = \left.{\frac}{\partial}{\partial x}\, K_D (t,x)\right|_{x=L}$$
The analyticity is immediate from as is given $${\lvertC_x (E)\rvert} + {\lvertS_x(E)\rvert} \leq C\exp \bigl( x\sqrt{{\lvertE\rvert}}\bigr)$$ Moreover, since for $t<L$, $$\lim_{E\to -\infty}\, {\frac}{C_t(E)}{C_L(E)} =0 \qquad\text{and}\qquad
\lim_{E\to -\infty}\, {\frac}{S_L(E)}{C_L(E)} =0$$ we have .
By and Theorem \[T4.5\], ${\frac}12 \Delta (-{\frac}{d^2}{dx^2}+V)$ is in ${{\mathcal R}}_L$. Moreover, the only terms contributing to $K(x,x+L)$ come from $C_L (-{\frac}{d^2}{dx^2}+V)$ and $K_D (L,L) S_L(-{\frac}{d^2}{dx^2}+V)$. By , , and , $$K(x,x+L) = -\tfrac14 \int_x^{x+L} V(y)\, dy + \tfrac12 \biggl( \tfrac12 \int_0^L
V_0(y)\, dy \biggr)$$ which, given the periodicity of $V_0$, is .
There is a second proof of . It depends on this analog of Naĭman’s lemma:
[L4.8]{} If $V$ is $L_{{\text{\rm{loc}}},{\text{\rm{unif}}}}^1$ and $-{\frac}{d^2}{dx^2}+V$ commutes with $S_L+S_{-L}$, then $${\label}{4.48a}
V(x+L)=V(x)$$
Suppose first that $V$ is bounded. Then $S_L + S_{-L}$ leaves $D(-{\frac}{d^2}{dx^2})$ invariant and commutes with it, so $S_L + S_{-L}$ commutes with $V$. If $f$ is supported in a small neighborhood of $x_0$, $(x_0-\delta,
x_0 +\delta)$ with ${\lvert\delta\rvert} < L/2$, then $(S_L+S_{-L})(Vf)$ is two separate pieces $V(x-L) f(x-L)$ supported near $x_0+L$ and $V(x+L)f(x+L)$ supported near $x_0-L$, while $V(S_L+S_{-L})f$ is two pieces $V(x) f(x-L)$ and $V(x)f(x+L)$. Since the pieces are disjoint, $$V(x) f(x-L) =V(x+L)f(x-L)$$ which implies .
For general $V$, take $g\in {{\mathcal{C}}}_0^\infty({{\mathbb{R}}})$ with $\int g(x)\, dx =1$ and note that $\int g(x) S_x (-{\frac}{d^2}{dx^2}+V) S_{-x}\, dx$ is $-{\frac}{d^2}{dx^2}
+g*V$ and it commutes with $S_L+S_{-L}$ also. But $g*V$ is bounded, so it is periodic. follows by using an approximate $\delta$-function.
Block Jacobi Matrices and Matrix Orthogonal Polynomials
=======================================================
[s5]{} What the magic formula suggests is that the Jacobi matrix $J$ has parameters that approach an isospectral torus if and only if $\Delta(J)$ approaches $S^p + S^{-p}$. $\Delta(J)$ is a matrix of width $2p+1$ (i.e., $\Delta(J)_{k\ell}=0$ if $k-\ell\notin \{0,\pm 1, \dots,
\pm p\}$) and $S^p + S^{-p}$ is a matrix with $1$’s at the extremes.
A matrix of width $2\ell+1$ has the structure of a tridiagonal matrix if rewritten in terms of $\ell\times\ell$ blocks and $S^\ell + S^{-\ell}$ corresponds to $B_k={{\boldsymbol{0}}}$, $A_k={{\boldsymbol{1}}}$, the identity matrix, so $\Delta(J)\sim S^\ell + S^{-\ell}$, at the matrix level, approaches the ‘free case.’ This will allow us to reduce our main theorems to matrix analogs of the theorems on perturbations of the free case.
Of course, the association of the block matrix to orthogonal polynomials is critical—the orthogonality will be with respect to a matrix-valued measure. There is a huge literature on MOPRL (see, e.g., [@CFMV03; @DJLP; @DS2002; @DS02; @Dur99; @DD02; @DLR96; @DLR01; @DLR04; @DLRS99; @JDP96; @MS93; @MY03; @YM2001; @YMP01]) and MOPUC (see, e.g., [@BakCon; @DG92; @DGK78; @DGK3; @DGK2; @DGK79; @DGK81; @Ger81; @Lev; @MRG89; @Rod90; @newVA; @YK]). In this section, our main purpose is to set notation and discuss the important notion of equivalent families of block Jacobi matrices, a notion discussed more explicitly in [@DPSprep].
Given a semi-infinite complex matrix $M=\{m_{ij}\}_{1\leq i,j\leq\infty}$ and $\ell=1,2,\dots$, we define the $\ell\times\ell$ block decomposition as the family of $\ell\times\ell$ matrices $\{M_{qr}\}_{1\leq q,r\leq \infty}$ by $${\label}{5.1}
(M_{qr})_{ij} = m_{\ell(q-1)+i,\, \ell(r-1)+j} \qquad i,j=1,\dots\ell$$
A block Jacobi matrix is an $M$ where $${\label}{5.2}
M_{qr} = \begin{cases}
B_q &\text{if } r=q\geq 1 \\
A_q & \text{if } r=q+1,\, q\geq 1 \\
A_{q-1}^\dag &\text{if } r=q-1, \, q\geq 2 \\
0 & {\lvertq-r\rvert}\geq 2
\end{cases}$$ with each $A_q$ invertible and each $B_q$ Hermitian. Here, following [@OPUC1 Section 2.13], we use $^\dagger$ for Hermitian adjoint; this is to avoid confusion with the Szegő dual $\Phi_n^*$ appearing in OPUC.
We will start writing ${{\mathcal J}}$ for such matrices.
In analogy with the scalar case, one may be tempted to require $${\label}{5.3}
A_q >0$$ but to include $\Delta(J)$, we do not want to do that exclusively. If holds, we say that ${{\mathcal J}}$ is of type $1$.
If instead $${\label}{5.4}
A_1 \dots A_n >0$$ for all $n$, we say ${{\mathcal J}}$ is of type $2$.
An $\ell\times\ell$ matrix, $K$, is said to be in ${{\mathcal L}}$ if it is lower triangular with strictly positive diagonal elements, that is, $${\label}{5.5}
K_{ij} = \begin{cases}
0 & \text{if } i<j \\
>0 & \text{if } i=j
\end{cases}$$ If each $A_q\in{{\mathcal L}}$, we say that ${{\mathcal J}}$ is of type $3$.
The calculations in Section \[s2\] and \[s3\] show:
[P5.1]{}
If $\Delta$ is the discriminant of a periodic Jacobi matrix, $J_0$, of period $\ell$, then for any Jacobi matrix, $J$, $\Delta(J)={{\mathcal J}}$ is a block Jacobi matrix of type 3.
If $\Delta$ is the discriminant of a periodic CMV matrix, ${{\mathcal{C}}}_0$, of even period $\ell$, then for any CMV matrix, ${{\mathcal{C}}}$, of $\Delta({{\mathcal{C}}})={{\mathcal J}}$ is a block Jacobi matrix of type 3.
We will see that distinct ${{\mathcal J}}$’s may correspond to the same measure. Indeed, in the scalar case, the $b_n$’s and ${\lverta_n\rvert}$’s are fixed by the measure, but the $\arg (a_n)$’s are arbitrary. Thus, we define
Two block Jacobi matrices, ${{\mathcal J}}$ and ${\widetilde }{{\mathcal J}}$, are called equivalent if and only if there is an $\ell\times\ell$ block diagonal unitary ${{\mathcal U}}={{\boldsymbol{1}}}\oplus U_2
\oplus U_3\oplus\cdots$ (we will use $U_1$ for ${{\boldsymbol{1}}}$) so that $${\label}{5.6}
{\tilde }{{\mathcal J}}={{\mathcal U}}{{\mathcal J}}{{\mathcal U}}^{-1}$$ This is equivalent to
[5.7]{} $$\begin{aligned}
{\widetilde }B_n &= U_n B_n U_n^{-1} {\label}{5.7a} \\
{\tilde }A_n &= U_n A_n U_{n+1}^{-1} {\label}{5.7b}\end{aligned}$$
We will be interested in equivalence classes of ${{\mathcal J}}$’s.
Each equivalence class of ${{\mathcal J}}$’s has exactly one element each of type 1, type 2, and type 3.
The Nevai class is the set of ${{\mathcal J}}$’s for which $${\label}{5.8}
B_n\to 0 \qquad A_n^\dagger A_n^{ }\to{{\boldsymbol{1}}}$$
The following is immediate from :
[P5.3]{} If some ${{\mathcal J}}$ is in the Nevai class, so are all equivalent ${{\mathcal J}}$’s.
Damanik, Pushnitski, and Simon [@DPSprep] prove that
[P5.4]{} If ${{\mathcal J}}$ is in the Nevai class and is type 1, type 2, or type 3, then $${\label}{5.9}
A_n\to {{\boldsymbol{1}}}$$
We will sometimes need the MOPRL, the matrix orthogonal polynomials. What we describe here are the left OPs. There are also right OPs (see [@DPSprep]), which we do not need here. An $\ell$-dimensional matrix-valued measure is a positive scalar measure $d{\eta}_t(x)$ and a nonnegative $\ell\times \ell$ matrix-valued function $M(x)$. The matrix-valued measure $${\label}{5.10}
d{\eta}(x) = M(x)\, d{\eta}_t(x)$$ can always be normalized by $${\label}{5.11}
{\text{\rm{Tr}}}(M(x))=\ell$$
We will always assume $d{\eta}$ is normalized, that is, $${\label}{5.12}
\int d{\eta}(x)={{\boldsymbol{1}}}$$ The proper notion of nontriviality is a little subtle; it is discussed in detail in [@DPSprep]. For our purpose here, it is sufficient that ${\langle\!\langle}\cdot,\cdot{\rangle\!\rangle}_{\text{\rm{L}}}$ defined below is nondegenerate on polynomials.
If $f,g$ are two $\ell$-dimensional matrix-valued functions, we define $${\langle\!\langle}f,g{\rangle\!\rangle}_{\text{\rm{L}}}=\int g(x) M(x) f(x)^\dagger\, d{\eta}_t(x)$$ Note that this ‘inner product’ returns matrix values. Recall also that ${}^\dagger$ denotes the Hermitian conjugate of a matrix. The subscript ‘$L$’ is for ‘left’ and reflects the fact that if $C$ is an $\ell\times\ell$ matrix, then $$\begin{aligned}
{\langle\!\langle}f,Cg{\rangle\!\rangle}_{\text{\rm{L}}}&= C{\langle\!\langle}f,g{\rangle\!\rangle}_{\text{\rm{L}}}{\label}{5.13} \\
{\langle\!\langle}Cf,g{\rangle\!\rangle}_{\text{\rm{L}}}&= {\langle\!\langle}f,g{\rangle\!\rangle}_L C^\dagger {\label}{5.14}\end{aligned}$$ We will normally just write ${\langle\!\langle}\cdot,\cdot{\rangle\!\rangle}$ from now on.
Left orthonormal polynomials are of the form $$p_n(x) = \kappa_n x^n + \text{lower order}$$ with matrix coefficients, defined by $${\label}{5.15}
{\langle\!\langle}p_n,p_m{\rangle\!\rangle}=\delta_{nm} {{\boldsymbol{1}}}$$ So long as $d{\eta}$ is nontrivial, the $p_n$ exist. They are not unique since if $\{U_n\}_{n=1}^\infty$ are unitary $\ell\times\ell$ matrices, $${\label}{5.16}
{\tilde }p_n(x) = U_{n+1} p_n(x)$$ are also MOPRL. We demand $\kappa_0={{\boldsymbol{1}}}$, that is, $p_0(x)={{\boldsymbol{1}}}$, and so $U_1 ={{\boldsymbol{1}}}$.
$\{p_j\}_{j=0}^n$ are a left module basis for matrix polynomials of degree $n$, that is, if $f$ is any polynomial of degree $n$, then there are unique $\ell\times\ell$ matrices $f_0, \ldots, f_n$ so that $$f(x)=\sum_{j=0}^n f_j p_j(x)$$ Indeed, $${\label}{5.16a}
f_j ={\langle\!\langle}p_j,f{\rangle\!\rangle}$$
For $n=1,2,\ldots$, define $${\label}{5.17}
B_n={\langle\!\langle}p_{n-1}, xp_{n-1}{\rangle\!\rangle}\qquad A_n = {\langle\!\langle}p_n, xp_{n-1}{\rangle\!\rangle}$$ Then, since $xp_j =\sum_{\ell=0}^{j+1} C_\ell p_\ell$ implies ${\langle\!\langle}p_j, xp_n{\rangle\!\rangle}={\langle\!\langle}xp_j, p_n{\rangle\!\rangle}=0$ if $j\leq n-2$, we have $${\label}{5.18}
xp_n(x) =A_{n+1} p_{n+1}(x) + B_{n+1} p_n(x) + A_n^\dagger p_{n-1}(x)$$
This implies $A_{n+1} \kappa_{n+1} = \kappa_n$ so $${\label}{5.19}
\kappa_n = (A_1 \dots A_n)^{-1}$$ and the type $2$ condition is equivalent to $\kappa_n >0$.
Looking at , we see that holds for ${\tilde }p_n, p_n$ if and only if ${\tilde }A_n, {\widetilde }B_n$ are related to $A_n,B_n$ by . Jacobi matrix equivalence is just a ‘change of phase’ in the MOPRL.
Given a block Jacobi matrix, we can view it as acting on the Hilbert space $\ell^2
(\{1,2,\dots\},{{\mathbb{C}}}^\ell)$ with inner product $${\label}{5.20}
\langle f,g\rangle = \sum_{n=1}^\infty \langle f_n, g_n\rangle_{{{\mathbb{C}}}^\ell}$$ If $\{e_j\}_{j=1}^\ell$ is the standard basis of ${{\mathbb{C}}}^\ell$, then $\{\delta_{k;j}\}_{k=1}^\infty {}_{j=1}^{\ell}$, defined by $${\label}{5.21}
(\delta_{k;j})_n = \delta_{kn} e_j$$ is a basis. ${{\mathcal J}}$ acts on $\ell^2(\{1,2,\dots\},{{\mathbb{C}}}^\ell)$ via $${\label}{5.22}
({{\mathcal J}}f)_n = A_{n-1}^\dagger f_{n-1} + B_n f_n + A_n f_{n+1}$$ (with $A_0=0$).
The spectral measure for ${{\mathcal J}}$ is the $\ell\times\ell$ matrix-valued measure with $${\label}{5.23}
\langle \delta_{0;j}, f({{\mathcal J}}) \delta_{0;k}\rangle =
\int f(x)\, d{\eta}_{jk}(x)$$ for any scalar-valued function $f$. It is easy to see (e.g., [@DPSprep]) that this map from ${{\mathcal J}}$ to ${\eta}$ inverts the one given by forming the MOPRL and defining ${{\mathcal J}}$ by . Moreover, ${{\mathcal J}}$ and ${\widetilde }{{\mathcal J}}$ are equivalent if and only if $d{\tilde }{\eta}\equiv d{\eta}$.
The $m$-function is defined by $$\begin{aligned}
m(E) &= \int {\frac}{1}{x-E}\, d{\eta}(x) {\label}{5.24} \\
&= \langle \delta_{0;\,{{\boldsymbol{\cdot}}}}\, , ({{\mathcal J}}-E)^{-1} \delta_{0;\, {{\boldsymbol{\cdot}}}}\,\rangle {\label}{5.25}\end{aligned}$$ It is an $\ell\times\ell$ matrix-valued Herglotz function: $${\label}{5.26}
\operatorname{Im}E>0\Rightarrow \operatorname{Im}m(E)>0$$ that is, $\tfrac{1}{2i}(m-m^\dagger)$ is positive definite in the upper half-plane. For information on matrix Herglotz functions, see [@AG94; @AG95; @BT92a; @BT92b; @Fu76; @Ger77; @GT2000; @GKS98; @HS81; @Jo87; @Kr89; @KO78; @Sak92; @Wei87]. Obviously, by , $m$ is constant over equivalence classes.
As in the scalar case (see [@OPUC1 Section 1.2]), one has that for a.e. $x\in{{\mathbb{R}}}$, $\lim_{{\varepsilon}\downarrow 0} m(x+i{\varepsilon})\equiv m(x+i0)$ exists and $${\label}{5.27a}
d{\eta}_{\text{\rm{ac}}}(x) = \pi^{-1} \operatorname{Im}m (x+i0)\, dx$$ Here $${\label}{5.27b}
d{\eta}_{\text{\rm{ac}}}(x) = M(x) \, d{\eta}_{t;{\text{\rm{ac}}}}(x)$$ where $d{\eta}_{t;{\text{\rm{ac}}}}$ is the a.c. part of $d{\eta}_t$. Alternatively, $d{\eta}_{\text{\rm{ac}}}$ is the unique matrix-valued measure which is a.c. (i.e., ${\eta}_{\text{\rm{ac}}}(I) =0$ for any set with ${\lvertI\rvert}=0$) and where there is a set $K$ with ${\lvertK\rvert}=0$ so $({\eta}-{\eta}_{\text{\rm{ac}}}) ({{\mathbb{R}}}\setminus K)=0$.
Given a block Jacobi matrix, ${{\mathcal J}}$, by ${{\mathcal J}}^{(n)}$ we mean the matrix with the top $n$ (block matrix) rows and leftmost $n$ columns removed, that is, $${\label}{5.27x}
B_k^{(n)} = B_{k+n}^{} \qquad A_k^{(n)} = A_{k+n}^{ }$$ We write $m^{(n)}(z)$ for the $m$-function associated to ${{\mathcal J}}^{(n)}$. Equivalent ${{\mathcal J}}$’s do not have the same $m^{(n)}$ for $n\geq 1$ (although they are unitarily related). We see $m^{(0)}\equiv m$.
We will need the following result of Aptekarev–Nikishin [@AN84] (see also [@DPSprep]), a matrix analog of the well-known Jacobi–Stieltjes recursion for OPRL:
[T5.5]{} We have that $${\label}{5.27}
m^{(n)} (E)^{-1} = E -B_{n+1} -A_{n+1} m^{(n+1)} (E) A_{n+1}^\dagger$$ for $n=0,1,2,\dots$.
Next, we need to note the following analog of a well-known scalar result (see, e.g., [@Sim_Sturm]) proven in [@DPSprep]:
[T5.5A]{} Let ${{\mathcal J}}$ be a block Jacobi matrix with $\sigma_{\text{\rm{ess}}}({{\mathcal J}})
\subset [a,b]$. Then, for any ${\varepsilon}$, there is a $K$ so that for $k\geq K$, $${\label}{5.32}
\sigma({{\mathcal J}}^{(k)})\subset [a-{\varepsilon}, b+{\varepsilon}]$$
Finally, we need to look at poles and zeros of $\det (m(z))$. In the scalar $(\ell=1)$ case, poles occur precisely at eigenvalues of ${{\mathcal J}}$ and zeros at eigenvalues of ${{\mathcal J}}^{(1)}$, the once stripped Jacobi matrix. In that scalar case, these eigenvalues are distinct.
In the matrix case, ${{\mathcal J}}$ and ${{\mathcal J}}^{(1)}$ can have eigenvalues in common (as can be easily arranged by taking a direct sum of suitable scalar $J$’s) so there can be cancellations. We say a scalar meromorphic function, $f(z)$, has a zero/pole of order $k\in{{\mathbb{Z}}}$ at $z_0\in{{\mathbb{C}}}$ if $(z-z_0)^{-k} f(z)$ has a finite nonzero limit as $z\to z_0$. We will need the following result from [@DPSprep]:
[T5.6]{} Let $x_0\in{{\mathbb{R}}}$. Let $q_0$ be the multiplicity of $x_0$ as an eigenvalue of ${{\mathcal J}}$, and $q_1$ its multiplicity as an eigenvalue of ${{\mathcal J}}_1$. Then
$q_0 + q_1 \leq \ell$
$\det(m(z))$ has a zero/pole of order $q_1-q_0$.
We will also need the following result from Aptekarev–Nikishin [@AN84]:
[T5.8]{} Let ${{\mathcal J}}$ be a block Jacobi matrix with $\sigma_{\text{\rm{ess}}}({{\mathcal J}})=
[-2,2]$, $\sigma({{\mathcal J}})\setminus\sigma_{\text{\rm{ess}}}({{\mathcal J}})$ a finite set and with $g(x)=d{\eta}_{\text{\rm{ac}}}(x)/dx$ we have $${\label}{5.33}
\int (4-x^2)^{-1/2} \log (\det(g(x))\, dx > -\infty$$ Suppose ${{\mathcal J}}$ is type 2. Then $$\lim_{n\to\infty}\, A_1 \dots A_n$$ exists and is a strictly positive matrix.
A Denisov–Rakhmanov Theorem for MOPRL
=====================================
[s6]{} As preparation for proving Theorem \[T1.2\] in Section \[s7\], in this section we will prove
[T6.1]{} Let $d{\eta}$ be a nondegenerate $\ell\times\ell$ matrix-valued measure on ${{\mathbb{R}}}$ with associated block Jacobi matrix ${{\mathcal J}}$ of type 3 so that $$\begin{aligned}
{2}
&\text{\rm{(i)}} \qquad && \sigma_{\text{\rm{ess}}}({{\mathcal J}}) = [-2,2] {\label}{6.1} \\
&\text{\rm{(ii)}} \qquad && d{\eta}= f(x)\, dx + d{\eta}_{\text{\rm{s}}}{\label}{6.2}\end{aligned}$$ with $d{\eta}_{\text{\rm{s}}}$ singular and $${\label}{6.3}
\det(f(x)) >0$$ a.e. on $[-2,2]$. Then $${\label}{6.4}
B_n\to {{\boldsymbol{0}}}\qquad A_n\to {{\boldsymbol{1}}}$$
says the a.c. spectrum has multiplicity $\ell$.
If is replaced by the stronger $\sigma(J)=[-2,2]$ and type $3$ by type $2$, this is a theorem of Yakhlef–Marcellán [@YM2001]. We will prove Theorem \[T6.1\] by modifying their proof.
The shift from type $2$ to $3$ is easy on account of Proposition \[P5.4\]. By applying the argument of [@YM2001], we get ${\tilde }A_n\to{{\boldsymbol{1}}}$ for the equivalent ${\tilde }{{\mathcal J}}$ of type $2$, conclude the whole equivalence class is in the Nevai class, and see $A_n\to{{\boldsymbol{1}}}$. So we will only worry about the changes needed to go from $\sigma(J) =[-2,2]$ to $\sigma_{\text{\rm{ess}}}(J)=
[-2,2]$, where we follow Denisov’s approach for the scalar case [@DenPAMS].
[@YM2001] relies on a matrix version of Rakhmanov’s theorem proven by van Assche [@newVA]. We need to extend it slightly to allow a.c. spectrum on a large subset of $\partial{{\mathbb{D}}}$ rather than all of $\partial{{\mathbb{D}}}$:
[T6.2]{} Let $d\mu$ be an $\ell\times\ell$ matrix-valued measure on $\partial{{\mathbb{D}}}$ and let $\{\alpha_n\}_{n=0}^\infty$ denote its matrix Verblunsky coefficients . Suppose $${\label}{6.5}
d\mu = w(\theta)\, {\frac}{d\theta}{2w} + d\mu_{\text{\rm{s}}}$$ where $d\mu_{\text{\rm{s}}}$ is singular, and let $${\label}{6.6}
\Omega =\{\theta : \det (w(\theta)) >0\}$$ Then $${\label}{6.7x}
\limsup_{n\to\infty}\, \|\alpha_n\| \leq 2 \sqrt{2\ell} \, \biggl( 1 -
\biggl( {\frac}{{\lvert\Omega\rvert}}{2\pi}\biggr)^3\biggr)^{1/2}$$
1\. For notation on MOPUC, see [@DPSprep].
2\. Where we use $\{\alpha_n\}_{n=0}^\infty$, van Assche [@newVA] uses $\{H_n\}_{n=1}^\infty$ related to $\alpha_n$ by $${\label}{6.7}
H_n = -\alpha_{n-1}^\dagger$$
3\. We follow notation from [@newVA] and the variant of the scalar proof as in [@OPUC2 Section 9.1] where $a_n, b_n, c_n, d_n$ below all appear.
We define $$\begin{aligned}
a_n & = \| \alpha_n \| \\
b_{n,q} & = \frac{1}{2\pi} \int_0^{2\pi} \| [ \varphi_n^{\text{\rm{L}}}(e^{i\theta})
\varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} ] \, [ \varphi_n^{\text{\rm{L}}}(e^{i\theta})
\varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} ]^\dagger - I \| \, d\theta \\
c_{n,q} & = \frac{1}{2 \pi \ell} \int_0^{2\pi} {\text{\rm{Tr}}}( \varphi_n^{\text{\rm{L}}}(e^{i\theta})
\varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} [ \varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^\dagger ]^{-1}
\varphi_n^{\text{\rm{L}}}(e^{i\theta})^\dagger )^{1/2} \, d\theta \\
d_n & = \frac{1}{2 \pi \ell} \int_0^{2\pi} {\text{\rm{Tr}}}( [ \varphi_n^{\text{\rm{L}}}(e^{i\theta}) w(\theta) \varphi_n^{\text{\rm{L}}}(e^{i\theta})^\dagger ]^{1/2} ) \, d\theta\end{aligned}$$
[P6.3]{} For every $n \geq 0$, we have that $$\begin{aligned}
{2}
a_n & \leq b_{n,q} && \qquad \text{ for every } q \geq 1 \label{a} \\
b_{n,q}^2 & \leq 8 \ell^2 (1 - c_{n,q}) && \qquad \text{ for every } q \geq 1 \label{b} \\
d_n^2 & \leq \inf_{q \geq 1} c_{n,q} \label{c}\end{aligned}$$ Moreover, we have that $$\label{d}
\biggl( \frac{{\lvert\Omega\rvert}}{2\pi} \biggr)^{3/2} \leq \liminf_{n \to \infty} \, d_n$$ Consequently, $$\label{e}
\limsup_{n \to \infty} \, a_n \leq 2 \sqrt{2}\, \ell
\biggl( 1 - \biggl( \frac{{\lvert\Omega\rvert}}{2\pi} \biggr)^3 \biggr)^{1/2}$$
The second to last displayed formula on [@newVA p. 7] is . The estimates on the bottom half of [@newVA p. 12] show that $$\begin{aligned}
b_{n,q}^2 & = \frac{1}{4 \pi^2} \biggl( \int_0^{2\pi} \| [
\varphi_n^{\text{\rm{L}}}(e^{i\theta}) \varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} ] \,
[ \varphi_n^{\text{\rm{L}}}(e^{i\theta})
\varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} ]^\dagger - I \| \, d\theta \biggr)^2 \\
& \leq \frac{2 \ell}{\pi} \int_0^{2\pi} \| ( [ \varphi_n^{\text{\rm{L}}}(e^{i\theta})
\varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} ]\, [ \varphi_n^{\text{\rm{L}}}(e^{i\theta})
\varphi_{n+q}^{\text{\rm{L}}}(e^{i\theta})^{-1} ]^\dagger )^{1/2} - I\|^2 \, d\theta \\
& \leq \frac{2\ell}{\pi}\, ( 4 \pi \ell - 4 \pi \ell \, c_{n,q} ) \\
& = 8 \ell^2 (1 - c_{n,q})\end{aligned}$$ which is . The third displayed formula on [@newVA p. 14] is .
Now, mimicking the estimates on the bottom half of [@newVA p. 14], $$\begin{split}
\int_\Omega {\text{\rm{Tr}}}&( [ f(\theta) w(\theta) f(\theta)^\dagger ]^{1/4}
) \, d\theta \\
& \leq \biggl( 2 \pi \ell \int_\Omega {\text{\rm{Tr}}}( f(\theta)
\varphi_n^{\text{\rm{L}}}(e^{i\theta})^{-1} ( \varphi_n^{\text{\rm{L}}}(e^{i\theta})^\dagger )^{-1} f(\theta)^\dagger )\,
\frac{d\theta}{2\pi} \biggr)^{1/4} ( 2 \pi \ell \, d_n )^{1/2}
\end{split}$$ Taking $n \to \infty$, we see that $$\begin{split}
\int_\Omega {\text{\rm{Tr}}}& ([ f(\theta) w(\theta) f(\theta)^\dagger ]^{1/4})
\, d\theta \\
& \leq \biggl( 2 \pi \ell \, {\text{\rm{Tr}}}\int_\Omega f(\theta) d\mu(\theta)
f(\theta)^\dagger )^{1/4} \bigl( 2 \pi \ell \, \liminf_{n \to \infty}\, d_n \bigr)^{1/2}
\end{split}$$ Removing the singular part as in [@newVA], we obtain $$\begin{split}
\int_\Omega {\text{\rm{Tr}}}&( [ f(\theta) w(\theta) f(\theta)^\dagger ]^{1/4}) \,
d\theta \\
& \leq \biggl( 2 \pi \ell \, {\text{\rm{Tr}}}\int_\Omega f(\theta) w(\theta)
f(\theta)^\dagger d\theta \biggr)^{1/4} \bigl( 2 \pi \ell \, \liminf_{n \to \infty}\,
d_n \bigr)^{1/2}
\end{split}$$ Proceeding as in [@newVA pp. 15–16], it then follows that $${\lvert\Omega\rvert} \ell \leq ( 2 \pi \ell \,{\lvert\Omega\rvert} \ell )^{1/4} \bigl( 2 \pi \ell \, \liminf_{n
\to \infty}\, d_n\bigr)^{1/2}$$ which implies .
Putting these estimates together, $$a_n \leq b_{n,1} \leq 2 \sqrt{2} \, \ell (1 - c_{n,1})^{1/2} \leq 2 \sqrt{2}\, \ell (1 - d_n^2)^{1/2}$$ and hence $$\limsup_{n \to \infty}\, a_n \leq 2 \sqrt{2}\, \ell
\biggl( 1 - \biggl( \frac{{\lvert\Omega\rvert}}{2\pi}\biggr)^3 \biggr)^{1/2}$$ which is .
In particular, for $2\pi - {\lvert\Omega\rvert}$ small, $$\limsup_{n \to \infty}\, a_n = O \biggl( \biggl( 1 - \biggl( {\frac}{{\lvert\Omega\rvert}}{2\pi} \biggr)^3
\biggr)^{1/2} \biggr) = O (( 2\pi - {\lvert\Omega\rvert} )^{1/2})$$ as in the scalar case.
We have thus proven Theorem \[T6.2\]. To get Theorem \[T6.1\], we follow [@YM2001] using the analog of Denisov’s arguments for the case $\ell=1$.
By Proposition \[P5.4\], we need only prove for the type $2$ choice, for any ${\varepsilon}>0$, we have $${\label}{6.8}
\limsup \, (\|{\tilde }A_n-1\| + \|{\tilde }B_n\|) \leq {\varepsilon}$$ By the Szegő mapping and Geronimus connection formulae in [@YM2001], this holds by Theorem \[T6.2\] so long as for any ${\varepsilon}' >0$, we can find $k$ so $\sigma ({{\mathcal J}}^{(k)})
\subset [-2-{\varepsilon}', 2+{\varepsilon}']$, and this is true by Theorem \[T5.5A\].
A Denisov–Rakhmanov Theorem for Periodic OPRL
=============================================
[s7]{} Our main goal in this section is to prove Theorem \[T1.2\]. We will also prove the ‘hard’ half of Theorem \[T1.1\]. The simplicity of the proof shows the magic in the magic formula!
By a right limit of $J$, we mean a two-sided Jacobi matrix, $J_r$, (but with some $a$’s allowed to vanish) so that for some subsequence $n_j\to\infty$ and any $k\in{{\mathbb{Z}}}$, $${\label}{7.1}
a_{n_j+k}\to (a_r)_k \qquad b_{n_j+k}\to (b_r)_k$$ By our standing convention, Jacobi parameters are uniformly bounded, so by compactness, if $d_m ((a,b),{{\mathcal T}}_{J_0})\nrightarrow 0$, there exists a right limit $J_r\notin{{\mathcal T}}_{J_0}$. Thus, it suffices to show that any right limit $J_r$ has $J_r\in{{\mathcal T}}_{J_0}$.
By the hypotheses of Theorem \[T1.2\], the spectral mapping theorem, and the fact that $\Delta$ maps $\sigma_{\text{\rm{ess}}}(J_0)$ to $[-2,2]$ with a $p$-fold cover on $(-2,2)$, we see that $$\Delta(J)_{\text{\rm{ess}}}= [-2,2]$$ and $\Delta (J)$ has a.c. spectrum of multiplicity $p$. So thinking of ${{\mathcal J}}\equiv\Delta(J)$ as a block Jacobi matrix, ${{\mathcal J}}$ is of type $3$ and the hypotheses of Theorem \[T6.1\] apply. It follows that $A_n\to {{\boldsymbol{1}}}$, $B_n\to {{\boldsymbol{0}}}$. This means that $$\Delta (J_r) =S^p + S^{-p}$$ so by the magic formula (Theorem \[T2.1\]), $J_r\in{{\mathcal T}}_{J_0}$.
Rakhmanov’s theorem is often related to issues of $\operatorname*{w-lim}{\lvertp_n\rvert}^2\, d\mu$ and to the density of zeros. We note that there are also results of that genre here:
[T7.1A]{} If $J_0$ is a periodic Jacobi matrix of period $p$ and $J$ is a Jacobi matrix with bounded Jacobi parameters whose right limits all lie in ${{\mathcal T}}_{J_0}$ [[(]{}]{}in particular, if the hypotheses of Theorem \[T1.2\] hold[[)]{}]{}, then [[(]{}]{}with $d\mu$ the measure for $J$[[)]{}]{}
$${\label}{7.2}
\operatorname*{w-lim}_{n\to\infty}\, {\frac}{1}{p} \sum_{j=1}^p\, {\lvertp_{j+n}(x)\rvert}^2\, d\mu(x)=d\nu$$
the density of zeros for $J_0$.
The density of zeros of $p_n(x)$ converges to $d\nu$.
If $J_1\in{{\mathcal T}}_{J_0}$, then the spectral measure $d\mu_k^{(J_1)}$ associated to $\delta_k$ has period $p$ in $k$ since $J_1$ is periodic. Thus $\lim_{N\to\infty}{\frac}{1}{2N+1}
\sum_{{\lvertj\rvert}\leq N} d\mu_j^{(J_1)}={\frac}{1}{p}\sum_{j=1}^p d\mu_j^{(J_1)}$, but the limit is $d\nu$ by the discussion in Subsection \[ssTF\]. Since $${\label}{7.3}
\int x^\ell {\lvertp_j^{(J)}(x)\rvert}^2\, d\mu(x) = \langle \delta_j, J^\ell \delta_j\rangle$$ and $J_{j,j}, J_{j,j\pm 1}$ is very close to some $(J_1)_{j,j}$, $(J_1)_{j,j+1}$ for ${\lvertj-j_0\rvert}\leq M$ for fixed $M$ and $j_0\to\infty$, we see that moments of LHS of are close to moments of $d\nu$. This proves (a).
If $J^{(n)}$ denotes the top left $n\times n$ submatrix of $J$, then $$\int x^\ell\, d\nu_n ={\frac}{1}{n}\, {\text{\rm{Tr}}}((J^{(n)})^\ell)$$ so $$\lim_{n\to\infty}\, \int x^\ell\, d\nu_n = \lim_{n\to\infty}\, \int x^\ell
\biggl[ \, {\frac}{1}{n}\sum_{j=0}^{n-1} p_j(x)^2\, d\mu(x)\biggr]$$ and thus (a) implies (b).
We also have
[T7.1]{} If $d_m ((a,b),{{\mathcal T}}_{J_0})\to 0$, then $$\sigma_{\text{\rm{ess}}}(J) \subset \sigma_{\text{\rm{ess}}}(J_0)$$
By the magic formula, compactness, and the fact that every right limit of $J$ is in ${{\mathcal T}}_{J_0}$, we see that every right limit of $\Delta(J)$ is $S^p + S^{-p}$, that is, $A_n\to{{\boldsymbol{1}}}$, $B_n\to {{\boldsymbol{0}}}$. Thus, by Weyl’s theorem, $\sigma_{\text{\rm{ess}}}(\Delta(J))=[-2,2]$. Since $$\sigma_{\text{\rm{ess}}}(\Delta(J)) = \Delta (\sigma_{\text{\rm{ess}}}(J))$$ we see $\sigma_{\text{\rm{ess}}}(J)\subset \Delta^{-1} ([-2,2])=\sigma_{\text{\rm{ess}}}(J_0)$.
1\. Since $\Delta$ is $p$ to $1$, we cannot conclude that $\sigma_{\text{\rm{ess}}}(J_0)
=\sigma_{\text{\rm{ess}}}(J)$ from $\Delta (\sigma_{\text{\rm{ess}}}(J_0))=\Delta (\sigma_{\text{\rm{ess}}}(J))$.
2\. That $\sigma_{\text{\rm{ess}}}(J_0)\subset\sigma_{\text{\rm{ess}}}(J)$ is a simple trial function argument given that $J$ must have some right limits; see [@LastS; @LS_jdam].
Denisov–Rakhmanov Sets
======================
[new-s9]{} In this section, we want to show how one can take suitable limits of Theorem \[T1.2\] to get a ‘cheap’ proof of similar theorems in other nonperiodic cases. We will also present an insight into the proper general form of Denisov–Rakhmanov-type theorems.
The right limits we have discussed so far involve the weak product topology on the Jacobi parameters, so we will emphasize this fact by using the phrase ‘weak right limits’ in this section. We are also interested in limits in the $\ell^\infty$-topology for two-sided sequences, that is, $\{c_n^{(k)}\}_{n=-\infty}^\infty \to \{c_n^{(\infty)}\}_{n=-\infty}^\infty$ in this topology if and only if, as $k\to\infty$, $$\sup_n\, {\lvertc_n^{(k)}-c_n^{(\infty)}\rvert}\to 0$$
In terms of weak limits, we note the following:
[Pn9.1]{} Let ${{\mathcal{E}}}$ be a closed set. Let $J$ be a half-line Jacobi matrix with $${\label}{n9.1}
\Sigma_{\text{\rm{ac}}}(J)=\sigma_{\text{\rm{ess}}}(J)={{\mathcal{E}}}$$ Let $J_r$ be a weak right limit of $J$. Then $${\label}{n9.2}
\Sigma_{\text{\rm{ac}}}(J_r) =\sigma(J_r) ={{\mathcal{E}}}$$
Note that has $\sigma(J_r)$, not merely $\sigma_{\text{\rm{ess}}}(J_r)$.
By results in [@LastS], $$\sigma (J_r)\subset {{\mathcal{E}}}\subset\Sigma_{\text{\rm{ac}}}(J_r)$$ Since $\Sigma_{\text{\rm{ac}}}(J_r) \subset \sigma (J_r)$ trivially, holds.
Recall that a sequence $\{c_n\}_{n=-\infty}^\infty$ is called uniformly almost periodic (in the general theory of almost periodic functions, this defines ‘almost periodic’—we add ‘uniformly’ because the term is sometimes used in a weaker sense in the spectral theory literature) if and only if $\{c^{(\ell)}\}_{\ell=-\infty}^\infty$ given by $(c^{(\ell)})_n=
c_{n+\ell}$ has compact closure in the $\ell^\infty$-topology.
A set ${{\mathcal{E}}}$ is called [*essentially perfect*]{} if and only if ${{\mathcal{E}}}$ is closed, and for all $E\in {{\mathcal{E}}}$ and $\delta >0$, ${\lvert(E-\delta, E+\delta) \cap {{\mathcal{E}}}\rvert}>0$.
Essentially perfect sets are precisely the sets, ${{\mathcal{E}}}$, for which there is a purely a.c. measure $d{\eta}$ with ${\text{\rm{supp}}}(d{\eta})={{\mathcal{E}}}$.
A set ${{\mathcal{E}}}$ is said to be a [*Denisov–Rakhmanov set*]{} if and only if
${{\mathcal{E}}}$ is essentially perfect and bounded.
There is a set ${{\mathcal T}}_{{{\mathcal{E}}}}$ compact in the uniform topology so that for any bounded Jacobi matrix, $J$, for which holds, the set of right limits of $J$ lies in ${{\mathcal T}}_{{{\mathcal{E}}}}$.
The definition says nothing explicit about ${{\mathcal T}}_{{{\mathcal{E}}}}$ being a torus, but by Proposition \[Pn9.1\], if $J_r\in{{\mathcal T}}_{{{\mathcal{E}}}}$, then holds, and since ${{\mathcal T}}_{{{\mathcal{E}}}}$ is closed under translations, each $J_r$ in ${{\mathcal T}}_{{{\mathcal{E}}}}$ is almost periodic. By Kotani theory (see [@Kot; @Kot97; @S168; @OPUC2]), $\langle\delta_n, (J_r-E-i{\varepsilon})^{-1}\delta_n\rangle$ has real boundary values for a.e. $E$. In many cases and, in particular, if ${{\mathcal{E}}}$ is a finite union of closed intervals, Sodin–Yuditskii [@SY] (see also [@AntK; @BGHT]) proved there is a natural torus so that any almost periodic $J_r$ with real boundary values lies in this torus. Thus for such cases, that ${{\mathcal{E}}}$ is a Denisov–Rakhmanov set can be connected to approach to an isospectral torus. In particular, our Theorem \[T1.2\] implies the statement that $\sigma_{\text{\rm{ess}}}(J_0)$ is a Denisov–Rakhmanov set.
Given an essentially perfect set, ${{\mathcal{E}}}$, we define ${{\mathcal{D}}}({{\mathcal{E}}})$ to be the set of Jacobi matrices obeying .
The following two simple results will be the basis of our approximation theorems:
[Pn9.2]{} Let ${{\mathcal{E}}}$ be an essentially perfect set. Suppose there are uniformly compact sets $\{{{\mathcal T}}^{(n)}\}_{n=1}^\infty$ and ${{\mathcal T}}^{(\infty)}$ of two-sided Jacobi matrices so that
If $J_n\in{{\mathcal T}}^{(n)}$ and $J_n\to J_\infty$ weakly, then $J_\infty
\in{{\mathcal T}}^{(\infty)}$.
For any weak right limit point $J_r$ of some $J\in{{\mathcal{D}}}({{\mathcal{E}}})$, there is ${\tilde }J\in{{\mathcal T}}^{(n)}$ so $${\label}{n9.3}
\sup_{{\lvertj\rvert}\leq n}\, {\lverta_j^{(r)} -{\tilde }a_j\rvert} + {\lvertb_j^{(r)}-{\tilde }b_j\rvert} \leq {\frac}{1}{n}$$
Then ${{\mathcal{E}}}$ is a Denisov–Rakhmanov set.
Let $J_n$ be the ${\tilde }J$ guaranteed by . Then clearly, $J_n$ converges weakly to $J_r$ so, by (1), $J_r\in{{\mathcal T}}^{(\infty)}$. Since ${{\mathcal T}}^{(\infty)}$ is uniformly compact, ${{\mathcal{E}}}$ is a Denisov–Rakhmanov set.
[Pn9.3]{} Let $J_0$ be a fixed periodic Jacobi matrix with essential spectrum ${{\mathcal{E}}}_0$. Then for any $n$, there is a $\delta >0$ so that for any set ${{\mathcal{E}}}$ with $$\begin{aligned}
{2}
&\text{\rm{(a)}} \qquad && {{\mathcal{E}}}\subset \{E : {\text{\rm{dist}}}(E,{{\mathcal{E}}}_0)<\delta\} {\label}{n9.4} \\
&\text{\rm{(b)}} \qquad && {\lvert{{\mathcal{E}}}\rvert} > (1-\delta) {\lvert{{\mathcal{E}}}_0\rvert} {\label}{n9.5}\end{aligned}$$ and any $J\in{{\mathcal{D}}}({{\mathcal{E}}})$, we have that any right limit, $J_r$, obeys for some ${\tilde }J\in{{\mathcal T}}_{J_0}$.
Moreover, if $p$ is fixed and $C$ is a compact subset of $[(0,\infty)\times
{{\mathbb{R}}}]^p$, then $\delta$ can be picked to work for all $J_0 =\{(a_n,b_n)\}_{n=1}^p\in C$.
The uniformity claimed in the last statement comes from noting that choices can be made uniformly in the proof below.
Let $p$ be the period of $J_0$. We first claim that given $\delta_1$, we can find $\delta$ so if ${{\mathcal{E}}}$ obeys –, then $$\begin{gathered}
{\text{\rm{dist}}}(\Delta({{\mathcal{E}}}), [-2,2]) < \delta_1 {\label}{n9.6} \\
{\lvert\{x\in (-2,2) : \text{all $p$ solutions of $\Delta(E)=x$ lie in }{{\mathcal{E}}}\}\rvert} >
4-\delta_1 {\label}{n9.7x}\end{gathered}$$ This is immediate from the continuity of $\Delta$ and its derivatives.
Next, we note that given ${\varepsilon}_1$, we can find $\delta_1$ so that if ${{\mathcal J}}$ is a $p\times p$ block Jacobi matrix so that $$\begin{gathered}
\sigma_{\text{\rm{ess}}}({{\mathcal J}})\subset [-2-\delta_1, 2+\delta_1] \\
{\lvert\{E\in [-2,2] : {{\mathcal J}}\text{ has a.c. spectrum at $E$ of multiplicity $p$}\}\rvert}
>4-\delta_1\end{gathered}$$ then $$\limsup_{k,m\to\infty}\, {\lvert{{\mathcal J}}_{km} - (S^p + S^{-p})_{km}\rvert} <{\varepsilon}_1$$ The proof of this is identical to the proof of the matrix Denisov–Rakhmanov theorem.
Combining these steps, we are reduced to showing for any $n$ and ${\varepsilon}$, there is ${\varepsilon}_1$ so for all two-sided $J_r$ with ${\text{\rm{dist}}}(\sigma(J_r), {{\mathcal{E}}}) <{\varepsilon}$, we have that $${\label}{n9.7}
\sup_{k,m}\, {\lvert\Delta(J_r)-(S^p+S^{-p})_{km}\rvert} <{\varepsilon}_1$$ implies there is a ${\tilde }J\in{{\mathcal T}}_{J_0}$ so that holds. To do this, we first follow the proof of Theorem \[T2.1\] to note that for $n$, ${\varepsilon}_2$, and ${\varepsilon}_3$ fixed, we can find ${\varepsilon}_1$ so implies there is a $p$-periodic $J^\sharp$ such that $${\label}{n9.8}
\|\Delta (J^\sharp)-\Delta(J)\| <{\varepsilon}_2$$ and $$\sup_{{\lvertj\rvert}\leq n}\, {\lverta_j^{(r)}-a_j^\sharp\rvert} +{\lvertb_j^{(r)}-{\tilde }b_j\rvert}\leq{\varepsilon}_3$$
Finally, a compactness argument shows that for any $n$, we can find ${\varepsilon}_4$ so for any periodic $J^\sharp$ with $$\|\Delta(J^\sharp)-(S^p +S^{-p})\| <{\varepsilon}_4$$ there is a ${\tilde }J\in{{\mathcal T}}_{J_0}$ so that $$\|J^\sharp-{\tilde }J\| \leq {\frac}{1}{2n}$$ Putting these together implies .
[Tn9.4]{} Let $\ell_1,\ell_2, \dots$ be an arbitrary sequence in $(2,3,4,\dots)$. For any $\ell_1$-periodic Jacobi matrix $J^{(0)}$, there exist $k_2, k_3, \dots$ so that for any limit periodic $J$ with Jacobi coefficients $$\begin{aligned}
a_n &=a_n^{(0)} +\sum_{m=2}^\infty \operatorname{Re}[A_m e^{2\pi in/\ell_1\ell_2\dots\ell_m}] {\label}{n9.9} \\
b_n &=b_n^{(0)} +\sum_{m=2}^\infty \operatorname{Re}[B_m e^{2\pi in/\ell_1\dots\ell_m}] {\label}{9.10}\end{aligned}$$ obeying $${\label}{n9.11x}
{\lvertA_m\rvert} + {\lvertB_m\rvert} \leq k_m$$ we have that $\sigma(J)$ is a Denisov–Rakhmanov set.
The study of limit periodic discrete Schrödinger operators with small tails was initiated by Avron–Simon [@S147] and Chulaevsky [@Chu]. They prove purely a.c. spectrum.
As in [@S147; @Chu], one can pick the $k_m$’s so the spectrum is purely a.c. and so that the union of all isospectral tori for the periodic approximates lie in a fixed $\ell^\infty$ compact sets. This implies the limit periodic potentials also have compact isospectral sets, and within this compact set, weak convergence implies norm convergence so hypothesis (1) of Proposition \[Pn9.2\] holds. By decreasing the $k_m$’s if necessary, Proposition \[Pn9.3\], continuity of the spectrum in $\ell^\infty$ norm, and absolute continuity of periodic spectrum imply we can be sure that holds. Thus Proposition \[Pn9.2\] implies this theorem.
Our final theorem in this section is the following:
[Tn9.5]{} Fix $\ell$. Let ${{\mathcal G}}=\{(\alpha_1, \beta_1, \alpha_2, \dots,
\alpha_{\ell+1}, \beta_{\ell+1})\in{{\mathbb{R}}}^{2\ell+2} : \alpha_1 <\beta_1 < \alpha_2 < \beta_2 <
\cdots < \beta_{\ell+1}\}$. For $(\vec{\alpha}, \vec{\beta})\in{{\mathcal G}}$, define $${\label}{n9.11}
{{\mathcal{E}}}(\vec{\alpha},\vec{\beta})=\bigcup_{j=1}^{\ell+1}\, [\alpha_j,\beta_j]$$ Then $\{(\vec{\alpha},\vec{\beta}) : {{\mathcal{E}}}(\vec{\alpha},\vec{\beta})\text{ is a Denisov--Rahkmanov
set}\}$ contains a dense $G_\delta$.
1\. As we have seen, the ${{\mathcal{E}}}(\vec{\alpha},\vec{\beta})$ which arise from periodic problems are precisely those where the harmonic measure of each $e_j=[\alpha_j, \beta_j]$ is rational. In particular, if we fix $\vec{\alpha}$ and $\beta_{\ell+1}$, the set of $\beta$’s that are periodic is countable, and so certainly not a $G_\delta$. We show that the family that leads to Denisov–Rahkmanov sets is uncountable.
2\. It is a reasonable conjecture that every ${{\mathcal{E}}}$ is a Denisov–Rakhmanov set, so this result is weak. We include it because it is such a ‘cheap’ way to go beyond the periodic case using only that case.
For each $(\vec{\alpha},\vec{\beta})$, it is known [@SY] that there is an isospectral torus ${{\mathcal T}}$ of almost periodic $J$’s where (whole line) spectrum is precisely ${{\mathcal{E}}}(\vec{\alpha},\vec{\beta})$. It follows from the construction in [@SY] that if $(\vec{\alpha}^{(n)}, \vec{\beta}^{(n)})\in{{\mathcal G}}$ converge to $(\vec{\alpha}^{(\infty)},
\vec{\beta}^{(\infty)})\in{{\mathcal G}}$, then condition (1) of Proposition \[Pn9.2\] holds.
Let ${{\mathcal G}}_p$ be the subset of ${{\mathcal G}}$ coming from periodic problems—this is dense in ${{\mathcal G}}$. For $(\vec{\alpha}^{(0)},\vec{\beta}^{(0)})\in{{\mathcal G}}_p$, pick $J(\vec{\alpha}^{(0)},
\vec{\beta}^{(0)})$ periodic with ${{\mathcal{E}}}(\vec{\alpha}^{(0)},\vec{\beta}^{(0)})$ as spectrum and pick $\delta_n (\vec{\alpha}^{(0)},\vec{\beta}^{(0)})$ via Proposition \[Pn9.3\] requiring $\delta_n <{\frac}12\min({\lvert\beta_j^{(0)}-\alpha_j^{(0)}\rvert}, {\lvert\alpha_{j+1}^{(0)} -\beta_j^{(0)}\rvert})$. Let $U^{(n)}(\vec{\alpha}_0,\vec{\beta}_0)=\{(\vec{\alpha},\vec{\beta}) : E(\vec{\alpha},
\vec{\beta})$ obeys / for ${{\mathcal{E}}}= {{\mathcal{E}}}(\vec{\alpha},\vec{\beta})$, ${{\mathcal{E}}}_0=
{{\mathcal{E}}}(\vec{\alpha}^{(0)},\vec{\beta}^{(0)})$ and $\delta=\delta_n\}$, and let $$U^{(n)}= \bigcup_{{{\mathcal G}}_p} U^{(n)} (\alpha^{(0)},\beta^{(0)})$$ This is dense and open. Then $\cap_n U^{(n)}$ is a dense $G_\delta$ whose points, by construction and Proposition \[Pn9.2\], correspond to Denisov–Rakhmanov sets.
Sum Rules for MOPRL
===================
[s8]{} In this section, our main goal is to prove the following two theorems about block Jacobi matrices:
[T8.1]{} Let ${{\mathcal J}}$ be a block Jacobi matrix with $\ell\times\ell$ Jacobi parameters $\{A_n\}_{n=1}^\infty, \{B_n\}_{n=1}^\infty$ and matrix measure $d{\eta}$. Let $m(E)$ be given by and suppose $\sigma_{\text{\rm{ess}}}({{\mathcal J}})=[-2,2]$. Define for $z\in{{\mathbb{D}}}\setminus \{z=E+E^{-1} : E\in\sigma({{\mathcal J}})
\setminus [-2,2]\}$ $${\label}{8.1}
M(z) =-m(z+z^{-1})$$ Let $F,G$ be the functions $${\label}{8.2}
F(\beta + \beta^{-1}) = \tfrac14\, [\beta^2 -\beta^{-2}-\log \beta^2]$$ for $\beta\in{{\mathbb{R}}}\setminus [-1,1]$, that is, $E=\beta + \beta^{-1}\in{{\mathbb{R}}}\setminus [-2,2]$ and $${\label}{8.3}
G(a)=a^2 -1 - \log(a^2) \qquad a\in (0,\infty)$$ Then $\lim_{r\uparrow 1} M(re^{i\theta})$ exists for a.e. $\theta$ and $${\label}{8.4}
\begin{split}
{\frac}{1}{2\pi}\int \log &\biggl( {\frac}{\sin^\ell \theta}{\det(\operatorname{Im}M(e^{i\theta}))}\biggr)
\sin^2 \theta\, d\theta \\
&+ \sum_{E\in\sigma({{\mathcal J}})\setminus [-2,2]}\, F(E)
= \sum_{n=1}^\infty {\text{\rm{Tr}}}(\tfrac14\, B_n^2 + \tfrac12\, G(|A_n|))
\end{split}$$
1\. All terms are positive (since $F$ and $G$ are positive, this is evident for two terms; positivity of the integral will be seen below), so this sum rule always makes sense, although some terms may be $+\infty$.
2\. Recall that $|A_n|=\sqrt{A_n^\dagger A_n^{ }}$; although since the formula for $G(a)$ only involves $a^2$, one does not need to take a square-root.
3\. Because of the trace and absolute value, ${\text{\rm{Tr}}}({\frac}14 B_n^2 + {\frac}12 G({\lvertA_n\rvert}))$ is constant over equivalence classes of Jacobi matrix parameters.
4\. In the type 1 case, the RHS of is finite if and only if ${{\mathcal J}}-S^p-S^{-p}$ is Hilbert–Schmidt. This is also true when ${{\mathcal J}}$ is of type 3; see Proposition \[P9.12\].
[T8.2]{} Consider the three quantities: $$\begin{aligned}
Z({{\mathcal J}}) &= {\frac}{1}{4\pi}\int_0^{2\pi} \log
\biggl( {\frac}{\sin^\ell\theta}{\det(\operatorname{Im}M(e^{i\theta}))}\biggr)\, d\theta {\label}{8.5} \\
{{\mathcal{E}}}_0({{\mathcal J}}) &=\sum_{E\notin\sigma({{\mathcal J}})\setminus [-2,2]}\, \log ({\lvert\beta\rvert}) {\label}{8.6}\end{aligned}$$ where $\beta$ is related to $E$ by $${\label}{8.4x}
\beta\in{{\mathbb{R}}}\setminus [-1,1] \qquad E=\beta+\beta^{-1}$$ and $${\label}{8.5x}
A_0 ({{\mathcal J}}) = \lim_{N\to\infty} -\sum_{n=1}^N \log (\det({\lvertA_n\rvert}))$$ which we suppose exists but it may be $+\infty$ or $-\infty$. Then
If any two of $Z,{{\mathcal{E}}}_0, A_0$ are finite, then so is the third.
If all are finite, then $${\label}{8.6x}
Z(J)=A_0(J) + {{\mathcal{E}}}_0(J)$$
If all are finite, then $${\label}{8.7}
\lim_{N\to\infty} \, \sum_{n=1}^N \, {\text{\rm{Tr}}}(B_n)$$ exists.
We will prove (and actually use it to prove Theorem \[T1.3\]) that if ${{\mathcal{E}}}_0(J)<\infty$, then $Z(J)<\infty$ so long as $${\label}{8.8}
\underline{A}_0 ({{\mathcal J}}) =\liminf_N \biggl( -\sum_{n=1}^N \log (\det({\lvertA_n\rvert}))\biggr) <\infty$$
Theorem \[T8.1\] is a matrix-valued analog of the OPRL $P_2$ sum rule of Killip–Simon [@KS], and Theorem \[T8.2\] of the OPRL Case $C_0$ sum rule by Simon–Zlatoš [@SZ]. Both were refinements of sum rules of Case [@Case1; @Case2] who in turn was motivated by earlier KdV and Toda sum rules. Case only considered short-range ${\lverta_n-1\rvert} + {\lvertb_n\rvert}$, while [@KS; @SZ] considered the necessary techniques to go up to the borderline of validity. [@SZ] had some simplifications of [@KS], and [@Sim288] further simplified, although each of the later two proofs depends heavily on the earlier ones. Here, following Simon [@Sim288], we will prove a nonlocal step-by-step sum rule. As there, the key is a suitable representation theorem for meromorphic Herglotz functions—in this case, extended to matrix-valued functions.
For $a\in (-1,1)$, we define Blaschke factors as usual by $${\label}{8.9}
b(z,a) = \begin{cases} {\frac}{a-z}{1-az} & 0<a<1 \\
{\frac}{z-a}{1-az} & -1 < a\leq 0 \end{cases}$$
[P8.3]{} Let $f(z)$ be an $\ell\times\ell$ matrix-valued meromorphic function on ${{\mathbb{D}}}$ so that $$\begin{aligned}
{2}
& \text{\rm{(i)}} \qquad && \pm \operatorname{Im}f(z) >0 \text{ when } \pm\operatorname{Im}z >0 {\label}{8.10} \\
& \text{\rm{(ii)}} \qquad && \lim_{z\to 0}\, f(z) z^{-1} ={{\boldsymbol{1}}}{\label}{8.11}\end{aligned}$$ where $\operatorname{Im}f \equiv \frac1{2i}(f-f^\dagger)$. Then
For a.e. $\theta$, $\lim_{r\uparrow 1} f(re^{i\theta})\equiv
f(e^{i\theta})$ exists.
$\log {\lvert\det(f(e^{i\theta}))\rvert} \in \cap_{1\leq p <\infty}
L^p (\partial{{\mathbb{D}}}, d\theta/2\pi)$
All the zeros and poles of $\det (f(z))$ lie on $(-1,1)$ and are of finite order. Let $\{z_j\}_{j=1}^\infty$ and $\{p_j\}_{j=1}^\infty$ be those zeros and poles of $\det (f(z))$ repeated up to multiplicity [[(]{}]{}it can also happen that both sets are finite[[)]{}]{}. $z=0$ is not included in $\{z_j\}$. Then $${\label}{8.12}
B_\infty (z) =\lim_{r\uparrow 1} \, {\frac}{\prod_{{\lvertz_j\rvert}<r} b(z,z_j)}
{\prod_{{\lvertp_j\rvert}<r} b(z, p_j)}$$ exists and obeys:
$B_\infty$ is analytic and nonvanishing on ${{\mathbb{C}}}\setminus \{z_j\}
\cup \{p_j\}\cup\{z_j^{-1}\}\cup \{p_j^{-1}\}\cup \{\pm 1\}$
${\lvertB_\infty (e^{i\theta})\rvert}=1$ on $\partial{{\mathbb{D}}}\setminus\{\pm 1\}$
$${\label}{8.12a}
{\lvert\arg (B_\infty(z))\rvert}\leq 2\pi \ell$$
for ${\lvertz\rvert}<1$ with $\arg$ normalized by $\arg B_\infty (0)=0$.
We have the representation $${\label}{8.12b}
\det(f(z)) =z^\ell B_\infty (z) \exp\biggl( \int {\frac}{e^{i\theta}+z}{{e^{i\theta}-z}}\,
\log {\lvert\det (f(e^{i\theta}))\rvert}\, {\frac}{d\theta}{\pi}\biggr)$$
1\. It should be possible to prove that $0<\arg (B_\infty (z))
<\pi \ell$ for $\operatorname{Im}z>0$; we settle for the weaker result.
2\. is not central for a result of this type, but it is true in applications and simplifies the notation.
3\. This result for $\ell=1$ is in [@Sim288]. $\ell >1$ has some subtleties, but the basic strategy we use is that of [@Sim288].
We will prove this result in a sequence of lemmas:
[L8.4]{} $\det (f(z))$ is analytic and nonvanishing in $\Omega\equiv
\{z : z\in{{\mathbb{D}}}, \, \operatorname{Im}z>0\}$, and $\arg (\det(f(z)))$ can be chosen in that region to be continuous so that $${\label}{8.13}
0 < \arg (\det(f(z))) <\pi \ell$$
In $\Omega$, all matrix elements $\langle\varphi, f(z)\varphi\rangle$ are analytic and have a.e. boundary values (since they are scalar Herglotz functions), so by polarization, $f(z)$ is analytic on $\Omega$ and has a.e. boundary values. Thus $\det (f(z))$ as a polynomial in matrix elements is analytic on $\Omega$.
Consider $${\label}{8.14}
P(\lambda, z) = \det(\lambda{{\boldsymbol{1}}}-f(z))$$ which is a polynomial in $\lambda$ with analytic coefficients away from the poles of $f$. It follows, for $z$ near any $z_0$ about which $f$ is analytic, that the roots $P(\lambda,z)=0$ written as a function of $z$ are analytic functions in $(z-z_0)^{1/k}$ for some $k$ depending on $z_0$. It then follows that near any fixed $z_0$, all roots are analytic, that is, singularities are isolated.
Pick $x_0\in (0,{\varepsilon})$ so that $x_0 f(x_0) >0$, so all eigenvalues $\lambda_1(x_0),
\dots, \lambda_\ell(x_0)$ are in $(0,\infty)$. Let $z\in\Omega$ be a point about which all eigenvalues are analytic, and let $\gamma(z)$ be a simple closed path from $x_0$ to $z$ which avoids the discrete set where eigenvalues are not analytic and lie in $\Omega$ except for $x_0$ with, say, $\gamma(0)=x_0$, $\gamma(1)=z$. By analytically continuing eigenvalues, we get function $\{\lambda_j(z)\}_{j=1}^\ell$, so $\lambda_j (z)$ are all the eigenvalues of $f(\gamma(t))$ and $\lambda_j(0)\in
(0,\infty)$. By $\operatorname{Im}f>0$, $\operatorname{Im}\lambda_j(z) >0$, so if we define $\arg (\lambda_j
(z))$ with $\arg (\lambda_j(0))=0$, we have $$0 < \arg (\lambda_j(z)) <\pi$$ Thus $$\arg (\det(f(z))) = \sum_{j=1}^\ell \arg (\lambda_j(z))$$ normalized by $\arg (\det (f(x_0)))=0$ obeys .
By analyticity of $\det (f(z))$ and the fact that it is nonvanishing, $\arg(\det(f(z)))$ is uniquely defined as a continuous function on $\Omega$ with $\lim_{{\varepsilon}\downarrow 0}
\arg(\det(f(x_0 + i{\varepsilon})))=0$. By the above, holds at all points $z$ in $\Omega$ where all eigenvalues are analytic and so, by continuity and the open mapping theorem for analytic functions, all points.
[L8.5]{} Let $a<b$ lie in $(-1,1)$ so that both $a$ and $b$ are neither a zero nor a pole of $\det (f(z))$. Let $Z(a,b), P(a,b)$ be the number of zeros, poles of $\det(f(z))$ in $(a,b)$ counting multiplicity. Then $${\label}{8.15}
{\lvertZ(a,b)-P(a,b)\rvert} \leq\ell$$
By the argument principle, $2\pi (Z-P)$ is the change of $\arg (\det(f(z))$ along the circle through $a$ and $b$ centered at ${\frac}12 (a+b)$. By Lemma \[L8.4\], this is at most $2(\ell\pi)$.
[L8.6]{} The sets of zeros and poles [[(]{}]{}with multiplicity[[)]{}]{} of $\det(f(z))$, including the $\ell$-fold zero at $z=0$, can be written as $\ell$ subsets $z_j^{(k)}, p_j^{(k)}$ with $k=1, \dots, \ell$ and $-{\tilde }N_k < j < N_k$ [[(]{}]{}with $N_k$ and ${\tilde }N_k$ among $1,2,\dots, \infty$[[)]{}]{} so that $z_0^{(k)}=0$ and $${\label}{8.16}
z_j^{(k)} < p_j^{(k)} < z_{j+1}^{(k)}$$ for all allowed values of $j$.
1\. If there are infinitely many $z$ in $(-1,0)$ and in $(0,1)$, then ${\tilde }N_k=N_k=\infty$ for all $k$. The awkwardness requiring $N_k,{\tilde }N_k$ is only needed if there are finitely many zeros.
2\. To avoid notational complexity, we slightly lied if $N_k$ or ${\tilde }N_k$ is finite. If $N_k$ is finite, $z_j^{(k)}$ runs to $j=N_k$. $p_j^{(k)}$ can then run to either $N_k$ or $N_k-1$.
Construct $S_1, S_2, \dots, S_\ell$ as follows: Set $z_0^{(1)}=0$. Let $p_0^{(1)}$ be the first pole larger than $z_0^{(1)}$, $z_1^{(1)}$ the first zero larger than $p_0^{(1)}$, $p_1^{(1)}$ the next pole, etc. This either continues indefinitely, in which case we set $N_1=\infty$, or stops because there is no next zero or pole. Then do the same to the left of $0$, that is, $p_{-1}^{(1)}$ is the first pole smaller than $z_1^{(1)}$, etc. Clearly, the points in $S_1$ obey . Now remove the points of $S_1$ (or decrease their multiplicity by $1$) and repeat the construction (starting with $z_0^{(2)}=0$) to make $S_2, S_3, \dots, S_\ell$.
We claim that after we construct $\ell$ $S_j$’s, we have exhausted all the poles and zeros. Let us show this is true for $(0,1)$; the argument for $(-1,0)$ is similar (and since $0$ has multiplicity $\ell$, it is removed after $\ell$ steps).
Suppose ${\tilde }z$ is a zero that is left and it is closer to zero than any leftover zero or pole. If ${\tilde }z$ lies in some $(p_j^{(k)}, z_{j+1}^{(k)})$, $j=0,1,\dots$, we could have used it as $z_{j+1}^{(k)}$ so it cannot lie in any such interval. Put differently, there are only matched zeros and poles in $(0,{\tilde }z)\cap \cup_{j=1}^\ell S_j$. By the choice of ${\tilde }z$, there are no other poles in $(0,{\tilde }z)$. Thus, for small $\delta$, the interval $(-\delta, {\tilde }z+\delta)$ has $\ell+1$ extra zeros over poles, violating Lemma \[L8.5\]. So the closest leftover point is not a zero.
Suppose ${\tilde }p$ is a pole that is left and it is closer to zero than any other leftover zero or pole. As above, ${\tilde }p$ cannot lie in any $(z_j^{(k)}, p_j^{(k)})$, $j=0,1,\dots$, so there are only matched zeros and poles in $[0,{\tilde }p)\cap\cup_{j=1}^\ell
S_j$. But then, for small $\delta$, $(\delta, {\tilde }p+ \delta)$ has $\ell+1$ extra poles, violating Lemma \[L8.5\]. Thus $\cup_{j=1}^\ell S_j$ includes all zeros and poles.
[L8.7]{} The limit $B_\infty(z)$ of exists and obeys conditions [[(i)–(iii)]{}]{} of Proposition \[P8.3\][[(c)]{}]{}.
Renumber the $p_j^{(k)}$ into a single sequence $p_1, p_2, \dots$, so ${\lvertp_1\rvert} \leq {\lvertp_2\rvert} \leq \cdots$ and let $z_m$ be the correspondingly paired $z_{j+1}^{(k)}$ (paired to the $p_j^{(k)}$ that is $p$). Since $\{(p_j^{(k)},
z_{j+1}^{(k)})\}_{j=1}^{N_j}$ are disjoint subsets of $(0,1)$ for each fixed $k$, $$\sum_{j=1}^\infty \, {\lvertz_{j+1}^{(k)} - p_j^{(k)}\rvert} =
\sum_{j=1}^\infty z_{j+1}^{(k)} - p_j^{(k)} <1$$ so we see that $$\sum_{j=1}^\infty {\lvertz_j-p_j\rvert} \leq 2\ell$$ The existence of $B_\infty$ then follows by Proposition 13.8.2 of [@OPUC2], as do (i) and (ii).
To get (iii), we note that just taking the zeros and poles in a single $S_j$ yields a set obeying (13.8.5) and (13.8.6) of [@OPUC2]. So, by (13.8.10), that product has $\arg$ bounded by $2\pi$. The $\ell$-fold product thus obeys .
Given Lemma \[L8.7\], the proof is essentially that of Theorem 13.8.3 of [@OPUC2]. $${\label}{8.17}
g(z)\equiv {\frac}{\det(f(z))}{z^\ell B_\infty (z)}$$ is analytic and nonvanishing on ${{\mathbb{D}}}$ with $g(0)>0$ (since $B_\infty (0) >0$). Moreover, by and , $${\label}{8.18}
{\lvert\arg g(z)\rvert} \leq 4\pi \ell$$ so, by M. Riesz’ theorem, $\log (g(z))\in\cap_{p<\infty} H^p (\partial{{\mathbb{D}}})$ from which (a), (b) of the theorem are immediate and (d) follows from the Poisson representation for $\log (g(z))$ since $\log ({\lvertg(e^{i\theta})\rvert})=\log
({\lvert\det (f(e^{i\theta}))\rvert}$.
Now we turn to block Jacobi matrices where we obtain:
[T8.8]{}\
Let ${{\mathcal J}}$ be a block Jacobi matrix with $\sigma_{\text{\rm{ess}}}({{\mathcal J}})\subset [-2,2]$ and Jacobi parameters $\{A_n, B_n\}_{n=1}^\infty$. Let ${{\mathcal J}}^{(1)}$ denote this Jacobi matrix with the top row of blocks and left-most column of blocks removed. Let $m(E), m^{(1)}(E)$ be the $m$-functions given by . Let $M,M^{(1)}$ be defined on ${{\mathbb{D}}}$ by $${\label}{8.19}
M(z) =-m(z+z^{-1})$$ with poles at $\{p_i\}_{i=1}^N$ where $p_i + p_i^{-1}$ are eigenvalues of ${{\mathcal J}}$. We repeat each $p_i$ a number of times equal to the multiplicity of the eigenvalues [[(]{}]{}equivalently, the rank of the residue[[)]{}]{}. Let $\{z_i\}_{i=1}^{N-1}$ be the corresponding points for ${{\mathcal J}}^{(1)}$. Then
The Blaschke product, $B_\infty (z)$, defined by the $\{z_i\}
\cup\{p_i\}$ via exists and obeys [[(i)–(iii)]{}]{} of Proposition \[P8.3\][[(c)]{}]{}.
$M(z)$ and $M^{(1)}(z)$ have limits $M(e^{i\theta})$ and $M^{(1)}(e^{i\theta})$ as $r\uparrow 1$ for $z=re^{i\theta}$ for a.e. $\theta$ in $\partial{{\mathbb{D}}}$ and $${\label}{8.20}
\log\biggl[ {\frac}{\det (\operatorname{Im}M)(e^{i\theta})}{\det(\operatorname{Im}M^{(1)}(e^{i\theta}))}\biggr]
\in\bigcap_{1\leq p <\infty} L^p \biggl( \partial{{\mathbb{D}}}, {\frac}{d\theta}{2\pi}\biggr)$$
$${\label}{8.21}
\det\biggl( {\frac}{{\lvertA_1\rvert} M(z)}{z}\biggr)
= B_\infty (z) \exp\biggl( \int {\frac}{e^{i\theta}+z}{e^{i\theta}-z} \,
\log\biggl[ {\frac}{\det (\operatorname{Im}M(e^{i\theta}))}{\det (\operatorname{Im}M^{(1)}(e^{i\theta}))}\biggr]
\frac{d\theta}{4\pi}\biggr)$$
As in the case $\ell=1$, it can happen (although not in examples where sum rules are finite) that $\det (\operatorname{Im}M(e^{i\theta}))=\det (\operatorname{Im}M^{(1)}
(e^{i\theta}))=0$ for $\theta$ in a set of positive measure. (b) and (c) are shorthand for the more precise
For a.e. $\theta$, $\det (\operatorname{Im}M(e^{i\theta}))=0$ if and only if $\det (\operatorname{Im}M^{(1)}(e^{i\theta}))=0$.
There is an a.e. positive function $g(\theta)$ on $\partial{{\mathbb{D}}}$, equal to $\det (\operatorname{Im}M(e^{i\theta}))/\det (\operatorname{Im}M^{(1)}(e^{i\theta}))$ when the ratio is not $0/0$ so that and hold if the formal ratio is replaced by $g(\theta)$.
Given Proposition \[P8.3\], this is essentially identical to the proof of Theorem 13.8.4 of [@OPUC2] with care given to matrix issues. We begin by noting that for $n\to n+1$ first implies near $z=0$ $${\label}{8.22}
M^{(n+1)}(z)^{-1} = z^{-1} + O(1)$$ and then by that $${\label}{8.23}
\biggl( {\frac}{M^{(n)}(z)}{z}\biggr)^{-1} = 1-B_{n+1} z - (A_{n+1}^\dagger A_{n+1} -1)
z^2 + O(z^3)$$ Since $M^{(n)}(z)/z$ is near $1$ for $z$ small, we can compute its determinant using $${\label}{8.24}
\det(C) = \exp ({\text{\rm{Tr}}}(\log(C))$$ which holds if $\|C-1\|<1$. Thus $${\label}{8.25}
\begin{split}
\log &\det \biggl( {\frac}{M^{(n)}(z)}z \biggr) \\
&= {\text{\rm{Tr}}}(B_{n+1})z +
[{\text{\rm{Tr}}}\{ ([A_{n+1}^\dagger A_{n+1}-1]+ \tfrac12 B_{n+1}^2)\}]
z^2 + O(z^3)
\end{split}$$
In addition, implies $${\label}{8.26}
\operatorname{Im}[M(z)^{-1}] = \operatorname{Im}(z+z^{-1}) - A_1 \operatorname{Im}M_1(z) A_1^\dagger$$ so at points where $M(z)$ has radial limits (a.e. $\theta$, see below), $${\label}{8.27}
-[M(e^{i\theta})^\dagger]^{-1} \operatorname{Im}M(e^{i\theta}) [M(e^{i\theta})]^{-1} =
-A_1 \operatorname{Im}M(e^{i\theta}) A_1^\dagger$$ which, using (on account of $\det ({\lvertC\rvert})^2=\det (C^\dagger)\det (C)$) $${\lvert\det (A_1)\rvert} = \det ({\lvertA_1\rvert})$$ yields $${\label}{8.28}
{\lvert\det ({\lvertA_1\rvert} M(e^{i\theta}))\rvert}^2 =
{\frac}{\det (\operatorname{Im}M(e^{i\theta}))}{\det (\operatorname{Im}M_1 (e^{i\theta}))}$$
We now apply Proposition \[P8.3\] to $M(z)$ which obeys (since $\operatorname{Im}(z+z^{-1})<0$ on ${{\mathbb{D}}}$ and has a minus sign) and by .
By Theorem \[T5.6\], our $B_\infty (z)$ here (after perhaps canceling some zeros and poles) is the $B_\infty (z)$ of Proposition \[P8.3\]. (a) and (b) immediately follow from Proposition \[P8.3\]. We get from by using (noting has a $1/4\pi$ while a $1/2\pi$ on account of the square on the left side of ). We also use that if $c$ is a positive constant, $${\label}{8.29}
\exp\biggl( \int {\frac}{e^{i\theta}+z}{e^{i\theta}-z}\,
\log (c^2) {\frac}{d\theta}{4\pi}\biggr) =c$$
As in [@Sim288], we can get step-by-step $P_2$ (originally in [@KS]), $C_0$, $C_1$ (originally in [@SZ]) sum rules immediately from Taylor expansion of the $\log$ of . We let $\beta_j ({{\mathcal J}})$ be the numbers in $(-1,1)
\setminus \{0\}$ for which $E_j\equiv \beta_j + \beta_j^{-1}$ are eigenvalues of ${{\mathcal J}}$ counting multiplicities.
[T8.9]{} $\qquad$\
[[(i)]{}]{} $${\label}{8.30}
\begin{split}
{\frac}{1}{4\pi} &\int_0^{2\pi} \log \biggl( {\frac}{\det (\operatorname{Im}M^{(1)}(e^{i\theta}))}
{\det (\operatorname{Im}M (e^{i\theta}))} \biggr) d\theta \\
& -\sum_j \log ({\lvert\beta_j({{\mathcal J}})\rvert}) -
\log ({\lvert\beta_j ({{\mathcal J}}^{(1)})\rvert}) = -\log (\det{\lvertA_1\rvert})
\end{split}$$
[[(ii)]{}]{} $${\label}{8.31}
\begin{split}
-{\frac}{1}{2\pi} &\int_0^{2\pi} \log \biggl( {\frac}{\det (\operatorname{Im}M^{(1)}(e^{i\theta}))}
{\det (\operatorname{Im}M (e^{i\theta}))}\biggr) \cos\theta
\\
&+ \sum_j [\beta_j ({{\mathcal J}})
- (\beta_j^{-1} ({{\mathcal J}})^{-1})] - [\beta_j ({{\mathcal J}}^{(1)}) - (\beta_j ({{\mathcal J}}^{(1)}))^{-1}]
\equiv {\text{\rm{Tr}}}(B_1)
\end{split}$$
[[(iii)]{}]{} $${\label}{8.32}
\begin{split}
{\frac}{1}{4\pi} &\int \log \biggl( {\frac}{\det (\operatorname{Im}M^{(1)}(e^{i\theta}))}
{\det (\operatorname{Im}M (e^{i\theta}))}\biggr) \sin^2(\theta) \phi(\theta) \\
&+ \sum F (E_j({{\mathcal J}}))
-F(E_j ({{\mathcal J}}^{(1)})) = {\text{\rm{Tr}}}\tfrac14 \bigl( B_1^2 + \tfrac12\, G({\lvertA_1\rvert}) \bigr)
\end{split}$$ where $F$ is given by and $G$ by .
The $E_j({{\mathcal J}})$ in $(-\infty, -2)$ and $(2,\infty)$ and $E_j({{\mathcal J}}^{(1)})$ interlace in the $\ell=1$ case. In the general $\ell$ case, we have at most $\ell$ fewer eigenvalues of ${{\mathcal J}}^{(1)}$ on any $(-\infty, -E_0)$ or $(E_0,\infty)$ so, as in Lemma \[L8.6\], one can decompose into $\ell$ interlacing subsets. This and the monotonicity of functions like $F$ show the eigenvalue sums in – are conditionally convergent. Similarly, the integrals are always convergent.
Apply $\log$ to both sides of and take Taylor coefficients. The constant term is and the first derivative is . If $L(z)$ is the $\log$ of the left side and $R(z)$ of the right, then $$L(0) + \tfrac12\, L''(0) = R(0) + \tfrac12\, R''(0)$$ is .
The proofs of Theorems \[T8.1\]–\[T8.2\] are now identical to those of the scalar case; see, for example, the discussion of Theorems 13.8.6 and 13.8.8 of [@OPUC2]. In particular, $Z({{\mathcal J}})$ and $Q({{\mathcal J}})$ (the integral on the right of ) are negatives of relative entropies, and so, lower semi-continuous.
Szegő and Killip–Simon Theorems When All Gaps Are Open
======================================================
[s9]{} Our goal here is to prove Theorems \[T1.3\] and \[T1.4\]. Our strategy, of course, will be to translate Theorems \[T8.1\] and \[T8.2\] for $\Delta (J)$ to statements about $J$. Firstly, we need to relate the a.c. part of the matrix measure for $\Delta(J)$ to the a.c. part of the (scalar) measure for $J$. And secondly, to relate $\ell^2$ norms of coefficients of $\Delta(J)$ to the distance of $J$’s Jacobi parameters to the isospectral torus. We begin with the first question. Thus we take $${\label}{9.1}
d{\eta}_J(x)=\omega(x)\, dx + d{\eta}_{J,{\text{\rm{s}}}}(x)$$ with $d\mu_{J,{\text{\rm{s}}}}$ singular and $\omega$ supported precisely on $\sigma_{\text{\rm{ess}}}(J_0)$. By this assumption and the spectral mapping theorem, $\Delta(J)$ has a.c. spectrum precisely on $[-2,2]$ so the matrix measure for $\Delta(J)$ has the form $${\label}{9.2}
d{\eta}_{\Delta(J)}(E)=W(E)\, dE + d{\eta}_{\Delta(J),{\text{\rm{s}}}}(E)$$
[P9.1]{} Let $J_0$ be a periodic Jacobi matrix with period $p$ and $J$ a Jacobi matrix with Jacobi parameters $\{a_n, b_n\}_{n=1}^\infty$ and measure $d{\eta}_J$ of the form with $\omega$ supported on $\sigma_{\text{\rm{ess}}}(J_0)$. Let $\Delta$ be the discriminant for $J_0$ and $W(E)$ the a.c. part of the $p\times p$ matrix-valued measure $d{\eta}_{\Delta(J)}$ associated to $\Delta(J)$ [[(]{}]{}so $W$ is a $p\times p$ matrix[[)]{}]{}. Then for $E\in (-2,2)$ and $\Delta^{-1} (E) =
\{x_1, \dots, x_p\}$, $${\label}{9.3}
\det (W(E)) = \biggl( \prod_{j=1}^{p} a_j^{p-j} \biggr)^{-2}
\biggl( \prod_{j=1}^{p} a^{(0)}_j \biggr)^p
\biggl( \prod_{j=1}^p \omega(x_j) \biggr)$$
In the block Jacobi form, $d{\eta}_{\Delta(J)}$ has $jk$ matrix element equal to the spectral measure of the operator $\Delta(J)$ associated to $\delta_j,\delta_k$, that is, $\int F(x) (d{\eta}_{\Delta}(x))_{jk}
=\langle \delta_j, F(\Delta (J))\delta_k\rangle$. But $\delta_j=
p_{j-1}(J)\delta_1$. It follows that $$\begin{aligned}
W_{kj}(E) &= \sum_{\ell=1}^p \omega (x_\ell) \bigl( |\Delta'(x_\ell)| \bigr)^{-1} p_{k-1} (x_\ell) p_{j-1} (x_\ell) {\label}{9.4}
$$ Note that the factors of $1/\Delta'$ arise from the Jacobian $\tfrac{dE}{dx} = \Delta'(x)$. We can re-write as $W_{kj}(E) = (M\!AM^t)_{kj}$ where $A$ is the diagonal matrix $${\label}{9.7}
A_{\ell m} = \delta_{\ell m} \omega (x_\ell) \bigl( |\Delta'(x_\ell)| \bigr)^{-1}$$ and $M$ is the matrix $${\label}{9.6}
M_{k\ell} = p_{k-1}(x_\ell) \qquad k=1, \dots, p;\, \ell=1, \dots, p$$
Next we compute $\det(M)$; $\det(A)$ is easy. Note that $${\label}{9.9}
p_{k-1} (x_\ell) = \biggl(\, \prod_{j=1}^{k-1} a_j\biggr)^{-1} x_\ell^{k-1} +
\text{lower order}$$ Moreover, inductively one sees that the lower order terms can be neglected in the determinant—they can be removed by subtracting a multiple of rows above (i.e., smaller values of $k$). Thus, $${\label}{9.10x}
\det(M) = \biggl[\, \prod_{k=1}^p \biggl(\, \prod_{j=1}^{k-1} a_j\biggr)^{-1}\biggr]
\det (x_\ell^{k-1})= \biggl( \prod_{j=1}^{p} a_j^{p-j} \biggr)^{-1} \biggl( \prod_{j
> k} (x_j-x_k) \biggr)$$ by the well-known formula for Vandermonde determinants. This can be simplified further. The points $x_j$ are precisely the zeros of the polynomial $\Delta(x)-E$; hence, invoking , $$\Delta(x) - E = \biggl(\prod_{j=1}^p a^{(0)}_j \biggr)^{-1} \biggl(\prod_{k=1}^p (x-x_k) \biggr)$$ In this way we discover that $${\label}{9.10new}
\det(M)^2 = \biggl( \prod_{j=1}^{p} a_j^{p-j} \biggr)^{-2} \biggl(
\prod_{j=1}^{p} a^{(0)}_j \biggr)^p \biggl(\prod_{k=1}^p
|\Delta'(x_k)| \biggr)$$ Multiplying this by $\det(A)$ gives .
[C9.2&3]{} If $J_0$ has all gaps open and $\alpha >-1$, then $${\label}{9.11}
\int_{-2}^2 (4-E^2)^\alpha {\lvert\log \det (W(E))\rvert}\, dE <\infty$$ if and only if $${\label}{9.12}
\int_{\sigma_{\text{\rm{ess}}}(J_0)}{\text{\rm{dist}}}(x,{{\mathbb{R}}}\setminus\sigma_{\text{\rm{ess}}}(J_0))^\alpha
{\lvert\log \omega(x)\rvert}\, dx <\infty$$ When $\alpha=-\frac12$, the same conclusion holds even if some gaps are closed.
Since $\alpha >-1$, $(4-E^2)^\alpha$ (resp., ${\text{\rm{dist}}}(\dots)^\alpha$) are in $L^p$ for some $p>1$, so the $\log_+ (\quad)$ is always integrable and these conditions are equivalent to the integral without ${\lvert\cdot\rvert}$ being larger than $-\infty$.
Changing variables via $E=\Delta(x)$ and applying Proposition \[P9.1\] shows that holds if and only if $$\int_{\sigma_{\text{\rm{ess}}}(J_0)} |\log \omega(x) | \,(4-\Delta(x)^2)^\alpha \,|\Delta'(x)|\,dx < \infty$$
If all gaps are open then $|\Delta'(x)|$ is strictly positive on $\sigma_{\text{\rm{ess}}}(J_0)$, while $4-\Delta(x)^2$ is a polynomial with a simple zero at each band edge (and no others). This proves the first claim.
At a closed gap, $4-\Delta(x)^2$ has a double zero and $\Delta'(x)$ a simple zero. When $\alpha=-\frac12$, these cancel exactly.
Next we turn to the $\ell^2$ issue. Given any two-sided periodic matrix ${\tilde }J$ with Jacobi parameters $\{a_n, b_n\}_{n=1}^p$ and fixed periodic $J_0$, let $B_{J_0}({\tilde }J), A_{J_0}({\tilde }J)$ be the constant $p\times p$ blocks in $\Delta_{J_0} ({\tilde }J)$. We are heading towards showing that $\|B_{J_0}({\tilde }J)\|_2^2
+\|A_{J_0}({\tilde }J)-{{\boldsymbol{1}}}\|_2^2$ is comparable to ${\text{\rm{dist}}}((a_n,b_n)_{n=1}^p,{{\mathcal T}}_{J_0})$. This will be the key to showing $\ell^2$ tails in the matrix pieces of $\Delta_{J_0}(J)-S^p-S^{-p}$ for general $J$ is equivalent to .
The crucial fact will be that the polynomial coefficients of $\Delta_{J_0}({{\tilde }J})-\Delta_{J_0}(J_0)$ are comparable to ${\text{\rm{dist}}}((a_n,b_n)_{n=1}^p, {{\mathcal T}}_{J_0})$. For this we need the following, which is a simple application of the implicit function theorem and compactness:
[L9.3]{} Let $F$ be a $C^\infty$ map of an open set $U\subset{{\mathbb{R}}}^n$ to ${{\mathbb{R}}}^\ell$ with $\ell <n$. Suppose ${{\mathcal T}}=F^{-1}(y_0)$ is a smooth manifold of dimension $n-\ell$ and compact for some $y_0\in{{\mathbb{R}}}^\ell$, and $${\label}{9.13}
{\text{\rm{rank}}}((\nabla F)(x_0))=\ell$$ for all $x_0\in{{\mathcal T}}$. Then for any compact neighborhood, $K$, of ${{\mathcal T}}$, there are $c_K,d_K \in (0,\infty)$ so for all $x\in K$, $${\label}{9.14}
c_K{\lvertF(x)-y_0\rvert} \leq {\text{\rm{dist}}}(x,{{\mathcal T}})\leq d_K{\lvertF(x)-y_0\rvert}$$
One can restate in a more illuminated way in terms of the components $F_1, \dots, F_\ell$ of ${{\mathcal T}}$. Of course, $\nabla F_j(x_0)$ is orthogonal to ${{\mathcal T}}$ at $x_0$. The condition is equivalent to saying that $\{\nabla _j F(x_0)\}_{j=1}^\ell$ span the normal bundle to ${{\mathcal T}}$. This is equivalent to saying they are linearly independent. Notice that if $J_0$ has all gaps open, ${{\mathcal T}}_{J_0}$ is of dimension $p-1=2p-(p+1)$ and $\Delta_{J_0}$ is a polynomial of degree $p$, hence with $p+1$ coefficients. Thus the following shows we can use Lemma \[L9.3\]:
[T9.4]{} Suppose all gaps are open for some periodic $J_0$. Then at any point in ${{\mathcal T}}_{J_0}$, the gradients of the derivatives of the coefficients of $\Delta_J$ span the normal bundle of ${{\mathcal T}}_{J_0}$ in ${{\mathbb{R}}}^{2p}$.
$\Delta_{J_0}$ has the form $$\Delta_{J_0}(x)=(a_1 \dots a_p)^{-1} \prod_{j=1}^p (x-\lambda_j) =
\sum_{j=0}^p c_j x^j$$ where $\lambda_j$ are the roots. The coefficients thus obey $$\begin{aligned}
c_p^{-1} &= a_1 \dots a_p {\label}{9.15} \\
c_\ell c_p^{-1} &= \sum_{1\leq k_1 \leq \cdots\leq k_{p-\ell}\leq p}
\lambda_{k_1} \dots \lambda_{k_{1-\ell}} \equiv s_{k-\ell} \qquad \ell <p {\label}{9.16}\end{aligned}$$ It is well known that if $${\label}{9.17}
t_\ell =\sum_{j=1}^p \lambda_j^\ell$$ then $t_\ell$ is $\ell s_\ell$ plus a polynomial in $\{s_j\}_{j=1}^{\ell-1}$, so $\{\nabla t_j\}_{j=1}^\ell$ and $\{\nabla s_j\}_{j=1}^\ell$ span the same space. It follows that we need only show the gradients of $c_p^{-1}$ and $t_\ell$ span the normal bundle of ${{\mathcal T}}_{J_0}$.
Let $${\label}{9.18}
{{\mathcal M}}_0=\{(a_n,b_n)_{\ell=1}^p : a_1 a_2\dots a_p = a_1^{(0)} a_2^{(0)}\dots
a_p^{(0)}; \, b_1+\cdots + b_p =b_1^{(0)}+\cdots + b_p^{(0)}\}$$ We know ${{\mathcal M}}_0\supset{{\mathcal T}}_{J_0}$. Clearly, $\nabla c_p^{-1}$ and $\nabla t_1$ span the normal bundle to ${{\mathcal M}}_0$ since $t_1=\sum_{j=1}^p b_j$ (see ). Thus we need only show the projections of $\{\nabla t_\ell\}_{\ell=2}^p$ into the tangent space of ${{\mathcal M}}_0$ span the normal bundle of ${{\mathcal T}}_{J_0}$ in ${{\mathcal M}}_0$.
Studies of the Toda flows show that ${{\mathcal M}}_0$ is a symplectic manifold with $\{t_\ell\}_{\ell=2}^p$ Poisson commuting. Since the symplectic form on ${{\mathcal M}}_0$ is nondegenerate, to say $\{\nabla t_j\}_{j=2}^p$ span the normal bundle is the same as saying that the Hamiltonian flows generated by $\{t_j\}_{j=2}^p$ span the tangent bundle of ${{\mathcal T}}_{J_0}$, or equivalently, given $\dim ({{\mathcal T}}_{J_0})=p-1$, that these Hamiltonian flows are independent.
This independence is a theorem of van Moerbeke [@vMoer Theorem 5.2] or [@SimonNew].
[L9.5]{} Let $\chi_k$ be the projection onto the $k$-dimensional space spanned by $\{\delta_j\}_{j=1}^k$. For any compact subset, $K$, of period $p$ Jacobi matrices, there exist constants $c_K$ and $d_K$ in $(0,\infty)$ so for all $J\in K$, $${\label}{9.19}
c_K \biggl\|\, \sum_{\ell=0}^p \alpha_\ell J^\ell
\chi_{p+1}\biggr\|_2 \leq \biggl( \sum_{\ell=0}^p\,
{\lvert\alpha_\ell\rvert}^2 \biggr)^{\frac12} \leq d_K \biggl\|\,
\sum_{\ell=0}^p \alpha_\ell J^\ell \chi_{p+1}\biggr\|_2$$
$\{J^\ell \chi_{p+1}\}_{\ell=0}^p$ are independent since $J^\ell$ has strictly positive elements in the $\ell$-th diagonal and $\{J^k\}_{k<\ell}$ only has zero elements there. Hence, the matrix $${\label}{9.20b}
\left. {\text{\rm{Tr}}}(\chi_{p+1} J^\ell J^k \chi_{p+1})\right|_{\ell,k=0,
\dots, p}$$ is strictly positive so holds for each fixed $J$. The optimal constants are clearly continuous so uniformly bounded above and below on $K$.
[P9.6]{} Let $J_0$ be a periodic Jacobi matrix with all gaps open. For any compact neighborhood $K$ of ${{\mathcal T}}_{J_0}$ in $(0,\infty)^p\times{{\mathbb{R}}}^p$, there are constants $c_K$ and $d_K$ in $(0,\infty)$ so that for all $J\in K$, $$\begin{aligned}
c_K (\|A_{J_0}(J)-{{\boldsymbol{1}}}\|^2 + \|B_{J_0}(J)\|^2)^{1/2}
&\leq {\text{\rm{dist}}}(J,{{\mathcal T}}_{J_0}) \\
&\leq d_K (\|A_{J_0}(J)-{{\boldsymbol{1}}}\|^2 + \|B_{J_0}(J)\|^2)^{1/2}
$$
We have that $$\begin{aligned}
{\label}{9.21}
2\|A_{J_0}(J)-{{\boldsymbol{1}}}\|^2 + \|B_{J_0}(J)\|^2 & \le
\|[\Delta_{J_0}(J)-(S^p + S^{-p})]\chi_p\|^2 \\
\nonumber & \le 4\|A_{J_0}(J)-{{\boldsymbol{1}}}\|^2 + 2\|B_{J_0}(J)\|^2\end{aligned}$$ But by the magic formula, $${\label}{9.22}
\Delta_{J_0}(J_0)=S^p + S^{-p}$$ so $${\label}{9.23}
[\Delta_{J_0}(J)-(S^p + S^{-p})]\chi_{p+1} = \sum_{\ell=0}^p
c_\ell J^\ell \chi_{p+1}$$ where $c_\ell$ is the difference of coefficients for $J$ and $J_0$. By Lemma \[L9.5\], $${\label}{9.24}
\|[\Delta_{J_0}(J)-(S^p + S^{-p})]\chi_{p+1}\|^2
\sim\sum_{\ell=0}^p \, {\lvertc_\ell\rvert}^2$$ where $\sim$ means the ratio is bounded above and away from zero on compact subsets.
By Lemma \[L9.3\] and Theorem \[T9.4\], $${\label}{9.25}
\sum_{\ell=0}^p\, {\lvertc_\ell\rvert}^2 \sim {\text{\rm{dist}}}(J,{{\mathcal T}}_{J_0})^2$$ Combining this with proves the proposition.
Now we take a general $J$ not periodic and form $\Delta_{J_0}(J)$ which is a one-sided block Jacobi matrix with block elements $A_{n,J_0}(J), B_{n,J_0}(J)$.
[L9.7]{} $\Delta_{J_0}(J)_{k\ell}$ for $k\leq \ell$ depends only on $\{b_j\}_{j=k-\alpha}^{\ell+\alpha}$ and $\{a_j\}_{j=k-\alpha}^{\ell+\alpha -1}$ where $\alpha = \lfloor{\frac}12 (p-(\ell-k))\rfloor$ is the greatest integer less than or equal to ${\frac}12[p-(\ell-k)]$.
Each factor of $J$ changes index by at most one. In order to get from $k$ to $\ell$, $\ell-k$ steps are needed. The remainder cannot go below $\ell-\alpha$ or above $k+\alpha$ and get back to $k$ in $p$ steps.
[L9.8]{} Let $J$ have Jacobi parameters $\{a_n,b_n\}_{n=1}^\infty$. Let ${\tilde }J$ be periodic with period $p$ and suppose $b_n={\tilde }b_n$ for $kp-p\leq n \leq
kp+2p$ and $a_n={\tilde }a_n$ for $kp-p\leq n\leq kp + 2p-1$. Then $$A_{k,J_0}(J)=A_{J_0}({\tilde }J) \qquad
B_{k,J_0}(J)=B_{J_0}({\tilde }J)$$
Immediate from Lemma \[L9.7\].
[L9.9]{} Let $k\leq \ell$ and $\alpha = [{\frac}12(p-(\ell-k))]$. For any two $J$ and ${\tilde }J$ and any $K$, there is $C_K$ so that $${\label}{9.26}
{\lvert\Delta_{J_0}(J)_{k\ell} -\Delta_{J_0}({\tilde }J)_{k\ell}\rvert} \leq
C_K \sup_{k-\alpha\leq j\leq \ell+\alpha} [{\lvertb_j-{\tilde }b_j\rvert} +
{\lverta_j-{\tilde }a_j\rvert}]$$ so long as $${\label}{9.27}
\sup [{\lvertb_j\rvert} + {\lvert{\tilde }b_j\rvert} + {\lverta_j\rvert} + {\lvert{\tilde }a_j\rvert}]\leq K$$
Immediate from Lemma \[L9.7\] and the fact that $\Delta_{J_0}$ has matrix elements that are fixed (given $J_0$) polynomials in $a$’s and $b$’s.
[l.new1]{} [(a)]{} For any Jacobi matrix, $J$, and $\ell = 1,2,\ldots$, $m
= 1,2,\ldots$, $${\label}{f.new1a}
(J^\ell)_{m \, m+l} = a_m a_{m+1} \cdots a_{m+\ell-1}$$ and for $\ell = 2,3,\ldots$, $m = 1,2,\ldots$, $${\label}{f.new1b}
(J^\ell)_{m \, m+\ell-1} = a_m \cdots a_{m+\ell-2} \Bigl(
\sum_{j=0}^{\ell-1} b_{m+j} \Bigr)$$ [(b)]{} For $J_0$ periodic of period $p \ge 2$ and $m =
1,2,\ldots$, $$\begin{aligned}
{\label}{f.new1c} \Delta_{J_0}(J)_{m \, m+p} & = \frac{a_m \cdots a_{m+p-1}}{a^{(0)}_m \cdots a^{(0)}_{m+p-1}} \\
{\label}{f.new1d} \Delta_{J_0}(J)_{m \, m+p-1} & = \bigl( a^{(0)}_m
\cdots a^{(0)}_{m+p-1} \bigr)^{-1} (a_m \cdots a_{m+p-2}) \Bigl(
\sum_{j=0}^{p-1} (b_{m+j} - b^{(0)}_{m+j}) \Bigr)\end{aligned}$$
\(a) Since $J$ changes index by at most one, $$(J^\ell)_{m \, m+\ell} = (J_{m \, m+1}) \cdots (J_{m+\ell-1 \,
m+\ell})$$ proving , while $$(J^\ell)_{m \, m+\ell-1} = \sum_{j=0}^{\ell-1} (J^j)_{m \, m+j} \,
J_{m+j \, m+j} \, (J^{\ell+j-1})_{m+j \, m+\ell-1}$$ which, given , proves .
\(b) By , $$\Delta_{J_0}(J) = \bigl( a^{(0)}_1 \cdots a^{(0)}_p \bigr)^{-1}
\left[ J^p - \sum_{j=0}^{p-1} b^{(0)}_{j+1} J^{p-1} + O(J^{p-2})
\right]$$ which, given (a), $(J^{p-k})_{m \, m+p} = (J^{p-k})_{m \, m+p-1} =
0$ if $k = 2,3,\ldots$, and the periodicity of $a^{(0)}$ and $b^{(0)}$ yields and .
[l.new2]{} Suppose that $\Delta_{J_0}(J) -S^p - S^{-p}$ is a Hilbert–Schmidt operator on $\ell^2(\{0,1,2,\ldots\})$. Then, $$\begin{aligned}
{\label}{f.new2a} \sum_n (a_n a_{n+1} \cdots a_{n+p-1} - a^{(0)}_n
a^{(0)}_{n+1} & \cdots a^{(0)}_{n+p-1})^2 < \infty \\
{\label}{f.new2b} \sum_n \left( \sum_{j=0}^{p-1} (b_{n+j} - b^{(0)}_{n+j}) \right)^2 & < \infty \\
{\label}{f.new2c} \sum_n (a_{n+p} - a_n)^2 < \infty & \\
{\label}{f.new2d} \sum_n (b_{n+p} - b_n)^2 < \infty &\end{aligned}$$
For a Hilbert–Schmidt operator, any subset of matrix elements lies in $\ell^2$, so by , $$\sum_n \left| a_n \cdots a_{n+p-1} \bigl( a^{(0)}_n \cdots
a^{(0)}_{n+p-1} \bigr)^{-1} - 1 \right|^2 < \infty$$ which, given that $a^{(0)}_n \cdots a^{(0)}_{n+p-1}$ is $n$-independent, implies .
Similarly, implies if we note that $\{a_j\}$ bounded and $a_n \cdots a_{n+p-1} \to a^{(0)}_1 \cdots
a^{(0)}_p > 0$ implies $\inf a_j > 0$, so $$\inf_m \bigl( a^{(0)}_m \cdots a^{(0)}_{m+p-1} \bigr)^{-1} (a_m
\cdots a_{m+p-2}) > 0$$
Since the difference of $\ell^2$ sequences is $\ell^2$, implies (since $a^{(0)}_n$ is periodic) $$\sum_n (a_{n+p} - a_n)^2 (a_{n+1} \cdots a_{n+p-1})^2 < \infty$$ which, given that $\inf a_j > 0$, implies .
Similarly, since $$\sum_{j=0}^{p-1} (b_{n+1+j} - b_{n+j}) = b_{n+p} - b_n$$ implies .
Our next preliminary is to relate $A\in{{\mathcal L}}$ to $${\label}{9.36}
{\lvertA\rvert} = \sqrt{\displaystyle A^\dagger_{ }A}$$
[P9.12]{} The map $A\mapsto{\lvertA\rvert}$ from ${{\mathcal L}}$ to positive definite matrices is a diffeomorphism. In particular, for $A$’s in ${{\mathcal L}}$ with $\|A-1\| <{\frac}12$, there exist constants $C_1$ and $C_2$ so that $${\label}{9.37}
C_1 \|A-1\|_2 \leq \|\,{\lvertA\rvert}-1\|_2 \leq C_2 \|A-1\|_2$$
By , $A\mapsto{\lvertA\rvert}$ is a smooth map.
The inverse map (strictly $|A|^2\mapsto A$) is known as the Cholesky factorization; see [@GvL; @Wat]. Given $B>0$, apply the Gram–Schmidt procedure to the (linearly independent) columns of $B$ working from right to left. This gives a factorization $B=QA$ with $Q$ unitary and $A\in{{\mathcal L}}$. Note that $|A|=B$ and that because the columns of $B$ are linearly dependent, $B\mapsto A$ is also a smooth map.
[T9.11]{} Let $J_0$ be a two-sided $p$-periodic Jacobi matrix with all gaps open and let $\Delta_{J_0}$ denote its discriminant. For a Jacobi matrix with parameters $(a_n,b_n)$, the following are equivalent:
$\Delta_{J_0}(J) -S^p - S^{-p}$ is a Hilbert–Schmidt operator on $\ell^2(\{0,1,2,\ldots\})$.
$\sum_n {\text{\rm{Tr}}}\{B_n^2 + |A_n-1|^2\} < \infty$.
$\sum_n {\text{\rm{Tr}}}\{B_n^2 + (|A_n|-1)^2\} < \infty$.
$\sum_n {\text{\rm{Tr}}}\{B_n^2 + G(|A_n|)\} < \infty$.
$\sum_m d_m((a,b),{{\mathcal T}}_{J_0})^2 < \infty$.
$\sum_m \tilde d_m((a,b),{{\mathcal T}}_{J_0})^2 < \infty$.
Here we have adopted the abbreviations $A_n:=A_{n,J_0}(J)$ and $B_n:=B_{n,J_0}(J)$
(i)$\Leftrightarrow$(ii) amounts to the definition of the Hilbert–Schmidt norm.
(ii)$\Leftrightarrow$(iii) follows from Proposition \[P9.12\].
(iii)$\Leftrightarrow$(iv) Notice that $G$, defined in , obeys $$c_{\varepsilon}(x-1)^2 \leq G(x) \leq c'_{\varepsilon}(x-1)^2 \qquad \forall\ x\in ({\varepsilon}, {\varepsilon}^{-1})$$ Applying this to the eigenvalues of $|A_n|$ yields this equivalence.
(v)$\Leftrightarrow$(vi) is the $q=2$ case of Proposition \[P2.5\].
(vi)$\Rightarrow$(i) By Lemma \[L9.7\], each matrix element of $\Delta_{J_0}(J) -S^p - S^{-p}$ is a smooth function of $p$ consecutive pairs $(a_n,b_n)$; moreover, by the magic formula, all of these smooth functions vanish if the corresponding $p$-tuple belongs to ${{\mathcal T}}_{J_0}$. The implication now follows from the fact that smooth functions are Lipschitz.
(i)$\Rightarrow$(vi) Define $J^{(k)}$ to be the $p$-periodic Jacobi matrix that equals $J$ on block $k$, that is, $${\label}{9.31}
b_\ell^{(k)} =b_{kp+\ell} \qquad a_\ell^{(k)}=a_{kp+\ell}$$ for $\ell=1,2,\dots, p$. Obviously, $J^{(k)}=J$ on block $k$ and, by and , the difference on blocks $k-1$ and $k+1$ are in $\ell^2$, that is, $${\label}{9.32}
\sum_k \Bigl[\, \sup_{(k-1)p\leq j\leq (k+2)p-1}\, {\lvertb_j-b_j^{(k)}\rvert} +
{\lverta_j -a_j^{(k)}\rvert}\;\Bigr]^2 <\infty$$ Together with Lemma \[L9.9\] and Proposition \[P9.6\], this implies $${\label}{f.tnewa}
\sum_k \tilde d_{kp} (J^{(k)} , {{\mathcal T}}_{J_0})^2 < \infty$$ On the other hand, implies that for $j = 1,\ldots,p$, $${\label}{f.tnewb}
\sum_k \tilde d_{kp+j} (J,J^{(k)})^2 < \infty$$ By the triangle inequality, $$\tilde d_{kp+j} (J , {{\mathcal T}}_{J_0})^2 \le 2 d_{kp+j} (J,J^{(k)})^2 +
2 \tilde d_{kp} (J^{(k)} , {{\mathcal T}}_{J_0})^2$$ so and imply (vi).
We will refer to the three statements (i)–(iii) of Theorem \[T1.4\] simply by their numbers.
Suppose first that $$\label{1.52again}
d_m\bigl( (a,b),{{\mathcal T}}_{J_0})\bigr)\in\ell^2$$ Then (i) holds by Theorem \[T1.1\]. Moreover, by Theorem \[T9.11\] and the hypothesis that all gaps are open, the RHS of is finite. Therefore, the LHS is finite. Next we use this fact to prove (ii) and (iii).
As $\Delta'$ is nonvanishing at all band edges, $${\label}{9.38}
\sum_j F(\Delta(E_j)) <\infty \iff \sum_{j=1}^N {\text{\rm{dist}}}\bigl(E_j,\sigma_{\text{\rm{ess}}}(J)\bigr)^{3/2} <\infty$$ which verifies (iii). By Corollary \[C9.2&3\], $${\label}{9.39}
\text{Leftmost term in \eqref{8.4}} <\infty \iff \text{(ii) holds}$$ This completes the proof of (i)–(iii).
Conversely, if (i)–(iii) hold, then by , , and , we see that the RHS of is finite. By Theorem \[T9.11\], this implies .
Let $\beta_j$ be the $\beta$’s associated to $\Delta(J)$, that is, ${\lvert\beta_j\rvert} >1$, $\beta_j + \beta_j^{-1} =E_j$ with $E_j$ the eigenvalues of $\Delta(J)$ in ${{\mathbb{R}}}\setminus [-2,2]$. Then $\log {\lvert\beta_j\rvert}\sim
{\lvert\beta_j\rvert}-1$ as $\beta\to\pm 1$ small and ${\lvert\beta_j\rvert}-1 \sim ({\lvertE_j\rvert}-2)^{1/2}$. Therefore, $${\label}{9.40}
\text{\eqref{8.6}} <\infty \iff \sum_j ({\lvertE_j\rvert}-2)^{1/2}
\iff \text{\eqref{1.49}}$$
By Corollary \[C9.2&3\], $${\label}{9.41}
\text{\eqref{8.5}} <\infty \iff \text{\eqref{1.50}}$$
Finally, if $\{A_n, B_n\}_{n=1}^\infty$ are the $p\times p$ blocks in $\Delta(J)$, we have $A_n =U{\lvertA_n\rvert}$ for some $U$ with ${\lvert\det(U)\rvert}=1$, so $${\label}{9.42}
\det{\lvertA_n\rvert} ={\lvert\det(A_n)\rvert} = \prod_{j=(n-1)p+1}^{np}
\prod_{k=j}^{j+p=1} \biggl[ {\frac}{a_k}{a_k^{(0)}}\biggr]$$ by and . Thus $$\sum_{n=1}^N \log (\det({\lvertA_n\rvert})) - p\sum_{k=1}^{Np} \log \biggl[{\frac}{a_k}{a_k^{(0)}}\biggr]$$ is bounded. Thus is equivalent to .
By Theorem \[T8.2\], we see that when holds, then $\Leftrightarrow$ , and if they hold, has a limit.
Moreover, if they hold, the hypotheses of Theorem \[T1.4\] hold, so is true. That holds is a theorem of Peherstorfer–Yuditskii [@PY]; see also the remark below.
In the remainder of this section, we will describe an alternate approach to proving ; one based on combining the magic formula with Theorem \[T5.8\]. Unfortunately, because of the strong hypothesis on the discrete spectrum that appears in this theorem, we will not recover the full formulation from Theorem \[T1.3\].
Let $\tilde {{\mathcal J}}$ denote the (unique) type 2 block Jacobi matrix that is equivalent to ${{\mathcal J}}=\Delta_{J_0}(J)$, which is of type 3. Further, let us use $A_j$ and ${\tilde }A_j$ to denote the off-diagonal block entries of ${{\mathcal J}}$ and ${\tilde }{{\mathcal J}}$, respectively.
If we strengthen the hypothesis to finiteness of the discrete spectrum (i.e., finiteness of the set $\sigma({{\mathcal J}})\setminus\sigma_{\text{\rm{ess}}}({{\mathcal J}})$), then Theorem \[T5.8\] shows that implies the convergence of the product ${\tilde }A_1 \cdots {\tilde }A_n$ as $n\to\infty$. In view of and Proposition \[P9.12\], this convergence is inherited by the product $A_1 \cdots A_n$. Thus, it remains only to connect the convergence of this matrix product to the behavior of the sequences of parameters. This is the job of the next lemma. In the original version of this paper, it was only proved that the sequence $\{a_n\}$ was asymptotic to a fixed periodic sequence. The argument for the sequence $\{b_n\}$ was provided by one of the referees; we are most grateful for this.
Let $J_0$ be a $p$-periodic two-sided Jacobi matrix and let $\Delta=\Delta_{J_0}$ denote its discriminant. Let $J$ be a one-sided Jacobi matrix with parameters $\{a_n,b_n\}$. Suppose the product $A_n\cdots A_1$ converges to a non-singular matrix as $n\to\infty$. Here $A_n$ and $B_n$ denote the $p\times p$ block entries of $\Delta(J)$. Then the parameters of $J$ asymptotically converge to fixed periodic parameters in the sense of .
By applying the same affine transformation (i.e., $x\mapsto \alpha x+\beta$) to both $J$ and $J_0$, we may assume that the discriminant of $J_0$ takes the form $\Delta(x)=x^p+O(x^{p-2})$. This will significantly simplify some of the formulae that follow. Note also that this transformation makes $a^{(0)}_1\cdots a^{(0)}_p=1$ and $b^{(0)}_1 + \cdots + b^{(0)}_p=0$.
Let $A_n$ and $B_n$ denote the $p\times p$ block entries of $\Delta(J)$. Then $$\begin{aligned}
(A_n)_{k,k} &= a_{p(n-1)+k} \cdots a_{pn+k-1} \\
(A_n)_{k+1,k} &= a_{p(n-1)+k+1} \cdots a_{pn+k-1} \bigl[b_{p(n-1)+k+1} + \cdots + b_{pn+k}\bigr]\end{aligned}$$ as can be read off from Lemma \[l.new1\]. Using this and the lower-triangular structure of the matrices $A_j$, one may quickly deduce $$\begin{aligned}
(A_1\cdots A_n)_{k,k} &= \prod_{j=k}^{pn+k-1} a_j \label{AprodDiag}\\
(A_1\cdots A_n)_{k+1,k}
&= \sum_{r=1}^{n} (A_1)_{k+1,k+1} \cdots (A_{r-1})_{k+1,k+1} (A_r)_{k+1,k} (A_{r+1})_{k,k} \cdots (A_n)_{k,k} \notag\\
&= \Biggl( \prod_{j=k+1}^{pn+k-1} a_j \Biggr) \sum_{j=k+1}^{pn+k} b_j \label{AprodOffDiag}\end{aligned}$$
To see that the sequence $n\mapsto a_{pn+k}$ converges for each fixed $k\in\{1,\ldots,p\}$, one need only take ratios of for consecutive values of $k$ and the same $n$ (and also for $(n,k=p)$ and $(n+1,k=1)$), then send $n\to\infty$.
For the parameters $b_n$, one may proceed in a similar fashion: For example, when $2\leq k \leq p-1$, the fact that $$a_{k} \frac{(A_1\cdots A_n)_{k+1,k}}{(A_1\cdots A_n)_{k,k}} - a_{k-1} \frac{(A_1\cdots A_n)_{k,k-1}}{(A_1\cdots A_n)_{k-1,k-1}}
= b_{pn+k} - b_{k}$$ shows us that $b_{pn+k}$ converges as $n\to\infty$.
Szegő and Killip–Simon Theorems When Some Gaps Are Closed
=========================================================
[s10]{} Here we want to examine what might replace Theorems \[T1.3\] and \[T1.4\] if $J_0$ is periodic but with some closed gaps. The Szegő-type theorem is almost the same as Theorem \[T1.3\]:
[T10.1]{} Let $J_0$ be any two-sided periodic Jacobi matrix with Jacobi parameters $\{a_n^{(0)}, b_n^{(0)}\}_{n=-\infty}^\infty$, and $J$ a one-sided Jacobi matrix with Jacobi parameters $\{a_n, b_n\}_{n=1}^\infty$ and spectral measure $d{\eta}$. Suppose that holds, and that $${\label}{10.1}
\sum_{m=1}^N {\text{\rm{dist}}}(E_m,\sigma_{\text{\rm{ess}}}(J))^{1/2} <\infty$$ if $\{E_m\}_{m=1}^N$ is a labelling of the eigenvalues of $J$ outside $\sigma_{\text{\rm{ess}}}(J)$. Then $${\label}{10.2}
-\int_{\sigma_{\text{\rm{ess}}}(J_0)} \log \biggl( {\frac}{d{\eta}_{\text{\rm{ac}}}}{dx}\biggr) {\text{\rm{dist}}}(x, {{\mathbb{R}}}\setminus
\sigma_{\text{\rm{ess}}}(J_0))^{-1/2}\, dx <\infty$$ implies $${\label}{10.3}
\lim\biggl(\, \sum_{j=1}^{pN} \log \biggl( {\frac}{a_j}{a_j^{(0)}}\biggr)\biggr)$$ exists and lies in $(-\infty,\infty)$. Conversely, holds so long as $${\label}{10.4}
\limsup \biggl(\, \sum_{j=1}^N \log \biggl( {\frac}{a_j}{a_j^{(0)}}\biggr)\biggr) >-\infty$$ and then the limit in exists and lies in $(-\infty,\infty)$.
Moreover, if or holds, then there is $J_1\in{{\mathcal T}}_{J_0}$, so $${\label}{10.5}
d_m (J,J_1)\to 0$$
All that is missing is which we do not claim. However, since / imply (i)–(iii) of Theorem \[T10.3\] below, we have .
As noted, even with closed gaps, is equivalent to (see Corollary \[C9.2&3\]). Once one notes this, the proof of Theorem \[T1.3\] provides all the results stated as Theorem \[T10.1\].
Theorem \[T1.4\] used open gaps in two ways. First, in the translation of a matrix pseudo-Szegő condition, with $\alpha={\frac}12$ to the original spectral measure of $J$, and second, translating a Hilbert–Schmidt bound on $\Delta (J)-S^p-S^{-p}$ to $\ell^2$ approach to the isospectral torus. The second issue can be finessed if we leave things as a Hilbert–Schmidt condition, which reduces to a sum of translates of an explicit positive polynomial in the $a_n$’s and $b_n$’s being finite. As for translating with $\alpha={\frac}12$, the argument that proved Corollary \[C9.2&3\] translates immediately to
[L10.2]{} Suppose $\sigma(J_0)$ has closed gaps at $\{y_j\}_{j=1}^\ell
\subset\sigma(J_0)$. Then holds with $\alpha={\frac}12$ if and only if $${\label}{10.6}
\int_{\sigma_{\text{\rm{ess}}}(J_0)} {\text{\rm{dist}}}(x,{{\mathbb{R}}}\setminus\sigma (J_0))^{1/2}
\prod_{j=1}^\ell {\lvertx-y_j\rvert}^2 {\lvert\log (\omega(x))\rvert}\, dx <\infty$$
Plugging this into our proof of Theorem \[T1.4\] immediately yields
[T10.3]{} Let $J_0$ be a two-sided periodic Jacobi matrix with closed gaps at $\{y_j\}_{j=1}^\ell \subset \sigma(J_0)$ and let $J$ be a Jacobi matrix. Then $${\label}{10.7}
{\text{\rm{Tr}}}((\Delta_{J_0}(J)-S^p-S^{-p})^2) <\infty$$ if and only if
holds.
holds with ${\frac}12$ replaced by ${\frac}32$.
holds.
[E10.4]{} Take $J_0$ to be the two-sided free Jacobi matrix but think of it as period $2$. Then $$\Delta_J(x)=x^2-2$$ and a direct calculation of $J^2 -S^2-S^{-2}$ shows that is equivalent to the three conditions $$\begin{aligned}
\sum_n (a_n^2 + b_n^2 + a_{n+1}^2 -2)^2 &<\infty {\label}{10.8} \\
\sum_n (a_{n+1} (b_n + b_{n+1}))^2 &< \infty {\label}{10.9} \\
\sum_n (a_n a_{n+1}-1)^2 &< \infty {\label}{10.10}\end{aligned}$$ If $b_n=0$ and $${\label}{10.11}
a_n=1 + (-1)^n (n+1)^{-\beta}$$ then – hold if and only if $\beta >{\frac}14$ while, of course, requires $\beta >{\frac}12$. This is one of many known extensions of the $(J_0$-free) Killip–Simon theorem (see, e.g., Laptev et al. [@LNS2003], Kupin [@Kup2003], and Nazarov et al. [@NPVY]). Some of these results have MOPRL analogs which, via the magic formula, lead to variants of Theorems \[T1.4\] and \[T10.3\].
Eigenvalue Bounds for MOPRL
===========================
[s11]{} There are Birman–Schwinger kernels for MOPRL and it should be possible to extend the proofs of most bounds on the number of eigenvalues outside $[-2,2]$ or on moments of ${\lvertE_j\rvert}-2$ from the scalar to matrix case with optimal constants. But if one is willing to settle for less than optimal constants (but still not awful constants), there is a simple method to go from the scalar to matrix case. It depends on the following:
[T11.1]{} Let ${{\mathcal J}}$ be an $\ell\times\ell$ block Jacobi matrix in the Nevai class with Jacobi parameters $\{A_n, B_n\}_{n=1}^\infty$. Let $E_j^\pm ({{\mathcal J}})$ denote its eigenvalues counting multiplicity outside $[-2,2]$, that is, $E_1^+\geq E_2^+
\geq \cdots > 2 > - 2 > \cdots\geq E_2^- \geq E_1^-$. Let $J_\pm$ be the scalar Jacobi matrix with $a_n\equiv 1$ and $${\label}{11.1}
b_n^\pm =\pm \|B_n\| \pm \|A_{n-1}-1\| \pm \|A_n-1\|$$ and let $J_\pm^{(\ell)}$ be an $\ell$-fold direct sum of $J_\pm$. Then $${\label}{11.2}
{\lvertE_j^\pm ({{\mathcal J}})\rvert} \leq {\lvertE_j^\pm (J_\pm^{(\ell)})\rvert}$$
The matrix analog of the observation of Hundertmark–Simon [@HunS] extended to $2\ell\times 2\ell$ matrices (with $\ell\times\ell$ blocks) says that $${\label}{11.2x}
\begin{pmatrix}
-{\lvertA_n -{{\boldsymbol{1}}}\rvert} & {{\boldsymbol{1}}}\\
{{\boldsymbol{1}}}& -{\lvertA_n-{{\boldsymbol{1}}}\rvert}
\end{pmatrix}
\leq
\begin{pmatrix}
0 & A_n \\
A_n^\dagger & 0
\end{pmatrix}
\leq
\begin{pmatrix}
{\lvertA_n -{{\boldsymbol{1}}}\rvert} & {{\boldsymbol{1}}}\\
{{\boldsymbol{1}}}& {\lvertA_n -{{\boldsymbol{1}}}\rvert}
\end{pmatrix}$$ since $${\label}{11.3}
\left\| \begin{pmatrix}
0 & C \\
C^\dagger & 0
\end{pmatrix} \right\|^2 =
\left\| \begin{pmatrix}
C^\dagger C & 0 \\
0 & CC^\dagger
\end{pmatrix} \right\| = \|C\|^2$$ Thus writing $J_\pm^{(\ell)}$ as $\ell\times\ell$ blocks with each block a multiple of ${{\boldsymbol{1}}}$, $${\label}{11.4}
J_-^{(\ell)} \leq {{\mathcal J}}\leq J_+^{(\ell)}$$ from which is immediate.
[C11.2]{} For any block Jacobi matrix, ${{\mathcal J}}$, in Nevai class, $${\label}{11.5}
\sum_{j,\pm}\, (E_j^\pm ({{\mathcal J}})^2-4)^{1/2} \leq 2\ell \sum_n\, \|B_n\|
+ 4\ell \sum_n\, \|A_n-1\|$$
In particular, this implies if the RHS of is finite, so is the LHS.
Hundertmark–Simon [@HunS] proved $${\label}{11.6}
\sum_j\, (E_j^\pm (J_\pm)^2 -4)^{1/2} \leq \sum_n b_n^\pm$$ from which follows by .
The Analog of Nevai’s Conjecture
================================
[s12]{}
By , $J-J_0$ is trace class. Thus $J^\ell -J_0^\ell =\sum_{k=0}^{\ell-1} J^k (J-J_0) J^{\ell-1-k}$ is trace class, so $\Delta_{J_0}(J)-\Delta_{J_0}(J_0)=\Delta_{J_0}(J) -(S^p + S^{-p})$ is trace class.
It follows that if holds and $\Delta(J)$ has matrix Jacobi parameters $\{A_n,B_n\}_{n=1}^\infty$ that $${\label}{12.1}
\sum_{n=1}^\infty\, \|B_n\| + \sum_{n=1}^\infty \, \|A_n-1\|<\infty$$
By Corollary \[C11.2\], the eigenvalues $\Delta(J)$ obey $${\label}{12.2}
\sum_{j=1}^\infty\, ({\lvertE_j^\pm\rvert}-2)^{1/2} <\infty$$ also implies $${\label}{12.3}
\sum_{n=1}^\infty\, {\lvert\log (\det{\lvertA_n\rvert})\rvert} <\infty$$ We can apply Theorem \[T8.2\] and conclude that $Z(J)$ is finite, that is, $${\label}{12.4}
\int (4-E^2)^{-1/2} \log (\det(W(E)))\, dE >-\infty$$ By Corollary \[C9.2&3\], we obtain .
Perturbations of Periodic OPUC
==============================
[s13]{} In this final section, we want to present the translations of our results to OPUC. Since the magic formula maps periodic OPUC to MOPRL, the changes needed in the proofs will be minor, although for the analog of Theorem \[T1.4\], there is one significant change. It is interesting to note the sequence of mappings for the OPUC periodic Rakhmanov’s theorem. We map OPUC to MOPRL using the magic formula and them map that to MOPUC using the Szegő map.
The OPUC version of Theorem \[T1.1\] is already in Last–Simon [@LastS]. As for Theorem \[T1.2\]:
[T13.1]{} Let ${{\mathcal{C}}}_0$ be a two-sided periodic CMV matrix. Let ${{\mathcal{C}}}$ be an ordinary CMV matrix with Verblunsky coefficients $\{\alpha_n\}_{n=0}^\infty$. Suppose $${\label}{13.1}
\Sigma_{\text{\rm{ac}}}({{\mathcal{C}}}) = \sigma_{\text{\rm{ess}}}({{\mathcal{C}}}_0)$$ Then, as $m\to\infty$, $${\label}{13.2}
d_m (\alpha,{{\mathcal T}}_{{{\mathcal{C}}}_0}) \to 0$$
Let ${{\mathcal{C}}}_r$ be a right limit of ${{\mathcal{C}}}$. By Theorem \[T6.1\], $\Delta ({{\mathcal{C}}}_r)=S^p + S^{-p}$. Thus, by Theorem \[T3.1\], ${{\mathcal{C}}}_r\in{{\mathcal T}}_{{{\mathcal{C}}}_0}$.
As for the analog of Theorem \[T1.3\], if we drop discussion of $\ell^2$ convergence, it holds, similar to Theorem \[T10.1\].
[T13.2]{} Let ${{\mathcal{C}}}_0$ be a two-sided periodic CMV matrix with Verblunsky coefficients $\{\alpha_j^{(0)}\}_{j=-\infty}^\infty$, and ${{\mathcal{C}}}$ a one-sided CMV matrix with Verblunsky coefficients $\{\alpha_j\}_{j=0}^\infty$ and spectral measure $d\mu$. Suppose that $\sigma_{\text{\rm{ess}}}({{\mathcal{C}}})=\sigma({{\mathcal{C}}}_0)$ and $${\label}{13.3}
\sum_{m=1}^N {\text{\rm{dist}}}(E_m,\sigma_{\text{\rm{ess}}}({{\mathcal{C}}}))^{1/2} <\infty$$ where $\{E_m\}_{m=1}^N$ is a labelling of the eigenvalues of ${{\mathcal{C}}}$ outside $\sigma_{\text{\rm{ess}}}({{\mathcal{C}}}_0)$. Then $${\label}{13.4}
-\int_{\sigma_{\text{\rm{ess}}}({{\mathcal{C}}}_0)} \log \biggl( {\frac}{d\mu_{\text{\rm{ac}}}}{d\theta}\biggr)
{\text{\rm{dist}}}(\theta, {{\mathbb{R}}}\setminus\sigma_{\text{\rm{ess}}}({{\mathcal{C}}}_0))^{-1/2}\, {\frac}{d\theta}{2\pi} <\infty$$ implies $${\label}{13.5}
\lim_{N\to\infty}\biggl(\, \sum_{j=1}^{pN} \log \biggl( {\frac}{\rho_j}{\rho_j^{(0)}}\biggr)\biggr)$$ exists and lies in $(-\infty,\infty)$. Conversely, holds so long as $${\label}{13.6}
\limsup_{N\to\infty} \biggl(\, \sum_{j=1}^{N} \log \biggl( {\frac}{\rho_j}{\rho_j^{(0)}}\biggr)\biggr) >-\infty$$ and then the limit in exists and lies in $(-\infty,\infty)$.
1\. We have not stated that the $\alpha_n$ have a limit in ${{\mathcal T}}_{{{\mathcal{C}}}_0}$. We suspect that the methods of [@PY] extend to the OPUC but have not checked this and they do not explicitly mention it.
2\. Of course, if this theorem is applicable and ${{\mathcal{C}}}_0$ obeys the conditions of Theorem \[T13.3\] below, then we have a result on $\ell^2$ convergence to ${{\mathcal T}}_{{{\mathcal{C}}}_0}$.
3\. One can replace $\rho_j^{(0)}$ by the logarithmic capacity of $\sigma({{\mathcal{C}}}_0)$.
At open gaps, $\Delta'(e^{i\theta})\neq 0$, so is equivalent to $${\label}{13.7}
\sum_{\substack{ E\notin [-2,2] \\ E\in\sigma(\Delta({{\mathcal{C}}}))}}\, ({\lvertE\rvert}-2)^{1/2} <\infty$$ Moreover, by –, we have that $${\label}{13.8}
\log(\det{\lvertA_n\rvert}) = \log (\det(A_n)) = \log \biggl[\,\prod_{j=(n-1)p+1}^{np}
\prod_{k=j}^{j+p^{-1}} {\frac}{\rho_k}{\rho_k^{(0)}}\biggr]$$ Now just follow the proof of Theorem \[T10.1\].
In carrying over Theorem \[T1.4\] to OPUC, one runs into a serious roadblock: van Moerbeke’s theorem [@vMoer] that the Hamiltonian flows generated by the coefficients of the $t_j$ (given by ) are independent is not known for OPUC. Instead, we use a weaker result of Simon [@OPUC2 Section 11.10] that proves the derivatives of coefficients of $\Delta$ span the normal bundle for a dense open set that is ‘most’ of the points with all gaps open. We will call the isospectral tori in this dense open set the generic independence tori. Then by mimicking the arguments in Section \[s9\], we obtain
[T13.3]{} Let ${{\mathcal{C}}}_0$ be a two-sided CMV matrix in a generic independence torus with Verblunsky coefficients $\{\alpha_j^{(0)}\}_{j=-\infty}^\infty$. Let ${{\mathcal{C}}}$ be a CMV matrix with Verblunsky coefficients $\{\alpha_j\}_{j=0}^\infty$. Then $${\label}{13.9}
\sum_{m=0}^\infty d_m (\alpha, {{\mathcal T}}_{{{\mathcal{C}}}_0})^2 <\infty$$ if and only if
$\sigma_{\text{\rm{ess}}}({{\mathcal{C}}})=\sigma({{\mathcal{C}}}_0)$
For the eigenvalues $\{E_j\}_{j=1}^N$ not in $\sigma({{\mathcal{C}}}_0)$, $${\label}{13.10}
\sum_{j=1}^N {\text{\rm{dist}}}(E_j, \sigma ({{\mathcal{C}}}_0))^{3/2} <\infty$$
If $\mu$ is the spectral measure for ${{\mathcal{C}}}$, then $${\label}{13.11}
-\int_{\sigma({{\mathcal{C}}}_0)} \log \biggl( {\frac}{d\mu_{\text{\rm{ac}}}}{d\theta}\biggr)
{\text{\rm{dist}}}(\theta,\partial{{\mathbb{D}}}\setminus \sigma_{\text{\rm{ess}}}({{\mathcal{C}}}_0))^{1/2}\, {\frac}{d\theta}{2\pi} <\infty$$
Our last result is a periodic OPUC version of Nevai’s conjecture.
[T13.4]{} Let ${{\mathcal{C}}}_0$ be a two-sided $p$-periodic CMV matrix and let ${{\mathcal{C}}}$ be a CMV matrix with Verblunsky coefficients $\{\alpha_j\}_{j=0}^\infty$. Then $${\label}{13.12}
\sum_{m=0}^\infty d_m (\alpha, {{\mathcal T}}_{{{\mathcal{C}}}_0}) <\infty$$ implies
All the above results assume the period $p$ is even. However, by using sieving (see Example 1.6.14 of [@OPUC1]) to map period $p$ to period $2p$, one can extend Theorems \[T13.1\] and \[T13.2\] to $p$ odd. In particular, we obtain the $p=1$ results of [@BHLL; @ABMV; @BCL] as very special cases of ours.
[100]{} [AK]{} N. I. Akhiezer and M. Krein, *Some Questions in the Theory of Moments*, Transl. Math. Monographs, Vol. 2, American Mathematical Society, Providence, R.I., 1962; Russian original, 1938.
[ABMV]{} M. P. Alfaro, M. Bello Hernández, J. M. Montaner, and J. L. Varona, [*Some asymptotic properties for orthogonal polynomials with respect to varying measures*]{}, J. Approx. Theory [**135**]{} (2005), 22–34.
[AG94]{} D. Alpay and I. Gohberg, [*Inverse spectral problem for difference operators with rational scattering matrix function*]{}, Integral Eqs. Oper. Theory [**20**]{} (1994), 125–170.
[AG95]{} D. Alpay and I. Gohberg, [*Inverse spectral problem for differential operators with rational scattering matrix functions*]{}, J. Diff. Eqs. [**118**]{} (1995), 1–19.
[AntK]{} A. J. Antony and M. Krishna, [*Almost periodicity of some Jacobi matrices*]{}, Proc. Indian Acad. Sci. Math. Sci. [**102**]{} (1992), 175–188.
[AN84]{} A. I. Aptekarev and E. M. Nikishin, [*The scattering problem for a discrete Sturm–Liouville operator*]{}, Math USSR Sb. [**49**]{} (1984), 325–355; Russian original in Mat. Sb. (N.S.) [**121(163)**]{} (1983), 327–358.
[S147]{} J. Avron and B. Simon, [*Almost periodic Schrödinger operators. I. Limit periodic potentials*]{}, Comm. Math. Phys. [**82**]{} (1981/82), 101–120.
[BakCon]{} M. Bakonyi and T. Constantinescu, *Schur’s Algorithm and Several Applications*, Pitman Research Notes in Math. [**261**]{}, Longman, Essex, U.K., 1992.
[BCL]{} D. Barrios Rolanía, B. de la Calle Ysern, and G. López Lagomasino, [*Ratio asymptotics of polynomials orthogonal with respect to varying Denisov-type measures*]{}, J. Approx. Theory [**139**]{} (2006), 223–256.
[BHLL]{} M. Bello Hernández and G. López Lagomasino, [*Ratio and relative asymptotics of polynomials orthogonal on an arc of the unit circle*]{}, J. Approx. Theory [**92**]{} (1998), 216–244.
[BT92a]{} S. V. Belyi and E. R. Tsekanovskii, [*Classes of operator R-functions and their realization by conservative systems*]{}, Sov. Math. Dokl. [**44**]{} (1992), 692–696.
[BT92b]{} S. V. Belyi and E. R. Tsekanovskii, [*Realization theorems for operator-valued R-functions*]{}, The Israel M. Glazman Memorial Volume (I. Gohberg and Yu. Lyubich, eds.), pp. 55–91, Operator Theory: Advances and Applications, [**98**]{}, Birkhäuser, Basel, 1992.
[Blu]{} O. Blumenthal, [*Über die Entwicklung einer willkürlichen Funktion nach den Nennern des Kettenbruches für $\int_{-\infty}^0
\frac{\varphi (\xi)\, d\xi}{z-\xi}$*]{}, Ph.D. dissertation, Göttingen 1898.
[BGHT]{} W. Bulla, F. Gesztesy, H. Holden, and G. Teschl, [*Algebro-geometric quasi-periodic finite-gap solutions of the Toda and Kac–van Moerbeke hierarchies*]{}, Mem. Amer. Math. Soc. [**135**]{} (1998), no. 641.
[BuFi]{} M. Buys and A. Finkel, [*The inverse periodic problem for Hill’s equation with a finite-gap potential*]{}, J. Diff. Eqs. [**55**]{} (1984), 257–275.
[CFMV03]{} M. J. Cantero, M. P. Ferrer, L. Moral, and L. Velázquez, [*A connection between orthogonal polynomials on the unit circle and matrix orthogonal polynomials on the real line*]{}, J. Comput. Appl. Math. [**154**]{} (2003), 247–272.
[Case1]{} K. M. Case, [*Orthogonal polynomials from the viewpoint of scattering theory*]{}, J. Math. Phys. [**15**]{} (1974), 2166–2174.
[Case2]{} K. M. Case, [*Orthogonal polynomials, II*]{}, J. Math. Phys. [**16**]{} (1975), 1435–1440.
[Chu]{} V. A. Chulaevsky, [*An inverse spectral problem for limit-periodic Schrödinger operators*]{}, Functional Anal. Appl. [**18**]{} (1984), 220–233; Russian original in Funktsional. Anal. i Prilozhen. [**18**]{} (1984), 63–66.
[CodLev]{} E. A. Coddington and N. Levinson, *Theory of Ordinary Differential Equations*, McGraw-Hill, New York-Toronto-London, 1955.
[DPSprep]{} D. Damanik, A. Pushnitski, and B. Simon, [*Matrix orthogonal polynomials*]{}, Surv. Approx. Theory [**4**]{} (2008), 1–85.
[Jost1]{} D. Damanik and B. Simon, [*Jost functions and Jost solutions for Jacobi matrices, I. A necessary and sufficient condition for Szegő asymptotics*]{}, Invent. Math. [**165**]{} (2006), 1–50.
[Jost2]{} D. Damanik and B. Simon, [*Jost functions and Jost solutions for Jacobi matrices, II. Decay and analyticity*]{}, Int. Math. Res. Not. [**2006**]{}, Article ID 19396, 32 pages, 2006.
[DJLP]{} E. Defez, L. Jódar, A. Law, and E. Ponsoda, [*Three-term recurrences and matrix orthogonal polynomials*]{}, Util. Math. [**57**]{} (2000), 129–146.
[DG92]{} P. Delsarte and Y. V. Genin, [*On a generalization of the Szegő–Levinson recurrence and its application in lossless inverse scattering*]{}, IEEE Trans. Inform. Theory [**38**]{} (1992), 104–110.
[DGK78]{} P. Delsarte, Y. V. Genin, and Y. G. Kamp, [*Orthogonal polynomial matrices on the unit circle*]{}, IEEE Trans. Circuits and Systems [**CAS-25**]{} (1978), 149–160.
[DGK3]{} P. Delsarte, Y. V. Genin, and Y. G. Kamp, [*Planar least squares inverse polynomials, I. Algebraic properties*]{}, IEEE Trans. Circuits and Systems [**CAS-26**]{} (1979), 59–66.
[DGK2]{} P. Delsarte, Y. V. Genin, and Y. G. Kamp, [*Schur parametrization of positive definite block-Toeplitz systems*]{}, SIAM J. Appl. Math. [**36**]{} (1979), 34–46.
[DGK79]{} P. Delsarte, Y. V. Genin, and Y. G. Kamp, [*The Nevanlinna–Pick problem for matrix-valued functions*]{}, SIAM J. Appl. Math. [**36**]{} (1979), 47–61.
[DGK81]{} P. Delsarte, Y. V. Genin, and Y. G. Kamp, [*Generalized Schur representation of matrix-valued functions*]{}, SIAM J. Algebraic Discrete Methods [**2**]{} (1981), 94–107.
[DenPAMS]{} S. A. Denisov, [*On Rakhmanov’s theorem for Jacobi matrices*]{}, Proc. Amer. Math. Soc. [**132**]{} (2004), 847–852.
[DS2002]{} H. Dette and W. J. Studden, [*Matrix measures, moment spaces and Favard’s theorem for the interval $[0,1]$ and $[0,\infty)$*]{}, Linear Algebra Appl. [**345**]{} (2002), 169–193.
[DS02]{} H. Dette and W. J. Studden, [*A note on the matrix valued q-d algorithm and matrix orthogonal polynomials on $[0,1]$ and $[0,\infty)$*]{}, J. Comput. Appl. Math. [**148**]{} (2002), 349–361.
[Dur99]{} A. J. Duran, [*Ratio asymptotics for orthogonal matrix polynomials*]{}, J. Approx. Theory [**100**]{} (1999), 304–344.
[DD02]{} A. J. Durán and E. Defez, [*Orthogonal matrix polynomials and quadrature formulas*]{}, Linear Algebra Appl. [**345**]{} (2002), 71–84.
[DLR96]{} A. J. Durán and P. López-Rodríguez, [*Orthogonal matrix polynomials: Zeros and Blumenthal’s theorem*]{}, J. Approx. Theory [**84**]{} (1996), 96–118.
[DLR01]{} A. J. Durán and P. López-Rodríguez, [*The matrix moment problem*]{}, Margarita Mathematica, pp. 333–348, Univ. La Rioja, Logroño, 2001.
[DLR04]{} A. J. Durán and P. López-Rodríguez, [*Orthogonal matrix polynomials*]{}, Laredo Lectures on Orthogonal Polynomials and Special Functions, pp. 13–44, Adv. Theory Spec. Funct. Orthogonal Polynomials, Nova Sci. Publ., Hauppauge, NY, 2004.
[DLRS99]{} A. J. Durán, P. López-Rodríguez, and E. B. Saff, [*Zero asymptotic behaviour for orthogonal matrix polynomials*]{}, J. Anal. Math. [**78**]{} (1999), 37–60.
[East]{} M. S. P. Eastham, *The Spectral Theory of Periodic Differential Equations*, Scottish Academic Press, Edinburgh, 1975.
[FIT]{} A. Finkel, E. Isaacson, and E. Trubowitz, [*An explicit solution of the inverse periodic problem for Hill’s equation*]{}, SIAM J. Math. Anal. [**18**]{} (1987), 46–53.
[Fu76]{} M. Fukushima, [*A spectral representation of ordinary linear difference equation with operator valued coefficients of the second order*]{}, J. Math. Phys. [**17**]{} (1976), 1064–1072.
[GI1]{} V. Georgescu and A. Iftimovici, [*Crossed products of $C^*$-algebras and spectral analysis of quantum Hamiltonians*]{}, Comm. Math. Phys. [**228**]{} (2002), 519–560.
[Ger81]{} J. S. Geronimo, [*Matrix orthogonal polynomials on the unit circle*]{}, J. Math. Phys. [**22**]{} (1981), 1359–1365.
[GBk1]{} Ya. L. Geronimus, *Polynomials Orthogonal on a Circle and Their Applications*, Amer. Math. Soc. Transl. [**1954**]{} (1954), no. 104, 79 pp.
[Ger77]{} Ya. L. Geronimus, [*Orthogonal polynomials*]{}, Engl. translation of the appendix to the Russian translation of Szegő’s book in “Two Papers on Special Functions," Amer. Math. Soc. Transl., Ser. 2, Vol. 108, pp. 37–130, American Mathematical Society, Providence, RI, 1977.
[GShpa]{} F. Gesztesy and B. Simon, [*Inverse spectral analysis with partial information on the potential, I. The case of an a.c. component in the spectrum*]{}, Helv. Phys. Acta [**70**]{} (1997), 66–71.
[GStams]{} F. Gesztesy and B. Simon, [*Inverse spectral analysis with partial information on the potential, II. The case of discrete spectrum*]{}, Trans. Amer. Math. Soc. [**352**]{} (2000), 2765–2787.
[GT2000]{} F. Gesztesy and E. Tsekanovskii, [*On matrix-valued Herglotz functions*]{}, Math. Nachr. [**218**]{} (2000), 61–138.
[GKS98]{} I. Gohberg, M. A. Kaashoek, and A. L. Sakhnovich, [*Pseudo-canonical systems with rational Weyl functions: explicit formulas and applications*]{}, J. Diff. Eqs. [**146**]{} (1998), 375–398.
[GSprep]{} L. Golinskii and B. Simon, Section 4.3 of [@OPUC1].
[GvL]{} G. H. Golub and C. F. Van Loan, *Matrix Computations*, third ed., Johns Hopkins Studies in the Mathematical Sciences, Johns Hopkins University Press, Baltimore, 1996.
[HS81]{} D. B. Hinton and J. K. Shaw, [*On Titchmarsh–Weyl $M(\lambda)$-functions for linear Hamiltonian systems*]{}, J. Diff. Eqs. [**40**]{} (1981), 316–342.
[HL]{} H. Hochstadt and B. Lieberman, [*An inverse Sturm–Liouville problem with mixed given data*]{}, SIAM J. Appl. Math. [**34**]{} (1978), 676–680.
[HunS]{} D. Hundertmark and B. Simon, [*Lieb–Thirring inequalities for Jacobi matrices*]{} J. Approx. Theory [**118**]{} (2002), 106–130.
[Iwa]{} K. Iwasaki, [*Inverse problem for Sturm–Liouville and Hill’s equations*]{}, Ann. Mat. Pura Appl. (4) [**149**]{} (1987), 185–206.
[JDP96]{} L. Jódar, E. Defez, and E. Ponsoda, [*Orthogonal matrix polynomials with respect to linear matrix moment functionals: theory and applications*]{}, Approx. Theory Appl. (N.S.) [**12**]{} (1996), 96–115.
[Jo87]{} R. A. Johnson, [*$m$-functions and Floquet exponents for linear differential systems*]{}, Ann. Mat. Pura Appl., Ser. 4 [**147**]{} (1987), 211–248.
[Kil02]{} R. Killip, [*Perturbations of one-dimensional Schrödinger operators preserving the absolutely continuous spectrum*]{}, Int. Math. Res. Not. [**(2002)**]{}, no. 38, 2029–2061.
[Kil-SFest]{} R. Killip, [*Spectral theory via sum rules*]{}, Spectral Theory and Mathematical Physics: A Festschrift in Honor of Barry Simon’s 60th Birthday (F. Gesztesy et al., eds.), pp. 907–930, American Mathematical Society, Providence, RI, 2007.
[KS]{} R. Killip and B. Simon, [*Sum rules for Jacobi matrices and their applications to spectral theory*]{}, Ann. of Math. (2) [**158**]{} (2003), 253–321.
[Kot]{} S. Kotani, [*Ljapunov indices determine absolutely continuous spectra of stationary random one-dimensional Schrödinger operators*]{}, Stochastic Analysis (Katata/Kyoto, 1982), pp. 225–247, North–Holland Math. Library, 32, North–Holland, Amsterdam, 1984.
[Kot97]{} S. Kotani, [*Generalized Floquet theory for stationary Schrödinger operators in one dimension*]{}, Chaos Solitons Fractals [**8**]{} (1997), 1817–1854.
[Kr89]{} A. M. Krall, [*$M(\lambda)$ theory for singular Hamiltonian systems with one singular point*]{}, SIAM J. Math. Anal. [**20**]{} (1989), 664–700.
[KO78]{} M. G. Krein and I. E. Ovčarenko, [*Inverse problems for Q-functions and resolvent matrices of positive Hermitian operators*]{}, Soviet Math. Dokl. [**19**]{} (1978), 1131–1134.
[Kup2003]{} S. Kupin, [*On sum rules of special form for Jacobi matrices*]{}, C. R. Math. Acad. Sci. Paris [**336**]{} (2003), 611–614.
[LNS2003]{} A. Laptev, S. Naboko, and O. Safronov, [*On new relations between spectral properties of Jacobi matrices and their coefficients*]{}, Comm. Math. Phys. [**241**]{} (2003), 91–110.
[LastS]{} Y. Last and B. Simon, [*Eigenfunctions, transfer matrices, and absolutely continuous spectrum of one-dimensional Schrödinger operators*]{}, Invent. Math. [**135**]{} (1999), 329–367.
[LS\_jdam]{} Y. Last and B. Simon, [*The essential spectrum of Schrödinger, Jacobi, and CMV operators*]{}, J. Anal. Math. [**98**]{} (2006), 183–220.
[Lev]{} N. Levinson, [*The Wiener RMS (root-mean square) error criterion in filter design and prediction*]{}, J. Math. Phys. Mass. Inst. Tech. [**25**]{} (1947), 261–278.
[MagWin]{} W. Magnus and S. Winkler, *Hill’s Equation*, Interscience Tracts in Pure and Applied Math., 20, Interscience Publishers, New York, 1966.
[M1]{} M. Măntoiu, [*$C^*$-algebras, dynamical systems at infinity and the essential spectrum of generalized Schrödinger operators*]{}, J. Reine Angew. Math. [**550**]{} (2002), 211–229.
[MRG89]{} F. Marcellán and I. Rodríguez, [*A class of matrix orthogonal polynomials on the unit circle*]{}, Linear Algebra and Applications (Valencia, 1987), Linear Algebra Appl. [**121**]{} (1989), 233–241.
[MS93]{} F. Marcellán and G. Sansigre, [*On a class of matrix orthogonal polynomials on the real line*]{}, Linear Algebra Appl. [**181**]{} (1993), 97–109.
[MY03]{} F. Marcellán and H. O. Yakhlef, [*Recent trends on analytic properties of matrix orthonormal polynomials*]{}, Orthogonal Polynomials, Approximation Theory, and Harmonic Analysis (Inzel, 2000), Electron. Trans. Numer. Anal. [**14**]{} (2002), 110–124 (electronic).
[Marc]{} V. A. Marchenko, *Sturm–Liouville Operators and Applications*, Birkhäuser, Basel, 1986.
[MNT85a]{} A. Máté, P. Nevai, and V. Totik, [*Asymptotics for the ratio of leading coefficients of orthonormal polynomials on the unit circle*]{}, Constr. Approx. [**1**]{} (1985), 63–69.
[McKT]{} H. P. McKean and E. Trubowitz, [*Hill’s operator and hyperelliptic function theory in the presence of infinitely many branch points*]{}, Comm. Pure Appl. Math. [**29**]{} (1976), 143–226.
[Nai]{} P. B. Naĭman, [*On the theory of periodic and limit-periodic Jacobian matrices*]{}, Soviet Math. Dokl. [**3**]{} (1962), 383–385; Russian original in Dokl. Akad. Nauk SSSR [**143**]{} (1962), 277–279.
[Nai2]{} P. B. Naĭman, [*On the spectral theory of non-symmetric periodic Jacobi matrices*]{}, Zap. Meh.-Mat. Fak. Har’kov. Gos. Univ. i Har’kov. Mat. Obšč.(4) [**30**]{} (1964), 138–151 (Russian).
[NPVY]{} F. Nazarov, F. Peherstorfer, A. Volberg, and P. Yuditskii, [*On generalized sum rules for Jacobi matrices*]{}, Int. Math. Res. Not. [**2005**]{}, no. 3, 155–186.
[Nev79]{} P. Nevai, [*Orthogonal polynomials*]{}, Mem. Amer. Math. Soc. [**18**]{} (1979), no. 213, 185 pp.
[Nev92]{} P. Nevai, [*Orthogonal polynomials, recurrences, Jacobi matrices, and measures*]{}, Progress in Approximation Theory (Tampa, FL, 1990), pp. 79–104, Springer Ser. Comput. Math., [**19**]{}, Springer, New York, 1992.
[NT89]{} P. Nevai and V. Totik, [*Orthogonal polynomials and their zeros*]{}, Acta Sci. Math. (Szeged) [**53**]{} (1989), 99–104.
[NTppt]{} P. Nevai and V. Totik, [*Denisov’s theorem on recurrence coefficients*]{}, J. Approx. Theory [**127**]{} (2004), 240–245.
[PF]{} L. Pastur and A. Figotin, *Spectra of Random and Almost-Periodic Operators*, Grundlehren der Mathematischen Wissenschaften \[Fundamental Principles of Mathematical Sciences\], 297, Springer-Verlag, Berlin, 1992.
[PY]{} F. Peherstorfer and P. Yuditskii, [*Asymptotic behavior of polynomials orthonormal on a homogeneous set*]{}, J. Anal. Math. [**89**]{} (2003), 113–154.
[Rab5]{} V. S. Rabinovich, [*Discrete operator convolutions and some of their applications*]{}, Math. Notes [**51**]{} (1992), 484–492.
[Rakh77]{} E. A. Rakhmanov, [*On the asymptotics of the ratio of orthogonal polynomials*]{}, Math. USSR Sb. [**32**]{} (1977), 199–213.
[Rakh83]{} E. A. Rakhmanov, [*On the asymptotics of the ratio of orthogonal polynomials, II*]{}, Math. USSR Sb. [**46**]{} (1983), 105–117.
[RT]{} J. Ralston and E. Trubowitz, [*Isospectral sets for boundary value problems on the unit interval*]{}, Ergod. Th. Dynam. Sys. [**8**]{} (1988), 301–358.
[RS4]{} M. Reed and B. Simon, *Methods of Modern Mathematical Physics, IV: Analysis of Operators*, Academic Press, New York, 1978.
[R07]{} C. Remling, [*The absolutely continuous spectrum of Jacobi matrices*]{}, Preprint (arXiv:0706.1101).
[Rod90]{} L. Rodman, [*Orthogonal matrix polynomials*]{}, Orthogonal Polynomials (Columbus, OH, 1989), pp. 345–362, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci. [**294**]{}, Kluwer, Dordrecht, 1990.
[Ryck]{} E. Ryckman, [*A strong Szegő theorem for Jacobi matrices*]{}, Comm. Math. Phys. [**271**]{} (2007), 791–820.
[Sak92]{} A. L. Sakhnovich, [*Spectral functions of a canonical system of order $2n$*]{}, Math. USSR Sb. [**71**]{} (1992), 355–369.
[SF]{} A. Sebbar and T. Falliero, [*Capacities and Jacobi matrices*]{}, Proc. Edinburgh Math. Soc. (2) [**46**]{} (2003), 719–745.
[Sh]{} J. A. Shohat, *Théorie Générale des Polinomes Orthogonaux de Tchebichef*, Mémorial des Sciences Mathématiques [**66**]{}, pp. 1–69, Paris, 1934.
[S168]{} B. Simon, [*Kotani theory for one dimensional stochastic Jacobi matrices*]{}, Comm. Math. Phys. [**89**]{} (1983), 227–234.
[Sim288]{} B. Simon, [*A canonical factorization for meromorphic Herglotz functions on the unit disk and sum rules for Jacobi matrices*]{}, J. Funct. Anal. [**214**]{} (2004), 396–409.
[Sim303]{} B. Simon, [*Meromorphic Szegő functions and asymptotic series for Verblunsky coefficients*]{}, Acta Math. [**195**]{} (2005), 267–285.
[OPUC1]{} B. Simon, *Orthogonal Polynomials on the Unit Circle, Part 1: Classical Theory*, AMS Colloquium Series, American Mathematical Society, Providence, RI, 2005.
[OPUC2]{} B. Simon, *Orthogonal Polynomials on the Unit Circle, Part 2: Spectral Theory*, AMS Colloquium Series, American Mathematical Society, Providence, RI, 2005.
[Sim\_Sturm]{} B. Simon, [*Sturm oscillation and comparison theorems*]{}, Sturm-Liouville Theory: Past and Present (W. Amrein, A. Hinz, and D. Pearson, eds.), pp. 29–43, Birkhäuser, Basel, 2005.
[Jost3]{} B. Simon, [*Meromorphic Jost functions and asymptotic series for Jacobi parameters*]{}, Funct. Anal. Appl. [**41**]{} (2007), 143–153.
[SimonNew]{} B. Simon, *Szegő’s Theorem and Its Descendants: Spectral Theory for $L^2$ Perturbations of Orthogonal Polynomials*, in preparation; to be published by Princeton University Press.
[SZ]{} B. Simon and A. Zlatoš, [*Sum rules and the Szegő condition for orthogonal polynomials on the real line*]{}, Comm. Math. Phys. [**242**]{} (2003), 393–423.
[SY]{} M. Sodin and P. Yuditskii, [*Almost periodic Jacobi matrices with homogeneous spectrum, infinite-dimensional Jacobi inversion, and Hardy spaces of character-automorphic functions*]{}, J. Geom. Anal. [**7**]{} (1997), 387–435.
[Sz15]{} G. Szegő, [*Ein Grenzwertsatz über die Toeplitzschen Determinanten einer reellen positiven Funktion*]{}, Math. Ann. [**76**]{} (1915), 490–503.
[Sz20]{} G. Szegő, [*Beiträge zur Theorie der Toeplitzschen Formen, I, II*]{}, Math. Z. [**6**]{} (1920), 167–202; Math. Z. [**9**]{} (1921), 167–190.
[Teschl]{} G. Teschl, *Jacobi Operators and Completely Integrable Nonlinear Lattices*, Mathematical Surveys and Monographs, 72, American Mathematical Society, Providence, RI, 2000.
[Toda]{} M. Toda, *Theory of Nonlinear Lattices*, second ed., Springer Series in Solid-State Sciences, 20, Springer, Berlin, 1989.
[newVA]{} W. Van Assche, [*Rakhmanov’s theorem for orthogonal matrix polynomials on the unit circle*]{}, J. Approx. Theory [**146**]{} (2007), 227–242.
[vMoer]{} P. van Moerbeke, [*The spectrum of Jacobi matrices*]{}, Invent. Math. [**37**]{} (1976), 45–81.
[Wat]{} D. S. Watkins, *Fundamentals of Matrix Computations*, second ed., Pure and Applied Mathematics, Wiley–Interscience, New York, 2002.
[Wei87]{} J. Weidmann, *Spectral Theory of Ordinary Differential Operators*, Lecture Notes in Math. [**1258**]{}, Springer, Berlin, 1987.
[Weyl]{} H. Weyl, [*Über beschränkte quadratische Formen, deren Differenz vollstetig ist*]{}, Rend. Circ. Mat. Palermo [**27**]{} (1909), 373–392.
[YM2001]{} H. O. Yakhlef and F. Marcellán, [*Orthogonal matrix polynomials, connection between recurrences on the unit circle and on a finite interval*]{}, Approximation, Optimization and Mathematical Economics (Pointe-à-Pitre, 1999), pp. 369–382, Physica, Heidelberg, 2001.
[YMP01]{} H. O. Yakhlef, F. Marcellán, and M. A. Piñar, [*Relative asymptotics for orthogonal matrix polynomials with convergent recurrence coefficients*]{}, J. Approx. Theory [**111**]{} (2001), 1–30.
[YK]{} D. C. Youla and N. N. Kazanjian, [*Bauer-type factorization of positive matrices and the theory of matrix polynomials orthogonal on the unit circle*]{}, IEEE Trans. Circuits and Systems [**CAS-25**]{} (1978), 57–69.
[^1]: $^1$ Department of Mathematics, Rice University, Houston, TX 77005. E-mail: [email protected]. Supported in part by NSF grants DMS-0500910 and DMS–0653720.
[^2]: $^2$ Department of Mathematics, UCLA, Los Angeles, CA 90095. E-mail: [email protected]. Supported in part by NSF grant DMS-0401277 and a Sloan Foundation Fellowship.
[^3]: $^3$ Mathematics 253-37, California Institute of Technology, Pasadena, CA 91125. E-mail: [email protected]. Supported in part by NSF grant DMS-0140592 and U.S.–Israel Binational Science Foundation (BSF) Grant No. 2002068
[^4]: Recall that $\sigma_{\text{\rm{ess}}}(J)$ is obtained from the spectrum of the Jacobi matrix $J$ by removing all isolated points.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The problem of guaranteed parameter estimation (GPE) consists in enclosing the set of all possible parameter values, such that the model predictions match the corresponding measurements within prescribed error bounds. One of the bottlenecks in GPE algorithms is the construction of enclosures for the image-set of factorable functions. In this paper, we introduce a novel set-based computing method called interval superposition arithmetics (ISA) for the construction of enclosures of such image sets and its use in GPE algorithms. The main benefits of using ISA in the context of GPE lie in the improvement of enclosure accuracy and in the implied reduction of number set-membership tests of the set-inversion algorithm.'
address:
- 'School of Information Science and Technology, ShanghaiTech University, 319 Yueyang Road, Shanghai 200031, China. ([{zhayl,wangk,meduardov, borish}@shanghaitech.edu.cn)]{}'
- 'Faculty of Chemical and Food Technology, Slovak University of Technology in Bratislava, Radlinskeho 9, Bratislava, Slovakia. ([[email protected]]{})'
author:
- Junyan Su
- Yanlin Zha
- Kai Wang
- 'Mario E. Villanueva'
- Radoslav Paulen
- Boris Houska
bibliography:
- 'references.bib'
title: Interval Superposition Arithmetic for Guaranteed Parameter Estimation
---
Set Arithmetics, Interval arithmetics, Guaranteed Parameter Estimation.
Introduction {#sec::intro}
============
In science and engineering, the behavior of processes and systems is often described using a mathematical model. Mathematical model development often follows three steps: model structure specification, design (and realization) of experiments, and estimation of unknown model parameters [@fra08]. In the last step, parameters are sought for which the model outputs match available measurements [@lju99].
One possible way of addressing the parameter estimation problem is the use of set-membership estimation [@mil91; @bai95], also called guaranteed parameter estimation (GPE). The GPE problem can be formulated as an identification of the set of all possible model parameter values which are not falsified by the plant measurements, within some prescribed error bounds. A set-inversion algorithm [e.g. SIVIA by @jau93] can be applied to find such set for nonlinear models. Here, the parameter set is successively partitioned into smaller boxes and using exclusion tests some of these boxes are eliminated, until a desired approximation is achieved. Since its advent, GPE has found various applications [see e.g., @mar00; @jau02; @lin07; @has15; @pau15-ima].
The complexity of the search procedure in SIVIA is proportional to the tightness of the interval enclosures. Considerable effort has then been invested into developing different set-arithmetics [@makino1996; @pau15-ima such as Taylor models] to produce tighter enclosures of image-set of nonlinear factorable functions. These techniques usually require computing and storing quantities such as sensitivity information.
Here, we propose an attempt to improve GPE algorithms using a novel non-convex set-arithmetic called Interval Superposition Arithmetic (ISA). This arithmetic operates over Interval Superposition models (ISM), representing a piecewise constant enclosure over a grid of the domain. Unlike a naive application of interval arithmetic (IA) over the grid, the computational and storage complexity of ISA is polynomial. Furthermore, it is able to exploit separable structures in the computational graph of a factorable function. Finally, unlike Taylor model arithmetics—which are based on local information—ISA is based on globally valid algebraic relations. As a result, ISMs are tighter than Taylor models—at least over large domains.
The rest of the paper is organized as follows, Section \[sec::GPE\] reviews GPE and set-inversion. Section \[sec::ISA\] presents an overview of ISA. An algorithm for intersecting ISMs with an interval—which forms the basis for a set-inversion algorithm—is presented in Section \[sec::SIVISA\]. It is important to notice that the intersection algorithm runs in polynomial time, but the complexity of computing an arbitrarily close approximation of the parameter set is exponential. The application of the proposed algorithm to a simple case study is shown in Section \[sec::case\]. Section \[sec::conclusion\] concludes the paper.
#### Notation {#notation .unnumbered}
The set of real valued compact interval vectors is denoted by $\mathbb I^{n} = \{[a,b]\subset \mathbb R^{n} \; | \;
a,b\in\mathbb R^{n},\, a\leq b\}$. Let $I=[a,b]\in\mathbb I$ and $c\in\mathbb R$, $c+I=I+c$ we have $[a+c,b+c]$. Similarly, $cI=Ic$ denotes $[ca,cb]$ if $c\geq 0$ ($[cb,ca]$ if $c<0$). The diameter of $I$ is denoted by $\operatorname{diam}(I)=b-a$. Interval operations are evaluated by IA [@Moore2009], e.g., $$\begin{aligned}
[a,b] + [c,d] &= [a+b,c+d]\;,\\
[a,b] * [c,d] &= [\min\{ac,ad,bc,bd\},\max\{ac,ad,bc,bd\}]\\
\exp([a,b]) & = [\exp(a),\exp(b)]\end{aligned}$$
Guaranteed parameter estimation {#sec::GPE}
===============================
We consider a system represented by the algebraic model $$\label{eq::model}
y = f(x)\;.$$ Here, $x\in\mathbb{R}^{n_x}$ denotes unknown parameter while $y\in\mathbb{R}^{n_y}$ the (observed) output variables. The model is described by the, posibly nonlinear, function $f:\mathbb{R}^{n_x}\to\mathbb{R}^{n_y}$.
Given $n_m\in\mathbb N$ measurements, $y^{\rm m}_{1},\ldots,y^{\rm m}_{n_m}\in
\mathbb{R}^{n_y}$, the GPE paradigm works under the assumption that true system outputs $y^{\circ}_{1},\ldots,y^{\circ}_{N}$ can be observed only within some bounded measurement bounds. Thus, for each $i\in\{1,\ldots,n_m\}$, we have $$y^{\circ}_{i}\in y^{\rm m}_{i}+[-\eta_{i},\eta_{i}]=:Y_{i}\in\mathbb I^{n_y}\,$$ with $\eta_{1},\ldots,\eta_{n_m}\geq 0$. The aim of GPE is to compute the set $$\label{eq::gpeset}
X_{\rm e} := \left\{x\in X_{0}\ |\ \forall i\in\{1,\ldots,N\}: f(x)\in Y_{i}
\right\}\;,$$ i.e., the set of parameters (within some admissible domain $X_{0}\in
\mathbb{I}^{n_x}$) for which the model outputs are consistent with all the uncertain observations $Y_{i}$.
Computing requires intersecting the preimage of $Y_{i}$ under $f$, with the initial parameter domain, i.e., $$X_{\rm e} = \left(\bigcap_{i=1}^{n_{m}} f^{-1}(Y_{i}) \right) \cap X_{0} \;.$$ This problem is intractable, in all but the simplest of cases, and thus one has to settle for approximations of this set. State-of-the-art algorithms for set inversion provide inner ($\mathbb X_{\rm int}$) and boundary ($\mathbb X_{\rm bnd}$) subpavings, i.e. lists of non overlapping interval vectors, satisfying $$\bigcup_{X\in\mathbb X_{\rm int}} X \subseteq X_{\rm e}
\subseteq \left( \bigcup_{X\in\mathbb X_{\rm int}} X \right) \cup
\left( \bigcup_{X\in\mathbb X_{\rm bnd}} X \right)\;.$$
In a nutshell, these algorithms work by subdividing the parameter domain $X_{0}$ into smaller boxes such that $X_{0} = \bigcup_{j} X_{j}$. Set arithmetics are then used to construct enclosures of $f$ on $X_{j}$, i.e. sets $\overline{Y}_{j}\subset
\mathbb R^{n_y}$ satisfying $$\label{eq::enclosure}
\overline{Y}_{j}\supseteq \{ f(x) \ |\ x\in X_{j} \}\;.$$ Using the information provided by the enclosure $Y_{j}$, the following set membership tests can be performed to classify the parameter boxes $X_{j}$ as interior or boundary boxes:
1. If $\overline{Y}_{j}\subseteq Y_{i}$ for all $i\in\{1,\ldots,n_m\}$, $X_{j}\in\mathbb{X}_{\rm int}$
2. Else, if $Y_{i}\cap f(X) = \emptyset$ for some $i\in\{1,\ldots,n_m\}$, $X_{j}\cap X_{\rm e}= \emptyset$
3. Else, $X\in\mathbb{X}_{\rm bnd}$.
Figure \[fig::sinv\] shows the result of the above process for the function $f=x_{1}^{3} + x_{2}^{3}$ over $X_{0} = [-3,3]^2$, with $Y = [-2,2]$. The set $X_0$ has been divided into $N=20$ equidistant pieces along each coordinate, resulting in 400 interval vectors $X_{j}$. The plot shows the set $\bigcup_{i=1}^{N^{n_x}} \left(X_{j} \times \overline{Y}_{j}\right)$, and its projection onto the $(x_1,x_2)$-space. The red and blue boxes belong to $\mathbb X_{\rm int}$ and $\mathbb X_{\rm bnd}$ respectively.
In practice, the domain $X_{0}$ is subdivided iteratively by bisecting boundary boxes, starting with $\mathbb X_{\rm bnd} = X_{0}$ and $\mathbb X_{\rm int}=\emptyset$. The bounding, set-membership, and bisection operations are repeated until a termination criterion, e.g. $$\label{eq::maxdiam}
\forall X\in\mathbb{X}_{\rm bnd}:\quad \operatorname{diam}(X)\leq \epsilon\;,$$ for a user-defined tolerance $\epsilon>0$, is met. 0
[SINV.pdf]{} (-5,49)[$y$]{} (80,0)[$x_{1}$]{} (25,0)[$x_{2}$]{}
One of the bottlenecks of set inversion algorithms is the over-conservatism of existing set-arithmetics, particularly over large domains. Hence we propose to approach this problem within a novel set-arithmetics paradigm.
Interval superposition arithmetic {#sec::ISA}
=================================
Interval superposition arithmetic is a novel enclosure method for the image set of nonlinear factorable functions. It propagates nonconvex sets, called interval superposition models, through the computational graph of the function. Unlike Taylor [@makino1996] and Chebyshev models [@bat04; @raj17], ISA does not rely on local approximation methods, instead relying on global algebraic properties and partially separable structures within the function.
Interval superposition models
-----------------------------
Consider an interval domain $X=\left[\underline{x}_1,\overline{x}_{1}\right]\times
\ldots\times\left[\underline{x}_{n_x},\overline{x}_{n_x}\right]$. Now, take a partition of $X$ into intervals of the form $$\label{eq::branching}
X_{i}^{j} = [ \underline{x}_{i}+(j-1)h_i,\underline{x}_{i}+jh_{i}] \quad\text{with}
\quad h_i=\frac{\overline{x}_i-\underline{x}_i}{N}\;,$$ for all $i\in\{1,\ldots,n_x\}$ and all $j\in\{1,\ldots,N\}$, with $N$ being a user-specified integer. An interval superposition model of a real-valued function $f:\mathbb{R}^{n_x}\to\mathbb{R}$ on $X$ is an interval valued function $\Gamma:X\times\mathbb{I}^{n_x\times N}\times\mathbb{I}^{n_x}\to\mathbb{I}$, given by $$\label{eq::ISM}
\Gamma(x,A,X) = \sum^{n_x}_{i=1}\sum^{N}_{j=1} A_{i}^{j}\varphi_{i}^{j}(x)\;,$$ with $$\varphi_i^{j}(x) =
\begin{cases}
1 &\text{if} \ x_{i}\in X_{i}^{j}, \\
0 &\text{otherwise.}
\end{cases}$$ Here, $A_{i}^{j} = \left[ \underline{A}^{i}_{j},\overline{A}^{i}_{j}\right]$ are the components of a matrix $$A =
\begin{pmatrix}
A^{1}_{1} & \ldots & A^{N}_{1} \\
\vdots & \ddots & \vdots \\
A^{1}_{n_x} & \ldots & A^{N}_{n_x}
\end{pmatrix} \in \mathbb{I}^{n_x\times N}\;,$$ which, for a fixed $X$, completely determines the enclosure function of $f$. Note that ISMs for functions $f:\mathbb{R}^{n_x}\to\mathbb{R}^{n_y}$ are defined by stacking ISMs for each $f_{i}$. The matrix $A$ is constructed such that $\Gamma(\cdot,A,X)$ is a piecewise constant enclosure function of $f$ over $X$, i.e. $$\forall x\in\mathbb{X}: \quad f(x)\in \Gamma(x,A,X)\;.$$
The name *interval superposition* is motivated by the structure of the enclosure function: At any $x\in X^{j}_{1}\times\ldots\times X^{j}_{n_x}$, the interval $\overline{Y} = \Gamma(x,A,X)$ is given by the Minkowski sum (or superposition) of $n_x$ interval functions $\sum^{N}_{j=1}A^{j}_{i}\varphi^{j}_{i}(x)$. The separable structure of ISMs allows for a storage complexity of order ${\rm\bf O}(n_xN)$, since only $n_{x}N$ intervals need to be stored, in the matrix $A$, to represent the $N^{n_x}$ pieces of the enclosure. In Figure \[fig::sinv\] the graph of an ISM, over a partition of $X$ (with $N=20$) is shown. Although this set consists of $400$ interval vectors (shown in red, white and blue), only 40 intervals are stored in the matrix $A$.
This separability also allows for the global minima and maxima of $\Gamma(\cdot,A,X)$ over $X$, $$\label{eq::rbounder}
\lambda(A) = \sum^{n_x}_{i=1}
\underbrace{\min_{j\in\{1,\ldots,N\}} \underline{A}^{j}_i}_{=:L(A_{i})}
\ \, \text{and} \ \,
\mu(A) = \sum^{n_x}_{i=1}
\underbrace{\max_{j\in\{1,\ldots,N\}} \overline{A}^{j}_i}_{=:U(A_{i})}\;,$$ to be computed with a complexity of order ${\rm\bf O}(n_xN)$. The interval $\left[\lambda(A),\mu(A)\right]$ denotes the range of ISM.
Arithmetic rules for interval superposition models
--------------------------------------------------
Interval superposition arithmetics propagates ISMs through the computational graph of a factorable function, defined by a finite recursive composition of atom operations from a finite library $\mathcal L = \{ \exp, \, \sin,\, +,\, *,\,\ldots\}$.
Consider the functions $g,h:X\to\mathbb{R}$, and a (possibly bivariate) atom operation $\alpha$. Let the interval matrices $A,B\in\mathbb{I}^{n_x \times N}$ be the respective parameters for ISMs of $g$ and $h$ over $X$. In ISA, a univariate composition rule is a map taking $A$ as an input and returning an interval matrix $C\in\mathbb{I}^{n_x\times N}$ parameterizin an ISM such that $$\forall x\in\mathbb{X}: \quad (\alpha \circ g)(x) \in \Gamma(x,C,X)\;.$$ Here, $\alpha \circ g$ denotes the composition of $\alpha$ and $g$.
Bivariate composition rules in ISA are defined analogously, with the map taking both $A$ and $B$ as inputs. Although such maps are specific for each atom operation $\alpha$, the main steps are outlined in Algorithms \[alg::univariate\] and \[alg::product\] for univariate compositions and bivariate products respectively. The addition rule in interval superposition arithmetic is simple. An interval superposition model of $g+h$ on $X$ is parameterized by the matrix $C=A+B$, with the sum computed componentwise using interval arithmetics.
------------------------------------------------------------------------
height 1pt
------------------------------------------------------------------------
**Input:** Matrix $A$ parameterizing $F_{h,X}$ and an atom operation $\alpha$.\
**Main Steps:**\
1. Choose, for all $i \in \{ 1, \ldots, n_x \}$, central points $a_i \in \mathbb R$ satisfying $$L(A_i) \, \leq \, a_i \, \leq \, U(A_i) \quad \text{and set} \quad \omega = \sum_{i=1}^{n_x} a_i \; .$$
2. Choose a suitable remainder bound $r_\alpha(A) \geq 0$ such that $$\begin{aligned}
\left| \sum_{i=1}^{n_x} \alpha( \omega + \delta_i ) - (n_x-1)\alpha(\omega) - \alpha \left( \omega + \sum_{i=1}^{n_x} \delta_i \right) \right| \; \leq \; r_\alpha(A) \notag
\end{aligned}$$ for all $\delta \in \mathbb R^{n_x}$ with $\forall i \in \{ 1, \ldots, n_x \}, \; \; L(A_i) \leq a_i + \delta_i \leq U(A_i)$.
3. Compute the interval valued coefficients $$C_i^j = \alpha \left(\omega - a_i + A_i^j \right) - \frac{n_x-1}{n_x} \alpha(\omega) \; .$$ for all $i \in \{ 1, \ldots, n_x\}$ and all $j \in \{ 1, \ldots, N \}$, where $\alpha \left(\omega - a_i + A_i^j \right)$ is evaluated in interval arithmetic.
4. Set $C_k^j \leftarrow C_k^j + r_{\alpha}(A) \cdot [-1,1] $ for all $j \in \{ 1, \ldots, N \}$ with $$k \in \operatorname*{argmax}_{i\in\{1,\ldots,n_x\}} \sum^{N}_{j=1} \overline{A}_{i}^{j} - \underline{A}_{i}^{j}$$.
**Output:** Matrix $C\in\mathbb{I}^{n_x\times N}$ parameterizing $\Gamma(\cdot,C,X)$ for $\alpha \circ g$.
------------------------------------------------------------------------
height 1pt
------------------------------------------------------------------------
height 1pt
------------------------------------------------------------------------
**Input:** Matrices $A$ and $B$ parameterizing $F_{h,X}$ and $F_{g,X}$.\
**Main Steps:**\
1. Compute the central points, $\forall i \in \{ 1, \ldots, n_x \}$ $$a_i = \frac{\mathrm{U}\left( A_i \right) + \mathrm{L}\left( A_i \right)}{2}\quad
\text{and} \quad
b_i = \frac{\mathrm{U}\left( B_i \right) + \mathrm{L}\left( B_i \right)}{2}\,$$ then set $$a = \sum_{i=1}^{n_x} a_i \; , \; \; b = \sum_{i=1}^n b_i \; , \;\;
c = \sum_{i=1}^{n_x} a_i b_i \;, \;\;\text{and} \;\;
\omega = \frac{ab - c}{n_x}\; .$$
2. Compute $\rho_i(A) = \frac{U(A_i)-L(A_i)}{2}$ and $\rho_i(B) = \frac{U(B_i)-L(B_i)}{2}$ for all $i \in \{ 1, \ldots, n_x \}$ as well as the associated remainder bound $$R(A,B) = \left( \sum_{i=1}^{n_x} \rho_i(A) \right) \left( \sum_{i=1}^{n_x} \rho_i(B) \right) - \sum_{i=1}^{n_x} \rho_i(A)\rho_i(B) \; .$$
3. Compute, for each $i \in \{ 1, \ldots, n_x\}$ and all $j \in \{ 1, \ldots, N \}$ $$C_{i}^j = \left( A_{i}^j + a - a_i \right) \left( B_{i}^j + b - b_i \right) - \left( a - a_i \right)\left(b - b_i \right) - \omega\;.$$
4. Set $C_k^j \leftarrow C_k^j + R(A,B) \cdot [-1,1] $ for all $j \in \{ 1, \ldots, N \}$ with $$k \in \operatorname*{argmax}_{i\in\{1,\ldots,n_x\}} \sum^{N}_{j=1} \overline{A}_{i}^{j} - \underline{A}_{i}^{j}$$.
**Output:** Matrix $C\in\mathbb{I}^{n_x\times N}$ parameterizing $\Gamma(\cdot,C,X)$, for $g*h$.
------------------------------------------------------------------------
height 1pt
\[thm::univariate\] Let $\Gamma(x,A,X)$ and be an ISM of $g$ on $X$. If the matrix $C\in\mathbb{I}^{n_x\times N}$ is computed using Algorithm \[alg::univariate\], then $\Gamma(x,C,X)$ is an ISM of $\alpha \circ g$ on $X$.
For the first statement, let $x\in X$ be an arbitrary point. Since $\Gamma(x,A,X) =
\sum^{n_x}_{i=1}\sum^{N}_{j=1}A^{j}_i\varphi^{j}_{i}(x)$ is an ISM of $g$, there exists a sequence $j_1,\ldots j_{n_x}\in\{1,\dots,N\}$ and points $y_{i}\in A^{j_i}_{i}$ satisfying $g(x)=\sum^{n_x}_{i=1}y_i$. Let $\delta_i = y_i -a_i$, with $\omega$ defined as in Algorithm \[alg::univariate\] one can write $$\small
\begin{aligned}
\alpha(g(x)) &= \hphantom{{}+{}} \alpha\left( \omega+\sum^{n_x}_{i=1}\delta_i \right)\\
&=\hphantom{{}+{}} \sum^{n_x}_{i=1}\left(\alpha(\omega+\delta_i)
-\frac{n_x-1}{n_x}\alpha(\omega)\right)\\
&\hphantom{{}={}}-\underbrace{\left( \sum^{n_x}_{i=1}\alpha(\omega+\delta_i)-(n_x-1)\alpha)\omega
-\alpha\left(\omega+\sum^{n_x}_{i=1}\delta_i\right) \right)}_{r_{\alpha}(A)[-1,1]}\,.
\end{aligned}$$ Since $\delta_i\in A_{i}^{j_i}-a_i$, we have $\alpha\left(\omega-a_i+A^{j_i}_i\right)$ and $$\small
\begin{aligned}
\alpha(g(x)) &\in \sum^{n_x}_{i=1}\left(\alpha\left(\omega-a_i+A^{j_i}_i\right)
-\frac{n_x-1}{n_x}\alpha(\omega)\right) + r_{\alpha}(A)[-1,1]\\
&=\sum^{n_x}_{i=1} C_{i}^{j_i}\;,
\end{aligned}$$ which implies the statement of the theorem.
\[thm::product\] Let $\Gamma(x,A,X)$ and $\Gamma(x,B,X)$ be ISMs of $g$ and $h$, respectively, on $X$. If $C\in\mathbb{I}^{n_x\times N}$ is computed using Algorithm \[alg::product\], then $\Gamma(x,C,X)$ is an ISM of $g * h$ on $X$.
A proof of Thm. \[thm::product\] proceeds along the same lines as the proof of Thm. \[thm::univariate\] and its omitted for the sake of brevity.
The construction of remainder bounds and central points used in Algorithm \[alg::univariate\] exploits globally valid algebraic properties, called addition theorems, of common univariate operations. As an example, for the exponential function, the addition theorems $e^{\omega+\delta_i} = e^{\omega}e^{\delta_i}$ and $e^{\omega+\sum^{n_x}_{i=1}\delta_i} = e^{\omega}\prod^{n_x}_{i=1}e^{\delta_i}$, hold globally over the real numbers. Letting $t_i = e^{\delta_i}-1$, $r_{\alpha}(A)$ can be constructed by bounding the left-hand side of the expression in Step 2) of Algorithm \[alg::univariate\]. This yields the expression $$e^{\omega}\left| \sum^{n_x}_{i=1}t_i+1-\prod^{n_x}_{i=1}(1+t_i)\right|
\leq e^{\omega}\left( \prod^{n_x}_{i=1}(1+s_i) - \sum^{n_x}_{i=1} s_i-1 \right)$$ with $s_{i}=\max\left\{ e^{U(A_i)-a_i}-1,1-e^{L(A_i)-a_1} \right\}$. Choosing $a_i=\log\left(\frac{1}{2}\left( e^{U(A_{i})} + e^{L(A_{i})}\right)\right)$, minimizes $$s_{i} = \frac{e^{U(A_i)}-e^{L(A_i)}}{e^{U(A_i)}+e^{L(A_i)}}\;.$$ The technical derivations for the remainder bounds $r_{\alpha}(A)$ and the central points $a_i$ for other atom operations can be found in [@Zha2016].
The final ingredient for an arithmetic of interval superposition models is the construction of a (trivial) ISM for the input variables $x_i$. As each variable is independent of the rest, the coefficients can be set as $A^{j}_{k} = 0$ for all $k\neq i$ and all $j\in\{1,\ldots,N\}$. The $i$th row of $A$ is then initialized as $A^{j}_{i}=X^{j}_{i}$ for each $j\in\{1,\ldots,N\}$.
Notice that composition rules in ISA have a computational complexity of ${\rm \bf O}(n_xN)$. Unfortunately, global statements regarding the model accuracy can only be given whenever $f$ is separable, i.e. $f(x) = \sum^{n_x}f_{j}(x_j)$ for some factorable functions $f_{1},\ldots,f_{n_x}$. In this case, the error is of order ${\rm \bf O}\left(\frac{1}{N}\right)$ over all bounded domains $X\subset\mathbb{I}^{n_x}$.
ISA-Based Set-inversion Algorithm {#sec::SIVISA}
=================================
This section proposes a novel search strategy based on ISA for addresing GPE. It has as its core computing the intersection of an ISM with an interval.
Consider an ISM, of the function $f$ over $X$, parameterized by $A\in\mathbb{I}^{n_x\times N}$. The direct way of computing the intersection between this ISM and $Y=[\underline{y},\overline{y}]$ is to compute the value of the ISM at each interval $X_{j_1} \times \ldots \times X_{j_N}$ in the partition of $X$. This requires computing all possible superpositions of coefficients $A^{i}_{j}$. Such approach, while straightforward, is unfortunately not efficient since its computational complexity is ${\rm \bf O}\left( N^{n_x} \right)$.
As it turns out, computing an over approximation of the desired intersection can be done by testing only certain selected combinations. The proposed approach, requires sorting the components $A^{j}_{i}=[\underline{A}^{j}_{i},
\overline{A}^{j}_{i}]$ of the rows $A_i$ of the matrix $A$ in both decreasing and increasing orders. The corresponding permutations are denoted by the functions $\overline{\pi}_{i},\underline{\pi}_{i}:\{1,\ldots,N\}\to\{1,\ldots,N\}$ satisfying $$\overline{A}^{\overline{\pi}_{i}(1)}_i \geq \overline{A}^{\overline{\pi}_{i}(2)}_i \geq \ldots \geq \overline{A}^{\overline{\pi}_{i}(N)}_i$$ and $$\underline{A}^{\underline{\pi}_{i}(1)}_i \leq \underline{A}^{\underline{\pi}_{i}(2)}_i \leq \ldots \leq \underline{A}^{\underline{\pi}_{i}(N)}_i \; .$$ In the following, we use the shorthand $\overline \Pi = ( \overline \pi_1, \ldots, \overline \pi_{n_x} )$ and $\underline \Pi = ( \underline \pi_1, \ldots, \underline \pi_{n_x} )$. The main pre-processing step for computing a set inversion is outlined in Algorithm \[alg:Intersect\].
------------------------------------------------------------------------
height 1pt
------------------------------------------------------------------------
\[alg:Intersect\] **Input:** Parameters $A$ and $X$ of the input model and an interval $Y$\
**Main Step:**\
1. Sort each $A_i$ to obtain the permutations $\underline \Pi$ and $\overline \Pi$.
2. Choose a finite number $n_J$ of intervals $\underline J_k = [0, \underline j_k]$ with index vectors $\underline j_k \in \{1, \ldots, N \}^{n_x}$ such that $$\forall k\in \{1,\dots,n_J\}, \quad \sum_{i=1}^{n_x} \underline A_i^{{\underline \pi}_i((\underline j_k)_i)} \leq \underline y$$
3. Choose a finite number $n_J$ of intervals $\overline J_k = [0, \overline j_k]$ with index vectors $\overline j_k \in \{1, \ldots, N \}^{n_x}$ such that $$\forall k\in \{1,\dots,n_J\}, \quad \sum_{i=1}^{n_x} \overline A_i^{{\overline \pi}_i((\overline j_k)_i)} \geq \overline y$$
**Output:** Permutations $\underline \Pi,\overline \Pi$ and intervals $\underline J = (\underline J_1, \ldots, \underline J_{n_J})$, $\overline J = (\overline J_1, \ldots, \overline J_{n_J})$.\
------------------------------------------------------------------------
height 1pt
\[thm::Intersect\] Let $\underline \Pi,\overline \Pi$ and $\underline J = (\underline J_1, \ldots, \underline J_{n_J})$, $\overline J = (\overline J_1, \ldots, \overline J_{n_J})$ be computed by Algorithm \[alg:Intersect\]. Define $$\underline \Xi = \bigcup_{k \in \{ 1, \ldots, n_J\}} \bigcup_{j \in \underline J_k} \Xi_1^{{\underline \pi}_1(j_1)} \times \ldots \times \Xi_{n_x}^{{\underline \pi}_{n_x}(j_{n_x})}$$ and $$\overline \Xi = \bigcup_{k \in \{ 1, \ldots, n_J\}} \bigcup_{j \in \overline J_k} \Xi_1^{{\overline \pi}_1(j_1)} \times \ldots \times \Xi_{n_x}^{{\overline \pi}_{n_x}(j_{n_x})}$$ with $\Xi_i^j = [ \underline x_i + (j-1) h_i, \underline x_i + j h_i]$ and $h_i = \frac{\overline x_i - \underline x_i}{N}$. Then, $$X \setminus \left( \underline \Xi \cup \overline \Xi \right) \supseteq \mathbb X_\mathrm{int} \cup \mathbb X_\mathrm{bnd} \; .$$
By construction, the function $f$ takes values larger than $\overline y$ on all interval boxes $\Xi_1^{{\underline \pi}_1(j_1)} \times \ldots \times \Xi_{n_x}^{{\underline \pi}_{n_x}(j_{n_x})}$ for any $j \in \overline J_{k}$. Similarly, $f$ takes smaller values than $\underline y$ on all intervals $\Xi_1^{{\overline \pi}_1(j_1)} \times \ldots \times \Xi_{n_x}^{{\overline \pi}_{n_x}(j_{n_x})}$ for any $j \in \underline J_{k}$. Consequently, the union of all of these boxes cannot possibly contain a point of $\mathbb X_\mathrm{int} \cup \mathbb X_\mathrm{bnd}$, which is the statement of the theorem.
Theorem \[thm::Intersect\] provides a constructive procedure for finding the desired outer approximation of the set $\mathbb X_\mathrm{int} \cup \mathbb X_\mathrm{bnd}$. Notice that the computational complexity of Algorithm \[alg:Intersect\] is of order $O( n_x N \log(N) )$, because we need to sort the intervals along all coordinate directions. The associated storage complexity is of order $O( n_x N )$. Finally, we have to keep in mind, however, that computing and storing the sets $\underline \Xi$ and $\overline \Xi$ is expensive in general, as these sets may be composed of an exponentially large amount of sub-intervals. Nevertheless, it is not necessary to store these sets explicitly as long as we store the permutation matrices $\underline \Pi$ and $\overline \Pi$ as well as the boxes $\underline J$ and $\overline J$, which uniquely represent the set $X \setminus \left( \underline \Xi \cup \overline \Xi \right)$.
Notice that there are various heuristics possible for refining the above procedure. However, the corresponding methods are analogous to the implementation in SIVIA and based on state-of-the-art branching techniques. Thus, the proposed technique based on Algorithm 3 can be embedded in an exhaustive search procedure, if one wishes to approximate the set $\mathbb X_\mathrm{int} \cup \mathbb X_\mathrm{bnd}$ with any given accuracy.
Numerical Examples {#sec::case}
==================
This section illustrates some of the benefits of ISA as a bounding method for the range of factorable functions, as well as its application to GPE. Algorithms 1, 2, and a set-inversion algorithm based on Algorithm 3 were implemented in the programming language `Julia`. For comparison, a basic SIVIA algorithm was also implemented in `Julia`. The termination for both algorithms was based on . All results were obtained on an Intel Xeon CPU X5660 with 2.80GHz and 16GB RAM.
Bounding a nonlinear function: ISMs vs TMs
------------------------------------------
Consider the nonlinear factorable function $$f(x) = e^{\sin(x_1)+\sin(x_2)\cos(x_2)}$$ over the domain $X = [0,1]\times[0,\overline{x}_{2}]$. Here, $\overline{x}_{2}\in[0.1,20]$ denotes a parameter which controlling the diameter of the domain. In order to measure the quality of an arithmetic, we used the Hausdorff distance between the range of $f$, $f(X)={f(x)|x\in X}$, and an enclosure set $\overline Y \supseteq f(X)$. This distance is given by $$d_{\rm H}(f(X),\overline{Y}) = \max_{y\in \overline Y} \min_{x\in f(X)} || x-y||_{\infty}\;.$$
Figure \[fig::enclosure\] shows the overestimation of enclosures in the form of Taylor models of orders 1 and 2 as well as interval superposition models with $N=1$, $N=10$, and $N=100$ as a function of the domain parameter $\overline{x}_2$. Although the Hausdorff distance between $f(X)$ and $\overline{Y}$ does not increase monotonically with $\overline{x}_2$, the rough trend observed on the plot is that the overestimation increases with the size of the domain. Furthermore, the plot shows that interval superposition models outperform Taylor models over large domains. One aspect that is not shown in the figure is that over small domains, e.g. over $[0,10^{-1}]^2$, enclosures based on Taylor models outperform those constructed using interval superposition arithmetics.
[testB-clean.pdf]{} (5,74)[$d_{\rm H}\left(f(X),\overline Y \right)$]{} (94,6)[$\overline x_2$]{}
(88,49)[$N = 1$]{} (88,43)[$N = 10$]{} (88,34)[$N = 100$]{}
(20,59) (6,59)
(4,-4)[$1$]{} (22,-4)[$5$]{} (42,-4)[$10$]{} (64,-4)[$15$]{} (85,-4)[$20$]{}
(-10,6)[$10^{-2}$]{} (-9.9,18)[$10^{-1}$]{} (-2.5,30)[$1$]{} (-7,42.5)[$10^1$]{} (-7.2,55)[$10^2$]{} (-7.2,67)[$10^3$]{}
Guaranteed parameter estimation via ISMs
----------------------------------------
We consider a reaction system given : $$\label{eq::odes}
\begin{alignedat}{2}
\dot{z_1}(t) &= - (x_1 + x_3) z_1(t) + x_2 z_2(t), \quad &&z_1(0) = 1, \\
\dot{z_2}(t) &= x_1 z_1(t) - x_2 z_2(t), &&z_2(0) = 0\;,
\end{alignedat}$$ with $y(t) = z_2(2)$ [@pau15-ima]. The output variable, can be represented as the factorable function $$y(t) = \frac{e^{\frac{-t\rho}{2}}x_1 (e^{\frac{t\sigma}{2}} - e^{\frac{ - t\sigma}{2}})}{\sigma}$$ with $\sigma = \sqrt{x_1^2 + x_2^2 + x_3^2 + 2x_1p_2 + 2x_1x_3 - 2x_2x_3}$ as well as $\rho = x_1 + x_2 + x_3$. In the following, we fix $x_3 = 0.35$ and consider $n_m = 15$ measurements corresponding to the time instants $t_i=1,2,\ldots,15$. Process measurements were obtained by simulating with $x = (0.6,0.15,0.35)^T$, rounding to the second significant digit. Measurement errors of $\pm 10^{-3}$ were added to these values.
The performance of the proposed GPE algorithm using ISA was tested against a standard SIVIA. We have interval superposition models with $N=2,\,10,\,20$. Figure \[fig::results\] shows a summary of the results of the GPE algorithm using ISMs with $N=2$. The left plot, shows an approximation of the set $X_e$. The plot shows the inner partition (in red) for $\epsilon=10^{-5}$ and the boundary partitions for $\epsilon=10^{-4}$ (light blue) and $\epsilon-10^{-5}$ (dark blue). The central and right plots show, respectively, a comparison of the number of iterations and CPU time against the tolerance $\epsilon$—for SIVIA (solid red line) and ISM-based set-inversion with $N=2$ (solid black line), $N=10$ (dotted black line), and $N=20$ (dashed black line). In terms of the number of iterations and number of boundary boxes (not shown), ISM-based set-inversion (for all $N$) outperforms SIVIA. This is due to the fact that ISA is able to detect and exploit structures in the factorable function to remove redundant boxes. On the contrary, with respect to the CPU time, SIVIA outperforms the proposed algorithm. This can be traced back to the fact that the the cost per iteration is larger for ISA. Furthermore, the implementation is still at prototype stage and requires further refinement in terms of computing remainder bounds and memory management in the algorithms.
[rowplots3.pdf]{}
(30,2)[$x_1$]{} (5.5,-1)[$0.598$]{} (14.5,-1)[$0.600$]{} (23.5,-1)[$0.602$]{}
(4.5,19)[$x_2$]{} (-2,1.5)[$0.1485$]{} (-2,6.5)[$0.1495$]{} (-2,11.5)[$0.1505$]{} (-2,16.5)[$0.1515$]{}
(64,2)[$\epsilon$]{} (36.5,-1)[$10^{-5}$]{} (42.5,-1)[$10^{-4}$]{} (49,-1)[$10^{-3}$]{} (55,-1)[$10^{-2}$]{} (61.5,-1)[$10^{-1}$]{}
(38.5,19)[Iterations]{} (35,1)[$10^{0}$]{} (35,4)[$10^{1}$]{} (35,7)[$10^{2}$]{} (35,10.5)[$10^{3}$]{} (35,13.5)[$10^{4}$]{} (35,16.5)[$10^{5}$]{}
(98.5,2)[$\epsilon$]{} (70.5,-1)[$10^{-5}$]{} (76.5,-1)[$10^{-4}$]{} (83.0,-1)[$10^{-3}$]{} (89.5,-1)[$10^{-2}$]{} (95.5,-1)[$10^{-1}$]{}
(73,19)[CPU Time]{} (68,1)[$10^{-2}$]{} (68,4)[$10^{-1}$]{} (69,7)[$10^{0}$]{} (69,10.5)[$10^{1}$]{} (69,13.5)[$10^{2}$]{} (69,16.5)[$10^{3}$]{}
Conclusion {#sec::conclusion}
==========
This paper presented Interval superposition arithmetics, a novel set-arithmetic for computing enclosures of the image set of factorable functions and its use in guaranteed parameter estimation. The main advantage of ISA is its polynomial storage and computational complexity. [The core routine behind the proposed GPE method is the intersection of an interval superposition model and an interval. Although the proposed intersection routine has a computational complexity of order $O(n_x N \log(N))$, computing an arbitrarily accurate approximation of the parameter set requires exponential run time. Our numerical examples illustrate the advantages of ISA over other set arithmetics when constructing enclosures for factorable functions—particularly over large domains. We have also shown how the proposed technique can be used to solve a GPE problem. Although the number of iterations is reduced when using ISA, the overal CPU time is larger than SIVIA. This suggest that, although ISA can improve certain aspects of GPE algorithms, there is still much room for improvement. Improved ISA-based algorithms for constructing approximations of inverse-image sets in polynomial run-time will be investigated in future work.]{}
Acknowledgments {#acknowledgments .unnumbered}
===============
[This work was supported by: National Science Foundation China (NSFC), Grant 61473185; ShanghaiTech University, Grant F-0203-14-012; and Slovak Research and Development Agency, project APVV SK-CN-2015-0016 “CN-SK cooperation: Verified Estimation and Control of Chemical Processes”. RP also acknowledges: Slovak Research and Development Agency, project APVV 15-0007; and European Commission, grant agreement 790017 (GuEst). ]{}
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'M. Murgia'
- 'A. Crapsi'
- 'L. Moscadelli '
- 'L. Gregorini'
date: 'Received; Accepted'
title: 'Radio continuum and CO emission in star-forming galaxies.'
---
14[$B_{1.4}$]{}
Introduction
============
One of the major goals of the studies of external galaxies is understanding the relationship between the star formation rate (SFR) and the physical condition in the interstellar medium. Several indicators have been suggested to estimate the SFR (of massive stars) in galaxies. These include the U-band magnitude, the strength of Balmer lines emission, the far-infrared (FIR) emission and the radio luminosity; the rates inferred from the different indicators span almost four orders of magnitude going from $10^{-2}$ to $10^{2}$ . Cram et al. (1998) checked the consistency between the SFR deduced from the U-band, H$_{\alpha}$, FIR and radio luminosity using a sample of 700 local galaxies. They noted that there are systematic differences between these various indicators. In particular they suggested that the H$_{\alpha}$ luminosity may underestimate the SFR by approximately an order of magnitude when the SFR is $\geq 20$ . They concluded that the radio continuum luminosity at decimeter wavelengths of a star forming galaxy provides a better way to estimate the current rate of star formation. The radio continuum emission at 1.4 GHz from a star-forming galaxy is mainly synchrotron radiation produced from relativistic electrons accelerated by supernovae explosions (Lequeux 1971). Indeed the radio continuum luminosity appears to be directly proportional to the rate of formation of supernovae (Condon 1992). This view is reinforced by the tight correlation existing between the radio luminosity and the FIR for spiral galaxies (see e.g. Condon 1992). Since the radio continuum at 1.4 GHz does not suffer significant extinction, the radio luminosity constitutes a very useful tool to determine the current SFR in a spiral galaxy.
Since the discovery that stars form in molecular clouds, it is essential to determine, not only the rate, but also the efficiency of conversion of the interstellar gas in stars; i.e. the star formation efficiency (SFE). The SFE measures the formation rate of young stars per unit of mass of gas available to form those stars. Determining the SFE is important to distinguish a situation in which a high SFR indicates a higher efficiency in converting gas in stars rather than a higher gas quantity.
The CO molecule luminosity and the virial mass of giant molecular clouds correlate very well in our Galaxy and in other nearby spirals (Young & Scoville 1991 and references therein). The comparison of different SFR tracers with the mass of molecular clouds provides indeed an important tool to investigate the behaviour of the SFE within and among galaxies. Many studies have been concerned with the behaviour of the star formation process on global scales, averaged over the entire star-forming disk. These works showed that the disk-averaged star formation process is well described by a Schmidt (1959) law of the type $\Sigma_{\rm SFR} \propto \Sigma_{\rm gas}^{N}$, where $\Sigma_{\rm SFR}$ and $\Sigma_{\rm gas}$ are the observable surface density of SFR and total (atomic + molecular) gas density, respectively, and the exponent $N$ typically ranges from 1.3 to 1.5 (Kennicutt 1998).
An interesting development of these global studies, the investigation of the behaviour of the SFE [*within*]{} the disks of the individual galaxies, provides much physical insight into the star formation process itself. The extragalactic CO survey of the Five College Radio Astronomy Observatory (Young et al. 1995, hereafter FCRAO CO Survey) provided a uniform database of CO data for 300 galaxies at a resolution of 45, opening the possibility to extend the study of the Schmidt relationship of the SFR versus the H$_{2}$ density over the same physical regions well inside the galaxy disks. Since the star formation process involves the molecular gas directly, some authors recognized that the determination of the Schmidt law assumes a clear physical meaning if restricted to this gas component. Moreover, in the considered regions the molecular gas is dominant over the atomic one and, contrary to this latter, its azimuthally averaged distribution follows closely the radial profiles of the main SFR indicators (Tacconi and Young 1986, Young & Scoville 1991). Rownd & Young (1999; hereafter RY99) conducted an H$_{\alpha}$ imaging of 121 of these galaxies, determining the local relationship between the SFR and the molecular gas. They found a correlation between these two quantities and concluded that for face-on spiral, in general, there are no strong SFE gradients across the star-forming disks. The majority of large SFE variations they found are seen between adjacent disk points, reflecting regional differences in the SFE, and any radial gradients are at most a secondary effect. In contrast, they pointed out that consistent radial variations (up to an order of magnitude or more) of the SFE exist within many highly inclined galaxy disks. They attributed the decreasing SFE towards the centers of these galaxies to a large amount of dust extinction on the luminosity.
Adler et al. (1991) found a correlation between the radio continuum flux density at 20 cm and the CO line emission on global scales for a sample of 31 spiral galaxies. They also studied the relationship of these two quantities within the disks of 8 nearby well resolved spiral galaxies, finding that their ratio is constant both inside the same galaxy and from galaxy to galaxy.\
The work we present here is complementary to the analysis of RY99 and extends that of Adler et al. (1991). We combined the radio continuum images at 1.4 GHz from the NRAO VLA Sky Survey (NVSS, Condon et al. 1998) with the FCRAO CO survey to study the relationship between the radio continuum and the molecular gas point-to-point within the disks of 180 star-forming spiral galaxies. It is important to stress that we are comparing two homogeneous data set with the same angular resolution of 45.\
The paper is organized as follows: in Sect. 2 and Sect. 3 we present the sample used and we describe the data analysis, respectively. In Sect. 4 we present the results of the statistical analysis and in Sect. 5 we discuss the results obtained.\
We use a Hubble constant H$_0$=50 km s$^{-1}$Mpc$^{-1}$ throughout the paper.
Sample selection
================
To investigate the star formation law within the disks of normal galaxies, we combined the data from two public surveys. We use the NVSS for the radio continuum and the FCRAO CO survey for the molecular gas, respectively.
The NVSS was performed at 1.4 GHz with the Very Large Array (VLA) in D configuration. It has an angular resolution of 45 (FWHM), a noise level of 0.45 mJy/beam (1$\sigma$) and covers all the sky north of declination –40. The shortest baseline is 35 m, corresponding to $\simeq$167$\lambda$, therefore structures up to about 10 in angular size are properly imaged.
The FCRAO CO survey comprises 300 galaxies observed along the major axis of the disk for a total of 1412 locations. Most of the galaxies in the survey are spirals or irregulars north of declination –25. At the frequency of the CO $J=1-0$ transition (115.27 GHz) the FWHM of the 14-m FCRAO telescope is 45. The weakest line detected depends on the width of the line, and hence on the velocity field within the beam. The uncertainties on the individual line intensity vary from galaxy to galaxy. A conservative estimate of the rms noise, including the calibration, baseline removal, and the rms noise per channel is about 25% (the median signal-to-noise ratio is 4).
We note that the two surveys have uniform sensitivity and identical angular resolution. This fact circumvents the difficulties deriving from the comparison of data from multiple instruments or studies which are subtly incompatible either because of inconsistent signal-to-noise ratios or unmatched resolution.
The original FCRAO CO survey includes 300 galaxies selected from the RC2 (de Vaucouleurs et al. 1976) or the [*IRAS*]{} database satisfying at least one of the following criteria: i) $B^{0}_{\rm T} < 13.0$, ii) $S_{60} > 5$ Jy or iii) $S_{100} > 10$ Jy. Although the FCRAO CO survey is not a complete sample in terms of flux-density or volume limit, the observed galaxies cover a wide range of luminosity, morphology and environments. For this reason they represent an ideal database to study the behaviour of the star formation process and the molecular gas in a wide variety of conditions.
Since our interest was primarily to investigate the behaviour of star formation within the galaxy disks, we have selected, from the FCRAO CO survey, a sub-sample of 180 objects for which there were at least three different observations of the CO line in the disk.
{width="18cm"}
Distributions of morphological types, angular diameters, distances and linear diameters of the sample along with the corresponding distributions for the FCRAO CO survey are shown in Fig. \[fig1\].
The galaxies in our sample have morphological types ranging between S0 and I0. Most of them (89%) are spiral galaxies with morphological type between Sa-Sd.
The majority of the galaxies (about 90%) have an optical angular diameter (from RC2) $D_{25}\leq$10, the median angular diameter being about 5. This ensures that for all of them the NVSS properly recovered the flux density of the extended structures (see above). The remaining 20 galaxies have an optical angular diameter from 10 to 25. For these galaxies it is possible that the NVSS missed a significant fraction of flux from the extended structure. Because of its outstanding angular size, NGC598 (M33, $D_{25} = 62\arcmin$) had been excluded from the analysis.
The distances of the galaxies in our sample (taken from Young et al. 1995) span from the Local Group up to about 80 Mpc. Over 87 galaxies are at the distance of the Virgo Cluster (20 Mpc).
The linear diameters range from $\sim 4$ to $\sim 100$ kpc. The median value is 31 kpc.
Our selection excluded most of the galaxies with an angular diameter less than 3 , i.e. the intrinsically small galaxies (linear diameter smaller than 30 kpc) and the more distant ones.
The complete list of the galaxies in our study (including the galaxy name, Hubble type, inclination, angular diameter and distance) is available in electronic format at http://www.ira.bo.cnr.it/ $\sim$crapsi[\_]{}s/RADIOCO/Article/tab\_art.txt.
Data analysis
=============
For each galaxy in our sample we extracted the radio surface brightness, $B_{1.4}$, from the NVSS images . We averaged the NVSS pixel values within circular regions of 45 diameter, centered on the same pointing positions of the FCRAO survey. Following the same formalism adopted by RY99 for the emission, the radio brightness is expressed in terms of the ratio between the radio luminosity at 1.4 GHz, $L_{1.4}$, and the de-projected surface, $\Sigma$, intercepted by the beam on the galaxy disk: $$\frac{L_{1.4}~({\rm W~Hz^{-1}})}{\Sigma~(\rm kpc^{2})}= 3.2
\times 10^{18} ~B_{1.4}\left(\frac{\rm mJy}{\rm beam}\right)~\cos~i
\label{lsigmab}$$ where $i$ is the inclination angle of the galaxy with respect to the plane of the sky. The surface luminosity has the advantage to be distance independent as opposed to the global luminosity.
Since most of continuum radio luminosity of normal galaxies is produced by relativistic electrons accelerated by supernovae explosion and the supernova rate ($\nu_{\rm SN}$) is directly related to the SFR of massive stars, a relation is expected between the radio luminosity and the star formation rate. In the following we indicate with SFR the formation rate of stars with mass $M\geq 5 M_{\odot}$. Condon (1992) calibrated empirically the $\nu_{\rm SN}$-SFR relation using the supernova rate and the radio luminosity of our Galaxy: $$L_{\nu}~({\rm W~Hz^{-1}})=5.3\times10^{21}\nu_{\rm GHz}^{-\alpha}
\cdot{\rm SFR} ~({\rm M_{\odot}yr^{-1}})
\label{condon92}$$ where a Miller-Scalo (1979) initial mass function (IMF) is assumed and $\alpha$ is the radio spectral index ($S_{\nu}\propto \nu^{-\alpha}$). Condon (1992) noted that this relation should apply to the majority of normal galaxies in order to be consistent with the tight FIR/radio correlation widespreadly observed.
Assuming for the radio spectral index a typical value of $\alpha = 0.8$, from Eq. (\[lsigmab\]) and Eq. (\[condon92\]) the relation between the star formation rate per unit surface, $\Sigma_{\rm SFR}$, and the radio brightness at 1.4 GHz is found to be: $$\Sigma_{\rm SFR}\left({\rm \frac{M_{\odot}}{yr~kpc^{2}}}\right)=
8.0\times10^{-4}~B_{1.4}\left(\frac{\rm mJy}{\rm beam}\right)~\cos~i
\label{sfrb}$$
In order to derive the surface density from the FCRAO CO survey integrated CO intensity , we used the formula (RY99):
$$\Sigma_{{\rm H_{2}}}({\rm M_{\odot}~pc^{-2}})=8.5~I_{\rm CO}
({\rm K~km~s^{-1}})~\cos~i
\label{sigmah2}$$
In this formula it is assumed a constant linear conversion factor between the CO luminosity and mass equal to $X_{\rm CO}=2.8\times
10^{20}~{\rm ~H_{2}~cm^{-2}[K~km~s^{-1}]^{-1}}$ (Bloemen et al. 1986).\
A realistic estimate of the uncertainties in both $\Sigma_{\rm SFR}$ and $\Sigma_{{\rm H_{2}}}$ surface densities should consider several systematic effects. The relation between the SFR and the radio luminosity is based on many not well proved assumptions, such as the IMF thresholds and slope and the extrapolation of the $\nu_{\rm SN}$-SFR Milky Way relation to other galaxies. Cram et al. (1999) pointed out that different modelling of these parameters introduce scaling uncertainties up to a factor of 2. The dominant errors in the gas density are the variation on the CO/conversion factor. These variations can be as high as $\pm$40% for luminous spiral galaxies as those studied in this work (Devereux & Young 1991). Despite these uncertainties, the data provide very strong constrains on the form of the SFE because of the wide ranges of SFR and gas densities explored.
Consistently with the definition given by RY99, the radio SFE is calculated by the ratio of to . In terms of our observables ($B_{1.4}$ and ) the SFE is expressed by $${\rm SFE~(yr^{-1})}\equiv \frac{\Sigma_{\rm SFR}}{\Sigma_{\rm H_{2}}}=
9.4\times 10^{-11} ~\frac{B_{1.4}}{I_{\rm CO}}
\label{SFE}$$ where $B_{1.4}$ and are measured in mJy/beam and K km s$^{-1}$, respectively.
The SFE defined by Eq. (\[SFE\]) gives the fraction of molecular gas converted to massive stars per year. Since the typical lifetime of the synchrotron radiating electrons is shorter than $10^{8}$ yr (Condon 1992), the SFR inferred from the radio luminosity traces a stellar population not older than this timescale (hereafter $\tau_{8}$). The percentage of molecular gas consumed over all this period is $${\rm SFE_{\tau_{8}} (\%)}=0.94 ~\frac{B_{1.4}}{I_{\rm CO}}
\label{SFEperc}$$
{width="17.5cm"}
{width="17.5cm"}
Results
=======
The data analysis presented in Sec. 3 allows one to compare the SFR and the molecular gas content to determine the star formation efficiency along the galaxy disks.
Overlays of optical (grayscale) and radio continuum (contours) images for eight selected galaxies[^1] are shown in the left columns of Figs. \[fig2\] and \[fig3\], where the circles indicate the positions and the beam size of the CO observation. In the middle columns we plot $B_{1.4}$ versus reporting also the corresponding values of $\Sigma_{\rm SFR}$ and $\Sigma_{{\rm H_{2}}}$ in the upper and right axis, respectively. The panels in the right columns show the radio brightness ($B_{1.4}$), the CO integrated intensity () and their ratio as a function of distance from the galaxy center. The convention is that radius is positive for positive right ascension (or declination) pointing shifts. The optical images are taken from the red Palomar Digitized Sky Survey. The NVSS radio contours start at 0.9 mJy beam$^{-1}$ ($2\sigma$) and are spaced by a factor of $\sqrt{2}$. In the plots, error bars and arrows indicate respectively measurement uncertainties and upper limits ($2\sigma$). In the middle column panels, the reference lines represent the mean (solid) and standard deviation (short-dashed) of the SFE computed using all the detections in the whole sample (see Sect. 4.2). In right column panels, the dashed and continuous lines show the $B_{1.4}$ and trends, respectively.
![Histogram of star formation efficiency at 628 pointing positions for the entire sample. Only points with a signal-to-noise ratio greater than 2$\sigma$ have been considered. Mean and dispersion refer to the logarithmic values of , see text. Values greater than 100% indicate present SFRs that cannot be supported for more than $10^{8}$ yr.[]{data-label="fig4"}](fig4.ps){width="9cm"}
The SFE within galaxy disks
---------------------------
Most of the galaxies in our sample are at distances between 10 and 40 Mpc, the corresponding linear resolution of the 45 beam is between 2.2 and 8.7 kpc, i.e. so large to sample the contribution from many molecular clouds complexes. In the following, one should keep in mind that the star-forming regions we are observing are comparable with the width of the spiral arms only for the nearest objects.
The galaxies shown in Figs. 2 and 3 are representative of the diversity of behaviour seen in the distributions of $B_{1.4}$ and within galaxy disks.
The most striking feature is the linear correlation between these two quantities observed for many edge-on and face-on galaxies (see Fig. 2). In these cases, $B_{1.4}$ and present the same scaling from the galaxy center outward, resulting in a constancy of the SFE along the disk. However, there are clear examples of disks characterized by systematic SFE trends (see Fig. 3). A 14 gradient steeper (flatter) with respect to one implies a SFE decreasing (increasing) with radius (e.g. NGC5236 and NGC5247). Calculating the SFR from non-thermal radio continuum allows us to include in the analysis high inclined galaxies, such as NGC1055 and NGC3079, which generally suffer from extinction in the optical band (see Sect. 5.1). By fitting a power law of the form $B_{1.4}\propto {I_{\rm CO}}^{N}$, we found that the fraction of linear ($0.5< N <2$) correlations is 67%, while the fractions of super-linear ($N>2$) and sub-linear ($N<0.5$) correlations are 23% and 10%, respectively. We examine the composite correlation including all the pointings in the sample in Sect. 4.3.
![Histograms of star formation efficiency sorted by galaxy type. Only points with a signal-to-noise ratio greater than 2$\sigma$ have been considered. Mean and dispersion refer to the logarithmic values of , see text.[]{data-label="fig5"}](fig5.ps){width="9cm"}
The SFE variation among galaxies
--------------------------------
Having established a correlation between the radio continuum surface brightness and the CO line intensity within galaxies, we investigated the variation of the SFE among galaxies. Fig. 4 shows the distribution of the for the galaxy sample. varies from about 0.1 up to more than 100%. Values greater than 100% indicate present SFRs that cannot be supported for more than $10^{8}$ yr. Since the distribution of logarithmic values is more Gaussian we calculated the mean and the dispersion of $\log$(). The mean of $\log$() is approximately 3.5% of the gas consumed per 10$^{8}$ yr and the rms of the distribution is slightly less than a factor of 3. mean and dispersion of the sample are traced in middle plots of Figs. 2 and 3 as solid and dashed reference lines, respectively.
We examined also the variation of among galaxies compared to the morphological type (see Fig. 5).The mean star formation efficiency varies weakly (about 25%) with the morphological type going from S0 to Scd galaxies.
![Histograms of star formation efficiency for nuclear (top panel) and disk (bottom panel) pointings. The solid line and solid portion of the histograms indicate normal and Seyfert galaxies, respectively. Three galaxies have nuclear greater than 100%; namely NGC2655, NGC3310 and NGC4151. NGC2655 and NGC4151 are classified as Sy2 and Sy1.5 and have nuclear of 100% and 170%, respectively. The galaxy NGC3310 is a peculiar starburst galaxy characterized by an exceptional nuclear of about 135%[]{data-label="fig6"}](fig6.ps){width="9cm"}
Larger SFE close to the galaxy centers might be attributed to the presence of an active nucleus (AGN). Fig. 6 shows the distribution of disk and nuclear in Seyfert and normal galaxies. Seyferts have a slightly higher than normal galaxies. The mean of the log , for Seyfert and normal galaxies, is: 4.1% and 3.3% in the disks and 5.5% and 4.3% in the nuclei, respectively. Most Seyferts have a nuclear comparable with normal galaxies. Only in few cases the nuclear emission is dominated by the radio source related with the AGN, e.g. NGC1068 (Wilson & Ulvestad 1987), NGC4151 (Pedlar et al. 1993), NGC2655 (Keel & Hummel 1988) and NGC4388 (Irwin et al. 2000). We conclude that the AGN-related emission affects only marginally the estimate of the in the nuclear pointings for most Seyfert in our sample.
We further investigate the behaviour of the SFE with respect to the galaxy inclination and size, and beam linear resolution. Fig. 7 shows the maximum variation of the , defined as SFE$_{\rm max}$/SFE$_{\rm min}$, inside each galaxy. Most galaxies present SFE variation up to a factor 6, the median variation being a factor of 2.5.
![Star formation efficiency variations within each galaxy as a function of inclination (top panel) and linear resolution of the 45 beam (bottom panel). Seyfert galaxies are distinguished with open dots. The majority of galaxies have internal SFE variation less than a factor of 3. The starburst galaxies NGC3034 (M82) and NGC520 and the HII galaxy NGC6503, have exceptional internal SFE variations (larger than a factor of 30).[]{data-label="fig7"}](fig7.ps){width="9cm"}
Fig. 7 shows that the internal SFE variations are not strong function of the galaxy inclination or linear resolution of the observing beam. Considerable SFE variations occur also for size of the beam greater than 5 kpc, i.e. smoothing over large regions of the galaxy disks. This fact was already noted by RY99. Ten galaxies show an internal SFE variation greater than a factor of 10. These are: the circumnuclear starbursts IC342, NGC253, NGC520, NGC660, NGC2146 and NGC3034 (see Kennicutt 1998); the Seyfert galaxies NGC2841 and NGC3368; the peculiar galaxy NGC3628; the HII galaxy NGC6503. M82, NGC520 and NGC6503, show exceptional internal SFE variations larger than a factor of 30. In particular, the nearby starburst M82 show a variation of about 2.6 order of magnitude.
Finally, we investigated the behaviour of the star formation efficiency as a function of the distance from the galaxy centers. Fig. 8 shows the SFE as a function of radius for all galaxies. The SFE is found to be approximately constant at all radii.
![Star formation efficiency as a function of radius for all galaxies. Limits are not shown. All the detections are represented as small points. Bold points and error bars represent the median value and the 50% of objects in each bin, respectively.[]{data-label="fig8"}](fig8.ps){width="8.5cm"}
The composite radio Schmidt law
-------------------------------
Our data offer the possibility to determine the [*spatially resolved*]{} Schmidt law using the radio continuum as SFR indicator for a sample of galaxies six times larger than the one of Adler et al. (1991). Fig. 9 shows the relationship between the SFR and the surface densities for all the pointings of the sample. Dashed and dotted reference lines indicate the mean and standard deviation of the for all the detection in the sample (see Sect. 4.1). The galaxies NGC4302, NGC4526, NGC4710 and NGC4866 have been omitted since they have inclination $i=90\degr$. A clear correlation is evident over more than three orders of magnitude, albeit a consistent scatter of 0.5 dex is present. We stress once more that, since this is a correlation between two brightness, we can rule out any subtle effect introduced by the distance. This result confirms the correlation found by Adler et al. (1991) on global scales. The best weighted fit of the radio Schmidt law, $\Sigma_{{\rm SFR}}=a \times {\Sigma_{{\rm H_{2}}}}^{N}$, is: $a=2.6(\pm1.2)\times 10^{-4}$ and $N=1.3\pm 0.1$. The value for the exponent of the radio Schmidt law is fully consistent with both the global (e.g. Kennicutt 1998) and local (RY99) slopes found using the line emission as SFR indicator. This exponent for the Schmidt law indicates a SFE which weakly increases with the SFR.
![Spatially resolved radio Schmidt law for all the detections in the sample. Arrows indicate upper limits at 2$\sigma$. The solid line is the best fit of the composite radio Schmidt law: $\Sigma_{{\rm SFR}} \propto {\Sigma_{{\rm H_{2}}}}^{1.3}$. Dashed and dotted lines refer to the mean and standard deviation of the for all detections in the sample: 3.5% and a factor of 2.8 respectively. The galaxies NGC4302, NGC4526, NGC4710 and NGC4866 have been omitted since they have inclination $i=90\degr$.[]{data-label="fig9"}](fig9.ps){width="8.5cm"}
Discussion and conclusions
==========================
The most striking result of our study is the overall consistency of the SFE obtained from the 20 cm radio continuum as SFR indicator instead of the emission (RY99). Basically we found all the essential features observed by RY99: the constancy of the SFE both within and among galaxies; the weak dependence of the SFE on morphological type and linear size of the observing beam; the extent and slope of the composite Schmidt law. Nevertheless, some important differences arise. In this section we present in details the differences between the two works, discussing their nature and implications.
![ Comparison of SFRs surface densities deduced from 1.4 GHz luminosity (horizontal axis) and emission (vertical axis) for 102 galaxies of our sample in common with RY99. The various panels show different values of galaxy inclinations $i$, as indicated. There are 453 detections, 65 of which are upper limits (at 2$\sigma$). The reference dashed lines indicate 1:1 relations. The subsample of face-on galaxies ($i \leq 40\degr$) shows the best correlation, with a dispersion of a factor of 2.3. SFRs surface densities deduced by the for highly inclined galaxies ($i \geq 70\degr$) are systematically underestimated for $\Sigma_{\rm SFR} {\rm 1.4~GHz} \geq 3\times 10^{-3} ~{\rm M_{\odot}~yr
^{-1}kpc^{-2}}$.[]{data-label="fig10"}](fig10.ps){width="8.5cm"}
Star formation rate surface densities
-------------------------------------
Cram et al. (1998) checked the consistency between the SFR deduced from the radio luminosity with the rates predicted by other indicators, in particular by the luminosity. They concluded that the rates deduced from the radio continuum and the luminosities, although in broad agreement, are affected by two systematic deviations: respectively at low ($\leq 0.1 ~{\rm M_{\odot}yr^{-1}}$) and high ($\geq 20 ~{\rm M_{\odot}yr^{-1}}$) star formation rates, the luminosity overestimate and underestimate the SFR deduced by the radio luminosity. The deviation at low rates can be attributed in part to problems related to the zero-point luminosity calibration. They suggest that the deviation at high SFR could be attributed to a larger amount of extinction by dust in those objects undergoing strong star formation or to particular IMFs that favour low mass supernovae progenitors rather than high mass stars responsible for the .
By comparing our data set with that of RY99 we have the possibility to extend the consistency check between the surface star formation rate densities deduced from the radio continuum and emission over the same regions of galaxy disks.
![ Ratio of SFRs surface densities deduced from and 1.4 GHz luminosity for highly inclined (NGC1055 and NGC3079) and starburst galaxies (NGC2146, NGC3034). These examples dramatically illustrate that the SFRs surface density inferred from the emission towards the nuclei of these objects are underestimated by more than one order of magnitude.[]{data-label="fig11"}](fig11.ps){width="9cm"}
Using Eq. (2) of Cram et al. (1998), we calculated the SFR surface density from the brightness reported by RY99 through the formula: $$\Sigma_{\rm SFR}\left({\rm \frac{M_{\odot}}{yr~kpc^{2}}}\right) =
2.6\times10^{-2}~\mu({\rm H_{\alpha}})
\label{hasfr}$$ where the (de-projected) surface brightness, $\mu({\rm H_{\alpha}})$, is measured in ${\rm L_{\odot}~pc^{-2}}$. Eq. (2) of Cram et al. (1998) accounts for a correction of 1.1 magnitude in the luminosity to compensate for the mean extinction in (Kennicutt 1983a). Fig. 10 shows the direct comparison between the radio and SFRs inferred respectively from Eq. (\[sfrb\]) and Eq. (\[hasfr\]) for 453 position within 102 galaxies of the sample. All galaxies are shown in the top panel of Fig. 10. In general there is a broad agreement between the two SFR indicators. Considering the distribution of the logarithm values of the ratio $R=\Sigma_{\rm SFR(H_{\alpha})}/\Sigma_{\rm SFR(1.4 GHz)}$, we have that $R$ have a mean of 0.83 with a standard deviation of 2.8, i.e. the is underestimating the surface SFR as compared to continuum radio emission. Sorting the sample by galaxy inclination indicates that this effect is strongly correlated with galaxy orientation. For the face-on subsample ($i < 40\degr$) $R$ has a mean of 1.1 and a dispersion of a factor 2.3. At intermediate inclinations ($40\degr \leq i < 70\degr$) mean and dispersion of $R$ are 0.8 and a factor of 3, respectively. Finally, for highly inclined galaxies ($i \geq 70\degr$) mean and dispersion of $R$ are 0.7 and a factor of 2.9, respectively. In particular, SFR surface densities deduced by the for highly inclined galaxies are systematically underestimated for $\Sigma_{\rm SFR(1.4~GHz)}$ greater than $\sim 3\times 10^{-3} ~{\rm M_{\odot}~yr^{-1}kpc^{-2}}$. Fig. 11 shows two edge-on spirals, NGC3079 and NGC1055 (see Figs. 2 and 3) and two well known starburst galaxies, NGC2146 and M82, in which SFRs deduced by the emission are underestimated by more than a factor of about 10. As a consequence, the SFE variations deduced for these objects by RY99, on the basis of observations, can be attributed to extinction, as suspected by these authors.
These results have two important implications: i) the close correlation observed between $\Sigma_{\rm SFR(1.4~GHz)}$ and $\Sigma_{\rm SFR(H_{\alpha})}$ for face-on galaxies reinforces the use of the radio luminosity as SFR indicator not only on global but also on local scales; ii) extinction could significantly affect estimates based on emission for high SFRs in edge-on galaxies.
Although the star formation rates deduced from the emission are systematically underestimated compared to those deduced from the radio continuum, the mean SFE reported by RY99 for their entire sample (121 galaxies) is 4.3%, i.e. higher than the mean SFE deduced from the radio continuum for our entire sample of 180 galaxies. However, considering the 103 galaxies we have in common with RY99, the mean SFE deduced from the emission is about 3.1% in good agreement with our value.
![ Ratio of SFRs surface densities deduced from 1.4 GHz luminosity and emission as functions of the linear scale of the beam (top panel) and distance from galaxy center (bottom panel). Highly inclined galaxies ($i> 70\degr$) are represented with open dots.[]{data-label="fig12"}](fig12.ps){width="9cm"}
Non-thermal radio continuum scale-length
----------------------------------------
Discrete supernova remnants (SNR) themselves account only for $<10\%$ of the radio luminosity of normal galaxies (Ilovaisky & Lequeux 1972). Most of cosmic rays escape from their parent SNRs and diffuse along the galaxy disk during their characteristic lifetimes which are much larger than typical ages of SNRs ($\sim 10^{5}$ yr). Therefore one expects the non-thermal radio continuum emission to be smoothed over a larger region around the parent stellar population with respect to the emission. On scale smaller than the cosmic rays characteristic diffusion scale, $D_{\rm CR}$, the non-thermal radio continuum cannot provide a reliable estimate of the local SFR. On the other hand, averaging over regions larger than the typical spiral arm width leads to a dilution of the emission. Fig. 12 (top panel) shows the ratio between the SFR surface densities deduced from and from radio luminosity versus linear size of the observing beam. One notes that, apart for edge-on galaxies (where the is affected by extinction), the ratio $\Sigma_{\rm SFR(1.4~GHz)}$ to $\Sigma_{\rm SFR(H_{\alpha})}$ tends to increase with the beam size. In particular, the ratio is smaller or greater than one for beam sizes respectively below and above $\sim 5$ kpc. This is consistent with a scenario in which the characteristic linear scale of both cosmic rays diffusion scale and arm size is about 3 kpc. Higher resolution radio images are required to determine the lowest linear scale at which the radio continuum correlates with the emission and thus providing an estimate of $D_{\rm CR}$.
We found also that the ratio $\Sigma_{\rm SFR(1.4~GHz)}$ to $\Sigma_{\rm SFR(H_{\alpha})}$ on average does not depend on the distance from the galaxy centers (Fig. 12, bottom panel).
The nature of Schmidt law scatter
---------------------------------
The dispersion (standard deviation) of the composite radio Schmidt law shown in Fig. 9 is a factor of 2.8, which is smaller than the corresponding dispersion of a factor 4 of the Schmidt law (RY99). This is probably because the radio continuum does not suffer the effects of dust extinction. The noisiness of the correlation could reflect real variations of the mean Schmidt law, indicating that it should be regarded at best as a first order approximation of the star formation process, or it could be due to the many assumptions involved in the calculation of gas and SFR densities.
Variations of the $X_{\rm CO}$ factor from galaxy to galaxy can be advocated to explain a part of the correlation scatter. These could introduce uncertainties up to a factor of 2 in the gas density scale (Kennicutt 1998). Another possibility is that the extrapolation of the proportionality between CO luminosity and virial mass of giant molecular clouds observed in our own Galaxy and in nearby galaxies, which is the basis of molecular mass determinations in this and similar works, does not hold for all spiral galaxies.
![ Radio Schmidt law for Sa, Sbc and Sc (top panel) and S0-Sab, Scd, Sd-Sm and merging-irregular (bottom panel); limits are not shown. Dashed and dotted reference lines indicate the mean and the dispersion of the two sub-samples, respectively. The scatter of the Schmidt law is considerably reduced excluding extreme morphological types. The scatter of the correlations shown in top and bottom panels is a factor of 2.4 and 3.8, respectively. In top panel, the deviation from the correlation of the four points belonging to NGC1068 is due to the AGN-related radio emission of this Seyfert galaxy (see Sect. 4.2).[]{data-label="fig13"}](fig13.ps){width="8.75cm"}
However, by comparing SFR surface densities deduced from and from radio continuum luminosity we showed that, even in the face-on subsample, the $\Sigma_{\rm SFR(H_{\alpha})}-\Sigma_{\rm SFR(1.4~GHz)}$ relation itself is affected by a scatter of at least a factor of 2. Hence, the uncertainties of the SFR indicators alone can account for a consistent fraction of the Schmidt law scatter.
Fig. 13 shows the Schmidt law separately for Sb, Sbc and Sc and S0-Sab, Scd and Sd-Sm galaxies (only detections with signal-to-noise ratio greater than 2$\sigma$ are considered). The two subsamples have the same mean SFE of 3.5% but very different dispersions of a factor of 2.4 for the former and 3.8 for the latter, i.e. the scatter of the Schmidt law is considerably reduced excluding extreme morphological types. RY99 also found that the Schmidt law is considerably tightened with the exclusion of the irregular galaxies and mergers. These SFE variations around the mean Schmidt law can be attributed to the particular physical conditions and/or environments experienced by these objects (see e.g. starburst galaxies), but they could also be induced by observational effects further amplified by poor statistic. In fact, Scd galaxies which are characterized by the lower mean SFE in our sample (see also Fig. 5) behave consistently with other morphological types excluding IC342 for which the NVSS misses flux density from the extended structure, see Sec. 2.
Molecular gas consumption timescale
-----------------------------------
The star formation efficiency can be expressed in terms of gas consumption (or cycling) timescale. The molecular gas consumption timescale, $\tau_{\rm H_{2}}$, indicates the time needed to convert all the molecular gas in stars given a constant SFE. Expressed in Gyr, $\tau_{\rm H_{2}}$ is given by $$\tau_{\rm H_{2}}({\rm Gyr})=\frac{10}{\rm SFE_{\tau 8} (\%)}$$ The mean star formation efficiency found in this work for spiral galaxies yields a molecular gas consumption timescale of about $\tau_{\rm H_{2}}
\simeq 2.8$ Gyr. Given the dispersion of the SFE in the sample, most galaxies have gas cycling timescale between about 1 and 9 Gyr. At the lower end of the distribution we found galaxies with a $\sim 1\%$, such as NGC4535 (see Fig. 2). The corresponding molecular gas consumption timescale is almost comparable with the Hubble time. The shortest gas cycling timescale corresponds to galaxies with exceptional SFE like NGC3310 (see Fig. 3). Star formation rates in these galaxies are so high, compared to their gas supplies, that the gas cycling timescale is $\tau_{\rm H_{2}} \le 0.1$ Gyr.
It is interesting to investigate the behaviour of the so-called starburst galaxies with respect to the gas cycling timescale. In literature, galaxies have been classified as starbursts according to different criteria. Heckman et al. (1998) define a galaxy as starburst when it is hosting a star-forming event that dominates its bolometric luminosity, i.e. on the basis of the magnitude of the SFR. Alternatively, Shu (1987) and Young (1987) proposed a classification based on the efficiency of the star formation. In this latter definition a galaxy with a high SFR is not defined as starburst if the mass of gas available is enough to sustain the star formation rate and vice versa. Following RY99 we selected the galaxies hosting a region in which the SFE is enhanced by a factor of three compared to the mean of the sample; for these starburst regions $\tau_{\rm H_{2}} \le 1$ Gyr. Fig. 14 shows the usual $\Sigma_{{\rm SFR}} -
\Sigma_{{\rm H_{2}}}$ plane for all the detection in the sample along with three reference lines indicating the gas cycling timescales $\tau_{\rm H_{2}}=0.1$, 1 and 10 Gyr.
![ $\Sigma_{{\rm SFR}} - \Sigma_{{\rm H_{2}}}$ plane for all the detection in the sample (limits are not shown) along with three reference lines corresponding to different molecular gas consumption timescales, as indicated. Pointings of known starburst galaxies are represented by open dots. []{data-label="fig14"}](fig14.ps){width="9cm"}
Most galaxies known in literature as “starbursts” have consumption timescales comparable with those of normal spiral galaxies. In some cases, e.g. M82, starburst galaxies host both regions characterized by SFE lower and higher than the mean of the sample. On the other hand, galaxies with normal or even low SFR surface densities, such as the highly inclined galaxy NGC4631, should be regarded as starburst according to our selection criterion based on the SFE. Finally, there are some galaxies for which the SFE is so high that the gas cycling timescale is $\tau_{\rm H_{2}} \le 0.1$ Gyr, e.g. the starburst NGC3310. NGC3310 is one of the best examples of a local UV-bright starburst (see Conselice et al. 2000 and reference therein). The exceptional high SFE of this galaxy indicates that its interstellar medium is truly affected by an extraordinary star formation event.
Conclusions
-----------
The results of this work are the following:
1. There is a tight correlation between the 20 cm non-thermal radio continuum and the CO line intensity in a representative sample of 180 spiral galaxies. The correlation holds within and among the galaxies.
2. The mean star formation efficiency, i.e. the ratio of the radio SFR to the molecular gas densities, for our sample is $0.035\times
10^{-8}$ yr$^{-1}$ with a dispersion of a factor of 3. This corresponds to convert 3.5% of the available gas into stars on a time of $10^{8}$ yr.
3. The comparison of SFRs surface densities deduced from 1.4 GHz luminosity and from the emission for 102 galaxies, reveals that $\Sigma_{\rm SFR(1.4~GHz)}$ and $\Sigma_{\rm SFR(H_{\alpha})}$ are closely correlated for face-on galaxies ($i \leq 40\degr$), reinforcing the use of the radio radio luminosity as SFR indicator not only on global but also on local scales. SFRs surface densities deduced by the luminosity for highly inclined galaxies ($i \geq 70\degr$) are systematically underestimated for $\Sigma_{\rm SFR(1.4~GHz)} \geq 3\times 10^{-3} ~{\rm M_{\odot}~yr
^{-1}kpc^{-2}}$.
4. The star formation efficiency varies weakly (less than 25%) with the Hubble morphological type.
5. The variation of the SFE within individual galaxy disks is less than a factor of 3. The largest variations are found in starburst galaxies.
6. The SFE is found to be approximately constant as a function of distance from the galaxy centers.
7. The composite radio Schmidt law, star formation versus molecular gas content, extends for more than 3 order of magnitude with an exponent of 1.3.
8. Most galaxies known in literature as “starbursts” have consumption timescales comparable with those of normal spiral galaxies. In some cases, e.g. M82, starburst galaxies host both regions characterized by a SFE lower and higher than the mean of the sample. Furthermore, there are some galaxies for which the SFE is so high that the gas cycling timescale is $\tau_{\rm H_{2}} \le 0.1$ Gyr, e.g. NGC3310.
We thank R. Fanti, G. Grueff and M. Johnson who carefully read the manuscript and provided useful comments. We acknowledge the Italian Ministry for University and Scientific Research (MURST) for partial financial support (grant Cofin99-02-37). The National Radio Astronomy Observatory is operated by Associated Universities, Inc., under contract with National Science Foundation.
Adler, D.S., Allen, R.J., Lo, K.Y., 1991, ApJ 238, 475 Bloemen, J.B.G.L., Strong, A.W., Mayer-Hasselwander, H.A., et al., 1986, A&A, 154, 25 Condon, J.J., 1992, ARA&A, 30, 575 Condon, J.J., Cotton, W.D., Greisen, E.W., et al., 1998, AJ, 115, 1693 Conselice, C.J., Gallagher, J.S., Calzetti, D., et al., 2000, AJ, 119, 79
Cram, L., Hopkins, A., Mobasher, B., et al., 1998, ApJ, 507, 155 de Vaucouleurs, G., de Vaucouleurs, A., Corwin, J.R., 1976, Second Refrence Catalogue of Bright Galaxies (Austin: Univ. of Texas Press) (RC2) Devereux, N.A., & Young, J.S., 1991, ApJ, 371, 515 Heckman, T., Carmelle, R., Leitherer, C., et al., 1998, ApJ, 503, 646 Ilovaisky, S.A., & Lequeux, J., 1972, A&A, 20, 347 Irwin, J.A., Saikia, D.J., English, J., 2000, AJ, 119, 1592 Keel, W.C., & Hummel, E., 1988, A&A, 194, 90 Kennicutt, R.C., 1983a, ApJ, 272, 54 Kennicutt, R.C., 1998, ApJ, 498, 541 Lequeux, J., 1971, A&A, 15, 42 Miller, G.E., & Scalo, J.M., 1979, ApJS, 41, 513 Pedlar, A., Kukula, M.J., Longley, D.P.T., et al., 1993, MNRAS, 263, 471 Rownd, B.K., & Young, J.S., 1999, AJ, 118, 670 Schmidt, M., 1959, ApJ, 129, 243 Shu, F.H., 1987, in Star Formation in Galaxies, ed. C.J. Lonsdale Persson (NASA CP - 2466)(Washington: NASA), 263 Tacconi, L., & Young, J.S., 1986, ApJ, 308, 600 Wilson, A.S., & Ulvestad, J.S., 1987, ApJ, 319, 105 Young, J.S., 1987, in Star Formation in Galaxies, ed. C.J. Lonsdale Persson (NASA CP - 2466)(Washington: NASA), 197 Young, J.S., & Scoville, N.Z., 1991, ARA&A, 29, 581 Young, J.S., Xie, S., Tacconi, L., et al., 1995, ApJS, 98, 219
[^1]: The plots for all the galaxies in the sample can be found at the web page:\
http://www.ira.bo.cnr.it/$\sim$crapsi[\_]{}s/RADIOCO/Catalog
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The problem of a Klein-Gordon particle moving in equal vector and scalar Rosen-Morse-type potentials is solved in the framework of Feynman’s path integral approach. Explicit path integration leads to a closed form for the radial Green’s function associated with different shapes of the potentials. For $q\leq -1$, and $\frac{1}{2\alpha }\ln \left\vert q\right\vert
<r<+\infty $, the energy equation and the corresponding wave functions are deduced for the $l$ states using an appropriate approximation to the centrifugal potential term. When $-1<q<0$ or $q>0$, it is shown that the quantization conditions for the bound state energy levels $E_{n_{r}}$ are transcendental equations which can be solved numerically. Three special cases such as the standard radial Manning-Rosen potential $(\left\vert
q\right\vert =1)$, the standard radial Rosen-Morse potential $(V_{2}\rightarrow -V_{2},q=1)$ and the radial Eckart potential $(V_{1}\rightarrow -V_{1},q=1)$ are also briefly discussed.
Keywords: Rosen-Morse potential; Manning-Rosen potential; Green’s function; Path integral; Bound states.
PACS Nos. 03.65.-w; 03.65.Db
\*Corresponding author, E-mail: [email protected]
author:
- |
A Khodja, A Kadja, F Benamira and L Guechi\*\
Laboratoire de Physique Théorique, Département de Physique,
- 'Faculté des Sciences Exactes, Université des frères Mentouri,'
- 'Constantine 1, Route d’Ain El Bey, Constantine, Algeria'
title: 'Path integral solution for a Klein-Gordon particle in vector and scalar deformed radial Rosen-Morse-type potentials'
---
Introduction
============
The purpose of the present paper is to further study the problem of a spinless relativistic particle of mass $M$ and charge $(-e)$ in interaction with a mixed central potential consisting of a vector potential $V_{q}\left( r\right) $ and a scalar potential $S_{q}\left( r\right) $. The scalar potential is added to the rest mass and the whole can be interpreted as an effective position dependent mass. The potentials $V_{q}\left(
r\right) $ and $S_{q}\left( r\right) $ are equal and of the form (see, for example Ref. [@Kadja] and references therein)$$V_{q}\left( r\right) =S_{q}\left( r\right) =-\frac{V_{1}}{\cosh
_{q}^{2}(\alpha r)}-V_{2}\tanh _{q}(\alpha r), \label{a.1}$$where $V_{1}$, $V_{2}$, $\alpha $ and $q$ are four ajustable real parameters, $V_{1}$ and $V_{2}$ describe the depth of the potential well, $\alpha $ is the screening parameter related to the range of the potential characterized by the length $\alpha ^{-1}$. The expression (\[a.1\]) is defined in terms of $q-$deformed hyperbolic functions$$\sinh _{q}x=\frac{e^{x}-qe^{-x}}{2},\text{ \ \ \ \ }\cosh _{q}x=\frac{e^{x}+qe^{-x}}{2},\text{ \ \ \ \ }\tanh _{q}x=\frac{\sinh _{q}x}{\cosh _{q}x}\text{\ } \label{a.1a}$$which have been introduced for the first time by Arai [@Arai]. The deformation parameter $q$ being a non-zero real number. The expression ([a.1]{}) represents a variant of the deformed Schiöberg potential [Amrouche]{} or the Tietz potential [@Zhang] which serve as modeling potentials for diatomic molecules. In the case where $q=1$, the potentials (\[a.1\]) reduce to the usual radial Rosen-Morse potential that has been investigated from different points of view in the last decade [Akoshile,Ikhdair1,Setare,Ibrahim,Aguda,Chen,Hamzavi,Tan,Liu]{}. Also, for $-1\leq q<0$ or $q>0$, various methods have been used to solve the Klein-Gordon and Dirac equations [@Jia; @Yang; @Wang; @Ikhdair2; @Ghoumaid] with these same potentials. In particular, for $\left\vert q\right\vert =1$ and for a light modification of expression (\[a.1\]), the relativistic rotational-vibrational energies and the radial wave functions have been approximately calculated with the help of the supersymmetric WKB approach and through the resolution of the Klein-Gordon equation [@He]. More recently, the usefulness of the Manning-Rosen potential and the Rosen-Morse potential for calculations of the vibrational partition function to study the thermodynamic properties of diatomic molecules has been emphasized by Jia and co-workers [@Wang2; @Song].
In this study, we shall present a complete treatment of the bound state problem for the potentials (\[a.1\]) via the path integral approach for any real value of parameter $q\neq 0$.
The organization of the paper is as follows: in section 2, we formulate the radial Green’s function associated with any $l-$wave in the framework of Feynman’s path integral. In section 3, we construct the radial Green’s function associated with the deformed Manning-Rosen potential $\left(
q<0\right) $. When $q\leq -1$ and $\frac{1}{\alpha }\ln \left\vert
q\right\vert <r<+\infty $, we use an appropriate approximation to the centrifugal potential term to calculate the expression of the radial Green’s function for a state of orbital quantum number $l$. From this, we shall then obtain the equation for the energy spectrum and the normalized eigenfunctions. For $-1<q<0$, the $q-$deformed Manning-Rosen potential is converted into the standard Manning-Rosen potential which is defined on a half-line. In this case, the radial Green’s function, for $l=0$, is evaluated in a closed form by using the perturbation method which consists in adding a Dirac $\delta -$function perturbation to the standard Manning-Rosen potential and making the strength of this perturbation infinitely repulsive to create a totally reflecting boundary. In this fashion, we obtain the radial Green’s function for a particle moving on a half-line. From the poles of the Green’s function, we derive a transcendental equation for the energy levels. In section 4, the $q-$deformed Rosen-Morse potential $(q>0)$ is worked out in a similar way. We first transform the path integral associated with this potential into the one of the standard Rosen-Morse potential on a half-line and by means of the same Dirac $\delta -$function perturbation trick, we calculate the radial Green’s function and also obtain a transcendental equation for the $s-$state energy levels. In section 5, the standard radial Manning-Rosen potential $(\left\vert q\right\vert =1)$, the standard radial Rosen-Morse potential $(V_{2}\rightarrow -V_{2},$ $q=1)$ and the radial Eckart potential $(V_{1}\rightarrow -V_{1},$ $q=1)$ are considered as special cases. Section 6 will be a conclusion.
Green’s function
================
The Green’s function associated with a particle moving in a vector potential and a scalar potential with spherical symmetry can develop into partial waves [@Peak] in spherical coordinates:
$$G\left( \overset{\rightarrow }{r^{\prime \prime },t^{\prime \prime }},\overset{\rightarrow }{r^{\prime },t^{\prime }}\right) =\frac{1}{r^{\prime
\prime }\text{ }r^{\prime }\text{ }}\sum_{l=0}^{\infty }\frac{2l+1}{4\pi }G_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime
})P_{l}\left( \cos \Theta \right) ,\,\, \label{a.2}$$
where the radial Green’s function is given by
$$\begin{aligned}
G_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime }) &=&\frac{1}{2i}\dint\limits_{0}^{\infty }d\Lambda \text{ }\left\langle
r^{\prime \prime },t^{\prime \prime }\right\vert \exp \left[ \frac{i}{2}\left[ -P_{r}^{2}+(P_{0}-V_{q})^{2}\right. \right. \notag \\
&&\left. \left. -\frac{l(l+1)}{r^{2}}-(M+S_{q})^{2}\right] \Lambda \right]
\left\vert r^{\prime },t^{\prime }\right\rangle , \label{a.3}\end{aligned}$$
and $P_{l}\left( \cos \Theta \right) $ is the Legendre polynomial of degree $l$ in $\cos \Theta =\frac{\overrightarrow{r}^{\prime \prime }.\overrightarrow{r}^{\prime }}{r^{\prime \prime }r^{\prime }}=\cos \theta
^{\prime \prime }\cos \theta ^{\prime }+\sin \theta ^{\prime \prime }\sin
\theta ^{\prime }\cos (\phi ^{\prime \prime }-\phi ^{\prime }).$
In Feynman’s formulation [@Hibbs; @Kleinert], the radial Green’s function $G_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime })$ is explicitly expressed in the form of a path integral,$\ $
$$G_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime })=\frac{1}{2i}\dint\limits_{0}^{\infty }d\Lambda P_{l}(r^{\prime \prime },t^{\prime
\prime },r^{\prime },t^{\prime };\Lambda ), \label{a.4}$$
where
$$\begin{aligned}
P_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime };\Lambda
) &=&\text{ }\underset{N\rightarrow \infty }{\lim }\dprod\limits_{n=1}^{N}\left[ \dint dr_{n}dt_{n}\right] \text{ }\dprod\limits_{n=1}^{N+1}\left[
\dint \frac{d(P_{r})_{n}\text{ }d(P_{0})_{n}}{(2\pi )^{2}}\right] \notag \\
&&\times \exp \left[ i\text{ }\dsum\limits_{n=1}^{N+1}A_{1}^{n}\right] ,
\label{a.5}\end{aligned}$$
with the short time action $A_{1}^{n}$ given by
$$\begin{aligned}
A_{1}^{n} &=&-(P_{r})_{n}\Delta r_{n}+(P_{0})_{n}\Delta t_{n}+\frac{\varepsilon _{\Lambda }}{2}\text{ }\left[ -(P_{r})_{n}^{2}+\left(
(P_{0})_{n}-V_{q}(r_{n})\right) ^{2}\right. \notag \\
&&\left. -\frac{l(l+1)}{r_{n}^{2}}-(M+S_{q}(r_{n}))^{2}\right] , \label{a.6}\end{aligned}$$
in which the notation used is $\Delta r_{n}=r_{n}-r_{n-1},\Delta
t_{n}=t_{n}-t_{n-1},r_{n}=r(t_{n}),$ and $\varepsilon _{\Lambda }=\frac{\Lambda }{N+1}.$
Let us notice first that the integrations on the variables $t_{n}$ in the expression (\[a.5\]) give $N$ Dirac distributions $\delta \left(
(P_{0})_{n}-(P_{0})_{n+1}\right) $. Thereafter, performing the integrations on the variables $(P_{0})_{n}$, one finds that
$$(P_{0})_{1}=(P_{0})_{2}=...=(P_{0})_{N+1}=E. \label{a.7}$$
Consequently, the propagator (\[a.5\]) becomes
$$P_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime };\Lambda
)=\frac{1}{2\pi }\dint\limits_{-\infty }^{+\infty }dE\exp \left[
iE(t^{\prime \prime }-t^{\prime })\right] P_{l}(r^{\prime \prime },r^{\prime
};\Lambda ), \label{a.8}$$
with the kernel $P_{l}(r^{\prime \prime },r^{\prime };\Lambda )$ given by
$$P_{l}(r^{\prime \prime },r^{\prime };\Lambda )=\text{ }\underset{N\rightarrow \infty }{\lim }\dprod\limits_{n=1}^{N}\left[ \dint dr_{n}\right]
\dprod\limits_{n=1}^{N+1}\left[ \dint \frac{d(P_{r})_{n}}{2\pi }\right] \exp \left[ i\dsum\limits_{n=1}^{N+1}A_{2}^{n}\right] , \label{a.9}$$
where the new short time action is then$$\begin{aligned}
A_{2}^{n} &=&-(P_{r})_{n}\Delta r_{n}+\frac{\varepsilon _{\Lambda }}{2}\text{
}\left[ -(P_{r})_{n}^{2}+(E-V_{q}(r_{n}))^{2}\right. \notag \\
&&\left. -\frac{l(l+1)}{r_{n}^{2}}-(M+S_{q}(r_{n}))^{2}\right] .
\label{a.10}\end{aligned}$$Then, performing the integrations over the variables $(P_{r})_{n},$ we find
$$\begin{aligned}
P_{l}(r^{\prime \prime },r^{\prime };\Lambda ) &=&\frac{1}{\sqrt{2i\pi
\varepsilon _{\Lambda }}}\text{ }\underset{N\rightarrow \infty }{\lim }\dprod\limits_{n=1}^{N}\left[ \dint \frac{dr_{n}}{\sqrt{2i\pi \varepsilon
_{\Lambda }}}\right] \exp \left\{ i\dsum\limits_{n=1}^{N+1}\left[ \frac{\left( \Delta r_{n}\right) ^{2}}{2\varepsilon _{\Lambda }}\right. \right.
\notag \\
&&\left. \left. -\frac{\varepsilon _{\Lambda }}{2}\left( \left[
M+S_{q}(r_{n})\right] ^{2}-\left[ E-V_{q}(r_{n})\right] ^{2}+\frac{l(l+1)}{r_{n}^{2}}\right) \right] \right\} , \notag \\
&& \label{a.11}\end{aligned}$$
and substituting (\[a.8\]) into (\[a.4\]), we notice that the term depending on time $t$ does not contain the pseudo-time variable $\Lambda .$ Thus, we can rewrite the Green’s function (\[a.4\]) in the form: $$G_{l}(r^{\prime \prime },t^{\prime \prime },r^{\prime },t^{\prime })=\frac{1}{2\pi }\int_{-\infty }^{+\infty }dE\exp \left[ iE(t^{\prime \prime
}-t^{\prime })\right] G_{l}(r^{\prime \prime },r^{\prime }), \label{a.12}$$where
$$G_{l}(r^{\prime \prime },r^{\prime })=\frac{1}{2i}\int_{0}^{\infty }d\Lambda
P_{l}(r^{\prime \prime },r^{\prime };\Lambda ). \label{a.13}$$
By assuming that $V_{q}(r)=S_{q}(r),$ the radial Green’s function (\[a.13\]) reduces to
$$G_{l}(r^{\prime \prime },r^{\prime })=\frac{1}{2i}\int_{0}^{\infty }d\Lambda
\exp \left( i\widetilde{E}_{0}^{2}\Lambda \right) K_{l}(r^{\prime \prime
},r^{\prime };\Lambda ), \label{a.14}$$
where
$$\begin{aligned}
K_{l}(r^{\prime \prime },r^{\prime };\Lambda ) &=&\frac{1}{\sqrt{2i\pi
\varepsilon _{\Lambda }}}\underset{N\rightarrow \infty }{\lim }\dprod\limits_{n=1}^{N}\left[ \dint \frac{dr_{n}}{\sqrt{2i\pi \varepsilon
_{\Lambda }}}\right] \exp \left\{ i\dsum\limits_{n=1}^{N+1}\left[ \frac{\left( \Delta r_{n}\right) ^{2}}{2\varepsilon _{\Lambda }}\right. \right.
\notag \\
&&\left. \left. -\frac{\varepsilon _{\Lambda }}{2}\left( 2\left( E+M\right)
V_{q}(r_{n})+\frac{l(l+1)}{r_{n}^{2}}\right) \right] \right\} , \label{a.15}\end{aligned}$$
and
$$\widetilde{E}_{0}^{2}=\frac{E^{2}-M^{2}}{2}. \label{a.16}$$
The radial propagator (\[a.15\]) and the radial Green’s function ([a.14]{}) depend on the arbitrary real deformation parameter $q$. When $-1<q<0$ or $q>0$, the radial Green’s function (\[a.14\]) can be only evaluated exactly for the $s$ states, but, when $q\leq -1$, the $l-$state problem can be solved by using a proper approximation to the centrifugal potential term. So to undertake this study, three interesting cases must be distinguished according to the values of the deformation parameter $q$.
Deformed radial Manning-Rosen Potentials
========================================
First case: $q\leq -1.$
-----------------------
When $q\leq -1$, the potentials (\[a.1\]) are written in the form:
$$V_{q}\left( r\right) =S_{q}\left( r\right) =-\frac{V_{1}}{\sinh _{\left\vert
q\right\vert }^{2}(\alpha r)}-V_{2}\coth _{\left\vert q\right\vert }(\alpha
r). \label{a.17}$$
The motion takes place in the half-space $r>r_{0}=\frac{1}{2\alpha }\ln (-q)$. The figure 1 represents the variations with $(\alpha r)$ of the deformed Manning-Rosen potential (\[a.17\]) for three different $\left\vert
q\right\vert $ values. In order to construct the path integral for a state of orbital quantum number $l$, we first use the expression $$\frac{1}{r^{2}}\approx \alpha ^{2}\left( \frac{1}{3}+\frac{\left\vert
q\right\vert }{\sinh _{\left\vert q\right\vert }^{2}(\alpha r)}\right)
\label{a.18}$$as a good approximation for $\frac{1}{r^{2}}$ in the centrifugal potential term when $\left\vert q\right\vert \geq 1$ as it can be seen in Fig. 2 which contains a plot of $\frac{1}{(\alpha r)^{2}}$ and of $\left( \frac{1}{3}+\frac{\left\vert q\right\vert }{\sinh _{\left\vert q\right\vert }^{2}(\alpha
r)}\right) $ for some characteristic values of $\left\vert q\right\vert $. Note that the approximation (\[a.18\]) is equivalent to those of the literature [@Cui; @Guechi] for $\left\vert q\right\vert =1$. We next transform the variable $r\in \left] r_{0},+\infty \right[ $ into a new variable $x\in \left] 0,+\infty \right[ $ by
$$r=\frac{1}{\alpha }\left( x+\frac{1}{2}\ln \left\vert q\right\vert \right) .
\label{a.19}$$
In addition to the change of variable, we rescale the local time interval $\varepsilon _{\Lambda }$ to a new time interval $\alpha ^{-2}\varepsilon
_{s} $. Putting these considerations together, we can rewrite the Green’s function (\[a.14\]) as: $$G_{l}(r^{\prime \prime },r^{\prime })=-\frac{1}{2\alpha }G_{MR}^{l}\left(
x^{\prime \prime },x^{\prime };\widetilde{E}_{l}^{2}\right) , \label{a.20}$$with$$G_{MR}^{l}\left( x^{\prime \prime },x^{\prime };\widetilde{E}_{l}^{2}\right)
=i\int_{0}^{\infty }dS\exp \left( i\frac{\widetilde{E}_{l}^{2}}{\alpha ^{2}}S\right) K_{MR}^{l}\left( x^{\prime \prime },x^{\prime };S\right) ,
\label{a.21}$$where$$\widetilde{E}_{l}^{2}=\widetilde{E}_{0}^{2}-\frac{l(l+1)\alpha ^{2}}{6},
\label{a.22a}$$and$$K_{MR}^{l}\left( x^{\prime \prime },x^{\prime };S\right) =\int Dx(s)\exp
\left\{ i\int_{0}^{S}\left[ \frac{\overset{.}{x}^{2}}{2}-V_{MR}^{l}(x)\right]
ds\right\} \label{a.22b}$$is the propagator associated with the standard Manning-Rosen potential [Manning]{}$$V_{MR}^{l}(x)=-\widetilde{V}_{2}\coth x+\frac{\widetilde{V}_{1}}{\sinh ^{2}x};\text{ \ \ }x>0, \label{a.23}$$in which we have put$$\widetilde{V}_{1}=-(E+M)\frac{V_{1}}{\left\vert q\right\vert \alpha ^{2}}+\frac{l(l+1)}{2};\text{ \ \ \ \ \ \ \ }\widetilde{V}_{2}=(E+M)\frac{V_{2}}{\alpha ^{2}}. \label{a.24}$$
The Green’s function can be stated in a closed form as is known from the literature [@Grosche2; @Benamira1; @Benamira2]
$$\begin{aligned}
G_{MR}^{l}\left( x^{\prime \prime }\,\,,\,x^{\prime }\,;\widetilde{E\,}_{l}^{2}\,\right) &=&\frac{\Gamma (M_{1}-L_{E})\Gamma (L_{E}+M_{1}+1)}{\Gamma (M_{1}+M_{2}+1)\Gamma (M_{1}-M_{2}+1)} \notag \\
&&\times \left( \frac{2}{1+\coth x^{\prime }}.\frac{2}{1+\coth x^{\prime
\prime }}\right) ^{\frac{M_{1}+M_{2}+1}{2}} \notag \\
&&\times \text{ }\left( \frac{\coth x^{\prime }-1}{\coth x^{\prime }+1}.\frac{\coth x^{\prime \prime }-1}{\coth x^{\prime \prime }+1}\right) ^{\frac{M_{1}-M_{2}}{2}} \notag \\
&&\times \text{ }_{2}F_{1}\left( M_{1}-L_{E},L_{E}+M_{1}+1,M_{1}-M_{2}+1;\frac{\coth x_{>}-1}{\coth x_{>}+1}\right) \notag \\
&&\times \text{ }_{2}F_{1}\left( M_{1}-L_{E},L_{E}+M_{1}+1,M_{1}+M_{2}+1;\frac{2}{\coth x_{<}+1}\right) . \notag \\
&& \label{a.25}\end{aligned}$$
The quantities $L_{E},$ $M_{1}$ and $M_{2}$ are given by$$\left\{
\begin{array}{c}
L_{E}=-\frac{1}{2}+\frac{1}{2\alpha }\sqrt{(M+E)(M-E+2V_{2})+\frac{\alpha
^{2}}{3}l(l+1)}; \\
M_{1,2}=\sqrt{\left( l+\frac{1}{2}\right) ^{2}-2(M+E)\frac{V_{1}}{\alpha
^{2}\left\vert q\right\vert }}\pm \frac{1}{2\alpha }\sqrt{(M+E)(M-E-2V_{2})+\frac{\alpha ^{2}}{3}l(l+1)}.\end{array}\right. \label{a.26}$$
The energy spectrum for the bound states can be obtained from the poles of the Green’s function (\[a.25\]) or the poles of the Euler function $\Gamma (M_{1}-L_{E})$. These poles are given by $$M_{1}-L_{E}=-n_{r};\text{ \ }n_{r}=0,1,2,...\text{ .} \label{a.27}$$By inserting the values of $L_{E}$ and $M_{1}$ in (\[a.27\]), we obtain$$\begin{aligned}
M^{2}-E_{n_{r},l}^{2} &=&\frac{(M+E_{n_{r},l})^{2}V_{2}^{2}}{\alpha
^{2}\left( n_{r}+\frac{1}{2}+\sqrt{\left( l+\frac{1}{2}\right)
^{2}-2(M+E_{n_{r},l})\frac{V_{1}}{\alpha ^{2}\left\vert q\right\vert }}\right) ^{2}} \notag \\
&&+\alpha ^{2}\left( n_{r}+\frac{1}{2}+\sqrt{\left( l+\frac{1}{2}\right)
^{2}-2(M+E_{n_{r},l})\frac{V_{1}}{\alpha ^{2}\left\vert q\right\vert }}\right) ^{2} \notag \\
&&-\frac{\alpha ^{2}}{3}l(l+1). \label{a.28}\end{aligned}$$The corresponding wave functions are of the form:$$\begin{aligned}
u_{n_{r},l}^{q\leq -1}(r) &=&r\Psi _{n_{r},l}^{q\leq -1}(r) \notag \\
&=&N_{n_{r},l}\left( 1-\left\vert q\right\vert e^{-2\alpha r}\right)
^{\delta _{l}}\left( \left\vert q\right\vert e^{-2\alpha r}\right)
^{w_{l}}P_{n_{r}}^{(2w_{l},2\delta _{l}-1)}\left( 1-2\left\vert q\right\vert
e^{-2\alpha r}\right) , \notag \\
&& \label{a.29}\end{aligned}$$where$$\left\{
\begin{array}{c}
w_{l}=\frac{1}{2\alpha }\sqrt{M^{2}-E_{n_{r},l}^{2}-2\left(
M+E_{n_{r},l}\right) V_{2}+\frac{\alpha ^{2}}{3}l(l+1)}; \\
\delta _{l}=\frac{1}{2}+\sqrt{\left( l+\frac{1}{2}\right)
^{2}-2(M+E_{n_{r},l})\frac{V_{1}}{\alpha ^{2}\left\vert q\right\vert }}.\end{array}\right. \label{a.30}$$The normalization constant $N_{n_{r}}$ follows from the condition$$\int_{\frac{1}{2\alpha }\ln \left\vert q\right\vert }^{+\infty }\left\vert
u_{n_{r},l}^{q\leq -1}(r)\right\vert ^{2}dr=1. \label{a.31}$$The calculation leads to$$N_{n_{r},l}=\left[ \frac{4\alpha w_{l}(n_{r}+w_{l}+\delta _{l})}{n_{r}+\delta _{l}}\frac{n_{r}!\Gamma (n_{r}+2w_{l}+2\delta _{l})}{\Gamma
(n_{r}+2w_{l}+1)\Gamma (n_{r}+2\delta _{l})}\right] ^{\frac{1}{2}}.
\label{a.32}$$
Second case: $-1<q<0$
---------------------
In this case, the potential (\[a.17\]) is defined in the interval $\mathbb{R}
^{+}$. A plot of this potential for three different $\left\vert q\right\vert
$ values is shown in Fig. 3. The transformation $r=\frac{1}{\alpha }(x+\frac{1}{2}\ln \left\vert q\right\vert )$ converts $r\in \left] 0,+\infty \right[ $ into $x\in \left] -\frac{1}{2}\ln \left\vert q\right\vert ,+\infty \right[ $. This means that the kernel (\[a.22b\]) is the propagator describing the evolution of a particle in the presence of a Manning-Rosen-type potential on the half-line $x>x_{0}=$ $-\frac{1}{2}\ln \left\vert q\right\vert $. As we can not perform a direct path integration to evaluate the propagator ([a.22b]{}), the problem can be solved by a trick that consists in incorporating an auxiliary term potential defined by a Dirac $\delta $ function in equation (\[a.22b\]) to form an impenetrable barrier [Clark]{} at $x=x_{0}$. Since the approximation (\[a.18\]) is not suitable for $0<\left\vert q\right\vert <1$, we limit ourselves to the evaluation of the Green’s function associated with the $s-$waves. So the Green’s function (\[a.21\]), for $l=0$, becomes $$G_{MR}^{\delta }\left( x^{\prime \prime },x^{\prime };\widetilde{E}_{0}^{2}\right) =i\int_{0}^{\infty }dS\exp \left( i\frac{\widetilde{E}_{0}^{2}}{\alpha ^{2}}S\right) K_{MR}^{\delta }\left( x^{\prime \prime
},x^{\prime };S\right) , \label{a.33}$$where
$$K_{MR}^{\delta }\left( x^{\prime \prime },x^{\prime };S\right) =\int
Dx(s)\exp \left\{ i\int_{0}^{S}\left[ \frac{\overset{.}{x}^{2}}{2}-V_{MR}^{\delta }(x)\right] ds\right\} . \label{a.34}$$
The path integral (\[a.34\]) is the propagator of a particle which moves in a potential of the form:
$$V_{MR}^{\delta }(x)=V_{MR}^{0}(x)-\eta \delta (x-x_{0}), \label{a.35}$$
where $V_{MR}^{0}(x)$ is the expression of the potential (\[a.23\]) for $l=0$. As is quite clear, given the complicated form of the potential ([a.35]{}), the calculation of the Green’s function (\[a.33\]) can not be performed directly. We propose to apply the perturbation approach in order to express $\exp \left( i\eta \int_{s^{\prime }}^{s^{\prime \prime }}\delta
(x-x_{0})ds\right) $ in power series. Then, the propagator (\[a.34\]) can be written as:
$$\begin{aligned}
K^{\delta }\left( x^{\prime \prime }\,\,,\,x^{\prime }\,;\,S\,\right)
&=&K_{MR}^{0}\left( x^{\prime \prime }\,\,,\,x^{\prime }\,;\,S\,\right)
+\sum_{n=1}^{\infty }\frac{\left( i\eta \right) ^{n}}{n!}\overset{n}{\underset{j=1}{\Pi }}\left[ \dint_{s_{i}}^{s_{j+1}}ds_{j}\dint_{-\infty
}^{\infty }dx_{j}\right] \notag \\
&&\times \text{ }K_{MR}^{0}\left( x_{1}\,\,,\,x^{\prime
}\,;\,s_{1}-s_{i}\,\right) \delta \left( x_{1}-x_{0}\right) \text{ }K_{MR}^{0}\left( x_{2}\,\,,x_{1}\,;\,s_{2}-s_{1}\,\right) \notag \\
&&\times ...\times \delta \left( x_{n-1}-x_{0}\right) \text{ }K_{MR}^{0}\left( x_{n}\,\,,x_{n-1}\,;\,s_{n}-s_{n-1}\,\right) \notag \\
&&\times \delta \left( x_{n}-x_{0}\,\right) \text{ }K_{MR}^{0}\left(
x^{\prime \prime }\,\,,\,x_{n}\,;\,S-s_{n}\,\right) \notag \\
&=&K_{MR}^{0}\left( x^{\prime \prime }\,\,,x^{\prime }\,;\,S\,\right)
+\sum_{n=1}^{\infty }\left( i\eta \right)
^{n}\dint_{s_{i}}^{s_{f}}ds_{n}\dint_{s_{i}}^{s_{n}}ds_{n-1}...\dint_{s_{i}}^{s_{2}}ds_{1} \notag \\
&&\times \text{ }K_{MR}^{0}\left( x_{0}\,\,,x^{\prime
}\,;\,s_{1}-s_{i}\,\right) K_{MR}^{0}\left(
x_{0}\,\,,\,x_{0}\,;\,s_{2}-s_{1}\,\right) \times ... \notag \\
&&\times \text{ }K_{MR}^{0}\left(
x_{0}\,\,,\,x_{0}\,;\,s_{n}-s_{n-1}\,\right) K_{MR}^{0}\left( x^{\prime
\prime }\,,x_{0}\,;S-s_{n}\,\,\right) , \label{a.36}\end{aligned}$$
where we ordered the time as follows: $s^{\prime
}=s_{0}<s_{1}<s_{2}<...<s_{n}<s_{n+1}=s^{\prime \prime }$. To perform the successive integrations on the variables $s_{j}$ in equation (\[a.36\]), we insert (\[a.36\]) into (\[a.33\]) and use the convolution theorem of Fourier transformation, then we obtain
$$G_{MR}^{\delta }\left( x^{\prime \prime }\,,\,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) =G_{MR}^{0}\left( x^{\prime \prime }\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) -\frac{G_{MR}^{0}\left( x^{\prime \prime
}\,,\,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) G_{MR}^{0}\left(
x_{0}\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) }{G_{MR}^{0}\left(
x_{0}\,,\,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) -\frac{1}{\eta }},
\label{a.37}$$
where $G_{MR}^{0}\left( x^{\prime \prime }\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) $ is the Green’s function (\[a.25\]) associated with the standard Manning-Rosen potential (\[a.23\]), for $l=0$.
If we now take the limit $\eta \rightarrow -\infty $, the physical system will be forced to move in the potential $V_{MR}^{0}\left( x\right) $ bounded by an infinitely repulsive barrier [@Grosche2; @Clark] located at $x\,\,=\,x_{0}$. In this case, the Green’s function is then given by:
$$\begin{aligned}
\tilde{G}_{MR}^{0}\left( x^{\prime \prime }\,\,,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) &=&\lim_{\eta \rightarrow -\infty }G_{MR}^{\delta }\left(
x^{\prime \prime }\,\,,\,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) \notag
\\
&=&G_{MR}^{0}\left( x^{\prime \prime }\,\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) -\frac{G_{MR}^{0}\left( x^{\prime \prime }\,\,,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) G_{MR}^{0}\left( x_{0}\,\,,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) }{G_{MR}^{0}\left( x_{0}\,\,,\,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) }\text{ }. \notag \\
&& \label{a.38}\end{aligned}$$
Finally, when $-1<q<0$, the radial Green’s function of our problem is expressed as:$$G_{0}(r^{\prime \prime },r^{\prime })=-\frac{1}{2\alpha }\tilde{G}_{MR}^{0}\left( x^{\prime \prime }\,\,,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) . \label{a.39}$$
The energy spectrum is determined from the poles of the expression ([a.38]{}), i.e., by the equation $G_{MR}^{0}\left( x_{0}\,\,,\,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) =0$, or as well by the transcendental equation$$_{2}F_{1}(\delta +w-p,\delta +p+w,2w+1;\left\vert q\right\vert )=0,
\label{a.40}$$where the quantities $\delta $, $p$ and $w$ are defined as$$\left\{
\begin{array}{c}
\delta =\frac{1}{2}\left( 1+\sqrt{1-\frac{8(E_{n_{r}}+M)V_{1}}{\alpha
^{2}\left\vert q\right\vert }}\right) ; \\
p=\frac{1}{2\alpha }\sqrt{(M+E_{n_{r}})(M-E_{n_{r}}+2V_{2})}; \\
w=\frac{1}{2\alpha }\sqrt{(M+E_{n_{r}})(M-E_{n_{r}}-2V_{2})}.\end{array}\right. \label{a.41}$$The equation (\[a.40\]) can be solved numerically to determine the discrete energy levels of the particle. The corresponding wave functions are of the form:$$\begin{aligned}
u_{n_{r}}^{-1<q<0}(r) &=&r\Psi _{n_{r}}^{-1<q<0}(r)=C\left( 1-\left\vert
q\right\vert e^{-2\alpha r}\right) ^{\delta }\left( \left\vert q\right\vert
e^{-2\alpha r}\right) ^{w} \notag \\
&&\times \text{ }_{2}F_{1}(\delta +w-p,\delta +p+w,2w+1;\left\vert
q\right\vert e^{-2\alpha r}), \label{a.42}\end{aligned}$$where $C$ is a constant factor. Note that these wave functions well satisfy the boundary conditions$$u_{n_{r}}^{-1<q<0}(r)\underset{\text{ }r\rightarrow 0}{\rightarrow }0\text{,
\ \ } \label{a.43}$$and$$u_{n_{r}}^{-1<q<0}(r)\underset{r\rightarrow \infty }{\rightarrow }0\text{.}
\label{a.44}$$
Deformed radial Rosen-Morse potentials
======================================
For $q>0$, the potentials (\[a.1\]) have the form of the deformed Rosen-Morse potential which is defined in the interval $\mathbb{R}
^{+}$. Figure 4 contains a plot of the deformed radial Rosen-Morse potential for six different $q$ values. In order to bring back the integral (\[a.14\]), for $l=0$, to a solvable form, we proceed as in the previous case. We perform the following coordinate transformation:
$$r\in
\mathbb{R}
^{+}\rightarrow x\in \left] -\frac{1}{2}\ln q,+\infty \right[ \label{a.45}$$
defined by$$r=\frac{1}{\alpha }\left( x+\frac{1}{2}\ln q\right) . \label{a.46}$$
After changing $\varepsilon _{\Lambda }$ into $\alpha
^{-2}\varepsilon _{s}$ or $\Lambda $ into $\alpha ^{-2}S$, we can write ([a.14]{}), for the $s$ states, in the following form:$$G_{0}(r^{\prime \prime },r^{\prime })=-\frac{1}{2\alpha }\tilde{G}_{RM}^{0}\left( x^{\prime \prime }\,\,,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) , \label{a.47}$$where $$\tilde{G}_{RM}^{0}\left( x^{\prime \prime }\,\,,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) =i\int_{0}^{\infty }dS\exp \left( i\frac{\widetilde{E}_{0}^{2}}{\alpha ^{2}}S\right) K_{RM}^{0}\left( x^{\prime \prime },x^{\prime
};S\right) , \label{a.48}$$and$$K_{RM}^{0}\left( x^{\prime \prime },x^{\prime };S\right) =\int Dx(s)\exp
\left\{ i\int_{0}^{S}\left[ \frac{\overset{.}{x}^{2}}{2}-\widetilde{V}_{2}\tanh x+\frac{\widetilde{V}_{1}}{q\cosh ^{2}x}\right] ds\right\} .
\label{a.49}$$The constants $\widetilde{V}_{1}$ and $\widetilde{V}_{2}$ are given by$$\widetilde{V}_{1}=(E+M)\frac{V_{1}}{\alpha ^{2}};\text{ \ \ \ \ \ \ \ }\widetilde{V}_{2}=-(E+M)\frac{V_{2}}{\alpha ^{2}}. \label{a.50}$$The propagator (\[a.49\]) has the same shape as the path integral relative to the potential originally introduced by Rosen and Morse to discuss the vibrational states of the polyatomic molecules [@Rosen] . The Rosen-Morse potential is defined for $x\in
\mathbb{R}
$, but in this case we have transformed the path integral for the potentials (\[a.1\]) into a path integral for a standard Rosen-Morse-type potential via the transformation $r\rightarrow r(x)$ which converts $r\in
\mathbb{R}
^{+}\rightarrow x\in ]-(1/2)\ln q,+\infty \lbrack $. This means that the motion of the particle takes place in the half-space $x>x_{0}=-(1/2)\ln q$. Then, to calculate the Green’s function relative to the $s-$waves, we proceed as in the previous case and we obtain
$$\begin{aligned}
\tilde{G}_{RM}^{0}\left( x^{\prime \prime }\,\,,x^{\prime }\,;\,\tilde{E}\,_{0}^{2}\right) &=&G_{RM}^{0}\left( x^{\prime \prime }\,\,,\,x^{\prime
}\,;\,\tilde{E}_{0}^{2}\,\right) -\frac{G_{RM}^{0}\left( x^{\prime \prime
}\,\,,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) G_{RM}^{0}\left(
x_{0}\,\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) }{G_{RM}^{0}\left(
x_{0}\,\,,\,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) },\text{ } \notag \\
&& \label{a.51}\end{aligned}$$
where $G_{RM}^{0}\left( x^{\prime \prime }\,\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) $ is the Green’s function associated with the standard Rosen-Morse potential [@Rosen]$$V_{RM}(x)=\widetilde{V}_{2}\tanh x-\frac{\widetilde{V}_{1}}{q\cosh ^{2}x};\text{ \ \ \ \ }x\in
\mathbb{R}
. \label{a.52}$$It is known that the solution by the path integral for this potential leads to the following expression of the Green’s function [Grosche2,Benamira1,Benamira2]{}
$$\begin{aligned}
G_{RM}^{0}\left( x^{\prime \prime }\,\,,\,x^{\prime }\,;\,\tilde{E}_{0}^{2}\,\right) &=&\frac{\Gamma (M_{1}-L_{E})\Gamma (L_{E}+M_{1}+1)}{\Gamma (M_{1}+M_{2}+1)\Gamma (M_{1}-M_{2}+1)} \notag \\
&&\times \text{ \ }\left( \frac{1-\tanh x^{\prime }}{2}\frac{1-\tanh
x^{\prime \prime }}{2}\right) ^{\left( M_{1}+M_{2}\right) /2} \notag \\
&&\times \text{ \ }\left( \frac{1+\tanh x^{\prime }}{2}\frac{1+\tanh
x^{\prime \prime }}{2}\right) ^{\left( M_{1}-M_{2}\right) /2} \notag \\
&&\times \text{ \ }_{\text{ }2}F_{1}\text{ }\left( M_{1}-L_{E},\text{ }L_{E}+M_{1}+1,\text{ }M_{1}-M_{2}+1;\text{ }\frac{1+\tanh x_{>}}{2}\right)
\notag \\
&&\times \text{ \ }_{\text{ }2}F_{1}\text{ }\left( M_{1}-L_{E},\text{ }L_{E}+M_{1}+1,\text{ }M_{1}+M_{2}+1;\text{ }\frac{1-\tanh x_{<}}{2}\right) ,
\notag \\
&& \label{a.53}\end{aligned}$$
with the notation$$\left\{
\begin{array}{c}
L_{E}=-\frac{1}{2}+\frac{1}{2}\sqrt{1+8\frac{(E+M)V_{1}}{\alpha ^{2}q}}; \\
M_{1,2}=\frac{1}{2\alpha }\left( \sqrt{(M+E)(M-E+2V_{2})}\pm \sqrt{(M+E)(M-E-2V_{2})}\right) .\end{array}\right. \label{a.54}$$
The bound state energy levels are determined from the poles of the Green’s function (\[a.51\]), i.e., by the equation $G_{RM}^{0}\left(
x_{0}\,\,,\,x_{0}\,;\,\tilde{E}_{0}^{2}\,\right) =0,$ or as well by the following quantization condition which is a transcendental equation involving the hypergeometric function
$$_{2}F_{1}\left( p+w-\delta +1,p+w+\delta ,2p+1;\frac{1}{1+q}\right) =0,
\label{a.55}$$
where the parameters $\delta $, $p$ and $w$ are defined by$$\left\{
\begin{array}{c}
\delta =\frac{1}{2}\left( 1+\sqrt{1+8\frac{(E_{n_{r}}+M)V_{1}}{\alpha ^{2}q}}\right) ; \\
p=\frac{1}{2\alpha }\sqrt{(M+E_{n_{r}})(M-E_{n_{r}}-2V_{2})}; \\
w=\frac{1}{2\alpha }\sqrt{(M+E_{n_{r}})(M-E_{n_{r}}+2V_{2})}.\end{array}\right. \label{a.56}$$The equation (\[a.55\]) can be also solved numerically.
Using the Green’s function (\[a.53\]) for the Rosen-Morse potential and the link between (\[a.47\]) and (\[a.53\]), we show that the wave functions corresponding to the bound states have the form:
$$\begin{aligned}
u_{n_{r}}^{q>0}(r) &=&r\Psi _{n_{r}}^{q>0}(r)=C\left( \frac{1}{1+qe^{-2\alpha r}}\right) ^{p}\left( \frac{q}{q+e^{2\alpha r}}\right) ^{w}
\notag \\
&&\times \text{ }_{2}F_{1}\left( p+w-\delta +1,p+w+\delta ,2p+1;\frac{1}{1+qe^{-2\alpha r}}\right) , \label{a.57}\end{aligned}$$
where $C$ is a constant factor. These wave functions are physically acceptable since they satisfy the boundary conditions $$u_{n_{r}}^{q>0}(r)\underset{\text{ }r\rightarrow 0}{\rightarrow }0,
\label{a.58}$$and
$$u_{n_{r}}^{q>0}(r)\underset{\text{ }r\rightarrow \infty }{\rightarrow }0.
\label{a.59}$$
Particular cases
================
First case: standard radial Manning-Rosen potentials
----------------------------------------------------
If we take $\left\vert q\right\vert =1$, the potentials (\[a.17\]) turn to the standard radial Manning-Rosen potential
$$V\left( r\right) =S\left( r\right) =-\frac{V_{1}}{\sinh ^{2}(\alpha r)}-V_{2}\coth (\alpha r). \label{a.60}$$
The energy equation and the normalized wave functions of the bound states can be deduced from expressions (\[a.28\]) and (\[a.29\])
$$\begin{aligned}
M^{2}-E_{n_{r},l}^{2} &=&\frac{(M+E_{n_{r},l})^{2}V_{2}^{2}}{\alpha
^{2}\left( n_{r}+\frac{1}{2}+\sqrt{\left( l+\frac{1}{2}\right)
^{2}-2(M+E_{n_{r},l})\frac{V_{1}}{\alpha ^{2}}}\right) ^{2}} \notag \\
&&+\alpha ^{2}\left( n_{r}+\frac{1}{2}+\sqrt{\left( l+\frac{1}{2}\right)
^{2}-2(M+E_{n_{r},l})\frac{V_{1}}{\alpha ^{2}}}\right) ^{2} \notag \\
&&-\frac{\alpha ^{2}}{3}l(l+1), \label{a.61}\end{aligned}$$
$$\begin{aligned}
u_{n_{r},l}^{\left\vert q\right\vert =1}(r) &=&r\Psi _{n_{r},l}^{\left\vert
q\right\vert =1}(r) \notag \\
&=&\widetilde{N}_{n_{r},l}\left( 1-e^{-2\alpha r}\right) ^{\widetilde{\delta
}_{l}}\left( e^{-2\alpha r}\right) ^{w_{l}}P_{n_{r}}^{(2w_{l},2\widetilde{\delta }_{l}-1)}\left( 1-2e^{-2\alpha r}\right) , \notag \\
&& \label{a.62}\end{aligned}$$
where
$$\left\{
\begin{array}{c}
w_{l}=\frac{1}{2\alpha }\sqrt{M^{2}-E_{n_{r},l}^{2}-2\left(
M+E_{n_{r},l}\right) V_{2}+\frac{\alpha ^{2}}{3}l(l+1)}; \\
\widetilde{\delta }_{l}=\frac{1}{2}+\sqrt{\left( l+\frac{1}{2}\right)
^{2}-2(M+E_{n_{r},l})\frac{V_{1}}{\alpha ^{2}}}.\end{array}\right. \label{a.63}$$
and
$$\widetilde{N}_{n_{r},l}=\left[ \frac{4\alpha w_{l}(n_{r}+w_{l}+\widetilde{\delta }_{l})}{n_{r}+\widetilde{\delta }_{l}}\frac{n_{r}!\Gamma
(n_{r}+2w_{l}+2\widetilde{\delta }_{l})}{\Gamma (n_{r}+2w_{l}+1)\Gamma
(n_{r}+2\widetilde{\delta }_{l})}\right] ^{\frac{1}{2}}. \label{a.64}$$
Second case: standard radial Rosen-Morse potentials
---------------------------------------------------
By setting $q=1$, and by changing $V_{2}$ in $\left( -V_{2}\right) ,$ the expression (\[a.1\]) becomes the so-called standard Rosen-Morse potential
$$V\left( r\right) =S\left( r\right) =-\frac{V_{1}}{\cosh ^{2}(\alpha r)}+V_{2}\tanh (\alpha r). \label{a.65}$$
The energy levels $E_{n_{r}}$ are deduced from (\[a.55\]) by the transcendental equation
$$_{2}F_{1}\left( p+w-\delta +1,p+w+\delta ,2p+1;\frac{1}{2}\right) =0,
\label{a.66}$$
and the non-normalized wave functions (\[a.57\]) become in this case:
$$\begin{aligned}
u_{n_{r}}^{q=1}(r) &=&r\Psi _{n_{r}}^{q=1}(r)=C\left( \frac{1}{1+e^{-2\alpha
r}}\right) ^{p}\left( \frac{1}{1+e^{2\alpha r}}\right) ^{w} \notag \\
&&\times \text{ }_{2}F_{1}\left( p+w-\delta +1,p+w+\delta ,2p+1;\frac{1}{1+e^{-2\alpha r}}\right) , \label{a.67}\end{aligned}$$
where $p$ is replaced by $w$ and conversely.
Third case: radial Eckart potentials
------------------------------------
By taking $q=1$, and by changing $V_{1}$ in $\left( -V_{1}\right) $, the potentials (\[a.1\]) reduce to the Eckart potential
$$V\left( r\right) =S\left( r\right) =\frac{V_{1}}{\cosh ^{2}(\alpha r)}-V_{2}\tanh (\alpha r). \label{a.68}$$
The quantization condition of energy levels and the non-normalized wave functions can be derived from equations (\[a.55\]) and (\[a.57\]). They are written respectively,
$$_{2}F_{1}\left( p+w-\overline{\delta }+1,p+w+\overline{\delta },2p+1;\frac{1}{2}\right) =0, \label{a.69}$$
and
$$\begin{aligned}
u_{n_{r}}^{q=1}(r) &=&r\Psi _{n_{r}}^{q=1}(r)=C\left( \frac{1}{1+e^{-2\alpha
r}}\right) ^{p}\left( \frac{1}{1+e^{2\alpha r}}\right) ^{w} \notag \\
&&\times \text{ }_{2}F_{1}\left( p+w-\overline{\delta }+1,p+w+\overline{\delta },2p+1;\frac{1}{1+e^{-2\alpha r}}\right) , \label{a.70}\end{aligned}$$
with $\overline{\delta }=\frac{1}{2}+\sqrt{1-8\frac{(E_{n_{r}}+M)V_{1}}{\alpha ^{2}}}.$
Conclusion
==========
In this article, we have solved the problem of a relativistic spinless particle in the presence of equal vector and scalar $q-$deformed radial Rosen-Morse-type potentials by path integration. This problem has only partially been discussed through the resolution of the Klein-Gordon equation [@Jia]. However, a complete solution can be given for any deformation parameter $q\neq 0$. As we have shown, the path integral for the Green’s function associated with this mixture of equal vector and scalar potentials can not be evaluated in a unified manner whatever the value of the deformation parameter. When $q\leq -1$ and $\frac{1}{\alpha }\ln
\left\vert q\right\vert <r<\infty $, the radial Green’s function for any $l$ state is directly calculated by using an appropriate approximation to the centrifugal potential term. The energy equation and the corresponding wave functions are then obtained. For $-1<q<0$ or $q>0$, we have limited ourselves to the evaluation of the Green’s functions for the $s$ waves $(l=0) $. We have shown that the transformation of the Green’s function relative to the starting potential defined on the interval $\mathbb{R}
^{+}$ into a path integral associated with a Manning-Rosen or Rosen–Morse-type potential reduces the problem to that of a particle forced to move in a half-space $x>x_{0}$. This problem with the Dirichlet boundary conditions is treated by using the perturbation approach. In both cases, the quantization conditions are transcendental equations involving the hypergeometric function which can be solved numerically to determine the energy levels of the bound states. The radial wave functions, expressed in terms of the hypergeometric functions are also derived.
[99]{} A Kadja, F Benamira and L Guechi *Indian J. Phys.* **91** 259 (2016)
A Arai *J. Phys. A: Math. Gen.* **34** 4281 (2001)
A Amrouche, A Diaf and M Hachama *Can. J. Phys.* **95** 25 (2017)
C S Jia, Y F Diao, X J Liu, P Q Wang, J Y Liu and G D Zhang *J. Chem. Phys.* **137** 014101 (2012)
K J Oyewumi and C O Akoshile *Eur. Phys. A* **45** 311 (2010)
S M Ikhdair *J. Math. Phys.* **51** 023525 (2010)
M R Setare and S Haidari *Phys. Scr.* **81** 015201 (2010)
T T Ibrahim, K J Oyewumi and S M Wyngaardt *Eur. Phys. J. Plus* **127** 100 (2012)
E V Aguda, *Can. J. Phys.* **91** 689 (2013)
T Chen, S R Lin and C S Jia *Eur. Phys. J. Plus* **128** 69 (2013)
S M Ikhdair and M Hamzavi *Chin. Phys. B* **22** 040302 (2013)
M S Tan, S He and C S Jia *Eur. Phys. J. Plus* **129** 264 (2014)
J Y Liu, X T Hu and C S Jia *Can. J. Chem.* **92** 40 (2014)
L Z Yi, Y F Diao, J Y Liu and C S Jia *Phys. Lett. A* **333** 212 (2004)
X Q Zhao, C S Jia and Q B Yang *Phys. Lett. A* **337** 189 (2005)
X C Zhang, Q W Liu, C S Jia and L Z Wang *Phys. Lett. A* **340** 59 (2005)
S M Ikhdair *J. Quantum Inf. Sc.* **1** 73 (2011)
A Ghoumaid, F Benamira and L Guechi *J. Math. Phys.* **57** 024102 (2016)
C S Jia, T Chen and S He *Phys. Lett. A* **377** 682 (2013)
C S Jia, L H Zhang and C W Wang *Chem. Phys. Lett.* **667** 211 (2017)
X Q Song, C W Wang and C S Jia *Chem. Phys. Lett.* **673** 50 (2017)
D Peak and A Inomata *J. Math. Phys.* **10** 1422** **(1969)
R P Feynman and A Hibbs *Quantum Mechanics and Path Integrals* (New York : Mc Graw Hill) (1965)
H Kleinert *Path Integrals in Quantum Mechanics, Statistics Polymer Physics and Financial Markets* ( Singapore : World Scientific) (2009)
C S Jia, T Chen and L G Cui *Phys. Lett. A* **373** 1621 (2009)
A Ghoumaid, F Benamira and L. Guechi *Can. J. Phys.* **91** 120 (2013)
M F Manning and N Rosen *Phys. Rev.* **44** 953 (1933)
C Grosche* J. Phys. A: Math. Gen.* **38** 2947 (2005)
F Benamira, L Guechi, S Mameri and M A Sadoun *J. Math. Phys.* **48** 032102 (2007)
F Benamira, L Guechi, S Mameri and M A Sadoun *J. Math. Phys.* **51** 032301 (2010)
T E Clark, R Menikoff and D H Sharp *Phys. Rev.* **D** **22** 3012 (1980)
N Rosen and P M Morse *Phys. Rev.* **42** 210 (1932)
Figure captions
Fig. 1. A plot of the Manning-Rosen potential (18) with $V_{2}=V_{1}/4$, for different values of $\left\vert q\right\vert \geq 1$.
Fig. 2. A plot of the expression $\left( 1/3+\left\vert q\right\vert /\sinh
_{\left\vert q\right\vert }^{2}\left( \alpha r\right) \right) $ compared to $\left( 1/\left( \alpha r\right) ^{2}\right) $, for different values of $\left\vert q\right\vert >0$.
Fig. 3. A plot of the Manning-Rosen potential (18) with $V_{2}=V_{1}/4$, for different values of $0<\left\vert q\right\vert <1$.
Fig. 4. A plot of the Rosen-Morse potential (1) with $V_{2}=V_{1}/4$, for different values of $q>0$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
A recent paper (gr-qc/9909017) criticizes our work on the structure of spacetime foam. Its authors argue that the quantum uncertainty limit for the position of the quantum clock in a gedanken timing experiment, obtained by Wigner and used by us, is based on unrealistic assumptions. Here we point out some flaws in their argument. We also discuss their other comments and some other issues related to our work, including a simple connection to the holographic principle. We see no reason to change our cautious optimism on the detectability of spacetime foam with future refinements of modern gravitational-wave interferometers like LIGO/VIRGO and LISA.
Key words: quantum spacetime foam; clocks; gravitational-wave detectors; foundations, theory of measurement, miscellaneous theories
PACS: 42.50.L; 04.80.N; 06.30.F; 03.65.Bz
address:
- |
(a) Center for Theoretical Physics, Laboratory for Nuclear Science and Department of Physics,\
Massachusetts Institute of Technology, Cambridge, MA 02139\
- |
(b) Institute of Field Physics, Department of Physics and Astronomy,\
University of North Carolina, Chapel Hill, NC 27599-3255\
author:
- '**Y. Jack Ng$^{(a),(b)}$[^1] and H. van Dam$^{(b)}$**'
title: '**On Wigner’s clock and the detectability of spacetime foam with gravitational-wave interferometers**'
---
Introduction
============
In their recent paper[@anos] “On the detectability of quantum spacetime foam with gravitational-wave interferometers,” Adler, Nemenman, Overduin, and Santiago claim that the way we use Wigner’s quantum clock [@wigner] in a gedanken timing experiment is not justified, thus casting doubt on the detectability of spacetime foam with gravitational-wave interferometers. In particular, they claim that the quantum uncertainty limit for the position of the quantum clock is actually much smaller than that obtained by Wigner and used by us. Since we [@nvd1; @nvd2] were the first to propose using Wigner’s clock to explore the quantum structure of spacetime and to conclude that classical spacetime breaks down into “quantum foam” in a manner quite different from the canonical picture [@MTW], we feel a special obligation to respond to the criticism and to clarify the physics behind our proposal. We will show that the arguments by Adler et al. are invalid.
But first, we should make it clear that we merely want to find out what the low-energy limit of quantum gravity can tell us about the structure of spacetime. For that purpose, it suffices to employ the general principles of quantum mechanics and general relativity. We have little to say about, and for this work, have no use for, the correct theory of quantum gravity (be it string theory, Ashtekar variables/loop-gravity formalism, or something else). We have in mind the low-energy limit of quantum gravity as manifested in the low-frequency spectrum of the displacement noise levels registered in the gravitational-wave interferometers.
In the next section, we recapitulate our previous work [@nvd1; @nvd2; @nvd3] on spacetime measurements and spacetime foam. In Section III, we respond to each of the four objections against our work raised by Adler et al. In Section IV, we answer some further questions which we think the readers may ask. We offer our conclusions in Section V. We point out that our spacetime foam model[@nvd1; @nvd2; @nvd3] is consistent with the holographic principle[@tHooft].
Space-time measurements and the foaminess of spacetime
=======================================================
Suppose we want to measure the distance between two separated points A and B. To do this, we put a clock (which also serves as a light-emitter and receiver) at A and a mirror at B. A light signal is sent from A to B where it is reflected to return to A. If the clock reads zero when the light signal is emitted and reads $t$ when the signal returns to A, then the distance between A and B is given by $l = ct/2$, where $c$ stands for the speed of light. The next question is: What is the uncertainty (or error) in the distance measurement? Since the clock at A and the mirror at B are the agents in measuring the distance, the uncertainty of distance $l$ is given by the uncertainties in their positions. We will concentrate on the clock, expecting that the mirror contributes a comparable amount to the uncertainty in the measurement of $l$. Let us first recall that the clock is not stationary; its spread in speed at
time zero is given by the Heisenberg uncertainty principle as $$\delta v = \frac{\delta p}{m} \gsim \frac{\hbar}{2m\delta l},
\label{ineq1}$$ where $m$ is the mass of the clock. This implies an uncertainty in the distance at time $t$, $$\delta l(t) = t \delta v \gsim \left(\frac{\hbar}{m \delta l(0)}\right)
\left(\frac{l}{c}\right),
\label{ineq2}$$ where we have used $t/2 = l/c$ (and we have dropped an additive term $\delta l(0)$ from the right hand side since its presence complicates the algebra but does not change any of the results). Minimizing $(\delta
l(0) + \delta l(t))/2$, we get the quantum mechanical uncertainty relation $$\delta l^2 \gsim \frac{\hbar l}{mc}.
\label{ineq3}$$
Next, we make use of the principle of equivalence, by exploiting the equality of the inertial mass and the gravitational charge of the clock, to eliminate the dependence on $m$ in the above inequality. This will promote the quantum mechanical uncertainty relation to a quantum gravitational uncertainty relation, making the uncertainty expression useful. Let the clock at A be a light-clock consisting of two parallel mirrors (each of mass $m/2$), a distance of $d$ apart, between which bounces a beam of light. On the one hand, the clock must tick off time fast enough such that $d/c \lsim \delta
l/c$, in order that the distance uncertainty is not greater than $\delta
l$: $\delta l \gsim d$. On the other hand, $d$ is necessarily larger than the Schwarzschild radius $Gm/c^2$ of the mirrors ($G$ is Newton’s constant) so that the time registered by the clock can be read off at all: $d \gsim \frac{Gm}{c^2}$. From these two requirements, it follows that $$\delta l \gsim \frac{Gm}{c^2},
\label{ineq4}$$ the product of which and Eq. (\[ineq3\]) yields the (low-energy) quantum gravitational uncertainty relation[@karol] $$\delta l \gsim (l l_P^2)^{1/3},
\label{ineq5}$$ where $l_P = (\frac{\hbar G}{c^3})^{1/2}$ is the Planck length. The intrinsic uncertainty in space-time measurements just described can be interpreted as inducing an intrinsic uncertainty in the space-time metric $g_{\mu \nu}$. Noting that $\delta l^2 = l^2 \delta g$ and using Eq. (\[ineq5\]) we get $$\delta g_{\mu \nu} \gsim (l_P/l)^{2/3}.
\label{delg}$$
The fact that there is an uncertainty in the space-time metric means that space-time is foamy. Eq. (\[ineq5\]) and Eq. (\[delg\]) constitute our model of spacetime foam. We note that even on the size of the whole observable universe ($\sim 10^{10}$ light-years), Eq. (\[ineq5\]) yields a fluctuation of only about $10^{-15}$ m. We further note that, according to our spacetime foam model, space-time fluctuations lead to decoherence phenomena. The point is that the metric fluctuation $\delta g$ induces a multiplicative phase factor in the wave-function of a particle (of mass $m$) $$\psi \rightarrow e^{i \delta \phi} \psi,
\label{wf}$$ given by $$\delta \phi = \frac{1}{\hbar} \int m c^2 \delta g^{00} dt.
\label{phase}$$ One consequence of this additonal phase is that a point particle with mass $m > m_P$ ($m_P \equiv \hbar / c l_P$ is the Planck mass) is a classical particle (i.e., it suffices to treat it classically).
Though the fluctuations that space-time undergoes are extremely small, recently Amelino-Camelia has argued (convincingly, we think) that modern gravitational-wave interferometers may soon be sensitive enough to test our model of space-time foam.[@amca] The idea is fairly simple. Due to the foaminess of space-time, in any distance measurement that involves an amount of time $t$, there is a minute uncertainty $\delta l \sim (ct l_{P}^2)^{1/3}$. But measuring minute changes in (the) relative distances (of the test masses or the mirrors) is exactly what a gravitational-wave interferometer is designed to do. Hence, the intrinsic uncertainty in a distance measurement for a time $t$ manifests itself as a displacement noise (in addition to other sources of noises) that infests the interferometers $$\sigma \sim (ct l_{P}^2)^{1/3}.
\label{noise}$$ We can write the displacement noise in terms of its Fourier transform, the associated displacement amplitude spectral density $S(f)$ of frequency $f$. For a frequency-band limited from below by the time of observation $t$, $\sigma$ is given in terms of $S(f)$ by[@radeka] $$\sigma^2 = \int_{1/t}^{f_{max}}[S(f)]^2 df.
\label{spden}$$ For the displacement noise given by Eq. (\[noise\]), the associated $S(f)$ is $$S(f) \sim f^{-5/6} (c l_{P}^2)^{1/3}.
\label{SD}$$ Since we are considering only the low-energy limit of quantum gravity, we expect this formula for $S(f)$ to hold only for frequencies much smaller than the Planck frequency ($c/l_P$).
We can now use the existing noise-level data[@abram] obtained at the Caltech 40-meter interferometer to put a bound on $l_P$. In particular, by comparing Eq. (\[SD\]) with the observed noise level of $3\times
10^{-19} {\rm mHz}^{-1/2}$ near 450 Hz, which is the lowest noise level reached by the interferometer, we obtain the bound $l_{P} \lsim 10^{-29}$ m which, of course, is consistent with the known value $l_P \sim 10^{-35}$ m. Since $S(f)$ goes like $f^{-5/6}$ according to Eq. (\[SD\]), we can look forward to the LISA generation of gravitational-wave interferometers for improvement by optimizing the performance at low frequencies. (We hope that the gain by going to lower frequencies will not be offset by other factors such as a much larger arm length of the interferometers.)
Reply to the comments by Adler et al
====================================
In this section we reply to the four points raised by Adler et al. in their paper [@anos] “On the detectability of quantum spacetime foam with gravitational-wave interferometers.”
\(1) Ref. [@anos] claims that if Wigner’s clock is quantum mechanical but not free, then the uncertainty limit becomes much smaller than that (Eq. (\[ineq3\])) obtained by Wigner and used by us. In particular, Adler et al. give the example of a quantum clock bound in a harmonic oscillator potential. These authors err in neglecting the fact that the clock is then bound to something. We can now consider that something to be part of the clock (after all, we have already considered the light emission and detection devices as part of our clock), and proceed with our argument presented in Section II.
\(2) Ref. [@anos] claims that Wigner’s limit is based on another unrealistic assumption: that the clock does not interact with the environment. In particular, Adler et al. point out that, if the clock is sufficiently large or complex, it will interact with its enviroment in such a way that its wave function decoheres. In addition, these authors claim, such interactions may localize or “collapse” the wave function, resulting in the clock wave function that does not spread linearly over macroscopic times, as opposed to what we have used in Eq. (\[ineq2\]). We admit that Adler et al. have raised a good point. But the question of wave function decoherence in the context of fundamental spacetime measurements is quite subtle. We think it is much more reasonable that the phenomenon of enviroment-induced decoherence is an outcome (rather than an input) of quantum spacetime measurements at the fundamental level, with gravity being the universal agent of quantum decoherence as argued in Section II and as emphasized by us[@nvd1; @nvd2].
\(3) Adler et al. observe that the existing noise-level data[@abram] obtained at the Caltech 40-meter interferometer can be used to place a lower limit on the effective mass of the hypothetical clock in Eq. (\[ineq3\]). They find the effective clock mass to be larger than 3 grams which, according to them, is such a remarkably large mass that it hardly seems plausible as a fundamental property of spacetime. This time, these authors err in forgetting that the length scale involved in the Caltech 40-meter interferometer measurement is macroscopic and has nothing to do with fundamental length scales. To appreciate this point, one can use Eq. (\[ineq3\]) and Eq. (\[ineq5\]) to show that the optimum mass for Wigner’s clock (optimum in the sense that it yields the smallest uncertainty in distance measurements) is given by $$m \sim m_P (l/l_P)^{1/3}.
\label{optm}$$ Thus the optimum mass of the quantum clock depends on $l$, the distance in the distance measurement. If $l$ is macroscopic, the optimum mass is much larger than the Planck mass. On the other hand, if we dare (recall that we expect our result to be valid only for the low-energy domain of quantum gravity) to use Eq. (\[optm\]) in the measurement of a microscopic distance approaching the fundamental length scale $l_P$, the optimum mass of the hypothetical clock would approach $m_P$, the fundamental mass scale. Therefore, the relatively large mass of the clock found in Ref. [@anos] is to be expected since the distance involved in the Caltech interferometer measurement at 450 Hz is huge compared to the Planck length.
The above three objections raised by Adler et al. are all directed at the quantum uncertainty limit (Eq. (\[ineq3\])) obtained by Wigner and used by us. There is a way, albeit an indirect one, to show that the uncertainty limit (Eq. (\[ineq3\])) actually should be quite palatable even to those who believe that the intrinsic uncertainty in distance measurements is independent of the distance being measured and is given simply by the Planck length. All it takes is to use Eq. (\[ineq3\]) as the starting point. But for the bound on $m$, instead of Eq. (\[ineq4\]), one uses $$l \gsim \frac{Gm}{c^2},
\label{cons}$$ which is nothing but the mathematical statement of the obvious observation that, to measure the distance from A to B, point B should not be inside the Schwarzschild radius of the clock at A. Then one finds $$\delta l \gsim l_P,
\label{Wheeler}$$ the canonical uncertainty[@MTW] in distance measurements. Thus the only question remaining is whether the more restrictive bound on $m$ given by Eq. (\[ineq4\]) is also correct. This brings us to the next comment.
\(4) Adler et al. note that the presence of the measurement clock system certainly produces a distortion of spacetime, but Eq. (\[ineq4\]) tells us that it also produces an uncertainty in spacetime distances of about the same amount. (This fact has not escaped our attention. See Ref. [@nvd1] and [@nvd2].) They contend that Eq. (\[ineq4\]) must be wrong. In particular, taking the spinning Earth as the quantum clock, they assert that, with Eq. (\[ineq4\]), one would conclude that objects in the vicinity of the Earth have a minimum intrinsic position uncertainty of roughly the Schwarzschild radius of the Earth, which is about 1 cm, and this is manifestly false by many orders of magnitude. Our reply is simply that one cannot use the spinning Earth, excellent as it is as a clock for daily lives, as Wigner’s quantum clock in the gedanken timing experiment. For one thing, the spinning Earth, by itself, cannot function as a clock. One also needs the Sun or the stars, for example, thus complicating the already huge timing device. As a clock, the spinning Earth is not very accurate. Let us imagine building a telescope with opening as large as the earth. For visible light, its resolving power is about $10^{-14}$ radians. The Earth and the telescope rotate by about $10^{-4}$ radians per sec. Hence, the spinning Earth, as a clock, cannot be precise beyond the $10^{-10}$ sec. level, which can be translated to yield a distance measurement accuracy to about 1 cm, hardly the precision needed for spacetime measurements at the fundamental level. Intuitively, it is also clear that a quantum spacetime measurement cannot tolerate the use of a monstrously huge and massive clock like the spinning Earth which causes such a distortion in the geometry of spacetime that it completely overwhelms the uncertainty in distance measurements. Recall that even on the size of the observable universe, the end result of our analysis yields a distance fluctuation of only about $10^{-15}$ m which is much smaller than the Schwarzschild radius of the Earth. It is true that we have merely used the light-clock as a model clock and there may be more ideal clocks to use; but due to its simplicity, the light-clock fits the bill of a quantum clock for the gedanken timing experiment at the fundamental level.
Comments on some other questions
================================
In this section, we comment on four more questions that we think some of our readers may ask.
\(1) In Section II, we require our light-clock to tick off time fast enough such that $d/c \lsim \delta l/c$, implying that $d/c$ is the smallest unit of time for our light-clock. Some readers may well ask whether it is not possible to have smaller units of time by taking fractions of $d/c$. If it is possible, then the inequality $d/c \lsim \delta l/c$ needs not hold. Our reply is that, to be accurate, $d/c$ is indeed the smallest unit of time for our light-clock. To make an analogy, one does not use a minute-clock to time a 100-m dash which takes only about 10 sec, a fraction of a minute.
\(2) Recall that our use of the light-clock in the gedanken timing experiment yields Eq. (\[ineq4\]). One may wonder if the ensuing result (Eq. (\[ineq5\])) is not just an artifact of our model clock. Thus it is logical to ask whether it is not possible to replace our light-clock with some other types of clocks such that all those inequalities (including the Schwarzschild bound) no longer hold. In the absence of explicit examples, it is hard to draw any conclusions. But let us consider a clock made of a small object revolving around a black-hole just outside its event-horizon. (And let us ignore the gravitational radiation problem.) Timing is provided by the periodicity of the motion. Then there is no analog of $d$, the separation of mirrors in the light-clock, and it follows that those inequalities are no longer valid, so goes the hypothetical argument. The trouble with this argument is that actually there is an analog of $d$, given by the size of the orbit around the black-hole; and mass of the clock here is that of the black-hole. Therefore it follows that those inequalities in Eqs. (\[ineq4\]) and (\[ineq5\]) (as an order of magnitude estimate) still hold.
\(3) In Eq. (\[spden\]), we have used $1/t$ as the lower limit of integration; but what if the lower limit is actually a multiple (call it n) of $1/t$? The answer is that, since Eq. (\[noise\]) holds only up to a multiplicative factor of order 1, a short calculation shows that so long as the multiple n is no more than 2 orders off unity, Eq. (\[spden\]) stands as it is.
\(4) One may worry that the metric fluctuations given by Eq. (\[delg\]) yield an unacceptably large fluctuation in energy density. Since we asked ourselves this very question and answered it in Ref. [@nvd2] already, we will be very brief here. But let us generalize the discussion to metric fluctuations of the form parametrized by $a$ with $0 < a \leq 1$ $$\delta g \gsim (l_P/l)^{a}
\label{gdelg}$$ (corresponding to distance uncertainties of $\delta l \gsim l^{1-a}
l_P^{a}$ and displacement amplitude spectral densities of $S(f) \sim
c^{1-a} l_P^{a} f^{a-3/2}$). Models with larger spacetime fluctuations are parametrized by smaller values of $a$. We note that, for our model of spacetime foam, $a=2/3$, while, for the canonical model, $a=1$. The case $a=1/2$ corresponds to the model of spacetime foam considered in Ref. [@GAC]. Regarding the metric fluctuation as a gravitational wave quantized in a spatial box of volume $V$, one finds[@nvd2] that the energy density is given by $$\rho \sim \left(\frac{m_P c^2} {V}\right),
\label{rho1}$$ for $1/2 < a \leq 1$. Thus the energy density associated with metric fluctuations given by Eq. (\[gdelg\]) is obviously small in the large volume ($V >> l_P^3$) limit which we have assumed. Note that the energy density is of the form given by Eq. (\[rho1\]) and holds, as an order of magnitude estimate (consistent with what we have been using), independent of the parameter $a$ so long as $a$ is not too close to 1/2. For $a=1/2$, one gets $$\rho \sim \left(\frac{m_P c^2}{V}\right) ln
\left(\frac{V^{1/3}}{l_P}\right).
\label{rho2}$$ For $0 < a < 1/2$, one finds $$\rho \sim \left(\frac{m_P c^2}{V}\right)
\left(\frac{V^{1/3}}{l_P}\right)^{1-2a}.
\label{rho3}$$ The trend is clear: in general, larger spacetime fluctuations cost more energy. Note that the energy density $\rho$ associated with metric fluctuations (Eq. (\[gdelg\])) is the smallest for the range of $a$ which includes the canonical model and our model of spacetime foam.
Conclusions
===========
In Ref. [@anos], Adler et al. raise four objections to our work on spacetime measurements and spacetime foam. Three of the objections are related to the question whether the quantum uncertainty limit obtained by Wigner and used by us is valid. These authors also criticize the gravitational uncertainty limit (Eq. (\[ineq4\])) obtained by us; they conclude that it is an artifact of our choice of a particular type of hypothetical clock and is, therefore, non-fundamental in nature. While they have raised some good points, we believe their argument is flawed (as shown in Section III). We agree that the question of an ideal quantum clock is not yet settled. But it is just inappropriate to use the spinning Earth as a quantum clock in a fundamental spacetime measurement. History has taught us that fundamental physics is best explored with simple devices; our light-clock is a simple device.
Since all the criticism by Adler et al. is related to issues of clocks, perhaps a better argument for a spacetime foam different from the canonical model is one that does not use clocks. As shown in Section IV, the energy density associated with spacetime quantum fluctuations takes on the smallest (and comparable) values for those spacetime foam models with the parameter $a$ in the range $1/2 < a \leq 1$ so long as $a$ is not too close to 1/2. So, it is possible that Nature chooses to have a larger spacetime fluctuation (than that predicted by the canonical model) at a comparable cost of energy. This argument is very loose, but hopefully we have made our point. Only future experiments can tell which value of $a$ (i.e., which spacetime foam model) Nature picks. At present, if we assume that the distance uncertainty expressions given above are not off by more than an order of magnitude, a short calculation shows that the existing data provided by the Caltech 40-meter interferometer rule out models with $a < 0.54$. We can expect more stringent bounds on $a$ with modern gravitational-wave interferometers.
There is one theoretical consideration which sets our model of spacetime foam ($a=2/3$) apart from the others. It is its connection to the holographic principle[@tHooft] which asserts that the number of degrees of freedom of a region of space is bounded (not by the volume but) by the area of the region in Planck units. To see that, let us consider a region of space with linear dimension $l$. According to the conventional wisdom, the region can be partitioned into cubes as small as $l_{P}^3$. It follows that the number of degrees of freedom of the region is bounded by $(l/l_P)^3$, i.e., the volume of the region in Planck units. But according to our spacetime foam model[@nvd1; @nvd2; @nvd3], the smallest cubes inside that region have a linear dimension of order $(l l_{P}^2)^{1/3}$. Accordingly, the number of degrees of freedom of the region is bounded by $[l/(ll_{P}^2)^{1/3}]^3$, i.e., the area of the region in Planck units, as stipulated by the holographic principle. Thus one may even say that the holographic principle has its origin in the quantum fluctuations of spacetime. Evidence for our spacetime foam model would lend experimental support for the holographic principle.
Finally we recall that spacetime (metric) fluctuations can be regarded as a kind of quantized gravitational waves. It is uncanny that, through future refinements, modern gravitational-wave interferometers like LIGO, VIRGO, and LISA, which are designed to detect gravitational waves from neutron stars, supernovae, black-holes, and the like, may also be able to detect, as a by-product, a very different kind of gravitational waves — the kind that encodes the quantum fluctuations of spacetime.
[**Acknowledgments**]{}\
One of us (YJN) thanks R. Weiss for a useful discussion. This work was supported in part by the U.S. Department of Energy under \#DF-FC02-94ER40818 and \#DE-FG05-85ER-40219, and by the Bahnson Fund of the University of North Carolina at Chapel Hill. Part of the work was carried out by YJN while he was on leave of absence at MIT. He thanks the faculty at the Center for Theoretical Physics for their hospitality.
R. J. Adler, I. M. Nemenman, J. M. Overduin, and D. I. Santiago, gr-qc/9909017.
E. P. Wigner, Rev. Mod. Phys. [**29**]{} (1957) 255; H. Salecker and E. P. Wigner, Phys. Rev. [**109**]{} (1958) 571.
Y. J. Ng and H. van Dam, Mod. Phys. Lett. [**A9**]{} (1994) 335; in Proc. of the Fundamental Problems in Quantum Theory, eds. D. M. Greenberger and A. Zeilinger, Ann. New York Acad. Sci. [**755**]{} (1995) 579.
Y. J. Ng and H. van Dam, Mod. Phys. Lett. [**A10**]{} (1995) 2801; Europhys. Lett. [**38**]{} (1997) 401.
C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation (W. H. Freeman, 1973), pp. 1190-1194.
Y. J. Ng and H. van Dam, gr-qc/9906003, to appear in Found. Phys.
G. ’t Hooft, in Salamfest 1993, p.284, gr-qc/9310026; L. Susskind, J. Math. Phys. [**36**]{} (1995) 6377.
A very different way to obtain this result can be found in F. Karolyhazy, IL Nuovo Cimento [**A42**]{} (1966) 390. See also N. Sasakura, Prog. Theor. Phys. [**102**]{} (1999) 169, and Preprint hep-th/0001161.
G. Amelino-Camelia, Nature [**398**]{} (1999) 216.
V. Radeka, Ann. Rev. Nucl. Part. Sci. [**38**]{} (1988) 217.
A. Abramovici, et. al., Phys. Lett. [**A218**]{} (1996) 157.
G. Amelino-Camelia, Mod. Phys. Lett. [**A9**]{} (1994) 3415. Also see L. Diosi and B. Lukacs, Phys. Lett. [**A142**]{} (1989) 331.
[^1]: Corresponding author. E-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Crowd counting from unconstrained scene images is a crucial task in many real-world applications like urban surveillance and management, but it is greatly challenged by the camera’s perspective that causes huge appearance variations in people’s scales and rotations. Conventional methods address such challenges by resorting to fixed multi-scale architectures that are often unable to cover the largely varied scales while ignoring the rotation variations. In this paper, we propose a unified neural network framework, named Deep Recurrent Spatial-Aware Network, which adaptively addresses the two issues in a learnable spatial transform module with a region-wise refinement process. Specifically, our framework incorporates a Recurrent Spatial-Aware Refinement (RSAR) module iteratively conducting two components: i) a Spatial Transformer Network that dynamically locates an attentional region from the crowd density map and transforms it to the suitable scale and rotation for optimal crowd estimation; ii) a Local Refinement Network that refines the density map of the attended region with residual learning. Extensive experiments on four challenging benchmarks show the effectiveness of our approach. Specifically, comparing with the existing best-performing methods, we achieve an improvement of 12% on the largest dataset WorldExpo’10 and 22.8% on the most challenging dataset UCF\_CC\_50.'
author:
- |
Jérôme Lang\
Laboratoire d’Analyse et Modélisation des Systèmes pour l’Aide à la Décision (LAMSADE)\
[email protected]
- |
Lingbo Liu$^1$, Hongjun Wang$^1$, Guanbin Li$^1$[^1], Wanli Ouyang$^2$, Liang Lin$^1$\
$^1$ School of Data and Computer Science, Sun Yat-sen University, Guangzhou, China\
$^2$ School of Electrical and Information Engineering, The University of Sydney, Sydney, Australia\
{liulingb,wanghq8}@mail2.sysu.edu.cn, [email protected],\
[email protected], [email protected]
bibliography:
- 'counting.bib'
title: 'Crowd Counting using Deep Recurrent Spatial-Aware Network'
---
Introduction
============
{width="0.925\columnwidth"}
\[fig:challenge\]
Crowd counting, which aims at estimating the total number of people in unconstrained crowded scenes, has received increasing research interests in recent years due to its potential application in many real-world scenarios, such as video surveillance [@xiong2017spatiotemporal] and traffic monitoring [@zhang2017fcn]. Despite recent progress, crowd counting remains a challenging issue, especially in the face of extreme occlusions, changes in lighting and camera perspective.
In recent two years, deep convolutional neural networks have been widely used in crowd counting and have made substantial progress [@sindagi2017cnn; @onoro2016towards; @sam2017switching; @xiong2017spatiotemporal; @zhang2015cross]. The success of most existing deep models stems from the effective modeling of the scale variation of people/heads and the varying degrees of regional crowd density. However, all of these methods, without exception, obtain the density estimation of the whole image by merging the prediction results of a number of pre-designed fixed sub-network structures (working on the whole image or image patches). Specifically, they are either proposed to fuses the features from multiple convolutional neural networks with different receptive fields to handle the scale variation of people groups [@zhang2016single], or directly divide crowd scene into multiple non-overlapping patches and provide a pool of regression networks for each patch selection [@sam2017switching]. Although these strategies can, to a certain extent, improve the adaptability of the prediction to images of diverse density regions or people of various scales, the limited enumeration of a fixed number of tailor-designed networks or several sizes of receptive fields can not well cope with all the scenarios after all. Most importantly, none of these algorithms take into account the impact of different pose and photographic angles on crowd density estimation while crafting their network structures. As shown in Figure \[fig:challenge\], the camera viewpoints in various scenes create different perspective effects and may result in large variation of scales, in-plane and out-plane rotation of people.
To address the aforementioned concerns, we propose a Deep Recurrent Spatial-Aware Network for crowd counting. The core of our network is a Recurrent Spatial-Aware Refinement (RSAR) module, which recurrently conducting refinement on an initial crowd density map through adaptive region selection and residual learning. Specifically, the RSAR module consists of two alternately performed components: i) a Spatial Transformer Network is incorporated in each LSTM step for simultaneously region cropping and warping, which allows the network to adaptively cope with the various degrees of congestion, people scale and rotation variation in the same scene; ii) a Local Refinement Network refines the density map of the selected region. In general, the main contributions of this work are three-fold.
- We provide an adaptive mode to simultaneously handle the effect of both scale and rotation variation by introducing a spatial transform module for crowd counting. To the best of our knowledge, we are the first to address the issue of the rotation variation on this task.
- We propose a novel deep recurrent spatial-aware network framework to recurrently select a region (with learnable scale and rotation parameters) from an initial density map for refinement, dependent on feature warping and residual learning.
- Extensive experiments and evaluations on several public benchmarks show that our proposed method achieves superior performance in comparison to other state-of-the-art methods. Specifically, compared with the existing best-performing methods, we achieve an improvement of 12% on the WorldExpo’10 dataset and 22.8% on the most challenging UCF\_CC\_50 dataset.
Related Work
============
**Deep learning Methods for Crowd Counting:** Inspired by the significant progress of deep learning on various computer vision tasks[@zhu2017learning; @chen2017recurrent; @li2017face; @chen2016disc; @li2017instance]), many researchers also have attempted to adapt the deep neural network to the task of crowd counting and achieved great success. Most of the existing methods addressed the scale variation of people with multi-scale architectures. Boominathan et al. [@boominathan2016crowdnet] proposed to tackle the issue of scale variation using a combination of shallow and deep networks along with an extensive data augmentation by sampling patches from multi-scale image representations. HydraCNN [@onoro2016towards] proposed to learn a non-linear regression model which used a pyramid of image patches extracted at multiple scales to perform the final density prediction. A pioneering work was proposed by Zhang et al. [@zhang2016single], in which they utilized multi-column convolutional neural networks with different receptive fields to learn scale-robust features. Sam et al. [@sam2017switching] proposed a Switching-CNN to map a patch from the crowd scene to one of the three regression networks, each of which can handle the particular range of scale variation. CP-CNN [@sindagi2017generating] proposed a Contextual Pyramid CNN to generate high-quality crowd density estimation by incorporating global and local contextual information into the multi-column networks. However, the above-mentioned methods have two significant limitations. First, as a neural network with the fixed static receptive field can only handle a limited enumeration of scale variation, these methods do not scale well to large scale change and cannot well cope with all level of scale variation in diverse scenarios. Second, they did not take the rotation variation of people into consideration, which limits the models’ robustness towards camera perspective variations. To the best of our knowledge, we are the first work to simultaneously address the issue of large scale and rotation variations in an adaptively learnable mode on this task.
**Deep learning Methods for Crowd Counting:** Spatial transformer Network (STN) [@jaderberg2015spatial] is a sub-differentiable sampling-based module, which is designed to spatially transform its input map to an output map that corresponds to a subregion of the input map and can be hence regarded as an effective region selection mechanism. It is convenient to incorporate a spatial transformer layer to the convolutional neural network and train it with the standard back-propagation algorithm. A parameter matrix is used to determine the location of the subregion, as well as its resize scale and the rotation angle. Recently, the spatial transformer has been applied to various computer vision tasks, e.g., multi-label image recognition [@wang2017multi] and saliency detection [@kuen2016recurrent]. To the best of our knowledge, our work is the first in successfully using multiple iterations of STN within a LSTM framework for crowd counting.
{width="1.8\columnwidth"}
\[fig:network-structure\]
Proposed Method
===============
As illustrated in Figure \[fig:network-structure\], we propose a novel Deep Recurrent Spatial-Aware Network for crowd counting, which is composed of two modules, including a Global Feature Embedding (GFE) module and a Recurrent Spatial-Aware Refinement (RSAR) module. Specifically, the GFE module takes the whole image as input for global feature extraction, which is further used to estimate an initial crowd density map. And then the RSAR module is applied to iteratively locate image regions with a spatial transformer-based attention mechanism and refine the attended density map region with residual learning. For convenience in the following, we denote the crowd density map in the $i$-th iteration as ${M_i}$. Noted that ${M_0}$ is the initial crowd density map.
Global Feature Embedding
------------------------
\[sec:coarse\_map\_generation\] The goal of the Global Feature Embedding module is to transform the input image into high-dimensional feature maps, which is further used to generate an initial crowd density map of the image. Inspired by previous works [@zhang2016single; @sindagi2017generating], we develop our GFM module in a multi-scale mode. As shown in Figure \[fig:mcnn\](a), the GFE module is composed of three columns of CNNs, each of which has seven convolutional layers with different kernel sizes and channel numbers as well as three max-pooling layers. Given an image ${I}$, we extract its global feature $g$ by feeding it into GFM and concatenating the outputs of all the columns. After obtaining the global feature $g$, we generate the initial crowd density map ${M_0}$ of image ${I}$ using a convolutional layer with a kernel size of ${1\times1}$, which can be expressed as: $$g=GFM(I),
M_0 = Conv(g),$$ where ${Conv}$ denotes the convolution operation. As shown in Figure \[fig:result\_visual\], the initial crowd density is too rough to estimate the number of people in the image. To further improve the quality of crowd density map, we propose a Recurrent Attentive Refinement (RSAR) module to iteratively refine the density map, which will be described in the next subsection.
Recurrent Spatial-Aware Refinement
----------------------------------
In this section, we propose a novel Recurrent Attentive Refinement (RSAR) module to iteratively refine the crowd density map. Our proposed RSAR consists of two alternately performed components: i) a Spatial Transformer Network dynamically locates an attentional region from the crowd density map; ii) a Local Refinement Network refines the density map of the selected region with residual learning. A high-quality crowd density map with accurately estimated crowd number would be acquired after a refinement of ${n}$ iterations.
### Attentional Region Localization
In the $i$-th iteration, we first determine which image region to be refined with a spatial transformer network. We encode ${M_{i-1}}$ into a 512-dimensions feature with a fully-connected layer and put it into a Long Short-Term Memory (LSTM) [@hochreiter1997long] layer, which can be formulated as: $$c_i, h_i = \textbf{LSTM}(c_{i-1}, h_{i-1}, FC(M_{i-1})),$$ where ${c_i, h_i}$ are the memory cell and hidden state of current iteration and FC is the fully-connected layer. The LSTM is utilized for capturing the past information of the density maps and transformation actions. We define the transformation matrix ${T_i}$ of the spatial transformer as: $$T_i={
\left[ \begin{array}{ccc}
\theta_{11}^i & \theta_{12}^i & \theta_{13}^i\\
\theta_{21}^i & \theta_{22}^i & \theta_{23}^i
\end{array}
\right ]}, \label{T}$$ which allows cropping, translation, scale and rotation to be applied to the input map. We take the hidden state ${h_i}$ to calculate the parameters of transformation matrix ${T_i}$ with a fully-connected layers. Then we extract a region density map $r_i$ from the whole density map ${M_{i-1}}$ according to the transformation matrix ${T_i}$, which can be expressed as: $$r_i=ST(M_{i-1}, T_i),$$ where ST denotes the spatial transformer. The region map $r_i$ will be resized to a given size ${w*h}$ by bilinear interpolation.
### Region Density Map Refinement
Inspired by the previous works\[Sindagi and Patel, 2017a; 2017b\] which have validated the importance of global contextual modeling in crowd counting, we also take into consideration the global context when performing attended density region refinement. Intuitively, the global context contains the density level and density distribution of the given image. We encode the global feature ${g}$ described in section \[sec:coarse\_map\_generation\] with two stacked fully-connected layers. The former one has 256 neurons and the latter one has $w*h$ neurons. We construct the global context map ${c_g}$ by reshaping the output of the last FC layer to a new tensor with a size of $w*h$.
After obtaining the region map $r_i$ and global context map ${c_g}$, we can perform crowd map refinement on the attentional region with a Local Refinement Network, which learns the residual between the region map $r_i$ and the ground truth of its corresponding region. As shown in Figure \[fig:mcnn\](b), the Local Refinement Network also consists of three columns of CNNs. It takes the concatenation of $r_i$ and ${c_g}$ as the input and calculates the residual map of the attentional region. Finally, we can get a new crowd density map ${M_i}$ by adding the inverse transformed residual map to the ${M_{i-1}}$, which can be described as: $$M_i=M_{i-1} + IST(LRN(r_i, c_g), T_i^{-1}),$$ $$T_i^{-1}= r*{
\left[ \begin{array}{ccc}
\theta_{22}^i & -\theta_{12}^i & \theta_{12}^i \theta_{23}^i -\theta_{13}^i \theta_{22}^i\\
-\theta_{21}^i & \theta_{11}^i & \theta_{13}^i \theta_{21}^i - \theta_{11}^i \theta_{23}^i
\end{array}
\right ]},$$ $$r=\frac{1}{\theta_{11}^i \theta_{22}^i - \theta_{12}^i \theta_{21}^i},$$ where IST is the inverse spatial transformer that can transform the residual map back to the attentional region according to the inverse of ${T_i}$, denoted as ${T_i^{-1}}$.
{width="0.925\columnwidth"}
\[fig:mcnn\]
Networks Optimization
---------------------
\[sec:optimization\] For the existence of fully-connected layers in our model, we need to ensure that all the input images are of the same resolutions. Therefore, when optimizing the networks, we transform all training images or patches to the same resolutions and the details will be described in section \[sec:experiments\]. Suppose there are N training images after data augmentation, and each image $I_i$ is annotated with a set of 2D points ${L_i}$ = {${l_1,...,l_{C(i)}}$}, where $C(i)$ is the total number of the labeled pedestrian in the image. We generate ground truth density map ${D_i}$ of image ${I_i}$ with the specific strategy introduced by Lempitsky et al. [@lempitsky2010learning], which ensures that the sum of the ground truth density map ${\sum_{p\in I_i} D_i(p)}$ over the entire image ${I_i}$ matches its crowd count with some negligible deviation.
Our networks are trained in an end-to-end manner and the overall loss function is defined as: $$Loss= \sum_{i=1}^{N} \| M_i^0 - D_i\|^2 + \sum_{i=1}^{N} \| M_i^n - D_i\|^2, \label{loss}$$ where $M_i^0$ and ${M_i^n}$ are the initial map and the refined density map for image $I_i$. We adopt the TensorFlow [@abadi2016tensorflow] toolbox to implement our crowd counting network. The filter weights of all convolutional layers and fully-connected layers are initialized by truncated normal distribution with a deviation equal to 0.01. The learning rate is set to $ 10^{-4} $ initially and multiplied by 0.98 every 1K training iterations. The batch size is set to 1. We optimize our networks parameters with Adam optimization [@kingma2014adam] by minimizing the loss function Eq.(\[loss\]).
Experiments
===========
\[sec:experiments\] In this section, we first compare our method with recent state-of-the-art methods of the crowd counting task on four public challenging datasets. We further conduct extensive ablation studies to demonstrate the effectiveness of each component of our model.
Evaluation Metric
-----------------
Following the existing work [@zhang2016single; @sindagi2017generating], we adopt the mean absolute error (MAE) and mean squared error (MSE) as metrics to evaluate the accuracy of crowd counting estimation, which are defined as:
$$MAE = \frac{1}{N} \displaystyle\sum_{i=1}^{N} \|p_i - \hat{p}_i\|,
MSE = \sqrt{\frac{1}{N} \displaystyle\sum_{i=1}^{N} \| p_i - \hat{p}_i \|^2},$$
where ${N}$ is the total number of testing images, ${p_i}$ and ${\hat{p}_i}$ are the ground truth count and estimated count of the $i$-th image respectively. As discussed in section \[sec:optimization\], ${\hat{p_i}}$ can be calculated by summing over the estimated crowd density map.
Evaluations and Comparisons
---------------------------
{width="1.75\columnwidth"}
\[fig:result\_visual\]
[**ShanghaiTech [@zhang2016single].**]{} This dataset contains 1,198 images of unconstrained scenes with a total of 330,165 annotated people. And it is divided into two parts: Part A with 482 images crawled from the Internet, and Part B with 716 images taken from the busy shopping streets. For each part, we resize the images to its maximal resolution when training and testing.
We compare our method with five recently published state-of-the-art methods on this dataset, i.e., Zhang et al. [@zhang2015cross], MCNN [@zhang2016single], Cascaded-MTL [@sindagi2017cnn], Switching-CNN [@sam2017switching], CP-CNN [@sindagi2017generating]. As shown in Table \[tab:Result\_ShanghaiTech\], our proposed method outperforms other competing methods on both parts of the ShanghaiTech dataset. Specifically, our method achieves a significant improvement of 49.7% in MAE and 39.5% in MSE over the existing best-performing algorithm CP-CNN on Part B.
[**UCF\_CC\_50 [@idrees2013multi].**]{} As an extremely challenging benchmark, this dataset contains 50 annotated images of diverse scenes collected from the Internet. Except for different resolutions, aspect ratios and perspective distortions, the images of this dataset also suffer from a wide range of person counts, varying from 94 to 4,543. Following the standard protocol discussed in [@idrees2013multi], we split the dataset into five subsets and perform a five-fold cross-validation. When training, we randomly crop some regions with a range of $[0.5, 0.9]$ from the original images and resize them to $1024*768$. The testing images are directly resized to the same resolution.
We perform comparison against several state-of-the-art methods based on deep-learning, including Zhang et al. [@zhang2015cross], MCNN [@zhang2016single], Walach et al. [@walach2016learning], Cascaded-MTL [@sindagi2017cnn], Switching-CNN [@sam2017switching], CP-CNN [@sindagi2017generating] and ConvLSTM-nt [@xiong2017spatiotemporal]. As shown in Table \[tab:Result-UCF\], our method achieves superior performance in comparison to other competing methods on UCF\_CC\_50 dataset. Specifically, our method outperforms the existing best-performing method ConvLSTM-nt by 22.8% over the MAE metric.
Method MAE MSE
----------------------------- ---------- ----------
[@zhang2015cross] 467 498.5
[@zhang2016single] 377.6 509.1
[@walach2016learning] 364.4 341.4
[@sindagi2017cnn] 322.8 341.4
[@sam2017switching] 318.1 439.2
[@sindagi2017generating] 295.8 320.9
[@xiong2017spatiotemporal] 284.5 297.1
Ours [****]{} [****]{}
: Performance evaluation of different methods on the UCF\_CC\_50 dataset.
\[tab:Result-UCF\]
[**MALL [@chen2012feature].**]{} This dataset was captured by a publicly accessible surveillance camera in a shopping mall with more challenging lighting conditions and glass surface reflections. The video sequence consists of 2,000 frames with a resolution of ${320*240}$. Following the same setting as [@chen2012feature], we use the first 800 frames for training and the remaining 1,200 frames for evaluation. The images are kept the original resolution when training and testing.
We compare our method with Gaussian Process Regression [@chan2008privacy], Ridge Regression [@chen2012feature], Cumulative Attribute Regression [@chen2013cumulative], Count forest [@pham2015count] and ConvLSTM [@zhang2016single]. The results of all methods are summarized in Table \[tab:Result-MALL\]. Our still-image based method outperforms the existing best method ConvLSTM, which utilized extra temporal dependencies from surveillance video to estimate the crowd count.
Method MAE MSE
------------------------ ---------- ----------
[@chan2008privacy] 3.72 20.1
[@chen2012feature] 3.59 19.0
[@chen2013cumulative] 3.43 17.7
[@pham2015count] 2.50 10.0
[@zhang2016single] 2.24 8.5
Ours [****]{} [****]{}
: Performance evaluation on the MALL dataset. []{data-label="tab:Result-MALL"}
[**WorldExpo’10 [@zhang2015cross].**]{} This dataset contains 1,132 video sequences captured by 108 surveillance cameras during the Shanghai WorldExpo in 2010. The training set consists of 3,380 annotated frames from 103 scenes, while the testing images are extracted from other five different scenes with 120 frames per scene. The images are kept the original resolution for training and testing. When testing, the Region of Interest (ROI) maps for each scene are provided and we only consider the crowd count under the ROI.
We compare our method with seven recently published state-of-the-art methods, including Chen et al. [@chen2013cumulative], Zhang et al. [@zhang2015cross], MCNN [@zhang2016single], Shang et al. [@shang2016end], Switching-CNN [@sam2017switching], CP-CNN [@sindagi2017generating] and ConvLSTM [@xiong2017spatiotemporal]. As demonstrated in Table \[tab:Result\_WorldExpo\], our method achieves state-of-the-art performance with respect to the average MAE of five scenes. Specifically, our model gets the lowest MAE in Scene 2 and Scene 5, which are the two most challenging scenes in the testing set.
Method Scene1 Scene2 Scene3 Scene4 Scene5 Average
----------------------------- ---------- ---------- ---------- ---------- ---------- ----------
[@chen2013cumulative] [****]{} 55.9 [****]{} 11.3 [****]{} 16.5
[@zhang2015cross] 9.8 [****]{} 14.3 22.2 3.7 12.9
[@zhang2016single] 3.4 20.6 12.9 13 8.1 11.6
[@shang2016end] 7.8 15.4 14.9 11.8 5.8 11.7
[@sam2017switching] 4.4 15.7 [****]{} 11 5.9 9.4
[@sindagi2017generating] 2.9 14.7 10.5 [****]{} 5.8 [****]{}
[@xiong2017spatiotemporal] 7.1 15.2 15.2 13.9 [****]{} 10.9
Ours [****]{} [****]{} 10.3 [****]{} 3.7 [****]{}
\[tab:Result\_WorldExpo\]
----------- -------------- -------------- -------------- --------------
MAE MSE MAE MSE
Basic 83.1 120.4 20.1 31.8
T 75.9 109.2 16.5 21.9
T + S 72.9 103.1 14.6 21.4
T + S + R [**69.3**]{} [**96.4**]{} [**11.6**]{} [**19.5**]{}
----------- -------------- -------------- -------------- --------------
: Comparison of the performance of our model with different constraints of the spatial transformer on ShanghaiTech dataset. T, S, and R correspond to translation, scale, and rotation respectively.
\[tab:components\]
Ablation Study
--------------
In this section, we perform extensive ablation studies on ShanghaiTech dataset and demonstrate the effects of several components in our framework.
[**Effectiveness of Spatial Transformation:**]{} The transformation matrix defined in Eq.(\[T\]) allows translation, rotation and scale to be applied to the input map. To validate the effectiveness of each of them, we train our model with three different constraints of spatial transformer: i) spatial transformation only with translation by directly setting $\theta_{11}$, $\theta_{22}$ to 1 and $\theta_{12}$, $\theta_{21}$ to 0, denoted as ${T}$; ii) spatial transformation with translation and scale but without rotation by directly setting $\theta_{12}$, $\theta_{21}$ to 0, denoted as ${T+S}$; iii) spatial transformation with translation, rotation and scale, denoted as ${T+S+R}$. The performance of these variants are summarized in Table \[tab:iteration\] and the performance of initial density map is denoted as “Basic”. We can see that directly locating a region without scale and rotation operation for refinement can reduce the MAE from 83.1 to 75.9 on ShanghaiTech Part A. When dynamically resizing the local regions, the MAE on Part A can be reduced to 73.2. When simultaneously resizing and rotating the local regions, we can get the best performance on ShanghaiTech dataset. The experiment has demonstrated the effectiveness of the spatial transformation for crowd density map refinement.
[**Effectiveness of Global Context:**]{} In order to validate the effectiveness of the global context, we train a variant of our model that refines the local region map without the global context. As shown in Table \[tab:context\], the performance would drop over all metrics when removing the global context. Actually, the global context contains the information of density level and density distribution, which is beneficial for the crowd counting inference.
-------------------- -------------- -------------- -------------- --------------
MAE MSE MAE MSE
W/O Global Context 74.44 100.1 15.7 24.9
W/ Global Context [**69.3**]{} [**96.4**]{} [**11.6**]{} [**19.5**]{}
-------------------- -------------- -------------- -------------- --------------
: Effectiveness verification of global context on the ShanghaiTech dataset.[]{data-label="tab:context"}
[**Effectiveness of Recurrent Refinement:**]{} To validate the superiority of the recurrent refinement mechanism of our method, we train our model with different refinement iterations. As shown in Table \[tab:iteration\], the performance gradually increases with more iterations and it slightly drops until 40 steps. Our method gets the best performance with 30 iterations, therefore we set the refinement iteration $n$ to 30 in our model for all datasets. Figure \[fig:result\_visual\] shows the visual comparison with the refined maps and the initial maps and we can see that our method can generate more accurate crowd density maps after multi-step refinements.
---------- -------------- -------------- -------------- --------------
MAE MSE MAE MSE
$n$ = 0 83.1 120.4 20.1 31.8
$n$ = 10 82.1 111.3 18.8 30.4
$n$ = 20 73.2 113.1 11.2 19.3
$n$ = 30 [**69.3**]{} [**96.4**]{} [**11.1**]{} [**18.2**]{}
$n$ = 40 74.12 110.33 14.27 20.78
---------- -------------- -------------- -------------- --------------
: ShanghaiTech dataset experimental results on the variants of our model using different refinement steps. Our method has the best performance when the density map is refined by $n=30$ steps.
\[tab:iteration\]
Conclusion
==========
In this paper, we introduce a novel Deep Recurrent Spatial-Aware Network for crowd counting which simultaneously models the variations of crowd density as well as the pose changes in a unified learnable module. It can be regarded as a general framework for crowd map refinement. Extensive experiments on four challenging benchmarks show that our proposed method achieves superior performance in comparison to the existing state-of-the-art methods. In our future research, we plan to delve into the research of incorporating our model in other existing crowd flow prediction framework.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was supported by State Key Development Program under Grant 2016YFB1001004, National Natural Science Foundation of China under Grant 61622214 and Grant 61702565, and Guangdong Natural Science Foundation Project for Research Teams under Grant 2017A030312006. This work was also sponsored by CCF-Tencent Open Research Fund. Wanli Ouyang is supported by the SenseTime Group Limited.
[^1]: Corresponding author is Guanbin Li.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The paper considers the convergence to equilibrium for measure solutions of the spatially homogeneous Boltzmann equation for hard potentials with angular cutoff. We prove the exponential sharp rate of strong convergence to equilibrium for conservative measure solutions having finite mass and energy. The proof is based on the regularizing property of the iterated collision operators, exponential moment production estimates, and some previous results on the exponential rate of strong convergence to equilibrium for square integrable initial data. We also obtain a lower bound of the convergence rate and deduce that no eternal solutions exist apart from the trivial stationary solutions given by the Maxwellian equilibrium. The constants in these convergence rates depend only on the collision kernel and conserved quantities (mass, momentum, and energy). We finally use these convergence rates in order to deduce global-in-time strong stability of measure solutions.'
author:
- Xuguang Lu and Clément Mouhot
bibliography:
- 'Biblio-LM.bib'
title: 'On Measure Solutions of the Boltzmann Equation part II: Rate of Convergence to Equilibrium '
---
\[theorem\][Lemma]{} \[theorem\][Conjecture]{} \[theorem\][Corollary]{} \[theorem\][Proposition]{} \[theorem\][Definition]{}
\[theorem\][Remark]{} \[theorem\][Remarks]{} \[theorem\][Examples]{} \[theorem\][Example]{}
0.1in
[**Mathematics Subject Classification (2000)**]{}: 35Q Equations of mathematical physics and other areas of application \[See also 35J05, 35J10, 35K05, 35L05\], 76P05 Rarefied gas flows, Boltzmann equation \[See also 82B40, 82C40, 82D05\].
[**Keywords**]{}: Boltzmann equation; spatially homogeneous; hard potentials; measure solutions; equilibrium; exponential rate of convergence; eternal solution; global-in-time stability.
Introduction {#sec1}
============
The Boltzmann equation describes evolution of a dilute gas. Investigations of the spatially homogeneous Boltzmann equation have made a lot of progresses in the last decades and it is hoped to provide useful clues for the understanding of the complete (spatially inhomogeneous) Boltzmann equation. The complete equation is more realistic and interesting to physics and mathematics but remains still largely out of reach mathematically and will most likely need long term preparations and efforts. For review and references of these areas, the reader may consult for instance [@Villani-handbook; @MR2765747; @partI].
The present paper is a follow-up to our previous work [@partI] on measure-valued solutions[^1] to the spatially homogeneous Boltzmann equation for hard potentials. In this second part, we prove that, under some angular cutoff assumptions (which include the hard sphere model), solutions with measure-valued initial data having finite mass and energy converge strongly to equilibrium in the exponential rate $e^{-{\lambda}t}$, where ${\lambda}>0$ is the spectral gap of the corresponding linearized collision operator. This sharp exponential rate was first proved in [@Mcmp] for initial data with bounded energy, and belonging to $L^1$ (for the hard sphere model) or to $L^1\cap L^2$ (for all hard potentials with cutoff). The core idea underlying our improvement of this result to measure solutions is that instead of considering a one-step iteration of the collision integral which produces the $L^1\cap L^2$ integrability for the hard sphere model (as first observed by Abrahamsson [@MR1697495], elaborating upon an idea in [@MR1697562]), we consider a *multi-steps* iteration which produces the $L^1\cap L^{\infty}$ integrability for all hard potentials with angular cutoff. This, together with approximation by $L^1$ solutions through the Mehler transform, and the property of the exponential moment production, enables us to apply the results of [@Mcmp] and obtain the same convergence rate $e^{-{\lambda}t}$ for measure solutions. We also obtain a lower bound of the convergence rate and establish the global in time strong stability estimate. As a consequence we prove that, for any hard potentials with cutoff, there are no eternal measure solutions with finite and non-zero temperature, apart from the Maxwellians.
The spatially homogeneous Boltzmann equation
--------------------------------------------
The spatially homogeneous Boltzmann equation takes the form $$\label{(B)}
\frac{{\partial}}{{\partial}t}f_t(v)=Q(f_t,f_t)(v), \quad (v,t)\in
\mathbb{R}^N\times(0,\infty),\quad N\ge 2$$ with some given initial data $f_t(v)|_{t=0}=f_0(v) \ge 0$, where $Q$ is the *collision integral* defined by $$\label{(1.1)}
Q(f,f)(v)={\int\!\!\!\!\int}_{\mathbb{R}^N\times
\mathbb{S}^{N-1}}B(v-v_*,{\sigma})\Big(f(v')f(v_*')-f(v)f(v_*)\Big)
\, {\rm d}\sigma \, {\rm d}v_*.$$ In the latter expression, $v,v_*$ and $v', v_*'$ stand for velocities of two particles after and before their collision, and the microscopic conservation laws of an elastic collision $$\label{(1.3)}
v'+v_*'=v+v_*,\quad |v'|^2+|v_*'|^2 = |v|^2 + |v_*|^2.$$ induce the following relations: $$\label{(1.2)}
v'=\frac{v+v_*}{2}+\frac{|v-v_*|}{2}{\sigma},\quad
v_*'=\frac{v+v_*}{2}-\frac{|v-v_*|}{2}{\sigma}$$ for some unit vector ${\sigma}\in \mathbb{S}^{N-1}$.
The collision kernel $B(z,{\sigma})$ under consideration is assumed to have the following *product form* $$\label{(1.4)}
B(z,{\sigma})=|z|^{{\gamma}}b\left(\frac{z}{|z|}\cdot{\sigma}\right)\, ,\quad \gamma>0$$ where $b$ is a nonnegative Borel function on $[-1,1]$. This corresponds to the so-called *inverse power-law interaction potentials* between particles, and the condition $\gamma>0$ corresponds to the so-called *hard potentials*. Throughout this paper we assume that the function $b$ satisfies *Grad’s angular cutoff*: $$\label{(Grad)}
A_0:=\int_{{\mathbb S}^{N-1}}b\left(\frac{z}{|z|}\cdot{\sigma}\right){\rm d}{\sigma}=\left|{\mathbb S}^{N-2}\right|
\int_{0}^{\pi}b(\cos\theta)\sin^{N-2}\theta\,{\rm d}\theta<\infty$$ and it is always assumed that $A_0>0$, where $|{\mathbb S}^{N-2}|$ denotes the Lebesgue measure of the $(N-2)$-dimensional sphere ${\mathbb S}^{N-2}$ (recall that in the case $N=2$ we have ${\mathbb S}^{0}=\{-1,1\}$ and $|{\mathbb S}^{0}|=2$ ). This enables us to split the collision integral as $$Q(f,g)=Q^{+}(f,g)-Q^{-}(f,g)$$ with the two bilinear operators $$\begin{aligned}
\label{(Q+fg)}
&& Q^{+}(f,g)(v)={\int\!\!\!\!\int}_{{\mathbb
R}^N\times{\mathbb S}^{N-1}} B(v-v_*,{\sigma})f(v')g(v_*') \, {\rm
d}{\sigma}\, {\rm d}v_*,
\\
&&\label{(Q-fg)} Q^{-}(f,g)(v)=A_0f(v)\int_{{\mathbb
R}^N}|v-v_*|^{{\gamma}} g(v_*) \, {\rm d}v_*.\end{aligned}$$ which are nonnegative when applied to nonnegative functions.
The bilinear operators $Q^{\pm}$ are bounded from $L^1_{s+{\gamma}}({\mathbb R}^N) \times L^1_{s+{\gamma}}({{\mathbb R}^{N}})$ to $L^1_{s}({\mathbb R}^N)$ for $s\ge 0$, where $L^1_s({\mathbb R}^N)$ is a subspace of $L^1_0({{{\mathbb R}^N}}):= L^1({{{\mathbb R}^N}})$ defined by $$\label{(1-Lspace)}
f\in L^1_s({\mathbb R}^N) \quad \Longleftrightarrow \quad \|f\|_{L^1_s}:=
\int_{{\mathbb R}^N}{\langle}v{\rangle}^{s} |f(v)|\, {\rm d}v<\infty.$$ where we have used the standard notation $$\forall \, v \in \mathbb R, \quad {\langle}v{\rangle}:=\sqrt{1+|v|^2}.$$ Since in the equation , $f=g=f_t$, by replacing $$B(v-v_*,{\sigma}) \quad \mbox{ with } \quad
\frac{1}{2}\big[B(v-v_*,{\sigma})+B(v-v_*,-{\sigma})\big]$$ one can assume without loss of generality that the function $b$ is even: $b(-t)=b(t)$ for all $t\in [-1,1]$. This in turn implies that the polar form of $Q^+$ satisfies $$\label{(1-Qcommut)}
Q^{+}(f,g) \equiv Q^{+}(g,f).$$
The definition of the solutions {#sec:notion-solutions}
-------------------------------
The equation is usually solved as an integral equation as follows. Given any $0\le f_0\in L^1_2({\mathbb R}^N)$, we say that a nonnegative Lebesgue measurable function $(v,t)\mapsto f_t(v)$ on $[0,\infty)\times{\mathbb R}^N$ is a mild solution to if for every $t\ge 0$, $v\mapsto f_t(v)$ belongs to $L^1_2({\mathbb R}^N)$, $\sup_{t\ge 0}\|f_t\|_{L^1_2}<\infty$, and there is a Lebesgue null set $Z_0$ (which is independent of $t$) such that $$\begin{cases}\displaystyle
\forall\,t\in[0,\infty),\quad \forall\, v\in {\mathbb
R}^N\setminus Z_0, \quad
\int_{0}^{t}Q^{\pm}(f_{\tau},f_{\tau})(v)\, {\rm
d}\tau<\infty,\vspace{0.2cm} \\ \displaystyle
\forall\,t\in[0,\infty), \quad \forall\,v\in
{\mathbb R}^N\setminus Z_0, \quad
f_t(v)=f_0(v)+\int_{0}^{t}Q(f_{\tau},f_{\tau})(v)\, {\rm
d}\tau.
\end{cases}$$
The bilinear operators $(f,g) \mapsto Q^{\pm}(f,g)$ can now be extended to measures. For every $s\ge 0$, let ${\mathcal B}_s({\mathbb
R}^N)$ with the norm $\|\cdot\|_s$ be the Banach space of real Borel measures on ${\mathbb R}^N$ defined by $$\label{(1-norm)}
F\in {\mathcal B}_s({\mathbb
R}^N) \quad \Longleftrightarrow \quad \|F\|_s:=\int_{{\mathbb R}^N }{\langle}v{\rangle}^s \, {\rm d}|F|(v)<\infty,$$ where the positive Borel measure $|F|$ is the total variation of $F$. This norm $\|\cdot\|_s$ can also be defined by duality: $$\label{(dual)}
\|F\|_s=\sup_{{\varphi}\in C_c({{{\mathbb R}^N}}),\,\|{\varphi}\|_{L^{\infty}}\le 1}
\left|\int_{{{{\mathbb R}^N}}}{\varphi}(v){\langle}v{\rangle}^s\, {\rm d}F(v)\right|.$$ The latter form is convenient when dealing with the difference of two positive measures. The norms $\|\cdot\|_s$ and $\|\cdot\|_{L^1_s}$ are related by $$\label{(norm)}
\|F\|_s=\|f\|_{L^1_s}\quad
\mbox{if}\quad {\rm d}F(v)=f(v)\, {\rm d}v.$$
For any $F,G\in {\mathcal B}_{s+{\gamma}}({\mathbb R}^N)$ ($s\ge 0$), we define the Borel measures $Q^{\pm}(F,G)$ and $$Q(F,G)=Q^{+}(F,G)-Q^{-}(F,G)$$ through Riesz’s representation theorem by $$\label{(1-Q+meas)}
\int_{{\mathbb R}^N}\psi(v)\, {\rm d}Q^{+}(F,G)(v)=
{\int\!\!\!\!\int}_{{\mathbb R}^N\times{\mathbb R}^N }L_B[\psi](v,v_*)
\, {\rm d}F(v)\, {\rm d}G(v_*),$$ $$\label{(1-Q-meas)}
\int_{{\mathbb R}^N }\psi(v)\, {\rm d}Q^{-}(F,G)(v)=
A_0{\int\!\!\!\!\int}_{{\mathbb R}^N\times{\mathbb R}^N}
|v-v_*|^{{\gamma}}\psi(v)d F(v)\, {\rm d}G(v_*)$$ for all bounded Borel functions $\psi$, where $$\label{(1-LB)}
L_B[\psi](v,v_*)=|v-v_*|^{{\gamma}}\int_{{\mathbb S}^{N-1} }b({\bf n}\cdot {\sigma})
\psi(v')\,\, {\rm d}{\sigma},\quad {\bf n}=\frac{v-v_*}{|v-v_*|}$$ and in case $v=v_*$ we define ${\bf n}$ to be a fixed unit vector ${\bf e}_1$. It is easily shown (see Proposition 2.3 of [@partI]) that the extended bilinear operators $Q^{\pm}$ are also bounded from ${\mathcal B}_{s+{\gamma}}({\mathbb R}^N)\times {\mathcal
B}_{s+{\gamma}}({\mathbb R}^N)$ to ${\mathcal B}_{s}({\mathbb R}^N)$ for $s\ge 0$: if $F,G\in{\mathcal B}_{s+{\gamma}}({\mathbb R}^N)$ then $Q^{\pm}(F,G)\in{\mathcal B}_{s}({\mathbb R}^N)$ and $$\label{(1-Qbound)}
\left\|Q^{\pm}(F,G)\right\|_{s}\le 2^{(s+{\gamma})/2}A_0\left(\|F\|_{s+{\gamma}}\|G\|_0+\|F\|_{0}
\|G\|_{s+{\gamma}}\right),$$ $$\label{(1-Q-differ-bound)}
\left\|Q^{\pm}(F,F)-Q^{\pm}(G,G)\right\|_{s}\le
2^{(s+{\gamma})/2}A_0\left(\|F+G\|_{s+{\gamma}}\|F-G\|_{0}+\|F+
G\|_{0}\|F-G\|_{s+{\gamma}}\right).$$
Let us finally define the cone of positive distributions with $s$ moments bounded: $${\mathcal B}_s^+({\mathbb R}^N) :=\left\{ F\in {\mathcal B}_s({\mathbb
R}^N)\,|\, F\ge 0\right\}.$$ We can now define the notion of solutions that we shall use in this paper. We note that the condition $\gamma\in(0,2]$ as assumed in the following definition is mainly used for ensuring the existence of solutions.
Let $B(z,{\sigma})$ be given by with ${\gamma}\in (0,2]$ and with $b$ satisfysing the condition . Let $\{F_t\}_{t\ge 0}\subset {\mathcal B}_2^{+}(\mathbb{R}^N)$. We say that $\{F_t\}_{t\ge 0}$, or simply $F_t$, is a [*measure strong solution*]{} of equation if it satisfies the following:
- $\sup\limits_{t\ge 0}\|F_t\|_2<\infty,$
- $t\mapsto F_t\in C([0,\infty);{\mathcal
B}_2(\mathbb{R}^N))\cap C^1([0,\infty);{\mathcal
B}_0(\mathbb{R}^N))$ and $$\label{def-strong}
\forall \, t\in[0,\infty), \quad \frac{\, {\rm d}}{\, {\rm d}t}F_t=Q(F_t,F_t).$$
Furthermore $F_t$ is called a [*conservative solution*]{} if $F_t$ conserves the mass, momentum and energy, i.e. $$\forall\, t\ge 0, \quad \int_{\mathbb{R}^N}\left( \begin{array}{c} 1 \\ v \\
|v|^2 \end{array} \right)\, {\rm d}F_t(v)= \int_{\mathbb{R}^N}\left( \begin{array}{c} 1 \\ v \\
|v|^2 \end{array} \right)\, {\rm d}F_0(v).$$
Observe that and imply the strong continuity of $t\mapsto F_t\in C([0,\infty);{\mathcal
B}_2(\mathbb{R}^N))$ and therefore the strong continuity of $t\mapsto Q(F_t,F_t) \in C([0,\infty);{\mathcal
B}_0(\mathbb{R}^N))$. Hence the differential equation is equivalent to the integral equation $$\label{def-strong-bis}
\forall \, t\ge 0, \quad F_t=F_0+\int_{0}^{t}Q(F_\tau, F_\tau)\, {\rm d}\tau,$$ where the integral is taken in the sense of the Riemann integration or more generally in the sense of the Bochner integration. Recall also that here the derivative ${\rm d} \mu_t/{\rm d}t$ and integral $\int_{a}^{b}\nu_t\, {\rm d}t$ are defined by $$\left(\frac{\, {\rm d} }{\, {\rm d}t}\mu_t\right)(E)=\frac{\, {\rm d}}{\, {\rm d}t} \mu_t(E),\quad
\left(\int_{a}^{b}\nu_t \, {\rm d}t\right)(E)= \int_{a}^{b}\nu_t (E)
\, {\rm d}t$$ for all Borel sets $E\subset \mathbb{R}^N$.
Recall of the main results of the first part {#sec:recall-main-results}
--------------------------------------------
The following results concerning moment production and uniqueness of conservative solutions which will be used in the present paper are extracted from our previous paper [@partI]. The following properties (a) and (b) are a kind of “gain of decay” property of the flow stating and quantifying how moments of the solutions become bounded for any positive time even they are not bounded at initial time; the following properties (c)-(d)-(e) concern the stability of the flow.
\[(theo1.0)\] Let $B(z,{\sigma})$ be defined in with ${\gamma}\in (0,2]$ and with the condition . Then for any $F_0\in {\mathcal
B}^{+}_2(\mathbb{R}^N)$ with $\|F_0\|_0> 0$, there exists a unique conservative measure strong solution $F_t$ of equation satisfying $F_t|_{t=0}=F_0$. Moreover this solution satisfies:
- $F_t$ satisfies the moment production estimate: $$\begin{aligned}
\label{(1.12)}&&
\forall\,t>0,\quad \forall\,
s\ge 0, \quad \|F_t\|_{s}\le {\mathcal K}_s \left(1+\frac{1}{t}
\right)^{\frac{(s-2)^{+}}{{\gamma}}}\end{aligned}$$ where $(x-y)^+=\max\{x-y,0\}$, $$\begin{aligned}
\label{(1.14)}&&{\mathcal K}_s:=
{\mathcal K}_s(\|F_0\|_0,\|F_0\|_2)=\|F_0\|_2\left[2^{s+7}\frac{\|F_0\|_2}{\|F_0\|_0}
\left(1+\frac{1}{16\|F_0\|_2A_2{\gamma}}\right)\right]^{\frac{(s-2)^{+}}{{\gamma}}}
\\
&&
\label{(1-A2)}
A_2:=\left|\mathbb{S}^{N-2}\right|\int_{0}^{\pi}
b(\cos\theta)\sin^N\theta\,\, {\rm d}\theta.\end{aligned}$$
- If ${\gamma}\in (0,2)$ or if $$\label{(1-gamma 2)}
{\gamma}=2\quad {\rm and}\quad \exists\, 1<p<\infty\quad {\rm s.t.}\quad
\int_{0}^{\pi}[b(\cos\theta)]^{p}\sin^{N-2}\theta\,\, {\rm
d}\theta<\infty$$ then $F_t$ satisfies the exponential moment production estimate: $$\begin{aligned}
\label{(1.13)}
\forall\,t>0, \quad \int_{\mathbb{R}^N}e^{\alpha(t)\langle v\rangle ^{{\gamma}}} \, {\rm d}F_t(v)\le
2\|F_0\|_0\end{aligned}$$ where $$\begin{aligned}
&&\label{(1-alpha)}
\alpha(t)=2^{-s_0} \frac{\|F_0\|_0}{\|F_0\|_2}\left(1-e^{-\beta
t}\right), \quad \beta =16\|F_0\|_2A_2{\gamma}>0,\end{aligned}$$ and $1<s_0<\infty$ depends only on the function $b$ and ${\gamma}$.
- Let $G_t$ be a conservative measure strong solutions of equation on the time-interval $[\tau, \infty)$ with an initial datum $G_t|_{t=\tau}=G_\tau\in{\mathcal
B}^{+}_2(\mathbb{R}^N)$ for some $\tau\ge 0$. Then:
- If $\tau=0$, then $$\label{(1.21)}
\forall \, t\ge 0, \quad \left\|F_t-G_t\right\|_{2}\le \Psi_{F_0}\left(\left\|F_0-G_0\right\|_2\right) e^{C(1+t)},\quad$$ where $$\label{(1.20)}
\Psi_{F_0}(r)=r+r^{1/3}+ \int_{|v|>r^{-1/3}}|v|^2\, {\rm d}F_0(v),\quad
r>0, \qquad
\Psi_{F_0}(0)=0,$$ and $C={\mathcal R}({\gamma},A_0,A_{2},\|F_0\|_0, \|F_0\|_2)$ is an explicit positive continuous function on $({\mathbb R}_+ ^*)^5$.
- If $\tau>0$, then $$\label{(1.21*)}
\forall \, t\in [\tau,
\infty), \quad \left\|F_t-G_t\right\|_{2} \le \left\|F_\tau-G_\tau\right\|_2 e^{C_{\tau}(t-\tau)},$$ where $$C_{\tau}:= 4\left({\mathcal
K}_{2+{\gamma}}+\|F_0\|_2\right)\left(1+\frac{1}{\tau}\right),$$ and ${\mathcal K}_{2+{\gamma}}$ is defined by (\[(1.14)\]) with $s=2+{\gamma}$.
- If $F_0$ is absolutely continuous with respect to the Lebesgue measure, i.e. $${\rm d}F_0(v)=f_0(v)\, {\rm d}v \quad \mbox{ with } \quad 0\le
f_0\in L^1_2(\mathbb{R}^N),$$ then $F_t$ is also absolutely continuous with respect to the Lebesgue measure: ${\rm d}F_t(v)=f_t(v)\, {\rm d}v$ for all $t\ge 0$, and $f_t$ is the unique conservative mild solution of equation with the initial datum $f_0$.
- If $F_0$ is not a single Dirac distribution, then there is a sequence $f_{k,t}$, $k \ge 1$, of conservative mild solutions of equation with initial data $0\le f_{k,0}\in
L^1_2(\mathbb{R}^N)$ satisfying $$\label{(1.22)}
\int_{\mathbb{R}^N} \left( \begin{array}{c} 1 \\ v \\
|v|^2 \end{array} \right) f_{k,0}(v) \, {\rm d}v=\int_{\mathbb{R}^N}
\left( \begin{array}{c} 1 \\ v \\
|v|^2 \end{array} \right) {\rm
d}F_0(v),\quad k=1,2,\dots$$ such that $$\label{(1.23)}
\forall\,{\varphi}\in C_b(\mathbb{R}^N),\quad
\forall\,t\ge 0, \quad \lim_{k\to\infty}\int_{\mathbb{R}^N}{\varphi}(v) f_{k,t}(v)\, {\rm d}v
=\int_{\mathbb{R}^N}{\varphi}(v)\, {\rm d}F_t(v).$$ Besides, the initial data can be chosen of the form $f_{k,0}=I_{n_k}[F_0],\, k=1,2,3,\dots$ where $\{I_{n_k}[F_0]\}_{k=1}^{\infty}$ is a subsequence of the Mehler transforms $\{I_n[F_0]\}_{n=1}^{\infty}$ of $F_0$.
\[remark1.1\]
1. In the physical case, $N=3$ and $0<{\gamma}\le
1$, the moment estimates and also hold for conservative weak measure solutions of equation without angular cutoff (see [@partI]).
2. The *Mehler transform* $$I_n[F](v):=e^{Nn}\int_{{{{\mathbb R}^N}}}M_{1,0,T}\left(e^n\left(v-u-\sqrt{1-e^{-2n}}\,(v_*-u)\right)\right)
\, {\rm d}F(v_*) \in L^1_2({\mathbb R}^N)$$ of a measure $F\in {\mathcal B}_2^{+}({{{\mathbb R}^N}})$ (which is not a single Dirac distribution) will be studied in Section 4 (after introducing other notations) where we shall show that $I_n[F]$ has a further convenient property: $$\lim_{n \to \infty} \|I_n[F]-M\|_2 = \|F-M\|_2$$ and thus it is a useful tool in order to reduce the study of properties of measure solutions to that of $L^1$ solutions. Here $M$ is the Maxwellian (equilibrium) having the same mass, momentum, and energy as $F$, see - below.
Normalization
-------------
In most of the estimates in this paper, we shall try as much as possible to make explicit the dependence on the basic constants in the assumptions. But first let us study the reduction that can be obtained by scaling arguments.
Under the assumption , it is easily seen that $F_t$ is a measure solution of equation with the angular function $b$ if and only if $t\mapsto F_{A_0^{-1}t}$ is a measure solution of equation with the scaled angular function $A_0^{-1}b$. Therefore without loss of generality we can assume the normalization $$\label{(Grad-1)}
A_0=\left|{\mathbb S}^{N-2}\right|
\int_{0}^{\pi}b(\cos\theta)\sin^{N-2}\theta\,{\rm d}\theta=1.$$
Next given any $\rho>0$, $u\in{{{\mathbb R}^N}}$ and $T>0$, we define the bounded positive linear operator ${\mathcal N}_{\rho,u,T}$ on ${\mathcal
B}_2({{{\mathbb R}^N}})$ as follows: for any $F\in {\mathcal B}_2({{{\mathbb R}^N}})$, there is a unique ${\mathcal N}_{\rho,u,T}(F)\in{\mathcal
B}_2({{{\mathbb R}^N}})$ such that (thanks to Riesz representation theorem), $$\begin{gathered}
\forall \, \psi \mbox{ Borel function s.t. } \sup_{v\in{{{\mathbb R}^N}}}|\psi(v)|{\langle}v{\rangle}^{-2}
<+ \infty, \\ \int_{{{{\mathbb R}^N}}}\psi(v)\, {\rm d}{\mathcal N}_{\rho,u,T}(F)(v)=
\frac{1}{\rho}\int_{{{{\mathbb R}^N}}}\psi\left(\frac{v-u}{\sqrt{T}}\right)\, {\rm d}F(v).\qquad \qquad \qquad\end{gathered}$$ We call ${\mathcal N}_{\rho,u,T}$ the *normalization operator* associated with $\rho,u,T$. The inverse ${\mathcal
N}_{\rho,u,T}^{-1}$ of ${\mathcal N}_{\rho,u,T}$ is given by ${\mathcal N}_{\rho,u,T}^{-1}={\mathcal N}_{1/\rho,-u/\sqrt{T},1/T}$, i.e. $$\int_{{{{\mathbb R}^N}}}\psi(v)\, {\rm d}{\mathcal N}_{\rho,u,T}^{-1}(F)(v)=
\rho\int_{{{{\mathbb R}^N}}}\psi\left(\sqrt{T}\,v+ u\right)\, {\rm d}F(v).$$ It is easily seen that for every $F\in {\mathcal B}_2({{{\mathbb R}^N}})$ $$\begin{aligned}
&& \label{(1.34)}\left\|{\mathcal N}_{\rho,u,T}(F)\right\|_{0}=\frac{1}{\rho}\|F\|_0,
\\ &&
\label{(1.35)} \left\|{\mathcal N}_{\rho,u,T}(F)\right\|_{2}\le C_{\rho,|u|,T}\|F\|_2,\\
&&\label{(1.36)}
\left\|{\mathcal N}_{\rho,u,T}^{-1}(F)\right\|_{2}\le C_{1/\rho, |u|/\sqrt{T}, 1/T}\|F\|_2\end{aligned}$$ where $$\begin{aligned}
&&\label{(1.35c)}
C_{\rho,|u|,T}=
\frac{1}{\rho}\max\left\{ 1+\frac{|u|^2+|u|}{T}\, ; \ \frac{1+|u|}{T}\right\},\\
&& \label{(1.36c)}
C_{1/\rho,|u|/\sqrt{T},1/T}=
\rho\max\left\{ 1+|u|^2+\sqrt{T}|u|\, ; \ T+\sqrt{T}|u|\right\}.\end{aligned}$$
We then introduce the subclass ${\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ of ${\mathcal B}_2^{+}({{{\mathbb R}^N}})$ by $$\label{(1.33)}
\begin{cases}\displaystyle
F\in{\mathcal
B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})\Longleftrightarrow
F\in {\mathcal B}_2^{+}({{{\mathbb R}^N}}) \quad{\rm and}\vspace{0.2cm} \\
\displaystyle
\int_{{{{\mathbb R}^N}}}\, {\rm d}F(v)=\rho,\quad
{\color {red}{{\frac}{1}{\rho}}}\int_{{{{\mathbb R}^N}}}v\, {\rm d}F(v)=u,\quad
\frac{1}{N\rho}\int_{{{{\mathbb R}^N}}}|v-u|^2\, {\rm d}F(v) =T.
\end{cases}$$ In other words, $F\in{\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ means that $F$ has the mass $\rho$, mean-velocity $u$, and the kinetic temperature $T$. It is obvious that $F_t$ conserves mass, momentum, and energy is equivalent to that $F_t$ conserves mass, mean-velocity , and kinetic temperature.
When restricting ${\mathcal N}_{\rho,u,T}$ on ${\mathcal
B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$, it is easily seen that $${\mathcal N}_{\rho,u,T}: {\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}}) \to
{\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}}),\quad {\mathcal N}^{-1}_{\rho,u,T}:
{\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}})\to {\mathcal
B}_{\rho,u,T}^{+}({{{\mathbb R}^N}}).$$
Similarly we define $L^{1}_{\rho,u,T}({{{\mathbb R}^N}})$ by $$\label{(1-general)}
f\in L^{1}_{\rho,u,T}({{{\mathbb R}^N}})\Longleftrightarrow
\begin{cases}\displaystyle
0\le f\in L^1_2({{{\mathbb R}^N}}), \quad \int_{{{{\mathbb R}^N}}}f(v)\, {\rm d}v=\rho,
\vspace{0.2cm} \\ \displaystyle {{\frac}{1}{\rho}}\int_{{{{\mathbb R}^N}}}v f(v)\, {\rm
d}v=u,\quad \frac{1}{N\rho}\int_{{{{\mathbb R}^N}}}|v-u|^2 f(v)\, {\rm d}v =T.
\end{cases}$$ In this case, the normalization operator ${\mathcal N}= {\mathcal
N}_{\rho,u,T}: L^{1}_{\rho,u,T}({{{\mathbb R}^N}})\to L^{1}_{1,0,1}({{{\mathbb R}^N}})$ is written directly as $$\label{(1-norm-f)}
{\mathcal N}(f)(v)=\frac{T^{N/2}}{\rho}f\left(\sqrt{T}\, v+u\right).$$
Recall that the Maxwellian $M\in L^{1}_{\rho,u,T}({{{\mathbb R}^N}})$ is given by $$\label{eq:max}
M(v) :=\frac{\rho}{(2\pi T)^{N/2}}\exp\left(-\frac{|v-u|^2}{2T}\right).$$ For notational convenience we shall do not distinguish between a Maxwellian distribution $M\in {\mathcal B}_{\rho,u,T}^+({{{\mathbb R}^N}})$ and its density function $M\in L^{1}_{\rho,u,T}({{{\mathbb R}^N}})$: we write without risk of confusion that $$\label{eq:max-meas}
{\rm d}M(v)=M(v)\, {\rm d}v.$$
Due to the homogeneity of $z\mapsto B(z,{\sigma})=|z|^{{\gamma}}b(\frac{z}{|z|}\cdot{\sigma})$, we have $$L_B\left[\psi\left(\frac{\cdot-u}{\sqrt{T}}\right)\right]
(v,v_*)=T^{{\gamma}/2}L_B[\psi]\left(\frac{v-u}{\sqrt{T}},
\frac{v_*-u}{\sqrt{T}}\right)$$ and then by Fubini theorem we get (denoting simply ${\mathcal
N}={\mathcal N}_{\rho,u,T}$ when no ambiguity is possible) $$\forall\, F\in{\mathcal B}_2^+({{{\mathbb R}^N}}), \quad {\mathcal
N}\left(Q^{\pm}(F,F)\right)=\rho T^{{\gamma}/2} Q^{\pm}\left({\mathcal N}(F),
{\mathcal N}(F)\right).$$ Since ${\mathcal N}$ is linear and bounded, this implies that if $F_t$ is a measure strong solution of equation and $c=\rho T^{{\gamma}/2}$, then $$\frac{\, {\rm d}}{\, {\rm d}t}{\mathcal N}(F_{t/c})=
Q\left({\mathcal N}(F_{t/c}), {\mathcal N}(F_{t/c})\right).$$ This together with - leads to the following statement:
\[(prop1.1)\] Let $B(z,{\sigma})$ be defined by with ${\gamma}\in (0,2]$ and with the condition . Let $F_0\in{\mathcal
B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ with $\rho>0$, $u\in{{{\mathbb R}^N}}$ and $T>0$, and let $F_t$ be the unique conservative measure strong solution of equation with the initial datum $F_0$. Let $M\in
{\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ be the Maxwellian defined by , let ${\mathcal N}:={\mathcal N}_{\rho,u,T}$ be the normalization operator, and let $c=\rho T^{{\gamma}/2}$. Then:
- The normalization $t\mapsto {\mathcal N}(F_{t/c})$ is the unique conservative measure strong solution of equation with the initial datum ${\mathcal N}(F_0)\in{\mathcal
B}_{1,0,1}^{+}({{{\mathbb R}^N}})$.
- For all $t\ge 0$ $$\begin{cases}\displaystyle
\left\|F_{t}-M\right\|_0=\rho\|{\mathcal N}(F_{t})-{\mathcal
N}(M)\|_0 , \vspace{0.2cm} \\ \displaystyle
\left\|F_{t}-M\right\|_2
\le C_{1/\rho,|u|/\sqrt{T},1/T}\|{\mathcal N}(F_{t})-{\mathcal
N}(M)\|_2, \vspace{0.2cm}\\ \displaystyle
\left\|{\mathcal N}(F_{t})-{\mathcal N}(M)\right\|_2\le C_{\rho,|u|,T}
\left\|F_{t}-M\right\|_2
\end{cases}$$ where $C_{\rho,|u|,T}$ and $C_{1/\rho,|u|/\sqrt{T},1/T}$ are given in -.
Linearized collision operator and spectral gap
----------------------------------------------
For any nonnegative Borel function $W$ on ${{{\mathbb R}^N}}$ we define the weighted Lebesgue space $L^p({{{\mathbb R}^N}}, W)$ with $1\le p<\infty $ by $$f\in L^p({{{\mathbb R}^N}},W) \quad \Longleftrightarrow \quad \|f\|_{L^p(W)}
:=\left(\int_{{{{\mathbb R}^N}}}|f(v)|^p W(v)\, {\rm
d}v\right)^{1/p}<\infty.$$
Let $B(z,{\sigma})$ as defined in with ${\gamma}\in (0,2]$ and with $b$ satisfying . Let $M$ be the Maxwellian with mass $\rho>0$, mean velocity $u$ and temperature $T>0$ defined in , and let $$L_M: L^2\left({{{\mathbb R}^N}}, M^{-1}\right)\to L^2\left({{{\mathbb R}^N}}, M^{-1}\right)$$ be the linearized collision operator associated with $B(z,{\sigma})$ and $M(v)$, i.e. $$\label{(1-linear operator)}
L_M(h)(v)={\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb S}^{N-1}}}}B(v-v_*,{\sigma})M(v_*)
\left(h'+h_*'-h-h_*\right)\, {\rm d}{\sigma}\, {\rm d}v_*.$$ It is well-known that the spectrum ${\Sigma}(L_M)$ of $L_M$ is contained in $(-\infty, 0]$ and has a positive spectral gap $S_{b,{\gamma}}(\rho,\mu,T) >0$, i.e. $$S_{b,{\gamma}}(\rho,u,T) :=\inf\left\{ {\lambda}>0\,\,|\,\, -{\lambda}\in
{\Sigma}(L_M)\right\}>0.$$
Moreover by simple calculations, one has the following scaling property on this spectral gap $$S_{b,{\gamma}}(\rho,u,T)= \rho T^{{\gamma}/2} S_{b,{\gamma}}(1,0,1).$$
In the spatially homogeneous case, the study of the linearized collision operator goes back to Hilbert [@MR1511713; @MR0056184] who computed the collisional invariant, the linearized operator and its kernel in the hard spheres case, and showed the boundedness and “complete continuity” of its non-local part. Carleman [@Carleman] then proved the existence of a spectral gap by using Weyl’s theorem and the compactness of the non-local part proved by Hilbert. Grad [@Grad1; @Grad2] then extended these results to the case of hard potentials with cutoff. All these results are based on non-constructive arguments. The first constructive estimates in the hard spheres case were obtained only recently in [@Baranger-Mouhot] (see also [@MR2254617] for more general interactions, and [@MR2301289] for a review). Let us also mention the works [@WCUh:LBE:70; @Bobylev1975; @Boby:maxw:88] for the different setting of *Maxwell molecules* where the eigenbasis and eigenvalues can be explicitly computed by Fourier transform methods. Although these techniques do not apply here, the explicit formula computed are an important source of inspiration for dealing with more general physical models.
Main results
------------
In order to use the results obtained in [@Mcmp] (see also [@Vi03a; @MR2081030]) for $L^1$ solutions, we shall need the following additional assumptions for some of our main results: $$\begin{aligned}
\label{(b-upperbd)}&&\|b\|_{L^{\infty}}:=\sup_{t\in[-1,1]}b(t)<\infty,\\
&&\label{(b-lowerbd)} \inf_{t\in[-1,1]}b(t)>0.\end{aligned}$$ Recall that for the hard sphere model, i.e. $N=3, {\gamma}=1,$ and $b \equiv {\rm const}.>0$, the conditions - are satisfied.
The first main result of this paper is concerned with the upper bound of the rate of convergence to equilibrium when the dimension $N$ is greater or equal to $3$.
\[(theo1.1)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be given by with ${\gamma}\in(0, \min\{2, N-2\}]$ and with $b$ satisfying , , and . Let $\rho>0$, $u\in{{{\mathbb R}^N}}$ and $T>0$, and let $${\lambda}=S_{b,{\gamma}}(\rho,\mu,T)=S_{b,{\gamma}}(1, 0, 1)\, \rho \, T^{{\gamma}/2}>0$$ be the spectral gap for the linearized collision operator associated with $B(z,{\sigma})$ and the Maxwellian $M\in {\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$. Then for any conservative measure strong solution $F_t$ of the equation with $F_0\in {\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ we have: $$\forall\, t\ge 0, \quad \|F_t-M\|_2\le C\|F_0-M\|_2^{1/2} e^{-{\lambda}t}$$ where $$C:=C_0 \, C_{1/\rho,|u|/\sqrt{T},1/T}
\, \left(C_{\rho,|u|,T} \right)^{1/2}$$ with $C_{\rho,|u|,T}$ and $C_{1/\rho,|u|/\sqrt{T},1/T}$ given in and , and with some constant $C_0<\infty$ which depends only on $N$, ${\gamma}$, and the function $b$ (through the bounds , ).
\[remark1.UpBd\]
1. It should be noted that, in addition to the exponential rate, Theorem \[(theo1.1)\] also shows that for the hard potentials considered here, the convergence to equilibrium is *grossly determined*, i.e. the speed of the convergence only depends only on the collision kernel and the conserved macroscopic quantities (mass, momentum, energy). This is essentially different from those for non-hard potentials (i.e. ${\gamma}\le 0$), see for instance [@MR2546739].
2. Applying Theorem \[(theo1.1)\] to the normal initial data and the Maxwellian $F_0, M\in{\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}})$, and using $\left\|F_0-M\right\|_2^{1/2}\le \left(\left\|F_0\right\|_2+\left\|M\right\|_2\right)^{1/2}
=(2(1+N))^{1/2}$ we have $$\forall\, t\ge 0, \quad \left\|F_t-M\right\|_2\le C_0e^{-{\lambda}t},\quad
{\lambda}=S_{b,{\gamma}}(1,0,1)$$ where $C_0<\infty$ depends only on $N$, ${\gamma}$, and the function $b$. Then by normalization (using Proposition \[(prop1.1)\]) and the relation $S_{b,{\gamma}}(\rho,\mu,T)=S_{b,{\gamma}}(1, 0, 1)\rho T^{{\gamma}/2}$, we conclude that if $F_0, M\in{\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$, then for the same constant $C_0$ we have $$\label{(4.36)}
\left\|F_t-M\right\|_2\le
C_0 \, C_{1/\rho,|u|/\sqrt{T},1/T} \, e^{-{\lambda}t},\quad t\ge 0;\quad
{\lambda}=S_{b,{\gamma}}(\rho,u,T).$$ This estimate will be used in proving our next results Corollary \[cor:11+12\] and Theorem \[(theo1.3)\].
3. In general, in this paper we say that a constant $C$ depends only on some parameters $x_1, x_2, \dots, x_m$, if $C=C(x_1, x_2, \dots, x_m)$ is an explicit continuous function of $(x_1,x_2,\dots,x_m)\in I$ where $I\subset {\mathbb R}^m$ is a possible value range of the parameters $(x_1, x_2, \dots, x_m)$. In particular this implies that if $K$ is a compact subset of $I$, then $C$ is bounded on $K$.
The second main result is concerned with the lower bound of the rate of convergence to equilibrium.
\[(theo1.2)\] Let $B(z,{\sigma})$ be given by with ${\gamma}\in (0,2]$ and with the condition . Let $\rho>0$, $u\in{{{\mathbb R}^N}}$, $T>0$ and let $M\in{\mathcal
B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ be the Maxwellian. Then for any conservative measure strong solution $F_t$ of equation with initial data $F_0\in{\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ we have:
- If $0<{\gamma}<2$, then $$\forall\, t\ge 0, \quad \|F_t-M\|_0\ge
(4\rho)^{1-\alpha}\|F_0-M\|_0 ^{\alpha}
\exp\left(-\beta\,t^{\frac{2}{2-{\gamma}}}\right)$$ where $$\alpha=\left(\frac{2}{{\gamma}}\right)^{\frac{{\gamma}}{2-{\gamma}}} \quad \mbox{
and } \quad \beta= \left(1-\frac{{\gamma}}{2}\right) \left(2^{6}(N+1)^2\rho T^{{\gamma}/2}\right)^{\frac{2}{2-{\gamma}}}.$$
- If ${\gamma}=2$, then $$\forall\, t\ge 0, \quad \left\|F_t-M\right\|_0\ge
4\rho\left(\frac{\left\|F_{0}-M\right\|_0}{4\rho}\right)^{e^{\kappa\,t}}$$ with $\kappa=2^{6}(N+1)^2\rho T$.
\[remark1.LowBd\]
1. The lower bounds established with the norm $\|\cdot\|_0$ imply certain lower bounds in terms of the norm $\|\cdot\|_2$. In fact, on one hand, it is obvious that $\|F_t-M\|_2\ge \|F_t-M\|_0$. On the other hand, for the standard case $F_0, M\in {\mathcal B}^{+}_{1,0,1}({{{\mathbb R}^N}})$, applying the inequalities and $\log y\le \sqrt{y}$($y\ge 1$) we have $$\|F_0-M\|_0\ge \Big({\frac}{1}{4(N+1)}\|F_0-M\|_2\Big)^2.$$ Then, for the general case $F_0,M\in{\mathcal
B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$, we use part (II) of Proposition \[(prop1.1)\] (normalization) to deduce $$\left\|F_0-M\right\|_0\ge \rho\Big({\frac}{1}{4(N+1)}\cdot{\frac}{1}{
C_{1/\rho,|u|/\sqrt{T},1/T}}\|F_0-M\|_2\Big)^2.$$
2. To our knowledge, Theorem \[(theo1.2)\] is perhaps the first result concerning the lower bounds on the relaxation rate for the hard potentials. Of course –and in spite of the fact that the assumptions of Theorem \[(theo1.2)\] are weaker than those of Theorem \[(theo1.1)\]–, these lower bounds are very rough as compared with the corresponding upper bounds in Theorem \[(theo1.1)\]. The particular formula in these lower bounds come from limitations of the method we adopted. We conjecture that under the same assumptions on the initial data (i.e. assuming only that $F_0$ have finite mass, momentum and energy), the lower bounds have the same form ${\rm cst}.e^{-{\rm cst}. t}$ as the upper bounds. This may be investigated in the future.
Now let us state an important corollary of Theorem \[(theo1.1)\] and Theorem \[(theo1.2)\], which gives a positive answer (see the part (iii) below), for hard potentials, to the question of *eternal solutions* raised in [@Villani-handbook Chapter 1, subsection 2.9] (see also [@MR2946633]).
\[cor:11+12\] Under the same assumptions on $N,{\gamma}$ and $B(z,{\sigma})$ as in Theorem \[(theo1.1)\], let $F_0\in{\mathcal
B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ with $\rho>0$, $u\in{{{\mathbb R}^N}}$ and $T>0$, and let $M\in {\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ be the Maxwellian. Then we have:
- Let $F_t\in{\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ be the unique conservative measure solution of equation on $[0,\infty)$ with the initial datum $F_0$. If $F_0\neq M$, then $F_t\neq M$ for all $t\ge 0$. In other words, $F_t$ can not arrive at equilibrium state in finite time unless $F_0$ is an equilibrium.
- Let $F_t\in{\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ be a conservative backward measure strong solution of equation on an interval $(-t_{\infty}, 0]$ for some $0<t_{\infty}\le \infty$, i.e. $$\frac{\, {\rm d}}{\, {\rm d}t}F_t=Q(F_t,F_t),\quad t\in (-t_{\infty}, 0].$$
Then if $F_0\neq M$, then $(-t_{\infty}, 0]$ must be bounded, and if $F_0=M$, then $F_t\equiv M$ on $(-t_{\infty}, 0]$. In particular we have
- If a conservative measure strong solution $F_t$ of equation in ${\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ is *eternal*, i.e. defined for all $t \in \mathbb R$, then it has to be stationary and $F_t = M$ for all $t \in \mathbb R$.
The proof of this Corollary is easy and we would like to present it here.
Part (i) is follows simply from the lower bound in Theorem \[(theo1.2)\]. Part (iii) follows from part (ii). In fact let $F_t$ be an eternal solution of equation as defined in the part (iii) of the statement. Then $F_t$ is also a backward measure strong solution of equation on the unbounded time-interval $(-\infty, 0]$. By part (ii) we conclude that $F_0=M$ and thus $F_t\equiv M$ on $(-\infty,0]$. Then by the uniqueness of forward solutions we conclude that $F_t=M$ for all $t\in \mathbb R$.
To prove part (ii), we use the existence and the uniqueness theorem of conservative measure strong solutions (see Theorem \[(theo1.0)\]) to extend the backward solution $F_t$ to the whole interval $(-t_{\infty},\infty)$. Fix any $\tau\in (-t_{\infty},
0)$. Then $t\mapsto F_{\tau+t}$ is a conservative measure strong solution of equation on $[0,\infty)$ with the initial datum $F_{\tau}$. By using the upper bound of the convergence rate in Theorem \[(theo1.1)\] (see also ), together with the conservation of mass, momentum, and energy we have (with ${\lambda}=S_{b,{\gamma}}(\rho,u,T)$) $$\label{(5.11)}
\forall \, t\ge 0, \quad \left\|F_{\tau+t}-M\right\|_0\le Ce^{-{\lambda}t},$$ where $C>0$ only depends on $N,{\gamma}, b,\rho, u,T$. Taking $t=-\tau$ gives $$\left\|F_0-M\right\|_0\le Ce^{{\lambda}\tau}.$$ Thus if $\left\|F_0-M\right\|_0>0$, then $$-\tau\le \frac{1}{{\lambda}}
\log\left(\frac{C}{\left\|F_0-M\right\|_0}\right)<\infty.$$ Letting $\tau\to -t_{\infty}$ leads to $$t_{\infty}\le \frac{1}{{\lambda}}
\log\left(\frac{C}{\left\|F_0-M\right\|_0}\right)<\infty.$$ Next, applying Theorem \[(theo1.2)\], we have for all $t\ge 0$ $$\begin{aligned}
\label{(5.12)}
\left\|F_{\tau+t}-M\right\|_0 \ge & (4\rho)^{1-\alpha}
\left\|F_{\tau}-M\right\|_0^{\alpha}e^{-\beta\,t^{\frac{2}{2-{\gamma}}}}
& ({\rm if} \,\,\, 0<{\gamma}<2)\\ \label{(5.13)}
\left\|F_{\tau+t}-M\right\|_0 \ge & 4\rho\left (\frac{\left\|F_{\tau}-M\right\|_0}{4\rho}\right)^{e^{\kappa\,t}}
& ({\rm if} \,\,\, {\gamma}=2).\end{aligned}$$ Now suppose $\left\|F_0-M\right\|_0=0$. Then taking $t=-\tau$ so that $\left\|F_{\tau+t}-M\right\|_0=0$ we obtain from , that $\left\|F_{\tau}-M\right\|_0=0$. Since $\tau\in(-t_{\infty},0)$ is arbitrary, this shows that $F_{t}\equiv
M$ on $(-t_{\infty},0]$ and concludes the proof.
The third main result is concerned with the global-in-time stability of measure strong solutions.
\[(theo1.3)\] Let $N,{\gamma}$ and $B(z,{\sigma})$ satisfy the same assumptions in Theorem \[(theo1.1)\]. Let $\rho_0>0$, $u_0 \in{{{\mathbb R}^N}}$, $T_0>0$ and let $M\in{\mathcal
B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ be the Maxwellian.
Then for any conservative measure strong solutions $F_t$, $G_t$ of equation with $F_0\in{\mathcal
B}^{+}_{\rho_0,u_0,T_0}({{{\mathbb R}^N}})$, there are explicitable constants $\eta\in (0,1)$, $C \in (0,\infty)$ only depending on $N$, ${\gamma}$, $b$, $\rho_0$, $u_0$, $T_0$, such that $$\sup_{t\ge 0}\left\|F_t-G_t\right\|_2\le {\widetilde}{\Psi}_{F_0}\left(\left\|F_0-G_0\right\|_2\right)$$ where $$\forall \, r\ge 0, \quad {\widetilde}{\Psi}_{F_0}(r) := C\Big(r+\left[\Psi_{F_0}(r)\right]^{\eta}\Big)$$ with $\Psi_{F_0}(r)$ defined in .
Previous results and references
-------------------------------
Apart from the paper [@Mcmp] already mentioned concerning the sharp rate of relaxation for $L^1$ solutions in the case of hard spheres or hard potentials with cutoff, let us mention the many previous works that developed quantitative estimates on the rate of convergence [@CC92; @MR1263387; @CC94; @MR1725612; @MR1991033; @MR2546739; @MR2546739; @MR2546739]. Let us also mention the recent work [@GMM] obtaining sharp rates of relaxation for $L^1_v L^\infty_x$ solutions in the spatially inhomogeneous case in the torus.
Strategy and plan of the paper {#sec:strategy-plan-paper}
------------------------------
The rest of the paper is organized as follows: In Section 2 we give an integral representation for the one-step iterated collision operator $(f,g,h)\mapsto Q^{+}(f, Q^{+}(g,h))$ and prove an $L^p$ gain of integrability for this operator. This is a generalization of Abrahamsson’s result [@MR1697495] which is concerned with $N=3$ and ${\gamma}=1$. In order to obtain the required regularities of such iterated collision operators, the technical difficulty is to deal with small values of ${\gamma}$. In that case one needs multi-step iteration of $Q^+$. In Section 3 we use iteratively the previous multi-step estimates on $Q^+$ to give a series of positive decompositions $f_t=f^n_t+h^{n}_t$ for $t\in [t_0,\infty)$ with $t_0>0$, for an $L^1$ mild solution $f_t$. In this decomposition the $f^n_t$ are bounded (in $L^{\infty}({{{\mathbb R}^N}})$) and regular (they belong at least to $H^1({{{\mathbb R}^N}})$ for instance) when $n$ is large enough; whereas $h^{n}_t$ decays in $L^1$ norm exponentially fast as $t\to +\infty$. By approximation we then extend such positive decompositions to the measure strong solutions $F_t$. In Section 4 we first use the results of [@Mcmp] and those obtained in Section 3 to prove Theorem \[(theo1.1)\] for $L^1$ mild solutions, and then we use approximation by $L^1$ mild solutions to complete the proof of Theorem \[(theo1.1)\] for measure solutions. The proof of Theorem \[(theo1.2)\] is given in Section 5. In Section 6 we prove Theorem \[(theo1.3)\] which is an application of Theorem \[(theo1.0)\] and Theorem \[(theo1.1)\].
Throughout this paper, unless otherwise stated, we always assume that $N\ge 2$ as already indicated in equation .
$L^p$-estimates of the iterated gain term {#sec2}
=========================================
We introduce the weighted Lebesgue spaces $L^p_s({{{\mathbb R}^N}})$ for $1\le
p\le \infty, 0\le s<\infty$ as: $$\begin{cases} \displaystyle
f\in L^p_s({{{\mathbb R}^N}})
\Longleftrightarrow \|f\|_{L^p_s}
=\left(\int_{{{{\mathbb R}^N}}}{\langle}v{\rangle}^{ps}|f(v)|^p\, {\rm
d}v\right)^{1/p}<\infty,\quad 1\le p<\infty \vspace{0.2cm} \\
\displaystyle
f\in L^{\infty}_s({{{\mathbb R}^N}}) \Longleftrightarrow \|f\|_{L^{\infty}_s}
=\sup_{v\in{{{\mathbb R}^N}}}{\langle}v{\rangle}^{s}|f(v)|<\infty,\quad
p=\infty.
\end{cases}$$ In the case $s=0$, we denote $L^p_0({\mathbb R}^N)=L^p({\mathbb R}^N)$ as usual.
We shall use the following formula of change of variables. For any ${\bf n}\in {{{\mathbb S}^{N-1}}}$ and $\psi$ nonnegative measurable on ${{{\mathbb S}^{N-1}}}$: $$\label{(2.0)}
\int_{{{{\mathbb S}^{N-1}}}}\psi({\sigma})\, {\rm d}{\sigma}=\int_{-1}^{1}(1-t^2)^{(N-3)/2}
\left( \int_{{\mathbb S}^{N-2}({\bf n})}\psi\left(t{\bf
n}+\sqrt{1-t^2}\,{\omega}\right)\, {\rm d}^{\bot}{\omega}\right) \, {\rm d}t$$ where ${\mathbb S}^{N-2}({\bf n})=\{{\omega}\in{{{\mathbb S}^{N-1}}}\,|\,\, {\omega}\,
\bot\,{\bf n}\}$ and ${\rm d}^{\bot}{\omega}$ denotes the sphere measure element of ${\mathbb S}^{N-2}({\bf n})$.
For convenience we rewrite as follows: $$\label{(2.1)}\int_{{{{\mathbb S}^{N-1}}}}\psi({\sigma})\, {\rm d}{\sigma}=\int_{{\mathbb R}}\zeta(t)
\left( \int_{{\mathbb S}^{N-2}({\bf n})}\psi({\sigma}_{{\bf n}}(t,{\omega}))\,
{\rm d}^{\bot}{\omega}\right) \, {\rm d}t$$ where $$\label{(2.2)}
\forall \, t\in \mathbb R, \quad \zeta(t):=(1-t^2)^{\frac{N-3}{2}}{\bf 1}_{(-1,1)}(t)$$ $$\label{(2.3)}
{\sigma}_{{\bf n}}(t,{\omega}):=
\left\{\begin{array}{ll}-{\bf n} \quad & {\rm if} \quad t\le
-1\vspace{0.2cm} \\
\,\,t{\bf n}+\sqrt{1-t^2}\,{\omega}\quad
& {\rm if} \quad t \in (-1,1) \vspace{0.2cm} \\
\,\,\,\, {\bf n} \quad & {\rm if} \quad t\ge 1.
\end{array}\right.$$
\[(lem2.1)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be given by with $b$ satisfying . Let $ f\in
L^1_{{\gamma}}({{{\mathbb R}^N}})$ and $g, h \in L^1_{2{\gamma}}({{{\mathbb R}^N}})$. Then $Q^{+}(f, Q^{+}(g,h))\in L^1({{{\mathbb R}^N}})$ with the estimate $$\label{(2.11)}
\left\|Q^{+}\left(f, Q^{+}(g,h)\right)\right\|_{L^1}\le A_0^2\|f\|_{L^1_{{\gamma}}}
\|g\|_{L^1_{2{\gamma}}}\|h\|_{L^1_{2{\gamma}}}.$$
Moreover we have the following representation: for almost every $v\in{{{\mathbb R}^N}}$ $$\label{(2.4)}
Q^{+}\left(f, Q^{+}(g,h)\right)(v) ={\int\!\!\!\!\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N\times{\mathbb R}^N}}}K_{B}(v,v_*,w,w_*)f(v_*)g(w)h(w_*)
\, {\rm d}v_*\, {\rm d}w \, {\rm d}w_*$$ where $K_B: {\mathbb R}^{4N}\to [0,\infty)$ is defined by $$\label{(2.5)}
K_{B}(v,v_*,w,w_*) :=
\begin{cases} \displaystyle
\frac{2^N}{|v-v_*||w-w_*|}\zeta\left({\bf n}\cdot
\frac{2v-(w+w_*)}{|w-w_*|} \right ) \times \vspace{0.2cm} \\ \displaystyle
\qquad \qquad \int_{{\mathbb
S}^{N-2}({\bf n})} \frac{B(w-w_*,{\sigma})\,B(
w'-v_*,{\sigma}')}{|w'-v_*|^{N-2}} \, {\rm d}^{\bot}{\omega}\vspace{0.2cm}
\\ \displaystyle \qquad \qquad \qquad \qquad
\mbox{if} \quad |v-v_*||w-w_*|\neq 0,
\vspace{0.2cm} \\ \displaystyle
\,\, \, 0 \quad\qquad \qquad \qquad\mbox{if} \quad
|w-w_*||v-v_*|= 0,
\end{cases}$$ where the function $\zeta$ is given by , and $$\label{(2.6)}
{\bf n}:=\frac{v-v_*}{|v-v_*|},\quad
w'=\frac{w+w_*}{2}+\frac{|w-w_*|}{2}{\sigma},\quad {\sigma}'=
\frac{2v-v_*-w'}{|2v-v_*-w'|}$$ with $$\label{(2.7)}
{\sigma}= {\sigma}({\omega})={\sigma}_{{\bf
n}}(t,{\omega})
\quad \mbox{at} \quad t={\bf
n}\cdot\left( \frac{2v-(w+w_*)}{|w-w_*|} \right).$$
Inserting the formula of $B(z,{\sigma})$ into gives the more detailed expression of $K_B$: $$\begin{gathered}
\label{(2.9)}
K_{B}(v,v_*,w,w_*) = \\
\frac{2^N}{|w-w_*|^{1-{\gamma}}|v-v_*|}\zeta\left( {\bf
n}\cdot\frac{2v-(w+w_*)}{|w-w_*|} \right) \int_{{\mathbb
S}^{N-2}({\bf n})}
\frac{b\left(\frac{w-w_*}{|w-w_*|}\cdot{\sigma}\right)\,b\left(
\frac{w'-v_*}{|w'-v_*|}\cdot{\sigma}'\right)}{|w'-v_*|^{N-2-{\gamma}}} \,
{\rm d}^{\bot}{\omega}\end{gathered}$$ for $|w-w_*||v-v_*|\neq 0$. Also we note that $$K_{B}(v,v_*,w,w_*)>0 \Longrightarrow
|v-v_*||w-w_*|\neq 0 \quad \mbox{and} \quad
\left|{\bf n}\cdot\frac{2v-(w+w_*)}{|w-w_*|}\right|<1$$ which implies, by using the formula for $\sigma_{{\bf n}}(t,\omega)$ in this case and the value of $t$, that $(v-w')\cdot(v-v_*)=0$ and therefore by Pythagoras’ theorem $$\begin{aligned}
\displaystyle
&&\label{(2.10)}
|w'-v_*|=|w'-v + v -v_*|=\sqrt{|v-v_*|^2+|v-w'|^2}, \vspace{0.2cm} \\
\displaystyle
&&\label{(2.10*)} |w'-v_*|\ge |v-v_*|.
\end{aligned}$$
We shall use the following formula of change of variables (see [@Villani-handbook Chapter 1, Sections 4.5-4.6]): For every nonnegative measurable function $\psi$ on ${\mathbb R}^{4N}$, one has $$\begin{gathered}
\label{(2-change)}
{\int\!\!\!\!\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N\times{\mathbb S}^{N-1}}}}B(v-v_*,{\sigma}) \psi(v',v_*',
v,v_*)
\, {\rm d}{\sigma}\, {\rm d}v_* \, {\rm d}v\\
={\int\!\!\!\!\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N\times{\mathbb S}^{N-1}}}}B(v-v_*,{\sigma}) \psi(v,v_*,v',v_*') \, {\rm d}{\sigma}\, {\rm d}v_* \, {\rm d}v.\end{gathered}$$
We can assume that $f,g,h$ are all nonnegative. Applying , , and recalling definition of $L_B[{\varphi}]$ (see ) we have, for any nonnegative measurable function ${\varphi}$ on ${{{\mathbb R}^N}}$, $$\begin{cases}\displaystyle
\int_{{{{\mathbb R}^N}}}Q^{+}(f,
Q^{+}(g,h))(v){\varphi}(v) \, {\rm d}v =\int_{{{{\mathbb R}^N}}} f(v_*) \left( \int_{{{{\mathbb R}^N}}}
Q^{+}(g,h)(w)L_B[{\varphi}](w,v_*)\, {\rm d}w \right) \, {\rm
d}v_*, \vspace{0.2cm} \\
\displaystyle
\int_{{{{\mathbb R}^N}}} Q^{+}(g,h)(w)L_B[{\varphi}](w,v_*) \, {\rm
d}w={\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}L_B\left[L_B[{\varphi}](\cdot, v_*)\right](w,w_*)g(w)h(w_*) \,
{\rm d}w \, {\rm d}w_*,
\end{cases}$$ and so $$\begin{gathered}
\label{(2.12)}
\int_{{{{\mathbb R}^N}}}Q^{+}(f, Q^{+}(g,h))(v){\varphi}(v) \, {\rm d}v \\
={\int\!\!\!\!\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N\times{\mathbb R}^N}}}L_B\left[L_B[{\varphi}](\cdot, v_*)\right](w,w_*)f(v_*)g(w)h(w_*) \, {\rm d}v_* \, {\rm d}w
\, {\rm d}w_*.\end{gathered}$$ Taking ${\varphi}=1$ and using the inequalities $$\label{(2.12WV)}
\begin{cases}
|w-w_*|\le {\langle}w{\rangle}{\langle}w_*{\rangle},\vspace{0.2cm}\\ \displaystyle
|w'-v_*|\le \frac{|w+w_*|}{2}+\frac{|w-w_*|}{2}+|v_*|
\le {\langle}w{\rangle}{\langle}w_*{\rangle}{\langle}v_*{\rangle},
\end{cases}$$ we obtain $$L_B\left[L_B[1](\cdot, v_*)\right](w,w_*)
\le A_0^2{\langle}v_*{\rangle}^{{\gamma}}{\langle}w{\rangle}^{2{\gamma}}{\langle}w_*{\rangle}^{2{\gamma}}$$ and it follows from that $$0\le Q^{+}\left(f, Q^{+}(g,h)\right)\in L^1({{{\mathbb R}^N}})$$ and so holds true.
Comparing with , it appears that in order to prove the integral representation we only need to prove that the following identity $$\label{(2.14)}
\forall\, 0\le
{\varphi}\in C_c({{{\mathbb R}^N}}), \quad L_B\left[L_B[{\varphi}](\cdot,
v_*)\right](w,w_*)=\int_{{{{\mathbb R}^N}}}K_B(v,v_*,w,w_*){\varphi}(v)\, {\rm d}v$$ holds for all $w,w_*,v_*\in{{{\mathbb R}^N}}$ satisfying $$\label{(2.13)}
0\neq \left|\frac{w+w_*}{2}-v_*\right|\neq \frac{|w-w_*|}{2}\neq 0.$$
Observe that $$\begin{cases} \displaystyle
L_B\left[L_B[{\varphi}](\cdot,v_*)\right](w,w_*)=
L_B\left[L_B[{\varphi}(v_*+\cdot)](\cdot,0)\right](w-v_*,w_*-v_*),\vspace{0.2cm} \\
\displaystyle
K_B(v,v_*,w,w_*)=K_B(v-v_*,0,w-v_*,w_*-v_*).
\end{cases}$$ By replacing respectively ${\varphi}(v_*+\cdot)$, $w-v_*$ and $w_*-v_*$ with ${\varphi}(\cdot)$, $w$ and $w_*$, we can assume without loss of generality that $v_*=0$. That is, in order to prove , we only need to prove $$\label{(2.15)}
\forall\, 0\le {\varphi}\in C_c({{{\mathbb R}^N}}), \quad
L_B\left[L_B[{\varphi}](\cdot,0)\right](w,w_*)=\int_{{{{\mathbb R}^N}}}K_B(v,0,w,w_*){\varphi}(v)\,
{\rm d}v.$$
To do this we first assume that $b \in C([-1,1])$ so that the use of the Dirac distribution is fully justified. We compute $$\label{(2.16)}
L_B\left[L_B[{\varphi}](\cdot, 0)\right](w,w_*) =\int_{{{{\mathbb S}^{N-1}}}}
B(w-w_*,{\sigma})\int_{{{{\mathbb S}^{N-1}}}}B(w',{\omega}){\varphi}\left(\frac{w'}{2}
+\frac{|w'|}{2}{\omega}\right)\, {\rm d}{\omega}\, {\rm d}{\sigma},$$ and, by using and with $v_*=0$, we have $w'\neq 0$ for all ${\sigma}\in{{{\mathbb S}^{N-1}}}$. Let ${\delta}= {\delta}(x)$ be the one-dimensional Dirac distribution. Applying the integral representation $$\forall \, \psi\in C((0,\infty)),\ \rho>0, \quad
\psi(\rho)=\frac{2}{\rho^{N-2}}\int_{0}^{\infty}
r^{N-1}\psi(r){\delta}(\rho^2-r^2)\, {\rm d}r$$ to the function $$\psi(\rho):= {\varphi}\left(\frac{w'}{2} +\rho{\omega}\right)$$ and then taking $\rho=|w'|/2$ and changing variable $r{\omega}=z$ we have $$\begin{gathered}
\int_{{{{\mathbb S}^{N-1}}}}B(w',{\omega}){\varphi}\left(\frac{w'}{2} +\frac{|w'|}{2}{\omega}\right) \, {\rm
d}{\omega}\\ =2\left|\frac{w'}{2}\right|^{-(N-2)} \int_{{{{\mathbb R}^N}}}
B\left(w',\frac{z}{|z|}\right){\varphi}\left(\frac{w'}{2}+ z\right) {\delta}\left(\frac{|w'|^2}{4}-|z|^2\right)
\, {\rm d}z.\end{gathered}$$ We then use the change of variable $z=v-w'/2$ and Fubini theorem: $$\begin{gathered}
\label{(2,17)}
L_B\left[L_B[{\varphi}](\cdot, 0)\right](w,w_*) = \\
\int_{{{{\mathbb R}^N}}}{\varphi}(v)\left(\int_{{{{\mathbb S}^{N-1}}}}
2 \left|\frac{w'}{2}\right|^{-(N-2)} B(w-w_*,{\sigma})
B\left(w',\frac{v-w'/2}{|v-w'/2|}\right)
{\delta}\left(\left|\frac{w'}{2}\right|^2-\left|v-\frac{w'}{2}\right|^2\right)
\, {\rm d}{\sigma}\right) \, {\rm d}v.\end{gathered}$$
We now assume that $v\neq 0$ and ${\bf n}=v/|v|$ satisfy $$\left|{\bf n}\cdot\frac{2v-(w+w_*)}{|w-w_*|}\right|\neq 1,
\quad \left|v-\frac{w+w_*}{4}\right|\neq \left|\frac{w-w_*}{4}\right|.$$ We deduce that $|v-w'/2|>0$ for all ${\sigma}\in{{{\mathbb S}^{N-1}}}$ and we compute using -- with ${\bf
n}=v/|v|$ that $$\begin{gathered}
\int_{{{{\mathbb S}^{N-1}}}}2 \left|\frac{w'}{2}\right|^{-(N-2)}
B(w-w_*,{\sigma}) B\left(w',\frac{v-w'/2}{|v - w'/2|}\right)
{\delta}\left(\frac{|w'|^2}{4}-\left|v-\frac{w'}{2}\right|^2\right)\, {\rm d}{\sigma}\\
=\int_{{\mathbb R}}\zeta(t)\,\left(\int_{{\mathbb S}^{N-2}({\bf
n})} 2 \left|\frac{w'}{2}\right|^{-(N-2)} B(w-w_*,{\sigma})
B\left(w',\frac{v-w'/2}{|v- w'/2|}\right)\Bigg|_{{\sigma}={\sigma}_{{\bf n}}(t,{\omega})}\, {\rm d}^{\bot}{\omega}\right)\\
\times {\delta}\left(|v|\frac{|w-w_*|}{2}t -v\cdot\left(v- \frac{w+w_*}{2}\right)\right)\, {\rm d} t\\
=\frac{2^N}{|v||w-w_*|}\zeta\left(\frac{v\cdot(
2v-(w+w_*))}{|v||w-w_*|}\right)\int_{{\mathbb
S}^{N-2}({\bf n})} |w'|^{-(N-2)} B(w-w_*,{\sigma})
B\left(w',\frac{2v-w'}{|2v-w'|}\right)\, {\rm d}^{\bot}{\omega}\end{gathered}$$ where ${\sigma}$ in the last line is given by . Thus we obtain $$\begin{gathered}
\label{(2.18)} \qquad L_B[L_B[{\varphi}](\cdot,
0)](w,w_*)
=\frac{2^N}{|w-w_*|}\int_{{{{\mathbb R}^N}}} \frac{{\varphi}(v)}{|v|}\zeta\left({\bf
n}\cdot \frac{2v-(w+w_*)}{|w-w_*|}\right) \\ \times\int_{{\mathbb
S}^{N-2}({\bf n})}|w'|^{-(N-2)} B(w-w_*,{\sigma})
B\left(w',\frac{2v-w'}{|2v-w'|}\right)\, {\rm d}^{\bot}{\omega}\, {\rm
d}v.\end{gathered}$$ This proves .
Finally, thanks to $N\ge 3$, we use standard approximation arguments in order to prove that still holds without the continuity assumption on the function $b$. We skip these classical calculations.
\[(lem2.2)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined in with $b$ satisfying . Let $1\le p,q\le \infty$ satisfy $1/p+1/q=1$.
Then, in the case where we have $$\label{(2.19)}
0<{\gamma}< N-2,\quad \frac{N-1}{N-1-{\gamma}}\le p< \frac{N}{N-1-{\gamma}},$$ the following estimate holds $$\label{(2.20)}
\left(\int_{{{{\mathbb R}^N}}}[K_{B}(v,v_*,w,w_*)]^p\, {\rm d}v\right)^{1/p} \le
C_{p}\|b\|_{L^{\infty}}^2|w-w_*|^{2{\gamma}-N/q}.$$
Second, in the case where we have $$\label{(2.21)}
{\gamma}\ge N-2,\quad 1\le p<N,$$ then the following estimate holds: $$\label{(2.22)}\left(\int_{{{{\mathbb R}^N}}}[K_B(v,v_*,w,w_*)]^{p} \, {\rm d}v\right)^{1/p}\le
C_{p}\|b\|_{L^{\infty}}^2\langle{v_*}\rangle^{2{\gamma}-N/q} \langle{w}\rangle^{2{\gamma}-N/q}\langle{w_*}\rangle^{2{\gamma}-N/q}.$$ The constants $C_p$ only depend on $N,{\gamma}, p$.
By replacing the function $b$ with $b/\|b\|_{L^{\infty}}$ we can assume for notation convenience that $\|b\|_{L^{\infty}}=1$. Fix $w,w_*, v_*\in{{{\mathbb R}^N}}$. To prove the lemma we may assume that $w\neq w_*$. Recall that $N\ge 3$ implies $\zeta(t)\le {\bf 1}_{(-1,1)}(t)$. Then from - we have $$\begin{gathered}
\label{(2.24)} K_B(v,v_*,w,w_*)\\
\le \frac{2^N|{\mathbb S}^{N-2}|
}{|w-w_*|^{1-{\gamma}}}\cdot\frac{1}{|v-v_*|^{N-1-{\gamma}}}{\bf
1}_{(-1,1)}\left({\bf n}\cdot
\frac{2v-(w+w_*)}{|w-w_*|}\right)\quad {\rm for}\quad
0<{\gamma}<N-2,\end{gathered}$$ whereas for ${\gamma}\ge N-2$ we have $$\begin{gathered}
\label{(2.25)}
K_B(v,v_*,w,w_*)\\
\le 2^{N}|{\mathbb S}^{N-2}|{\langle}v_*{\rangle}^{{\gamma}+2-N}{\langle}w{\rangle}^{2{\gamma}+1-N} {\langle}w_*{\rangle}^{2{\gamma}+1-N}\frac{1}{|v-v_*|}{\bf 1}_{(-1,1)}\left({\bf n}\cdot
\frac{2v-(w+w_*)}{|w-w_*|}\right)\end{gathered}$$ where we used $N-2\ge 1$ and the inequalities in .
Let us define $$J_{\beta}(w,w_*)=\int_{{{{\mathbb R}^N}}}\frac{1}{|v-v_*|^{\beta}}{\bf
1}_{(-1,1)}\left({\bf n}\cdot
\frac{2v-(w+w_*)}{|w-w_*|}\right) \, {\rm d}v.$$ We need to prove that $$\begin{aligned}
\label{(2.27)}&& J_{\beta}(w,w_*)\le
|{{{\mathbb S}^{N-1}}}|({\langle}v_*{\rangle}{\langle}w{\rangle}{\langle}w_*{\rangle})^{N-1-\beta}|w-w_*|\quad
\mbox{when} \quad 0<\beta<N-1,\\
&&
\label{(2.28)} J_{\beta}(w,w_*)
\le \frac{|{{{\mathbb S}^{N-1}}}|}{(N-\beta)}
|w-w_*|^{N-\beta} \quad \mbox{when} \quad N-1\le \beta<N.\end{aligned}$$ To do this we use the change of variable $$v=v_*+\frac{|w-w_*|}{2} r{\sigma}, \quad
{\rm d}v=\left|\frac{w-w_*}{2}\right|^{N} r^{N-1}\, {\rm d}r\, {\rm d}{\sigma}$$ to compute $$\label{(2.29)}J_{\beta}(w,w_*)
=\left|\frac{w-w_*}{2}\right|^{N-\beta}\int_{{{{\mathbb S}^{N-1}}}}I(u\cdot{\sigma})
\, {\rm d}{\sigma}$$ where $$I(u\cdot {\sigma})=\int_{0}^{\infty}r^{N-1-\beta}{\bf 1}_{\{|r+u\cdot {\sigma}|<1\}}
\, {\rm d}r,\quad u=\frac{2v_*-(w+w_*)}{|w-w_*|}.$$
Let us now estimate $I(u\cdot {\sigma})$ uniformly in ${\sigma}$. If $u\cdot
{\sigma}\ge 1$, then $I(u\cdot {\sigma})=0$. Suppose $u\cdot {\sigma}< 1$. If $0<\beta<N-1$, then $$I(u\cdot {\sigma}) \le
2(1+|u|)^{N-1-\beta}\le 2
\left(\frac{|\frac{w-w_*}{2}|+|\frac{w+w_*}{2}|+|v_*|}{|\frac{w-w_*}{2}|}\right)^{N-1-\beta}$$ which together with and the third inequality in gives . If $N-1\le \beta<N$, then $0<N-\beta\le 1$ so that $$I(u\cdot {\sigma})
\le\frac{2^{N-\beta}}{N-\beta}$$ which implies .
Now suppose is satisfied. Then using and with $$N-1\le \beta=(N-1-{\gamma})p<N$$ we obtain : $$\begin{gathered}
\left(\int_{{{{\mathbb R}^N}}}[K_{B}(v,v_*,w,w_*)]^p\, {\rm d}v\right)^{1/p}
\le C_p{|w-w_*|^{{\gamma}-1}}\left(J_{(N-1-{\gamma})p}(w,w_*)
\right)^{1/p}\\
\le C_{p}|w-w_*|^{{\gamma}-1}|w-w_*|^{\frac{N-p(N-1-{\gamma})}{p}}
=C_{p}|w-w_*|^{2{\gamma}-N/q}.\end{gathered}$$
Next suppose is satisfied. If $N-1\le p<N$, then by - with $\beta=p$ and using $|w-w_*|\le
{\langle}w{\rangle}{\langle}w_*{\rangle}$ we have $$\begin{gathered}
\left(\int_{{{{\mathbb R}^N}}}[K_{B}(v,v_*,w,w_*)]^p\, {\rm
d}v\right)^{1/p}
\le C_{p}\langle v_*\rangle ^{{\gamma}+2-N}\langle w\rangle^{2{\gamma}+1-N}
\langle w_*\rangle ^{2{\gamma}+1-N}
\left(J_{p}(w,w_*)\right)^{1/p}\\
\le C_{p}\langle v_*\rangle ^{{\gamma}+2-N}\langle w\rangle^{2{\gamma}-N/q} \langle
w_*\rangle ^{2{\gamma}-N/q}.\end{gathered}$$
Similarly if $1\le p<N-1$, then using - with $\beta=p$ we have $$\left(\int_{{{{\mathbb R}^N}}}[K_{B}(v,v_*,w,w_*)]^p\, {\rm d}v\right)^{1/p}\\
\le C_{p}\langle v_*\rangle ^{{\gamma}-(N-1)/q}\langle w\rangle^{2{\gamma}-N/q} \langle
w_*\rangle ^{2{\gamma}-N/q}.$$ Since ${\gamma}\ge N-2\ge 1$ and $1\le p<N$ imply $$\max\left\{{\gamma}+2-N, {\gamma}-\frac{N-1}{q}\right\}\le 2{\gamma}-N/q,$$ it follows that $$\max\left\{{\langle}v_*{\rangle}^{{\gamma}+2-N},\,{\langle}v_*{\rangle}^{{\gamma}-\frac{N-1}{q}}\right\}\le {\langle}v_*{\rangle}^{2{\gamma}-N/q}.$$ This concludes the proof of .
\[(lem2.3)\] Let $K(v,v_*)$ be a measurable function on ${{{\mathbb R}^N\times{\mathbb R}^N}}$ and let $$\forall \, v\in {{{\mathbb R}^N}}, \quad T(f)(v):=\int_{{{{\mathbb R}^N}}}K(v,v_*)f(v_*)\,
{\rm d}v_*.$$ Assume that $1\le r<\infty,\, 0\le s<\infty$ and that there is $0<A<\infty$ such that $$\label{(2.30)} \left(\int_{{{{\mathbb R}^N}}}|K(v,v_*)|^r \,
{\rm d} v\right)^{\frac{1}{r}}\le A{\langle}v_*{\rangle}^{s}\qquad {\rm
a.e.}\quad v_*\in {{{\mathbb R}^N}}.$$ Then $$\label{(2.31)}
\forall \, f\in L^1_s({{{\mathbb R}^N}}), \quad \|T(f)\|_{L^r}\le A\|f\|_{L^1_s}.$$ Furthermore let $1\le p,q\le \infty$ satisfy $$\frac{1}{p}=\frac{1}{q}+\frac{1}{r}-1$$ and assume that there is $0< B<\infty$ such that $$\label{(2.32)}
\left(\int_{{{{\mathbb R}^N}}}(|K(v,v_*)|{\langle}v_*{\rangle}^{-s})^{r} \, {\rm d}v_*\right)^{\frac{1}{r}}
\le B\qquad {\rm a.e.} \quad v\in {{{\mathbb R}^N}}.$$ Then $$\label{(2.33)}
\forall\,f\in L^q_s({{{\mathbb R}^N}}), \quad \|T(f)\|_{L^p}\le A^{\frac{r}{p}}
B^{1-\frac{r}{p}}\|f\|_{L^q_s}.$$
Let us define $$\forall \, v\in {{{\mathbb R}^N}}, \quad T_s(f)(v):=\int_{{{{\mathbb R}^N}}}K(v,v_*){\langle}v_*{\rangle}^{-s}f(v_*)\, {\rm d}v_*.$$ By Minkowski inequality and we have $$\label{(2.34)}
\forall\, f\in L^1({{{\mathbb R}^N}}), \quad \left\|T_s(f)\right\|_{L^r}\le A\|f\|_{L^1}.$$ Also by we have $$\label{(2.35)}
\forall\, f\in L^{r'}({{{\mathbb R}^N}}), \quad \left\|T_s(f)\right\|_{L^{\infty}}\le B\|f\|_{L^{r'}}$$ where $1\le r'\le \infty$ satisfies $1/r+1/{r'}=1$.
Let $\theta=1-r/p$. By assumption on $p,q,r$ we have $0\le \theta\le
1$ and $$\frac{1}{p}=\frac{1-\theta}{r}+\frac{\theta}{\infty},\quad
\frac{1}{q}=\frac{1-\theta}{1}+\frac{\theta}{r'}.$$ So by , and Riesz-Thorin interpolation theorem (see e.g. [@MR0304972 Chapter 5]) we have $$\label{(2.36)}
\forall\,f\in L^q({{{\mathbb R}^N}}), \quad \left\|T_s(f)\right\|_{L^p}\le A^{1-\theta}B^{\theta}\|f\|_{L^q}.$$ Now if we set $(f)_s(v_*)={\langle}v_*{\rangle}^sf(v_*)$, then $$T(f)=T_s((f)_s), \quad \|(f)_s\|_{L^1}=\|f\|_{L^1_s}, \quad
\|(f)_s\|_{L^q}=\|f\|_{L^q_s}$$ and thus - follow from -.
In order to highlight structures of inequalities, we adopt the following notional convention: [*Functions $f,g,h$ appeared below are arbitrary members in the classes indicated. Whenever the notation (for instance) $\|f\|_{L^p_s}$ appears, it always means that $f\in L^p_s({\mathbb R}^N)$ with the norm $\|f\|_{L^p_s}$; and if $\|f\|_{L^p_s}$, $\|f\|_{L^q_k}$ appear simultaneously, it means that $f\in L^p_s({\mathbb R}^N)\cap
L^q_k({\mathbb R}^N)$.*]{}
\[(lem2.4)\] Let $0<\alpha<N, 1\le \alpha q<N,1<p\le \infty$ with $1/p+1/q=1$. Then $$\sup_{v\in{{{\mathbb R}^N}}}\int_{{{{\mathbb R}^N}}}\frac{|f(v_*)|}{|v-v_*|^{\alpha}} \,
{\rm d}v_* \le
2\left(\frac{\left|{{{\mathbb S}^{N-1}}}\right|}{N-\alpha q}\right)^{\frac{\alpha }{N}} \|f\|_{L^1}^{1-\frac{\alpha q
}{N}} \|f\|_{L^p}^{\frac{\alpha q }{N}}.$$
This follows from Hölder inequality and a minimizing argument.
\[(lem2.5)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined in with the condition . For any $w,w_*\in{{{\mathbb R}^N}}$ with $w\neq w_*$, let $$\label{(2.37)}
\forall \, v\in{{{\mathbb R}^N}}, \quad T_{w,w_*}(f)(v):=\int_{{{{\mathbb R}^N}}}K_{B}(v,v_*,w,w_*)f(v_*)\, {\rm d}v_*$$ for nonnegative measurable or certain integrable functions $f$ as indicated below.
- Suppose $0<{\gamma}< N-2$. Let $p_1=(N-1)/(N-1-{\gamma})$. Then $$\label{(2.38)}\|T_{w,w_*}(f)\|_{L^{p_1}}\le C_{p_1}\|b\|_{L^{\infty}}^2
\|f\|_{L^1}{\langle}w{\rangle}^{{\gamma}}{\langle}w_*{\rangle}^{{\gamma}}.$$ Also if $1< p, q<\infty$ satisfy $$\frac{1}{p}=
\frac{1}{q}+\frac{1}{p_1}-1$$ then $$\label{(2.39)}
\left\|T_{w,w_*}(f)\right\|_{L^{p}}\le
C_{p}\|b\|_{L^{\infty}}^2 \|f\|_{L^q_{1}} \frac{{\langle}w{\rangle}^{1-\frac{p_1}{p}}{\langle}w_*{\rangle}^{1-\frac{p_1}{p}}}{
|w-w_*|^{1-{\gamma}-\frac{1}{p}}}.$$ And if $1< p\le
\infty, 1\le q< N/(N-1-{\gamma})$ satisfy $1/p+1/q=1$, then $$\label{(2.40)}
\left\|T_{w,w_*}(f)\right\|_{L^{\infty}}\le \frac{C_{p}\|b\|_{L^{\infty}}^2}{|w-w_*|^{1-{\gamma}}}
\|f\|_{L^1}^{1-\frac{N-1-{\gamma}}
{N}q} \|f\|_{L^{p}}^{\frac{N-1-{\gamma}}{N}q}.$$
- Suppose ${\gamma}\ge N-2$. Let $1< p<N$, $1/p+1/q=1$. Then $$\label{(2.41)}
\left\|T_{w,w_*}(f)\right\|_{L^p}\le
C_{p}\|b\|_{L^{\infty}}^2\|f\|_{L^1_{2{\gamma}-N/q}}{\langle}w{\rangle}^{2{\gamma}-N/q}{\langle}w_*{\rangle}^{2{\gamma}-N/q}.$$ Furthermore if $N/(N-1)< p<N$, then $$\label{(2.42)}
\left\|T_{w,w_*}(f)\right\|_{L^{\infty}}\le C_{p}\|b\|_{L^{\infty}}^2
\|f\|_{L^1_{{\gamma}+2-N}}^{1-\frac{q}{N}} \|f\|_{L^p_{{\gamma}+2-N}}^{\frac{q}{N}} {\langle}w{\rangle}^{2{\gamma}+1-N}{\langle}w_*{\rangle}^{2{\gamma}+1-N}.$$
The constants $C_p<\infty$ only depend on $N,{\gamma}, p$.
As in the proof of Lemma \[(lem2.2)\] we can assume that $\|b\|_{L^{\infty}}=1$.
[*Case [(i)]{}.*]{} Suppose $0<{\gamma}<N-2$. By Lemma \[(lem2.2)\] we have $$\label{(2.43)}
\left(\int_{{{{\mathbb R}^N}}}[K_B(v,v_*,w,w_*)]^{p_1}\, {\rm d}v\right)^{1/{p_1}}\le
C_{p_1}|w-w_*|^{2{\gamma}-N/{q_1}}$$ where $q_1=(p_1)/(p_1-1)=(N-1)/{\gamma}$. Since $$0< 2{\gamma}-N/{q_1}< {\gamma}\quad \mbox{ and } \quad |w-w_*|\le {\langle}w{\rangle}{\langle}w_*{\rangle},$$ follows from with $r=p_1, s=0$. Next recalling and the second inequality in we see that that $$K_B(v,v_*,w,w_*)> 0 \quad \Longrightarrow \quad \sqrt{1+|v-v_*|^2} \le
\sqrt{2}{\langle}w{\rangle}{\langle}w_*{\rangle}{\langle}v_*{\rangle}$$ so that $$\label{(2.44)}K_B(v,v_*,w,w_*){\langle}v_*{\rangle}^{-1}\le
\sqrt{2}{\langle}w{\rangle}{\langle}w_*{\rangle}\frac{K_B(v,v_*,w,w_*)}{\sqrt{1+|v-v_*|^2}}.$$ This together with and $(N-1-{\gamma})p_1=N-1$ gives $$\begin{gathered}
\label{(2.45)}
\left(\int_{{{{\mathbb R}^N}}}(K_B(v,v_*,w,w_*){\langle}v_*{\rangle}^{-1})^{p_1} \, {\rm d}v_*\right)^{1/{p_1}}\\
\le C_{p_1}\frac{{\langle}w{\rangle}{\langle}w_*{\rangle}}{
|w-w_*|^{1-{\gamma}}}\left(\int_{{{{\mathbb R}^N}}}\frac{{\rm
d}v_*}{(1+|v-v_*|^2)^{p_1/2} |v-v_*|^{N-1}}\right)^{1/{p_1}}
=C_{p_1}\frac{{\langle}w{\rangle}{\langle}w_*{\rangle}}{
|w-w_*|^{1-{\gamma}}}.\end{gathered}$$ Note that the above integral is finite since $p_1>1$. If we set $$A_{w,w_*}:=C_{p_1}
|w-w_*|^{2{\gamma}-N/{q_1}},\quad B_{w,w_*}:=C_{p_1}\frac{{\langle}w{\rangle}{\langle}w_*{\rangle}}{ |w-w_*|^{1-{\gamma}}}$$ then we see from and that Lemma \[(lem2.3)\] can be used for $T_{w,w_*}(f)$ with $r=p_1$ and $s=1$, and thus for all $f\in L^q_1({{{\mathbb R}^N}})$ $$\left\|T_{w,w_*}(f)\right\|_{L^p} \le
\left(A_{w,w_*}\right)^{\frac{p_1}{p}}\left(B_{w,w_*}\right)^{1-\frac{p_1}{p}}\|f\|_{L^q_1}
=C_p\frac{{\langle}w{\rangle}^{1-\frac{p_1}{p}}{\langle}w_*{\rangle}^{1-\frac{p_1}{p}}}
{|w-w_*|^{1-{\gamma}-\frac{1}{p}}}\|f\|_{L^q_1}$$ where we have computed (using the definitions of $p_1, q_1$) $$\frac{p_1}{p}\left(2{\gamma}-\frac{N}{q_1}\right)-\left(1-\frac{p_1}{p}\right)(1-{\gamma})
=\frac{1}{p}+{\gamma}-1.$$ This proves . To prove we use to get $$\left|T_{w,w_*}(f)(v)\right|\le \frac{2^N|{\mathbb S}^{N-2}|}{|w-w_*|^{1-{\gamma}}}
\int_{{{{\mathbb R}^N}}}\frac{|f(v_*)|\, {\rm d}v_*}{|v-v_*|^{N-1-{\gamma}}}.$$ Since $1\le (N-1-{\gamma})q<N$, it follows from Lemma \[(lem2.4)\] that $$\left\|T_{w,w_*}(f)\right\|_{L^{\infty}}\le \frac{C_{p}}{|w-w_*|^{1-{\gamma}}}
\|f\|_{L^1} ^{1-\frac{N-1-{\gamma}}
{N}q} \|f\|_{L^p} ^{\frac{N-1-{\gamma}}{N}q}.$$
[*Case [(ii)]{}.*]{} Suppose ${\gamma}\ge N-2$. In this case we recall the inequality . Let $1< p<N$ and $1/p+1/q=1$. Then applying Lemma \[(lem2.3)\] to $T_{w,w_*}(f)$ with $r=p, s=2{\gamma}-N/q$ gives . Finally suppose $N/(N-1)< p<N$. Recalling and using we have $$\left|T_{w,w_*}(f)(v)\right|\le C_{N,{\gamma}}{\langle}w{\rangle}^{2{\gamma}+1-N} {\langle}w_*{\rangle}^{2{\gamma}+1-N}
\int_{{{{\mathbb R}^N}}} \frac{{\langle}v_*{\rangle}^{{\gamma}+2-N}}{|v-v_*|}|f(v_*)|\, {\rm
d}v_*.$$ Since $q=p/(p-1)<N$, it follows from Lemma \[(lem2.4)\] that $$\int_{{{{\mathbb R}^N}}} \frac{{\langle}v_*{\rangle}^{{\gamma}+2-N}}{|v-v_*|}|f(v_*)|\, {\rm d}v_*\le C_{p}
\|f\|_{L^1_{{\gamma}+2-N}}^{1-\frac{q}{N}}
\|f\|_{L^p_{{\gamma}+2-N}}^{\frac{q}{N}}.$$ This proves .
Let $f,g,h$ be nonnegative measurable functions on ${{{\mathbb R}^N}}$. Define for any $s\ge 0$ $$(f)_s(v):={\langle}v{\rangle}^sf(v).$$ Then applying the inequality ${\langle}v{\rangle}\le {\langle}v'{\rangle}{\langle}v_*'{\rangle}$ we have $$\begin{cases} \displaystyle
\left(Q^{+}(f,g)\right)_s\le Q^{+}((f)_s, (g)_s),\vspace{0.2cm} \\ \displaystyle
\left(Q^{+}(f, Q^{+}(g, h))\right)_s \le Q^{+}\left((f)_s,
Q^{+}((g)_s, (h)_s)\right),
\end{cases}$$ and so on and so forth. Consequently we have for all $s\ge 0$ and $1\le p\le
\infty$: $$\label{(2.46)}
\left\|Q^{+}(f, Q^{+}(g, h))\right\|_{L^p_s}\le
\left\|Q^{+}\left((f)_s,Q^{+}\left((g)_s, (h)_s\right)\right)\right\|_{L^p}$$ provided that the right hand side makes sense.
Now we are going to prove the $L^{p}$ and $L^{\infty}$ boundedness of the iterated operator $Q^{+}(f, Q^{+}(g,h))$. Let $$\label{(2.47)}
N_{{\gamma}}=\left\{\begin{array}{ll} \left[\frac{N-1}{{\gamma}}\right] & {\rm
if}\,\,\,0<{\gamma}<N-2
\\ \\
\,\,\,\,\,\,1 & {\rm if}\,\,\,{\gamma}\ge N-2
\end{array}\right.$$ where $[x]$ denotes the largest integer not exceeding $x$.
\[(theo2.1)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined by with the condition . Given any $s\ge 0$ we have:
- Suppose $0<{\gamma}<N-2$. Let $N_{{\gamma}}$ be defined in and let $$p_n=\frac{N-1}{N-1-n{\gamma}} \in (1,\infty],\quad n=1,2,\dots,N_{{\gamma}}.$$ Then
- For all $n=1,2,\dots,N_{{\gamma}}$, we have $$\label{(2.48)}
\left\|Q^{+}(f, Q^{+}(g,h))\right\|_{L^{p_{1}}_s}\le
C_{p_1}\|b\|_{L^{\infty}}^2\|f\|_{L^1_{s}}
\|g\|_{L^1_{s+{\gamma}}} \|h\|_{L^1_{s+{\gamma}}}.$$
- If $1\le n\le N_{{\gamma}}-1$, then $$\label{(2.49)}
\left\|Q^{+}(f, Q^{+}(g,h))\right\|_{L^{p_{_{n+1}}}_s}\le
C_{p_{_{n+1}}}\|b\|_{L^{\infty}}^2\|f\|_{L^{p_n}_{s+1}}
\|g\|_{L^1_{s+{\gamma}_1}} \|h\|_{L^1_{s+{\gamma}_1}}^{1-\theta_n}
\|h\|_{L^{p_n}_{s+{\gamma}_1}}^{\theta_n}$$ where $$\label{(2.50)}
{\gamma}_1=\max\{{\gamma}\, ,\ 1\},\quad
\theta_n=\frac{1}{N}\left(1-\frac{N-2}{n}\right)^{+},\quad n\ge 1.$$
- Finally if $n=N_{{\gamma}}$, then $$\label{(2.51)}
\left\|Q^{+}(f,Q^{+}(g,h))\right\|_{L^{\infty}_s}\le
C_{\infty}\|b\|_{L^{\infty}}^2 \|f\|_{L^1_s} ^{1-\alpha_1} \|f\|_{L^{p_{_{N_{{\gamma}}}}}_s}^{\alpha_1}
\|g\|_{L^1_{s+{\gamma}_*}} \|h\|_{L^1_{s+{\gamma}_*}}^{1-\alpha_2}
\|h\|_{L^{p_{_{N_{{\gamma}}}}}_s}^{\alpha_2}$$ where ${\gamma}_*=({\gamma}-1)^{+}$ and $$\label{(2.52)}
0<\alpha_1 := \left( \frac{N-1}{{\gamma}N_{{\gamma}}}\right) \left(
\frac{N-1-{\gamma}}{N} \right) <1,\quad 0\le \alpha_2 :=\left(
\frac{N-1}{{\gamma}N_{{\gamma}}}\right) \left( \frac{(1-{\gamma})^{+}}{N} \right)<1.$$
- Suppose ${\gamma}\ge N-2$. Let $1< p<N$, $1/p+1/q=1$. Then $$\label{(2.53)}
\left\|Q^{+}(f, Q^{+}(g,h))\right\|_{L^p_s}\le
C_{p}\|b\|_{L^{\infty}}^2\|f\|_{L^1_{s+2{\gamma}-N/q}}
\|g\|_{L^1_{s+2{\gamma}-N/q}}\|h\|_{L^1_{s+2{\gamma}-N/q}}.$$ Furthermore if $N/(N-1)<p<N$, then $$\label{(2.54)}
\left\|Q^{+}\left(f, Q^{+}(g,h)\right)\right\|_{L^{\infty}_s}
\le C_{p}\|b\|_{L^{\infty}}^2 \|f\|_{L^1_{s+{\gamma}+2-N}}^{1-\frac{q}{N}}
\|f\|_{L^p_{s+{\gamma}+2-N}}^{\frac{q}{N}}
\|g\|_{L^1_{s+2{\gamma}+1-N}}\|h\|_{L^1_{s+2{\gamma}+1-N}}.$$
All the constants $C_p<\infty$ only depend on $N,{\gamma},p$.
Observe that in the case $\gamma \ge N-2$, the iterated operator maps (forgetting about the weights) $L^1 \times L^1 \times L^1$ to $L^p$ for $1<p<N$. This can be recovered heuristically from Lions’ theorem in [@MR1284432] and [@MR2081030] that states that $Q^+$ maps $L^1 \times H^s$ to $H^{s+(N-1)/2}$ for $s \in \mathbb R$. Then $L^1$ is contained in $H^{-N/2-0}$ and applying twice Lions’ theorem one gets that the iterated operator maps $L^1\times L^1 \times L^1$ to $H^{2(N-1)/2-N/2-0} = H^{N/2-1-0}$. And the Sobolev embedding for the space $H^{N/2-1}$ is precisely $L^N$.
The proof is a direct application of the inequalities obtained in Lemma \[(lem2.4)\] and Lemma \[(lem2.5)\]. We can assume that $f,g,h $ are all nonnegative. And because of , we need only to prove the theorem for $s=0$. By the integral representation of $Q^{+}(f, Q^{+}(g,h))$ and the definition of $T_{w,w_*}$ we have $$Q^{+}(f,Q^{+}(g,h))(v)={\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}T_{w,w_*}(f)(v)g(w) h(w_*)\,
{\rm d}w \, {\rm d}w_*,$$ and therefore $$\label{(2.55)}
\left\|Q^{+}\left(f, Q^{+}(g,h)\right)\right\|_{L^p}\le
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}\left\|T_{w,w_*}(f)\right\|_{L^p}g(w) h(w_*) \, {\rm
d}w \, {\rm d}w_*$$ for all $1\le p\le\infty$.
[*Case [(i)]{}.*]{} Suppose $0<{\gamma}<N-2$. By we have $$\left\|T_{w,w_*}(f)\right\|_{L^{p_1}}\le
C_{p_1}\|b\|_{L^{\infty}}^2\|f\|_{L^1}{\langle}w{\rangle}^{{\gamma}}{\langle}w_*{\rangle}^{{\gamma}}$$ and so by $$\left\| Q^{+}(f, Q^{+}(g,h)) \right\|_{L^{p_1}}\le
C_{p_1}\|b\|_{L^{\infty}}^2\|f\|_{L^1}\|g\|_{L^1_{{\gamma}}}\|h\|_{L^1_{{\gamma}}}.$$ This proves (with $s=0$).
- Suppose $1\le n\le N_{{\gamma}}-1$. By definition of $p_n$ we have $$\frac{1}{p_{n+1}}=\frac{1}{p_n}+\frac{1}{p_1}-1,\quad
n=1,2,\dots,N_{{\gamma}}-1.$$ By and we have $$\label{(2.56)}
\begin{cases}\displaystyle
\left\|T_{w,w_*}(f)\right\|_{L^{p_{n+1}}}\le C_{p_{n+1}}\|b\|_{L^{\infty}}^2
\frac{{\langle}w{\rangle}^{1-\frac{p_1}{p_{n+1}}}{\langle}w_*{\rangle}^{1-\frac{p_1}{p_{n+1}}}}{|w-w_*|^{1-{\gamma}-\frac{1}{p_{n+1}}}}\|f\|_{L^{p_n}_1},
\vspace{0.2cm} \\ \displaystyle
\|Q^{+}(f,Q^{+}(g,h))\|_{L^{p_{_{n+1}}}} \vspace{0.2cm} \\
\displaystyle \mbox{ } \qquad
\le C_{p_{_{n+1}}}\|b\|_{L^{\infty}}^2\|f\|_{L^{p_n}_1}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}\frac{{\langle}w{\rangle}^{1-\frac{p_1}{p_{n+1}}}{\langle}w_*{\rangle}^{1-\frac{p_1}{p_{n+1}}}}{|w-w_*|^{1-{\gamma}-\frac{1}{p_{n+1}}}}g(w)h(w_*) {\rm
d}w \, {\rm d}w_*.
\end{cases}$$ Let $q_n\ge 1$ be defined by $$\frac{1}{q_n}+\frac{1}{p_n}=1, \quad \mbox{i.e.} \quad
q_n=\frac{N-1}{n{\gamma}}.$$ We have $$1-{\gamma}-\frac{1}{p_{n+1}}=\frac{n+2-N}{N-1}{\gamma}, \quad
q_n\left(1-{\gamma}-\frac{1}{p_{n+1}}\right)=1-\frac{N-2}{n}.$$ If $n\le N-2$, then $1-{\gamma}-1/p_{n+1}\le 0$ and using $|w-w_*|\le {\langle}w{\rangle}{\langle}w_*{\rangle}$ we have $$\frac{{\langle}w{\rangle}^{1-\frac{p_1}{p_{n+1}}}{\langle}w_*{\rangle}^{1-\frac{p_1}{p_{n+1}}}}{|w-w_*|^{1-{\gamma}-\frac{1}{p_{n+1}}}}
\le {\langle}w{\rangle}^{{\gamma}}{\langle}w_*{\rangle}^{{\gamma}}$$ and so by we have $$\left\|Q^{+}\left(f, Q^{+}(g,h)\right)\right\|_{L^{p_{_{n+1}}}} \le C_{p_{_{n+1}}}
\|b\|_{L^{\infty}}^2\|f\|_{L^{p_n}_1}
\|g\|_{L^1_{{\gamma}}}\|h\|_{L^1_{{\gamma}}}.$$
If $n>N-2$, then $0<q_n(1-{\gamma}- 1/p_{n+1})<1$ so that applying Lemma \[(lem2.4)\] (with $q=q_{n},\,
\alpha=1-{\gamma}-1/p_{n+1}$) and recalling the definition of $\theta_n$ we have $$\label{(2.57)}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}\frac{{\langle}w{\rangle}^{1-\frac{p_1}{p_{n+1}}}{\langle}w_*{\rangle}^{1-\frac{p_1}{p_{n+1}}}}
{|w-w_*|^{1-{\gamma}-\frac{1}{p_{n+1}}}}g(w)h(w_*) \, {\rm d}w \, {\rm d}w_* \le
C_{p_{_{n+1}}}\|g\|_{L^1_1}(\|h\|_{L^1_1})^{1-\theta_n}(\|h\|_{L^{p_n}_1})^{\theta_n}$$ and thus (with $s=0$) follows from and .
- Now let $n=N_{{\gamma}}$. Let us recall that $$q_{N_{\gamma}}=\frac{N-1}{N_{{\gamma}}{\gamma}}, \quad
N_{{\gamma}}> \frac{N-1}{{\gamma}}-1>0,$$ hence $$\label{(2.58)}
q_{N_{{\gamma}}}(N-1-{\gamma})< N-1.$$ Using and (for the $L^{\infty}$ norm) we have $$\label{(2.59)}
\begin{cases} \displaystyle
\left\|T_{w,w_*}(f)\right\|_{L^{\infty}}\le \frac{C_{\infty}\|b\|_{L^{\infty}}^2}{|w-w_*|^{1-{\gamma}}}
\|f\|_{L^1}^{1-\frac{N-1-{\gamma}}
{N}q_{N_{{\gamma}}}} \|f\|_{L^{p_{N_{{\gamma}}}}}^{\frac{N-1-{\gamma}}{N}
q_{N_{{\gamma}}} }, \vspace{0.2cm} \\ \displaystyle
\left\|Q^{+}\left(f,Q^{+}(g,h)\right)\right\|_{L^{\infty}} \vspace{0.2cm} \\ \displaystyle
\mbox{ } \qquad \le C_{\infty}\|b\|_{L^{\infty}}^2 \|f\|_{L^1} ^{1-\frac{N-1-{\gamma}}
{N}q_{N_{{\gamma}}}} \left( \|f\|_{L^{p_{N_{{\gamma}}}}}\right)^{\frac{N-1-{\gamma}}{N} q_{N_{{\gamma}}}}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}\frac{g(w)h(w_*)}{|w-w_*|^{1-{\gamma}}} \, {\rm d}w \, {\rm
d}w_*.
\end{cases}$$ If $0<{\gamma}<1$, then from and $N\ge 3$ we have $0<q_{N_{{\gamma}}}(1-{\gamma})<N-1$ so that using Lemma \[(lem2.3)\] gives $$\label{(2.60)}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}\frac{g(w)h(w_*)}{|w-w_*|^{1-{\gamma}}} \, {\rm d}w \,
{\rm d}w_*
\le C_{{\gamma}}\|g\|_{L^1}
\|h\|_{L^1}^{1-\frac{1-{\gamma}}{N}q_{N_{{\gamma}}}}
\|h\|_{L^{p_{N_{{\gamma}}}}}^{\frac{1-{\gamma}}{N}q_{N_{{\gamma}}}}.$$ If $1\le {\gamma}<N-2$, then $|w-w_*|^{{\gamma}-1}\le {\langle}w{\rangle}^{{\gamma}-1} {\langle}w_*{\rangle}^{{\gamma}-1} $ and so $$\label{(2.61)}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}} \frac{g(w)h(w_*)}{|w-w_*|^{1-{\gamma}}} \, {\rm
d}w \, {\rm d}w_* \le\|g\|_{L^1_{{\gamma}-1}}\|h\|_{L^1_{{\gamma}-1}}.$$ Thus (with $s=0$) follows from , and .
[*Case [(ii)]{}.*]{} Suppose ${\gamma}\ge N-2$ and let $1<p<N$, $1/p+1/q=1$. Then (with $s=0$) follows from and . Furthermore if $N/(N-1)<p<N$, then (with $s=0$) follows from (for the $L^{\infty}$ norm) and .
Iteration and Decomposition of Solutions {#sec3}
========================================
We begin by the study of the process of iteration of the collision operator and decomposition of solutions though the following lemma. Roughly speaking the strategy of the decomposition is the following. We use the Duhamel representation formula to decompose the flow associated with the equation into two parts, one of which is more regular than the initial datum, while the amplitude of the other decreases exponentially fast with time, and we repeat this process in order to increase the smoothness, starting each time a new flow having the smoother part of the previous solution as initial datum. Each time we start a new flow, we depart from the true solution, since the initial datum is not the real solution, and we keep track of this error through a Lipschitz stability estimate. Finally the times of the decomposition are chosen in such a way that the time-decay of the non-smooth parts dominates the time-growth in these Lipschitz stability errors.
\[(lem3.1\] Let $B(z,{\sigma})$ be defined in with ${\gamma}\in(0,2]$ and with the condition . Let $f_t\in
L^1_2({{{\mathbb R}^N}})$ be a mild solution of equation . Let us define $$\label{(3.1)} \forall \, s, t\ge 0, \quad
E_{s}^{t}(v):=\exp\left(-\int_{s}^{t}
\int_{{{{\mathbb R}^N}}}|v-v_*|^{{\gamma}}f_{\tau}(v_*) \, {\rm d}v_* \, {\rm
d}\tau \right).$$
Given any $t_0\ge 0$ we also define for all $t\ge t_0$ $$\begin{aligned}
\label{(3.2)}&& f^{0}_t(v)=f_t(v),\quad h^0_t(v) =0,\\
&&\label{(3.3)}f_{t}^{n}(v)
=\int_{t_0}^{t}E_{t_1}^{t}(v)\int_{t_0}^{t_1}
Q^{+}\left( f^{n-1}_{t_1},\,Q^{+}(f^{n-1}_{t_2}, f^{n-1}_{t_2})E_{t_2}^{t_1}
\right)(v) \, {\rm d}t_2 \, {\rm d}t_1, \\
&&\label{(3.4)}
h_t^{1}(v)=f_{t_0}(v)E_{t_0}^{t}(v)+\int_{t_0}^{t}
Q^{+}\left(f_{t_1},\,f_{t_0}E_{t_0}^{t_1}\right)(v)E_{t_1}^{t}(v)\, {\rm d}t_1,\\
&&\label{(3.5)}h_t^{n}(v)=h_t^{1}(v) +\int_{t_0}^{t}E_{t_1}^{t}(v)
\int_{t_0}^{t_1}Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2},
\,h^{n-1}_{t_2})E_{t_2}^{t_1}\right)(v) \, {\rm d}t_2 \, {\rm d}t_1 \\
\nonumber &&
\qquad\,\,\, +\int_{t_0}^{t}E_{t_1}^{t}(v) \int_{t_0}^{t_1}
Q^{+}\left(f_{t_1},\,Q^{+}(f^{n-1}_{t_2},
h^{n-1}_{t_2})E_{t_2}^{t_1} \right)(v)\, {\rm d}t_2 \, {\rm d}t_1
\nonumber\\
&& \qquad\,\,\, +\int_{t_0}^{t}E_{t_1}^{t}(v)
\int_{t_0}^{t_1}Q^{+}\left(h^{n-1}_{t_1},\,
Q^{+}\left(f^{n-1}_{t_2}, f^{n-1}_{t_2}\right)E_{t_2}^{t_1}\right)(v)
\, {\rm d}t_2 \, {\rm d}t_1
\nonumber\end{aligned}$$ for $n=1,2,3, \dots$
Then $f_t^n\ge 0$, $h_t^n\ge 0$ and there is a null set $Z\subset
{{{\mathbb R}^N}}$ which is independent of $t$ and $n$ such that for all $v\in
{{{\mathbb R}^N}}\setminus Z$ $$\label{(3.6)}
\forall\, t\in[t_0,\infty),\quad n=1,2,3,\dots, \quad
f_t(v)=f_{t}^{n}(v)+h_t^{n}(v).$$
In the following we denote by $Z_0, Z_1, Z_2, \dots \subset
{{{\mathbb R}^N}}$ some null sets (i.e. ${\rm meas}(Z_n)=0$) which are independent of the time variable $t$. The decomposition is based on the Duhamel representation formula for the solution $f_t$: for all $v\in {{{\mathbb R}^N}}\setminus Z_0$ $$\label{(3.7)}
\forall \, t\ge t_0, \quad f_t(v)=
f_{t_0}(v)E_{t_0}^t(v)+\int_{t_0}^{t} Q^{+}(f_{t_1},
f_{t_1})(v)E_{t_1}^t(v)\, {\rm d}t_1.$$ Here we note that in the definition of $E_{s}^{t}(v)$ we have used the assumption , i.e., $A_0=1$. Applying to $f_{t}$ at time $t=t_1$ and inserting it into the second argument of $Q^{+}(f_{t_1}, f_{t_1})$ we obtain for all $t\ge t_0$ and all $v\in {{{\mathbb R}^N}}\setminus Z_1$ $$\label{(3.8)}
f_t(v)= h_t^1(v) +\int_{t_0}^{t}E_{t_1}^{t}(v)\int_{t_0}^{t_1}
Q^{+}\left( f_{t_1},\,Q^{+}(
f_{t_2}, f_{t_2})E_{t_2}^{t_1}\right)(v)
\, {\rm d}t_2 \, {\rm d}t_1.$$ That is, we have the decomposition $$\forall\, t\ge t_0,\quad \forall\, v\in {{{\mathbb R}^N}}\setminus Z_1, \quad
f_t(v)=h_t^{1}(v)+f_{t}^{1}(v).$$
Suppose for some $n\ge 1$, the decomposition $f_t(v)=
h^{n}_t(v)+f_t^{n}(v)$ holds for all $t\ge t_0$ and all $v\in
{{{\mathbb R}^N}}\setminus Z_n$. Let us insert $f_{t_2}=h_{t_2}^n+f_{t_2}^n$ and $f_{t_1}=h_{t_1}^n+f_{t_1}^n$ into $ Q^{+}\left( f_{t_1},\,Q^{+}(
f_{t_2}, f_{t_2})E_{t_2}^{t_1}\right)$ in the following way: $$\begin{cases} \displaystyle
Q^{+}\left( f_{t_1},\,Q^{+}( f_{t_2}, f_{t_2})E_{t_2}^{t_1} \right)=
Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}, h_{t_2}^n)E_{t_2}^{t_1}\right)
+Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}, f_{t_2}^n)E_{t_2}^{t_1}\right),
\vspace{0.2cm} \\ \displaystyle Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2},
f_{t_2}^n)E_{t_2}^{t_1}\right)
=Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}^n,
h_{t_2}^n)E_{t_2}^{t_1}\right)+
Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}^n,
f_{t_2}^n)E_{t_2}^{t_1}\right), \vspace{0.2cm} \\ \displaystyle
Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}^n, f_{t_2}^n)E_{t_2}^{t_1}\right)
= Q^{+}\left(h_{t_1}^n,\,Q^{+}(f_{t_2}^n, f_{t_2}^n)E_{t_2}^{t_1}
\right)+Q^{+}\left(f_{t_1}^n,\,Q^{+}(f_{t_2}^n,
f_{t_2}^n)E_{t_2}^{t_1}\right).
\end{cases}$$ Then $$\begin{gathered}
Q^{+}\left( f_{t_1},\,Q^{+}(
f_{t_2}, f_{t_2})E_{t_2}^{t_1}\right)=
Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}, h_{t_2}^n)E_{t_2}^{t_1}\right)
+Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}^n, h_{t_2}^n)E_{t_2}^{t_1}\right) \\
+Q^{+}\left(h_{t_1}^n,\,Q^{+}(f_{t_2}^n, f_{t_2}^n)E_{t_2}^{t_1}
\right)+Q^{+}\left(f_{t_1}^n,\,Q^{+}(f_{t_2}^n,
f_{t_2}^n)E_{t_2}^{t_1}\right).\end{gathered}$$ Inserting this into yields $$\forall\, t\ge t_0,\quad\forall\,\in {{{\mathbb R}^N}}\setminus Z_{n+1}, \quad
f_t(v)= h^{n+1}_t(v)+f_t^{n+1}(v).$$ This proves the lemma by induction, and the null set $Z$ can be chosen $Z=\bigcup_{n=1}^{\infty}Z_n$.
Note that the above iterations make sense since $f^0_t
= f_t\ge 0$ and, by induction, all $f^n_t$ are nonnegative. Note also that if $t_0>0$, then, by moment production estimate (Theorem \[(theo1.0)\]), we have $$\forall \, t \ge t_0,\quad f_t\in \bigcap_{s\ge 0} L^1_s({{{\mathbb R}^N}}).$$ This enables us to use moment estimates for $Q^{+}(f, Q^{+}(g,h))$: $$\label{(3.9)} \forall \, s\ge 0, \quad
\left\|Q^{+}\left(f, Q^{+}(g,h)\right)\right\|_{L^1_s} \le
\|f\|_{L^1_{s+{\gamma}}}\|g\|_{L^1_{s+2{\gamma}}}\|h\|_{L^1_{s+2{\gamma}}}.$$
Before we can show the regularity property of $f^n_t$ and the exponential decay (in norm) of $h^{n}_t$ we need further preparation.
Recall that the Sobolev space $H^{s}({{{\mathbb R}^N}})$ $(s>0)$ is a subspace of $f\in
L^2({{{\mathbb R}^N}})$ defined by $$f\in H^{s}({{{\mathbb R}^N}}) \Longleftrightarrow \|f\|_{H^{s}}=\|\widehat{f}\|_{L^2_s}
=\left(\int_{{{{\mathbb R}^N}}}{\langle}\xi{\rangle}^{2s}|\widehat{f}(\xi)|^2 \, {\rm d}\xi\right)^{1/2}
<\infty$$ where $\widehat{f}(\xi)$ is the Fourier transform of $f$: $$\widehat{f}(\xi)={\mathcal F}(f)(\xi)=\int_{{{{\mathbb R}^N}}}f(v)e^{-{\rm i}\xi\cdot v}
\, {\rm d}v.$$ As usual we denote the *homogeneous* seminorm as $$\|f\|_{\dot{H}^{s}}=\|\widehat{f}\|_{\dot{L}^2_{s}}
=\left(\int_{{{{\mathbb R}^N}}}|\xi|^{2s}|\widehat{f}(\xi)|^2 \, {\rm
d}\xi\right)^{1/2}.$$ The norm and seminorm are related by $$\label{(3.13-1)}
\| f \|_{\dot{H^s}} \le \|f\|_{H^{s}}\le (2\pi)^{N/2}2^{s/2}\|f\|_{L^2}+
2^{s/2}\|f\|_{\dot{H}^{s}}.$$
It is easily proved (see [@MR1663589 pp. 416-417]) that if the angular function $b$ satisfies $$\label{(3.12)}
\|b\|_{L^2}^2:=\left|{\mathbb
S}^{N-2}\right|\int_{0}^{\pi}b(\cos\theta)^2\sin^{N-2}\theta
\, {\rm d}\theta <\infty$$ then $Q^{+}: L^2_{N+{\gamma}}({{{\mathbb R}^N}})\times L^2_{N+{\gamma}}({{{\mathbb R}^N}})\to L^2({{{\mathbb R}^N}})$ is bounded with $$\label{(3.13-2)}\|Q^{+}(f,g)\|_{L^2}\le C\|b\|_{L^2}\|f\|_{L^2_{N+{\gamma}}}\|g\|_{L^2_{N+{\gamma}}}$$ where $C<\infty$ only depends on $N$ and ${\gamma}$. This together with , and the estimate of $\|Q^{+}(f,g)\|_{\dot{H}^{s}}$ obtained in [@MR1639275; @MR1663589] for $s=(N-1)/2$ leads to the following lemma.
\[(lemBD)\] Let $B(z,{\sigma})$ be defined in with the condition . Then $Q^{+}:
L^2_{N+{\gamma}}({{{\mathbb R}^N}})\times L^2_{N+{\gamma}}({{{\mathbb R}^N}})\to
H^{\frac{N-1}{2}}({{{\mathbb R}^N}})$ is bounded with the estimate $$\|Q^{+}(f,g)\|_{H^{\frac{N-1}{2}}}\le
C\|b\|_{L^2}\|f\|_{L^2_{N+{\gamma}}} \|g\|_{L^2_{N+{\gamma}}}$$ where $C<\infty$ only depends on $N,{\gamma}$.
The following lemma will be useful to prove the $H^1$-regularity of $f^n_t$ in the decomposition $f_t=f_t^n+h_t^n$.
\[(lem3.2)\] Let $B(z,{\sigma})$ be defined in with $\gamma\in(0,2]$ and with the condition . Let $F_t\in {\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}})$ be a conservative measure strong solution of equation . Then for any $t_0>0$ we have $$\label{(3.10)}
\forall\,t\ge t_0, \ \forall\, v\in{{{\mathbb R}^N}},\quad \int_{{{{\mathbb R}^N}}}|v-v_*|^{{\gamma}}\, {\rm d}F_t(v_*)
\ge a{\langle}v{\rangle}^{{\gamma}}\ge a\qquad$$ where $$\label{(3.11)} a:=\left[{\mathcal
K}_4\Big(1+\max\{1,\,1/t_0\}\Big)^{2/{\gamma}}\right]^{-(2-{\gamma})/2}$$ and ${\mathcal K}_4={\mathcal K}_4(1,1+N)\,(>1)$ is the constant in . In particular if $t_0\ge 1$, then $a$ is independent of $t_0$.
Moreover for any $t_0\le t_1\le t<\infty$, let $E_{t_1}^{t}(v)$ be defined as in for the measure $F_\tau$, i.e. $$E_{t_1}^{t}(v):=\exp\left(-\int_{t_1}^{t}
\int_{{{{\mathbb R}^N}}}|v-v_*|^{{\gamma}}\, {\rm d}F_\tau(v_*) \, {\rm
d}\tau \right).$$ Then for any $f\in L^{\infty}({{\mathbb R}^N}) \cap L^1_2({{\mathbb R}^N})\cap H^1({{{\mathbb R}^N}})$ we have $fE_{t_1}^{t}\in H^1({{{\mathbb R}^N}})$ and $$\label{(3-H-regularity)}
\left\|f E_{t_1}^{t}\right\|_{H^1({{\mathbb R}^N})}\le C\left[\|f\|_{L^{\infty}({{\mathbb R}^N})}+\|f\|_{L^1_2({{\mathbb R}^N})}
+\|f\|_{H^1({{\mathbb R}^N})} \right] e^{-a(t-t_1)}(1+t-t_1)$$ where $C$ only depends on $N,{\gamma}$.
Let $$L_s(F_t)(v):=\int_{{{{\mathbb R}^N}}}|v-v_*|^{s}\, {\rm d}F_t(v_*).$$ By conservation of mass, momentum and energy, we have $L_2(F_t)(v)=N+|v|^2>{\langle}v{\rangle}^2$ and so the inequality is obvious when ${\gamma}=2$. Suppose $0<{\gamma}<2$. In this case we use the inequality $|v-v_*|\le {\langle}v{\rangle}{\langle}v_*{\rangle}$ and the moment production estimate with $s=4$ to get $$L_4(F_t)(v)\le {\langle}v{\rangle}^4\left\|F_t\right\|_{4} \le {\langle}v{\rangle}^4{\mathcal
K}_4\left(1+1/t_0\right)^{\frac{2}{{\gamma}}}.$$ Then from the decomposition $2={\gamma}\cdot{\frac}{2}{4-{\gamma}}+4\cdot{\frac}{2-{\gamma}}{4-{\gamma}}$ and using Hölder inequality we have $$\begin{aligned}
{\langle}v{\rangle}^2&<& L_2(F_t)(v) \le
[L_{{\gamma}}(F_t)(v)]^{\frac{2}{4-{\gamma}}} [L_4(F_t)(v)]^{\frac{2-{\gamma}}{4-{\gamma}}}\\
&\le &
[L_{{\gamma}}(F_t)(v)]^{\frac{2}{4-{\gamma}}}{\langle}v{\rangle}^{\frac{4(2-{\gamma})}{4-{\gamma}}}\left[{\mathcal
K}_4(1+1/t_0)^{\frac{2}{{\gamma}}}\right]^{\frac{2-{\gamma}}{4-{\gamma}}}.\end{aligned}$$ This gives $$\forall\,t\ge t_0, \quad {\langle}v{\rangle}^{{\gamma}}\le L_{{\gamma}}(F_t)(v)\left[{\mathcal
K}_4(1+1/t_0)^{\frac{2}{{\gamma}}}\right]^{\frac{2-{\gamma}}{2}}\le \frac{1}{a}
L_{{\gamma}}(F_t)(v)$$ and follows.
The proof of is based on the following a priori estimates. First of all we have $$|{\partial}_{ v_j}E_{t_1}^{t}(v)|^2\le {\gamma}^2 e^{-2a(t-t_1)}(t-t_1)\int_{t_1}^{t}{\rm d}\tau\int_{{{{\mathbb R}^N}}}|v-v_*|^{2({\gamma}-1)}{\rm d}F_{\tau}(v_*)$$ where we used Cauchy-Schwartz inequality, $\|F_{\tau}\|_0=1$, and $E_{t_1}^{t}(v)\le e^{-a(t-t_1)}$.
[*Case 1: $0<{\gamma}<1$.*]{} In this case we have $-N<2({\gamma}-1)<0$ so that $$\int_{{{{\mathbb R}^N}}}|f(v)|^2|v-v_*|^{2({\gamma}-1)}\, {\rm d}v
\le C\|f\|_{L^{\infty}}^2
+\|f\|_{L^2}^2$$ hence $$\sum_{j=1}^N\|f {\partial}_{v_j}E_{t_1}^{t}(v) \|_{L^2}^2\le C \left(\|f\|_{L^{\infty}}^2
+\|f\|_{L^2}^2 \right)e^{-2a(t-t_1)}(t-t_1)^2.$$
[*Case 2: $ {\gamma}\ge 1$.*]{} Since ${\gamma}\le 2$, this implies $|v-v_*|^{2({\gamma}-1)}\le {\langle}v{\rangle}^2 {\langle}v_*{\rangle}^2$. Then recalling $\|F_{\tau}\|_2=1+N$ and $f\in L^{\infty}({{{\mathbb R}^N}})$ we have $$\int_{{{{\mathbb R}^N}}}\left(\int_{{{{\mathbb R}^N}}}|f(v)|^2
|v-v_*|^{2({\gamma}-1)}\, {\rm d}v\right)\, {\rm d}F_{\tau}(v_*)
\le(1+N)\|f\|_{L^{\infty}} \|f\|_{L^1_2}$$ which shows that $$\sum_{j=1}^N\left\|f {\partial}_{v_j}E_{t_1}^{t}\right\|_{L^2}^2
\le C \|f\|_{L^{\infty}} \|f\|_{L^1_2} e^{-2a(t-t_1)}(t-t_1)^2.$$
Combing the two cases and using $\|f\|_{L^{\infty}}\|f\|_{L^1_2}\le
\frac{1}{2}\|f\|_{L^{\infty}}^2+\frac{1}{2}\|f\|_{L^1_2}^2$ we obtain $$\left\|f E_{t_1}^{t} \right\|_{H^1({{\mathbb R}^N})}^2
\le C\left(\|f\|_{L^{\infty}({{\mathbb R}^N})}^2+\|f\|_{L^1_2({{\mathbb R}^N})}^2+
\|f\|_{H^1({{\mathbb R}^N})}^2\right)e^{-2a(t-t_1)}(1+t-t_1)^2.$$
A full justification requires standard smooth approximation arguments, for instance one may replace $f$ and $|v-v_*|^{{\gamma}}$ with $f*\chi_{{\varepsilon}}$ and $({\varepsilon}^2+|v-v_*|^2)^{{\gamma}/2}$ respectively, and then let ${\varepsilon}\to 0^{+}$, etc., where $\chi_{{\varepsilon}}(v)={\varepsilon}^{-N}\chi({\varepsilon}^{-1}v)$ and $\chi\ge 0$ is a smooth mollifier. We omit the details here.
\[(theo3.1)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined by with $\gamma\in(0, 2]$ and with the conditions -. Let $f_t\in L^1_{1,0,1}({{{\mathbb R}^N}})$ be a conservative mild solution of equation .
Then for any $t_0>0$, the positive decomposition $f_t=f^n_t+h^n_t$ given in (3.1)-(3.5) on $[t_0,\infty)$ satisfies the following estimates for all $s\ge 0$: $$\begin{aligned}
\label{(3.14)}&&
\sup_{n\ge N_{{\gamma}}+1,\, t\ge t_0} \left\|f_{t}^{n}\right\|_{L^{\infty}_s}\le
C_{t_0, s}\\
&&\label{(3.15)}\sup_{n\ge N_{{\gamma}}+2,\, t\ge t_0} \left\|f_{t}^{n}\right\|_{H^1}\le
C_{t_0}\\
&&\label{(3.16)}
\forall\, t\ge t_0,\ \forall\,n\ge 1, \quad \|h^n_t\|_{L^1_s}\le C_{t_0,s,n}e^{-\frac{a}{2}(t-t_0)} \\
&&\label{(3.15*)}
\forall\, t_1,t_2\ge t_0, \quad \sup_{n\ge
1}\left\|f^n_{t_1}-f^n_{t_2}\right\|_{L^1_s},\quad
\sup_{n\ge 1}\left\|h^n_{t_1}-h^n_{t_2}\right\|_{L^1_s}\le
C_{t_0,s}|t_1-t_2|.\end{aligned}$$ Here $N_{{\gamma}}$ is defined by , $a=a_{t_0}>0$ is given in , and $C_{t_0}, C_{t_0,s}, C_{t_0,s,n}$ are finite constants depending only on $N$, ${\gamma}$, the function $b$, $\max\{1,\,
1/t_0\}$, $s$, as well as $n$ in the case of $C_{t_0,s,n}$. In particular if $t_0\ge 1$, all these constants are independent of $t_0$.
Let $$\!|\!|\!|f\!|\!|\!|_{s}=\sup_{t\ge
t_0}\|f_{t}\|_{L^1_s},\quad s\ge 0.$$ By using Theorem \[(theo1.0)\], and the fact that $$\|f_0\|_{L^1_0}=1, \quad \|f_0\|_{L^1_2}=1+N$$ we have with ${\mathcal K}_s={\mathcal K}_s(1,1+N)$ that $$\forall\, s\ge 0, \quad \!|\!|\!|f\!|\!|\!|_{s}\le {\mathcal K}_s
\Big(1+\max\{1,\,1/t_0\}\Big)^{(s-2)^{+}/{\gamma}}.$$ We first prove and . To do this it suffices to prove the following estimates -:
$\bullet $ For all $s\ge 0$ and all $t\ge t_0$ $$\label{(3.17)}
\forall \, n \ge 1, \quad \left\|h^n_{t}\right\|_{L^1_s} \le
\!|\!|\!|f\!|\!|\!|_{s+(2n-1){\gamma}} ^{2n}(1+t-t_0)^{2n-1}e^{-a(t-t_0)}.$$
$\bullet $ If $0<{\gamma}<N-2$, we then define $$p_n:=\frac{N-1}{N-1-n{\gamma}}, \quad n=1,2,3,\dots,N_{{\gamma}},$$ and then for all $s\ge 0$ $$\begin{aligned}
\label{(3.18)}&&
\max_{1\le n\le N_{{\gamma}}}\sup_{t\ge t_0}\|f_{t}^{n}\|_{L^{p_{n}}_s}
\le C_{a} \, \!|\!|\!|f\!|\!|\!|_{s+s_1}^{\beta_*},\\
&&\label{(3.19)}
\sup_{n\ge N_{{\gamma}}+1}\sup_{t\ge t_0}\|f_{t}^{n}\|_{L^{\infty}_s}\le
C_{a} \, \!|\!|\!|f\!|\!|\!|_{s+s_1}^{\beta^*},\end{aligned}$$ where $s_1=N_{{\gamma}}+{\gamma}$ and $0<\beta_*, \beta^*<\infty$ depend only on $N$ and ${\gamma}$.
$\bullet $ If ${\gamma}\ge N-2$ and $1<p<N$, then for all $s\ge 0$, $$\begin{aligned}
\label{(3.20)} &&
\sup_{t\ge t_0}\|f_{t}^{1}\|_{L^p_s}\le C_{a,p} \,
\!|\!|\!|f\!|\!|\!|_{s+2{\gamma}-N/q} ^3,\\
&&\label{(3.21)}
\sup_{n\ge 2}\sup_{t\ge t_0}\|f_{t}^{n}\|_{L^{\infty}_s}\le
C_{a} \, \!|\!|\!|f\!|\!|\!|_{s+s_1}^{3+4/N},\end{aligned}$$ where $s_2=3{\gamma}+2-3N/2$.
In the following we denote by $C, C_{*}, C_{*,*}$ the finite positive constants (larger than $1$) that only depend on $N,{\gamma}, A_2,\|b\|_{L^{\infty}},$ and on the arguments “$*,*$”; they may have different values in different places.
By definition of $E_{s}^{t}$ (see ) and Lemma \[(lem3.2)\] we have $$\label{(3.22)}
\forall \, t_0\le t_2\le t_1\le t, \quad
\begin{cases} \displaystyle
E_{t_1}^{t} \le e^{-a(t-t_1)}, \vspace{0.2cm} \\ \displaystyle
E_{t_2}^{t_1} E_{t_1}^{t} \le e^{-a(t-t_2)}.
\end{cases}$$ We then deduce from and $0\le f^{n}_t\le f_t $ that $$\begin{aligned}
\label{(3.23)}&& h_t^{n}(v)\le
h_t^{1}(v)+ 2\int_{t_0}^{t}\int_{t_0}^{t_1}
e^{-a(t-t_2)}Q^{+}\left(f_{t_1},\,Q^{+}( f_{t_2},\,h^{n-1}_{t_2})
\right)(v)\, {\rm d}t_2 \, {\rm d}t_1\\
&&\nonumber
\qquad \,\,\,\, +\int_{t_0}^{t}\int_{t_0}^{t_1}
e^{-a(t-t_2)}Q^{+}\left(h^{n-1}_{t_1},\,Q^{+}(f_{t_2}, f_{t_2})\right)(v)\, {\rm
d}t_2 \, {\rm d}t_1,\\
\label{(3.24)} && f^{n}_t(v) \le \int_{t_0}^{t}\int_{t_0}^{t_1}
e^{-a(t-t_2)} Q^{+}\left(f^{n-1}_{t_1},\,Q^{+}( f^{n-1}_{t_2},
f^{n-1}_{t_2})\right)(v)\, {\rm d}t_2 \, {\rm d}t_1.\end{aligned}$$
Next by definition of $h_t^{1}$ in and using and $\|f_{t}\|_{L^1_s}\ge \|f_t\|_{L^1}=1$ we have $$\|h_t^{1}\|_{L^1_s} \le
\!|\!|\!|f\!|\!|\!|_{s+{\gamma}}^2(1+t-t_0)e^{-a(t-t_0)}.$$ Suppose holds for some $n\ge 1$. Using for $h^{n+1}_t$ we have $$\begin{gathered}
\|h^{n+1}_t\|_{L^1_s} \le
e^{-a(t-t_0)} \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}}^2(1+t-t_0)+2 \,
\!|\!|\!|f\!|\!|\!|_{s+2{\gamma}}^2
\int_{t_0}^{t}\int_{t_0}^{t_1}\|h^{n}_{t_2}\|_{L^1_{s+2{\gamma}}}
e^{-a(t-t_2)} \, {\rm d}t_2 \, {\rm d}t_1
\\
+ \!|\!|\!|f\!|\!|\!|_{s+2{\gamma}}^2
\int_{t_0}^{t}\|h^{n}_{t_1}\|_{L^1_{s+{\gamma}}}\int_{t_0}^{t_1}
e^{-a(t-t_2)} \, {\rm d}t_2 \, {\rm d}t_1
\\
= \!|\!|\!|f\!|\!|\!|_{s+2{\gamma}}^2e^{-a(t-t_0)}\left(1+t-t_0+
\int_{t_0}^{t} \int_{t_0}^{t_1}e^{a(t_2-t_0)}\left(2\|h^{n}_{t_2}\|_{L^1_{s+2{\gamma}}}+\|h^{n}_{t_1}\|_{L^1_{s+{\gamma}}}
\right)\, {\rm d}t_2 \, {\rm
d}t_1 \right)\end{gathered}$$ and by inductive hypothesis on $h^{n}_t$ we have for all $t_0\le
t_2\le t_1$, $$\begin{gathered}
2\left\|h^{n}_{t_2}\right\|_{L^1_{s+2{\gamma}}}+\left\|h^{n}_{t_1}\right\|_{L^1_{s+{\gamma}}}
\le 2\left(\|f\|_{s+2{\gamma}+(2n-1){\gamma}}\right)^{2n}(1+t_2-t_0)^{2n-1}e^{-a(t_2-t_0)}
\\
+ \left(\|f\|_{s+{\gamma}+(2n-1){\gamma}}\right)^{2n}(1+t_1-t_0)^{2n-1}e^{-a(t_1-t_0)}\\
\le
3 \, \!|\!|\!|f\!|\!|\!|_{s+(2n+1){\gamma}}
^{2n}(1+t_1-t_0)^{2n-1}e^{-a(t_2-t_0)}.\end{gathered}$$ So $$\|h^{n+1}_t\|_{L^1_s} \le
\!|\!|\!|f\!|\!|\!|_{s+(2n+1){\gamma}}^{2(n+1)}e^{-a(t-t_0)}\left(
1+t-t_0+3 \int_{t_0}^{t}(1+t_1-t_0)^{2n-1}(t_1-t_0)\, {\rm d}t_1
\right).$$ It is easily checked that $$\forall \, t \ge t_0, \quad 1+t-t_0+3\int_{t_0}^{t}(1+t_1-t_0)^{2n-1}(t_1-t_0) \, {\rm d}t_1\le
(1+t-t_0)^{2n+1}.$$ Thus $$\|h^{n+1}_t\|_{L^1_s} \le
\!|\!|\!|f\!|\!|\!|_{s+(2n+1){\gamma}}^{2(n+1)}(1+t-t_0)^{2n+1}e^{-a(t-t_0)}.$$ This proves .
Now we are going to prove -. First of all by and the inequality $$\int_{t_0}^{t}\int_{t_0}^{t_1}e^{-a(t-t_2)}\, {\rm d}t_2 \, {\rm d}t_1 \le \frac{1}{a^2}$$ we have $$\label{(3.25)}
\sup_{t\ge t_0}\|f^n_t\|_{L^p_s}\le \frac{1}{a^2} \left( \sup_{t_1\ge t_2\ge t_0}
\left\|Q^{+}(f^{n-1}_{t_1},\,Q^{+}( f^{n-1}_{t_2},
f^{n-1}_{t_2}) )\right\|_{L^p_s} \right)$$ for all $s\ge 0, 1\le p\le \infty$, provided that the right hand side makes sense.
[*Case 1: $0<{\gamma}<N-2$.*]{} We first prove that $$\label{(3.26)}
\forall\,s\ge 0,\quad \sup_{t\ge t_0}\|f^n_t\|_{L^{p_{n}}_s}
\le C_{a,n}\left( \!|\!|\!|f\!|\!|\!|_{s+n-1+{\gamma}_1}\right)^{\beta_n},\qquad
n=1,2,\dots, N_{{\gamma}}$$ where ${\gamma}_1=\max\{{\gamma}, 1\}$ and $$\beta_n:=2(N+1)\left(1+\frac{1}{N}\right)^{n-1} +1-2N.$$
By part (i) of Theorem \[(theo2.1)\] we have $$\forall \, t_1\ge t_2\ge t_0, \quad
\left\|Q^{+}\left(f_{t_1},\,Q^{+}(f_{t_2}, f_{t_2})\right)\right\|_{L^{p_1}_s} \le
C_{1} \|f_{t_1}\|_{L^1_{s}}\|f_{t_2}\|_{L^1_{s+{\gamma}}}^2 \le
C_{1} \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}_1}^3.$$ Using with $p=p_1$ and $n=1$ (recalling $f^{(0)}_t(v)=f_t(v)$) gives $$\forall \, s\ge 0, \quad \sup_{t\ge t_0}\left\|f^1_t\right\|_{L^{p_{1}}_s} \le
C_{a,1} \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}_1}^3.$$ Since $\beta_1=3$, this proves that the inequality in holds for $n=1$.
Suppose the inequality in holds for some $1\le n\le
N_{{\gamma}}-1$. Then we compute using $0\le f^n_t\le f_t$ and part (I) of Theorem \[(theo2.1)\] that, for all $s\ge 0$, $$\begin{gathered}
\label{(3.27)}
\left\|Q^{+}\left(f^{n}_{t_1},\,Q^{+}( f^n_{t_2},
f^n_{t_2})\right)\right\|_{L^{p_{_{n+1}}}_s}
\le C_{n}\|f^n_{t_1}\|_{L^{p_n}_{s+1}}
\|f^n_{t_2}\|_{L^1_{s+{\gamma}_1}}
\|f^n_{t_2}\|_{L^1_{s+{\gamma}_1}}^{1-\theta_n} \|f^n_{t_2} \|_{L^{p_n}_{s+1}}^{\theta_n} \\
\le C_{n} \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}_1}^{2-\theta_n} \left(\sup_{t\ge
t_0}\|f_{t}^{n}\|_{L^{p_n}_{s+1}}\right)^{1+\theta_n}.\end{gathered}$$ By inductive hypothesis on $f^n_t$ we compute $$\label{(3.28)}
\!|\!|\!|f\!|\!|\!|_{s+{\gamma}_1}^{2-\theta_n} \left(\sup_{t\ge
t_0}\|f_{t}^{n}\|_{L^{p_n}_{s+1}}\right)^{1+\theta_n} \le C_{a,n} \,
\!|\!|\!|f\!|\!|\!|_{s+n+{\gamma}_1}^{2-\theta_n+\beta_n(1+\theta_n)}.$$ Also by definition of $\theta_n$ and $\beta_n$ it is easily checked that $2-\theta_n+\beta_n(1+\theta_n)<\beta_{n+1}$. It then follows from , and that $$\forall\, s\ge 0, \quad \sup_{t\ge t_0}\|f^{n+1}_t\|_{L^{p_{_{n+1}}}_s}\le C_{a,n+1} \,
\!|\!|\!|f\!|\!|\!|_{s+n+{\gamma}_1}^{\beta_{n+1}}.$$ This proves that the inequality in holds for all $n=1,2,\dots,N_{{\gamma}}$. From and $N_{{\gamma}}-1+{\gamma}_1<
N_{{\gamma}}+{\gamma}=s_1$, we obtain with $\beta_*=\beta_{_{N_{{\gamma}}}}$.
Next let us prove . By Theorem \[(theo2.1)\] (see ,) and using with $n=N_{{\gamma}}$ we have $$\begin{gathered}
\left\|Q^{+}(f^{N_{{\gamma}}}_{t_1},\,Q^{+}(
f^{N_{{\gamma}}}_{t_2}, f^{N_{{\gamma}}}_{t_2})
)\right\|_{L^{\infty}_s}
\le C\|f^{N_{{\gamma}}}_{t_1}\|_{L^1_s}^{1-\alpha_1}
\|f^{N_{{\gamma}}}_{t_1}\|_{L^{p_{_{N_{{\gamma}}}}}_s}^{\alpha_1}
\|f^{N_{{\gamma}}}_{t_2}\|_{L^1_{s+{\gamma}}}
\|f^{N_{{\gamma}}}_{t_2}\|_{L^1_{s+{\gamma}}}^{1-\alpha_2}\|f^{N_{{\gamma}}}_{t_2}
\|_{L^{p_{_{N_{{\gamma}}}}}_s}^{\alpha_2}
\\
\le
C\left(\!|\!|\!|f\!|\!|\!|_{s+s_1}\right)^{3+(\beta_{_{N_{{\gamma}}}}-1)(\alpha_1+\alpha_2)}.\end{gathered}$$ This together with gives $$\label{(3.29)}
\sup_{t\ge t_0}\left\|f^{{N_{{\gamma}}}+1}_t\right\|_{L^{\infty}_s} \le
C_{a}\, \!|\!|\!|f\!|\!|\!|_{s+s_1} ^{\eta},\quad
\eta:=3+(\beta_{_{N_{{\gamma}}}}-1)(\alpha_1+\alpha_2).$$ Using with $p=\infty$, Theorem \[(theo2.1)\], and $$\|f^{N_{{\gamma}}+k}_t\|_{L^{p_{_{N_{{\gamma}}}}}_s}
\le \!|\!|\!|f\!|\!|\!|_{s}^{1/p_{N_{{\gamma}}}}
\left\|f^{N_{{\gamma}}+k}_t\right\|_{L^{\infty}_s}^{1/q_{N_{{\gamma}}}}$$ together with the $L^{\infty}_s$-boundedness for $k=1$, we deduce by induction on $k$ that, for all $s\ge 0$, $$\begin{gathered}
\label{(3.30)}
\sup_{t\ge t_0}\left\|f^{N_{{\gamma}}+k+1}_t\right\|_{L^{\infty}_s} \le
\frac{1}{a^2}\sup_{t_1\ge t_2\ge t_0}\left\|Q^{+}\left(f^{N_{{\gamma}}+k}_{t_1},\,Q^{+}\left( f^{N_{{\gamma}}+k}_{t_2},
f^{N_{{\gamma}}+k}_{t_2}\right)\right)\right\|_{L^{\infty}_s}
\\
\le C_a\sup_{t_1\ge t_2\ge t_0}\Bigg\{\left\|f^{N_{{\gamma}}+k}_{t_1}\right\|_{L^1_s}^{1-\alpha_1}
\left\|f^{N_{{\gamma}}+k}_{t_1}\right\|_{L^{p_{_{N_{{\gamma}}}}}_s}^{\alpha_1}\\
\times
\left\|f^{N_{{\gamma}}+k}_{t_2}\right\|_{L^1_{s+{\gamma}}}
\left\|f^{N_{{\gamma}}+k}_{t_2}\right\|_{L^1_{s+{\gamma}}}^{1-\alpha_2}
\left\|f^{N_{{\gamma}}+k}_{t_2}\right\|_{L^{p_{N_{{\gamma}}}}_s}^{\alpha_2}\Bigg\}
\\
\le C_a \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}}^{3-(\alpha_1+\alpha_2)/{q_{N_{{\gamma}}}}} \left(\sup_{t\ge
t_0}\|f^{N_{{\gamma}}+k}_t\|_{L^{\infty}_s}\right)^{(\alpha_1+\alpha_2)/{q_{N_{{\gamma}}}}}
\\
=C_a \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}} ^{3-{\delta}} \left(\sup_{t\ge
t_0}\|f^{N_{{\gamma}}+k}_t\|_{L^{\infty}_s}\right)^{{\delta}}<\infty,\quad
k=1,2,3,\dots\end{gathered}$$ where $${\delta}:=\frac{\alpha_1+\alpha_2}{q_{_{N_{{\gamma}}}}}=\frac{N-1-{\gamma}+(1-{\gamma})^{+}}{N}
\,\,\,\,(<1\,).$$ Now fix any $s\ge 0$ and let us define $$A=C_a \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}} ^{3-{\delta}} \quad \mbox{ and } \quad Y_k=\sup_{t\ge
t_0}\left\|f^{N_{{\gamma}}+k}_t\right\|_{L^{\infty}_s}.$$ Then, from , $$Y_{k+1}\le AY_k^{{\delta}},\quad k=1,2,\dots$$ which gives $$Y_{k+1}\le A^{1+{\delta}+\cdots+{\delta}^{k-1}}Y_{1}^{{\delta}^{k}}=
A^{\frac{1-{\delta}^{k}}{1-{\delta}}}Y_{1}^{{\delta}^{k}}\le A^{\frac{1}{1-{\delta}}}Y_{1},\quad
k=1,2,\dots$$ It follows from and ${\gamma}<s_1$ that $$\sup_{t\ge
t_0}\left\|f^{N_{{\gamma}}+k+1}_t\right\|_{L^{\infty}_s}=Y_{k+1} \le
\left(C_{a} \, \!|\!|\!|f\!|\!|\!|_{s+{\gamma}}^{3-{\delta}}\right)^{\frac{1}{1-{\delta}}}
C_{a} \, \!|\!|\!|f\!|\!|\!|_{s+s_1}^{\eta} \le
C_{a} ^{1+1/(1-\delta)} \, \!|\!|\!|f\!|\!|\!|_{s+s_1}^{\frac{3-{\delta}}{1-{\delta}}+\eta}$$ for all $k=1,2,3,\dots$ This gives with $\beta^*= (3-{\delta})/(1-{\delta})+\eta$.
[*Case 2: ${\gamma}\ge N-2$.*]{} By Theorem \[(theo2.1)\] we have for any $1<p<N$ and $s\ge 0$ $$\left\|Q^{+}\left(f_{t_1},\,Q^{+}( f_{t_2}, f_{t_2})\right)\right\|_{L^p_s}\le C_{p}
\|f_{t_1}\|_{L^1_{s+{\gamma}-N/q}} \|f_{t_2}\|_{L^1_{s+2{\gamma}-N/q}}^2 \le C_{p}
\, \!|\!|\!|f\!|\!|\!|_{s+2{\gamma}-N/q}^3.$$ This together with with $n=1$ implies that $$\forall\,s\ge 0, \quad \sup\limits_{t\ge t_0}\|f^1_t\|_{L^p_s}\le C_{a,p}
\, \!|\!|\!|f\!|\!|\!|_{s+2{\gamma}-N/q}^3.$$ This proves . In particular for $p=2$ we have $$\label{(3.31)}
\forall\, s\ge 0, \quad \sup_{t\ge t_0}\|f^1_t\|_{L^2_s}\le C_{a}
\, \!|\!|\!|f\!|\!|\!|_{s+2{\gamma}-N/2}^3.$$ Then using with $p=\infty$, Theorem \[(theo2.1)\] with $$p=q=2\in \left(N/(N-1),\, N\right),$$ and induction on $n$ starting from $n=1$ with the $L^2_s$-boundedness we have, for all $s\ge 0$, $$\begin{aligned}
&&\label{(3.32)} \sup_{t\ge
t_0}\left\|f^{n+1}_t\right\|_{L^{\infty}_s}\le \frac{1}{a^2} \sup_{t_1\ge
t_2\ge t_0} \left\|Q^{+}\left(f^{n}_{t_1},\,Q^{+}\left(f^{n}_{t_2},f^{n}_{t_2}\right)\right)
\right\|_{L^{\infty}_s} \\
&&\le C_a \sup_{t\ge t_1\ge
t_0}\left(\left\|f^{n}_{t_1}\right\|_{L^1_{s+{\gamma}+2-N}}^{1-2/N}
\left\|f^{n}_{t_1}\right\|_{L^2_{s+{\gamma}+2-N}}^{2/N}\left\|f^{n}_{t_2}\right\|_{L^1_{s+2{\gamma}+1-N}}^2\right)
\nonumber \\
&&\nonumber\le C_a \, \!|\!|\!|f\!|\!|\!|_{s+2{\gamma}+1-N})^{3-2/N}
\left(\sup_{t\ge t_0}\left\|f^{n}_{t}\right\|_{L^2_{s+{\gamma}+2-N}}\right)^{2/N}<\infty,\quad
n=1,2,3,\dots\end{aligned}$$ Taking $n=1$ and using and $2{\gamma}+1-N< 3{\gamma}+2-3N/2=:s_2$ we obtain $$\label{(3.33)}
\sup_{t\ge t_0}\left\|f^2_t\right\|_{L^{\infty}_s}\le C_{a} \,
\!|\!|\!|f\!|\!|\!|_{s+s_2}^{3+4/N}.$$ Further, using $$\forall \, n\ge 2, \quad \left\|f^{n}_{t}\right\|_{L^2_{s+{\gamma}+2-N}} \le
\!|\!|\!|f\!|\!|\!|_{s+2{\gamma}+4-2N}^{1/2}
\left\|f^{n}_{t}\right\|_{L^{\infty}_s} ^{1/2}$$ and $2{\gamma}+4-2N\le 2{\gamma}+1-N\le s_2$ (because ${\gamma}\ge N-2\ge 1$) we get from that $$\sup_{t\ge t_0}\left\|f^{n+1}_t\right\|_{L^{\infty}_s}\le C_{a}
\, \!|\!|\!|f\!|\!|\!|_{s+s_2}^{3-1/N} \left(\sup_{t\ge
t_0}\|f^{n}_{t}\|_{L^{\infty}_{s}}\right)^{1/N},\quad
n=2,3,\dots$$ By iteration we deduce, as shown above with ${\delta}=1/N$, and using that $$\begin{aligned}
&&
\sup_{t\ge t_0}\|f^{n+1}_t\|_{L^{\infty}_s}\le \left(C_{a}
\, \!|\!|\!|f\!|\!|\!|_{s+s_2}^{3-1/N}\right)^{\frac{1-{\delta}^{n-1}}{1-{\delta}}}
\left(\sup_{t\ge t_0}\|f^2_t\|_{L^{\infty}_s}\right)^{{\delta}^{n-1}}\\
&&\le C_a \left(\!|\!|\!|f\!|\!|\!|_{s+s_2}\right)^{
(3-\frac{1}{N})\frac{1-{\delta}^{n-1}}{1-{\delta}}+(3+4/N){\delta}^{n-1}}\le
C_{a}\left(\!|\!|\!|f\!|\!|\!|_{s+s_2}\right)^{3+4/N},\quad
n=2,3,\dots\end{aligned}$$ This proves .
Now let us prove the $H^1$-regularity of $f^{n}_t$ for $n\ge N_{{\gamma}}+2$. For notation convenience we denote $$Q_{t_1, t_2}^{n-1}(v):=Q^{+}\left(f^{n-1}_{t_1},\,Q^{+}\left(f^{n-1}_{t_2},
f^{n-1}_{t_2}\right)E_{t_2}^{t_1}\right)(v).$$ The iteration formula is then written $$\label{(3-iteration)} \forall \, t\ge t_0, \quad
f_{t}^{n}(v) =\int_{t_0}^{t}E_{t_1}^{t}(v)\int_{t_0}^{t_1}Q_{t_1,
t_2}^{n-1}(v)\, {\rm d}t_2 \, {\rm d}t_1.$$ Applying Theorem \[(theo2.1)\] and the $L^{\infty}_s$ estimate in we have $$\left\|Q_{t_1, t_2}^{n-1} \right\|_{L^{\infty}} \le
e^{-a(t_1-t_2)}\left\|Q^{+}\left(f^{n-1}_{t_1},\,Q^{+}(f^{n-1}_{t_2},
f^{n-1}_{t_2})\right)\right\|_{L^{\infty}} \le
C_{t_0}e^{-a(t_1-t_2)}.$$ Also by $f^{n-1}_t\le f_t$ we have $$\left\|Q_{t_1, t_2}^{n-1}\right\|_{L^1_2} \le
e^{-a(t_1-t_2)}\left\|Q^{+}\left(f^{n-1}_{t_1},\,Q^{+}\left(f^{n-1}_{t_2},
f^{n-1}_{t_2}\right)\right)\right\|_{L^1_2} \le C_{t_0}e^{-a(t_1-t_2)}.$$ And using Lemma \[(lemBD)\], and the $L^{\infty}_s$ estimate in we have $$\begin{gathered}
\left\|Q_{t_1,t_2}^{n-1}\right\|_{H^1}\le
\left\|Q_{t_1,t_2}^{n-1}\right\|_{H^{\frac{N-1}{2}}} \le C
\left\|f^{n-1}_{t_1}\right\|_{L^2_{N+{\gamma}}} \left\|Q^{+}(f^{n-1}_{t_2},
f^{n-1}_{t_2})E_{t_2}^{t_1}\right\|_{L^2_{N+{\gamma}}}\\
\le C\left\|f^{n-1}_{t_1}\right\|_{L^2_{N+{\gamma}}}\left\|f^{n-1}_{t_2}\right\|_{L^2_{2(N+{\gamma})}}^2e^{-a(t_1-t_2)}
\le C_{t_0}e^{-a(t_1-t_2)}.\end{gathered}$$ Thus we conclude from Lemma \[(lem3.2)\] that $Q_{t_1,t_2}^{n-1}E_{t_1}^t\in H^1({{{\mathbb R}^N}})$ and $$\label{(3-Q-E)}
\left\|Q_{t_1,t_2}^{n-1}E_{t_1}^t\right\|_{H^1}\le C_{t_0}e^{-a(t_1-t_2)} e^{-a(t-t_1)}(1+t-t_1)
= C_{t_0}e^{-a(t-t_2)}(1+t-t_1).$$ Using Minkowski inequality to we then conclude from and the above estimates that $f^{n}_t\in
H^{1}({{{\mathbb R}^N}})$ and $$\left\|f^{n}_t\right\|_{H^1}\le\int_{t_0}^{t}\int_{t_0}^{t_1}
\left\|Q_{t_1,t_2}^{n-1} E_{t_1}^{t}\right\|_{H^1}\, {\rm d}t_2 \, {\rm d}t_1\le
C_{t_0}\int_{t_0}^{t}\int_{t_0}^{t_1}e^{-a(t-t_2)}(1+t-t_1)\, {\rm
d}t_2 \, {\rm d}t_1\le C_{t_0}.$$ This proves .
Finally let us prove . To do this we rewrite $f^n_t$ as follows (recall definition of $f^n_t$ in ) $$f_{t}^{n}(v)
=E_{t_0}^t(v)\int_{t_0}^{t}E_{t_1}^{t_0}(v) \int_{t_0}^{t_1}
Q^{+}\left( f^{n-1}_{t_1},\,Q^{+}\left(f^{n-1}_{t_2}, f^{n-1}_{t_2}\right)E_{t_2}^{t_1}
\right)(v)\, {\rm d}t_2 \, {\rm d}t_1$$ and recall that $$E_{s}^t(v) =\exp\left(-\int_{s}^{t}L_{{\gamma}}(f_{\tau})(v)\, {\rm d}\tau
\right),\quad
L_{{\gamma}}(f_{\tau})(v):=\int_{{{{\mathbb R}^N}}}|v-v_*|^{{\gamma}}f_\tau(v_*) \, {\rm
d}v_*.$$ Then it is easily seen that the function $t\mapsto f^n_t(v)$ is absolutely continuous on every bounded subinterval of $[t_0,\infty)$ and $$\frac{{\partial}}{{\partial}t}f_{t}^{n}(v)
=\int_{t_0}^{t} Q^{+}\left(f^{n-1}_{t},\,Q^{+}\left(
f^{n-1}_{t_2}, f^{n-1}_{t_2}\right)E_{t_2}^{t}\right)(v)\, {\rm d}t_2
-L_{{\gamma}}(f_t)(v)f^n_t(v),\quad {\rm a.e.}\,\,\, t\ge t_0.$$ Since $$\begin{gathered}
\left\| Q^{+}\left( f^{n-1}_{t},\,Q^{+}\left(
f^{n-1}_{t_2}, f^{n-1}_{t_2}\right)E_{t_2}^{t}\right)\right\|_{L^1_s} \le
\left\| Q^{+}(f_{t},\,Q^{+}( f_{t_2},
f_{t_2}))\right\|_{L^1_s}e^{-a(t-t_2)} \\
\le
\left\|f_{t}\right\|_{L^1_{s+{\gamma}}}
\left\|f_{t_2}\right\|_{L^1_{s+2{\gamma}}}^2 e^{-a(t-t_2)}
\le C_{t_0,s}e^{-a(t-t_2)}\end{gathered}$$ and $$\left\|L_{{\gamma}}(f_t)f^n_t\right\|_{L^1_s}\le
\left\|f_t\right\|_{{\gamma}}\left\|f_t\right\|_{L^1_{s+{\gamma}}} \le
C_{t_0,s}$$ it follows that $$\left\|\frac{{\partial}}{{\partial}t}f_{t}^{n}\right\|_{L^1_s}\le
C_{t_0,s} \quad \mbox{ a.e. } t\ge t_0.$$ Thus, by the absolute continuity of $t\mapsto f^n_t(v)$, we deduce that $$\forall \, t_1, t_2 \ge t_0, \quad \left\|f^n_{t_1}-f^n_{t_2}\right\|_{L^1_s}\le
C_{t_0,s}|t_1-t_2|.$$
On the other hand, from $$\forall\,t\ge
t_0, \quad f_t(v)=f_{t_0}(v)+\int_{t_0}^{t}\left[Q^{+}(f_{\tau},f_{\tau})(v)
-L_{{\gamma}}(f_{\tau})(v)f_{\tau}(v)\right]\, {\rm d}\tau$$ we also have $\left\|f_{t_1}-f_{t_2}\right\|_{L^1_s}\le
C_{t_0,s}|t_1-t_2|$ for all $t_1,t_2\ge t_0$. Thus the function $t\mapsto h^n_t=f_t-f^n_t$ satisfies the same estimate. This proves and completes the proof of the theorem.
\[(corol3.2)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined in with $\gamma \in
(0,2]$ and with the conditions ,. Let $F_t\in B^{+}_{1,0,1}({{{\mathbb R}^N}})$ be a conservative measure strong solution of equation . Then for any $t_0>0$, $F_t$ can be decomposed as $$\label{(3.45)}
\forall \, t\ge t_0, \quad \, {\rm d}F_t(v)=f_t(v)\, {\rm d}v+\,
{\rm d}\mu_t(v),$$ with $$0\le f_t\in \bigcap_{s\ge 0}L^{\infty}_s\cap H^1({{{\mathbb R}^N}}), \quad
\mu_t\in \bigcap_{s\ge 0}{\mathcal B}^{+}_s({{{\mathbb R}^N}}),$$ satisfying for all $s\ge 0$ $$\begin{aligned}
&&\label{(3.47)}
\sup_{t\ge t_0}\left\|f_t\right\|_{L^{\infty}_s}\le C_{t_0, s}, \qquad \sup_{t\ge t_0}\left\|f_t\right\|_{H^{1}}\le C_{t_0}\\
&&\label{(3.48)} \forall\, t\ge t_0, \quad \left\|\mu_t\right\|_s\le
C_{t_0, s}e^{-\frac{a}{2}(t-t_0)},
\\
&&\label{(3.46)} \forall\, t_1,t_2\in[t_0,\infty), \quad
\left\|f_{t_1}-f_{t_2}\right\|_{L^1_s},\,\,\left\|\mu_{t_1}-\mu_{t_2}\right\|_s\le
C_{t_0,s}|t_1-t_2|,\end{aligned}$$ where $a=a_{t_0}>0$ is given in and $C_{t_0},
C_{t_0,s}$ are finite constant depending only on $N$, ${\gamma}$, the function $b$, $\max\{1,\, 1/t_0\}$ and $s$.
By Theorem \[(theo1.0)\], there is a sequence $\{f_{k,t}\}_{k=1}^{\infty}\subset L^1_{1,0,1}({{{\mathbb R}^N}})$ of conservative mild solutions of equation such that $$\label{(3.49)}
\forall\,{\varphi}\in C_c({{{\mathbb R}^N}}),\quad
\forall\,t\ge 0, \quad \lim_{k\to\infty}\int_{{{{\mathbb R}^N}}}{\varphi}(v) f_{k,t}(v)\, {\rm d}v
=\int_{{{{\mathbb R}^N}}}{\varphi}(v)\, {\rm d}F_t(v).$$ Let $n_{{\gamma}}=N_{{\gamma}}+2$ with $N_{{\gamma}}$ defined in and consider the positive decompositions of $f_{k,t}$ $$\forall \, k=1,2,3,\dots \ \forall \, t\ge t_0, \quad
f_{k,t}(v)=f_{k,t}^{n_{{\gamma}}}(v)+h_{k,t}^{n_{{\gamma}}}(v),\quad$$ given by -. By Theorem \[(theo3.1)\] we have for all $s\ge 0$ $$\begin{aligned}
\label{(3.50)}&&\sup_{k\ge 1, t\ge t_0}\left\|f_{k,t}^{n_{{\gamma}}}\right\|_{L^{\infty}_s}\le
C_{t_0, s},\quad \sup_{k\ge 1,\,t\ge t_0}\left\|f_{k,t}^{n_{{\gamma}}}\right\|_{H^1}\le
C_{t_0},\\
&&\label{(3.51)}
\forall \, t\ge t_0, \quad \sup_{k\ge 1}\left\|h_{k,t}^{n_{{\gamma}}}\right\|_{L^1_s}\le C_{t_0, s}
e^{-\frac{a}{2}(t-t_0)}, \\
&&\label{(3.52)}\forall\, t_1, t_2\ge t_0, \quad
\sup_{k\ge 1}\left\|f^{n_{{\gamma}}}_{k,t_1}-f^{n_{{\gamma}}}_{k,t_2}\right\|_{L^1_s},\,\,
\sup_{k\ge 1}\left\|h^{n_{{\gamma}}}_{k,t_1}-h^{n_{{\gamma}}}_{k,t_2}\right\|_{L^1_s}\le
C_{t_0,s}|t_1-t_2|.\end{aligned}$$ From , it is easily seen that for every $t\ge t_0$, $\{f^{n_{{\gamma}}}_{k,t}\}_{k=1}^{\infty}$ is relatively compact in $L^1({{{\mathbb R}^N}})$. Moreover by using the density of rational times, a diagonal process and , one can prove that there is a common subsequence $\{f^{n_{{\gamma}}}_{k_j,t}\}_{j=1}^{\infty}$ (where $\{k_j\}_{j=1}^{\infty}$ is independent of $t$) and a function $0\le f_t\in
L^1({{{\mathbb R}^N}})$, such that $$\label{(3.53)}
\forall\,t\ge t_0, \quad \left\|f^{n_{{\gamma}}}_{k_j,t}-f_t\right\|_{L^1}
\xrightarrow[j\to\infty]{} 0.$$ Since $h^{n_{{\gamma}}}_{k_j,t}=f_{k_j,t}-f^{n_{{\gamma}}}_{k_j,t}$, it follows from and the weak convergence that for every $t\ge t_0$, $h^{n_{{\gamma}}}_{k_j,t}$ converges weakly to some $\mu_t\in{\mathcal B}_0^{+}({{{\mathbb R}^N}})$ as $j\to\infty$, i.e. $$\label{(3.54)} \forall\,{\varphi}\in C_c({{{\mathbb R}^N}}),\
\forall\,t\ge t_0, \quad \lim_{j\to\infty}\int_{{{{\mathbb R}^N}}}{\varphi}(v)
h_{k_j,t}^{n_{{\gamma}}}(v)\, {\rm d}v =\int_{{{{\mathbb R}^N}}}{\varphi}(v)\, {\rm
d}\mu_t(v).$$ This leads to the decomposition . The inequalities , , follow easily from , , , , and the equivalent version of measure norm $\|\cdot\|_s$.
Rate of Convergence to Equilibrium {#sec4}
===================================
This section is devoted to the proof of Theorem \[(theo1.1)\]. We first recall the results in [@Mcmp] on the exponential rate of convergence to equilibrium for $L^1$ mild solutions.
\[(theo4.1)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined in with ${\gamma}\in (0, \min\{2, N-2\}]$ and with the conditions --. Let ${\lambda}=S_{b,{\gamma}}(1,0, 1)>0$ be the spectral gap of the linear operator $L_M$ in associated with $B(z,{\sigma})$ and the Maxwellian $M(v)=(2\pi)^{-N/2}e^{-|v|^2/2}$ in $L^1_{1,0,1}({{{\mathbb R}^N}})$.
Let $f_0\in L^1_{1,0,1}({{{\mathbb R}^N}})\cap L^2({{{\mathbb R}^N}})$ and let $f_t\in
L^1_2({{{\mathbb R}^N}}) $ be the unique conservative solution of equation with the initial datum $f_0$. Then there is a constant $0<C<\infty$, which depends only on $N$, ${\gamma}$, the function $b$, and on $\left\|f_0\right\|_{L^2}$, such that $$\forall\, t\ge 0, \quad \left\|f_t-M\right\|_{L^1}\le Ce^{-{\lambda}t}.$$ In the important case of hard sphere model (i.e. $N=3$, ${\gamma}=1$, and $b={\rm const.}$), the assumption “$ f_0\in L^1\cap L^2$" can be relaxed to “$ f_0\in L^1$" and the same result holds with the constant $C$ depending only on $N$, ${\gamma}$, and the function $b$.
\[(lem4.1)\] Suppose $N\ge 3$ and let $B(z,{\sigma})$ be defined in with ${\gamma}\in (0, \min\{2, N-2\}]$, and with the conditions --. Let ${\lambda}=S_{b,{\gamma}}(1,0, 1)>0$ be the spectral gap of the linear operator $L_M$ in associated with $B(z,{\sigma})$ and the Maxwellian $M(v)=(2\pi)^{-N/2}e^{-|v|^2/2}$ in $L^1_{1,0,1}({{{\mathbb R}^N}})$. Let $\alpha>0$, $m(v)=e^{-\alpha |v|^{{\gamma}}}$.
Then there are some explicitable finite constants ${\varepsilon}>0, C>0$ depending only on $N$, ${\gamma}$, the function $b$, and $\alpha$, such that if $f_t$ with the initial datum $f_0\in
L^1_{1,0,1}({{{\mathbb R}^N}})\cap L^1({{{\mathbb R}^N}}, m^{-2})$ is a conservative solution to equation satisfying $$\forall\, t\in[0,\infty), \quad \left\|f_t-M\right\|_{L^1(m^{-2})}\le
{\varepsilon}$$ then $$\forall\, t\in[0,\infty), \quad \left\|f_t-M\right\|_{L^1(m^{-1})}\le
C\left\|f_0-M\right\|_{L^1(m^{-1})}e^{-{\lambda}t}.$$
\[remark4.1\]
1. The original version of Theorem \[(theo4.1)\] and Lemma \[(lem4.1)\] in [@Mcmp] were proved for the class $L^1_{\pi^{N/2}, 0,1/2}({{{\mathbb R}^N}})$, i.e. for the Maxwellian $M(v)=M_{\pi^{N/2},0,1/2}(v)=e^{-|v|^2}$. According to Proposition \[(prop1.1)\] (normalization), these are equivalent to the present version. In fact let $g_t\in L^1_{\pi^{N/2},0,1/2}({{{\mathbb R}^N}}) $, $f_{t}\in L^1_{1,0,1}({{{\mathbb R}^N}})$ have the relation $$f_t(v)=(2\pi)^{-N/2}g_{t/c}(v/\sqrt{2}),\quad{\rm i.e.}\quad
g_t(v)=(2\pi)^{N/2}f_{ct}(\sqrt{2}\,v),\quad t\ge 0$$ where $c=\pi^{N/2}2^{-{\gamma}/2}$. Then $f_t$ is a conservative solution of equation if and only if $g_t$ is a conservative solution of equation . And we have $$\forall \, t\ge 0, \quad
\left\|f_{t}-M_{1,0,1}\right\|_{L^1}=\pi^{-N/2}\left\|g_{t/c}-M_{\pi^{N/2},0,1/2}\right\|_{L^1},$$ and $$S_{b,{\gamma}}(\pi^{N/2},0,1/2)=S_{b,{\gamma}}(1,0,1)\pi^{N/2}2^{-{\gamma}/2}.$$
2. In order to prove the exponential rate of convergence to equilibrium, it was introduced in [@Mcmp] the modified linearized collision operator $${\mathcal L}_m(h)=m^{-1}
ML_M(mM^{-1}h),\quad m(v)=e^{-a|v|^s}$$ with $M(v)=e^{-|v|^2}$, $a>0$ and $0<s<2$. It is proved in [@Mcmp] that ${\mathcal L}_m$ and $L_M$ has the same spectrum, but ${\mathcal L}_m$ is available to connect the exponential moment estimates of solutions. The proof of the original version of Theorem \[(theo4.1)\] in [@Mcmp] used additional technical assumptions: the angular function $b$ is convex and nondecreasing in $(-1,1)$, and the constant $s$ in $m(v)=e^{-a|v|^s}$ satisfies $0<s<{\gamma}/2$. These assumptions were only used to prove the exponential moment estimate of the form (see Lemma 4.7 of [@Mcmp]). According to Theorem \[(theo1.0)\] in Section 1, these additional assumptions on the function $b$ can be removed and the restriction $0<s<{\gamma}/2$ can be relaxed to $0<s\le {\gamma}$. In particular one can choose $s={\gamma}$.
3. In [@Mcmp] it was actually assumed that ${\gamma}\in(0,1]$; however this assumption was only used three times:
- The first and second times are in [@Mcmp Proof of Proposition 2.3, pp.643-645]:
- first in “$|v-v_*|^{{\gamma}}\sin^{N-2}\theta/2\le
(|v-v_*|\sin\theta/2)^{{\gamma}}=|v-v'|^{{\gamma}}$ since $N-2\ge 1\ge {\gamma}$” \[a key step in obtaining basic properties of the linearized collision operator $L_M$\], but here the condition $0<{\gamma}\le 1$ can be relaxed to $0<{\gamma}\le N-2$ for any $N \ge 3$;
- second in “$\|{\mathcal I}_{{\delta}}L^{+}\|_{L^2(M)}=O({\delta}^{2-{\gamma}})\to 0$ as ${\delta}\to 0^+$”, with ${\mathcal I}_{{\delta}}={\widetilde}{\Theta}_{{\delta}}* {\bf 1}_{\{|\cdot|\le
{\delta}^{-1}\}}$ as defined in [@Mcmp p.639]. In this place, if we assume $0<{\gamma}\le 2$ and let ${\bf 1}_{\{|\cdot|\le {\delta}^{-1}\}}$ be modified as ${\bf 1}_{\{|\cdot|\le {\delta}^{-1/2}\}}$, i.e. we redefine ${\mathcal I}_{{\delta}}={\widetilde}{\Theta}_{{\delta}}* {\bf 1}_{\{|\cdot|\le
{\delta}^{-1/2}\}}$, then the same proof in [@Mcmp] also yields $\|{\mathcal I}_{{\delta}}L^{+}\|_{L^2(M)}=O({\delta}^{2-{\gamma}/2})\to 0$ as ${\delta}\to 0^+$.
- The third time was in applying a regularity result from [@MR2081030]: in the latter paper the condition $0<{\gamma}<2$ was used to ensure the existence and the uniqueness of the mild solution of equation constructed in [@MR1697562]. However since by Theorem \[(theo1.0)\], the existence and the uniqueness of the mild solution of equation have been proven for all $0<{\gamma}\le 2$, the results of [@MR2081030] mentioned above holds also for ${\gamma}=2$.
Therefore the present assumption $0<{\gamma}\le \min\{2, N-2\}$ satisfies all requirements and so the above Theorem \[(theo4.1)\] and Lemma \[(lem4.1)\] hold true. Of course in the physical case, $N=3$, there is no improvement on ${\gamma}$.
4. Let $f_t$ be a conservative mild solution of equation on $[\tau,\infty)$ with the initial datum $f_{\tau}$. Applying Theorem \[(theo4.1)\] and Lemma \[(lem4.1)\] to the solution $t\mapsto f_{t+\tau}$ on $[0,\infty)$, the exponential terms for the decay estimates of $\left\|f_t-M\right\|_{L^1}$ on the time interval $[\tau,\infty)$ is given by $e^{-{\lambda}(t-\tau)}$.
To prove Theorem \[(theo1.1)\], we need further preparation.
\[(lem4.2)\] Let $0\le f\in L^1_{k+l}({{{\mathbb R}^N}})\cap L^2_1({{{\mathbb R}^N}})\cap
H^s({{{\mathbb R}^N}})$ with $k\ge 0$, $l>0$, $s>0$. Let $${\mathcal N}(f)={\mathcal N}_{\rho,u,T}(f)\in L^1_{1,0,1}({{{\mathbb R}^N}})$$ be the normalization of $f$ defined in ,, and suppose that $|\rho-1|+|u|+|T-1|\le 1/2$. Then $$\label{(4.2)}
\left\|f-{\mathcal N}(f)\right\|_{L^1_k}\le
C_{N,k,l,s}(f) \left(
|\rho-1|+|u|+|T-1|\right)^{\frac{sl}{(1+s)(k+N+2l)}}$$ where $$C_{N,k,l,s}(f):=C_{N,k,l}\max\left\{\|f\|_{L^1_{k+l}},\,
\|f\|_{L^2_1},\, \|f\|_{H^s}\right\}$$ and $C_{N,k,l}<\infty$ only depends on $N,k,l$.
Recall that ${\mathcal N}(f)=\rho^{-1}T^{N/2}f(\sqrt{T}\, v+u)$. Let ${\mathcal N}_1(f)=T^{N/2}f(\sqrt{T}\, v+u)$. Then $$\left\|f-{\mathcal N}(f)\right\|_{L^1_k}\le\left\|f-{\mathcal N}_1(f)\right\|_{L^1_k}
+2|\rho-1| \left\|{\mathcal N}_1(f)\right\|_{L^1_k}$$ where we used $|1-\rho^{-1}|\le 2|\rho-1|$ because $1/2\le \rho\le
3/2$. We need to prove that $$\label{(lem4.1)}\left\|f-{\mathcal
N}_1(f)\right\|_{L^1_k}\le C_{N} \left\|f-{\mathcal
N}_1(f)\right\|_{L^1_{k+l}} ^{\frac{k+N}{k+N+2l}}
\left\|\widehat{f}-\widehat{{\mathcal
N}_1(f)}\right\|_{L^2}^{\frac{2l}{k+N+2l}}.$$ Let $h=f-{\mathcal N}_1(f)$, $R\in(0,\infty)$. We have $$\|h\|_{L^1_k}= \int_{{\langle}v{\rangle}<R}{\langle}v{\rangle}^k |h(v)|\, {\rm d}v+\int_{{\langle}v{\rangle}\ge
R}{\langle}v{\rangle}^k |h(v)|\, {\rm d}v\le
C_N \|h\|_{L^2}R^{\frac{k+N}{2}}+\|h\|_{L^1_{k+l}}\frac{1}{R^{l}}.$$ Minimizing the right hand side with respect to $R\in (0,\infty)$ leads to $$\|h\|_{L^1_k}\le 2 C_N ^{\frac{2 l}{k+N+2l}} \|h\|_{L^1_{k+l}}^{\frac{k+N}{k+N+2l}}
\|h\|_{L^2}^{\frac{2l }{k+N+2l}}$$ which gives by Plancherel theorem $\|h\|_{L^2}=(2\pi)^{-N/2}\|\widehat{h}\|_{L^2}$.
Since $1/2\le T\le 3/2$ and $|u|\le 1/2$ imply $$1+ \left|\frac{v-u}{\sqrt{T}}\right|^2\le 4(1+|v|^2),$$ it follows that $$\left\|{\mathcal
N}_1(f)\right\|_{L^1_{k+l}}=\int_{{{{\mathbb R}^N}}}\left(1+\left|\frac{v-u}{\sqrt{T}}\right|^2\right)^{(k+l)/2}
f(v)\, {\rm d}v \le 2^{k+l}\|f\|_{L^1_{k+l}}$$ and thus $$\label{(4.3)}
\left\|f-{\mathcal N}(f)\right\|_{L^1_k}\\
\le C_{N,k,l} \|f\|_{L^1_{k+l}} ^{\frac{k+N
}{k+N+2l}} \left\|\widehat{f}-\widehat{{\mathcal N}_1(f)}\right\|_{L^2}^{\frac{2l
}{k+N+2l}}+2^{k+1}|\rho-1|\|f\|_{L^1_k}.$$ Next we compute $$\begin{aligned}
&& \widehat{{\mathcal N}_1(f)}(\xi)
=e^{{\rm i}T^{-1/2}\xi\cdot u}\widehat{f}(T^{-1/2}\xi),\\ \\
&& \left|1-e^{{\rm i}T^{-1/2}\xi\cdot u}\right|\le
2\left(1+|T^{-1/2}\xi|^2\right)^{s/2}
\max\{|u|,\, |u|^s\},\\ \\
&& \left|\widehat{f}(\xi)-\widehat{{\mathcal N}_1(f)}(\xi)\right|\le
\left|\widehat{f}(\xi)-\widehat{f}(T^{-1/2}\xi)\right|+2\left|\widehat{f}(T^{-1/2}\xi)\right|
\left(1+|T^{-1/2}\xi|^2\right)^{s/2}\max\{|u|,\,
|u|^s\},\end{aligned}$$ hence $$\label{(4.7)}
\left\|\widehat{f}-\widehat{{\mathcal N}_1(f)}\right\|_{L^2} \le
\left\|\widehat{f}-\widehat{f}\left(T^{-1/2}\cdot\right)\right\|_{L^2}+2^{1+N/4} \|f\|_{H^s}\max\{|u|,\,
|u|^s\}.$$ Write $\xi=(\xi_1,\xi_2, \dots,\xi_N), v=(v_1,v_2,\dots,v_N)$, and $f_j(v)=v_j f(v), j=1,2,\dots,N$. Then $$\widehat{f}(\xi)-\widehat{f}\left(T^{-1/2}\xi\right)
=-{\rm
i}\int_{\frac{1}{\sqrt{T}}}^{1}\sum_{j=1}^{N}\widehat{f_j}(t\xi)
\xi_j \, {\rm d}t.$$ By Cauchy-Schwarz inequality and $$1/2\le T\le 3/2 \quad \Longrightarrow \quad
\left|\frac{1}{\sqrt{T}}-1\right|\le |T-1|$$ we have $$\left|\widehat{f}(\xi)-\widehat{f}\left(T^{-1/2}\xi\right)\right| \le
|T-1|^{1/2}\left(\int_{1\wedge \frac{1}{\sqrt{T}}}^{1\vee
\frac{1}{\sqrt{T}}}\sum_{j=1}^{N} |\widehat{f_j}(t\xi)|^2 \, {\rm
d}t\right)^{1/2}\,|\xi|$$ where $a \wedge b=\min\{a,b\}$ and $a\vee b=\max\{a,b\}$. Let $1/p+1/q=1$ with $p=(1+s)/s$ and $q=1+s$. Then $$\begin{gathered}
\left|\widehat{f}(\xi)-\widehat{f}\left(T^{-1/2}\xi\right)\right|^2=
\left|\widehat{f}(\xi)-\widehat{f}\left(T^{-1/2}\xi\right)\right|^{2/p}
\left|\widehat{f}(\xi)-\widehat{f}\left(T^{-1/2}\xi\right)\right|^{2/q}\\
\le |T-1|^{1/p}\left(\int_{1\wedge \frac{1}{\sqrt{T}}}^{1\vee \frac{1}{\sqrt{T}}}\sum_{j=1}^{N}
\left|\widehat{f_j}(t\xi)\right|^2 \, {\rm d}t\right)^{1/p}\,|\xi|^{2/p}
\left|\widehat{f}(\xi)-\widehat{f}\left(T^{-1/2}\xi\right)\right|^{2/q}.\end{gathered}$$ It follows from Hölder inequality and $q/p=s$ that $$\begin{gathered}
\left\|\widehat{f}-\widehat{f}\left(T^{-1/2}\cdot\right)\right\|_{L^2}^2\\
\le |T-1|^{1/p}
\left(\int_{1\wedge \frac{1}{\sqrt{T}}}^{1\vee \frac{1}{\sqrt{T}}}\sum_{j=1}^{N}
\left(\int_{{{{\mathbb R}^N}}}\left|\widehat{f_j}(t\xi)\right|^2 \, {\rm d}\xi\right)\, {\rm d}t
\right)^{1/p}
\left\|\widehat{f}-\widehat{f}\left(T^{-1/2}\cdot\right)\right\|_{\dot{L}^2_s}
^{2/q}\end{gathered}$$ where ${\dot{L}^2_s}$ denotes the weighted $L^2$ space with the *homogeneous* weight $|\xi|^{2s}$. Since, by Plancherel theorem, $$\begin{gathered}
\int_{1\wedge \frac{1}{\sqrt{T}}}^{1\vee \frac{1}{\sqrt{T}}}
\left(\sum_{j=1}^{N} \int_{{{{\mathbb R}^N}}}\left|\widehat{f_j}(t\xi)\right|^2
\, {\rm d}\xi\right)\, {\rm d}t \\ =
(2\pi)^N\left( \int_{1\wedge \frac{1}{\sqrt{T}}}^{1\vee \frac{1}{\sqrt{T}}}t^{-N}{\rm
d}t \right) \left( \int_{{{{\mathbb R}^N}}}|v|^2 f(v)^2\, {\rm d}v \right) \le C_N\|f\|_{L^2_1}^2\end{gathered}$$ and $$\left\|\widehat{f}-\widehat{f}\left(T^{-1/2}\cdot\right)\right\|_{\dot{L}^2_s}\le
\left\|\widehat{f}\right\|_{\dot{L}^2_s}+
\left\|\widehat{f}\left(T^{-1/2}\cdot\right)\right\|_{\dot{L}^2_s}\le
\left(1+2^{(N+2s)/4}\right)\|f\|_{H^s},$$ it follows that $$\label{(4.8)}
\left\|\widehat{f}-\widehat{f}\left(T^{-1/2}\cdot\right)\right\|_{L^2} \le C_N \|f\|_{
L^2_1}^{1/p} \|f\|_{H^s}^{1/q}|T-1|^{1/(2p)}.$$ Thus we get from , , and $1/p=s/(1+s)$ that $$\label{(4.9)}
\left\|\widehat{f}-\widehat{{\mathcal N}_1(f)}\right\|_{L^2}\le
C_N\max\left\{ \|f\|_{L^2_1},\ \|f\|_{H^s}\right\}(|u|+
|T-1|)^{\frac{s}{2(1+s)}}$$ and so follows from and .
In order to apply existing results on $L^1$ solutions to our measure solutions, we shall use the *Mehler transform*, which is defined as follows:
Let $\rho>0$, $u\in {{{\mathbb R}^N}}$ and $T>0$. The *Mehler transform* $I_n[F]$ of $F\in {\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ is given by $$I_n[F](v)=e^{Nn}\int_{{{{\mathbb R}^N}}}M_{1,0,T}\left(e^n\left(v-u-\sqrt{1-e^{-2n}}\,(v_*-u)\right)\right)
\, {\rm d}F(v_*),\quad n>0,$$ where $M_{1,0,T}(v)=(2\pi T)^{-N/2}\exp(- |v|^2/(2T))$.
The following lemma gives some basic properties of the Mehler transform that we shall use in the proof of Theorem \[(theo1.1)\].
\[(lem4.4)\] Given any $\rho>0$, $u\in{{{\mathbb R}^N}}$ and $T>0$. Let $F\in{\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ and let $M \in{\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ be the Maxwellian distribution. Then $I_n[F]\in L^1_{\rho,u,T}({{{\mathbb R}^N}}) $ and for any $0\le s\le 2$ $$\label{(4.10*)}\lim_{n\to\infty}\left\|I_n[F]-M\right\|_{L^1_s}=\|F-M\|_s.$$
Recall the basic formula of $I_n[F]$: $$\label{(4.12)}
\int_{{{{\mathbb R}^N}}}\psi(v)I_n[F](v)\, {\rm d}v=
\int_{{{{\mathbb R}^N}}}\left(\int_{{{{\mathbb R}^N}}}\psi\left(e^{-n} z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right)M_{1,0,T}(z)
\, {\rm d}z\right)\, {\rm d}F(v_*)$$ where $\psi$ is any Borel function on ${{{\mathbb R}^N}}$ satisfying $\sup\limits_{v\in{{{\mathbb R}^N}}}|\psi(v)|{\langle}v{\rangle}^{-2}<\infty$. This formula is easily proved by using Fubini theorem and change of variables. From it is easily deduced that $I_n[F]\in L^1_{\rho,u,T}({{{\mathbb R}^N}})$ for all $n>0$ and $$\label{(4.11)}
\lim_{n\to\infty} \int_{{{{\mathbb R}^N}}}{\varphi}(v)I_n[F](v) \, {\rm
d}v=\int_{{{{\mathbb R}^N}}}{\varphi}(v) \, {\rm d}F(v)$$ for all ${\varphi}\in C({{{\mathbb R}^N}})$ satisfying $\sup\limits_{v\in{{{\mathbb R}^N}}}|{\varphi}(v)|{\langle}v{\rangle}^{-2}<\infty$.
Let $0\le s\le 2$. Applying the dual version of the norm $\|\cdot\|_s$ and the convergence we have $$\label{(4.11*)}
\|F-M\|_s\le \liminf_{n\to\infty}\left\|I_n[F]-M\right\|_{L^1_s}.$$
On the other hand we shall prove that $$\label{(4.15)}
\limsup_{n\to\infty}\left\|I_n[F]-M\right\|_{L^1_s}\le \|F-M\|_s$$ also holds true, and this together with then proves .
To prove , we take $$\psi_n(v)={\langle}v{\rangle}^s{\rm sign}\left(I_n[F](v)-M(v)\right).$$ Then $$\begin{gathered}
\left\|I_n[F]-M\right\|_s=
\int_{{{{\mathbb R}^N}}}\psi_n(v)I_n[F](v)\, {\rm d}v-
\int_{{{{\mathbb R}^N}}}\psi_n(v)M(v)\, {\rm d}v
\\
=\int_{{{{\mathbb R}^N}}}\left(\int_{{{{\mathbb R}^N}}}\psi_n\left(e^{-n}
z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right)M_{1,0,T}(z) \, {\rm d}z\right)\,
{\rm d}(F-M)(v_*)
\\
+\int_{{{{\mathbb R}^N}}}\left(\int_{{{{\mathbb R}^N}}}\psi_n\left(e^{-n}
z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right)M_{1,0,T}(z) \, {\rm
d}z\right)M(v_*)\, {\rm d}v_*-\int_{{{{\mathbb R}^N}}}\psi_n(v)M(v)\, {\rm
d}v
\\
:=I_{n,1}+I_{n,2}.\end{gathered}$$ Let $h$ be the sign function of $F-M$, i.e., $h:\mathbb{R}^N\to \mathbb{R}$ is a real Borel function satisfying $${\rm d}(F-M)(v_*)=h(v_*)\, {\rm d}|F-M|(v_*) \quad \mbox{ and }\quad h(v_*)^2\equiv
1$$ (see e.g. [@MR0210528 Chapter 6]). Then $$I_{n,1}\le \int_{{{{\mathbb R}^N}}}\int_{{{{\mathbb R}^N}}}\left\langle e^{-n} z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right\rangle ^sM_{1,0,T}(z)
\, {\rm d}z\, {\rm d}|F-M|(v_*).$$ Since $$\forall\, (z,v_*)\in{{{\mathbb R}^N}}\times{{{\mathbb R}^N}}, \quad
\lim_{n\to\infty}\left\langle e^{-n} z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right\rangle
^s=\left\langle v_*\right\rangle ^s$$ and $$\left\langle e^{-n} z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right\rangle^s \le 3^s\left\langle
u\right\rangle^s\left\langle z{\rangle}^s{\langle}v_*\right\rangle^s$$ so that $$\int_{{{{\mathbb R}^N}}}\int_{{{{\mathbb R}^N}}}3^s{\langle}u{\rangle}^s{\langle}z{\rangle}^s{\langle}v_*{\rangle}^s M_{1,0,T}(z)
\, {\rm d}z\, {\rm d}|F-M|(v_*)
=3^s{\langle}u{\rangle}^s\left\|M_{1,0,T}\right\|_s\|F-M\|_s<\infty,$$ it follows from dominated convergence that $$\label{(4.13)}
\limsup_{n\to\infty}I_{n,1}\le\left\|M_{1,0,T}\right\|_{L^1}
\|F-M\|_s=\|F-M\|_s.$$
Next we prove that $\limsup_{n\to\infty}I_{n,2}\le 0$. We compute by changing variable that $$\begin{gathered}
\int_{{{{\mathbb R}^N}}}\left(\int_{{{{\mathbb R}^N}}}\psi_n\left(e^{-n}
z+u+\sqrt{1-e^{-2n}}\,(v_*-u)\right) \, M_{1,0,T}(z)
\, {\rm d}z\right)M(v_*)\, {\rm d}v_*
\\
=\frac{1}{(1-e^{-2n})^{N/2}}
\int_{{{{\mathbb R}^N}}}M_{1,0,T}(z)
\, {\rm d}z\int_{{{{\mathbb R}^N}}}\psi_n(v) \, M\left(\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}}) u}{\sqrt{1-e^{-2n}}}\right)\, {\rm d}v.\end{gathered}$$ So we get $$\begin{gathered}
I_{n,2}\\
=
\frac{1}{(1-e^{-2n})^{N/2}}
\int_{{{{\mathbb R}^N}}}M_{1,0,T}(z)
\, {\rm d}z\int_{{{{\mathbb R}^N}}}\psi_n(v)\left[M\left(\frac{v-e^{-n} z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}\right)-M(v)\right]\, {\rm d}v\\
+\left(\frac{1}{(1-e^{-2n})^{N/2}}-1\right)\int_{{{{\mathbb R}^N}}}\psi_n(v)M(v)\, {\rm d}v
\\
\le \frac{1}{(1-e^{-2n})^{N/2}}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}{\langle}v{\rangle}^s\left|M\left(\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}\right)-M(v)\right|\, M_{1,0,T}(z) \, {\rm d}v\, {\rm d}z\\
+\left(\frac{1}{(1-e^{-2n})^{N/2}}-1\right)\int_{{{{\mathbb R}^N}}}{\langle}v{\rangle}^sM(v)\, {\rm d}v\end{gathered}$$ and finally (since the last last term above clearly converges to zero) $$\begin{gathered}
\label{(4.14)}
\limsup_{n\to\infty} I_{n,2}\\
\le \limsup_{n\to\infty}
{\int\!\!\!\!\int}_{{{{\mathbb R}^N\times{\mathbb R}^N}}}{\langle}v{\rangle}^s\left|M\left(\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}\right)-M(v)\right|M_{1,0,T}(z)
\, {\rm d}v\, {\rm d}z.\end{gathered}$$ It is obvious that the integrand in the right hand side of tends to zero as $n\to\infty$. To find a dominated function for the integrand, we recall that $$\begin{gathered}
M\left(\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}\right)\\
=\frac{\rho}{(2\pi
T)^{N/2}}\exp\left(-\frac{1}{2T}\left|\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}-u\right|^2\right).\end{gathered}$$ Elementary calculation shows that $$\frac{1}{2T} \left|\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}-u\right|^2 \ge
\frac{|v-u|^2}{4T}-\frac{|z|^2}{4T}.$$ This gives $$M\left(\frac{v-e^{-n} z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}\right)M_{1,0,T}(z)
\le\frac{\sqrt{\rho}}{(2\pi T)^{N/2}}\sqrt{M(v)}\sqrt{M_{1,0,T}(z)}$$ and thus $$\begin{gathered}
{\langle}v{\rangle}^s \left|M\left(\frac{v-e^{-n}
z-(1-\sqrt{1-e^{-2n}})u}{\sqrt{1-e^{-2n}}}\right)-M(v)\right|M_{1,0,T}(z)
\\
\le \frac{\sqrt{\rho}}{(2\pi T)^{N/2}}{\langle}v{\rangle}^s\sqrt{M(v)}\sqrt{M_{1,0,T}(z)} +{\langle}v{\rangle}^sM(v)M_{1,0,T}(z).\end{gathered}$$ By dominated convergence we then conclude that the limit in the right hand side of is zero. Therefore $\limsup_{n\to\infty}I_{n,2}\le 0$ and $$\lim\sup\limits_{n\to\infty}\left\|I_n[F]-M\right\|_s\le
\lim\sup\limits_{n\to\infty}I_{n,1}+
\lim\sup\limits_{n\to\infty}I_{n,2} \le \|F-M\|_s.$$ This proves and completes the proof of the lemma.
Now we are ready to prove Theorem \[(theo1.1)\].
We first prove that the theorem holds true for all $L^1$ mild solutions in $L^1_{1,0,1}({{{\mathbb R}^N}})$. We shall then use approximation and normalization to extend it to general measure solutions.
[*Step 1*]{}. Let ${\lambda}=S_{b,{\gamma}}(1,0,1)>0$ be the spectral gap of the linearized operator $L_M$ associated with the kernel $B(z,{\sigma})$ and the Maxwellian $M(v)=(2\pi)^{-N/2} e^{-|v|^2/2}$ in $L^1_{1,0,1}({{{\mathbb R}^N}})$. Let $f_0\in L^1_{1,0,1}({{{\mathbb R}^N}})$ and let $f_t$ be the unique conservative mild $L^1$ solution of equation with the initial datum $f_0$. We shall prove that $$\label{(4.17)}
\forall\, t\ge
0, \quad \left\|f_t-M\right\|_{L^1_2}\le
C_0\left\|f_0-M\right\|_{L^1_2}^{1/2}e^{-{\lambda}t}$$ where the constant $0<C_0<\infty$ depends only on $N$, $\gamma$, and the function $b$.
To do this we use Theorem \[(theo3.1)\] to consider the positive decomposition: $$\forall \, t \ge 1, \quad f_t=g_t+h_t,$$ where $g_t=f^{n}_t\ge 0$ and $h_t=h^{n}_t\ge 0$ are given in - with $n=N_{{\gamma}}+2$ and $t_0=1$. In the following we denote by $c_i>0, C_{i}>0\,(i=1,2,\dots)$ some finite constants that depend only on $N, \gamma$, and the function $b$. By Theorem \[(theo3.1)\] with $t_0=1$ we have $$\begin{aligned}
&& \label{(4.20)}
\forall \, t\ge 1, \quad \left\|f_t-g_t\right\|_{L^1_2}=\left\|h_t\right\|_{L^1_2} \le
C_1e^{-\frac{1}{2}at},\\
&&\label{(4.21}
\sup_{t\ge 1}\left\{\left\|g_t\right\|_{L^1_{N+4}},\, \left\|g_t\right\|_{L^2_1},\,
\left\|g_t\right\|_{H^1}\right\} \le C_{2}.\end{aligned}$$ We can assume that $C_1\ge 1$. Let $\tau_0=(2/a)\log(8C_1)\,(>1)$ and let $$\begin{cases} \displaystyle
\rho_t=\int_{{{{\mathbb R}^N}}}g_t(v)\, {\rm d}v,\,\,\,
u_t=\frac{1}{\rho_t}\int_{{{{\mathbb R}^N}}}v g_t(v)\, {\rm d}v,\,\,\,
T_t=\frac{1}{N\rho_t}\int_{{{{\mathbb R}^N}}}|v-u_t|^2g_t(v) \, {\rm d}v,
\vspace{0.2cm} \\ \displaystyle
{\mathcal
N}(g_t)(v)=\frac{T_t^{N/2}}{\rho_t}g_t(\sqrt{T_t}\,v+u_t),\quad t\ge
\tau_0.
\end{cases}$$ Using the relation $$T_t=\frac{1}{N\rho_t}\int_{{{{\mathbb R}^N}}}|v|^2g_t(v)\, {\rm
d}v-\frac{|u_t|^2}{N}$$ we compute $$\label{(4.23)}
\forall \, t\ge \tau_0, \quad \left|\rho_t-1\right|+\left|u_t\right|+\left|T_t-1\right|\le
4\left\|f_t-g_t\right\|_{L^1_2}\le 4C_1e^{-\frac{1}{2}at}\le
\frac{1}{2}.$$ So by and applying Lemma \[(lem4.2)\] (with $k=2,
l=N+2, s=1$) we have $$\forall \, t\ge \tau_0, \quad \left\|g_t-{\mathcal N}(g_t)\right\|_{L^1_2}\le C_3 \left(4C_1 e^{-\frac{1}{2}a
t}\right)^{1/6}=C_4 e^{-c_1t}.$$ This together with gives $$\label{(4.24)}
\forall \, t\ge
\tau_0, \quad \left\|f_t-{\mathcal N}(g_t)\right\|_{L^1_2}\le\left\|f_t-g_t\right\|_{L^1_2}+
\left\|g_t-{\mathcal N}(g_t)\right\|_{L^1_2} \le C_5e^{-c_1t}.$$ Also by , $\sup\limits_{t\ge
1}\left\|g_t\right\|_{L^2}\le C_2$, $\tau_0>1$ and ${\mathcal N}(g_t)\in L^1_{1,0,1}({{{\mathbb R}^N}})$ we have $$\label{(4.25)}
C_6\le \inf_{t\ge \tau_0}\left\|{\mathcal
N}(g_t)\right\|_{L^2},\quad \sup_{t\ge \tau_0} \left\|{\mathcal
N}(g_t)\right\|_{L^2}\le C_7.$$ The second inequality follows from elementary calculations and the bounds $1/2\le \rho_t$ and $T_t\le 3/2$. To prove the first one, we consider some $R>0$ and write $$1=\int_{|v|<R}{\mathcal N}(g_t)(v)\, {\rm d}v+\int_{|v|\ge R}{\mathcal
N}(g_t)(v)\, {\rm d}v\le \left|{\mathbb
B}^N\right|^{1/2}R^{N/2}\left\|{\mathcal N}(g_t)\right\|_{L^2}
+R^{-2}N,$$ where $|{\mathbb B}^N|$ is the volume of the unite ball ${\mathbb
B}^N$. If we now fix $R=\sqrt{2N}$, then $$\frac{1}{2}\left|{\mathbb B}^N\right|^{-1/2}(2N)^{-N/4}\le\left\|{\mathcal
N}(g_t)\right\|_{L^2}$$ for all $t\ge \tau_0$ so that the first inequality in holds for $C_6=(1/2)|{\mathbb B}^N|^{-1/2}(2N)^{-N/4}$.
To prove we use the following technique of “moving solutions" as used in [@MR1697495] and [@Mcmp]. For any $\tau\ge \tau_0$, let $(t,v)\mapsto f_{t}^{(\tau)}(v)$ be the unique conservative solution on $[\tau,\infty)\times{{{\mathbb R}^N}}$ with the initial datum at time $t = \tau$: $$f^{(\tau)}_{t}|_{t=\tau}=f_\tau^{(\tau)}= {\mathcal N}(g_\tau).$$ On the one hand, by Theorem \[(theo4.1)\], we have $$\forall\, t\ge \tau, \quad \left\|f_{t}^{(\tau)}-M\right\|_{L^1}\le
C_{f_\tau^{(\tau)}}e^{-{\lambda}(t-\tau)}$$ where the coefficient $0<C_{f_\tau^{(\tau)}} <\infty$ depends only on $N$, $\gamma$, the function $b$, and $\|f_\tau^{(\tau)}\|_{L^2}$. Since implies $C_6\le \|f_\tau^{(\tau)}\|_{L^2}\le C_7$ for all $\tau\ge \tau_0$, it follows from Remark \[remark1.UpBd\]-(3) that $\sup\limits_{\tau\ge \tau_0}C_{f_\tau^{(\tau)}}\le C_8$, and thus for every $\tau\ge \tau_0$ we have $$\label{(4.26)}
\forall\, t\ge \tau, \quad \left\|f_{t}^{(\tau)}-M\right\|_{L^1}\le
C_8e^{-{\lambda}(t-\tau)}.$$ On the other hand using the stability estimate we have $$\label{(4.27)}
\forall \, t\ge\tau, \quad \left\|f_t-f^{(\tau)}_t\right\|_{L^1_2}\le \left\| f_{\tau}-f_\tau^{(\tau)}\right\|_{L^1_2} e^{c_2(t-\tau)}.$$ Since and $\tau\ge \tau_0$ imply $$\left\|f_{\tau}-f_\tau^{(\tau)}\right\|_{L^1_2}=\left\|f_{\tau}-{\mathcal
N}(g_\tau)\right\|_{L^1_2}\le C_5 e^{-c_1\tau},$$ it follows from and that $$\label{(4.28)}
\forall \, t\ge \tau, \quad \left\|f_t-M\right\|_{L^1}\le \left\|f_t-f_t^{(\tau)}\right\|_{L^1}+
\left\|f_t^{(\tau)}-M\right\|_{L^1}\le
C_5e^{-c_1\tau+c_2(t-\tau)}+C_8e^{-{\lambda}(t-\tau)}.$$ Now for any $$t\ge t_1:=\frac{c_1+c_2+{\lambda}}{c_2+{\lambda}}\tau_0, \quad
\mbox{we choose} \quad \tau =\tau(t)=\frac{c_2+{\lambda}}{c_1+c_2+{\lambda}} t.$$ Then $t>\tau(t)\ge \tau(t_1)=\tau_0$ and $$-c_1\tau(t)+c_2(t-\tau(t))=-\frac{c_1{\lambda}}{c_1+c_2+{\lambda}}t:=-c_3
t.$$ Thus applying with $t>\tau=\tau(t)$ (for all $t\ge t_1$) we obtain $$\label{(4.29)}
\forall\,t\ge t_1, \quad \left\|f_t-M\right\|_{L^1}\le (C_5+C_8)e^{-c_3 t}.$$ Now let $$m(v):=\exp\left(-\frac{\alpha(1)}{4}|v|^{{\gamma}}\right)$$ where $\alpha(t)>0$ is given in Theorem \[(theo1.0)\] for the initial datum $F_0\in {\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}})$ defined by ${\rm d}F_0(v)=f_0(v)\, {\rm d}v$. Then $$\label{(4-moment)}
\sup_{t\ge 1}\left\|f_t\right\|_{L^1(m^{-4})} \le 2,\quad
\left\|M\right\|_{L^1(m^{-4})}\le C_9.$$ Therefore, using Cauchy-Schwarz inequality and , we get $$\label{(4.30)}
\forall\, t\ge t_1, \quad
\left\|f_t-M\right\|_{L^1(m^{-2})}\le\left\|f_t-M\right\|_{L^1(m^{-4})}^{1/2}\left\|f_t-M\right\|_{L^1}^{1/2}
\le C_{10}e^{-c_4 t}.$$ Let ${\varepsilon}>0$ be the constant in Theorem 4.1 corresponding to $m(v)$, and let us choose $$t_2=\max\left\{t_1,\,\frac{1}{c_4}\log\left(\frac{C_{10}}{{\varepsilon}}\right)\right\}.$$ Then we deduce from that $$\forall\, t\ge t_2, \quad \left\|f_t-M\right\|_{L^1(m^{-2})}\le C_{10}e^{-c_4 t}\le {\varepsilon}.$$ It follows from Lemma \[(lem4.1)\] (see Remark \[remark4.1\]-(3)) that $$\label{(4.31)}
\forall\, t\ge t_2, \quad \left\|f_t-M\right\|_{L^1(m^{-1})}\le C_{11}\left\|f_{t_2}-M\right\|_{L^1(m^{-1})}e^{-{\lambda}(t-t_2)}.$$ Next, applying the elementary inequality $$1+|v|^2\le C e^{{\delta}|v|^{2\eta}},\qquad
\eta:=\frac{{\gamma}}{2},\quad {\delta}:=\frac{\alpha(1)}{4}$$ for some constant $C=C_{\eta,{\delta}}>0$, we have $$\label{(4.31*)}
\left\|f_t-M\right\|_{L^1_2}\le
C_{12}\left\|f_t-M\right\|_{L^1(m^{-1})}.$$ On the other hand, using the bound in , we have $$\label{(4.33)}
\left\|f_{t_2}-M\right\|_{L^1(m^{-1})}\le
\left\|f_{t_2}-M\right\|_{L^1(m^{-2})}^{1/2}
\left\|f_{t_2}-M\right\|_{L^1_2}^{1/2}\le
C_{13}\left\|f_{t_2}-M\right\|_{L^1_2}^{1/2}.$$ It follows from , and that $$\label{(4.34)}
\forall \, t\ge t_2, \quad \left\|f_t-M\right\|_{L^1_2}\le
C_{14}\left\|f_{t_2}-M\right\|_{L^1_2}^{1/2} e^{-{\lambda}t}.$$
It remains to estimate $\|f_t-M\|_{L^1_2}$ in terms of $\|f_0-M\|_{L^1_2}$ for $t\in[0,t_2]$. To do this we use the estimate in [@partI p. 3359 line 4] for the measure $H_t:=F_t-G_t$ where $F_t,
G_t$ are measure solutions of the equation . Here we define more precisely $F_t, G_t$ to be $${\rm d}F_t(v)=M(v)\, {\rm d}v \quad \mbox{ and } \quad {\rm
d}G_t(v)=f_t(v)\, {\rm d}v.$$ Then $\|H_t\|_2=\|M-f_t\|_{L^1_2}, \|f_t\|_{L^1_2}=\|M\|_{L^1_2}$, and thus (recalling $A_0=1$) $$\begin{gathered}
\forall \, t \in [r,\infty), \quad \left\|M-f_t\right\|_{L^1_2}\le \\
2\left\|(M-f_r)^{+}\right\|_{L^1_2}+
4\left(\|M\|_{L^1_{2+{\gamma}}}+\|M\|_{L^1_2}\right)
\int_{r}^{t}\left\|M-f_s\right\|_{L^1_{{\gamma}}}\, {\rm d}s.\end{gathered}$$ Since $t\mapsto f_t(v)\ge 0$ is continuous on $[0,\infty)$ for a.e. $v\in {\mathbb R}^N$, it follows from dominated convergence that $$2\left\|(M-f_r)^{+}\right\|_{L^1_2}\xrightarrow[r \to 0^+]{}
2\left\|(M-f_0)^{+}\right\|_{L^1_2}=\left\|f_0-M\right\|_{L^1_2},$$ where the last equality follows from $$|f_0-M|=f_0-M+2(M-f_0)^{+} \quad \mbox{ and } \quad
\|f_0\|_{L^1_2}=\|M\|_{L^1_2}.$$ Thus letting $r\to 0^+$ gives $$\forall \, t \in [0,\infty), \quad \left\|f_t-M\right\|_{L^1_2}\le
\left\|f_0-M\right\|_{L^1_2}+c_5\int_{0}^{t}\left\|f_s-M\right\|_{L^1_{{\gamma}}}\,
{\rm d}s.$$ Since $\|f_s-M\|_{L^1_{{\gamma}}}\le \|f_s-M\|_{L^1_2}$, it follows from Gronwall lemma that $$\label{(4.32)}
\forall \, t\ge 0, \quad \|f_t-M\|_{L^1_2}\le \|f_0-M\|_{L^1_2}
e^{c_5t}.$$
Inserting this estimate with $t=t_2$ into the right hand side of gives $$\label{(4.34*)}
\forall \, t\ge t_2, \quad \left\|f_t-M\right\|_{L^1_2}\le
C_{15}\left\|f_0-M\right\|_{L^1_2}^{1/2} e^{-{\lambda}t}.$$ Also because of and $\left\|f_{0}-M\right\|_{L^1_2}\le
2(1+N)$, we have $$\label{(4.35)}
\forall \, t \in [0,t_2], \quad \left\|f_t-M\right\|_{L^1_2}
\le C_{16}\left\|f_0-M\right\|_{L^1_2}^{1/2} e^{-{\lambda}t}.$$ Combining , we then obtain with $C_0=\max\{C_{15},\, C_{16}\}$.
[*Step 2.*]{} Let us now prove that holds also true for all measure solutions in ${\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}})$. Given any $F_0\in {\mathcal B}^{+}_{1,0,1}({{{\mathbb R}^N}})$, let $F_t$ be the unique conservative measure strong solution of equation with the initial datum $F_0$. By part (e) of Theorem \[(theo1.0)\] and Lemma \[(lem4.4)\], there is a sequence $f_{k,t}\in L^1_{1,0,1}({{{\mathbb R}^N}})$ of solutions with the initial data $f_{k,0}:=I_{n_k}[F_0]\in
L^1_{1,0,1}({{{\mathbb R}^N}})$ such that $$\begin{aligned}
\label{(4.18)}
&& \forall\,{\varphi}\in C_b({{{\mathbb R}^N}}),\quad
\forall\,t\ge 0, \quad \lim_{k\to\infty}\int_{{{{\mathbb R}^N}}}{\varphi}(v)
f_{k,t}(v)\, {\rm d}v
=\int_{{{{\mathbb R}^N}}}{\varphi}(v)\, {\rm d}F_t(v), \\
&&\label{(4.18-initial)}
\lim_{k\to\infty}\left\|f_{k,0}-M\right\|_{L^1_2}=
\left\|F_0-M\right\|_{2}.\end{aligned}$$ Using the formulation of the norm $\left\|\cdot\right\|_2$, we conclude from that $$\label{(4.18F)}
\forall \, t\ge 0, \quad \left\|F_t-M\right\|_2\le
\lim\sup_{k\to\infty}\left\|f_{k,t}-M\right\|_{L^1_2}.$$ On the other hand, applying to $f_{k,t}$, we have $$\label{(4.19)}
\forall \, t\ge
0,\ k=1,2,3,\dots, \quad \left\|f_{k,t}-M\right\|_{L^1_2}
\le C_0\left\|f_{k,0}-M\right\|_{L^1_2}^{1/2}e^{-{\lambda}t}.$$ Combining , and we obtain $$\label{(4-standard)}
\forall \, t\ge 0, \quad \left\|F_t-M\right\|_2\le
C_0\left\|F_0-M\right\|_2^{1/2}e^{-{\lambda}t}.$$
[*Step 3.*]{} Finally we show that for any $\rho>0$, $u\in{{{\mathbb R}^N}}$ and $T>0$, the theorem holds true for all measure solutions in ${\mathcal B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$. Let $F_0\in {\mathcal
B}^{+}_{\rho,u,T}({{{\mathbb R}^N}})$ and let $F_t$ be the unique conservative measure strong solution with the initial datum $F_0$. Let $M_{\rho,u,T}\in {\mathcal B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$ be the Maxwellian and let ${\mathcal N}={\mathcal N}_{\rho,u,T}$ be the normalization operator. By Proposition \[(prop1.1)\], the flow $t\mapsto {\mathcal N}(F_{t/c})$ is the unique conservative measure strong solution of equation with the initial datum ${\mathcal N}(F_0)\in{\mathcal B}_{1,0,1}^{+}({{{\mathbb R}^N}})$. Here $c=\rho
T^{{\gamma}/2}$. Since ${\mathcal N}(M_{\rho,u,T})\in {\mathcal
B}_{1,0,1}^{+}({{{\mathbb R}^N}})$ is the standard Maxwellian, it follows from the above result that (writing ${\mathcal N}(F_t)={\mathcal
N}(F_{ct/c})$) $$\forall\, t\ge 0, \quad \left\|{\mathcal N}(F_{t})-{\mathcal
N}(M_{\rho,u,T})\right\|_2 \le C_0\left\|{\mathcal
N}(F_{0})-{\mathcal N}(M_{\rho,u,T})\right\|_2^{1/2}e^{-{\lambda}ct}.$$ Then, applying Proposition \[(prop1.1)\], we have $$\begin{gathered}
\forall\, t\ge 0, \quad \left\|F_t-M_{\rho,u,T}\right\|_2\le
C_{1/\rho,|u|/\sqrt{T},1/T}\left\|{\mathcal N}(F_{t})-{\mathcal N}(M_{\rho,u,T})\right\|_2
\\
\qquad\qquad \qquad\quad \le C_0C_{1/\rho,|u|/\sqrt{T},1/T}\left\|{\mathcal N}(F_{0})-{\mathcal N}(M_{\rho,u,T})\right\|_2^{1/2}e^{-{\lambda}ct}\\
\le C_0C_{1/\rho,|u|/\sqrt{T},1/T}[C_{\rho,|u|,T}]^{1/2}\left\|F_{0}-M_{\rho,u,T}\right\|_2^{1/2}e^{-{\lambda}ct}.\qquad\,\,\end{gathered}$$ Since ${\lambda}c=S_{b,{\gamma}}(1,0,1)\rho T^{{\gamma}/2}=S_{b,{\gamma}}(\rho,u,T)$ is the spectral gap of the linearized operator $L_{M_{\rho,u,T}}$, this completes the proof of Theorem \[(theo1.1)\].
Lower Bound of Convergence Rate {#sec5}
================================
In this section we prove Theorem \[(theo1.2)\]. Recall that we assume here that ${\gamma}\in(0, 2]$ and that the function $b$ satisfies only .
We first prove the theorem for the standard case, i.e. assuming $F_0, M\in{\mathcal B}^{+}_{1,0,1}({{{\mathbb R}^N}})$. By $$\forall\, t_1,t_2\in[0,\infty), \quad
F_{t_2}=F_{t_1}+\int_{t_1}^{t_2}Q(F_{\tau},F_{\tau})\, {\rm d}\tau$$ and $Q(M,M)=0$, we have $$\label{(5.1)}
\forall \, t_1, t_2\in[0,\infty), \quad \Big|\left\|F_{t_2}-M\right\|_0 -\left\|F_{t_1}-M\right\|_0\Big|\le
\left| \int_{t_1}^{t_2}\left\|Q(F_{\tau},F_{\tau})-Q(M,M)\right\|_0
\, {\rm d}\tau\right|.$$ Using the inequalities in , $0<{\gamma}\le 2$, and the conservation of mass and energy (which implies $\left\|F_t\right\|_{{\gamma}}\le \left\|F_t\right\|_{2}=1+N$, etc.) we have $$\label{(5.2)}
\left\|Q(F_{t},F_{t})-Q(M,M)\right\|_0\le
2^{4}(1+N)\left\|F_{t}-M\right\|_{{\gamma}}.$$ Since $t\mapsto \left\|F_{t}-M\right\|_{{\gamma}}$ is bounded and, by Hölder inequality, $$\label{(5.3)}
\left\|F_{t}-M\right\|_{{\gamma}}\le
\left\|F_t-M\right\|_0^{1-{\gamma}/2}\left\|F_t-M\right\|_2^{{\gamma}/2},$$ it follows from - that $t\mapsto
\|F_t-M\|_0$ is Lipschitz continuous and $$\label{(5.4)}
\left|\frac{\, {\rm d}}{\, {\rm d}t}\left\|F_{t}-M\right\|_0\right|\le
2^{4}(N+1)\left\|F_t-M\right\|_0^{1-{\gamma}/2}\left\|F_t-M\right\|_2^{{\gamma}/2}
\quad {\rm a.e.}\quad
t\in(0,\infty).$$ Next, thanks to the exponential decay of the Maxwellian, we show that $\|F_t-M\|_2$ can be controlled by $\|F_t-M\|_0$ (see e.g. below). In fact we show that this property holds for all measure $F\in{\mathcal B}^{+}_{1,0,1}({{{\mathbb R}^N}})$. To do this, let $(M-F)^{+}$ be the positive part of $M-F$, i.e., $(M-F)^{+}={\frac}{1}{2}(|M-F|+ M-F)$. Then $|M-F|=F-M+2(M-F)^{+}$. Let $h$ be the sign function of $M-F$, i.e., $h(v)^2\equiv 1$ such that ${\rm d}(M-F)=h{\rm d}|M-F|$. Then ${\rm d}(M-F)^{+}={\frac}{1}{2}(1+h){\rm d}(M-F)$. From these we have $${\rm d}|M-F|={\rm d}(F-M)+2{\rm d}(M-F)^{+}\quad {\rm and}\quad {\rm d}(M-F)^{+}\le {\rm d}M
\label{(FM)}$$ where the inequality part is due to $F\ge 0$. Now since $F,M$ have the same mass and energy, it follows from and $|F-M|=|M-F|$ that $$\label{(5.5)}
\left\|F-M\right\|_0=2\left\|(M-F)^{+}\right\|_0,\quad
\left\|F-M\right\|_2=2\left\|(M-F)^{+}\right\|_2.$$ Let $0<{\delta}<1$. Applying Jensen inequality to the convex function $x\mapsto \exp({\delta}x/2)$ and the measure $(M-F)^{+}$ and assuming $\|(M-F)^{+}\|_0>0$ we have $$\label{(5.6)}
\frac{1}{\left\|(M-F)^{+}\right\|_0}\int_{\mathbb{R}^N}\exp({\delta}{\langle}v{\rangle}^2/2)\, {\rm d}(M-F)^{+}\ge
\exp\left(\frac{{\delta}}{2}\cdot\frac{\left\|(M-F)^{+}\right\|_2}{\left\|(M-F)^{+}\right\|_0}\right).$$ On the other hand we have $${\rm d}(M-F)^{+}(v)\le \, {\rm d}M(v)=\frac{1}{(2\pi)^{N/2}}
\exp(-|v|^2/2)\, {\rm d}v$$ and $$\label{(5.7)}
\int_{{\bf R}^N}\exp({\delta}{\langle}v{\rangle}^2/2)\, {\rm d}(M-F)^{+}(v)
\le \int_{{\bf R}^N}\exp({\delta}{\langle}v{\rangle}^2/2)\, {\rm d}M(v)=e^{{\delta}/2}\left(\frac{1}{1-{\delta}}\right)^{N/2}.$$ Let us now choose ${\delta}=\frac{1}{N+1}$. Then $$e^{{\delta}/2}\left(\frac{1}{1-{\delta}}\right)^{N/2}=e^{\frac{1}{2(N+1)}}
\left(1+1/N\right)^{N/2}<2$$ and thus, from -, we deduce $$\exp\left(\frac{1}{2(N+1)}\cdot\frac{\left\|F-M\right\|_2}{\left\|F-M\right\|_0}\right)\le
\frac{4}{\left\|F-M\right\|_0},$$ i.e. $$\label{(5.8)}
\left\|F-M\right\|_2\le 2(N+1)\left\|F-M\right\|_0\log
\left(\frac{4}{\left\|F-M \right\|_0}\right).$$ If we adopt the convention $x\log (1/x)=0$ for $x=0$, the inequality also holds for $\|F-M\|_0=0$.
Now let us go back to the solution $F_t$. To avoid discussing the case $\|F_t-M\|_0=0$ for some $t$, we consider $$U_{{\varepsilon}}(t):=\frac{\|F_t-M\|_0+{\varepsilon}}{4},\quad 0<{\varepsilon}<1.$$ Then $$\frac{\|F_t-M\|_0}{4}<U_{{\varepsilon}}(t)\le \frac{2+{\varepsilon}}{4}<\frac{2}{3}, \quad
0<{\varepsilon}<\frac{2}{3}.$$ Using the inequality $$\forall \, 0\le x\le y\le \frac{2}{3}, \quad x\log \frac{1}{x} \le 2
y\log \frac{1}{y}$$ and (with $F=F_t$), we then obtain $$\frac{\left\|F_t-M\right\|_2}{4}
\le 4(N+1)U_{{\varepsilon}}(t)\log \left(\frac{1}{U_{{\varepsilon}}(t)}
\right).$$ Thus by we deduce $$\label{(5.9)}
\left|\frac{\, {\rm d}}{\, {\rm d}t}U_{{\varepsilon}}(t)\right|\le
A U_{{\varepsilon}}(t)\left [\log \left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right ]^{{\gamma}/2}
\quad {\rm a.e.}\quad
t\in(0,\infty)$$ where $A=2^{6}(N+1)^2$.
[*Case 1: $0<{\gamma}<2$.*]{} In this case we have, by , $$\frac{\, {\rm d}}{\, {\rm d}t}
\left[\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right]^{1-{\gamma}/2}
=-(1-{\gamma}/2)\left[\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right]^{-{\gamma}/2}\frac{1}{U_{{\varepsilon}}(t)}\cdot
\frac{\, {\rm d}}{\, {\rm d}t}U_{{\varepsilon}}(t) \le (1-{\gamma}/2)A$$ for almost every $t\in(0,\infty)$. Observe that the function $$t\mapsto \left[\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right]^{1-{\gamma}/2}$$ is absolutely continuous on every bounded interval of $[0,\infty)$. It follows that $$\label{(5.10)}
\forall\, t\ge 0, \quad
\left[\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right]^{1-{\gamma}/2}
\le
\left[\log\left(\frac{1}{U_{{\varepsilon}}(0)}\right)\right]^{1-{\gamma}/2}+(1-{\gamma}/2)A
t.$$ Next, using the convexity inequality $$\forall \, x,y\ge 0, \quad \left(x+\left(1-\frac{\gamma}{2}
\right)y\right)^{\frac{1}{1-{\gamma}/2}} \le \frac{\gamma}{2} \left(
\frac{2 x}{{\gamma}}
\right)^{\frac{1}{1-{\gamma}/2}}+\left(1-\frac{{\gamma}}{2}\right)
y^{\frac{1}{1-{\gamma}/2}},$$ we have $$\left\{\left[\log\left(\frac{1}{U_{{\varepsilon}}(0)}\right)\right]^{1-{\gamma}/2}+\left(1-\frac{{\gamma}}{2}\right)A
t \right\}^{\frac{1}{1-{\gamma}/2}}\le \alpha
\log\left(\frac{1}{U_{{\varepsilon}}(0)}\right) +\beta_1 t^{\frac{2}{2-{\gamma}}}$$ where $$\alpha=\left(\frac{2}{{\gamma}}\right)^{\frac{{\gamma}}{2-{\gamma}}},\quad
\beta_1=\left(1-\frac{{\gamma}}{2}\right)
A^{\frac{2}{2-{\gamma}}}=\left(1-\frac{{\gamma}}{2}\right)
\left(2^{6}(N+1)^2\right)^{\frac{2}{2-{\gamma}}}.$$ Thus, from , we obtain $$\forall\, t\ge 0, \quad U_{{\varepsilon}}(t)\ge
U_{{\varepsilon}}(0)^{\alpha} \exp\left(-\beta_1 t^{\frac{2}{2-{\gamma}}}\right).$$ Using the definition of $U_{{\varepsilon}}(t)$ and letting ${\varepsilon}\to 0+$, we get finally $$\forall\, t\ge 0, \quad \frac{\|F_t-M\|_0}{4}\ge \left
(\frac{\|F_0-M\|_0}{4}\right)^{\alpha} \exp\left(-\beta_1
t^{\frac{2}{2-{\gamma}}}\right).$$ This concludes the proof of the standard case for $0<{\gamma}<2$.
[*Case 2: ${\gamma}=2$.*]{} In this case we have by with ${\gamma}=2$ that $$\frac{\, {\rm d}}{\, {\rm d}t}
\log\left(\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right)=-
\left[\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right]^{-1}\frac{1}{U_{{\varepsilon}}(t)}\cdot
\frac{\, {\rm d}}{\, {\rm d}t}U_{{\varepsilon}}(t)\le A,\quad {\rm a.e.}\quad
t\in(0,\infty).$$ Since the function $$t \mapsto \log\left(\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right)$$ is absolutely continuous on every bounded interval of $[0,\infty)$, it follows that for all $t>0$ $$\log\left(\log\left(\frac{1}{U_{{\varepsilon}}(t)}\right)\right)
\le \log\left(\log\left(\frac{1}{U_{{\varepsilon}}(0)}\right)\right)+A t,
\qquad {\rm i.e.}\quad
U_{{\varepsilon}}(t)
\ge (U_{{\varepsilon}}(0))^{e^{At}}.$$ Letting ${\varepsilon}\to 0+$ leads to $$\forall\, t\ge 0, \quad \frac{\|F_t-M\|_0}{4}\ge \left
(\frac{\|F_0-M\|_0}{4}\right)^{e^{At}}.$$ This prove the standard case for ${\gamma}=2$.
[*General non-normalized setting.*]{} The general case can be reduced to the standard case by using normalization. Let $F_0\in
{\mathcal B}_{\rho,u,T}^+({{{\mathbb R}^N}})$, $c=\rho T^{{\gamma}/2}$ and let $M\in
{\mathcal B}_{\rho,u,T}^+({{{\mathbb R}^N}})$ be the Maxwellian. Then, according to Proposition \[(prop1.1)\], the normalization $t\mapsto {\mathcal
N}(F_{t/c})\in {\mathcal B}^{+}_{1,0,1}({{{\mathbb R}^N}})$ is a conservative measure strong solution of equation with the initial datum ${\mathcal N}(F_0)$. Applying the above estimates and $\|F_{t}-M\|_0=\rho\|{\mathcal N}(F_{t})-{\mathcal N}(M)\|_0$ we obtain that if $0<{\gamma}<2$ then $$\begin{gathered}
\forall\, t\ge 0, \quad \|F_t-M\|_0 =\rho\|{\mathcal N}(F_{c^{-1}ct})-{\mathcal N}(M)\|_0
\\
\ge 4\rho\left(\frac{\|{\mathcal N}(F_0)-{\mathcal N}(M)\|_0}{4}\right)^{\alpha}
\exp\left(-\beta_1(ct)^{\frac{2}{2-{\gamma}}}\right)\\
=4\rho\left(\frac{\|F_0-M\|_0}{4\rho}\right)^{\alpha}\exp\left(-\beta\,t^{\frac{2}{2-{\gamma}}}
\right)\qquad \qquad \qquad \qquad \quad \qquad \qquad\end{gathered}$$ with $$\beta=\beta_1c^{\frac{2}{2-{\gamma}}}=
(1-\frac{{\gamma}}{2}) \left(2^{6}(1+N)^2\rho
T^{{\gamma}/2}\right)^{\frac{2}{2-{\gamma}}}.$$
Similarly if ${\gamma}=2$, then $$\|F_t-M\|_0\ge
4\rho\left (\frac{\|{\mathcal N}(F_{0})-{\mathcal N}(M)\|_0}{4}\right)^{e^{Act}}
= 4\rho\left (\frac{\|F_{0}-M\|_0}{4\rho}\right)^{e^{\kappa\,t}}$$ with $\kappa=Ac=2^{6}(N+1)^2\rho T$. This completes the proof.
Global in Time Stability Estimate {#sec6}
=================================
In the last section we prove the the global in time strong stability of the measure strong solutions of equation .
Let $F_t$ be a conservative measure strong solution of equation with the initial datum $F_0\in {\mathcal
B}_{\rho_0,u_0,T_0}^{+}({{{\mathbb R}^N}})$, and let $G_t$ be any conservative measure strong solution of equation with the initial datum $G_0$. Let $$\label{(6.1)}
D_0:=\min\left\{\frac{\rho_0}{2},\,
\left(
\frac{4\left\|F_0\right\|_2}{N\rho_0^2}
+
\frac{6}{N}\left(\frac{\left\|F_0\right\|_2}{\rho_0^2}\right)^2\right)^{-1}\frac{T_0}{2}
\right\}.$$ If $\left\|F_0-G_0\right\|_2\ge D_0$, then by conservation of mass and energy we have for all $t\ge 0$, $$\label{(6.2)}
\left\|F_t-G_t\right\|_2\le
\left\|F_0\right\|_2+\left\|G_0\right\|_2
\le 2\left\|F_0\right\|_2+\left\|G_0-F_0\right\|_2\le
\left(\frac{2\left\|F_0\right\|_2}{D_0}+1\right)\left\|G_0-F_0\right\|_2.$$
In the following we assume that $\left\|F_0-G_0\right\|_2< D_0$. By the uniqueness theorem, we can also assume that $\left\|F_0-G_0\right\|_2>0$. Due to our choice of $D_0$, we see that $G_0$ is non-zero and is not a Dirac distribution. Therefore let $\rho>0$, $u\in{{{\mathbb R}^N}}$, $T>0$ be the mass, mean velocity and temperature corresponding to $G_0$, i.e., $G_0\in {\mathcal
B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$. Using the condition $\left\|F_0-G_0\right\|_2< D_0$ and elementary estimates we have $$\begin{cases} \displaystyle
|\rho-\rho_0|\le \left\|G_0-F_0\right\|_2, \quad
0<\frac{\rho_0}{2}\le \rho\le \frac{3\rho_0}{2}, \vspace{0.2cm} \\ \displaystyle
|u-u_0|\le
\frac{2\left\|F_0\right\|_2}{\rho_0^2}\left\|G_0-F_0\right\|_2,
\vspace{0.2cm} \\ \displaystyle
|T-T_0|\le\left( \frac{4\left\|F_0\right\|_2}{N\rho_0^2} +
\frac{6}{N}\left(\frac{\left\|F_0\right\|_2}{\rho_0^2}\right)^2\right)\left\|G_0-F_0\right\|_2,
\vspace{0.2cm} \\ \displaystyle
0<\frac{T_0}{2}\le T\le \frac{3T_0}{2}.
\end{cases}$$
Let $M_{F_0}, M_{G_0}$ be the Maxwellians associated with $F_0, G_0$ respectively, i.e. $M_{F_0}\in{\mathcal
B}_{\rho_0,u_0,T_0}^{+}({{{\mathbb R}^N}})$, $M_{G_0}\in {\mathcal
B}_{\rho,u,T}^{+}({{{\mathbb R}^N}})$. In the following calculations the constants $0<C_i<\infty$ $(i=1,2,\dots,9)$ only depend on $N$, the function $b$, ${\gamma}$, $\rho_0$, $u_0$ and $T_0$, and we recall that $$\left\|F_0\right\|_2=\rho_0(1+NT_0+|u_0|^2).$$
We need to estimate $\left\|M_{G_0}-M_{F_0}\right\|_2$. Let us define $${\mathcal M}(\rho,u,T; v)=(2\pi)^{-N/2}\rho T^{-N/2}\exp\left(-\frac{|v-u|^2}{2T}\right)$$ and let us compute $$\begin{cases}\displaystyle
\frac{{\partial}}{{\partial}\rho}{\mathcal M}(\rho,u,T; v)
={\mathcal M}(1,u,T; v), \vspace{0.2cm} \vspace{0.2cm} \\
\displaystyle
\nabla_{u}{\mathcal M}(\rho,u,T; v)
={\mathcal M}(\rho,u,T; v) \, \frac{v-u}{T},
\vspace{0.2cm} \\ \displaystyle
\frac{{\partial}}{{\partial}T}{\mathcal M}(\rho,u,T; v)
=\left(-\frac{(N/2+1)}{T}+\frac{|v-u|^2}{2T^2}\right)
{\mathcal M}(\rho,u,T; v).
\end{cases}$$ If we set $$\forall \, \theta \in [0,1], \quad \rho(\theta)=\theta\rho+(1-\theta)\rho_0,
\,\,\,u(\theta)=\theta u+(1-\theta)u_0,\,\,\,
T(\theta)=\theta T+(1-\theta)T_0,$$ then $$\begin{gathered}
\left|{\mathcal M}(\rho,u,T; v)-{\mathcal M}(\rho_0,v_0,T_0; v)
\right|
\le |\rho-\rho_0|
\int_{0}^{1}{\mathcal M}(1,u(\theta),T(\theta); v)\, {\rm d}\theta\\
+|u-u_0|\int_{0}^{1}{\mathcal M}(\rho(\theta),u(\theta),T(\theta); v)
\frac{|v-u(\theta)|}
{T(\theta)}
\, {\rm d}\theta \\
+|T-T_0|\int_{0}^{1}{\mathcal M}(\rho(\theta),u(\theta),T(\theta); v)
\left(\frac{(N/2+1)}{T(\theta)}+\frac{|v-u(\theta)|^2}{2T(\theta)^2}\right)
\, {\rm d}\theta.\end{gathered}$$ We then deduce $$\begin{gathered}
\left\|M_{G_0}-M_{F_0}\right\|_2=\int_{{{{\mathbb R}^N}}}{\langle}v{\rangle}^2
\left|{\mathcal M}(\rho,u,T; v)-{\mathcal M}(\rho_0,u_0,T_0; v)
\right|\, {\rm d}v\\
\le C_1
\left(|\rho-\rho_0|+|u-u_0|+|T-T_0|
\right)\end{gathered}$$ and thus using the above estimates for $\rho-\rho_0$, $u-u_0$ and $T-T_0$, we obtain $$\label{(6.3)}
\left\|M_{G_0}-M_{F_0}\right\|_2\le C_2\left\|G_0-F_0\right\|_2.$$
Next from the above estimates we have $${\lambda}_0=S_{b,{\gamma}}(1,0,1) \rho_0 T_0^{{\gamma}/2},\quad {\lambda}=S_{b,{\gamma}}(1,0,1)
\rho T^{{\gamma}/2} \ge 2^{-1-{\gamma}/2}{\lambda}_0.$$ Then using the convergence estimate and recalling that $$C_{1/\rho,|u|/\sqrt{T},1/T}=\rho\max\left\{ 1+|u|^2+\sqrt{T}|u|,\,
T+\sqrt{T}|u|\right\},$$ we have $$\begin{cases} \displaystyle
\left\|F_t-M_{F_0}\right\|_2\le C_3e^{-{\lambda}_0t}, \vspace{0.2cm} \\ \displaystyle
\left\|G_t-M_{G_0}\right\|_2\le C_0C_{1/\rho,|u|/\sqrt{T},1/T}e^{-{\lambda}t}
\le C_{4} \exp\left(- 2^{-1-{\gamma}/2}{\lambda}_0 t\right).
\end{cases}$$ Thus $$\forall\, t\ge 0, \quad
\left\|F_t-M_{F_0}\right\|_2+\left\|G_t-M_{G_0}\right\|_2 \le
C_5e^{-C_6 t},$$ and it follows from that $$\label{(6.4)}
\forall\, t\ge 0, \quad \left\|F_t-G_t\right\|_2\le C_5e^{-C_6 t}+C_2\left\|F_0-G_0\right\|_2.$$ On the other hand by the stability estimate we have $$\label{(6.5)}
\forall\,t\ge 0, \quad \left\|F_t-G_t\right\|_2\le
\Psi_{F_0}\left(\left\|F_0-G_0\right\|_2\right)e^{C_{7}(1+t)}.$$
The remaining of the proof is concerning with balancing properly and .
[*Case 1: $\Psi_{F_0}(\left\|F_0-G_0\right\|_2)<1$.*]{} Note that $\left\|F_0-G_0\right\|_2>0$ implies $\Psi_{F_0}(\left\|F_0-G_0\right\|_2>0$. Let $$t_0=
\log\left\{\left(\frac{1}{\Psi_{F_0}(\left\|F_0-G_0\right\|_2)}\right)^{\frac{1}{C_6+C_7}}\right\}.$$ For every $t\ge 0$, if $t\le t_0$, then, using , $$\left\|F_t-G_t\right\|_2\le \Psi_{F_0}\left(\left\|F_0-G_0\right\|_2\right)e^{C_{7}(1+t_0)}
=e^{C_{7}}\left[\Psi_{F_0}\left(\left\|F_0-G_0\right\|_2\right)\right]^{\frac{C_{6}}{C_6+C_{7}}}.$$ If $t\ge t_0$, then, using , $$\left\|F_t-G_t\right\|_2\le C_5\Big[\Psi_{F_0}(\left\|F_0-G_0\right\|_2)\Big]^{\frac{C_6}{C_6+C_7}}
+C_2\left\|F_0-G_0\right\|_2.$$ Thus $$\label{(6.6)}
\forall\, t\ge 0, \quad \left\|F_t-G_t\right\|_2\le C_{8}
\Big[\Psi_{F_0}(\left\|F_0-G_0\right\|_2)\Big]^{\frac{C_6}{C_6+C_{7}}}
+C_2\left\|F_0-G_0\right\|_2.$$
[*Case 2: $\Psi_{F_0}(\left\|F_0-G_0\right\|_2)\ge 1$.*]{} In this case we have, by conservation of mass and energy and $\left\|F_0-G_0\right\|_2\le \rho_0/2\le \left\|F_0\right\|_2/2$ that $$\label{(6.7)}
\forall\, t\ge 0, \quad \left\|F_t-G_t\right\|_2\le
\frac{5}{2}\left\|F_0\right\|_2\le
\frac{5}{2}\left\|F_0\right\|_2\Big[\Psi_{F_0}(\left\|F_0-G_0\right\|_2)\Big]^{\frac{C_6}{C_6+C_{7}}}.$$
Combining , , and , we obtain $$\forall\, t\ge 0, \quad \left\|F_t-G_t\right\|_2\le C_{9}\left(
\left[\Psi_{F_0}\left(\left\|F_0-G_0\right\|_2\right)\right]^{\frac{C_6}{C_6+C_{7}}}
+\left\|F_0-G_0\right\|_2 \right).$$ This proves Theorem \[(theo1.3)\]
*Acknowledgements.* This work was started while the first author was visiting the University Paris-Dauphine as an invited professor during the autumn 2006, and the support of this university is acknowledged. The first author also acknowledges support of National Natural Science Foundation of China, Grant No.10571101, No.11171173. The second author’s research was supported by the ERC Starting Grant MATKIT.
[© 2015 by the authors. This paper may be reproduced, in its entirety, for non-commercial purposes.]{}
[^1]: As in our previous work [@partI], the “measure-valued solutions" will be also called “measure solutions”.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A unified viewpoint is presented in margin to the [“Work, dissipation and fluctuations in nonequilibrium physics”]{} Bruxelles 22-25 March, 2006, where the topics were discussed by various authors and it became clear the need that the very different viewpoints be consistently presented by their proponents'
author:
- Giovanni Gallavotti
title: 'Fluctuation relation, fluctuation theorem, thermostats and entropy creation in non equilibrium statistical Physics'
---
=1=1
The [*fluctuation relation*]{} is a general symmetry property of mechanical systems which should hold under the only assumption that the system motions are [*chaotic*]{}: it reflects the [*time reversal symmetry*]{}.
Time reversal symmetry means a smooth isometry $I$ of phase space which anticommutes with the time evolution map $S_t$: $IS_t=S_{-t}I$. Therefore the familiar operation of time reversal $T$ or, even better, $TCP$ would be always valid in fundamental models, [@Ga98], and therefore the time reversal symmetry can be an issue only if one deals with phenomenological models in which dissipation is empirically introduced: as in the case of Navier-Stokes equations for fluids, for instance.
The discovery, published in the paper [@ECM93], of the fluctuation relation has led to renewed efforts towards the formulation of a theory of nonequilibrium stationary states. In the paper a link is attempted with the earlier proposal [@Ru80] for the description of the probability distribution for chaotic stationary states in fluids. Although the paper was a real breakthrough, the original argument needed to be made precise. A direct connection with [@Ru80] was established and called “[*fluctuation theorem*]{}” in [@GC95]: where it was shown how the paradigm of chaotic evolution constituted by the hyperbolic (also called “Anosov”) systems allowed for a precise formulation of sufficient conditions under which the fluctuation relation held.
The latter proof has been considered interesting because in a sense the hyperbolic evolutions perform for chaotic systems the role plaid by the harmonic oscillators for ordered systems. From these works emerged the interest for Physics of two new fundamental concepts, the [*chaotic hypothesis*]{} and a general mechanical notion for the [*entropy creation*]{}.
The first is an extension of the proposals, [@Ru80], that identified the probability distributions forming the ensembles suitable to give the statistical properties of states of turbulent fluids with the special class of distributions, well known and studied in the theory of dynamical systems, called SRB distributions.
The extension can be formulated, [@Ru80; @GC95], in the form a hypothesis:
: [*Motion on the attracting set of a chaotic system can be regarded as “hyperbolic”.*]{}
This hypothesis is very ambitious as it should be seen as an extension of the [*ergodic hypothesis*]{}, which it implies if applied to a system which is isolated and subject only to conservative forces.
It allows us to state existence of time averages of mechanical observables $x\to F(x)$, identified with functions of the phase space point $x$ representing the microscopic state of the system, the existence of the limits $\lim_{T\to\io} \fra1T \sum_{j=0}^{T-1}
F(S^jx)$ for all initial data $x$ chosen in the vicinity of the attracting set, setting aside a set of $0$ phase space volume. It also implies that, the limit is [*independent*]{} of $x$ (apart from the zero volume possible exceptions) and therefore define a [*statistics*]{}, a probability distribution $\m$ such that
$$\lim_{T\to\io} \fra1T \sum_{j=0}^{T-1} F(S^jx)=\ig\m(dy) \,F(y)\Eq(e1.1)$$ In other words for chaotic systems motions have a well defined statistics, a probability distribution that allows us to define (in principle) the time averages of the observables.
The nontriviality of the above statements becomes perhaps more clear in the case of systems subject to steady dissipation. Their stationary states cannot be described by statistics which are not [*singular*]{}: which attribute probability $1$ to sets of $0$ phase space volume.
This is a seemingly odd situation: we are interested in data randomly chosen with a probability distribution with density on phase space and, yet, they evolve with a statistics which is singular. Such a situation, however, appears to be quite clearly correct as soon as simulations are attempted in [*virtually any*]{} system which exhibits chaos (positive Lyapunov exponents) and is subject to some kind of dissipation: hence it is simply accounted for by the chaotic hypothesis, see for instance [@BGG97].
However technically the hypothesis is far more rich of implications. In fact the distribution $\m$ is, in hyperbolic systems, identified with the SRB distribution, which is not well known but it should be viewed as a generalization of the microcanonical distribution for isolated systems. In a sense that can be made very precise it is the distribution that gives equal probability to cells into which the phase space can be imagined discretized, [@Ga95a; @Ga95c; @Ga99]. It allows to give a mathematical formula for the averages in (e1.1) and even to give a precise definition of [*coarse grained cells*]{} in phase space, [@Ga95a; @Ga01; @Ga06b].
Without entering here in more technical details this means that we have, from the theory of the SRB distributions for hyperbolic systems, expressions for the averages in terms of mechanical quantities. The latter can therefore be used to derive general relations between observables averages [*even though, as it is virtually always the case, we cannot hope to compute their actual vlaues*]{}.
Just as in equilibrium where we can write the averages as integrals with respect to the Liouville distribution on the energy surface (or the canonical one) but we can hardly compute them: nevertheless we can establish, by using the formal expressions, general relations which turn out to be extremely interesting precisely because of their generality. A celebrated example is Boltzmann’s [*heat theorem*]{}, the second law of equilibrium thermodynamics, as a consequence of the assumption that the statistics of motion (of an isolated system subject to conservative forces) is the microcanonical distribution.
The chaotic hypothesis has a rather general consequence which should be seen as a generalization, at any distance from equilibrium, of Onsager-Machlup fluctuations theory near equilibrium, [@OM53a; @OM53b; @Ga96a].
Consider a mechanical system of particles described by a generic equation of motion for the representative point $x$ in phase space
$$\dot x= f(x)\Eq(e1.2)$$ where $x$ denotes the position and velocity components. The equation will be a model describing a finite system on which external non conservative forces act.
Therefore the equation will be non Hamiltonian and phase space volume will not be conserved. As a consequence the [*divergence*]{}
$$\s(x)=-\sum_i\dpr_{x_i} f_i(x)\Eq(e1.3)$$ [*does not vanish*]{}. However in presence of time reversal symmetry it will be odd under time reversal $\s(x)=-\s(Ix)$. The system will be called [*dissipative*]{} if even the time average $\s_+=\lim_{T\to\io}
\fra1T\ig_0^T \s(S_tx)dt=\ig \m(dy)\s(y)$ of $\s$ is [*positive*]{}.
In a dissipative system, $\s_+>0$, in the stationary state described by the statistics $\m$, consider the probability that $f_j(S_t
x)\sim\f(t)$ for $t\in [-\fra12\tau,\fra12\tau]$ where $t\to\f(t)$ is a prescribed [*pattern*]{}. The symbol $\sim$ means that $|f(S_tx)-\f(t)|<\e$ for some very small $\e$ (see [@Ga97; @Ga00; @Ga02] for a quantitative form of the notion of “very small”). Define the [*dimensionless phase space contraction*]{} $\s(x)$ as the divergence of the equations of motion changed in sign. Suppose that the average (with respect to the statistics $\m$, the integral of $\s(x)$ with respect to $\m$, see (e1.1)) phase space contraction $\s_+$ is positive $\s_+>0$ and define $p=\fra1T\ig_{-\fra\t2}^{\fra\t2} \fra{\s(S_t x)}{\s_+}dt$. Suppose that $f_1(x),\ldots,f_n(x)$ are $n$ observables with defined parity under time reversal, suppose odd for definiteness: $f_i(Ix)=-
f_i(x)$. Then
\(1) In particular this holds for the single observable $\s(x)/\s_+$, [@GC95; @Ge98], and [*it is a theorem for hyperbolic systems*]{}. This is the form in which the fluctuation relation was discovered, [@ECM93], and the above is an [*extension*]{} of it under the same assumptions, [@Ga97; @Ga00].\
(2) The extension was found first in the special case $f_1=\s(x)/\s_+,
f_2=\dpr_E \s(x)$ where it has been shown to imply the Onsager reciprocity and Green-Kubo formulae, [@Ga96a].\
(3) Another particular case is obtained by considering the probability that the averages $\lis f_1,\ldots, \lis f_n$, over the time interval $(-\fra\t2,\fra\t2)$ of the considered observables, have a given value $a_1,\ldots, a_n$ with $p$ determined by $a_1,\ldots, a_n$. Then for $|p|<p^*$
$$\fra{P( \lis f_1\sim a_1,\ldots, \lis f_n\sim a_n)}{ P( \lis f_1\sim
-a_1,\ldots, \lis f_n\sim -a_n)}=e^{p\s_+ \t+O(1)}\Eq(e1.5)$$ which is a very surprising relation because of the arbitrariness of the observables $f_j$ [*which do not appear in the r.h.s.*]{} except through their function $p$. The above relation appeared recently in the context of Kraichnan’s theory of passive scalars in a case in which $p=\sum_i a_i$, [@CDG06].\
(4) A mahematically precise form of the theorem, [@GC95; @Ga95b], is to say that for $|p|<p^*$ the probability that $p\in\D$ has the form $\exp(\t\max_{p\in\D}\z(p)+O(1))$ and the function $\z(p)$, which for hyperbolic systems is known to be analytic and convex in a natural interval of definition $(-p^*,p^*)$ (and $-\io$ outside it) satisfies, for $|p|<p^*$ the symmetry property
$$\z(-p)=\z(p)-p\s_+\Eq(e1.6)$$\
(5) A further consequence is that the stationary average of $\exp
\ig_0^\t \s(S_tx)dt$ satisfies
$$\media{e^{\ig_0^\t \s(S_tx)dt}}\sim1\Eq(e1.7)$$ where $\sim1$ means that the quantity is [*bounded*]{} as $\t\to\io$ ([*Bonetto*]{}’s formula, [@Ga00 Eq.(9.10.4)]): see below for a hint to possible applications and for its similarity to Jarzynski’s formula, [@Ja97; @Ja99].\
(6) The chaotic hypothesis and the fluctuation relation have been tested quite extensively, starting with [@BGG97], and it has almost become a test of the correctness of the computer programs simulations of chaotic systems rather than a formula to be tested. The situation is quite different with experiments where a lot of difficulties arise, on a case by case basis, in setting up experiments and interpreting them. Nevertheless there have been several attempts and it can be hoped that more will come, [@CV03a; @FM04; @VCC04; @GC05; @CHGLPR03; @BCG06].\
(7) The formulae above hold for time evolutions described by maps iterations as well as for those described by differential equations. It is worth stressing that most works, in particular [@GC95], deal with discrete time evolutions. The case of continuous time is considered less frequently, and for the first time it has been formulated as a theorem in [@Ge98]. The case of maps is possibly closer to applications as observations are usually done when some timing events occurr and evolution appears as a map between timing events. However a close examination of the relations between the continuous and discrete cases reveals a number of delicate properties (particularly in the case in which singular forces may be acting, like Lennard-Jones type of repulsive cores) which if neglected may lead to errors, as exemplified in [@BGGZ05].
=2=1
The result in the previous section hints at another major point of the research in the last 20 years of the ’900’s: in the ’980’s a concrete model of a thermostat became necessary to perform simulations of molecular dynamics in systems out of equilibrium.
The “Nosé–Hoover”, the “isokinetic” or the “isoenergetic” thermostats are prominent examples that were put forward and employed to study a large number of problems: see [@EM90] for a thorough discussion of the related problems. One of the results was the discovery of the flutuation relation.
Another important byproduct was the identification of the (not yet defined at the time) [*entropy creation rate*]{} with the phase space contraction. Although the original authors quite clearly attributed to the words they employed a meaning close to the physical one suggested by the given names the general attitude was, it seems, to regard the thermostat models as unphysical and, consequently, to attribute little value to the concept of entropy creation as related in some way to the thermodynamic entropy.
The above (extension) of the fluctuation theorem and relation, (e1.4), suggests that one should give a [*fundamental physical sense to the phase space contraction*]{}, at least in finite, time reversible systems. Since as said above ultimately time reversal (or the equivalent, for our purposes, TCP symmetry) is a law of nature all models should either display the symmetry or be equivalent to symmetric models.
In my view this identification between phase space contraction and entropy creation rate is an important new development, [@EM90; @Ru97; @Ru99], that is still not fully appreciated as it should.
It has to be stressed that, although since more than a century we are familiar with the entropy of equilibria and its mechanical interpretation, no mechanical definition of entropy creation rate in a process out of equilibrium has been proposed (or, better, accepted). The above extended fluctuation relation is clearly saying that the independence of the ratio in (e1.5) means that the conditional probability that a pattern occurs in presence of an average (dimensionless) phase space contraction $p$ is the same as that of the reverse pattern in presence of the opposite average phase space contraction ($-p$).
In other words if the entropy creation rate is reversed during a time interval then evolution of the [*other*]{} observables “proceeds backwards” with the same likelyhood it had to “proceed forward” when the average entropy production was $p$. [*All that has to be done to reverse the time arrow is the reverse the entropy creation rate*]{}.
All this leads to say that the identification of entropy creation rate and phase space contraction has to be taken seriously. Its identity with what one would naturally call entropy creation in the many particular cases studied in the works summarized in [@EM90] and continuing since should not be considered a curious coincidence but as a new insight into the foundations of nonequilibrium statistical mechanics.
Note that we are saying that in nonequilibrium entropy creation is defined and identified with mechanical quantities: but entropy itself is not defined (yet?), not even in stationary states. It might even be not needed and not unambiguous, [@Ga00; @Ga01].
To illustrate further the new identification of entropy creation rate with the mechanical quantity expressed by the divergence of the equations of motion, (e1.3), and its physical interest I discuss a class of examples which, it seems to me, fully justify the mentioned identification, [@Ga96; @Ga06].
Consider a mechanical system $\CC_0$ in contact with the mechanical systems $\CC_i$, $i=1,2,\ldots$. Microscopically the state is is described by the positions ${\bf X}_0,{\bf X}_1,\ldots,{\bf X}_n$ of the $N_1,N_1,\ldots,N_n$ constituent particles. The systems interact via short range pair forces with potential energies $U_a({\bf X}_i)$, $U_a({\bf X_i},{\bf X_0})$. There is no direct interaction between the particles in $\CC_a,\,a>0$.
The systems in $\CC_a,\, a>0$ should be thought as “thermostats” acting on the system $\CC_0$ across the separating walls via their mutual pair interactions.
The thermostats [*temperature*]{} $T_a$ is defined to be proportional to the kinetic energy via the Boltzmann’s constant $k_B$. Setting $K_a\=\fra12\sum_a\dot{\V X}_{a}^2\defi \fra32 k_B T_a\,N_a$, it is supposed constant and kept such by the action of suitable (phenomenological) forces on the $i$-th particle in $\CC_a$ of the form $-\a_a \dot{\bf X}_{ai}$. The $\a_a$ can be taken
$$\a_a = \fra{W_a-\dot U_a}{3 N_a k_B T_a}\Eq(e2.1)$$ where where the work performed by the system on the thermostats particles $-\sum \dot{\V X}_a\cdot\Dpr_{\V X_a}U_a(\V X_0,\V X_a)$ can be called $W_a=\dot Q_a$ = heat given to the thermostat $\CC_a$ by the system in $\CC_0$. The external forces $\V F_i$ are assumed to be purely positional.
Given the above dynamical model (for heat transport) remark that it is [*reversible*]{} and time reversal is just the usual velocity reversal (because the thermostat forces are [*even*]{} under global velocities change). Furthermore the divergence of the total phase space volume can be immediately computed and turns out to be $\s(\V X)=\s_0(\V
X)+\dot U(\V X)$ with
$$\s_0(\V X)=\sum_{a=1}^n \fra{\dot Q_a}{k_B T_a}\,\fra{3N_a-1}{3 N_a}=
\sum_{a=1}^n \fra{\dot Q_a}{k_B T_a}\Eq(e2.2)$$ where $U(\V X)=\sum_{a=1}^n \fra{\dot U_a}{k_B T_a}\,\fra{3N_a-1}{3
N_a}$, and $O(N_a^{-1})$ has been neglected in the last equality in (e2.2).
When computing time averages the “extra term” $\dot U$ will not contribute because being a time derivative its average will be $\fra1T(U(S_T\V X)-U(\V X))$ and therefore will give a vanishing contribution for large $T$ and the average of $\sum_{a=1}^n \fra{\dot
Q_a}{k_B T_a}=\s_0$ will be the average of $\s$. If the interaction $U$ is bounded also the fluctuations of the averages of $\s$ and $\s_0$ will coincide. In the cases in which the interactions are not bounded (Lennard-Jones repulsive cores) care has to be exercised in the fluctuations analysis: the picture does not change except in a rather well understood, trivial, way and this will not be discussed here, [@CV03; @BGGZ05].
Note that $\s_0$ is a “boundary term”, in the sense that it depends on the forces through the boundaries and the forces are supposed short range. Thus the question arises whether such kind of thermostats and short range interactions can lead to stationary states: this is not obvious but the “efficiency” of such thermostats has been investigated in molecular dynamics simulations, [@AES01; @GG06], leading (not surpringly) to the result that the thermostat mechanism in Fig.1 can lead to stationary states (even in presence of additional positional forces stirring the particles in $\CC_0$).
Since the thermotats are regarded in equilibrium the above expression shows that $\s(\V X)$ can be “legitimately” called the entropy increase of the reservoirs: so the mechanical notion of phase space contraction acquires a clear physical meaning: and this is a no small achievement of a long series of works based on simulations of molecular dynamics, [@EM90; @Ru99].
=3=1
\(1) Other studies of fluctuations have been proposed: they are rather different and apply to systems which are not stationary. The object of study are initial data [*sampled within an equilibrium distribution*]{} of a Hamiltonian system and subsequently evolved with the equations of motion of a dissipative time reversible system.
Then the phase space contraction averaged over a time $\t$, $a\defi\fra1\t\ig_0^\t\s(S_t x)dt$, will be such that the probability $P_0(a)$ [*with respect to the initial equilibrium distribution*]{} for $a$ to have a given value is such that
$$\fra{P_0(a)}{P_0(-a)}=e^{a \t}.\Eq(e3.1)$$ This is an exact identity, immediately following from the definitions. It involves [*no error terms*]{}, unlike the “similar” (e1.6) that can be written also as $\fra{P(p)}{P(-p)}=e^{p\s_+\t+O(1)}$, with $P$ the [*probability with respect to the stationary distribution*]{}, which is [*singular*]{} with respect to the equilibrium distributions if $\s_+>0$.
It has been claimed that, being valid for all times, it implies the fluctuation relation, (e1.6), for stationary states (at least when the stationary state exists). This would imply a simple, direct and [*assumptionless*]{} derivation of the fluctuation theorem in (e1.7) and should hold in spite of the fact that in [@GC95] an assumption about the chaotic nature of the motions is needed to derive it, together with a rather detailed understanding of the nature of chaotic systems.
However a derivation of the fluctuation theorem (e1.6) from (e3.1) involves considering (e3.1) [*after*]{} the limit $\t\to\io$ has been performed: a rather unclear procedure (note that the [*r.h.s.*]{} depends on $\t$). Leaving aside the logical consitency problems it should be kept in mind that in the stationary state, at least in the interesting cases in which $\s_+>0$ and there is dissipation, the statistics of motion will be controlled by a distribution that has nothing to do with the initial equilibrium distribution in which the averages in (e3.1) are considered.
Therefore the claim is incorrect and it is no surprise that some kind of chaos has to be present to obtain the fluctuation relation (e1.6). In fact one can give examples of simple systems in which (e3.1) holds for all times, the system evolves towards a stationary state and nevertheless the (e1.6) [*does not hold*]{}, [@CG99]. The confusion has crept into the literature and even affected experiments: this can only be explained by a certain lack of attention to the literature due to the urge to find an easy way of testing the large fluctuations in real systems (fluids, granular materials, or even biological systems).
Another aspect of (e3.1) is that it involves $a$ rather than $p=a/\s_+$. This is clearly a matter of convention: however care has to be exercised because the fluctuation relation (e1.6) is valid for $|p|<p^*$ with $p^*\ge1$ being a physically nontrivial quantity, [@Ga95b]. In terms of $a$ this means that it is valid for $|a|<\s_+$. Overlooking this fact has led to think that it should hold for all $a$’s and has led to errors in the literature. The errors are particularly noticeable in cases in which $\s_+$ is close to zero, when the interpretation of simulations becomes quite difficult because long time scales become relevant. It has to be noted that $\s_+^{-1}$ is a time scale diverging as $\s_+\to0$, see the discussion in [@BGGZ05].
\(2) A different, interesting, fluctuation result is [*Jarzynski’s formula*]{} which provides the means of computing the free energy difference between two [*equilibrium states*]{} at the same temperature.
Imagine to extract samples $x=(p,q)$ in phase space with a canonical probability distribution $\m_0(dpdq)= Z_0^{-1}e^{-\b H_0(p,q)}dpdq$, with $Z_0$ being the canonical partition function, and let $S_{0,t}(p,q)$ be the solution of the Hamiltonian [*time dependent*]{} equations $\dot p=-\dpr_q H(p,q,t),\dot q=\dpr_p H(p,q,t)$ for $0\le
t\le1$. Let $H_1(p,q)\defi H(p,q,1)$, then, [@Ja97; @Ja99],
\(i) The reader will recognize in this [*exact identity*]{} an instance of the Monte Carlo method. Its interest lies in the fact that it can be implemented [*without actually knowing*]{} neither $H_0$ nor $H_1$ nor the [*protocol*]{} $H(p,q,t)$. To evaluate the difference in free energy bewteen two equilibrium states [*at the same temperature*]{} of a system that one can construct in a laboratory, when the system changes its energy function from $H_0$ to $H_1$ (not necessarily explicitly known), then “all one has to do” is\
(a) To fix a protocol, a procedure, to transform the forces acting on the system along a well defined [*fixed once and for all*]{} path from the initial values to the final values in a fixed time interval ($t=1$ in some units), and\
(b) Measure the energy variation $W$ generated by the machines implementing the protocol. This is a really measurable quantity at least when $W$ can be interpreted as the work done on the system, or related to it.\
(c) Then average of the exponential of $-\b W$ with respect to a large number of repetition of the protocol abd apply (e3.2). This can be useful even, and perhaps mainly, in biological experiments.
\(ii) Imagine a protocol consisting in lifting a container with a gas in equilibrium to height $z$: the Hamiltonian changes by $Mgz$, if $M$ is the total mass and $g$ gravity constant. Eq. (e3.2) of course is correct, being an identity, and gives a free energy variation equal to $\b M g z$ while normally one would say that the free energy, and every other thermodynamic quantity, should have remained unchanged. Whether or not there has been a free energy variation really depends on what one is interested in studying. Thus if the interest is in measuring free energy variations in a biology experiment care has to be given (and is actually given) to make sure that the protocol followed does not introduce spurious, quite hidden, forms of work. This makes, once more, clear that the application of a mathematical identity to real systems requires careful examination of the conclusions drawn.
The two formulae (e3.2) and (e1.7) bear some similarities but are, however, quite different:
\(1) the $\ig_0^\t \s(S_tx)\, dt$ in (e1.7) is an entropy creation rather than the energy variation $W$.
\(2) the average in (e1.7) is over the SRB distribution of a stationary state, in general out of equilibrium, rather than on a canonical equilibrium state.
\(3) the (e1.7) says that $\media{ e^{-\ig_0^\t
\e(S_tx)\,dt}}_{SRB}$ is bounded as $\t\to\io$ rather than being $1$ exactly unlike (e3.2) which holds without corrections, [@Ja97; @Ja99].
The (e3.2) has proved useful in various equilibrium problems (to evaluate the free energy variation when an equilibrium state with Hamiltonian $H_0$ is compared to one with Hamiltonian $H_1$); hence it has some interest to investigate whether (e1.7) can have some consequences.
If a system is in a steady state and produces entropy at rate $\s_+$ (a living organism feeding on a background) the fluctuation relation (e1.6) and its consequence Bonetto’s formula, (e1.7), gives us informations on the fluctuations of entropy production, of heat produced, and (e1.7) [*could be useful*]{}, for instance, to check that all relevant heat transfers have been properly taken into account. This suggests that the fluctuation relation for stationary states could have some applications even in experiments in biology and be a valuable complement to (e3.2).
\(3) Finally the identification of entropy creation and pase space contraction suggests a possible quantitative measure for “how irreversible” is a transformation between two different stationary states (equilibrium or not). Since physical processes are often accompanied by volume changes with time $V\to V_t$ it is natural to allow them and to change the definition of phase space contraction (e2.2) to, [@Ga06],
-5mm $$\s^\G(\V X)=\s_0(\V X)+\dot U-N\fra{\dot V_t}{V_t}\Eq(e3.3)$$ where $N$ is the number of particles in the volume $V$.
Then one can try to define the “irreversibility time scale” $\t(\P)$ for a process $\P$ measuring the time scale over which the process manifests its irreversibility. Suppose that in the process $\P$ the parameters controlling the forces change with time from an initial value $\V F_0$ to a final one $\V
F_\io$; to be definite suppose $$\V F(t)=\V F_0+(1-e^{-\g t})(\V F_\io-\V F_0)\Eq(e3.4)$$ The (e3.4) allows us to consider the [*quasi instantaneous*]{} changes ($\g\to\io$) as well as the [*quasi static*]{} ones ($\g\to0$).
Starting the $N$–particles system in the stationary state $\m_0$ with parameters $\V F_0$ it evolves to $\m_t$ and, eventually, to the stationary state $\m_\io$ with parameters $\V F_\io$.
Let $\m_{srb,t}=$ be the SRB distribution with parameters $\V F(t)$ “frozen” at value taken at time $t$. Then if $\s^{srb}_t$ is the entropy creation rate in the “frozen” state $\m_{srb,t}$ an [*irreversibility time scale*]{} for $\P$ could be defined as
-8mm $$\t(\P)^{-1}=\fra1{N^2}\ig_0^\io
\Big(\media{\s^\G_t}_{\m_t}-\media{\s^{srb}_{t}}_{SRB,t}\Big)^2
dt\Eq(e3.5)$$
-6mm which can be checked to give the “expected results” in simple cases like the Joule expansion, see [@Ga06]. Quasi instantaneous processes $\P$ have a short $\t(\G)$, meaning that irreversibility becomes noticeable immediately, while quasi static processes have a long $\t(\P)$ indicating the opposite situation.
\(4) Although the above analysis is restricted to particle systems it can be extended to more general systems, in particular to fluids and turbulence. Not surprisingly as turbulence has been a source of inspiration for the development of the above ideas, [@Ga02; @Ga06].
[[R-0.4mm0.5mm-0.4mm V-0.3mm0.5mm-0.4mm E-.3mm 0.5mm]{}]{}
[37]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ().
, ****, ().
, ****, ().
, ****, ().
, ** (, , ).
, ** (, , ).
, ****, ().
, ****, ().
, , ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , , , , pp. ().
, , , ().
, , , , ****, ().
, ** (, , ).
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ().
, ****, ().
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Hang Li,$^{1,2,\dagger}$ Jian-Peng Dou,$^{1,2,\dagger}$ Xiao-Ling Pang,$^{1,2}$ Chao-Ni Zhang,$^{1,2}$\
Zeng-Quan Yan,$^{1,2}$ Tian-Huai Yang,$^{1,2}$ Jun Gao,$^{1,2}$ Jia-Ming Li,$^{3,\ast}$ Xian-Min Jin$^{1,2,\ast}$\
\
\
\
\
\
\
\
\
\
\
title: Multipartite Entanglement of Billions of Motional Atoms Heralded by Single Photon
---
Quantum entanglement is of central importance to quantum computing, quantum metrology, quantum information as well as the nature of quantum physics. Quantum theory does not prevent entanglement from being created and observed in macroscopic physical systems, in reality however, the accessible scale of entanglement is still very limited due to decoherence effects. Recently, entanglement has been observed among atoms from thousands to millions level in extremely low-temperature and well-isolated systems. Here, we create multipartite entanglement of billions of motional atoms in a quantum memory at room temperature, and certify the genuine entanglement via $M$-separability witness associated with photon statistics. The information contained in a single photon is found strongly correlated with the excitation shared by the motional atoms, which intrinsically address the large system and therefore stimulate the multipartite entanglement. Remarkably, our heralded and quantum memory built-in entanglement generation allows us to directly observe the dynamic evolution of entanglement depth and further to reveal the effects of decoherence. Our results verify the existence of genuine multipartite entanglement among billions of motional atoms at ambient condition, significantly extending the boundary of the accessible scale of entanglement. Besides probing the quantum-to-classical transition in an entirely new realm, the developed abilities of manipulating such a large-scale entanglement may enhance a wide spectrum of applications for emerging quantum technologies.\
Introduction. {#introduction. .unnumbered}
-------------
Quantum technologies, incorporating quantum entanglement [@EPR_1935] into communication [@np_QCm_Gisin; @QIPC_2005], simulation [@S.; @Lloyd_Science; @J.Zhang_Nature; @K.; @Eckert_NPhysics; @P.; @Hauke_NPhysics; @X.-L.; @Qi_NPhysics], computation [@prl_cluster-QCp; @science_OQC; @nature_QCp] and metrology [@review_quantum; @metrology], exert great advantages beyond classical approaches. For a large-scale multipartite entangled systems, the dimension of Hilbert space will be exponentially expanded as entangled particles increase, which inspires novel approaches of quantum computing or direct simulation for classically intractable problems [@QIPC_2005; @nature_QCp]. The ability to access large-scale and more practical multipartite entanglement has been regarded as a benchmark for quantum information processing, like the road map towards quantum supremacy [@quantum; @supremacy]. However, decoherence resulting from strong internal interactions and coupling with environment makes entanglement fragile, which also limits the expansion of the scale of multipartite entanglement, especially reaching the level of macroscopic physical systems.
So far, significant experimental progresses have been made in realizing different classes of multipartite entanglement in different artificially-engineered quantum systems. Greenberger-Horne-Zeilinger state, A well-known multipartite entanglement that shares the correlation on the creation and annihilation excited on all the particles, has already been generated at a scale up to 12 qubits in superconducting systems [@PRL.122.110501], 18 qubits in photonic systems [@PRL.120.260502], and 20 qubits in ions systems [@PRX.8.021012]. The exponentially low efficiency in simultaneously detecting many particles restricts the achievable scale. An interesting way to enhance the collective correlation is to create twin Fock entanglement state in a Bose-Einstein condensate (BEC) system through quantum phase transitions [@science_BEC]. W state is another representative multipartite entanglement that shares the correlation of the creation excited in one particle and annihilation in all other particles. The requirement of detecting only one excitation for W state is free from the exponential inefficiency of coincidence measurement, therefore can be more easily achieved at large scale, especially in atomic ensembles [@science_40atoms; @nature_680atoms; @nature_3000atoms; @NC_million-atoms].
The achieved large-scale entanglement states, however, have to be prepared and detected in the systems that are maintained at extremely low temperature and well isolated with environment to eliminate decoherence effects. The decoherence and noise induced by the motion and collision of room-temperature atoms are apparently harmful [@nPhys_msmemory; @collision_JWPan], and therefore were avoided in purpose in previous endeavors of observing large-scale entanglement. Though being more challenging, it would be more desirable to explore whether large-scale entanglement can exist in ambient condition and shared by many more motional atoms, not only for the fundamental interest of probing the boundary of quantum to classical transition, but also for future real-life quantum technologies.
Here, we experimentally demonstrate a multipartite entanglement of billions of motional atoms in a quantum memory operated at room temperature. The multipartite entanglement W state in a hot atomic vapor cell is heralded by registering a Stokes photon through far off-resonance spontaneous Raman scattering (SRS). In order to certify and quantify the entanglement scale, we convert the shared excitation of W state into an anti-Stokes photon by applying another interrogation pulse, and reveal a entanglement depth up to billions of atoms by the witness constructed with the correlated photon statistics. The far off-resonance configuration endorses the broadband feature of our scheme allowing to be operated at a high data rate. Furthermore, our heralded and quantum memory built-in fashion of entanglement generation allows us to directly observe the dynamic evolution of entanglement depth in a dissipative environment.
Experimental implement and results. {#experimental-implement-and-results. .unnumbered}
-----------------------------------
To herald multipartite entanglement W state, we adopt the SRS regime as proposed in the Duan-Lukin-Cirac-Zoller (DLCZ) protocol originally aiming at realizing applicable quantum repeaters [@nature_DLCZ]. However, we have to conceive a far off-resonance scheme to avoid the huge fluorescence noise in room-temperature atomic ensemble [@Dou_npj; @Dou_FORD; @Pang_hybrid], which does not exist in cold ensembles and diamonds [@Chou_cold; @science_diamond]. The energy levels of the $\Lambda $-type configuration are shown in Figure 1a. This process will generate the product state of correlated photon-atoms pairs, which can be expressed as [@science_diamond] $$\label{eq1}
\left | \psi _{s} \right \rangle= \left [ 1+\varepsilon_{s} S^{\dagger}a^{\dagger}\right ]\left | vac \right \rangle$$ where $\varepsilon_{s}$ is the excitation probability of Stokes photon, $\left |vac\right \rangle=\left |vac_{opt}\right \rangle\bigotimes \left |vac_{ato}\right \rangle$ is the initial product state of photon-atoms system, $S$ and $a$ are the annihilation operators of spin wave and Stokes photon, respectively. Here, we set the intensity of control light pulse so weak that the excitation probability $\varepsilon_{s}$ is much smaller than unity. Therefore, we can ignore the higher-order terms in the creating of $\left | \psi _{s} \right \rangle$ with extremly small probability [@nature_DLCZ; @PRL_kimble]. With the creation operators acting on the initial state of atomic ensemble, the $W$ state is written as [@nPhys_msmemory; @Dou_FORD] $\left| W_{1}\right\rangle {\rm{ = }}\frac{1}{{\sqrt N }}\sum\limits_{j = 1}^N {{e^{i\Delta{\vec k}\cdot {{\vec r}_j}}}} \left| {{g_1}{g_2}...{s_j}...{g_N}} \right\rangle $, where $N$ is the number of involving atoms, $\Delta{\vec k}$ is the wave-vector of spin wave, ${\vec r}_j$ is the position information of $j_{th}$ excited atom. This generating process of entanglement is shown in Figure 1b, and the W state can be heralded through the detection of one scattering Stokes photon. $\left| W_{1}\right\rangle$ contains only one excitation shared by all motional atoms illustrated in Figure 1c, where every atom posesses the equal probability being excited with spin up.
It is inevitable that the SRS in the generating process may produce high-order excitations with a comparably low probability, but such terms would change the structure of our desired multipartite entanglement. The entangled ensemble with more than one excitation event can be generally expressed as $\left| W_{2}\right\rangle {\rm{=}}\sqrt{\frac{2}{N(N-1)}}\sum\limits_{i<j}^N {{e^{i\Delta{\vec k}\cdot\left({{\vec r}_j+{\vec r}_i}\right)}}}\left|{{g_1}{g_2}...{s_i}...{s_j}...{g_N}}\right\rangle$, and higher-order events have negligible contributions. To certify and qualify the W state, we need to apply another optical probe pulse to convert the shared single excitation in atomic entanglement state into an anti-Stokes photon, as shown in Figure 1a and Figure 1b. In order to obtain the information of entanglement depth, we analyze the photon number statistics of the correlated Stokes and anti-Stokes photons via a Hanbury Brown-Twiss interferometer (see Methods). Due to the decoherence effects, as Figure 1d shows, the atomic ensemble with high-order excitations evolves to several subgroups, where each part shares only single excitation [@NC_million-atoms]. Suppose that the whole atomic ensemble is in a pure state, containing $M$ separable parts, which can be expressed in a product form $\left |\psi \right \rangle=\left |\psi _{1}\right \rangle\otimes \left |\psi _{2}\right \rangle...\otimes \left |\psi _{M} \right \rangle$, where $M$ indicates the number of separable subgroups, $\left |\psi _{i}\right \rangle(i=1,...,M)$ represents each separable group that may contain individual multipartite entanglement, while different subgroups are independent from the others (see Methods). Then, we can define entanglement depth as $D=N/M$, with $N$ is the number of total atoms participating in interaction [@NC_million-atoms].
In order to quantify the multipartite entanglement, we adopt the entanglement witnesses incorporated with photon number statistics of the correlated photons pair. Such witness is efficient especially when the vacuum component of the state is dominant [@NC_million-atoms]. For each given $M$ value, we ought to determine the lower bound of the entanglement state with $D$ particles entangled (see Methods). The witness operator can be expressed as $$\label{eq5}
\omega ^{M}=\left |W_{2} \right \rangle\left \langle W_{2} \right |-p_{2}^{bound}(p_{1},M)$$ where the two key parameters $p_{1}$ and $p_{2}$ are projective probabilities in the forms of $\left | \left \langle W_{1}|\psi \right \rangle \right |^{2}$ , $\left | \left \langle W_{2}|\psi \right \rangle \right |^{2}$, and $p_{2}^{bound}(p_{1},M)$ stands for the theoretical minimal value of $p_{2}$ under the condition of fixed $p_{1}$ and $M$ value. For the density matrix $\rho$ of experimental state, $tr(\rho \omega ^{M} ) < 0$ means that the entanglement depth is at least $D=N/M$. In actual experiments, $p_{1}$ should be defined as the conditional probability of detecting a correlated anti-Stokes photon with a heralded forward Stokes photon, and the probability $p_{2}$ stands for the probability of two excitation events, which is deduced by the autocorrelation function $g_{AS_{1}-AS_{2}|S}^{(2)}={2p_{2}}/{p_{1}^{2}}$ measured by a Hanbury Brown-Twiss interferometer.
The decoherence effects can be revealed by the observation of entanglement depth’s evolution via adjusting the delay of optical probe pulse in our quantum memory built-in configuration. The $p_{1}$ can be influenced by the retrieval efficiency of quantum memory, the photon loss of the channels and detectors. Therefore, the final experimental datas should be handled as the following two stages: the raw measured datas and the processed datas after subtracting the loss of the channels and detectors. The latter reflects the actual entanglement state at the moment just after applying the probe pulse. The $M$ values of entanglement state evolving with storage time are showed in Figure 2a, where the energy of the light pulses is 225 $pJ$. The data point below the boundary curve of fixed $M$ value indicates that there is an entanglement depth at least $N/M$. Our experimental datas show that the number of entanglement subgroups increases as the memory time elapsing, i.e. the decoherence effects caused by the thermal motions of atoms [@nPhys_msmemory; @Dou_FORD] will tremendously influence the structure of entanglement, which is consistent with the physical picture depicted in Figure 1d.
During the process of verifying the existence of entanglement, the collective enhancement effect contributes to the transducing of Stokes photon due to the phase coherence of $W$ state [@nature_DLCZ]. The variation of $p_{1}$ with the delay of the probe pulse is shown in Figure 3a. Due to the decoherence of phase mainly resulting from the motions of warm atoms, the effects of collective enhancement become deteriorative, which results in the exponential decay of $p_{1}$. What is more, we measure the cross-correlation between the correlated photons and the autocorrelation of the retrieved anti-Stokes photon as shown in Figure 3b. The degrade of quantum correlation and single-photon characteristic implies the variation of the structure of multipartite entanglement according to the relation of $p_{2}$ and $g_{AS_{1}-AS_{2}|S}^{(2)}$ (see Methods), which are consistent with the deduced $M$ values in Figure 2. Our results well exhibit the transition from quantum to classical in multipartite entanglement of billions of motional atoms heralded by single photon.
From another perspective, we can demonstrate how the entanglement depth varies with the delay time. In order to determine the informations of entanglement depth, the number of total caesium atoms involved in the interaction is the key parameter that should be measured precisely. The atomic density and total atomic number can be obtained by fitting the measured transmission rates of light with different frequency according to the absorption model (see Methods). The results show that there are nearly at least 8.85 billion motional atoms sharing the one excitation constituting the W state. As is shown in Figure 3c, despite the fast decrease of entanglement depth resulting from the increased noise and the destructive effects of decoherence, there are still considerable entanglement depth in the warm atomic ensemble after storing for the time of microseconds level.
It is also accessible to manipulate the size of large-scale entanglement state in the macroscopic ensemble by changing our experimental parameters. There are several elements that will influence the entanglement depth, such as the beam waist, the detuning, and the energy of addressing light. For the far off-resonance DLCZ protocol, we have chosen a “sweet point" for the detuning, which has been experimentally demonstrated to have the lowest unconditional noise [@Dou_npj; @Dou_FORD]. As for the beam waist, it is not appropriate to utilize too large beam waist to provide enough addressing energy. Note that there is no problem for using larger beam waist given that strong addressing light is equipped, whose advantage is that there will be more atoms involved in the creation of entanglement state. In our experiment, we choose a beam waist of $100 \mu m$ for providing a sufficient excitation rate. What dominantly influences the $p_1$ probability in our experiment is the energy of the addressing light pulse, which determines the excitation probability of the Stokes photon during the SRS process.
The $M$ values of multipartite entanglement created by different pulse energies, 115.5$pJ$, 225$pJ$, 330$pJ$ respectively, are shown in Figure 4a. The relation among entanglement depth and excitation probability, light pulse energy is also analyzed and shown in Figure 4b. The results show that the stronger light pulse energy has a higher $p_{1}$ probability because of the higher converting efficiency of $W$ state, but has a smaller $M$ value, which indicates that the structure of multipartite entanglement has not been deteriorated by noises. Actually, for conveniently and efficiently evaluating the scale of entanglement, the aforementioned definition of entanglement depth only delivers a lower bound of the actual entanglement scale [@NC_million-atoms], since the scale of genuine entanglement depth should be associated with the largest size of all subgroups. Thus, it is reasonable that the same lower bound of entanglement depth is observed with different excitation probabilities.
Discussion and Conclusion. {#discussion-and-conclusion. .unnumbered}
--------------------------
In summary, we have experimentally demonstrated a multipartite entanglement of billions of motional atoms heralded by single photon. With the quantum memory built-in and broadband capacities, we have efficiently displayed the dynamical evolution of entanglement depth and decoherence effects, as well as showing the transition from quantum to classical in the realm of multipartite entanglement. Our work has certified that quantum entanglement can be observed in a macroscopic room-temperature atomic ensemble with motional atoms and demonstrated the accessibility of feasible manipulations of entanglement depth, which greatly expanded the bound of operating large-scale multipartite entanglement and may have potential applications for future quantum information science and technologies.
Creating a larger-scale multipartite entanglement beyond billions of atoms is possible, a larger beam waist and stronger energy of light pulse will be helpful with the prerequisite of well controlled levels of noise. What’s more, the larger beam waist can mitigate the detrimental effect of decoherence brought by the thermal motions of atoms, since there is a broader space to prevent warm atoms from escaping from the interaction area, which leads to a longer lifetime of multipartite entanglement. Recent works also show that the anti-relaxation coating of vapor cell will preserve the coherence for longer time [@balaba_coating; @polsik_DLCZ], which may be beneficial for improving the maintenance of the heralded multipartite entanglement. Remarkably, recent proposals and experimental developments about transferring the single collective excitation of electronic spins to noble-gas nuclear spins by spin exchange regime may exceedingly prolong the lifetimes of the W state even up to several hours [@prl_nuclear; @spin; @ofer_noble-gas].
The $W$ state with phase informations encoded in billions of atoms exists in the form of a spin wave, which resemble a tremendous networked quantum sensors with entanglement between each elements. The phase informations of spin wave are not only related to the position informations of motional atoms, but also sensitive to some other physical parameters related to atomic internal states, like magnetic field [@interfermeter_prl], which makes the $W$ state become a promising candidate for quantum sensing. Due to the collective enhancement effects in the readout of spin wave, the huge scale will become an advantage in enhanced metrology. The nonclassical correlations contained in the $W$ state among huge entangled particles may endow the meteorological gain over classical states [@PRL_Wgain], such as being beneficial to interferometry measurement for beating standard quantum limit [@review_quantum; @metrology]. Interestingly, $W$ state is also robust for the purpose of metrology, because the remaining particles are still entangled while one particle is trace out. Furthermore, the multipartite entanglement constructed between these quantum sensors may significantly enhance the precision of multiparameter estimation [@PRL_MPE].
Acknowledgments. {#acknowledgments. .unnumbered}
----------------
The authors thank Jian-Wei Pan for helpful discussions. This research was supported by the National Key R&D Program of China (2019YFA0308700, 2017YFA0303700), the National Natural Science Foundation of China (61734005, 11761141014, 11690033), the Science and Technology Commission of Shanghai Municipality (STCSM) (17JC1400403), and the Shanghai Municipal Education Commission (SMEC) (2017-01-07-00-02- E00049). X.-M.J. acknowledges additional support from a Shanghai talent program.\
Methods {#methods .unnumbered}
-------
#### Experimental details: {#experimental-details .unnumbered}
The $^{133}$Cs cell is 75-mm-long and is placed into a magnetic field shielding, which has been filled 10 $Torr$ Ne buffer gas to alleviate collisions between cesium atoms. In order to get a large optical depth, which means there are more atoms participating in the creation of entanglement, the $^{133}$Cs cell is heated to 61$^{\circ}$C. To generate high-speed light pulse with enough intensity shown in Figure 1, we have developed a system to satisfy the needs of tunable central frequency, broad bandwidth, and more importantly, control pulse generated in a programable fashion. We have also established a frequency locked system to stabilize the W state during the creating and certifying process. It should be noticed that our collinear scheme makes the correlated Stokes and anti-Stokes photons being coaxial propagating under the phase-matching condition, therefore six home-built Fabry-Pérot cavities with high performance are employed to split and retrieve Stokes and anti-Stokes photons. As for single cavity, the transmission rate of each cavity reaches around $95\%$ , and the extinction rate of each cavity for signal and noise is up to $500:1$.
#### Witness for $M$-separability: {#witness-for-m-separability .unnumbered}
In the certifying process, we apply an optical pulse to interrogate the state of atomic ensemble by analyzing the correlated photons statistics. Firstly, we assume that the atomic state can be described in a pure state, which can be decomposed into a $M$-separable form like equation $\left |\psi \right \rangle$ in the main text. Due to the imperfect experimental conditions and decoherence, the representation of state in every independent subgroup may be a superposition of many possible states[@Duan_25memory(2017)]. Since we consider at most two excitations in the SRS process, the specific form of each state in the subgroup can be expressed in following state $$\label{eq1}
\left |\phi _{i} \right \rangle=a_{i}\left |W _{0} \right \rangle+b_{i}\left |W _{1} \right \rangle+c_{i}\left |W _{2} \right \rangle$$ where $\left |W_{1} \right \rangle$, $\left |W_{2} \right \rangle$ are the Dicke states in each subgroup; $\left |W_{0} \right \rangle$ is a vacuum state. Thus, the whole state of ensemble is the product of all subgroups: $$\label{eq2}
\left |\Psi \right \rangle=\bigotimes_{i=1}^{M}a_{i}\left |W _{0} \right \rangle+b_{i}\left |W _{1} \right \rangle+c_{i}\left |W _{2} \right \rangle$$ The two probability $p_{1}$, $p_{2}$ can be calculated specifically, $$\label{eq3}
p_{1}=\frac{\left |\prod_{i=1}^{M}a_{i} \right |^{2}}{M}\left | \sum_{i} \frac{b_{i}}{a_{i}}\right |^{2}$$ $$\label{eq4}
p_{2}=\frac{\left |\prod_{i=1}^{M}a_{i} \right |^{2}}{M^{2}(1-\frac{1}{N})}\left | \sqrt{2}\sum_{i<j} \frac{b_{i}b_{j}}{a_{i}a_{j}}+\sqrt{1-\frac{1}{D}}\sum_{i} \frac{c_{i}}{a_{i}}\right |^{2}$$ where $D=\frac{N}{M}$ is the entanglement depth. Actually, the entanglement depth should be defined as the largest scale of all the entangled subgroups, however, we can take ${N}/{M}$ as the representation of entanglement depth to avoiding get vary large subgroup size[@NC_million-atoms].
In order to determine the entanglement depth in experiment, we need to determine the lower bound for $p_{2}$ with a fixed $M$ value. Obviously, we need the probability $p_{2}$ as low as possible in actual experiment, which means that the fidelity of target $W_{1}$ state is high. The bound can be calculated by $$\label{eq5}
p_{2}^{bound}(p_{1},M)=min{\left \{ p_{2}|\psi :p_{1}, M = const \right \}}$$ Note that there are constraints between the coefficients of superposition in equation (3), $\left | a_{i} \right |^{2}+\left | b_{i} \right |^{2}+\left | c_{i} \right |^{2} \leq 1 $, and we can take the approximation $|a_{i}|^{2}+|b_{i}|^{2}+|c_{i}|^{2}= 1$ owing to the neglectable higher excitations. Utilizing the Lagrange multiplier method deduced in the supplementary notes of [@NC_million-atoms], the conclusion is that the symmetric solution gives the global minimal value of $ p_{2}$ for $M\leq5$. This symmetric solution requires that $a_{i}=a, b_{i}=b, c=-\sqrt{1-a^{2}-b^{2}}$. In this condition, the optimal values for $p_{1}, p_{2}$ are the solution as follows $$\label{eq6}
p_{1}^{sym}=Ma^{2M-2}b^{2}$$ $$\label{eq7}
p_{2}^{sym}=a^{2M}(\frac{1}{\sqrt{2}}(M-1)\frac{b^{2}}{a^{2}}+\frac{c}{a})^{2}$$ The final form of function $p_{2}^{bound}(p_{1},M)$ is, $$\label{eq8}
p_{2}^{bound}(p_{1},M)=a^{2M}[\frac{1}{\sqrt{2}}\frac{p_{1}(M-1)}{M}a^{-2M}-\frac{1}{a}\sqrt{1-a^{2}-\frac{p_{1}}{M}a^{2-2M}}]^{2}$$ For fixed $p_{1}$ and $ M$, we need to calculate minimal value of $p_{2}$ with $a\in (0,1)$. The theoretical bound of equation (10) can be obtained by take all values of $p_{1}\in (0,1)$, which is shown in Figure 2a and Figure 3b with different $M$ values.
#### Number of atoms involved in the creation of multipartite entanglement: {#number-of-atoms-involved-in-the-creation-of-multipartite-entanglement .unnumbered}
Theoretically, the transmission rate of probe light passing through the atomic ensemble has the following form [@phdOxford_Sprague; @absorption_model] $$\label{eq6}
T(\omega )=exp\left \{-\frac{2\pi nkLd^{2}}{h\varepsilon _{0}}\sum_{i=1}^{3}S_{i}l_{i}(\omega ) \right \}$$ where $n$ is the density of atoms, $k$ is the wave vector of probe light, $L$ is the length of our vapor cell, $d$ is the reduced dipole matrix element, $S_{i}$ is the strength of relative coupling from the hyperfine level $F=3$ of the ground state to $F'=2,3,4$ in the excited state[@Cs_Data], $l_{i}(\omega )$ is the normalization lineshape. For more precisely fitting, the normalization lineshape $l_{i}(\omega )$ should be considered as Vigot lineshape [@absorption_model]. More details about the theoretical absorption model and experimental fitting are in the supplementary notes. According to the fitting coefficients, the number density of atoms in the $^{133}$Cs cell is nearly $1.21\times 10^{18} m^{-3}$.
To determine how many atoms participating in the interaction, we need to know the volume of interaction area illuminated by light in the creating and certifying process, which can be evaluated with the volume of Gaussian light within the cell. The light beam used to generate multipartite entanglement is Gaussian beam, whose amplitude of electric field has the spatial dependence in following form $$\label{eq6}
|E(r)|=\frac{E_{0}}{\sqrt{1+\frac{z^{2}}{z_{w}^{2}}}}e^{-\frac{x^{2}+y^{2}}{W_{w}^{2}(1+\frac{z^{2}}{z_{w}^{2}})}}$$ where $z_{w}$ is Rayleigh length of laser beam, $W_{w}$ is the beam waist. Here, we only define the effective area of interaction by the amplitude decrease to ${1}/{10}$ of central magnitude in the Gaussian beam. Actually, the witness applied to certify our $W$ state doesn’t require equal amplitude for the entanglement state in $\left| W_{1}\right\rangle$ [@NC_million-atoms], therefore more atoms can be taken into calculations owing to the extension of Gaussian beam’s intensity. Therefore, the illuminated volume can be calculated as, $$\label{eq6}
V=\pi ln10\int_{-\frac{l}{2}}^{\frac{l}{2}}W_{w}^{2}(1+\frac{z^{2}}{z_{w}^{2}})dz$$ where $l=75.3\times10^{-3}m$ is the length of cesium cell, and beam waist $W_{w}=100\times10^{-6}m$, Rayleigh length $z_{w}=3.69\times10^{-2}m$. Finally, the total number of atoms $N$ involved into the creation of entanglement is $N=nV=8.85\times 10^{9}$.
[99]{}
A. Einstein, B. Podolsky, and N. Rosen, *Phys. Rev.* **47**, 777 (1935).
Nicolas Gisin, Rob Thew, Quantum communication. *Nature Photon.* **1**, 165-171 (2007).
P. Zoller, *et al*, Quantum information processing and communication. *Eur. Phys. J. D*. **36**, 203 (2005).
S. Lloyd, Universal quantum simulators. *Science*. **273**,1073 (1996).
J. Zhang, G. Pagano, P. W. Hess, *et al*, Observation of a many-body dynamical phase transition with a 53-qubit quantum simulator. *Nature (London)*. **551**, 601-604 (2017)
K. Eckert, *et al*, Quantum non- demolition detection of strongly correlated systems. *Nat. Phys*. **4**, 50-54 (2008).
P. Hauke, *et al*, Measuring multipartite entanglement through dynamic susceptibilities. *Nat. Phys*. **12**, 778-782 (2016).
X.-L. Qi, Does gravity come from quantum information? *Nat. Phys*. **14**, 984-987 (2018).
Robert Raussendorf, Hans J. Briegel, A One-Way Quantum Computer. *Phys. Rev. Lett*. **86**, 5188 (2001).
Jeremy L. O’Brien, Optical Quantum Computing. *Science*. **318**, 1567 (2007).
Ladd, T. D. *et al*, Quantum computers. *Nature* **464**, 45-53 (2010).
Pezzè, L., Smerzi, A., Oberthaler, M. K., Schmied, R. & Treutlein, Quantum metrology with non-classical states of atomic ensembles. *Rev. Mod. Phys*. **90**, 035005 (2018).
Preskill, John. Quantum computing and the entanglement frontier. arXiv:1203.5813
Ming Gong, *et al*, Genuine 12-Qubit Entanglement on a Superconducting Quantum Processor. *Phys. Rev. Lett*. **122**, 110501 (2019).
Xi-Lin Wang, *et al*, 18-Qubit Entanglement with Six Photons? Three Degrees of Freedom. *Phys. Rev. Lett*. **120**, 260502 (2018).
Nicolai Friis, *et al*. Observation of Entangled States of a Fully Controlled 20-Qubit System. *Phys. Rev. X*. **8**, 021012 (2018).
Xin-Yu Luo, *et al*. Deterministic entanglement generation from driving through quantum phase transitions. *Science*. **355**, 620 (2017).
Haas, F., Volz, J., Gehr, R., Reichel, J. & Estève, J. Entangled states of more than 40 atoms in an optical fiber cavity. *Science*. **344**, 180 (2014).
Hosten, O., Engelsen, N. J., Krishnakumar, R. & Kasevich, M. A. Measurement noise 100 times lower than the quantum-projection limit using entangled atoms. *Nature* **529**, 505 (2016).
McConnell, R., Zhang, H., Hu, J., Cuk, S. & Vuletic, V. Entanglement with negative Wigner function of almost 3,000 atoms heralded by one photon. *Nature* **519**, 439 (2015).
Florian Fröwis, *et al*. Experimental certification of millions of genuinely entangled atoms in a solid. *Nature Commun* **8**, 907 (2017).
Zhao, B. *et al*. A millisecond quantum memory for scalable quantum networks. *Nature Phys.* **5**, 95-99 (2009).
Manz, S., Fernholz, T., Schmiedmayer, J. & Pan, J.-W. Collisional decoherence during writing and reading quantum states. *Phys. Rev. A.***75**, 040101(R) (2007).
Duan, L. -M., Lukin, M. D., Cirac, J. I. & Zoller, P. Long distance quantum communication with atomic ensembles and linear optics. *Nature* **414**, 413-418 (2001).
Jian-Peng Dou, *et al*. Direct observation of broadband nonclassical states in a room-temperature light?matter interface. *npj Quantum Information* **31**, 55 (2018).
Dou, Jian-Peng and Yang, Ai-Lin, *et al*. A Broadband DLCZ Quantum Memory in Room-Temperature Atoms. *Comms. Phys.* **1**, 55 (2018).
Pang, X.-L. *et al*. A Hybrid Quantum Memory Enabled Network at Room Temperature. arXiv:1803.07122 (2018).
Chou, C. W.*et al*. Measurement-induced entanglement for excitation stored in remote atomic ensembles. *Nature* **438**, 828-832 (2005).
K. C. Lee *et al*. Entangling Macroscopic Diamonds at Room Temperature. *Science*. **334**, 1253 (2011).
C.W. Chou, S.V. Polyakov, A. Kuzmich, H. J. Kimble. Single-Photon Generation from Stored Excitation in an Atomic Ensemble. *Phys. Rev. Lett*. **92**, 213061 (2004).
M. V. Balabas, *et al*. High quality anti-relaxation coating material for alkali atom vapor cells. *Opt. Express.* **18**(6), 5825 (2010).
Michael Zugenmaier, *et al*. Long-lived non-classical correlations towards quantum communication at room temperature. *Comms. Phys.* **1**, 76 (2018).
A. Dantan, *et al*. Long-Lived Quantum Memory with Nuclear Atomic Spins. *Phys. Rev. Lett*. **95**, 123002 (2005).
Or Katz, *et al*. Quantum interface for noble-gas spins. arXiv:1905.12532 (2019).
Bing Chen, *et al*. Atom-Light Hybrid Interferometer. *Phys. Rev. Lett*. **115**, 043602 (2015).
Luca Pezzé, Augusto Smerzi. Entanglement, Nonlinear Dynamics, and the Heisenberg Limit. *Phys. Rev. Lett*. **102**, 100401 (2009).
Timothy J. Proctor, Paul A. Knott, and Jacob A. Dunningham, Multiparameter Estimation in Networked Quantum Sensors. *Phys. Rev. Lett*. **120**, 080501 (2018).
Yunfei Pu. *et al*. Experimental entanglement of 25 individually accessible atomic quantum interfaces. *Sci. Adv.* **4**, 3931 (2018).
Michael Roger Sprague. Quantum Memory in Atomic Ensembles. Ph.D. thesis, University of Oxford, (2014).
D. A. Steck, Cesium D Line Data (2003), URL http://steck.us/alkalidata.
P. Siddons, C. S. Adams, C. Ge, and I. G. Hughes, *J. Phys. B* **41**, 155004 (2008).
Duan, L.-M., Cirac, J. I. & Zoller, P. Three-dimensional theory for interaction between atomic ensembles and free-space light. *Phys. Rev. A* **66**, 023818 (2002).
Nunn. J, Quantum Memory in Atomic Ensembles. Ph.D. thesis, University of Oxford, 481-487 (2008).
![**The schematic diagram of creation and verification of multipartite entanglement.** **a.** The energy levels of creating and verifying W state. Solid lines represent three-level $\Lambda$-type configuration of atoms, two ground states label $\left|g\right\rangle$ ($6S_{1/2}, F=3$) with electronic spin down and $\left|s\right\rangle$ ($6S_{1/2}, F=4$) with electronic spin up, which are hyperfine ground states of cesium atoms (splitting is $\Delta g = 9.2{\rm{GHz}}$) ; excited state labels $\left| {\rm{e}} \right\rangle $ ($6P_{3/2}, F^{'}=2,3,4,5$). The shaded area between energy levels represent broad virtual energy levels induced by the short pump and probe laser pulse (2ns). **b.** The experimental scheme of creating and certifying the multipartite entanglement. The Hanbury Brown-Twiss interferometer is used for analyzing the statistics of the correlated Stokes and anti-Stokes photons, which can further reveal the informations of entanglement depth. **c.** The representation of the dashed arrow means that every atom has equal probability to be spin up or down, which is the main feature of the W state. **d.** The decoherence effects may change the structure of the multipartite entanglement. We take the two excitations event as example to show the evolutions of entanglement depth. The distributed cesium atoms with different colors illustrate the entanglement distribution with several subgroups. The set of same colored atoms is genuine multipartite entanglement, while two sets with different colors do not have the relationship of entanglement.[]{data-label="f1"}](figure1.pdf){width="1\linewidth"}
![**Dynamical evolution of the multipartite entanglement.** The verification results of witness and dynamical evolution of $M$ values with different storage time. The different $M$ values of corresponding theoretical lower bound curves of witness are 10 to $10^{6}$ from right to left. The $M$ values for experimental datas are 5, 6, 10, 14, 92 and 1000 in the right subgragh. Error bars are derived by Poisson distribution of photon number statistics from avalanche photo diodes.[]{data-label="f2"}](figure2.pdf){width="1.0\linewidth"}
![ **Decoherence of the multipartite entanglement.** **a.** The variations of $p_{1}$ probabilities. **b.** The crosscorrelation $g_{S-AS}^{(2)}$ between correlated stokes photon and anti-stokes photon is in the left side (as pink dots show); the autocorrelation $g_{AS_{1}-AS_{2}|S}^{(2)}$ of the retrieved anti-Stokes photon by a heralded Stokes photon is in the right side (as blue dots show). Error bars are derived by Poisson distribution of photon number statistics from avalanche photo diodes. **c.** The evolution of entanglement depth varies with the storage time of quantum memory.[]{data-label="f3"}](figure3.pdf){width="1.0\linewidth"}
![ **Measured entanglement depth with different excitation probabilities.** **a.** The $M$ values with different excitation probabilities are 14, 5, 5, and the corresponding energy are $E_{1}=115.5$ $pJ$, $E_{2}=225$ $pJ$, $E_{3}=330$ $pJ$. The time interval between creation and verification light pulse is 30 ns. Error bars are derived by Poisson distribution of photon number statistics from avalanche photo diodes. **b.** The visulized entanglement depth for comparason with different excitation probabilities. The red dots represent the relations of the excitation probability and pulse energy, and the size of circle taking the red dot as center stands for the scale of entanglement depth.[]{data-label="f3"}](figure4.pdf){width="0.85\linewidth"}
Supplementary Materials: Multipartite Entanglement of Billions of Motional Atoms Heralded by Single Photon {#supplementary-materials-multipartite-entanglement-of-billions-of-motional-atoms-heralded-by-single-photon .unnumbered}
==========================================================================================================
I. Number density of atoms in the atomic ensemble {#i.-number-density-of-atoms-in-the-atomic-ensemble .unnumbered}
==================================================
For determine number density of our vapor cell, we have measured the absorption curve of sweeping frequency. Therefore, it is necessary to build up a theoretical model to fit the absorption curve. The absorption of light beam through a atomic vapor is described by the Beer-Lambert law: $$\label{eq1}
I(z)=I_{0}exp[{-\alpha (\nu,T_{0} )z}]$$ where $I_{0}$ is the beam intensity in the entrance of cell, and absorption coefficient $\alpha (\nu,T_{0} )$is dependent on frequency $\nu$ and temperature $T_{0}$. During the experiment, we use beam pulse, whose intensity is weak enough, so the absorption coefficient is independent of intensity. However, there is not only single transitions existing, but many absorption transitions occurring in Cs atoms. As a result, the transmission $T(\omega )$ should be extended as: $$\label{eq2}
T(\omega)=exp[-(\sum\alpha _{i})L ]$$ where $\sum S_{i}\alpha _{i}$ is total absorption coefficient, $L$ is length of vapor cell. In our system, considering the selection rules of transitions, the possible transition mainly occurs between hyperfine level $F=3$ of the ground state$(6S_{1/2})$ to $F'=2,3,4$ in the excited state$(6P_{3/2})$. So, these absorption coefficients can be denoted as $\alpha _{32}$, $\alpha _{33}$ and $\alpha _{34}$, finally the actual transmission can be expressed in following form: $$\label{eq3}
T(\omega)=exp[-(\alpha _{32}+\alpha _{33}+\alpha _{34})L ]$$
The relative strengths of transition $S_{32}$, $S _{33}$ and $S _{34}$ has been regarded as usually used constants of database, which can be obtained in [@Cs_Data]. The absorption coefficients $\alpha _{32}$, $\alpha _{33}$ and $\alpha _{34}$, are related to the electric susceptibilities $\chi _{32}$, $\chi _{33}$ and $\chi _{34}$ [@absorption_model]: $$\label{eq4}
\chi _{i}(\omega ) =S_{i}d^{2}N\frac{2\pi }{h\varepsilon _{0}}l_{i}(\omega )$$ where $i$ stands for notations of $32$, $33$and $34$; $S_{i}$ is the relative strength of different transitions, which can also be denoted as $S_{32}$, $S _{33}$ and $S _{34}$; d is the matrix element of transition; N is the number density of atoms; $l_{i }(\omega )$ is the lineshape factor of transition $i$, which can be derived from the optical Bloch equations of two-level atom[@Loudon]. And then, we can get the expression of absorption coefficient: $$\label{eq5}
\alpha _{i}(\omega )=k\chi _{i}(\Delta )=kS_{i}d^{2}N\frac{2\pi }{h\varepsilon _{0}}l_{i }(\omega )$$ where k is the wave number of probe light, since the range of sweeping frequency in measurement is small, the wave number can be viewed as constant.
Taking equation $(5)$ into $(3)$, we can get the specific form of transimission: $$T(\omega )=exp\left \{-\frac{2\pi NkLd^{2}}{h\varepsilon _{0}}(S_{32}l_{32 }(\omega ) +S_{33}l_{33 }(\omega )+S_{34}l_{34 }(\omega )) \right \}
$$ Note that the lineshape factor of transition $l_{i }(\omega )$ should take Doppler lineshape into consideration for better agreeing well with the experimental datas.Therefore, considering Doppler effects and collision broadening mechanism, the total lineshape is Vigot type[@Loudon; @absorption_model], which is the convolution of Lorentz lineshape and Doppler lineshape in the form of equation $(7)$. $$\label{eq7}
l_{i}(\omega )=[f_{\Gamma }\otimes g_{\sigma }](\omega )=\int_{-\infty }^{+\infty }f_{\Gamma }(\Delta -\delta _{\omega })g_{\sigma }(\delta _{\omega })d\delta _{\omega }$$ where $\delta _{\omega }=kv_{z}=\frac{\omega}{c}v_{z}$ is Doppler shift of frequency, $v_{z}$ is the velocity of atom along the direction of wave vector $k$, $c$ is the velocuty of light in vaccum; $\Delta=\omega-\omega_{i}$ is the detune of probe light from the resonant frequency; $f_{\Gamma }(\omega)$, $g_{\sigma }(\omega)$ are normalized Lorentz lineshape and Doppler lineshape, respectively, whose expressions as equation$(8), (9)$. $$\label{eq8}
f_{\Gamma }(\omega )=\frac{\frac{\Gamma}{2\pi } }{\Delta ^{2}+(\frac{\Gamma }{2})^{2}}$$ $$\label{eq9}
g_{\sigma }(\omega )=\frac{1}{\sigma \sqrt{2\pi }}exp[-\frac{1}{2}(\frac{\delta_{\omega} }{\sigma })^{2}]$$ where $\Gamma$ is the linewidth of Lorentz lineshape, and $\sigma$ is the linewidth of Doppler lineshape. Inserting equation$(8), (9)$ into $(7)$, we can get the function of Voigt lineshape: $$\label{eq10}
l_{i}(\omega )=\frac{2\omega }{\pi \sqrt{2\pi }\Gamma \sigma c}\int_{-\infty }^{+\infty }\frac{1}{1+(\frac{\omega -\omega _{i}-\frac{\omega }{c}v_{z}}{\frac{\Gamma }{2}})^{2}}exp[-\frac{1}{2}(\frac{v_{z}}{\frac{c}{\omega }\sigma })^{2}]dv_{z}$$ To simplify the form of equation $(10)$, we can make a variable replacement $\omega ^{'}=\omega -\frac{\omega }{c}v_{z}-\omega_{i}$, and get the expression of Voigt lineshape again: $$\label{eq11}
l_{i}(\omega )=\frac{2}{\pi \sqrt{2\pi }\Gamma \sigma }\int_{-\infty }^{+\infty }\frac{1}{1+(\frac{\omega ^{'}}{ \frac{\Gamma}{2}})^{2}}exp[-\frac{1}{2}(\frac{\Delta- \omega^{'}}{\sigma })^{2}]d\omega ^{'}$$ where $\Delta= \omega-\omega_{i}$.
To obtain the specific expression of total transmission, we can replace $l_{i}(\omega)$ in equation $(6)$ with equation $(11)$ : $$\label{eq12}
\begin{split} T(\omega )=&exp\left \{ \frac{4d^{2}NkL}{\sqrt{2\pi }h\varepsilon _{0}\Gamma \sigma }[ \int_{-\infty }^{+\infty }(S_{32}\cdot exp[-\frac{1}{2}(\frac{\omega-\omega _{32} -\omega ^{'}}{\sigma })^{2}] \right. \\
&\left. + S_{33}\cdot exp[-\frac{1}{2}(\frac{\omega-\omega _{33} -\omega ^{'}}{\sigma })^{2}] +S_{34}\cdot exp[-\frac{1}{2}(\frac{\omega-\omega _{34} -\omega ^{'}}{\sigma })^{2}])\times \frac{1}{1+(\frac{\omega ^{'}}{\frac{\Gamma }{2}})^{2}} d\omega ^{'}] \right \}
\end{split}$$ And we can simplify the fitting equatuion$(12)$ by define fitting parameters $k_{1}$, $k_{2}$ and $k_{3}$, so we can get fitting equation $(13)$. $$\label{eq13}
\begin{split}
T(\omega )=&exp\left \{ k_{1}[ \int_{-\infty }^{+\infty }(S_{32}\cdot exp[-\frac{1}{2}(\frac{\omega-\omega _{32} -\omega ^{'}}{k_{2} })^{2}]+\right. \\
&\left.S_{33}\cdot exp[-\frac{1}{2}(\frac{\omega-\omega _{33} -\omega ^{'}}{k_{2} })^{2}] +S_{34}\cdot exp[-\frac{1}{2}(\frac{\omega-\omega _{34} -\omega ^{'}}{k_{2} })^{2}])\times \frac{1}{1+(\frac{\omega ^{'}}{k_{3}})^{2}} d\omega ^{'}] \right \}
\end{split}$$ where $$\label{eq14}
k_{1}=\frac{4d^{2}NkL}{\sqrt{2\pi }h\varepsilon _{0}\Gamma \sigma }$$ $$\label{eq15}
k_{2}= \sigma$$ $$\label{eq16}
k_{3}=\frac{\Gamma }{2}$$
![**Fitting of transmission rate curve**. The fitting result of transmission rate of the weak probe light with different detuning.[]{data-label="f2"}](figure5.pdf "fig:"){width="0.95\columnwidth"}\
Due to the actual experimental conditions, like inevitable noise and the errors of measurement, the measured transmission maybe a little different from theoretical values. We can add two extra parameters to transform fitting equation $(13)$ into $T(\omega )^{'}$. $$\label{eq17}
T(\omega )^{'}=k_{4}T(\omega )+k_{5}$$ The finally fitting result is showed in Fig s1. The key fitting parameters $k_{1}$, $k_{2}$ and $k_{3}$ are $0.6039$, $189.9895$ and $61.0204$. The unit of parameters $k_{2}$ and $k_{3}$ is $MHz$ in actual experiment. From equation $(14)-(16)$, we can get the relation between number density $N$ and fitting parameters: $$\label{eq17}
N=\frac{\sqrt{2\pi }h\varepsilon _{0}k_{1}k_{2}k_{3}}{2d^{2}kL}$$ where matrix element $d$ take the value of effective far-detuned dipole moment $2.1923\times 10^{-29} C\cdot m$[@Cs_Data], $k=\frac{\omega_{0}}{c}$ and $\omega_{0}=2\pi \cdot 351.726 THz$ is the frequency of cesium $D_{2}$ transition, the length of our cesium cell is $75.3mm$. Finally, the number density of atoms is evaluated as $1.2133\times 10^{18} m^{-3}$.
[99]{}
D. A. Steck, Cesium D Line Data (2003), URL http://steck.us/alkalidata.
P. Siddons, C. S. Adams, C. Ge, and I. G. Hughes, *J. Phys. B* **41**, 155004 (2008).
Rodney Loudon. *The Quantum Theory of Light*
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce the concept of a photonic Dirac monopole, appropriate for photonic crystals, metamaterials and 2D materials, by utilizing the Dirac-Maxwell correspondence. We start by exploring vacuum where the reciprocal momentum space of both Maxwell’s equations and the massless Dirac equation (Weyl equation) possess a magnetic monopole. The critical distinction is the nature of magnetic monopole charges, which are integer valued for photons but half-integer for electrons. This inherent difference is directly tied to the spin and ultimately connects to the bosonic or fermionic behavior. We also show the presence of photonic Dirac strings, which are line singularities in the underlying Berry gauge potential. While the results in vacuum are intuitively expected, our central result is the application of this topological Dirac-Maxwell correspondence to 2D photonic (bosonic) materials, as opposed to conventional electronic (fermionic) materials. Intriguingly, within dispersive matter, the presence of photonic Dirac monopoles is captured by nonlocal quantum Hall conductivity - i.e. a spatiotemporally dispersive gyroelectric constant. For both 2D photonic and electronic media, the nontrivial topological phases emerge in the context of massive particles with broken time-reversal symmetry. However, the bulk dynamics of these bosonic and fermionic Chern insulators are characterized by spin-1 and spin- skyrmions in momentum space, which have fundamentally different interpretations. This is exemplified by their contrasting spin-1 and spin- helically quantized edge states. Our work sheds light on the recently proposed quantum gyroelectric phase of matter and the essential role of photon spin quantization in topological bosonic phases.'
author:
- Todd Van Mechelen
- Zubin Jacob
bibliography:
- 'dirac\_monopole.bib'
title: 'Photonic Dirac monopoles and skyrmions: spin-1 quantization'
---
Introduction
============
Dirac’s pioneering paper [@Dirac60] showed that if magnetic monopoles are found in nature, their magnetic charges $Q$ would be quantized in units of the elementary charge $e$ of the electron, $$2\frac{eQ}{h}\in\mathbb{Z}.$$ $h$ being the Planck constant. This is the earliest example of topological quantization - fundamentally different from second quantization arising in quantum field theories. Although there exists no experimental proof of magnetic monopoles [@Preskill1984] to date, there is ample evidence of quantized topological charges in reciprocal (energy-momentum) space. Specifically, the appearance of such monopoles in the band structure of solids indicates the presence of quantized topological invariants, like the Chern number [@Thouless1982] and $\mathbb{Z}_2$ index [@Kane_2_2005]. Ultimately, experimental observables such as the quantum Hall conductivity can be traced back to the existence of this quantized topological charge [@Xu613; @Laughlin1981].
There have been significant efforts to construct synthetic gauge potentials that mimic these monopole physics in cold atoms [@Zoller2005] and spin ice [@Morris411]. One striking example is the realization of non-Abelian gauge theories with Yang-Lee monopoles [@sugawa_observation_2016]. The topological field theory of light has surfaced in knotted solutions of Maxwell’s equations [@Kedia2013; @Stone2016], as well as the uncertainty relations for photons [@BB2012]. Along side this, there have been important recent developments to formulate topological properties for photons utilizing photonic crystals and metamaterials [@Lu2014; @Rechtsman2013; @Khanikaev2013; @CTChan2016; @Wang2016; @Silveirinha2015; @Haldane2008; @Haldane2008_2; @HafeziM.2013; @ShanuiFan2017; @Shuang2015; @GLYBOVSKI20161; @Papasimakis:10]. The pioneering work in topological photonic crystals has shown the existence of edge states robust to disorder. In the previously explored scenarios, the photonic crystal unit cell is carefully structured to obtain an additional degree of freedom (artificial gauge field) - quite often realized on a graphene-like honeycomb lattice. This approach was first implemented by Haldane for spinless (scalar) electrons in his seminal paper on the parity anomaly [@Haldane1988]. However, it remains an open question whether robust topological photonic edge states can occur in atomic matter. The role of photon spin and its quantization is yet another unresolved problem since previous theories have focused exclusively on pseudo-spin- phenomena [@Lu622; @Chen2016; @Yangeaaq1221; @WangLuyang2016].
Our spin-1 theory [@Hu2018] is fundamentally different in this respect since we do not ignore the polarization (spin) state of the photon, which cannot be neglected for a real gauge (vector) field. In our case, the topological theory is manifestly bosonic as it is connected to the winding of the gauge field itself - not pseudo-spin degrees of freedom. Another fundamental aspect of our theory is the inclusion of dispersion within matter, i.e. frequency and momentum dependence of conductivity, such that topological invariants emerge naturally from the global behavior of optical constants. For example, it has been shown that nonlocal gyrotropic [@VanMechelen2018] and magnetoelectric [@van_mechelen_2017] media will host massless spin-1 quantized edge states with massive-like photons in the bulk. Thus, it is necessary to understand the concept of bosonic Dirac monopoles and the influence of integer spin in topological photonic phases of matter.
In this paper, we elucidate the fundamental difference between the magnetic monopoles appearing in Maxwell’s equations and the Dirac equation. Our work shows that a magnetic monopole appears for both photons and massless fermions in the reciprocal energy-momentum space - even for vacuum. Using a Dirac-Maxwell correspondence, we identify the bosonic and fermionic nature of magnetic monopole charge, which is inherently present in the relativistic theories of both particles. While the results in vacuum are expected, we apply this topological theory to 2D photonic (bosonic) materials, in contrast to conventional electronic (fermionic) materials. The specific 2D photonic materials considered in this paper are gyroelectric which possess antisymmetric components of the conductivity tensor. We exploit the Dirac-Maxwell correspondence to show how dispersive gyroelectric media can support topologically massive particles, which are interpreted as photonic skyrmions. However, the differences in spin between bosons and fermions alter the behavior of these bulk skyrmions as well as their corresponding Chern numbers. We then analyze the unique topological edge states associated with nontrivial spin-1 and spin- skyrmions, which exhibit opposing helical quantization. This clearly shows how the integer and half-integer nature of monopoles is ultimately tied to the differing bosonic and fermionic spin symmetries. Our work sheds light on the recently proposed quantum gyroelectric phase of matter [@VanMechelen2018] which supports unidirectional transverse electro-magnetic (TEM) edge states with open boundary conditions (vanishing fields at the edge) - unlike any known phase of matter till date.
In the context of geometric phases, the concept of magnetic charges has a rich history starting from the pioneering works of Pancharatam, Berry, Chiao and Wu [@chruscinski_geometric_2004]. Unification of these geometric phases for bosons and fermions was shown for massive 3D particles using a relativistic quantum field theory [@BB1987]. In this paper, our focus is massless 3D particles and topologically massive 2D particles [@Shi2018; @Mortensen2018; @Horsley2018], as well as the direct demonstration of gauge discontinuities in Maxwell’s and Weyl’s equations. Our derivation does not utilize quantum field theoretic techniques and appeals only to the spin representation of the two particles. We note that spin quantization is fundamentally different from topological charges encountered in real space for OAM beams [@barnett_natures_2016; @Gawhary2018], polarization singularities [@Doug2014] and polarization vortices [@Shabanov2008]. This is due to the central concept of gauge discontinuity in magnetic monopole quantization, which is related to the topological field theory of bosons and fermions. We function in momentum space of Maxwell’s equations as opposed to real space so our work is specifically suited to develop topological invariants in the band structure of photonic crystals and wave dispersion within metamaterials [@Sihvola2005; @Li2018]. One important application of our current technique is in uncovering unique electromagnetic phases of matter displaying the quantum gyroelectric effect (QGEE) [@VanMechelen2018]. Our unified perspective also sheds light on recent developments of quantized bosonic Hall conductivity [@Lu2012; @Senthil2013; @Metlitski2013; @Vishwanath2013] and topological bosonic phases of matter [@Wen2011; @Chen1604], as opposed to fermionic phases [@Kane2010].
Skyrmions have a storied past in condensed matter - appearing in both real and momentum space of topological systems. In real space, these localized topological defects were first discovered in chiral magnets and quantum Hall ferromagnets but have also been observed in Bose-Einstein condensates and superconductors [@han_skyrmions_2017]. The behavior of these magnetic skyrmions is intimately tied to the Dzyaloshinskii-Moriya (DM) interaction [@Nagaosa2013] which generates the nontrivial winding of the spin structure. In momentum space, skyrmions often characterize the monopoles arising in the band structure of solids and are emergent phenomena in topological insulators and superconductors [@bernevig2013topological]. By contrast, photonic skyrmions are a very recent field of interest. A classical optical analog of skyrmion-like behavior has been reported using surface plasmon polaritons [@Tsesseseaau0227]. This work focuses on photonic skyrmions in momentum space which ultimately govern the equations of motion of a topological electromagnetic field. The physics of these topological fields manifest in nontrivial windings of a spin-1 vector as opposed to a spin- vector.
**Note:** For clarity, all 3D vectors will be denoted with a vector arrow $\vec{A}=(A_x,~A_y,~A_z)$, while we reserve boldface for 2D vectors $\mathbf{A}=(A_x,~A_y)$. The manuscript theme is the Dirac-Maxwell correspondence which directly compares bosonic and fermionic topological field theories. Throughout, the subscript $s=1$ stands for spin-1 photons and the subscript $s=\sfrac{1}{2}$ denotes spin- electrons.
Three dimensions: massless particles
====================================
Dirac-Maxwell correspondence
----------------------------
The correspondence between Dirac’s and Maxwell’s equations is best expressed in the Riemann-Silberstein (R-S) basis [@Bialynicki-Birula2013; @Barnett2014], which utilizes a vector wave function for light. Using this representation, we develop a topological field theory of the vacuum photon. In the R-S basis, we combine the electric $\vec{E}$ and magnetic $\vec{H}$ fields into a complex superposition, $$\vec{\Psi}=\frac{1}{\sqrt{2}}(\vec{E}+i\vec{H}),$$ where $i=\sqrt{-1}$ is the imaginary unit and the electromagnetic fields are associated with plane waves. We strongly emphasize that relativity requires vectorial representations for spin-1 bosonic fields and spinor- representations for fermionic fields. Spin-0 particles constitute scalar fields while spin-2 particles, such as gravitons, are described by tensor fields. Therefore, to unravel the topological bosonic properties of light, we cannot work in a restricted subspace ignoring components of the electromagnetic field. Simultaneously, we do not describe polarizations separately. In the R-S basis, Maxwell’s equations in vacuum can be combined into a first-order wave problem as follows, $$\label{eq:Maxwell}
i\vec{k}\times\vec{\Psi} =H_1\vec{\Psi}=\omega\vec{\Psi},$$ which we label as spin $s=1$. Here, $\omega$ is the frequency of light and we consider dynamical fields over all frequencies and wave vectors, not simple static fields. We can thus unambiguously identify a Hamiltonian for light, $$\label{eq:Spin1}
H_1(\vec{k})=\vec{k}\cdot\vec{S}=k_xS_x+k_yS_y+k_zS_z.$$ $\vec{k}=(k_x,~k_y,~k_z)$ is the momentum of the plane wave in vacuum and $\vec{S}=(S_x,~S_y,~S_z)$ are the set of SO(3) antisymmetric matrices, $$\label{eq:SO3}
S_x=\begin{bmatrix}
0 & 0 & 0 \\
0 & 0 & -i \\
0 & i & 0 \\
\end{bmatrix}, \qquad S_y=\begin{bmatrix}
0 & 0 & i \\
0 & 0 & 0 \\
-i & 0 & 0 \\
\end{bmatrix}, \qquad S_z=\begin{bmatrix}
0 & -i & 0 \\
i & 0 & 0 \\
0 & 0 & 0 \\
\end{bmatrix}.$$ These operators obey the familiar Lie algebra $[S_i,S_j]=i\epsilon_{ijk}S_k$ which encode information about integer spin. Notice our photonic Hamiltonian $H_1=\vec{k}\cdot\vec{S}$ represents optical helicity, i.e. the projection of spin $\vec{S}$ along the direction of momentum $\vec{k}$. This is further clarified on direct comparison with massless Dirac fermions (Weyl fermions), which are the supersymmetric partners of the massless photon [@Stephen2001]. The Weyl equation is expressed as, $$\label{eq:Dirac}
H_{\sfrac{1}{2}}\psi=E\psi,$$ where the massless Dirac Hamiltonian $H_{\sfrac{1}{2}}$, corresponding to spin $s=\sfrac{1}{2}$, is identified with electronic helicity, $$\label{eq:Spin1half}
H_{\sfrac{1}{2}}(\vec{k})=\vec{k}\cdot\vec{\sigma}=k_x\sigma_x+k_y\sigma_y+k_z\sigma_z.$$ $\vec{\sigma}=(\sigma_x,~\sigma_y,~\sigma_z)$ are the Pauli matrices of SU(2) and obey the identical Lie algebra $[\sigma_i,\sigma_j]=2i\epsilon_{ijk}\sigma_k$, $$\label{eq:SU2}
\sigma_x=\begin{bmatrix}
0 & 1\\
1 & 0
\end{bmatrix}, \qquad \sigma_y=\begin{bmatrix}
0 & -i\\
i & 0
\end{bmatrix}, \qquad \sigma_z=\begin{bmatrix}
1 & 0\\
0 & -1
\end{bmatrix}.$$ Both particles are massless and satisfy an analogous helicity equation. However, the critical difference is revealed in the group operations of the particular particle; encapsulated by the SO(3) antisymmetric matrices for the spin-1 photon \[Eq. (\[eq:SO3\])\] and the SU(2) Pauli matrices for the spin- Weyl fermion \[Eq. (\[eq:SU2\])\].
Helical eigenstates
-------------------
We now solve for the eigenstates of the above Hamiltonians. As expected, Maxwell and Weyls’ equations possess two helical degrees of freedom. For the photon \[Eq. (\[eq:Spin1\])\], these are conventional right- and left-handed circular polarization, $$\label{eq:Circular}
H_1\vec{e}_\pm=\pm k ~\vec{e}_\pm, \qquad \vec{e}_\pm(\vec{k})=\frac{1}{\sqrt{2}}(\hat{\theta}\pm i\hat{\phi}),$$ where $\theta$ and $\phi$ are the spherical polar coordinates of $\vec{k}$ and $k=|\vec{k}|$ is the magnitude of the wave vector. The photon is massless and therefore linearly dispersing in vacuum $\omega_\pm =\pm k$. Similarly, the eigenstates of the Weyl equation \[Eq. (\[eq:Spin1half\])\] are comprised of two massless helical spinors, which are represented as, $$\label{eq:Spinor}
H_{\sfrac{1}{2}}\psi_\pm = \pm k~\psi_\pm, \qquad \psi_+(\vec{k})=\begin{bmatrix}
\cos (\theta/2) \\
\sin (\theta/2)e^{i\phi}
\end{bmatrix},\qquad \psi_-(\vec{k})=\begin{bmatrix}
\sin (\theta/2) \\
-\cos (\theta/2)e^{i\phi}
\end{bmatrix}.$$ Indeed, these states are also linearly dispersing $E_\pm=\pm k$. An important observation can be made in $\vec{e}_\pm$ and $\psi_\pm$. The eigenstates are ill-defined at the origin of the momentum space $\vec{k}=0$, since they are arbitrarily dependent on $\theta$ and $\phi$ at this point. In fact, by parameterizing $\theta$ as the inclination from $k_z$, the eigenstates are not well-behaved at the north $\theta=0$ or south $\theta=\pi$ poles either - they are multivalued at both points. Such discontinuous behavior is impossible to remove and results from choosing a particular gauge for the eigenstates. This is the underlying source for Dirac monopoles and strings. The linear dispersion (light cone) of the massless helical states is displayed in Fig. \[fig:LinearDispersion\](a).
![(a) Linear dispersion (light cone) of the 3D massless photon and electron $\omega_\pm=E_\pm=\pm k$. At the origin of the momentum space $\vec{k}=0$ sits a magnetic monopole with quantized charge. This singularity is often called a Weyl point and is quantized to the spin of the particle $Q_s=s$. Integer and half-integer spin quantization is connected to bosonic and fermionic statistics respectively. (b) Dirac monopoles (Berry curvature) $\vec{F}_s=\vec{\nabla}_k\times\vec{A}_s$ of the massless electron $Q_{\sfrac{1}{2}}=\sfrac{1}{2}$ and photon $Q_1=1$ in momentum space. The monopole charge acts as a source for the magnetic field $\vec{\nabla}_{k}\cdot\vec{F}_s=4\pi Q_s\delta^3(\vec{k})$ and arises due to the discontinuous behavior in the spin eigenstates. Notice that the flux through any surface enclosing the monopole is necessarily quantized $Q_s=(4\pi)^{-1}\oiint \vec{F}_s\cdot d^2\vec{k}$. This monopole is accompanied by a string of singularities in the underlying gauge potential $\vec{A}_s$. Any closed path around the equator of the string produces a quantized Berry phase $\gamma_{s}=\oint\vec{A}_s\cdot d\vec{k}= 2\pi Q_s$. The accumulated phase in $\vec{k}$-space is fundamentally tied to the spin of the particle $\mathcal{R}_s(2\pi)=\exp(i\gamma_s)=(-1)^{2s}$.[]{data-label="fig:LinearDispersion"}](linear_dispersion.png){width="\linewidth"}
Spin quantization in photonic Dirac monopoles and strings
---------------------------------------------------------
In vacuum $\vec{k}$-space, we discover a magnetic Dirac monopole for both Maxwell’s and Weyl’s equations but with intrinsic differences. This is demonstrated by first defining the magnetic flux in momentum space - i.e. the Berry curvature. For the photon, the Berry curvature of either right- or left-handed helicity can be found from the circular eigenstates derived in Eq. (\[eq:Circular\]), $$\vec{F}_1^\pm=-i\vec{\nabla}_{k}\times[{\vec{e}_\pm}^*\cdot(\vec{\nabla}_{k}~\vec{e}_\pm)].$$ For the massless electron, the analogous Berry curvature is found by evaluating the spinor eigenstates in Eq. (\[eq:Spinor\]), $$\vec{F}_{\sfrac{1}{2}}^\pm=-i\vec{\nabla}_{k}\times[\psi_\pm^\dagger\vec{\nabla}_{k}~\psi_\pm].$$ Here, $\vec{\nabla}_{k}=\sum_j \hat{j}~\partial_{k_j}$ is the gradient operator in 3D momentum space. Note that the Berry curvature is a vector in three dimensions but a scalar in two dimensions. On evaluating the Berry curvature for both particles with positive and negative helicities ($\pm$), we find that $\vec{F}_s^\pm=\pm \vec{F}_s$ possesses a Dirac monopole, $$\label{eq:DiracMonopole}
\vec{F}_s= Q_s\vec{F}.$$ $\vec{F}$ being the magnetic field of a Dirac monopole in $\vec{k}$-space, $$\vec{F}(\vec{k})=\frac{\vec{k}}{k^3}.$$ Note that $Q_s$ in Eq. (\[eq:DiracMonopole\]) is the topological magnetic charge which generates the magnetic field. This quantity is fundamentally different for the two particles, $$Q_s= s.$$ $s$ is precisely the spin of the particle, which takes integer $s=1$ or half-integer $s=\sfrac{1}{2}$ values for bosons or fermions respectively. We emphasize that the magnetic monopole charge is naturally quantized, $$Q_s=\frac{1}{4\pi}\oiint \vec{F}_s\cdot d^2\vec{k}.$$ The charge is located at the origin $\vec{k}=0$ of the momentum space, exactly where the eigenstates are ill-defined, and acts as a source for the magnetic field $\vec{\nabla}_{k}\cdot\vec{F}_s=4\pi Q_s\delta^3(\vec{k})$. Notice that the magnetic monopole charge of the photon, $$Q_1=2Q_{\sfrac{1}{2}}=1,$$ is exactly twice the electron due to integer spin. The monopole charge for each helicity has opposite signs $Q_s^\pm=\pm Q_s$. This ensures the net charge vanishes $Q^+_s+Q^-_s=0$ at the origin $\vec{k}=0$; as expected due to time-reversal symmetry in vacuum [@Young2012]. A visualization of the magnetic flux is shown in Fig. \[fig:LinearDispersion\](b).
We note that the photonic Dirac monopole is accompanied by a string of singularities in the underlying gauge potential. This Dirac string is unobservable as it is a gauge dependent phenomenon but sheds light on the fundamental differences between electrons and photons. The Berry gauge potential for the massless photon and electron can be evaluated using the eigenstates in Eq. (\[eq:Circular\]) and (\[eq:Spinor\]) respectively, $$\vec{A}_1^\pm=-i{\vec{e}_\pm}^*\cdot(\vec{\nabla}_{k}~\vec{e}_\pm), \qquad \vec{A}_{\sfrac{1}{2}}^\pm=-i\psi_\pm^\dagger\vec{\nabla}_{k}~\psi_\pm.$$ Upon solving for $\vec{A}_s^\pm =\pm \vec{A}_s$, we again find a clear dependence on the magnetic monopole charge $Q_s$ which is different for bosons and fermions, $$\label{eq:DiracString}
\vec{A}_s(\vec{k})=Q_s\frac{1-\cos\theta}{k\sin\theta}\hat{\phi},$$ and $\vec{F}_s=\vec{\nabla}_{k}\times\vec{A}_s$ reproduces the Berry curvature in Eq. (\[eq:DiracMonopole\]). The gauge potential is singular along the $k_z$-axis, at $\theta=0$ and $\pi$, where the eigenstates are multivalued. This line singularity that originates at the monopole and extends to infinity is known as a Dirac string. Fig. \[fig:LinearDispersion\](b) displays a visualization of the Dirac monopole and strings for both massless particles. We note that the above equations are traditionally found in the theory of magnetic charges in real space [@Preskill1984] - not momentum space. Following this, quantization of magnetic charge naturally emerges from the requirement of a single-valued wave function in the presence of singular (multivalued) gauge potentials. Our rigorous derivation is unique as it unifies the momentum space of Maxwell’s equations and the Weyl equation. This makes it ideally suited for extension to topological theories of band structure in photonic crystals and wave dispersion in metamaterials.
Berry phase
-----------
We now provide a detailed comparison of $\vec{k}$-space Pancharatnam-Berry phase (hereon called geometric phase) for photons and electrons, that arises from their corresponding spin properties. The geometric phase calculated for any closed path on the $\vec{k}$-sphere is gauge invariant, $$\gamma_s=\oint \vec{A}_s\cdot d\vec{k}=\iint \vec{F}_s\cdot d^2\vec{k}.$$ $\gamma_s$ is the geometric phase and is equivalent to the flux of Berry curvature $\vec{F}_s$ through a surface bounded by the path. In this case, we see that $\iint \vec{F}_s\cdot d^2\vec{k}=Q_s\iint d\Omega$ is exactly the solid angle $\Omega(C)$ traced along the $\vec{k}$-sphere, $$\gamma_s=Q_s\Omega(C),$$ where $C$ designates the bounded path. We now consider a closed path around a great circle of the $\vec{k}$-sphere (eg: the equatorial path $k_z=0$), which encloses the monopole. For massless particles, this is equivalent to rotating the fields back into themselves. The accumulated phase must be quantized, $$\label{eq:GeometricPhase}
\gamma_s=2\pi Q_s.$$ This is the momentum space manifestation of Dirac’s quantization condition $2Q_s\in \mathbb{Z}$ which ensures the massless particles acquire the same phase under a $2\pi$ or $-2\pi$ rotation. We clearly see that geometric phases in $\vec{k}$-space are dependent on the spin of the particle, $$\exp(i\gamma_s)=(-1)^{2Q_s}.$$ Notice that $\exp\left(i\gamma_{\sfrac{1}{2}}\right)=-1$ and $\exp\left(i\gamma_{1}\right)=+1$ are antisymmetric or symmetric under a $\pm 2\pi$ rotation depending on the spin $Q_s=s$. Ultimately, the geometric phase of $\gamma_{\sfrac{1}{2}}=\pi$ or $\gamma_1=2\pi$ is tied to the fermionic or bosonic statistics of the particle. We note that this geometric phase $\gamma_{\sfrac{1}{2}}=\pi$ is routinely encountered for massless Dirac fermions in graphene [@Mazrzari2011; @Novoselov2006]. However, the direct correspondence with spin-1 massless photons $\gamma_1=2\pi$ has not been pointed out to date. Our results suggest that a thin wire supporting Dirac fermions would yield Chiao-Tomita phases [@Chiao1986] exactly half the value of photons. We also note that spin-momentum locking is a universal property in photonics [@VanMechelen2016; @kalhor2016universal; @Pendharker:18] which arises entirely from the transversality $\vec{k}\cdot\vec{\Psi}=0$ of electromagnetic waves in vacuum. This phenomenon can be explained with causal boundary conditions on evanescent fields and does not necessarily require topological considerations [@Bliokh1448]. For example, conventional surface plasmon polaritons (SPPs) and waveguide modes show spin-momentum locking but these are not related to any topologically protected edge states or nontrivial phases.
Rotational symmetries
---------------------
The nuance behind integer and half-integer geometric phases \[Eq. (\[eq:GeometricPhase\])\] is explained more rigorously by considering the operations of the rotational (spin) groups. Maxwell’s equations \[Eq. (\[eq:Maxwell\])\] transform under the SO(3) group $\mathcal{R}_1(\alpha)=\exp\left(i\alpha\hat{n}\cdot\vec{S}\right)$, where $\alpha$ is the angle subtended about an axis $\hat{n}$. This is true for all vector fields. Conversely, the Weyl equation \[Eq. (\[eq:Dirac\])\] transforms under the SU(2) group $\mathcal{R}_{\sfrac{1}{2}}(\alpha)=\exp\left(i\alpha\hat{n}\cdot\vec{\sigma}/2\right)$, characteristic of spinors. Although SO(3) and SU(2) obey the same Lie algebra, the group representations are inequivalent. The distinction is evident under a cyclic rotation, $$\mathcal{R}_s(2\pi)=(-1)^{2s}.$$ Notice that the accumulated phase is different depending on the particle species. This is due to the fact that fermions are antisymmetric $\mathcal{R}_{\sfrac{1}{2}}(2\pi)=-1$ under rotations, while bosons are symmetric $\mathcal{R}_1(2\pi)=+1$ and this behavior is guaranteed by the spin-statistics theorem [@Pauli1940]. The difference fundamentally changes the interpretation of fermionic and bosonic topologies [@Regnault_2013].
Two dimensions: topologically massive particles
===============================================
Dirac-Maxwell correspondence
----------------------------
Up to this point, we have only considered the 3D dynamics of the vacuum photon and its analogies with the Weyl fermion. Now we shift to the 2D domain to harness these topological properties and elucidate the fundamental role of spin in nontrivial phases of matter. Nontrivial 2D materials are characterized by an integer topological invariant - the Chern number $C\in\mathbb{Z}$. In electronics, these materials are often called Chern insulators [@Jotzu2014] because they are insulating in the bulk but host metallic one-way edge states that are robust to disorder. In the long wavelength limit $k\approx 0$, the simplest fermionic Chern insulator is described by the 2D Dirac equation [@bernevig2013topological], $$\label{eq:FermionChern}
\mathcal{H}_{\sfrac{1}{2}}\psi=E\psi, \qquad \mathcal{H}_{\sfrac{1}{2}}(\mathbf{k})=v(k_x\sigma_x+k_y\sigma_y)+\Lambda(k)\sigma_z.$$ Equation (\[eq:FermionChern\]) is essentially identical to the Weyl equation \[Eq. (\[eq:Spin1half\])\] except we have replaced the $z$-component of the momentum with a Dirac mass $k_z\to\Lambda(k)$. We have also introduced the Fermi velocity $v$ to characterize the effective speed of electrons within the material. It is easy to check that $\Lambda(k)$ breaks time-reversal symmetry but preserves rotational symmetry about the $z$-axis, $\mathcal{R}_{\sfrac{1}{2}}(\alpha)=\exp\left(i\alpha\sigma_z/2\right)$. The meat of the topological physics lies in this spatially dispersive Dirac mass [@shen2011topological], $$\Lambda(k)=\Lambda_0-\Lambda_2k^2.$$ $\Lambda_0=\Lambda(0)$ opens a band gap and $\Lambda_2$ accounts for the curvature of the energy bands. Importantly, when $\Lambda_0\Lambda_2>0$ there is so-called *band inversion* [@Zhu2012] and the effective mass changes sign within the dispersion $\Lambda(k_i)=0$, precisely at $k_i=\sqrt{\Lambda_0/\Lambda_2}$. The quadratic momentum dependence $k^2=\mathbf{k}\cdot\mathbf{k}$ is also crucial to regularize the long wavelength theory [@VanMechelen2018; @Ryu2010]. This means we can project the planar momentum space onto the surface of the Riemann sphere $\mathbb{R}^2\simeq S^2$, a necessary constraint for continuum topological field theories.
We now study the equivalent 2D dynamics of the photon - the bosonic Chern insulator. As anticipated, the 2D Maxwell theory is the supersymmetric partner of the 2D Dirac theory [@VanMechelen2018; @Dunne1999] and takes an analogous form, $$\label{eq:BosonChern}
\mathcal{H}_{1}\vec{\Psi}=\omega\vec{\Psi}, \qquad \mathcal{H}_{1}(\mathbf{k})=v(k_xS_x+k_yS_y)+\Lambda(k)S_z.$$ Equation (\[eq:BosonChern\]) is formally equivalent to the 3D Maxwell equation \[Eq. (\[eq:Spin1\])\] with the substitution of a mass term $k_z\to\Lambda(k)$. Here, $v=1/\sqrt{\varepsilon}$ is the effective speed of light which is governed by the dielectric permittivity $\varepsilon>1$. Like the Dirac equation \[Eq. (\[eq:FermionChern\])\], time-reversal symmetry is broken but rotational symmetry is preserved about the $z$-axis, $\mathcal{R}_1(\alpha)=\exp\left(i\alpha S_z\right)$. There is one caveat however; the photonic wave function $\vec{\Psi}$ is slightly altered since we only retain transverse-magnetic (TM) waves in two dimensions, $$\vec{\Psi}=\frac{1}{\sqrt{2}}\left(\sqrt{\varepsilon}E_x, ~\sqrt{\varepsilon}E_y, ~i H_z\right)=\frac{1}{\sqrt{2}}\left(\sqrt{\varepsilon}\mathbf{E},~ i H_z\right).$$ The transverse-electric (TE) component cannot couple to a 2D material as all electrical currents lie in the $x$-$y$ plane. Nevertheless, the underlying topological physics remain unchanged.
Dispersive transverse conductivity
----------------------------------
Our central result is that the above mentioned Maxwell Hamiltonian can possess a mass term arising from dispersion of optical constants. Still, one might question the seemingly ad hoc insertion of a photonic mass $\Lambda(k)$ for two reasons: 1. *Are Maxwell’s equations still gauge invariant?* 2. *Does this mass have any physical origin?* The answer is *yes* to both [@Dunne1999; @BOYANOVSKY1986483]. In fact, it is nothing but the Hall conductivity [@Hall1879], $$\varepsilon\Lambda(k)=\sigma_H(k)=\sigma_0-\sigma_2k^2.$$ Remarkably, our result shows that the Hall conductivity for 2D photons plays the exact same role as the Dirac mass for 2D electrons. We note that the Hall conductivity is related to the anti-symmetric components of the conductivity tensor. $\sigma_0=\sigma_H(0)$ is the conventional static (DC) component which opens a band gap in the vacuum dispersion. This property of low energy bandgap is fundamentally similar to the role of the Dirac mass for fermions. $\sigma_2$ is the nonlocal (momentum dependent) correction to $\sigma_H$ which dictates the curvature of the photonic bands. Until very recently, the momentum dependence of $\sigma_H$ had never been considered for topological purposes [@VanMechelen2018]. This type of behavior can also be generalized to its high-frequency (AC) equivalent in the context of nonlocal gyrotropy, but we restrict ourselves to the low-energy limit $\omega\approx 0$ for simplicity. In this limit, nonlocal Hall conductivity defines the quantum gyroelectric phase of matter.
![Spin expectation value $\hat{\mathcal{M}}_z(k)$ as a function of $k$. (a) $N=0$ skyrmion with no band inversion $\Lambda_0\Lambda_2<0$. The spin returns to initial state $\hat{\mathcal{M}}_z(0)=\hat{\mathcal{M}}_z(\infty)$ and the total winding is trivial. (b) $N=1$ skyrmion with band inversion $\Lambda_0\Lambda_2>0$. In this case, the spin flips direction $\hat{\mathcal{M}}_z(0)\neq\hat{\mathcal{M}}_z(\infty)$ and the total winding is nontrivial. $k_i$ labels the band inversion point where $\hat{\mathcal{M}}_z(k_i)=0$ passes through zero. This point must occur for the spin to flip directions and can only be removed at a topological phase transition.[]{data-label="fig:spin_texture"}](spin_texture.png){width="\linewidth"}
Spin-1 photonic skyrmions
-------------------------
The electronic \[Eq. (\[eq:FermionChern\])\] and photonic \[Eq. (\[eq:BosonChern\])\] Hamiltonians can be written in a more suggestive form by introducing the skyrmion spin vector $\vec{\mathcal{M}}=(\mathcal{M}_x,~\mathcal{M}_y,~\mathcal{M}_z)$, $$\label{eq:SkyrmionVector}
\mathcal{H}_1=\vec{\mathcal{M}}\cdot\vec{S}, \qquad \mathcal{H}_{\sfrac{1}{2}}=\vec{\mathcal{M}}\cdot\vec{\sigma}.$$ As we can see, this new vector $\vec{\mathcal{M}}$ has replaced the original 3D wave vector $\vec{k}$ in the massless equations and closely resembles the Zeeman interaction [@han_skyrmions_2017]. Indeed, the spin precesses about an axis formed by $\vec{\mathcal{M}}$, $$\dot{\vec{S}}=-i[\vec{S},\mathcal{H}_1]=g_1\vec{\mathcal{M}}\times\vec{S}.$$ It is important to reiterate that $\vec{S}$ represents spin-1 operators while $\vec{\sigma}$ is spin-. This is exemplified by the fact that bosonic (vector) particles possess gyromagnetic $g$-factors of $g_{1}=(Q_1)^{-1}=1$, while fermionic (spinor) particles have $g$-factors of $g_{\sfrac{1}{2}}=(Q_{\sfrac{1}{2}})^{-1}=2$, $$\dot{\vec{\sigma}}=-i[\vec{\sigma},\mathcal{H}_{\sfrac{1}{2}}]=g_{\sfrac{1}{2}}\vec{\mathcal{M}}\times\vec{\sigma}.$$ The Larmor frequency $\Omega_s$ is fundamentally different between the two. The skyrmions precess at different rates depending on the spin representation, $$\Omega_s=g_s\mathcal{M}, \qquad g_s=(Q_s)^{-1},$$ where $\mathcal{M}=|\vec{\mathcal{M}}|$ is the magnitude of the skyrmion vector.
Note though, the skyrmion vector $\vec{\mathcal{M}}(\mathbf{k})$ is a function of a 2D momentum $\mathbf{k}$ and actually describes a *parametric surface* $\vec{\mathcal{M}}(\mathbf{k})=\left(vk_x,~vk_y,~\Lambda(k)\right)=\left(v\mathbf{k},~\Lambda(k)\right)$. The eigenstates assume an identical form with the substitution of $\vec{k}\to\vec{\mathcal{M}}$, $$\label{eq:Skyrmion}
\mathcal{H}_1\vec{e}_\pm = \pm \mathcal{M}~\vec{e}_\pm, \qquad \mathcal{H}_{\sfrac{1}{2}}\psi_\pm = \pm \mathcal{M}~\psi_\pm.$$ $\vec{e}_\pm$ are the right- and left-handed helical eigenstates derived in Eq. (\[eq:Circular\]) and $\psi_\pm$ are the equivalent spinors in Eq. (\[eq:Spinor\]). The dispersion relation for each of the eigenstates reads, $$\omega_\pm(k)=E_\pm(k)=\pm \mathcal{M}(k)=\pm\sqrt{v^2k^2+\Lambda^2(k)},$$ which are clearly gapped since $\mathcal{M}(0)=|\Lambda_0|$. These states have acquired mass in 2D. The critical difference of these new eigenstates is that the polar coordinate $\theta$ no longer parametrizes the inclination from $k_z$. Instead, it is governed by the spatially dispersive mass $\tan\theta(k)=vk/\Lambda(k)$, which is a function of the *in-plane* momentum $\mathbf{k}$. We can understand this phenomenon more clearly by evaluating the spin expectation value along the $\hat{z}$ direction, $$\langle S_z\rangle_\pm={\vec{e}_\pm}^*\cdot S_z\vec{e}_\pm=\pm Q_1\hat{\mathcal{M}}_z,$$ where $\hat{\mathcal{M}}_z(k)=\Lambda(k)/\mathcal{M}(k)=\cos\theta(k)$ is a normalized vector. Notice the spin comes in units of bosonic charge $Q_1=1$, as we would expect for an integer particle $s=1$. Analogously, the half-integer skyrmion $s=\sfrac{1}{2}$ arises in units of fermionic charge $Q_{\sfrac{1}{2}}=\sfrac{1}{2}$, $$\left\langle \sigma_z/2\right\rangle_\pm=\psi^\dagger_\pm (\sigma_z/2)\psi_\pm=\pm Q_{\sfrac{1}{2}}\hat{\mathcal{M}}_z.$$ At $k=0$, the spin points directly along $\hat{\mathcal{M}}_z(0)=\textrm{sgn}[\Lambda_0]$. However, as the momentum increases, $\vec{\mathcal{M}}$ tilts away from the $z$-axis and in some cases can flip directions entirely $\hat{\mathcal{M}}_z(\infty)=-\textrm{sgn}[\Lambda_2]$. This is a nontrivial topology. A depiction of trivial and nontrivial $\hat{\mathcal{M}}_z(k)$ as a function of $k$ is presented in Fig. \[fig:spin\_texture\].
![Left: spin texture $\hat{\mathcal{M}}(\mathbf{k})$ as a function of $\mathbf{k}$ for trivial and nontrivial skyrmions. (a) $N=0$ skyrmion with no band inversion $\Lambda_0\Lambda_2<0$. As an example, we have let $v=0.5$, $\Lambda_0=4$ and $\Lambda_2=-2$. Since the spin returns to initial state within the dispersion $\hat{\mathcal{M}}_z(0)=\hat{\mathcal{M}}_z(\infty)$, the total winding is trivial. (b) $N=1$ skyrmion with band inversion $\Lambda_0\Lambda_2>0$. To demonstrate, we have let $v=0.5$, $\Lambda_0=4$ and $\Lambda_2=2$. In this case, the spin flips direction within the dispersion $\hat{\mathcal{M}}_z(0)\neq\hat{\mathcal{M}}_z(\infty)$ and the total winding is nontrivial. Right: spin texture $\hat{\mathcal{M}}$ of the skyrmion projected on the unit sphere. As the momentum varies over all possible values, $\hat{\mathcal{M}}(\mathbf{k})$ can perform either a (c) retracted or (d) full evolution over the unit sphere. This corresponds to a total solid angle of $\Omega=0$ or $4\pi$ respectively.[]{data-label="fig:spin"}](spin.png){width="\linewidth"}
### **An aside: the zero helicity (longitudinal) state** {#an-aside-the-zero-helicity-longitudinal-state .unnumbered}
For completeness, there is technically one additional eigenstate associated with the photonic Hamiltonian \[Eq. (\[eq:BosonChern\])\] - the zero helicity (longitudinal) state, $$\mathcal{H}_1\vec{e}_0=0, \qquad \vec{e}_0=\hat{\mathcal{M}}=\vec{\mathcal{M}}/\mathcal{M}.$$ $\vec{e}_0$ is a completely flat band $\omega_0=0$ and represents the electrostatic limit (irrotational fields). This band belongs to the Hilbert space but can be removed from the spectrum by enforcing $\mathbf{k}\cdot\mathbf{D}=0$ at zero frequency, which implies there is no static charge present. Moreover, since $\vec{e}_0={\vec{e}_0}^*$ can always be chosen real, the Chern number of this band necessarily vanishes $C_0=0$.
Skyrmion magnetic field
-----------------------
We are now ready to assess the Berry curvature. In two dimensions, the Berry curvature is a scalar and characterizes the “magnetic” flux through the planar momentum space $\mathbb{R}^2$. Since our long wavelength theory is regularized, this is equivalent to the flux through the Riemann sphere $S^2$. For the 2D photon, the Berry curvature $\mathcal{F}_1^\pm$ is found by varying the in-plane momentum $\mathbf{k}$ of the right- and left-handed eigenstates $\vec{e}_\pm$, $$\mathcal{F}^\pm_1=-i(\partial_{k_x}{\vec{e}_\pm}^*\cdot\partial_{k_y}\vec{e}_\pm-\partial_{k_y}{\vec{e}_\pm}^*\cdot\partial_{k_x}\vec{e}_\pm).$$ The Berry curvature $\mathcal{F}_{\sfrac{1}{2}}^\pm$ of the 2D electron $\psi_\pm$ is derived in a similar fashion, $$\mathcal{F}^\pm_{\sfrac{1}{2}}=-i(\partial_{k_x}\psi^\dagger_\pm\partial_{k_y}\psi_\pm-\partial_{k_y}\psi^\dagger_\pm\partial_{k_x}\psi_\pm).$$ Just like the 3D massless particles \[Eq. (\[eq:DiracMonopole\])\], the Berry curvature $\mathcal{F}^\pm_{s}=\pm \mathcal{F}_s$ comes in units of quantized magnetic charge $Q_s=s$, $$\mathcal{F}_{s}=Q_s\mathcal{F}.$$ This emergent magnetic field $\mathcal{F}$ is generated by the momentum dependent variations in the spin texture $\hat{\mathcal{M}}=\vec{\mathcal{M}}/\mathcal{M}$, $$\mathcal{F}=\hat{\mathcal{M}}\cdot(\partial_{k_x}\hat{\mathcal{M}}\times\partial_{k_y}\hat{\mathcal{M}})=\vec{F}\cdot d^2 \vec{\mathcal{M}}.$$ $\mathcal{F}$ is precisely the magnetic field of a skyrmion [@Nagaosa2013] and has several profound interpretations. Mathematically, its the Jacobian and dictates the degree of continuous mapping from the momentum space (the Riemann sphere) onto the unit sphere $\hat{\mathcal{M}}$, i.e. $S^2\to S^2$. In another context, it tells us the differential flux of the Dirac monopole $\vec{F}$ onto the parametric surface $\vec{\mathcal{M}}(\mathbf{k})$, $$\vec{F}=\frac{\vec{\mathcal{M}}}{\mathcal{M}^3}, \qquad d^2 \vec{\mathcal{M}}=\partial_{k_x}\vec{\mathcal{M}}\times\partial_{k_y}\vec{\mathcal{M}},$$ where $d^2\vec{\mathcal{M}}$ is the surface normal. As the momentum varies over all possible values, the spin vector $\vec{\mathcal{M}}(\mathbf{k})$ can enclose the monopole any number of times. *Hence, the total magnetic flux counts the number of $Q_s$ monopoles enclosed by the skyrmion spin vector $\vec{\mathcal{M}}$*, $$N=\frac{1}{4\pi}\iint_{\mathbb{R}^2}\mathcal{F}dk_xdk_y=\frac{1}{4\pi}\iint_{\mathbb{R}^2}\hat{\mathcal{M}}\cdot(\partial_{k_x}\hat{\mathcal{M}}\times\partial_{k_y}\hat{\mathcal{M}})dk_xdk_y, \qquad N\in \mathbb{Z}.$$ This is known as the skyrmion (or winding) number. Since the momentum space is bounded on the Riemann sphere $\mathbb{R}^2\simeq S^2$, the skyrmion number $N$ is guaranteed to be an integer. A visualization of the $\hat{\mathcal{M}}$ unit sphere for trivial and nontrivial skyrmions is displayed in Fig. \[fig:spin\].
Chern insulators
----------------
The Chern number $C_s$ is directly proportional to the skyrmion number $N$ but has a very different meaning depending on the particle species. It counts twice the total magnetic charge of the skyrmion, $$C_s=\frac{1}{2\pi}\iint_{\mathbb{R}^2}\mathcal{F}_sdk_xdk_y=\frac{Q_s}{2\pi}\iint_{\mathbb{R}^2}\mathcal{F}dk_xdk_y=2Q_sN.$$ For spin- skyrmions, the Chern number is an integer $C_{\sfrac{1}{2}}=N\in \mathbb{Z}$ and is indistinguishable from the skyrmion number itself. Spin-1 skyrmions are quite different by comparison; the Chern number is an *even* integer $C_1=2N\in 2\mathbb{Z}$. From fermionic Chern arguments, one would expect to always find an even number of photonic edge states - but this is not the case [@Lu2012; @Senthil2013; @Metlitski2013; @Vishwanath2013]. Although a widely held belief, the conventional bulk-boundary correspondence fails for spin-1 bosons [@Tiwari2018]. We will demonstrate this fact explicitly.
Utilizing our spin vector $\vec{\mathcal{M}}$ defined in Eq. (\[eq:SkyrmionVector\]), the skyrmion magnetic field $\mathcal{F}$ in circular polar coordinates $\mathbf{k}=k(\cos\phi,~\sin\phi)$ reads, $$\mathcal{F}(k)=\frac{vk[\Lambda(k)-vk\Lambda'(k)]}{[v^2k^2+\Lambda^2(k)]^{3/2}}=\sin\theta(k)\partial_k\theta(k)=-\partial_k\hat{\mathcal{M}}_z(k).$$ Due to rotational symmetry, $\mathcal{F}(k)$ depends only on the magnitude of $k$. The geometric interpretation is clear - it describes variations in the solid angle $\mathcal{F}(k)dk d\phi=d\Omega(\mathbf{k})$ traced by $\hat{\mathcal{M}}(\mathbf{k})$. Integrating the magnetic flux over all momenta, we acquire the skyrmion number, $$N=\frac{1}{2}[\cos\theta(0)-\cos\theta(\infty)]=\frac{1}{2}[\hat{\mathcal{M}}_z(0)-\hat{\mathcal{M}}_z(\infty)]=\frac{1}{2}\left(\mathrm{sgn}[\Lambda_0]+\mathrm{sgn}[\Lambda_2]\right).$$ When band inversion is present $\Lambda_0\Lambda_2>0$, the $z$-component of the spin vector $\hat{\mathcal{M}}_z(0)\neq\hat{\mathcal{M}}_z(\infty)$ flips directions within the dispersion. This represents north $\theta=0$ and south $\theta=\pi$ poles on the unit sphere, which means $\hat{\mathcal{M}}(\mathbf{k})$ traces out a full solid angle, regardless of the relative magnitudes of $\Lambda_0$ and $\Lambda_2$. This is equivalent to saying the parametric surface $\vec{\mathcal{M}}(\mathbf{k})$ always encloses a monopole $N=\pm 1$. In the trivial regime $\Lambda_0\Lambda_2<0$, the $z$-component $\hat{\mathcal{M}}_z(0)=\hat{\mathcal{M}}_z(\infty)$ returns to its initial state at either the north or south poles and $\vec{\mathcal{M}}(\mathbf{k})$ never encloses a monopole $N=0$. Consequently, the Chern number in the nontrivial phase equates to $C_s=\pm 2Q_s$, which is an integer for the electron $C_{\sfrac{1}{2}}=\pm 1$, but an even integer for the photon $C_1=\pm 2$. In conventional spin- and pseudo-spin- problems, a large Chern number $|C_{\sfrac{1}{2}}|>1$ corresponds to multiple gapless edge states within the bulk topological band gap. This is not true for spin-1 bosonic particles. For $|C_1|=2$ there is a *single* spin-1 quantized edge state within the topological band gap which is illustrated in Fig. (\[fig:dispersion\]).
![Dispersion relation of the bulk and gapless edge bands (black lines) of the topologically massive 2D particles. (a) The conventional fermionic Chern insulator is characterized by a spin- skyrmion (Dirac equation). (b) The bosonic Chern insulator is described by a spin-1 skyrmion (Maxwell’s equations). The bulk Chern number $C_s=2Q_sN$ depends on both the magnetic charge (spin) $Q_s=s$ and the skyrmion number $N\in\mathbb{Z}$. This corresponds to integer phases for electrons $C_{\sfrac{1}{2}}\in\mathbb{Z}$ but *even* integer phases for photons $C_1\in 2\mathbb{Z}$. At low energy, a band gap is formed at $E=\omega=0$ by a spatially dispersive effective mass $\Lambda(k)=\Lambda_0-\Lambda_2k^2$. (a) For the 2D electron, this is simply the Dirac mass. (b) For the 2D photon, this mass is equivalent to a nonlocal Hall conductivity $\varepsilon\Lambda(k)=\sigma_H(k)$. In the presence of band inversion $\Lambda_0\Lambda_2>0$, there is a point where the effective mass changes sign $\Lambda(k_i)=0$, precisely at $k_i=\sqrt{\Lambda_0/\Lambda_2}$. The massless helically quantized edge states touch the bulk bands at this point. This is known as the quantum anomalous Hall effect (QAHE) for electrons and the quantum gyroelectric effect (QGEE) for photons [@VanMechelen2018]. The flat longitudinal band $\omega_0=0$ is shown for completeness and represents the electrostatic limit (irrotational fields). However, this band can be removed from the spectrum by requiring that all static charges vanish.[]{data-label="fig:dispersion"}](dispersion.png){width="\linewidth"}
Topological edge states {#subsec:Edge}
-----------------------
We now solve for the topology protected edge states of both particles. We stress that for both spin-1 and spin- phases, there is exactly one unidirectional solution at the edge. This makes intuitive sense because a single monopole $N=\pm 1$ exists in the band structure. A nontrivial skyrmion $N=\pm 1$ corresponds to either a forward or backward propagating edge state - forward for $N=+1$ and backward for $N=-1$. We take the boundary in the $x$ dimension such that $k_y$ is still a good quantum number. We then look for solutions of the form $\vec{\Psi}^\textrm{e}_\pm(x,y)=\vec{\Psi}^\textrm{e}_\pm(x)e^{ik_y y}$ and $\psi^\textrm{e}_\pm(x,y)=\psi^\textrm{e}_\pm(x)e^{ik_y y}$ that satisfy the boundary condition at infinity $\vec{\Psi}^\textrm{e}_\pm(x=+\infty)=\psi^\textrm{e}_\pm(x=+\infty)=0$. We also impose topological *open boundary conditions* [@Hatsugai1993; @avila_topological_2013] at the interface, $$\vec{\Psi}^\textrm{e}_\pm(x=0^+)=\psi^\textrm{e}_\pm(x=0^+)=0.$$ If this constraint is satisfied simultaneously, a solution will exist at any interface (even vacuum) because the edge state is insensitive to fields in the $x<0$ region.
Substituting into the Hamiltonians \[Eq. (\[eq:FermionChern\]) and (\[eq:BosonChern\])\] and applying boundary conditions, the topological edge states emerge. For photonic spin-1 states we have, $$\vec{\Psi}_\pm^\textrm{e}(x)=\begin{bmatrix}
\sqrt{\varepsilon}E_x\\ \sqrt{\varepsilon}E_y\\ i H_z
\end{bmatrix}^\textrm{e}_\pm=\Psi_0\begin{bmatrix}
1\\ 0\\\mp i
\end{bmatrix}\left(e^{-\eta_1x}-e^{-\eta_2x}\right).$$ Carrying out the same procedure, the electronic spin- states are expressed as, $$\psi_\pm^\textrm{e}(x)=\psi_0\begin{bmatrix}
1 \\ \pm i
\end{bmatrix}\left(e^{-\eta_1x}-e^{-\eta_2x}\right).$$ The wave functions of the spin-1 and spin- particles appear quite similar. The fundamental difference lies in the fact that $\vec{\Psi}$ is a vector (bosonic) field and its polarization state is defined in *real space*. $\psi$ is a spinor (fermionic) field - its polarization state is more abstract as it lives in a complex space. Notice there are two characteristic decay scales for the edge states $\eta_{1,2}$, like a damped harmonic oscillator, but in spatial frequency. These are the quadratic roots of the secular equation, $$\Lambda_0+\Lambda_2(\eta^2-k_y^2) \mp v\eta=0.$$ If $\textrm{sgn}[\Lambda_0]=\textrm{sgn}[\Lambda_2]=+1$ the skyrmion number is $N=+1$ and only a forward propagating solution $(+)$ exists. On the other hand, if $\textrm{sgn}[\Lambda_0]=\textrm{sgn}[\Lambda_2]=-1$ the skyrmion is $N=-1$ and only the backward propagating solution $(-)$ is permitted. $\eta_{1,2}$ characterize the degree of confinement at a particular momentum $k_y$ and are solved straightforwardly, $$\eta_{1,2}(k_y) =\frac{1}{2|\Lambda_2|}\left[v\pm\sqrt{v^2+4\Lambda_2(\Lambda_2k_y^2-\Lambda_0)}\right].$$ The spatial width of the wave packet depends on the size of the band gap formed by $\Lambda_0$ and $\Lambda_2$. However, regardless of their relative magnitudes, as long as $\Lambda_0\Lambda_2>0$ a solution always exists within the band gap - they are topologically protected.
Intriguingly, the edge waves are also *helically quantized* along the direction of propagation $\hat{k}=\hat{y}$, $$S_y\vec{\Psi}_\pm^\textrm{e}=\pm Q_1\vec{\Psi}_\pm^\textrm{e}.$$ Note that $\hat{k}\cdot\vec{S}=S_y$ is the spin-1 helicity operator and the edge photon carries a discrete unit of bosonic charge $Q_1=1$. Likewise, the electronic edge wave carries a discrete unit of fermionic charge $Q_{\sfrac{1}{2}}=\sfrac{1}{2}$, $$(\sigma_y/2)\psi_\pm^\textrm{e}=\pm Q_{\sfrac{1}{2}}\psi_\pm^\textrm{e},$$ where $\hat{k}\cdot\vec{\sigma}/2=\sigma_y/2$ is the spin- helicity operator. For spin-1, helical quantization means the field is completely transverse to the momentum $\hat{k}\cdot\vec{\Psi}_\pm^\textrm{e}=0$ and the edge state behaves identically to a massless photon. This is known as the quantum gyroelectric effect (QGEE) [@VanMechelen2018]. Similarly, the edge electron behaves just like a helical Weyl fermion. Their dispersion relations read, $$\omega_\pm(k_y)=E_\pm (k_y)=\pm vk_y, \qquad -k_i<k_y<k_i.$$ No solution exists for $k_y\to -k_y$ and the edge states are back-scatter immune. Notice they are linearly dispersing (massless) such that the group velocity is constant $\partial_{k_y}\omega_\pm=\partial_{k_y}E_\pm=\pm v$. Moreover, the edge states are gapless and touch the bulk bands precisely at the band inversion point $k_i=\sqrt{\Lambda_0/\Lambda_2}$, where $\Lambda(k_i)=0$. At this particular momentum, one of the decay lengths becomes infinite $1/\eta(k_i)\to \infty$ and the edge states join the continuum of bulk bands. A diagram of the bulk and edge dispersion is shown in Fig. \[fig:dispersion\].
**Note**: It should be pointed out that the photonic edge states $\vec{\Psi}_\pm^\textrm{e}$ are ill-defined in the zero energy limit $\omega=k_y=0$, which is characteristic of all transverse waves. This is where the edge dispersion intersects the longitudinal band $\omega_0=0$. Since this state is removed from the spectrum (no static charges present), the electromagnetic field vanishes at this point. No zero modes exist for the photon. On the other hand, the electronic edge states $\psi_\pm^\textrm{e}$ have a smooth limit at $E=k_y=0$ and zero modes are permitted. This is yet another significant difference between bosons and fermions which is related to the fact that the Dirac equation can host Majorana bound modes [@Oreg2010]. Since photons are their own antiparticles, no such Majorana states are possible.
Conclusion
==========
In conclusion, we have introduced the concept of a photonic Dirac monopole appropriate for the field of spin photonics, topological photonic crystals and metamaterials. It shows magnetic monopole charge quantization in momentum space arising solely from spin-1 properties of the photon. We elucidated this phenomenon using a Dirac-Maxwell correspondence in the Riemann-Silberstein basis and applied this topological theory to 2D photonic materials. These topologically massive photons are interpreted as spin-1 skyrmions and arise from nonlocal Hall conductivity. Our work illuminates the role of photon spin in the recently proposed quantum gyroelectric phase of matter and topological bosonic phases [@van_mechelen_2017; @VanMechelen2018]. The edge states of such a topological phase exhibit spin-1 quantization as opposed to spin- quantization in fermionic phases of matter. This is ultimately connected to the presence of quantized monopole charges (bosonic- or fermionic-like [@Metlitski2013; @Vishwanath2013]) in the dispersion of bulk matter. Experimentally probing monopole charge in momentum space can shed light on fundamental symmetries in topological electrodynamics of photons and electrons.
Acknowledgements {#acknowledgements .unnumbered}
================
This research was supported by the Defense Advanced Research Projects Agency (DARPA) Nascent Light-Matter Interactions (NLM) Program and the National Science Foundation (NSF) \[Grant No. EFMA-1641101\].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We discuss the problem of CMB spectrum corruption during Galactic emission removing. A new technique of spatial–frequency data reduction is proposed. The technique gives us a possibility to avoid a spatial harmonics nonorthogonality. The proposed technique is applied to the two-year COBE DMR sky maps.
We exclude the harmonics with l=7, 9,13, 23 and 25 as having anomalous statistics noise behavior. One shows that procedure do not give systematic errors, if the data are statistically regular.
The spectral parameter of the power spectrum of primordial perturbation $n=1.84 \pm 0.29$ and quadrupole moment $Q_2=15.22
\pm 3.0$ are estimated. The power spectrum estimation results are inconsistent with the Harrison-Zel‘dovich $n=1$ model with the confidence 99%. It is shown a necessity of an increasing a survey sensitivity to reach a more reliable estimation of the cosmological signal.
author:
- 'Brukhanov A.A., Skulachev D.P., Strukov I.A., Konkina T.V.'
title: |
COBE Data Spatial–Frequency Analysis and\
CMB Anisotropy Spectrum.[^1]
---
*Space Research Institute, Moscow, Russia E-mail: [email protected], [email protected]$\;$*
Introduction.
=============
A large scale cosmic microwave background (CMB) anisotropy carries the information about a primordial metric perturbation amplitude and spectrum, it shape reflects the Universe evolution and structure. To encrypt the information one analyzes a CMB anisotropy spatial spectrum and compares it with a theoretical models with different types of the Universe evolution scenario. A sensitivity rising in the CMB investigations caused a detection of the anisotropy. As a result one may find prerequisites of a more detail examination of the physical conditions involved in the Universe birth and evolve. Being high sensitive and multi-frequent COBE experiment (Smoot et al., 1992, Bennett et al., 1994) gives a possibility to careful expanding of our knowledge about the CMB structure and spectrum.
Unfortunately, the Galaxy radiation predominates even at a millimeter wavelength range. This rises a problem of the radiation filtering. The most Galaxy radiation is coming from rather narrow region located along the Galactic plane. The main method to exclude the radiation is to “cut out” the Galactic plane vicinity and to analyze the rest regions on the sky map ( Strukov et al., 1991a, 1991b, Smoot et al., 1992, Bennett et al., 1994, Gorski, 1994, Gorski et al., 1994, Wright et al., 1994b). This method is well known but it works perfectly only if variance data analysis is applied. If we use the Galactic “cut out” in the case of spatial spectrum analysis we lose a spherical functions orthogonality on the rest sphere part. As a result the harmonics begin to influence each to other. The problem may be solved partly using a new basis which is orthogonal over the rest part of the sphere ( Gorski, 1994) or is orthogonal only to a monopole and dipole because the excluding of these two components effects the most spectrum modification. The spectra obtained this ways may differ from a real one (Bunn et al., 1994) and may cause the essential different conclusions when interpreting the results.
A modern multi-frequency experiments seemingly allow us to separate the blackbody CMB radiation component from the frequency dependent Galaxy foreground. But variations of the radiation spectral composition prevent the procedure from being performed with a desired accuracy.
We propose the method of the spatial spectrum analysis which allows to reach the most spatial resolution without any loss in the reciprocal harmonics orthogonality. The method implies that the CMB filtering stage is conducted not before but after the radio map is expanded to spherical harmonics.
There are two filtering stages in the proposed process. The first stage is a spatial filtering one when some components related with the Galactic plane and the Galaxy center are excluded from the total spatial spectrum. By this expedient the Galactic radiation magnitude is essentially reduced. And only after that, during the second filtering stage a “frequency cleaning” procedure is used. At this stage the CMB anisotropy is separated from the Galactic radiation using several frequency channels and some a priori Galaxy radiation model. The both stages nowhere are associated with any signal harmonics orthogonality losses.
The main method difficulty may be found at the second stage and is bound up with the problem of a correct Galactic radiation model selection. But using the first stage filtering we strongly reduce the effect of the model uncertainty on the net result.
As an initial data we used public released COBE DMR 2-years all sky maps from 31.5, 53, and 90 GHz frequency channels. In addition we exploited a 19.2 GHz balloon data kindly provided by the authors of the survey.
The Method.
============
A data reduction process practically always is accomplished by some types of the data censoring like weighing, excluding outliers, corrections for systematic errors, trends and so on. A censor method selection is based on a priori information about the analyzing sample. The censoring is aimed at the improvement of the estimation stability and at the approaching the real data statistic parameters to the used a priori models.
In our case the values on the observed radio map is a sum of a cosmological signal we try to measure, a receiver noise, the Galactic radiation, and perhaps some signal related with neglected systematic effects. It is the cosmological signal that we are interested in, so the other components remain to be carefully estimated, significantly reduced or to be completely excluded. In an ideal case (no Galactic radiation and no systematic effects) the procedure would not cause a significant change in the cosmological signal parameters.
In a Galactic coordinate system one may find a Galactic emission concentrated near the Galactic plane and around the Galaxy center. The radiation may be described using only a few spatial harmonics. If these harmonics are excluded the Galactic radiation is substantially reduced. In our case we excluded 23 harmonics only. At the same time the rest spatial spectrum consist of 649 harmonics. If a signal is normal distributed the 649 harmonics are quite sufficient for a stable signal spectrum parameter estimation.
To estimate the signal parameter adequately it is absolutely necessary to know (or to estimate) exactly a noise which may be found in any analyzed data. The most problem in the procedure is the presence of anomalous magnitudes in the noise spectrum. If the signal/noise ratio is not good enough the anomalies may cause a significant variation in the parameters to be estimated.
We emphasize that a cause of the anomalies may be some residual systematic errors. In this case it is very difficult to build up a statistical model of the anomaly and the accuracy of the signal estimation may drops below a point that can be tolerated.
If the number of the anomalous harmonics is not large in comparison with the total signal harmonics, the radical solution of the problem is an exclusion of the anomalous harmonics from the analysis. If data are regular, the excluding some anomalous (in the noise spectrum) harmonics is rather safe procedure. That is because the cosmological signal and the noise are Gauss type, and the signal parameters are weakly depended on both the fact of excluding and the criteria of anomalous outliers selection (in our case we rated to anomalous the outliers beyond the 3–sigma).
But if the neglected systematic errors are present in the data, this procedure causes the significant improvement of the sample regularity and causes the increasing of an estimation efficiency. At the same time if the excluding process significantly changes the estimated parameters, it may be a strong reason to assume the data is not regular, and consequently may indicates that it is necessary to use some type of procedure to reduce the irregularity.
It should be noted that using data reduction methods which break a spatial harmonics orthogonality causes a certain spectral smoothing and may reduce or even may camouflage the abnormal amplitudes. In this situation the risk of missing a really existing anomaly is run, and the signal estimation process may be accomplished with some additional errors.
The data reduction procedure we propose is the following. We use COBE A and B channels (Wright et al., 1994a) and produce the sum and the difference radio-maps as a half-sum and a half-difference of the A and B maps respectively. All data on the sum map are a sum of a signal and a noise. All data on the difference map are only the noise, the signal amplitude is zero.
We exclude from the sum map the monopole and dipole components, and the spectral components cased by the Galaxy emission. Then the sum map is additionally “cleaned” from the Galactic radiation. For this purpose we use 19 GHz and 31 GHz maps to determine the Galactic radiation spectral index. Then the 31 GHz map is subtracted from 53 GHz map, but with a weight corresponding the determined spectral index.
We assume the hypothesis the cosmological signal and the noise are normal distributed. The noise spectrum shape is assumed to be completely determined by the number of measurements in every point on the sky. Based on this we calculate the noise parameters and discover anomalous noise components. Then we use a Monte Carlo simulation to estimate signal magnitude, signal spectrum parameters, and the estimates confidence levels. The latter procedure is applied both to a total analyzed spectrum and to a censored spectrum with anomalous components excluded. In addition we test the determined estimations for stability to the procedure of quadrupole component subtracting.
Signal Analysis on the Sphere.
-------------------------------
The signal on the sphere may be represented as a spherical harmonics expansion $Y_{l,m}(\theta, \varphi)$: $$\Delta T(\theta, \varphi) = \sum a^m_l Y_{l,m}(\theta, \varphi)$$ In this case the signal variance on the sphere $\sigma^2$ is represented as:
$$\sigma^2 = \frac{1}{4 \pi} \sum_l \sum_m (a_l^m)^2 = \sum_l \Delta
T_l^2 \,$$ where $\Delta T_l^2$ is a $l$–th spherical harmonic power.
In any real experiment the analyzed signal $a^m_{l,signal}$ is affected by a transfer function $W_l$ (thereafter we use the COBE antenna transfer function (Wright et al., 1994a) as $W_l$) and some noise $a^m_{l,noise}$ is added:
$$\Delta T (\theta, \varphi) = \sum (W_l a^m_{l,signal}+a^m_{l,noise})
Y_{l,m}(\theta, \varphi) \,$$
The cosmological signal on the sphere is a sample of a random process. In a standard cosmological scenario the inflation creates adiabatic scalar and tensor Gauss fields. So the spectrum is assumed to be a sample of a normal random process with independent harmonics, $\langle a_l^m a_k^n \rangle
\sim \delta
(l,k) \delta (m,n)$.
If initial density perturbation spectrum is $(\delta \rho / \rho)^2 = A k^n$, one may find for the mean lowest multipoles in expansion $\Delta T(\theta, \varphi) $ (Bond, Efstathiou, 1987):
$$\langle (a_l^m)^2 \rangle \sim \frac{\Gamma(l+(n-1)/2)
\Gamma((9-n)/2)}
{\Gamma(l+(5-n)/2) \Gamma((3+n)/2)}
\label{approx_DTT}$$
A consideration of other physical phenomena ( evolution effects, acoustic oscillation, Silk effect etc.) causes a more complicated spectrum (White et al., 1994 and references therein). Hereafter we approximate a real spectrum with the (\[approx\_DTT\]). For a quantitative spectrum description we use only two parameters: a power spectrum index $n$ and quadrupole magnitude $Q^2_{rms}=\langle \Delta T^2_2 \rangle$ as an amplitude.
A correlation function on the sphere $C(\beta)=\langle(\Delta T(\vec q_1)) (\Delta
T(\vec q_2))\rangle $, $\vec q_1 \vec q_2=\cos(\beta)$ may be represented in terms of spherical functions: $$C(\beta)=\frac{1}{4 \pi} \sum P_l(\cos \beta ) \sum_m
(a_l^m)^2$$
Using the spherical harmonic expansion we may calculate the correlation between two maps (for example between 31.5 GHz and 53 GHz maps): $$\rho({\rm MAP1, MAP2}) = \frac{ \sum_{l,m} a_l^m(1) a_l^m(2) }
{\sqrt{ \sum_{l,m} (a_l^m(1))^2 \sum_{l,m} (a_l^m(2))^2 }}$$
Spectrum Parameters Analysis.
------------------------------
A cosmological model scenario gives us a prediction of a mean CMB spectrum fluctuation. Experimenters deal with the single sample of the scenario. In addition a noise is always present in a real data. In this connection there is the problem of a signal parameters estimation. The problem was discussed by Bunn et al., 1994 and Sazhin et al., 1995. Here we will not go into the problem, but will touch upon the estimation not only the signal, but the noise parameters too.
The maximum likelihood estimation is known to be the most effective. It may be illustrated if the signal spectrum has a given a priori shape.
Let the spectrum shape is $\langle (a_l^m)^2 \rangle =Q^2 F_{lm}$. If the amplitude is $Q$, the likelihood function may be described as: $$f(a_{lm},Q|F_l) = \prod_{lm} \frac{1}{\sqrt{2 \pi Q^2 F_{lm}}}
\exp \left( - \frac{ a_{lm}^2 } { 2 Q^2 F_{lm} } \right),$$
The maximum likelihood estimation $Q_{\rm ML}$ may be found if solve the equation maximizing the likelihood function: $$\frac{\partial \ln f(Q|F_l)}{\partial Q^2} = 0$$ $$Q^2_{\rm ML} = \frac {\sum (a^2_{lm}/F_{lm})}{ M }, \qquad
M=\sum_{l,m} 1 \,$$
where $M$ is the number of the analyzed harmonics.
The parameter is evident having $\chi^2_M$ distribution with $M$ degree of freedom, $Q^2_{\rm ML} \sim \chi^2_M$. The amplitude estimation calculated using the variance on the sphere $ \sigma^2
$ is: $$Q^2_{\rm POWER}=\frac{\sum \Delta T^2_l}{\sum \langle \Delta
T^2_l \rangle |{Q=1}}
= \frac{\sigma^2}{\sigma^2_{th}|Q=1} \sim \chi^2_{N_{eff}} \, .$$
It is clear that: $N_{eff}= \frac{(\sum_{lm} \Delta T^2_l)^2}{\sum_{lm} (\Delta
T^2_l)^2} \le M$. So for the Harrison-Zel’dovich spectrum $N_{eff} \approx 100$, $ M \approx 700$ пpи $l_{max} =25$. The maximum likelihood estimation accuracy is about 5% and the total power estimation is about 14%.
Using the approach we may determine the signal spatial spectrum parameters in the presence of noise. Maximum likelihood estimation offers to find a minimum of the functional:
$$\sum_{l,m} \ln(W_l^2 \langle(a_l(Q,n))^2 \rangle + \langle (a_{l,noise}^m)^2
\rangle ) + \frac{(b_l^m)^2}{W_l^2 \langle(a_l(Q,n))^2 \rangle +
\langle (a_{l,noise}^m)^2 \rangle}
\label{likhood}$$
relative $Q$ and $n$ with given spectrum shape and measured signal and noise sample spectra. Here $\langle(a_l(Q,n))^2 \rangle$ is the model signal spectrum, depending upon desired parameters and described as (\[approx\_DTT\]), $(b_l^m)^2$ is the measured spectrum (we describe it further), and $\langle(a_{l,noise}^m)^2 \rangle$ is a model noise spectrum (\[sp\_noise\]), its amplitude is found using the same approach but from the difference map .
In addition we are to take into account a procedure of the Galactic radiation “cleaning” and a procedure of a conversion the antenna temperatures into a thermodynamic scale. In all cases the summation is taken over multipoles chosen for the analysis. The necessary conditions for the described procedure is a knowledge of the measured signal and noise spectra.
COBE Data Noise Analysis.
--------------------------
There are a lot of papers concerning the COBE data noise analysis. Bennett et al., 1994, Lineweaver et al., 1994 discussed the correlation noise properties and showed a weak ( 0.45% ) correlation at 60$^\circ$. So in practice we may assume the noise is uncorrelated. The noise power estimation may be done using the difference maps.
As a first approximation assume that the noise is white. However the real observations have different accuracy on the sphere. The COBE orbit configuration was so that the points near the ecliptic poles were observed much longer than others. Let us consider the methods to take into account the measurement accuracy is not equal for different points.
Let the noise in every point on the sphere be normal with zero mean and the variance $\sigma^2(\theta,\varphi)=\sigma_0^2/N(\theta,\varphi)$, where $N(\theta,\varphi)$ is a number of measurements in the point with the $\theta$, $\varphi$ coordinates. Then we may propose the following noise spectrum model on the sphere. If there is a white noise sample $G=\sum n_l^m Y_l^m$, $\langle n_l^m
n_{l^
{\prime}}^{m^{\prime}} \rangle = 0$, $\langle (n_l^m)^2
\rangle =n^2=const$, the true noise on the sphere is a product of $G(\theta,\varphi)$ (here, and only here $n$ is a magnitude of the noise power spectrum, not a spectrum index), and $1/\sqrt{N(\theta,\varphi)}$: $$2
R(\theta,\varphi) = 1/\sqrt{N(\theta,\varphi)} G(\theta,\varphi) =
w(\theta,\varphi)G(\theta,\varphi)$$
According to Peebls (1980), assume $w(\Omega)Y_l^m(\Omega) =
\sum_{l^{ \prime} m^{\prime}} w_{l l^{\prime}}^{m m^{\prime}}
Y_{l^{\prime}}^{m^{\prime}}(\Omega)$, where $$w_{l l^{\prime}}^{m m^{\prime}} = \int d \Omega w(\Omega)
Y_{l^{\prime}}^{m^{\prime}}(\Omega) Y_l^m(\Omega) \$$
For the true noise spherical harmonics expansion $R=\sum a_{l,noise}^m Y_l^m$, coefficients are $a_{l,noise}^m=\sum_{l^{\prime} m^{\prime}} w_{l l^{\prime}}^{m
m^{\prime}}
n_{l^{\prime}}^{m^{\prime}}$.
Taking into account noncorrelatedness of white noise harmonics a covariation of $a_{l,noise}^m$ may be written as: $$\langle a_{l,noise}^m a_{l^*,noise}^{m^*} \rangle =\sum_{l^{\prime}
m^{\prime}}
\sum_{l^{\prime\prime} m^{\prime\prime}}
w_{l l^{\prime}}^{m m^{\prime}}
w_{l^* l^{\prime \prime}}^{m^* m^{\prime\prime}}
\langle n_{l^{\prime}}^{m^{\prime}} n_{l^{\prime \prime}}^{m^{\prime
\prime}} \rangle = \\ n^2 \sum_{l^{\prime} m^{\prime}} w_{l
l^{\prime}}^{m m^{ \prime}} w_{l^* l^{\prime}}^{m^* m^{\prime}}$$
And finally noise power spectrum is: $$\langle|a_{l,noise}^m|^2 \rangle =n^2 \sum_{l^{\prime} m^{\prime}}
|w_{l l^{\prime}}^{m
m^ {\prime}}|^2 = n^2 \int d \Omega w^2(\Omega)|Y_l^m(\Omega)|^2
\label{sp_noise}$$
We use not only unit or zero weights, but unrestricted values. Because of that our equation differs from Peebls’s one in the squared weights.
Then, using maximum likelihood approach to the difference maps we may calculate the amplitude of the spectrum given above.
The power spectrum model make it possible to analyze the noise spectrum measured from the difference maps. The analysis shows that in general the noise spectrum is well described both the model (\[sp\_noise\]) and the white noise model. It is due to a rather perfect COBE coverage of the celestial sphere. But the same analysis shows for the both models the existence of harmonics with an excess power. So in the noise spectrum of 31 GHz map one may find for $l=13$ the power is greater than predicted from (\[sp\_noise\]). The power excess is 3.4$\sigma_{l=13}$, where predicted multipole power variation $\sigma_l$ is calculated basing on the model and on the degrees of freedom for the analyzed component. Further we use a censoring of the data to exclude this and similar components from the analysis.
The noise spectral component with $l=25$ has anomalous high correlation between 31 and 53 GHz maps. After subtracting the Galaxy radiation the excess power is 3.3$\sigma_{l=25}$. Components with $l=25$ were excluded from the analysis too.
We find a significant power excess of several noise components on 53 GHz map. We exclude the components with power excess more than 3$\sigma_{l}$. After that we determine the spectrum parameters for the rest components and the procedure is repeated until the spectrum is free from the abnormal components. Thus we excluded the components with $l=7,\, 23,\, 9$ having the initial power excess of 3.11$\sigma_{l=7}$, 2.63$\sigma_{l=23}$, and 2.31$\sigma_{l=9}$.
In the final analysis we use spectral components from $l_{min}=2$ to $l_{max}=22$, except components $l=7,\, 9,\, 13$.
Such censoring effects only a weak power spectrum distortion if the signal amplitude distribution is normal. The matter is discovered by numerical modeling, the results are presented in Chapter 3.
During signal spectrum analysis we ignore a noise spectrum components correlation and use the spectrum shape only. Further modeling shows the same results if we use a pure white noise.
Selection of the Galactic Radiation Model
-----------------------------------------
Our approach is based on the spatial data filtering which reduces the influence of the Galactic spectrum index uncertainty. The accepted Galactic radiation model for the analyzed frequency region is assumed to involve a synchrotron component with frequency dependence as $T=T_0\nu^{\alpha}$, $\alpha=-3\pm 0.2$ and a bremsstrahlung component with spectral index $\alpha=-2.1$ (Bennett et al., 1992). If there are the space regions of a different nature of radiation on the line of sight, it may cause a spatial variation of the spectral index $\alpha$. Let us try to estimate roughly the contribution of such variation if we use some effective but constant value $\alpha_0$. Let the true value of the spectral index be $\alpha=\alpha_0+\Delta
\alpha$. After the frequency depended part is excluded from the data $T_1$ и $T_2$ of two frequency channels, one may derive: $$\delta T = T_1 \left(\frac{\nu_2}{\nu_1}\right)^{\alpha_0+\Delta
\alpha} - T_1
\left(\frac{\nu_2}{\nu_1}\right)^{\alpha_0} \approx
T_2 \left[ \left(\frac{\nu_2}{\nu_1}\right)^{\Delta \alpha}
-1 \right] \approx
T_2 \Delta \alpha \ln \left(\frac{\nu_2}{\nu_1}\right)$$
The residual signal power caused by the spectral index variation $\sigma^2_{\rm VAR}= \Delta \alpha^2 (\ln\nu_2 /\nu_1)^2
\sigma^2_{\rm MAP2} $. For the frequencies 31.5 and 53 GHz the residual RMS is smaller than $0.52 \Delta \alpha \sigma_{\rm MAP2}$. It can be seen that the less is Galactic radiation contribution, the less is the residual RMS on the more high frequency radio map.
It is evident the most Galactic radiation is concentrated near the Galaxy plane and near its center. If the spatial spectrum is represented in Galactic coordinate system the most power may be found in the components with $m=0$ for the even multipoles (Galaxy plane) and with $m=1$ for the odd multipoles (Galaxy center). So if we exclude this components the influence of the spectral index variation may be reduced drastically. At the same time this procedure does not break the multipoles orthogonality.
Table \[tab\_gal\] shows some results obtained after the spectrum analysis from $l_{min} =2$ to $l_{max}=25$.
In addition one may find at the last row the difference between 53 GHz map and the 31 GHz map scaled with a spectral index $\alpha
= -2.15$ after spatial filtering. Data are represented in antenna temperatures. One can see the resultant $\sigma_{Sky}$ may be explained either by a cosmological signal existence or by the spectral index variation within $\Delta \alpha \approx 0.5$.
Unfortunately the available data give no way to estimate the variation with required precision. For more reliable conclusions it is necessary to have the sensitivity many times better than COBE instrument has.
We tried to estimate the spectral index variation crude. For this purpose we use both the COBE data and 19.2 GHz survey (Boughn et al., 1992). We has converted latter data to a COBE beam shape (Wright et al., 1994a). After that we analyzed the frequency dependence of the most intensive spherical harmonics and determined the accuracy of the estimation.
The data are shown in Table \[indx\] where one can see correspondingly: analyzed frequency ranges, spectral index $\alpha$ for desired signal/noise ratio equal to 5, spectral index variation $\Delta\alpha_5$, and the predicted spectral index variance $\Delta\alpha_N$, calculated basing on the noise analysis. In addition there are shown the spectral index $\alpha_W$ and its variation weighted-mean for all harmonics with $m\ne0$ for $l=2k$ and with $m\ne1$ for $l=2k+1$.
It may be noticed the absolute value of the spectral index $\alpha_{19-31}$ is regular lower than $\alpha_{31-53}$. The contribution to the total spectrum of the spectral components with signal/noise ratio equal to 5 is more than 40/found the only spectral component with the spectral index like a synchrotron Galactic radiation has. This is $a_{2,2}$ component with the spectral index $\alpha_{19-31}=-2.86\pm 0.15$, $\alpha_{31-53}=-2.92\pm 0.22$.
The different maps correlation analysis gives the mean spectral index estimate in the region $\alpha=-(2.2 \div 2.3)$.
So we may assume the hypothesis the spectral index is constant and its measured variations are caused only by the instrumental noise.
On the basis of the analysis we can conclude for the most significant Galactic spectral components (with the exception of $a_{2,2}$ ) the Galactic radiation model may be assumed to be one-component, with the radiation cased only by an ionized hydrogen. So the $\Delta\alpha$ variations may be taken as zero.
Now we may determine the measured spectrum
$$b_l^m = \left(
k(53) - k(31) \left( \frac{53}{31.5}\right)^{-2.15}
\right)^{-1}
\left(
a_l^m(53\,GHz) - a_l^m(31.5\,GHz)
\left( \frac{53}{31.5} \right)^{-2.15}
\right) \, ,$$
where $k(31)=1./1.025724$, $k(53)=1./1.074197$ are the scale coefficients to transform the thermodynamic temperatures to antenna temperatures for 31.5 GHz and 53 GHz, $a_l^m(31.5\,GHz)$, $a_l^m(53\,GHz)$ are multipole coefficients for 31.5 GHz and 53 GHz maps.
Unfortunately, the COBE sensitivity is not good enough and the accuracy of the estimation is determined by the instrumental noise. We are pinning our hopes on the “Relict-2” space experiment. “Relict-2” instrument is preparing now, it has an order better sensitivity than COBE has, and it will give the possibility to analyze the Galactic spectral index radiation with high reliability and so to separate the cosmological signal from the Galactic radiation with high accuracy.
Results
=======
As a result of the analysis we obtained the spatial spectrum parameters estimates under the assumption that Galactic radiation is one-component. For the quadrupole we found: $$Q_2=15.22 \pm 3.0 \mu{\rm K}$$ and for the power spectrum index: $$n=1.84 \pm 0.29$$ We analyzed spectral components from $l_{min}=2$ to $l_{max}=22$ if $m\ne 0$ for $l=2k$, and $m\ne 1$ for $l=2k+1$, and (due to the noise anomalies) if $l \ne 7,9,13$.
The accuracy of the obtained parameters is derived using Monte-Carlo simulation. For all estimates we show RMS errors. The procedures of difference and sum maps producing and of noise parameters determining are included in the Monte-Carlo simulation. The simulation also takes into account the noise, its variations, and a cosmic variance.
We find the total power measured in chosen spatial frequency window as $(70.28\,\mu{\rm K})^2$, noise estimate as $(56.76\,\mu{\rm K})^2$, and on-sphere signal estimate as $(41.44\,\mu{\rm K})^2\pm(14.68\,\mu{\rm K})^2$. Total number of analyzed harmonics is $M=446$.
Table \[tbl\_n\] shows the results. In addition we include into the Table \[tbl\_n\] the data obtained with anomalous harmonics, and with excluded quadrupole.
We tested an effect of the harmonics excluding procedure on the signal estimation. In the case of the noise outliers caused by systematic errors and if all harmonics are orthogonal the excluding does not effect the estimation bias, but may only increase the errors in comparison to non-censored data. Simulation shows the $Q_2$ and $n$ variations are 0.9 and 0.11 respectively.
On the other hand, if the outliers are pure stochastic (i.e. we are dealing with a low-probability noise sample) the noise estimation bias may arise and, as a result may arise the signal estimation bias. We simulated this occasion for the signal and noise with the given spectrum and to examine noise outliers stronger than 2.3$\sigma$. After that we calculated the signal estimates both for censored and non-censored spectra.
After 1228 simulations we have found 500 events with the outliers. The corresponding estimates are: $\langle
n_{censor}-n_{full}\rangle=0.07\pm 0.08$ $\langle
Q_{censor}-Q_{full} \rangle=-0.4\pm 0.83$. It should be noted that the obtained estimates really are upper bounds rather than two-sided limits. It is because we exclude the components, which then are used in some linear combination and thus are additionally normalized. It may, in general, reduce the effect of spectrum censoring.
COBE data analysis shows (see Table \[tbl\_n\]) that the procedure of the anomalous harmonics excluding effects the significant change in $n$. If the instrumental noise is completely random the probability to obtain so large difference is smaller than $0.02\%$. One may see that the applying the anomalous harmonics censoring to COBE data causes a dramatically decreasing of the strange dependence of the results on whether or not the quadrupole is excluded. It may be an indirect support to the necessity of proposed data censoring.
Basing on the obtained parameters we may predict the signal in an experiment like it is conducted in Tenerife (Hancock et al., 1994). Assuming $5.5^{\circ}$ beam, $8.1^{\circ}$ antennae separation, and 3-point method of observation i.e. $\Delta T = T_0
-0.5(T_1 + T_2)$, we may calculate the following data for the spectrum (\[approx\_DTT\]) type using previously obtained $Q_2$ and $n$, excluding components with anomalous noise and applying the spatial filtering: $\sigma_{Tenerif} = 54.82 \, \mu{\rm K}$.
The same analysis, but without excluding anomalous components shows two estimates:
$\sigma_{Tenerif} = 37.22 \, \mu{\rm K}$ with quadrupole included, and: $\sigma_{Tenerif} = 34.47 \, \mu{\rm K}$ with quadrupole excluded.
We do not use a $4^{\circ}$ binning in the analysis. The binning may decrease the obtained data in some degree. The binning is used in Tenerife experiment and the results are (Hancock S. et al., 1994):
$\sigma_s=49\pm 10 \, \mu{\rm K}$ for 33 GHz channel, and $\sigma_s=42\pm 9 \, \mu{\rm K}$ for a sum of the 15 and 31 GHz channels. So the cosmological signal spectrum with the parameters we obtained is in a good agreement with the observations which are more sensitive than COBE for high spatial harmonics.
Conclusions.
=============
The systematic effects exclusion problem is met practically in any experiment. In the case of CMB anisotropy observation the Galaxy radiation is the primary effect. The method usually used to suppress the radiation is to “cut off” the Galaxy plane region. It causes the problem of the monopole and dipole exclusion. The latter in its turn effects the additional systematic errors but at the less level.
The orthogonal basis when coupled with the spatial-frequency filtering of the Galactic radiation allows to avoid the systematic errors mentioned above. In addition the proposed approach make it possible to analyze and to exclude more fine effects usually caused errors in spectrum parameters measurements. If an instrumental noise is normal the data censoring could not cause a significant difference between the spectral parameters derived from censored and non-censored data sets. Being detected the difference may be result from either neglected residual effects or a noise nonnormality. Being in the context of the normal distribution function we are forced to exclude the spectral components with an anomalous noise behavior.
The cosmological spectrum parameters we obtained differ in some degree from the parameters derived in previous works used the same initial COBE data. The difference is due to the anomalous noise harmonics exclusion rather than spatial–frequency Galactic radiation filtering. In its turn the harmonics exclusion is possible because of the harmonics orthogonality conservation.
A number of investigators (Table \[rez\]) announced the strange end result sensitivity to a quadrupole component. Most likely it is also attributed to anomalous noise harmonics influence.
The results based on 2-year COBE data and obtained by several authors are shown in Table \[rez\]. In addition we would like to remind the result of $n=1.7$, derived by (Hancock S. et al., 1994) after a comparison of 1-year COBE and Tenerife data.
So we obtained the estimate of $n$ more precise and somewhat higher than it is derived by other investigators.
To the best of our knowledge, nobody has investigated the COBE noise (A–B) maps in detail. Gorski et al. (1994) used only the mean noise parameters assumed they are normal. Wright et al. (1994b) used the noise spectrum, but the used basis was not complete orthogonal. Bennett et al. (1994) worked by the help of a correlation function but only on the part of the sphere. In the latter two cases the basis function orthogonality is lost and it is impossible to analyze noise spectrum in detail.
The next result we obtained and that differs from previously published is the rejection of the trivial (with $n=1$) Harrison-Zel’dovich spectrum on the confidence level of 99%.
As a result there are very unlikely the models with high $\Lambda$–term and a more probable are the open models with $\Omega <1$, if we assume the initial density perturbations are determined by the power law index $n=1$ (Kamionkowski M., Spergel D.N., 1994). Our result supports the existence of a barionic entropy models. At the same time there are not ruled out the models having $n>1$ in analyzed scales and having more complicated inflation potential (Starobinsky 1992, White et al., 1994 and references therein).
It must be emphasized that by now the spectrum parameters determining accuracy is not enough to draw more deep inferences. Moreover we can see the obtained results being very sensitive to the accuracy of the noise determining and to the used specific estimation procedure.
We hope the 4-year COBE investigation circle will be accessible in the near future. Unfortunately even the 4-year data will not improve the situation drastically. First, it is necessary to increase the instrumental sensitivity as a minimum an order. This will be reached in the planned experiment “RELICT-2”. So the accuracy of the $n$ estimation will be about 5-7% and will be determined the cosmic variance rather than instrumental noise (Sazhin et al., 1995). Second, it is necessary to enhance the anisotropy measurement angle resolution coupled with a sky coverage increasing (Scott et al., 1994). Unfortunately, the modern middle and small scale investigations give us the information only about a few point on the sky and so the results are strongly contaminated by the sample variance.
Authors thank Boughn S.P., Cheng E.S., Cottingham D.A., and Fixen D.J. for providing the 19 GHz data. The COBE data sets were developed by the NASA/GSFC under guidance of COBE Science Working Group and were provided by the NSSDC. This work was supported partly by RFFI grants N93-02-930, 93-02-931, ISF grants N MO6000, MO6300. We would like to thank A. Klypin and M. Sazhin for helpful discussions and useful comments.
References.
============
[=-1cm]{}
Bunn et al. (Bunn E., Hoffman Y., Silk J.)// Astrophys.J. 1994. V.425. P.359.
Bennett et al. (Bennet C.L., Smooth G.F., Hinshaw G., Wright E.L. et.al.)// Astrophys.J.(Letters) 1992. V.396. L7.
Bennett et al. (Bennet C.L., Kogut A., Hinshaw G., Banday A.J. et.al.)// Astrophys.J. 1994. V.436. P.423.
Boughn et al. (Boughn S.P., Cheng E.S., Cottingham D.A., Fixen D.J.)// Astrophys.J. (Letters) 1992. V.391. L49.
Bond, Efstathiou (Bond J.R., Efstathiou G.)// Monthly Not.Roy.Astron.Soc. 1987. V.226, P.655
Gorski (Gorski Krzysztof. M.)// Astrophys.J.(Letters) 1994. V.430. L85.
Gorski et al. (Gorski K.M., Hinshaw G., Banday A.J., Bennet C.L. et.al.)// Astrophys.J.(Letters) 1994. V.430. L89.
Kamionkowski, Spergel (Kamionkowski M., Spergel D.N.)// Astroph.J., 1994, V.432, P.7-16
Lineweaver (Lineweaver C.H., Smooth G.F., Bennet C.L., Wright E.L. et.al.)// Astrophys.J. 1994. V.436. P.452.
Peebls (Peebls P.J.E.)// The Large-Scale Structure of the Universe, Princeton: Princeton Univ.Press, 1980.
Wright et al. (Smooth G.F., Kogut A., Hinshaw G., Tenorio L. et.al.)// Astrophys.J. 1994а. V.420. P.1
Wright et al. (Wright E.L., Smooth G.F., Bennet C.L., Lubin P.M.)// Astrophys.J. 1994б. V.436. P.443.
White et al. (White M., Scott D., Silk J.)// Ann. Rev. Astron. and Astrophys. 1994. V.32. P.319.
Hancock et al. (Hancock S., Davies R.D., Lasenby A.N., Gutierrez de la Cruz C.M. et.al.)// Nature. V.367. P.333.
Sazhin et al. (Sazhin M.V., Brukhanov A.A, Strukov I.A., Skulachev D.P.) // Pis’ma v astronomichesky zhurnal, 1995, v.21, p.403 (in Russian, in English see Astronomy Letters ( Soviet Astronomy Letters), 1995, v.21, p.358-365)
Scott et al. (Scott D., Srednicki M., White M.)// Astrophys.J.(Letters) 1994. V.421. L5.
Smoot et al. (Smoot G.F., Bennett C.L., Kogut A., Wright E.L., et.al.)// Astrophys.J.(Letters) 1992. V.396. L1.
Starobinsky (Starobinsky A.A) // Pis’ma v zhurnal experim. i teoret. fiziki, 1992. v.55. p.477 (in Russian)
Strukov et al. (Strukov I.A., Brukhanov A.A, Skulachev D.P., Sazhin M.V.)// Pis’ma v astronomichesky zhurnal 1992a. v.18. p.387 (in Russian, in English see Soviet Astronomy Letters, 1992, v. 18, p. 153 )
Strukov et al. (Strukov I.A., Brukhanov A.A, Skulachev D.P., Sazhin M.V.)// Monthly Not. Roy. Astron. Soc. 1992b. V.258, P.37p.
Sugiyama, Silk (Sugiyama N., Silk J.)// Phys.Rev.Letters. 1994. V.73. P.509.
Table Caption
=============
[=-1cm]{}
[**Table \[indx\].**]{} Mean spectral indexes and its variations determined for spherical harmonics at different frequency ranges. The first column – the frequency ranges where the spectral index $\alpha$ is analyzed. The second column – spectral index determined from the most significant spectral components with the signal/noise ratio lager than 5, $\alpha_5$. The third column – its variation, $\Delta\alpha_5$. The forth column – spectral index variance predicted from noise analysis, $\Delta\alpha_N$. The fifth column – the mean weighted spectral index $\alpha_W$ for all harmonics with $m\ne0$ for $l=2k$, and $m\ne1$ for $l=2k+1$. The sixth column – its variance $\Delta\alpha_W$.
[**Table \[tab\_gal\].**]{} Comparative result of an influence of spatial filtering (i.e. excluding the components with $m=0$ for $l=2k$ and $m=1$ for $l=2k$ from $l=2$ to $l=25$ inclusively) to a signal amplitude for different frequencies. The last row shows the signal amplitude after the spatial – frequency filtering assuming the Galactic radiation spectral index as corresponded to ionized hydrogen. $\sigma_{(A+B)/2}$ – amplitude, calculated as a half-sum of two maps (i.e. signal plus noise), $\sigma_{(A-B)/2}$ – noise, calculated as a half-difference of two maps, $\sigma_{Sky}$ – signal amplitude estimation on the sky.
[**Table \[tbl\_n\].**]{} Spectrum analysis after spatial–frequency Galactic radiation filtering. The influence of quadrupole excluding is shown. In addition it is shown the results after noise anomalous harmonics are excluded. The first row is corresponded to spectrum from $l=2$ to $l=22$ with the abnormal harmonics excluded. The second row – the same but in addition the quadrupole is excluded. The third row shows the results if all harmonics from $l=2$ to $l=25$ are used. The forth row shows in this case the influence of quadrupole exclusion. $\sigma_{A+B}^2$ – total measured power within given spatial window, $\sigma_{A-B}^2$ – noise power estimation obtained within the same window from a difference map, $\sigma_{Sky}^2$ – cosmological signal power estimation, ($ \sigma_{Sky}^2=\sigma_{A+B}^2 - \sigma_{A-B}^2 $ ), $Q_2$, $n$ – estimations of parameters for spectrum (\[approx\_DTT\]), $M$ – total number of analyzed harmonics. In all cases we assume that $m\ne0$ for $l=2k$ and $m\ne1$ for $l=2k+1$.
[**Table \[rez\].**]{} Results of CMB anisotropy spatial spectrum analysis obtained by different authors basing on COBE 2-years data. For the comparison our result is shown. It is shown how a quadrupole excluding does influence on the end result.
------------- ------------ ------------------ ------------------ ------------ ------------------
frequency $\alpha_5$ $\Delta\alpha_5$ $\Delta\alpha_N$ $\alpha_W$ $\Delta\alpha_W$
(GHz)
19.2 – 31.5 -2.13 0.36 0.38 -2.12 0.69
31.5 – 53.0 -2.27 0.28 0.36 -2.19 0.76
------------- ------------ ------------------ ------------------ ------------ ------------------
\[indx\]
------------------ ----------- -------------------- -------------------- ----------------
frequency spatial $\sigma_{(A+B)/2}$ $\sigma_{(A-B)/2}$ $\sigma_{Sky}$
(GHz) filtering $(\mu K)$ $(\mu K)$ $(\mu K)$
yes 641. 95.92 634.70
31.5
no 254.79 94.52 239.83
no 191.64 33.04 188.80
53.
yes 83.32 32.20 76.86
frequency
“cleaning”
с $\alpha=-2.15$ yes 50.78 45.44 22.67
------------------ ----------- -------------------- -------------------- ----------------
\[tab\_gal\]
---- ------------ ------------------ ------------------ --------------------------- ----------------- ---------------- -----
Quadrupole $\sigma_{A+B}^2$ $\sigma_{A-B}^2$ $\sigma_{Sky}^2$ $Q_2$
NN analysis $\mu K^2$ $ \mu K^2$ $ \mu K^2$ $ \mu K$ $n$ $M$
1 yes $(70.28)^2$ $(56.76)^2$ $(41.44)^2 \pm (14.68)^2$ $15.22 \pm 2.9$ $1.84\pm 0.29$ 446
2 no $(68.92)^2$ $(56.47)^2$ $(39.51)^2 \pm (14.65)^2$ $15.55\pm 3.8 $ $1.81\pm 0.37$ 442
3 yes $(82.91)^2$ $(74.19)^2$ $(37.03)^2 \pm (17.48)^2$ $18.03 $ $1.31$ 649
4 no $20.3 $ $1.12$ 645
---- ------------ ------------------ ------------------ --------------------------- ----------------- ---------------- -----
\[tbl\_n\]
[|c|c|c|]{} Author& $n$ & $Q_{rms-PS}$\
\
& &\
Bennett et al.(1994)& $1.42_{-0.55}^{+0.49}$ & $12.8_{-3.3}^{+5.2} $\
& &\
Gorski et al.(1994)& $1.22_{-0.52}^{+0.43}$ & $17.0_{-5.2}^{+7.5}
$\
& &\
Wright et al.(1994b)& $1.39_{-0.39}^{+0.34}$ &\
& &\
this work & $1.84 \pm 0.29$ & $15.22 \pm 2.9$\
& &\
\
& &\
Bennett et al.(1994)& $1.11_{-0.55}^{+0.60}$ & $15.8_{-5.2}^{+7.5} $\
& &\
Gorski et al.(1994)& $1.02_{-0.59}^{+0.53}$ & $20.0_{-6.5}^{+10.5} $\
& &\
Wright et al.(1994a), $l=3-30$& $1.25_{-0.45}^{+0.40}$ &\
& &\
Wright et al.(1994b), $l=3-19$& $1.46_{-0.44}^{+0.39}$ &\
& &\
this work & $1.81 \pm 0.37$ & $15.55 \pm 3.8$\
& &\
\[rez\]
[^1]: accepted to Pis’ma v Astronomicheskij Zhurnal, 1996, v.22, in Russian
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Dany Lachance-Quirion,$^{1}$ Samuel Piotr Wolski,$^{1}$ Yutaka Tabuchi,$^{1}$\
Shingo Kono,$^{1}$ Koji Usami,$^{1}$ Yasunobu Nakamura$^{1,2\ast}$\
\
\
\
\
\
title: |
Entanglement-based single-shot detection\
of a single magnon with a superconducting qubit
---
The recent development of hybrid systems based on superconducting circuits has opened up the possibility of engineering sensors of quanta of different degrees of freedom. Quantum magnonics, which aims to control and read out quanta of collective spin excitations in magnetically-ordered systems, furthermore provides unique opportunities for advances in both the study of magnetism and the development of quantum technologies. Using a superconducting qubit as a quantum sensor, we report the detection of a single magnon in a millimeter-sized ferromagnetic crystal with a quantum efficiency of up to $\mathbf{0.71}$. The detection is based on the entanglement between a magnetostatic mode and the qubit, followed by a single-shot measurement of the qubit state. This proof-of-principle experiment establishes the single-photon detector counterpart for magnonics.
Quantum sensing aims to exploit the fragility of quantum states to external perturbations for the development of novel sensors. Quantum-enhanced sensing has now become one of the leading applications of quantum technologies [@Degen2017; @Boss2017]. Entanglement can be harnessed in quantum sensing to indirectly probe a system of interest through a well-controlled auxiliary mode acting as the sensor [@Johnson2010; @Zhao2012a; @Kono2018]. Such a task requires careful engineering to integrate existing quantum technologies into sensors able to detect various physical quantities.
The recent development of hybrid quantum systems provides a natural platform to engineer such quantum sensors [@Kurizki2015]. Indeed, the combination of systems that harness complementary features for quantum technologies opens up the possibility of sensing one degree of freedom through another well-controlled system. One of the main challenges for this application lies in achieving high-fidelity control and readout of the quantum sensor in a hybrid device. Hybrid systems based on superconducting circuits [@Blais2004; @Devoret2013] offer a versatile platform to overcome this challenge. Recent demonstrations include the measurement of the coherence of a bulk acoustic wave resonator [@Chu2017] and the creation and characterization of quantum states of phonons [@Satzinger2018; @Chu2018].
Quantum magnonics provides another promising architecture for developing quantum sensors based on hybrid systems [@Tabuchi2015; @Tabuchi2016; @Lachance-Quirion2017; @Lachance-Quirion2019]. In quantum magnonics, magnetostatic modes inmagnetically-ordered solid-state systems are coherently coupled to superconducting qubits. Here, we combine high-fidelity control and readout of a superconducting qubit to demonstrate a sensor able to faithfully detect single magnons, the quanta of excitations in magnetostatic modes, via entanglement between the two systems. Our demonstration brings the equivalent of the single-photon detector to the emerging field of magnon spintronics [@Chumak2015] and establishes a novel quantum technology for magnetism.
To realize the single-magnon detector, we use a hybrid system composed of a spherical ferrimagnetic crystal of yttrium iron garnet (YIG), a transmon-type superconducting qubit, and a three-dimensional microwave cavity [@Tabuchi2015; @Tabuchi2016; @Lachance-Quirion2017; @Lachance-Quirion2019]. As schematically represented in Fig. 1A, this system hosts three modes of interest: the uniform magnetostatic mode, or Kittel mode, in the ferromagnetic crystal of tunable frequency $\omega_\mathrm{m}/2\pi$; the qubit of frequency $\omega_\mathrm{q}/2\pi\approx7.92$ GHz; and a microwave cavity mode of frequency $\omega_\mathrm{c}/2\pi\approx8.45$ GHz. The Kittel mode and the superconducting qubit are respectively coupled to the cavity mode through magnetic-dipole [@Huebl2013; @Tabuchi2014] and electric-dipole couplings [@Blais2004; @Koch2007; @Paik2011]. These interactions lead to an effective beam-splitter interaction between the Kittel mode and the qubit [@Tabuchi2015; @Tabuchi2016; @Lachance-Quirion2017; @Lachance-Quirion2019]. This coherent interaction enters the strong coupling regime with a coupling strength $g_\mathrm{q-m}/2\pi=7.13$ MHz, much larger than the decay rates of each system (Fig. 1B).
Due to the strong coherent coupling, a strong dispersive interaction between the Kittel mode and the qubit can be engineered [@Lachance-Quirion2017; @Lachance-Quirion2019]. This dispersive interaction, of strength $\chi_\mathrm{q-m}$, is described by the Hamiltonian $$\hat\mathcal{H}_\mathrm{q-m}^\mathrm{disp}/\hbar=\frac{1}{2}\left(2\chi_\mathrm{q-m}\hat c^\dagger\hat c\right)\hat\sigma_z,
\label{eq:dispersive}$$ where $\hat c$ ($\hat c^\dagger$) annihilates (creates) a magnon in the Kittel mode, and $\hat\sigma_z=|e\rangle\langle e|-|g\rangle\langle g|$, with $|g\rangle$ ($|e\rangle$) the ground (excited) state of the qubit. The qubit–magnon dispersive interaction leads to a shift of the qubit frequency by $2\chi_\mathrm{q-m}$ for each magnon in the Kittel mode. To characterize the dispersive interaction, we perform Ramsey interferometry on the qubit while continuously driving the Kittel mode on resonance at $\omega_\mathrm{m}/2\pi\approx7.79$ GHz, far-detuned from the qubit (Figs. 1C and D). As shown in Fig. 1E, the qubit spectrum, obtained from the Fourier transform of the Ramsey oscillations, indicates that the qubit frequency is shifted by $2\chi_\mathrm{q-m}/2\pi=-3.82$ MHz in the presence of a single magnon, a quantity larger than the linewidths $\gamma_\mathrm{m}/2\pi=1.61$ MHz of the Kittel mode and $\gamma_\mathrm{q}/2\pi=0.33$ MHz of the qubit, therefore reaching the strong dispersive regime [@Gambetta2006; @Schuster2007; @Lachance-Quirion2017; @Lachance-Quirion2019; @Sletten2019a].
The single-magnon detection protocol is enabled by the possibility of entangling the Kittel mode and the qubit [@Johnson2010]. Indeed, through the strong dispersive interaction, the qubit can be excited conditionally on the Kittel mode being in the vacuum state $|0\rangle$ [@Johnson2010; @Kirchmair2013]. The effect of the conditional excitation $\hat X_\pi^0$, with the qubit initially in the ground state $|g\rangle$ and the Kittel mode in an arbitrary magnon state $|\psi\rangle=\sum c_{n_\mathrm{m}}|n_\mathrm{m}\rangle$, is given by $$\hat X_\pi^0|g\psi\rangle=c_0|e0\rangle+\sum_{n_\mathrm{m}>0}c_{n_\mathrm{m}}|gn_\mathrm{m}\rangle,
\label{eq:conditional_excitation}$$ where $|in_\mathrm{m}\rangle=|i\rangle\otimes|n_\mathrm{m}\rangle$ is the state of the composite system with $|i=g,e\rangle$ and $|n_\mathrm{m}\rangle$ being the qubit states and the magnon Fock states, respectively. From Eq. (\[eq:conditional\_excitation\]), measuring the qubit in the ground state indicates the presence of *at least* a single magnon in the Kittel mode. The detection protocol represented schematically in Fig. 2A is composed of the entangling operation $\hat X_\pi^0$ and a readout of the qubit state. The fidelity of the entangling gate is mainly determined by the duration $\tau_\pi$ of the excitation, hereafter called the detection time [@Johnson2010; @Kirchmair2013]. Indeed, the excitation is conditional only if $\tau_\pi$ is such that the spectral width $\propto1/\tau_\pi$ is smaller than the amplitude of the shift per excitation $2\left|\chi_\mathrm{q-m}\right|$. The state of the qubit is read out using the high-power readout technique [@Reed2010], enabling single-shot readout with a fidelity $\mathcal{F}_\mathrm{r}\approx0.9$ without the use of near-quantum-limited amplifiers (Fig. 2B).
To benchmark the detection protocol, a coherent state of magnons $|\beta\rangle$ is initially prepared through a displacement operation $\hat D(\beta)=e^{\beta\hat c^\dagger-\beta^*\hat c}$ (Fig. 2A). The detection probability $p_g(\overline{n}_\mathrm{m})$ is then related to the magnon population $\overline{n}_\mathrm{m}$ through the probability $p_{n_\mathrm{m}\geq1}=1-e^{-\overline{n}_\mathrm{m}}$ of having at least a single magnon in the Kittel mode. More specifically, the detection probability is given by $$p_g(\overline{n}_\mathrm{m})=\eta\left(1-e^{-\overline{n}_\mathrm{m}}\right)+p_g(0),
\label{eq:detection_probability}$$ where $\eta$ and $p_g(0)$ are respectively the quantum efficiency and the dark-count probability, both critical figures in respect of evaluating the performance of the detector. Figure \[fig:Single\_shot\_detection\]C shows the detection probability $p_g(\overline{n}_\mathrm{m})$ obtained experimentally for a detection time $\tau_\pi=200$ ns. Fitting the data to Eq. (\[eq:detection\_probability\]), a quantum efficiency $\eta=0.71$ and a dark-count probability $p_g(0)=0.24$ are determined. Considering these values, if the Kittel mode is in the vacuum state $|0\rangle$, the probability that the detector *does not click* is $1-p_g(0)=0.76$ (ideally $1$). When the Kittel mode is in the Fock state $|1\rangle$, the detector *clicks* with a probability $\eta+p_g(0)=0.95$ (ideally $1$). These results constitute the first demonstration of the single-shot detection of a single magnon, thus bringing the equivalent of the single-photon detector to the field of magnonics.
Signatures of the mechanisms limiting the performance of the single-magnon detector are obtained by measuring the dark-count probability and the quantum efficiency for different detection times $\tau_\pi$ (Fig. 3). As shown in Fig. 3A, the dark-count probability $p_g(0)$ increases with the detection time $\tau_\pi$ due to the finite qubit relaxation time $T_1=0.80~\mu$s and coherence time $T_2^*=0.97~\mu$s. Furthermore, initialization and readout errors set a lower bound on the dark-count probability at $\approx0.08$. Figure \[fig:Detection\_efficiency\]B shows that the quantum efficiency $\eta$ increases for larger detection times due to an increase in the selectivity of the entangling operation between the qubit and the Kittel mode. For longer detection times, decoherence of the qubit limits the efficiency, leading to an optimal detection time at $\tau_\pi\approx200$ ns. Two relevant upper bounds on the quantum efficiency are satisfied (Fig. \[fig:Detection\_efficiency\]B). First, as the magnons are detected by using the qubit as the quantum sensor, the quantum efficiency is bounded by the qubit readout fidelity $\mathcal{F}_\mathrm{r}\approx0.9$. Secondly, the dark-count probability $p_g(0)$ sets an upper limit on the quantum efficiency at $\eta\leq1-p_g(0)$ through the probability $\eta+p_g(0)$ of detecting the single magnon Fock state $|1\rangle$.
As shown in Fig. 3, numerical simulations of the detection protocol are in good agreement with the experimental results without any fitting parameters (see supplementary materials). Therefore, we use the numerical model to determine the effect of qubit initialization, control, readout, and entangling errors on the dark-count probability $p_g(0)$ and detection inefficiency $1-\eta$ (Table \[tab:Error\_budget\]). Notably, qubit decoherence constitutes the primary source of error limiting the performance of the detector. A dark-count probability below $0.03$ and a quantum efficiency above $0.96$ should be within experimental reach with an improved single-magnon detector (see supplementary materials).
---------------------- ------------------------ --------------
Source of error
Dark-count probability Inefficiency
$p_g(0)$ $1-\eta$
Qubit initialization $0.032$ $0.023$
Qubit decoherence $0.15$ $0.21$
Qubit readout $0.024$ $0.061$
Entanglement $-$ $0.039$
Total $0.22$ $0.33$
Experiment $0.24$ $0.29$
---------------------- ------------------------ --------------
: **Error budget**. Contributions from different sources of error determined from numerical simulations for a detection time $\tau_\pi=200$ ns. The total error is not equal to the sum of the listed errors due to additional errors and multiple error processes (see supplementary materials). \[tab:Error\_budget\]
The performances of the detector can also be improved, without any hardware modifications, by considering an alternative detection scheme. Instead of detecting the presence of *at least* one magnon ($n_\mathrm{m}=1,2,\dots$) with the protocol of Fig. \[fig:Single\_shot\_detection\]A, the presence of *exactly* one magnon ($n_\mathrm{m}=1$) can be detected using the conditional operation $\hat X_\pi^1$ that excites the qubit only if there is exactly a single magnon in the Kittel mode [@Johnson2010; @Narla2016]. In the limit where the probability of having more than one magnon is negligible, both protocols detect the presence of a single magnon. Experimentally, the conditional excitation $\hat X_\pi^1$ is realized by attempting to excite the qubit at its frequency with a single magnon in the Kittel mode, $\omega_\mathrm{q}^1$. As shown in Fig. \[fig:Detection\_efficiency\]C, the detection of exactly a single magnon enables us to reduce the dark-count probability by half to $0.12$. Indeed, in this scheme, qubit decoherence does not contribute significantly to the dark-count probability as the qubit is never actually excited in the absence of magnons. Nevertheless, as shown in Fig. \[fig:Detection\_efficiency\]D, the quantum efficiency is very similar for both schemes (see supplementary materials). A good agreement between the experimental and numerical results is found without any fitting parameters, highlighting a good understanding of the physics at play.
The high-fidelity detection of a single magnon, corresponding to a precession of the magnetization vector of the millimeter-sized ferromagnetic crystal with an angle of $\sim10^{-17}$ degrees, represents a significant advance for magnonics and quantum technologies based on magnetism. The magnon detection can be made quantum non-demolition (QND) with a QND readout of the qubit state [@Johnson2010; @Kono2018]. The relaxation and coherence times of superconducting qubits in quantum magnonics, currently limiting the performance of the single-magnon detector, could be enhanced by reducing internal losses of the microwave cavity modes.
Near-term applications include the heralded probabilistic creation of quantum states of magnons, a critical step towards the development of a magnon-based quantum transducer [@Hisatomi2016; @Lachance-Quirion2019]. Furthermore, the single-magnon detector could help to uncover weak magnon excitation processes such as the potential excitation of magnons from galactic axions [@Flower2019]. In the longer term, the development of planar devices [@McKenzie-Sell2019] for the integration of single-magnon detectors could represent the ultimate limit to the conversion between magnons and electrical signals for emerging technologies such as magnon spintronics [@Chumak2015].
[10]{}
C. L. Degen, F. Reinhard, P. Cappellaro, [*Reviews of Modern Physics*]{} [ **89**]{}, 035002 (2017).
J. M. Boss, K. S. Cujia, J. Zopes, C. L. Degen, [*Science*]{} [**356**]{}, 837 (2017).
B. R. Johnson, [*et al.*]{}, [*Nature Physics*]{} [**6**]{}, 663 (2010).
N. Zhao, [*et al.*]{}, [*Nature Nanotechnology*]{} [**7**]{}, 657 (2012).
S. Kono, K. Koshino, Y. Tabuchi, A. Noguchi, Y. Nakamura, [*Nature Physics*]{} [**14**]{}, 546 (2018).
G. Kurizki, [*et al.*]{}, [*Proceedings of the National Academy of Sciences*]{} [**112**]{}, 3866 (2015).
A. Blais, R.-S. Huang, A. Wallraff, S. M. Girvin, R. J. Schoelkopf, [ *Physical Review A*]{} [**69**]{}, 062320 (2004).
M. H. Devoret, R. J. Schoelkopf, [*Science*]{} [**339**]{}, 1169 (2013).
Y. Chu, [*et al.*]{}, [*Science*]{} [**358**]{}, 199 (2017).
K. J. Satzinger, [*et al.*]{}, [*Nature*]{} [**563**]{}, 661 (2018).
Y. Chu, [*et al.*]{}, [*Nature*]{} [**563**]{}, 666 (2018).
Y. Tabuchi, [*et al.*]{}, [*Science*]{} [**349**]{}, 405 (2015).
Y. Tabuchi, [*et al.*]{}, [*Comptes Rendus Physique*]{} [**17**]{}, 729 (2016).
D. Lachance-Quirion, [*et al.*]{}, [*Science Advances*]{} [**3**]{}, e1603150 (2017).
D. Lachance-Quirion, Y. Tabuchi, A. Gloppe, K. Usami, Y. Nakamura, [*Applied Physics Express*]{} [**12**]{}, 070101 (2019).
A. V. Chumak, V. I. Vasyuchka, A. A. Serga, B. Hillebrands, [*Nature Physics*]{} [**11**]{}, 453 (2015).
H. Huebl, [*et al.*]{}, [*Physical Review Letters*]{} [**111**]{}, 127003 (2013).
Y. Tabuchi, [*et al.*]{}, [*Physical Review Letters*]{} [**113**]{}, 083603 (2014).
J. Koch, [*et al.*]{}, [*Physical Review A*]{} [**76**]{}, 042319 (2007).
H. Paik, [*et al.*]{}, [*Physical Review Letters*]{} [**107**]{}, 240501 (2011).
J. Gambetta, [*et al.*]{}, [*Physical Review A*]{} [**74**]{}, 042318 (2006).
D. I. Schuster, [*et al.*]{}, [*Nature*]{} [**445**]{}, 515 (2007).
L. R. Sletten, B. A. Moores, J. J. Viennot, K. W. Lehnert, [*Physical Review X*]{} [**9**]{}, 021056 (2019).
G. Kirchmair, [*et al.*]{}, [*Nature*]{} [**495**]{}, 205 (2013).
M. D. Reed, [*et al.*]{}, [*Physical Review Letters*]{} [**105**]{}, 173601 (2010).
A. Narla, [*et al.*]{}, [*Physical Review X*]{} [**6**]{}, 031036 (2016).
R. Hisatomi, [*et al.*]{}, [*Physical Review B*]{} [**93**]{}, 174427 (2016).
G. Flower, J. Bourhill, M. Goryachev, M. E. Tobar, [*Physics of the Dark Universe*]{} [**25**]{}, 100306 (2019).
L. McKenzie-Sell, [*et al.*]{}, [*Physical Review B*]{} [**99**]{}, 140414(R) (2019).
J. R. Johansson, P. D. Nation, F. Nori, [*Computer Physics Communications*]{} [**184**]{}, 1234 (2013).
H.-P. Breuer, F. Petruccione, [*[The Theory of Open Quantum Systems]{}*]{} (Oxford University Press, 2002).
M. Boissonneault, J. M. Gambetta, A. Blais, [*Physical Review Letters*]{} [**105**]{}, 100504 (2010).
D. Rist[è]{}, J. van Leeuwen, H.-S. Ku, K. Lehnert, L. DiCarlo, [*Physical Review Letters*]{} [**109**]{}, 050507 (2012).
P. Magnard, [*et al.*]{}, [*Physical Review Letters*]{} [**121**]{}, 060502 (2018).
T. Walter, [*et al.*]{}, [*Physical Review Applied*]{} [**7**]{}, 054020 (2017).
K. Juliusson, [*et al.*]{}, [*Physical Review A*]{} [**94**]{}, 063861 (2016).
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors would like to thank Arjan van Loo for fruitful discussions and Jacob Koenig for carefully reading the manuscript. This work is partly supported by JSPS KAKENHI (26220601, 18F18015), JST ERATO (JPMJER1601), FRQNT Postdoctoral Fellowships, and MEXT Monbukagakusho Scholarship. D.L.-Q. is an International Research Fellow of JSPS. **Authors contributions:** D.L.-Q. and S.P.W performed the experiments and analyzed the data; D.L.-Q. performed the numerical simulations with assistance from S.K; Y.T and Y.N. conceived the hybrid system; and K.U. and Y.N. advised on all efforts. All authors contributed to discussions and production of the manuscript.
Supplementary materials {#supplementary-materials .unnumbered}
=======================
Supplementary Text\
Figs. S1 to S11\
Tables S1 to S5\
References *(30-36)*
![**Strong dispersive regime of quantum magnonics.** (**A**) Interaction of strength $g_\mathrm{q-m}$ between the Kittel mode (with frequency $\omega_\mathrm{m}$) of a spherical ferrimagnetic crystal of YIG and a superconducting qubit ($\omega_\mathrm{q}$), engineered through magnetic- and electric-dipole couplings to a microwave cavity mode ($\omega_\mathrm{c}$). The ferrimagnetic sphere is magnetized with an external magnetic field $\mathbf{B}_0$. (**B**) Normalized qubit spectrum measured as a function of the coil current. Dashed lines are guides for the eye. Right: qubit spectrum measured at $\omega_\mathrm{q}\approx\omega_\mathrm{m}$. The line shows a fit to the data. (**C**) Ramsey interferometry protocol to probe the qubit in the presence of a continuous excitation of $\overline{n}_\mathrm{m}$ magnons in the Kittel mode. (**D**) Probability $p_e$ of the qubit being in the excited state $|e\rangle$ as a function of the free evolution time $\tau$ in the presence of $\overline{n}_\mathrm{m}=0.53$ magnons in the Kittel mode. (**E**) Normalized qubit spectrum, obtained from the Fourier transform of $p_e$, indicating a strong dispersive interaction between the Kittel mode and the qubit. The black line shows a fit to the data. The blue (red) line and shaded area show the spectral component corresponding to the magnon vacuum state $|0\rangle$ (Fock state $|1\rangle$). \[fig:Strong\_dispersive\_regime\]](Strong_dispersive_regime.pdf)
![**Single-shot detection of single magnons.** (**A**) Protocol to detect the presence of at least a single magnon. The Kittel mode and the qubit are entangled through the qubit excitation $\hat X_\pi^0$ conditional on the Kittel mode being in the vacuum state. To characterize the detection protocol, a coherent state of magnons is initially prepared through a displacement operation $\hat D(\beta)$. The state of the qubit is read out at the end of the protocol. (**B**) Histograms of the demodulated qubit readout signal for $10^5$ single shots when preparing the qubit in the ground state $|g\rangle$ (excited state $|e\rangle$) obtained with the high-power readout technique. The vertical dashed line (dotted line) indicates the demodulated signal corresponding to the qubit occupying the ground state $|g\rangle$ (excited state $|e\rangle$). (**C**) Detection probability $p_g(\overline{n}_\mathrm{m})$ as a function of the magnon population $\overline{n}_\mathrm{m}$. The solid black line shows a fit to Eq. (\[eq:detection\_probability\]), indicating a magnon detection efficiency $\eta=0.71$ and a dark-count probability $p_g(0)=0.24$ (dashed black line) for a detection time $\tau_\pi=200$ ns. Error bars are smaller than the symbols. \[fig:Single\_shot\_detection\]](Single_shot_detection_fig.pdf)
![**Characterization of the single-magnon detector.** (**A** and **B**) Dark-count probability $p_g(0)$ (A) and quantum efficiency $\eta$ (B) as a function of the detection time $\tau_\pi$. Results from numerical simulations are shown as solid lines. In (A), the dashed line shows the dark-count probability due to initialization and readout errors. In (B), the dotted and dot-dashed lines indicate the limits on the quantum efficiency set by readout errors and the dark-count probability, respectively. Error bars are smaller than the symbols. (**C** and **D**) Dark-count probability (C) and quantum efficiency (D) for the detection of at least a single magnon ($n_\mathrm{m}\geq1$) and of exactly a single magnon ($n_\mathrm{m}=1$) for a detection time $\tau_\pi=200$ ns. \[fig:Detection\_efficiency\]](Detection_efficiency.pdf)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Mesoscopic systems provide us a unique experimental stage to address nonequilibrium quantum statistical physics. By using a simple tunneling model, we describe the electron exchange process via a quantum coherent conductor between two reservoirs, which yields the fluctuation theorem (FT) in mesoscopic transport. We experimentally show that such a treatment is semiquantitatively validated in the current and noise measurement in an Aharonov-Bohm ring. The experimental proof of the microreversibility assumed in the derivation of FT is presented.'
author:
- 'Shuji Nakamura$^{1}$, Yoshiaki Yamauchi$^{1}$, Masayuki Hashisaka$^1$, Kensaku Chida$^1$, Kensuke Kobayashi$^1$'
- 'Teruo Ono$^1$ Renaud Leturcq$^2$, Klaus Ensslin$^3$, Keiji Saito$^4$, Yasuhiro Utsumi$^5$, Arthur C. Gossard$^6$'
title: Fluctuation Theorem and Microreversibility in a Quantum Coherent Conductor
---
Introduction
============
Since the 1980’s mesoscopic conductors have been serving as an ideal stage to investigate the quantum scattering problem both theoretically and experimentally, because the quantum transport through a single site can be precisely probed in electronic measurement [@DattaETMS]. The Landauer-B[ü]{}ttiker formalism embodies this advantage of mesoscopic physics, as was successfully applied to the Aharonov-Bohm ring, the quantum point contact, and the quantum dot, through which the mesoscopic physics has been established \[see Fig. 1(a)\]. Not only the current averaged over for a certain time ($\langle I \rangle$), but also the current fluctuation $\langle(\delta I)^2 \rangle$ due to the partition process is treated in the same framework [@ButtikerPRB1992; @MartinPRB1992; @BlanterPR2000]. Actually, the quantum shot noise measurement was successfully demonstrated, for example, to provide the direct proof of the fractional charge [@de-PicciottoNature1997; @SaminadayarPRL1997] and the Cooper pair [@JehlNature2000] by looking at how carriers are scattered at a mesoscopic conductor.
These days the mesoscopic transport is invoking much interest from another point of view. As the electron transport can be viewed as the electron exchange process between the reservoirs via the conductor as shown in Fig. 1(b), it serves as a well-defined test stage for nonequilibrium quantum statistical physics [@EspositoRMP2009]. The unique advantage of this approach lies in that the degree of nonequilibrium can be finely tuned by the bias voltage applied to the conductor. In addition, many events, namely numerous electron exchange processes, can be monitored, which enables us to perform precise measurement.
![(a) Schematic picture of the mesoscopic conductor coupled to the two reservoirs in the nonequilibrium regime. $\mu_L$ and $\mu_R$ are the chemical potentials of the left and the right reservoirs, respectively. Mesoscopic transport based on the Landauer-B[ü]{}ttiker picture is schematically shown. When a conductor is biased, electrons injected from one of the reservoirs are either transmitted or reflected at the conductor, which yields finite current fluctuation \[$\langle(\delta
I)^2\rangle\neq 0$\]. (b) The transport can be also viewed as the electron exchange process between the two reservoirs.](FluctuationMicroreversibility_Fig1.eps){width=".99\linewidth"}
To quantitatively address the above topic, the fluctuation theorem (FT) [@EvansPRL1993] is believed to play a central role [@EspositoRMP2009; @TobiskaPRB2005; @AndrieuxJCP2004; @AndrieuxNJP2009; @SaitoPRB2008; @ForsterPRL2008; @Forster_arXiv; @UtsumiPRB2010; @CampisiPRL2010; @AltlandPRB2010]. Based on microscopic reversibility (“microreversibility” or detailed balance), this relation exactly links the probabilities of the production and consumption of the entropy in a given system that is coupled to the reservoir. FT corresponds to a microscopic expansion of the macroscopic second law of thermodynamics and is proven to yield the linear-response theory [@GallavottiPRL1996], and the Onsager-Casimir relations [@SaitoPRB2008]. FT was experimentally proved to be valid in classical systems such as a colloidal particle in fluid [@WangPRL2002] and a resistor [@GarnierPRE2005]. Although it was extended to the quantum regime [@KurchanArxiv], an experimental check in this regime was still lacking. More recently, FT was theoretically addressed in the mesoscopic transport [@TobiskaPRB2005; @AndrieuxJCP2004; @AndrieuxNJP2009; @UtsumiPRB2010; @CampisiPRL2010; @AltlandPRB2010] even in the presence of the magnetic field [@SaitoPRB2008; @ForsterPRL2008; @Forster_arXiv] and was indeed shown to be relevant in the analysis [@UtsumiPRB2010] of the electron counting experiments [@GustavssonPRL2006; @FujisawaScience2006]. While the incoherent tunneling events across the quantum dot(s) were investigated in the above experiments, the validity of FT in the quantum coherent regime was left to be addressed.
Recently, we experimentally showed the presence of nontrivial relations between the nonlinear response and the nonequilibrium fluctuation in the coherent transport of an Aharonov-Bohm (AB) ring [@NakamuraPRL2010]. When the current $I$ and the current fluctuation (current noise power spectral density) $S$ are expanded in the Taylor series as a polynomial of the bias voltage $V$, $$I(V, B) = G_1(B) V + \frac{1}{2!} G_2(B) V^2 + \frac{1}{3!} G_3(B) V^3 + \cdots,
\label{EqPolyI}$$ and $$S(V, B) = S_0(B) + S_1(B) V + \frac{1}{2!} S_2(B) V^2 + \cdots,
\label{EqPolyS}$$ we showed that there are proportional relations of $S_1^S \propto G_2^S$ and $S_1^A \propto G_2^A$. Here, the coefficients that are symmetrized ($S$) or antisymmetrized ($A$) with respect to the magnetic field reversal are defined as $$G_2^{S, A} (B)\equiv G_2 (B) \pm G_2 (-B),$$ and $$S_1^{S, A} (B)\equiv S_1 (B) \pm S_1 (-B)$$ (take $+$ and $-$ for $S$ and $A$, respectively). This result is beyond the consequence of the fluctuation-dissipation theorem $S_0(B) = 4 k_B T
G_1 (B)$. Our observation semiquantitatively agrees with the theoretical prediction on the basis of FT [@SaitoPRB2008] and provides an evidence of FT in the nonequilibrium quantum regime.
In this paper we expand the above work to further support our previous report. In Sec. II, based on a simple tunneling model, we derive FT in an applicable form to simple mesoscopic conductors. In Sec. III, we discuss the breakdown of the Onsager-Casimir reciprocity in the nonequilibrium regime in the presence of the magnetic field. Then, as a fundamental aspect of FT in mesoscopic transport, we show that the validity of the microreversibility can be directly addressed in a quantum regime.
Fluctuation Theorem in a Mesoscopic System
==========================================
Zero magnetic field case
------------------------
We explain FT by using the simplest setup and deduce the aforementioned nonequilibrium fluctuation relations. We consider a mesoscopic conductor, say a quantum point contact, where the two quantum wires are coupled by tunneling. While more systematic and general derivation for these relations is performed by using a cumulant generating function [@AndrieuxNJP2009; @UtsumiPRB2010; @CampisiPRL2010; @AltlandPRB2010; @SaitoPRB2008; @ForsterPRL2008; @Forster_arXiv], the present simplest model is sufficiently instructive to treat here. We assume that no energy relaxation takes place inside the conductor, which is fulfilled in many mesoscopic devices smaller than the energy relaxation length such as a quantum dot, chaotic cavity, ring, and so on. First we treat the zero-magnetic field case to show $S_1 = 2 k_B T
G_2$ in Eqs. (\[EqPolyI\]) and (\[EqPolyS\]). The relations between the coefficients in the current and the current noise are schematically shown in Fig. 2.
![ (color online) (a) Current-voltage characteristic as a function of the bias voltage $V$. While Ohm’s law holds around $V = 0$, the current $I$ is not linear at large $V$ and the $I$-$V$ characteristics can be decomposed into a polynomial of $V$ with coefficients $G_1$, $G_2$, $G_3 \dots$ as in Eq. (\[EqPolyI\]). In this study we address $G_2$. This schematic graph shows the total current $I$, the $G_1V$ contribution, and the $1/2! G_2 V^2$ contribution, in the solid, dashed, and dotted curves, respectively. The case with a negative $G_2$ is shown. (b) Similarly, the current noise spectral density $S$ (shown in the solid curve) can be expressed in a polynomial form of $V$ as in Eq. (\[EqPolyS\]). The Johnson-Nyquist relation tells that $S(V = 0) =
S_0 = 4k_BTG_1$. This schematic graph shows the case with a negative $S_1$ in the dashed curve. Here we address the coefficient of the term linear in $V$ ($S_1$) and the relation between $S_1$ and $G_2$. ](FluctuationMicroreversibility_Fig2.eps){width=".99\linewidth"}
The present system is described by the following Hamiltonian: $$\begin{aligned}
H &=& H_L + H_R + H_{LR},\end{aligned}$$ where $H_L$ and $H_R$ are the Hamiltonian of the left and right quantum wires and $H_{LR}$ is the tunneling part between them. The initial density matrix is decoupled into the equilibrium states of each wire, where the left and right wires are assumed to have equal temperature $1/\beta = k_BT$ and have chemical potentials $\mu_L$ and $\mu_R$, respectively. Then the whole density matrix is described by $$\begin{aligned}
\hat{\rho}_{\rm initial} &=& \sum_{n_L , n_R } \rho_{n_L, n_R }
|n_L , n_R\rangle \langle n_L , n_R | , \\
\rho_{n_L, n_R } &=&
{e^{-\beta [E_{n_L} - \mu_{L} \, n_{L } ] } \over Z_{L}}
{e^{-\beta [E_{n_R} - \mu_{R} \, n_{R } ] } \over Z_{R}},\end{aligned}$$ where $Z_L$ and $Z_R$ are the normalization factors, and $|n_L , n_R
\rangle$ defines the state that $n_L$ and $n_R$ electrons are present inside the left and right wires with the eigenenergies $E_{n_L}$ and $E_{n_R}$ of $H_L$ and $H_R$, respectively.
The probability of finding the state $|n_L ' , n_R '\rangle $ after a certain time $\tau$ starting from the initial state $|n_L , n_R \rangle$ is expressed as $$\begin{aligned}
P_{(n_L , n_R )\to (n_L' , n_R')}
\!=\! |\langle n_L' , n_R' | e^{{-i\tau \over \hbar}H } | n_L , n_R \rangle |^2
\rho_{n_L , n_R} . ~\end{aligned}$$ The microreversibility or the time reversal symmetry is given by $$\begin{aligned}
|\langle n_L' , n_R' | e^{{-i\tau \over \hbar } H } | n_L , n_R \rangle |^2 \!=\!
|\langle n_L , n_R | e^{{-i\tau \over \hbar } H } | n_L', n_R'\rangle |^2 .~\end{aligned}$$ Here for simplicity we assume that $| n_L , n_R \rangle$ and $| n_L' ,
n_R' \rangle$ has the time reversal symmetry as the electron numbers are the good quantum number, while in the general treatment [@SaitoPRB2008] this assumption is not necessary.
As the electron number conservation $n_{L}- n_L' = -(n_R - n_R') $ and the energy conservation are satisfied at very large $\tau$, $E_{n_L'}
-E_{n_L} \approx -(E_{n_{R}' } - E_{n_R } ) $. Using the microreversibility and the conservation laws, we find the relation $$\begin{aligned}
P_{(n_L ,n_R)\to (n_L' , n_R') }
&=& P_{(n_L' ,n_R')\to (n_L , n_R) } e^{A (n_L - n_{L}')}, \label{dft}\end{aligned}$$ where $A$ is an affinity $A=\beta (\mu_L - \mu_R)$. The probability that the number of the transmitted electron is $Q$, is defined as $P(Q)
= \sum_{n_L , n_R , n_L' , n_R'}P_{(n_L ,n_R)\to (n_L' , n_R') } \delta
(Q - (n_L - n_{L}'))$. Therefore, FT is obtained as the direct consequence of the microreversibility $$\begin{aligned}
P(Q) &=& P(-Q) e^{AQ} \label{ft}.\end{aligned}$$ This microreversibility ensures the following sum rule, which is called “global detailed balance” in Ref. \[\], $$\begin{aligned}
\langle e^{AQ} \rangle = 1,
\label{EqGlobal}\end{aligned}$$ since $1 = \sum_Q P(Q) = \sum_Q P(-Q) e^{AQ} = \langle e^{AQ} \rangle$. Here, $\langle\cdots\rangle$ denotes the expectation $\langle\cdots\rangle \equiv \sum_Q \cdots P(Q)$.
Now let us discuss the higher order correlations between the current and its noise power, which are the central topic in the present paper. With FT (\[ft\]), we find the following identity $$\begin{aligned}
\langle Q \rangle &=& \sum_Q Q P(Q) =
- \sum_Q Q P(Q) e^{-A Q} \nonumber \\
&=&
-\langle Q \rangle + A \langle Q^2 \rangle -{A^2\over 2! }\langle Q^3 \rangle +
\cdots . \end{aligned}$$ Furthermore, we note that $\langle Q^n \rangle$ can be expanded in the Taylor series of $A$ with the coefficients $\langle Q^n \rangle_m$ ($n,
m$ integer) $$\begin{aligned}
\langle Q^n \rangle &=& \langle Q^n \rangle_0 + A \langle Q^n \rangle_1
+ {A^2\over 2!} \langle Q^n \rangle_2 + \cdots .\end{aligned}$$ Comparing order by order with respect to $A$, we find infinite number of relationships among these coefficients, some of which are given as $$\begin{aligned}
\langle Q^2 \rangle_0 &=& 2 \langle Q \rangle_1 , \label{1st}\\
\langle Q^2 \rangle_1 &=& \langle Q \rangle_2 . \label{2nd}\end{aligned}$$
Averaged current $I$ and current noise power $S$ are defined as $I=\langle Q \rangle /\tau$ and $S = 2(\langle Q^2 \rangle - \langle Q
\rangle^2 )/\tau~$ [@NoteFactor2]. The first relation (\[1st\]) is equivalent to the fluctuation dissipation relations [@GallavottiPRL1996] $$\begin{aligned}
S_0 = 4 k_{\rm B} T G_1,\end{aligned}$$ and the second relation (\[2nd\]) is to $$\begin{aligned}
S_1 = 2 k_{\rm B} T G_2.
\label{S1G2}\end{aligned}$$ This relation is beyond the fluctuation-dissipation relation and directly links the nonlinearity and the nonequilibrium of the system.
Finite magnetic field case
--------------------------
At $B\neq 0$, the microreversibility requires that the probability $P(Q,B)$ should satisfy [@SaitoPRB2008] $$\begin{aligned}
P(Q,B) = P(-Q, -B)\exp(AQ).
\label{PQB}\end{aligned}$$ $P(Q,B)$ is now decomposed to the symmetric and antisymmetric parts regarding the magnetic field reversal; $P_{\pm} (Q) \equiv P(Q,B) \pm
P(Q,-B)$, which fulfill $$\begin{aligned}
P_{\pm}(Q) = \pm P_{\pm}(-Q)e^{AQ}.\end{aligned}$$ Although the symmetric part $P_{+}(Q)$ produces the same fluctuation relations as $P(Q)$ does, the antisymmetric probability gives rise to a nontrivial result. By considering the antisymmetrized number of charges exchanged between the reservoirs, $$\begin{aligned}
\langle Q_{-} \rangle \equiv \sum_Q Q P_{-}(Q) = \sum_Q Q P_{-}(Q) e^{-AQ}\end{aligned}$$ and defining $\langle Q_{-}^n \rangle_m$ with nonnegative integers $n$ and $m$ as the coefficients in the Taylor expansion of the above $\langle
Q_{-} \rangle$ with regard to $A$, we obtain $$\begin{aligned}
\langle Q_{-}^3 \rangle_0 = 2\langle Q_{-}^2 \rangle_1.
\label{Asym1}\end{aligned}$$ By noting the following relation, which is the consequence of the normalization condition $\sum_Q P(Q)=1$, $$\begin{aligned}
0 = \sum_Q P_{-}(Q) = - \sum_Q P_{-}(Q) e^{-AQ},
\label{EqGlobalQ-}\end{aligned}$$ we obtain $$\begin{aligned}
3\langle Q_{-} \rangle_2 - 3\langle Q_{-}^2 \rangle_1 + \langle
Q_{-}^3 \rangle_0 = 0.
\label{Asym2}\end{aligned}$$
The current that is antisymmetrized with regard to the $B$ reversal is defined as $I(V, B)-I(V, -B) = \langle Q_{-} \rangle/\tau$ and the current noise power is also defined in the same way, Eqs. (\[Asym1\]) and (\[Asym2\]) yields $$\begin{aligned}
S_1^A =C_0^A/2k_BT
\label{WMicroRev}\end{aligned}$$ and $$\begin{aligned}
S_1^A-2k_BTG_2^A = C_0^A/3k_BT,
\label{WOMicroRev}\end{aligned}$$ respectively. Here, $C_0^A$, which originates from the term $\langle
Q_{-}^3 \rangle_0$, is the antisymmetric part of the third cumulant at equilibrium. These two yield the antisymmetric relation expressed by $$\begin{aligned}
S_1^A = 6k_BTG_2^A.\end{aligned}$$
The above deduction totally relies on the microreversibility as is the case in a systematic derivation [@SaitoPRB2008]. Recently, however, an interesting possibility of the broken microreversibility in mesoscopic conductors is pointed out [@ForsterPRL2008; @Forster_arXiv]. It was discussed that, because of the global detailed balance expressed by Eq. (\[EqGlobal\]), the sum rule Eq. (\[EqGlobalQ-\]) and hence Eq. (\[Asym2\]) hold true without microreversibility, even if we do not resort to the relation $P_{-} (Q) = -P_{-} (-Q) \exp(AQ)$ (Eq. (\[PQB\])). In this case, Eq. (\[Asym1\]) and the resultant Eq. (\[WMicroRev\]) are no more valid and only Eq. (\[WOMicroRev\]) is expected. To address this issue experimentally is the main motivation of the present paper.
The conventional current and shot noise formulas in the Landauer-B[ü]{}ttiker framework can be also expressed in the polynomial form of $V$ {see Eqs. (39) and (61) in Ref. \[\]}. By taking the energy-dependent transmission into account, a relation similar to Eq. (\[S1G2\]) holds true. However, this approach, which is based on the transmission defined in the equilibrium, fails to explain the nonlinear conductance that is not symmetric with respect to the magnetic field reversal. Indeed, it is established theoretically and experimentally [@SanchezPRL2004; @SpivakPRL2004; @WeiPRL2005; @AngersPRB2007; @LeturcqPRL2006] that due to electron-electron interactions induced in biased mesoscopic conductors, the Onsager-Casimir reciprocal relations are broken, leading to finite $G_2^A$. We will show the experimental data regarding this below.
Magnetic field asymmetry and microreversibility
===============================================
Experiment
----------
We used an Aharonov-Bohm (AB) ring as a typical coherent conductor. Figure 3(a) shows an atomic force microscope (AFM) image of the AB ring fabricated by local oxidation using an AFM [@HeldAPL1998] on a GaAs/AlGaAs heterostructure two-dimensional electron gas (2DEG) (the electron density $3.7 \times 10^{11}$ cm$^{-2}$, the mobility $2.7
\times 10^5$ cm$^2$/Vs and the electron mean free path $2.7$ $\mu$m). The two-terminal current and noise measurement setup in a dilution refrigerator is also shown in Fig. 3(a). The in-plane gates defined by the oxide lines are grounded in the present measurement. The 2DEG has a back gate to tune the electron density and the conductance of the AB ring can be modulated by the back gate voltage $V_g$ and the magnetic field $B$ by the AB effect. Figure 3(b) shows the image plot of the conductance as a function of $V_g$ and $B$. The upper panel of Fig. 3(b) presents the conductance at $V_g =-0.09$ V displaying clear AB oscillations with an oscillation period being 25 mT in agreement with the ring radius of 230 nm [@LeturcqPRL2006; @YamauchiPRB2009]. The conductance of the ring ranges between 1.3 and 1.7 in units of $2e^2/h
\sim (12.9$ k$\Omega)^{-1}$ with typical visibility of $\sim 0.13$. The presence of electron interferences guarantees the coherent electron transport in the device.
![ (color online) (a) Atomic force microscope (AFM) image of the AB ring fabricated by local oxidation using an AFM [@HeldAPL1998] on a GaAs/AlGaAs 2DEG with the experimental setup for the two-terminal current and noise measurements. (b) Image plot of the conductance of the ring as a function of $V_g$ and $B$ (in the lower panel). The upper panel shows the magnetoconductance at $V_g=-0.09$ V as indicated by a white line in the lower panel. (c) Conductance as a function of $V_g$ at $B = \pm 25$ mT \[indicated by dashed lines in the lower panel of (b)\]. (d) Equilibrium noise ($S_0$) as a function of $V_g$ at $B = \pm
25$ mT. ](FluctuationMicroreversibility_Fig3.eps){width=".99\linewidth"}
In addition to the dc measurement, we performed the noise measurement as follows \[also see Fig. 3(a)\]. The voltage fluctuation across the sample on the resonant circuit, whose resonant frequency is about 3.0 MHz with the bandwidth of $\sim 140$ kHz, is extracted as an output signal of the cryogenic amplifier [@de-PicciottoNature1997; @DiCarloRSI2006; @HashisakaPRB2008; @NakamuraPRB2009; @NakamuraPRL2010]. The time-domain signal is then captured by the two-channel digitizer, and is converted to spectral density data via FFT. To increase the resolution of the noise spectrum, we performed the cross-correlation technique by using two sets of resonant circuit and amplifier. The sample was placed in a dilution refrigerator whose base temperature is 45 mK and the electron temperature in the equilibrium was 125 mK as deduced from the thermal noise. By numerically fitting the obtained resonant peak, the current noise power spectral density $S$ is obtained as performed in Ref. .
In the analysis of the current $I$ and the current noise $S$ as polynomials of $V$, the bias window was set to $|eV| \leq 50$ $\mu$eV, which corresponds to $4.6 k_B T$ at $T=125$ mK. In this bias range, the Joule heating is expected to be negligible as seen in previous shot noise measurements for mesoscopic devices [@HashisakaPRB2008; @NakamuraPRB2009]. The coefficients in Eqs. (\[EqPolyI\]) and (\[EqPolyS\]) are deduced from the numerical fitting to the obtained current and current noise. The polynomial fitting for $I$ and $S$ was performed by taking up to the fifth order of $V$ for $I$ and up to the fourth order of $V$ for $S$ into account, respectively. The analysis up to third or seventh order of $V$ for $I$ and second order of $V$ for $S$ yields results consistent with those presented below. We note that the measurement was carefully performed at several different $V_g$’s and $B$’s, and all the results are in a quantitative agreement with each other within the experimental accuracy of the present work.
Results and Discussions
-----------------------
Figure 3(c) shows the zero-bias conductance $G_1$ obtained at $B=25$ mT and $B=-25$ mT at 125 mK as a function of the back gate voltage $V_g$. Since $V_g$ modulates the electron density in the ring hence the interference pattern, the conductance fluctuates as $V_g$ varies. As the Onsager-Casmir reciprocity tells, $G_1$ behaves similarly at $B=25$ and $-25$ mT as $V_g$ changes. The correlation factor ($CF$) between the two, which is the covariance of the two divided by the product of their standard deviations, is 0.91. Similarly, as shown in Fig. 3(d) the gate-dependent thermal noises ($S_0$) at $B=25$ mT and $-25$ mT lap over each other with $CF= 0.68$. Also we note that the proportionality between $G_1$ and $S_0$ indicates that $S_0 = 4 k_BTG_1$ holds with an electron temperature of $T= 125$ mK. The coefficients of the first term in Eqs. (\[EqPolyI\]) and (\[EqPolyS\]) satisfy the Onsager-Casimir reciprocity as a fundamental property in the equilibrium.
{width=".99\linewidth"}
Next we discuss the coefficients in the second term of Eqs. (\[EqPolyI\]) and (\[EqPolyS\]). Figures 4(a) and (b) show $G_2$ and $S_1$ at $B=25$ mT and $-25$ mT, respectively. It is remarkable that unlike the equilibrium property ($G_1$ and $S_0$), $G_2$ and $S_1$ are not symmetric with respect to the magnetic field reversal. Indeed $CF$’s between the traces for the negative and positive fields are as low as $0.20$ and $0.38$ for $G_2$ and $S_1$, respectively. Regarding $G_2$, the presence of this asymmetry was reported recently as the signature of the electron-electron correlation effect induced in a biased mesoscopic conductor [@WeiPRL2005; @AngersPRB2007; @LeturcqPRL2006] The noise measurement clearly tells that $S_1$ is also not symmetric with respect to the magnetic field reversal.
Figures 4(c) and (d) show $G_2^S$ and $G_2^A$ in the left axis as a function of $V_g$, respectively, where $S_1^S$ and $S_1^A$ are superposed in the right axis. Clearly, there appears strong correlation between $G_2^S$ and $S_1^S$ and between $G_2^A$ and $S_1^A$ with $CF =
0.84$ and $0.85$, respectively. As the theory predicts that $S_1^S =
2k_BTG_2^S$ and $S_1^A = 6k_BTG_2^A$, Fig. 4(e) and (f) shows the plots to compare between the theory and the experiment. The dotted lines are the prediction. As is consistent with the previous report [@NakamuraPRL2010], the symmetric part deviates from the theory while the antisymmetric part in Fig. 4(f) is in better agreement with the theory than the symmetric one in Fig. 4(e). For the presented data set, $S_1^S / 2k_BTG_2^S =6.00^{+0.94}_{-0.98}$ and $S_1^A
/6k_BTG_2^A = 1.61^{+0.22}_{-0.20}$, being statistically consistent with the previous report [@NakamuraPRL2010; @PassingJCCCB1983]. The reason for the observed considerable deviation from the theory in the symmetric part is not yet clear. We note that in a double-quantum dot experiment performed in the incoherent regime [@FujisawaScience2006] similar large discrepancy between the prediction based on FT was reported, where the back action of the nonequilibrium quantum point contact attached to the dots to detect their charge states explains the observation [@UtsumiPRB2010]. In the present case, as no such detector is present, further effort is necessary to solve this problem.
Regarding the amplitude of $G_2^S$ and $G_2^A$, the experiment on the nonlinear transport in the AB ring fabricated on the conventional 2DEG was reported before [@AngersPRB2007]. The radius of their ring is about three times larger than ours. They measured the temperature dependence of the amplitudes $G_2^S$ and $G_2^A$ and found that the amplitudes rapidly decrease as temperature increases from 30 mK to 1 K. Similar temperature dependence was observed in the present ring. At the lowest temperature, the amplitude of $G_2^S$ and $G_2^A$ in the present case is slightly larger but falls in the same range of their result.
{width=".99\linewidth"}
Now let us discuss the microreversibility in the present system. In the presence of the magnetic field, the possibility of the absence of the microreversibility in the nonequilibrium was recently pointed out [@ForsterPRL2008]. While the antisymmetric relation is only given by Eq. (\[WOMicroRev\]), the restriction of the microreversibility simultaneously requires the relation of Eq. (\[WMicroRev\]). Thus we can basically obtain $C_0^A$ from the experimental data in two ways; by calculating $C_0^A = 3k_BT
(S_1^A-2k_BTG_2^A)$ which holds true regardless of the microreversibility condition and by calculating $C_0^A = 2k_BTS_1^A$ validated only with the microreversibility. Figure 5(a) shows the obtained $C_0^A$ from the data set shown in Fig. 4. Clearly as a function of $V_g$, $C_0^A$ calculated in two ways coincide each other. Figure 5(b) shows the result for the data reported in Fig. 3 in Ref. where the field is swept with a fixed $V_g$. In this case, too, $C_0^A$ obtained in two ways almost perfectly equal each other.
In Fig. 5 (c), we plot $3k_BT (S_1^A-2k_BTG_2^A)$ vs. $2k_BTS_1^A$ at 125 mK and 450 mK. As we have seen that $S_1^A /6k_BTG_2^A$ deviates from unity, the slope is slightly different from unity. However, within the accuracy of the present experiment, we may claim that two values are the same. This tells us that in the present experiment the assumption of the microreversibility is valid.
Finally, we note that the present demonstration gives a single example of the validity of the microreversibility in the nonequilibrium quantum regime in the presence of the magnetic field. This fundamental topic should be experimentally addressed in many systems such as electron interferometers [@ForsterPRL2008; @Forster_arXiv; @LimPRB2010], the quantum dot [@SanchezPRB2009], and the macroscopic inhomogeneous system [@NagaevPRL2010].
Conclusions
===========
We show that the fluctuation theorem is semiquantitatively valid in the description of the quantum transport in mesoscopic systems. Unlike the conventional scattering theory, this description gives a nontrivial relation between the nonlinearity and the nonequilibrium in the presence of the magnetic field. The direct test of the validity of the microreversibility was also addressed. Since the fluctuation theorem does not directly give the physical interpretation of the current through the device as the Landauer-B[ü]{}ttiker formalism does, both descriptions are complementary to each other. We believe that by combining these two pictures, nonequilibrium properties in mesoscopic systems in the presence of the interaction effect will be further addressed.
We appreciate fruitful discussions from Markus B[ü]{}ttiker, Masahito Ueda, Takeo Kato, and Hisao Hayakawa. This work is partially supported by KAKENHI, Yamada Science Foundation, SCAT, Matsuo Science Foundation, Strategic International Cooperative Program the Japan Science and Technology Agency (JST), and the German Science Foundation (DFG).
[99]{}
See, e.g., S. Datta, [*Electronic Transport in Mesoscopic Systems*]{} (Cambridge University Press, Cambridge, England, 1995); Y. Imry, [*Introduction to Mesoscopic Physics*]{} (Oxford UP, New York, 1997).
M. B[ü]{}ttiker, Phys. Rev. B [**46,**]{} 12485 (1992).
Th. Martin, R. Landauer, Phys. Rev. B [**45,**]{} 1742 (1992).
Y. M. Blanter, and M. B[ü]{}ttiker, Phys. Rep. [**336,**]{} 1 (2000).
R. de-Picciotto, M. Reznikov, M. Heiblum, V. Umansky, G. Bunin, and D. Mahalu, Nature **389**, 162 (1997).
L. Saminadayar, D.C. Glattli, Y. Jin, B. Etienne, Phys. Rev. Lett. **79** 2526 (1997).
X. Jehl, M. Sanquer, R. Calemczuk, and D. Mailly, Nature **405,** 50 (2000).
M. Esposito, U. Harbola, and S. Mukamel, Rev. Mod. Phys. [**81,**]{} 1665 (2009).
D. J. Evans, E. G. D. Cohen, and G. P. Morriss, Phys. Rev. Lett. [**71,**]{} 2401 (1993).
D. Andrieux, and P. Gaspard, J. Chem. Phys. [**121,**]{} 6167 (2004); J. Stat. Mech. P01011 (2006); J. Stat. Phys. [**127,**]{} 107 (2007).
J. Tobiska, and Yu. V. Nazarov, Phys. Rev. B [**72,**]{} 235328 (2005).
D. Andrieux, P. Gaspard, T. Monnai, and S. Tasaki, New J. Phys. [**11,**]{} 043014 (2009).
K. Saito, and Y. Utsumi, Phys. Rev. B [**78,**]{} 115429 (2008); Y. Utsumi and K. Saito, [*ibid.*]{} [**79,**]{} 235311 (2009).
H. F[ö]{}rster, and M. B[ü]{}ttiker, Phys. Rev. Lett. [**101,**]{} 136805 (2008).
H. F[ö]{}rster, and M. B[ü]{}ttiker, AIP Conference Proceedings [**1129,**]{} 20th International Conference on Noise and Fluctuations, M. Macucci and G. Basso, eds. (Melville, New York, 2009). p. 443.
Y. Utsumi, D. S. Golubev, M. Marthaler, K. Saito, T. Fujisawa, and G. Sch[ö]{}n, Phys. Rev. B [**81,**]{} 125331 (2010).
M. Campisi, P. Talkner, and P. H[ä]{}nggi, Phys. Rev. Lett. [**105,**]{} 140601 (2010).
A. Altland, A. De Martino, R. Egger, and B. Narozhny Phys. Rev. B [**82,**]{} 115323 (2010).
G. Gallavotti, Phys. Rev. Lett. [**77,**]{} 4334 (1996).
G. M. Wang, E. M. Sevick, E. Mittag, D. J. Searles, and D. J. Evans, Phys. Rev. Lett. [**89,**]{} 050601 (2002).
N. Garnier and S. Ciliberto, Phys. Rev. E [**71,**]{} 060101(R) (2005).
J. Kurchan, arXiv:cond-mat/0007360 (unpublished).
S. Gustavsson, R. Leturcq, B. Simovic, R. Schleser, T. Ihn, P. Studerus, K. Ensslin, D. C. Driscoll, and A. C. Gossard, Phys. Rev. Lett. [**96,**]{} 076605 (2006).
T. Fujisawa, T. Hayashi, R. Tomita, and Y. Hirayama, Science [**312,**]{} 1634 (2006).
S. Nakamura, Y. Yamauchi, M. Hashisaka, K. Chida, K. Kobayashi, T. Ono, R. Leturcq, K. Ensslin, K. Saito, Y. Utsumi, and A. C. Gossard, Phys. Rev. Lett. [**104,**]{} 080602 (2010).
The factor 2 is artificially introduced so that the resultant expression for $S_0$ and $G_1$ is consistent with the classical expression of the Johnson-Nyquist relation.
D. S[á]{}nchez, and M. B[ü]{}ttiker, Phys. Rev. Lett. [**93,**]{} 106802 (2004); M. L. Polianski, and M. B[ü]{}ttiker, Phys. Rev. Lett. [**96,**]{} 156804 (2006); [*ibid*]{} Phys. Rev. B [**76,**]{} 205308 (2007).
B. Spivak, and A. Zyuzin, Phys. Rev. Lett. [**93,**]{} 226801 (2004).
J. Wei, M. Shimogawa, Z. Wang, I. Radu, R. Dormaier, D.H. Cobden, Phys. Rev. Lett. [**95,**]{} 256601 (2005); D. M. Zumb[ü]{}hl, C. M. Marcus, M. P. Hanson, and A.C. Gossard, [*ibid.*]{} [**96,**]{} 206802 (2006); C. A. Marlow, R. P. Taylor, M. Fairbanks, I. Shorubalko, and H. Linke, [*ibid.*]{} [**96,**]{} 116801 (2006); B. Brandenstein-K[ö]{}th, L. Worschech, and A. Forchel, Appl. Phys. Lett. [**95,**]{} 062106 (2009).
L. Angers, E. Zakka-Bajjani, R. Deblock, S. Gueron, H. Bouchiat, A. Cavanna, U. Gennser, and M. Polianski, Phys. Rev. B [**75,**]{} 115309 (2007).
R. Leturcq, D. S[á]{}nchez, G. G[ö]{}tz, T. Ihn, K. Ensslin, D. C. Driscoll, and A. C. Gossard, Phys. Rev. Lett. [**96,**]{} 126801 (2006); R. Leturcq, R. Bianchetti, G. G[ö]{}tz, T. Ihn, K. Ensslin, D.C. Driscoll, A.C. Gossard, Physica E [**35,**]{} 327-331 (2006).
Y. Yamauchi, M. Hashisaka, S. Nakamura, K. Chida, S. Kasai, T. Ono, R. Leturcq, K. Ensslin, D. C. Driscoll, A. C. Gossard, and K. Kobayashi, Phys. Rev. B [**79,**]{} 161306(R) (2009).
R. Held, T. Vancura, T. Heinzel, K. Ensslin, M. Holland, and W. Wegscheider, Appl. Phys. Lett. [**73,**]{} 262 (1998).
L. DiCarlo, Y. Zhang, D. T. McClure, C. M. Marcus, L. N. Pfeiffer, and K. W. West, Rev. Sci. Instrum. [**77,**]{} 073906 (2006); M. Hashisaka, Y. Yamauchi, S. Nakamura, S. Kasai, K. Kobayashi, and T. Ono, J. Phys. Conf. Series [**109,**]{} 012013 (2008).
M. Hashisaka, Y. Yamauchi, S. Nakamura, S. Kasai, T. Ono, and K. Kobayashi, Phys. Rev. B [**78,**]{} 241303(R) (2008).
S. Nakamura, M. Hashisaka, Y. Yamauchi, S. Kasai, T. Ono, and K. Kobayashi, Phys. Rev. B [**79,**]{} 201308(R) (2009).
To estimate the error bars of the coefficients between the two values, Passing-Bablok regression was adopted since both $S_1$ ($S_1^S$ or $S_1^A$) and $G_2$ ($G_2^S$ or $G_2^A$) have statistical uncertainties, where the conventional linear regression is not justified to estimate the error bars. Here the error bar indicates the 95% confidence interval. H. Passing, and W. A. Bablok, J. Clin. Chem. Clin. Biochem. [**21,**]{} 709 (1983).
J. S. Lim, D. S[á]{}nchez, and R. L[ó]{}pez, Phys. Rev. B [**81,**]{} 155323 (2010).
D. S[á]{}nchez, Phys. Rev. B 79, 045305 (2009).
K. E. Nagaev, O. S. Ayvazyan, N.Yu. Sergeeva, and M. B[ü]{}ttiker, Phys. Rev. Lett. [**105,**]{} 146802 (2010).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present the observations of a double-decker filament to study its formation, triggering, and eruption physics. It is observed that the double-decker filament was formed by splitting of an original single filament. During the splitting process, intermittent bright point bursts are observed in the filament channel, which resulted in the generation of the upper filament branch. The eruption of the newly formed double-decker filament was possibly triggered by two recurrent two-sided loop jets in the filament channel and the continuous mass unloading from the upper filament body. The interaction between the first jet and the filament directly resulted in the unstable of the lower branch and the fast rising phase of the upper branch. The second jet occurred at the same site about three hours after the first one, which further disturbed and accelerated the rising of the lower filament branch. It is interesting that the rising lower branch overtook the upper one, and then the two branches probably merged into one filament. Finally, the whole filament erupted violently and caused a large-scale coronal mass ejection, leaving behind a pair of flare ribbons and two dimming regions on the both sides of the filament channel. We think that the intermittent bursts may directly result in the rearrangement of the filament magnetic field and therefore the formation of the double-decker filament, then the recurrent jets further caused the fully eruption of the entire filament system. The study provides convincing evidence for supporting the scenario that a double-decker filament can be formed by splitting a single filament into two branches.'
address:
- 'Yunnan Observatories, Chinese Academy of Sciences, Kunming, China, 650216'
- 'State Key Laboratory of Space Weather, Chinese Academy of Sciences, Beijing 100190, China'
- 'University of Chinese Academy of Sciences, Beijing, China,100049'
- 'Center for Astronomical Mega-Science, Chinese Academy of Sciences, Beijing, 100012, China'
author:
- Zhanjun Tian
- Yuandeng Shen
- Yu Liu
title: 'Formation and Eruption of a Double-decker Filament Triggered by Micro-bursts and Recurrent Jets in the Filament Channel'
---
Sun: atmosphere;Sun: activity; Sun: filaments ;Sun: magnetic fields
Introduction
============
Solar prominences are thread-like clouds consisting of relatively cool (5000-8000K) and dense (10$^{10}$-10$^{11}$cm$^{-3}$) magnetized plasma suspended in the hot tenuous corona [e.g., @hiraya85; @foukal04]. Prominences and filaments are the same entities with the former over the limb and the latter against the solar disk. In the following, we use the term “filament” throughout the paper. Filaments are always observed along photospheric magnetic polarity inversion lines (PIL) that denote filament channels. The magnetic field of filament channels are approximately aligned with PILs, which may [**be**]{} caused by the photospheric activities such as supergranular shearing, convecting motions, and flux cancellations [@wang07]. In extreme ultraviolet (EUV) observations, filament channels are darker than their surroundings. Besides the main spine along the filament channel, a filament often hosts a few lateral feet or barbs. In high resolution observations, a typical filament shows more fine structures such as bubbles and ubiquitous plasma flows in every parts of the filaments [@berger08; @berger10; @shen15].
Although it has been widely accepted that filaments are supported by coronal magnetic fields against gravity, the structure and formation process of the magnetic fields in and around filaments remains unclear. Different magnetic topologies have been proposed to understand the magnetic structures of filaments, such as the sheared arcade [e.g., @anti94], and/or flux rope [e.g., @amari00] models. So far, the formation mechanism of filaments are still unclear. Various scenarios have been proposed to explain the formation of filaments. For example, filaments can be formed by emerging from the photosphere [e.g., @2009ApJ...697.1529F; @2008SoPh..247..103Y], rearranging of an existing magnetic structure due to photospheric flux cancellation and/or coronal reconnection [e.g., @2015ApJ...803...86Y; @2016ApJ...830...16Y; @2016ApJ...816...41Y; @yan15].
The eruption of filaments are the most striking phenomena in the solar atmosphere, and they can cause devastating influence on the near-Earth space environment, especially associated with solar flares or/and coronal mass ejections (CMEs) [e.g., @lin03; @hud06]. The eruption of filaments are often accompanied by various solar phenomena, such as flares [e.g., @2003ApJ...593L.137L; @2011RAA....11..594S; @2012ApJ...745....9Y; @2015ApJ...805...48B], CMEs [e.g., @2012ApJ...750...12S; @2012SoPh..279..115Y], coronal dimmings [e.g., @2011SoPh..270..551Y; @2011NewA...16..276B], global EUV waves [e.g., @2012ApJ...754....7S; @2017ApJ...851..101S; @2013ApJ...775...39Y]. Filaments supported by different magnetic fields have been explained with different kinds of physical mechanisms. Besides the well-known “Lin-Forbes” model [@lin00], “tether-cutting" model [e.g., @2001ApJ...552..833M; @2016ApJ...818L..27C], and “breakout" model [e.g., @1999ApJ...510..485A; @2012ApJ...750...12S; @2016ApJ...820L..37C], there are still many other theoretical filament eruption models based on the loss of equilibrium. For example, filament eruptions can be formed directly from slow evolution of the magnetic fields leading to “catastrophic loss of equilibrium" in the relevant magnetic structures [e.g., @forbes90; @lin00; @lin03; @lin04; @low01]. In addition, the helical kink instability [e.g., @toeroe04; @liu09], the torus instability [e.g., @kliem06; @schr08], and the mass-unloading mechanism are also important for destabilizing filaments [e.g., @low03; @bi14; @2014NewA...26...23X; @qin17; @jenkin18].
Recently, @liu12 reported so-called “double-decker filament”, which is composed by two filaments in the same filament channel but separated in height. The two filament components are named as upper branch (UB) and lower branch (LB), respectively. The authors proposed two possible scenarios for explaining the formation of the double-decker filaments. One is the emerge filament model in which a new filament emerges below the old one to compose a double-decker filament. The other is the splitting model in which a single filament splits into two filaments for internal reconnection. This model is motivated by partial eruption proposed by [@gil01]. In the partial eruption scenario, the reconnection within a single stretched flux rope or flux bundles vertically splits along the original rope axis into two ropes with the same handedness. Furthermore, the splitting often occurs before or during the eruption in these cases, which means that the two new branches can not be survival for much longer time. [@kliem14] further demonstrated that such a special filament configuration can be stable if an external toroidal (shear) field component exceeding a threshold value. If the external toroidal field component decreases sufficiently, then both filaments turned unstable. In addition, the authors also suggested that the transfer of magnetic flux and current to the UB is the key mechanism responsible for the loss of equilibrium for double-decker filaments and this phenomenon was also discovered by other researchers [e.g., @zhu14; @zhang14]. However, the authors did not provide any believable observational evidence for supporting their models.
In the present paper, we present unambiguous evidence to support the splitting model of double-decker filaments. The event occurred on 2017 November 16 in the quiet-Sun region on the south of NOAA active region AR12687 (S07E43), in which a quiescent filament first split into two branches due to some small-scale explosive bursts in the filament channel, then it reached a new equilibrium state and lasted for a few hours. During this time interval, two recurrent two-sided loop jets in the filament channel and mass unloading from the UB are observed, which are thought to be the trigger agents corresponding to the eruption of the newly formed double-decker filament. The used observations are introduced in Section 2, results are given in Section 3, conclusions and discussion are presented in the last section.
Observations
=============
The main data used in this study were taken by the Atmospheric Imaging Assembly [AIA; @lemen12] and the Helioseismic and Magnetic Imager [HMI; @schou12] on board the [*Solar Dynamics Observatory*]{} [[SDO]{}; @pesnel12]. The AIA provides full-disk EUV and ultraviolet (UV)-visible images in ten channels. The pixel size of the AIA images is of $0.6''$, and the cadences of the EUV and UV-visible images are 12 and 24 s, respectively. The HMI provides continuous full-disk line-of-sight (LOS) magnetograms that have a cadence of 45 s and a measuring precision of 10 Gauss. The CME associated with the filament eruption is studied by using the white-light images taken by the Large Angle and Spectrometric Coronagraphs [[LASCO]{}; @bru95] on board the [*Solar and Heliospheric Observatory*]{} ([*SOHO*]{}).
Results
=======
An overview of the pre-eruption condition of the event is shown in [Figure \[fo\]]{} with the AIA 304 Å and the HMI LOS magnetograms. It can be seen that the filament resided in a filament channel that was in the northwest-southeast direction. The outline of the filament is overlaid in the HMI magnetogram, which clearly shows that the filament is a quiescent filament that located in the quiet-Sun region, and the magnetic fields on both sides of the filament are of opposite magnetic polarities (see [Figure \[fo\]]{} (b)). The close up view of the small region located in the filament channel is plotted in panel (c) of [Figure \[fo\]]{}, which shows a small bipolar magnetic region, in which obvious flux cancellation is detected between the positive and negative polarities. This special region was also the source region of some transient bursts and the subsequent two recurrent two-sided loop jets. The quiescent filament can be best seen in the AIA 304 Å images, which with its eastern part close to the disk limb and the western part on the disk (see [Figure \[fo\]]{} (d)). The filament was split into two branches (see [Figure \[fo\]]{} (f)), i.e., the so-called double-decker filament. The newly formed double-decker filament erupted at about 18:00:00 UT, which resulted in a pair of flare ribbons and double dimming regions on the solar surface, and a large-scale CME in the outer corona. The detail evolution processes for the formation, triggering, and eruption of the filament are described in following sections.
Formation of the Double-decker Filament
----------------------------------------
To study the formation process of the double-decker filament, we trace back the filament to at about 05:00:00 UT, that is about 13 hours before the start of the violent filament eruption. At 05:47:29 UT, it is obvious that the filament was a single long dark feature in the filament channel (see the dotted line in [Figure \[fo\]]{} (d)). During the following several hours, some intermittent bright point bursts are observed around the bipolar magnetic region in the filament channel. The arrow labelled “BP” in panel (e) of [Figure \[fo\]]{} points to a typical bright point burst. It is observed that the filament started to split into two branches right after an obvious bright point burst in the bipolar magnetic region at about 07:06:29 UT. One can see that the filament kept stable at the original location, a loop-like filament thread started to separate from the west section ( the right of “BP") of the original filament and showed a southward slow rising motion (see the arrow labelled “LF” in panel (e) of [Figure \[fo\]]{}). Here, we define the stable part of the filament as the LB, while the rising loop-like part as the UB. The rising motion of the UB made the separation distance to the LB increasing in time and the separation slowly slipped to the east section. About two hours later, it is easy to distinguish the two branches in the AIA 304 Å images, due to the increased separation distance between them (see [Figure \[fo\]]{} (f)). The positions of the UB and LB at the time of 12:40:29 UT (see panel (a) in [Figure \[jets\]]{} ) are all overlaid on the HMI LOS magnetogram in [Figure \[fo\]]{} (b), when the separation was well developed. One can see that the [**projection position of**]{} UB located on the positive magnetic region on the [**southern**]{} side of the filament channel (see the orange contour in [Figure \[fo\]]{} (b)), which suggests that the UB had rose to a higher altitude. The slow rising motion of the east section of UB lasted for about 3 hours with a projected speed of about [0.68 km s${}^{-1}$]{}. For more details about the formation of the double-decker filament and the evolution of the magnetic bipolar, one can see the online animations (animation1.mpg, animation2.mpg) associated with [Figure \[fo\]]{}.
The evolution pattern described above suggests that the double-decker filament was formed by splitting the original single filament, due to some intermittent bright point bursts in the filament channel. The intermittent bright point bursts may represent some kind of magnetic reconnections among field lines that rooted in the source region, where the positive- and negative-polarity flux meet together. Those magnetic reconnections can cause the rearrangement of the magnetic field distribution in the filament and its surroundings [e.g., @wang07; @wang13]. More discussions about the relations between the two phenomena, bursts and splitting, are displayed in Section \[c\_d\].
Triggering and Eruption of the Filament
---------------------------------------
At about 12:30:00 UT, a two-sided loop jet [e.g., @alex99; @tian17; @ning16] occurred at the bipolar magnetic region in the filament channel, whose arms were in opposite directions along the filament channel (see the arrow labelled with “J1” in [Figure \[jets\]]{} (a) and (b)). The eastward and westward arms of the jet directly interacted with LB, and bright plasma blobs are observed propagating along the filament threads (see [Figure \[jets\]]{} (a) and (b)). The interaction of the jet and the LB may directly inject plasma into the filament, which can further cause the unstable of LB due to the increased mass with momentum [@guoj10]. Additionally, this is similar to the study performed by [@2005ApJ...631L..93L], where the authors observed that filaments are generated by the injection of recurrent jets, and suggested that there should be a direct link between the filament axial fields and the large-scale background fields along which the ejecta can be driven into the filament. It is interesting that the UB experienced a faster rising phase relative to the previous splitting stage right after the start time of the jet, which suggests that the jet not only disturbed the LB but also the dynamics of the entire filament system.
It is noted that the rising of UB stopped at a plateau height at about 14:00:00 UT and lasted for about three hours until the start of the fast eruption of the filament at about 17:00:00 UT. During this time interval, intermittent mass drainage is observed along the western leg of UB (see the blue arrow in [Figure \[jets\]]{} (e)), which might be important for the loss of equilibrium of the filament [e.g., @low03; @bi14; @qin17; @jenkin18]. At about 15:20:00 UT, the second two-sided loop jet occurred at the same site as the first one, and the morphology of the jet is shown with the AIA 171 Å and 304 Å images (see the arrow labelled with “J2” in [Figure \[jets\]]{} (c) and (d)). In contrast to the first jet, the eastward arm resembles a loop-like feature, and it was suppressed within a short distance, while the westward arm ejected to a large distance up to the UB’s west end. Due to the disturb caused by the interaction between the second jet and LB, the east section of LB started to rise in the south [**projection**]{} direction and cut off from the west section of LB which remained in the original height. For brevity, we refer the east section of the lower branch as “LB" in the following. It is observed that the rising speed of LB was faster than that of the UB, which caused the interaction and probably merging of the two branches at about 16:00:00 UT. Such an interaction and merging between two filaments have also been reported in previous studies [e.g., @su07; @2017ApJ...838..131Y]. However, due to the lack of stereoscopic observations, we are unable to distinguish the merging of the two branches were true or not, since the projection effect can also cause this in observations if the two separated branches were all moving in the line-of-sight. About one hour after the interaction of the two branches, the newly merged filament started the fast rising and eruption phases at about 17:00:00 UT. During the eruption, continuous mass drainage with many bright blobs are observed along the western leg of the erupting filament, which further reinforced the acceleration of the slow rising filament. The acceleration of the filament lasted for about one hour, and finally the entire filament lost equilibrium and erupted totally at about 18:00:00 UT. For more details about the two-sided loop jets and their interaction to the filament, one can see the online animation (animation3.mpg) associated with [Figure \[jets\]]{}.
During and after the eruption of the filament, a pair of bright flare ribbons are observed on the both sides of the filament channel. In addition, two obvious dimming regions appeared at the outer sides of the two flare ribbons (see [Figure \[jets\]]{} (f)). These low coronal phenomena are frequently observed in a large number of studies, and they are thought to be the typical eruption characteristics accompanying with filament eruptions. The present filament eruption further resulted in a large-scale CME in the outer corona, which was recorded by LASCO C2 and C3 coronagraphs. The morphology of the CME is displayed in [Figure \[jets\]]{} (g) – (i) with the composited images made from AIA 304 Å (inner), LASCO C2 (middle), and LASCO C3 (outer) observations. To better show the CME, we use the running difference images of LASCO C2 and C3 in the composited images. Here, a running difference image is obtained by subtracting a image by the previous one in time, in which moving features can be observed more clearly. The CME was at an average speed of [610 km s${}^{-1}$]{} within the FOV of C2 and C3, and its first appearance times in the FOVs of LASCO C2 and C3 were at 20:18:00 UT and 22:06:07 UT, respectively.
The entire evolution process of the filament is also studied by using time-distance diagrams. To obtain a time-distance diagram, we first obtain the one-dimensional intensity profiles along a specified path at different times, and then a two-dimensional time-distance diagram can be generated by stacking the obtained one-dimensional intensity profiles in time. To study the kinematics of the two recurrent two-sided loop jets, Time-distance diagrams are made from AIA 304 and 171 Å observations along the filament axis (see panels (a) and (b) in [Figure \[td\]]{}), respectively. The ejection speed of the two jets are measured by applying a linear fit to the inclined bright features that represent the arms of the jets. The measurement results indicate that the speeds of the eastward (westward) arms of the first jet is about [-96.5 (84.3) km s${}^{-1}$]{} in the 304 Å time-distance diagram, while the speed of the westward arm of the second jet is about [112.3 km s${}^{-1}$]{}. The eastward arm of the second jet was confined within a short distance, thus that we do not measure its speed. In the 171 Å time-distance diagram, the corresponding ejection speeds of the jets are relatively slower than those obtained in the 304 Å time-distance diagram.
Panels (c) and (d) in [Figure \[td\]]{} are time-distance diagrams made from AIA 304 Å images along the two dotted lines across the filament axis (see [Figure \[jets\]]{} (e)), respectively. These time-distance diagrams show the time evolution of the filament more clearly. The time-distance diagram in [Figure \[td\]]{} (c) shows the evolution of the UB better. One can see that the filament underwent a slow rising phase with a speed of about [0.68 km s${}^{-1}$]{}. Right after the occurrence of the first two-sided loop jet, the UB started a faster constant rising phase with a speed of about [2.3 km s${}^{-1}$]{}. The change of the rising speed obviously indicates the influence resulted from the first jet. Interestingly, the western part of UB stopped to rise during the time interval from about 14:00:00 UT to 16:30:00 UT (see the blue box in [Figure \[td\]]{} (c)), which suggests that the rising UB reached a new equilibrium state. In addition, the mass drainage from the filament can also be observed in the time-distance diagram during this time interval. Finally, the filament started its violent eruption process after 17:00:00 UT. It is clearly that the eruption of the filament underwent a slow and a following fast eruption phases, and the corresponding average speeds are about [2.9 and 25.6 km s${}^{-1}$]{}, respectively.
The time-distance diagram in [Figure \[td\]]{} (d) shows the evolution of the both branches and their interaction process clearly, in which the LB and UB are indicated with arrows labelled with letters “LB” and “UB”, respectively. It can be seen that the LB kept stable before the first jet, then it showed a little rise during the time interval of the two jets. After the second jet, the LB underwent a fast rising phase and therefore led to the interaction and merging of the two filament branches during the time interval from 15:30:00 UT to 17:00:00 UT. The fast rising of the LB manifested the influence caused by the second jet. The merged filament started its eruption process after 17:00:00 UT, and the two acceleration phases are also clear in this time-distance diagram.
Magnetic Flux Variation
-----------------------
The magnetic flux variation around the eruption source region of the two-sided loop jets are studied in detail, and the results are plotted in [Figure \[flux\]]{}. The positive and negative magnetic fluxes within the blue box region as shown in [Figure \[fo\]]{} (c) are plotted as red and blue curves in [Figure \[flux\]]{} (a), respectively. A time-distance diagram ([Figure \[flux\]]{} (b)) along the red dashed line as shown in [Figure \[fo\]]{} (c) is also generated to show the spatial evolution of the positive and negative magnetic polarities in the eruption source region of the jets. As shown in [Figure \[flux\]]{} (a), the positive magnetic flux displayed a moderate and irregular fluctuation at a level of about $4 \times 10^{18}$ Mx during the observation time interval of our study. Whereas the magnitude of negative flux showed a continuous decreasing trend from $16 \times 10^{18}$ to $4 \times 10^{18}$ Mx except for the short period of time after first jet, corresponding to an average loss rate of $\backsim 2 \times 10^{18}$ Mx hr$^{-1}$, which is consistent with the results reported in @wang13.
Here, we focus on the two short time intervals around the beginnings of the two jets to analyze the causal relationship between the magnetic flux variations and the observed coronal jets. The start times of the jets are marked in [Figure \[flux\]]{} with two vertical dashed lines. As one can see that the positive (negative) flux before the start time of the first jet showed a rapid increase (decrease) period of about half an hour, then it changed to decrease (increase) trend right after the start time of the jet. The variation of the fluxes around the start time of the second jet showed a similar evolution pattern. These variation trends of the magnetic fluxes suggest the emergence (cancellation) of the positive (negative) magnetic flux before the start of the jets. However, right after the start of the jets the variation trend of the positive (negative) flux changed into cancellation (emergence). In the time-distance diagram ([Figure \[flux\]]{} (b)), the convergence motion between the positive and negative magnetic polarities is obviously during the entire observation time interval. The cancellation and emergence of the magnetic fluxes can also be clearly identified. Therefore, the eruption of the jets were possibly triggered by the alternating emergence and cancellation of the positive and negative magnetic fluxes.
Conclussions & Discussions {#c_d}
===========================
We present the observational analysis of a double-decker filament close NOAA active region AR12687 on 2017 November 16, by using the high temporal and spatial resolution data taken by the [*SDO*]{}. For the first time, we report the detailed evolution of the double-decker filament, including the formation, triggering, and eruption processes. This study provides a convincing example for supporting the prediction that a double-decker filament can be formed by splitting an original single filament into two filament branches.
The observational results indicate that the formation of the double-decker filament in the present case was due to the splitting of an original single filament. This physical picture is consistent with one of the possible formation mechanisms proposed by @liu12. At the very beginning, it is observed that a quiescent filament resided in a filament channel about 05:00:00 UT. In the following several hours, some intermittent bright point bursts are observed around the small bipolar magnetic regions in the filament channel. Simultaneously, the paired opposite polarities in the eruption source region showed obvious converging motion and magnetic cancellation between them. The bright point bursts represent energy releasing events attributed to magnetic reconnection, which can convert short field lines into upward longer and downward shorter loops [@van89]. So we can image that the short filament threads rooted in the positive and negative polarities in the eruption source region could be converted to longer filament threads via magnetic reconnection [@tian17]. The filament underwent a long splitting process of about 6 hours, in which the LB remained at the original height while the loop-like UB showed obvious slow rising movement to the south [**projection**]{} direction at a speed of about [0.68 km s${}^{-1}$]{}. For the splitting of a single filament into two branches, the model proposed in @liu12 did not discuss the detailed physical process. [@gil01] proposed a model in which the reconnection occurs inside the filament, this can cause the splitting of the filament. Another possibility for the filament splitting could be caused by mixed kink instability and magnetic reconnection process in the bipolar, which can effectively transfer and rearrange the magnetic field distribution in the filament. The mixed kink instability are proposed by @mei18 in their parametric study on kink instabilities. According to the linear theory of MHD instabilities for cylindrical plasma columns [@goedbloed10], complex twist turns distribution inside the realistic filament may allow appearance of external and internal kink instabilities in the meanwhile, and the mixed kink refer to this situation. The resultant evolution of filaments demonstrate some interesting and complicated behaviors, specifically, the splitting process of filaments. Since we did not observed reconnection characteristics in the filament body, the splitting of the present filament was not suitable for the model proposed by [@gil01]. We think that the splitting of the filament in the present case seems more suitable for the mechanism of mixed kink instability proposed by @mei18, in which the intermittent bright point bursts underneath the filament may provide successive disturbances to the filament system and therefore trigger or accelerate the external and internal kink instabilities of the filament.
The triggering of the filament eruption is analyzed in detail. It is found that the triggering of the filament eruption was tightly associated with the eruption of two recurrent two-sided loop jets originated from the small bipolar region. The variations of the magnetic fluxes suggest that the recurrent jets are possibly caused by alternating flux emergence and cancellation activities in the bipolar region. It is observed that the first jet directly interacted with the LB, and right after the beginning of the jet the UB was accelerated to a faster rising phase at a speed of [2.3 km s${}^{-1}$]{}. The second jet occurred at about 15:20:00 UT and interacted with the LB again, which directly resulted in the rising of the LB. Obviously, the rising of the LB was faster than the upper one. Therefore, the two filament branches interacted and probably merged during the time interval from 15:30:00 UT to 17:00:00 UT, during which substantial mass drainage from the UB is also observed. Finally, the filament erupted totally at about 17:00:00 UT, which caused a large-scale CME at an average speed of [610 km s${}^{-1}$]{}. In addition, low coronal features such as the pair of flare ribbons and the double dimming regions are also observed to confirm the eruption of the filament.
Based on the observational results, the temporal and spatial relationship between the recurrent jets and the onset of the filament eruption suggest that the quasi-equilibrium state of the newly formed double-decker filament was destroyed by the jet activities and the mass unloading of the UB. The recurrent two-sided loop jets can not only provide plasma material and momentum to the filament, but also can rearrange the filament magnetic field environment. On the other hand, the continuous mass unloading of the UB removed a part of weight from the filament. Consequently, the system can not keep balance between the downward gravity force and upward magnetic force. Therefore, the entire filament system will lose equilibrium and erupt when the upward force exceed the downward gravity. This is consistent with the models of mass unloading eruptions, which predict that significant material drainage may facilitate an eruption [@low03]. The model suggests that sufficiently large mass in filament can contribute to equilibrium. That is to say if mass removed, the magnetic environment would be free to expand and attempted to find a new equilibrium by increasing the height of the flux rope.
Solar jets are ubiquitous in the solar atmosphere, and they are often found to be associated with magnetic flux emergence and cancellation activities [e.g., @liu04; @2012ApJ...745..164S; @2017ApJ...851...67S]. Traditional theoretical studies proposed that a solar jet is caused by the magnetic reconnection between an emerging bipole and its ambient open fields [e.g., @1995Natur.375...42Y]. In particular cases, if an emerging magnetic bipole reconnects with the overlying horizontal field lines, the consequence should be the generation of a two-sided loop jet [@1995Natur.375...42Y]. In addition, two-sided loop jets can also be produced by the magnetic reconnection between adjacent filamentary threads [@tian17]. Recent high resolution observations indicate that many jets are dynamically associated with the eruption of mini-filaments [e.g., @2012ApJ...745..164S; @2017ApJ...851...67S; @2012NewA...17..732Y; @2014ApJ...796...73H; @2017ApJ...835...35H; @2017ApJ...842L..20L; @2018ApSS.363...26L]. In addition, it is also found that solar jets are tightly related to other large-scale solar eruption phenomena, in which solar jets often play a trigger role. For example, through interaction with other magnetic structures, jets can cause sympathetic CMEs [e.g., @2008ApJ...677..699J; @2012ApJ...745..164S] and filament oscillations [e.g., @2017ApJ...851...47Z]. In particularly, [@2005ApJ...631L..93L] found that some filaments can be quickly formed by trapping the cold material supplied by jets. In the present study, we observed the interaction of two-sided loop jets interacted with and provided mass to the double-decker filament, which triggered the loss of equilibrium and fully eruption of the filament system by rearranging the magnetic fields. So far, such kind of observations are still very scarce, more similar observational and theoretical studies are needed to figure out the detailed physical process during the jet-filament interaction period.
In summary, by analyzing the detailed evolution of a quiescent filament of about 15 hours, the observational results reveal the formation, triggering, and eruption details. The present study provides convincing evidence for supporting the scenario that a double-decker filament can be formed by splitting an original single filament due to some small-scale transient explosive events. The triggering of the filament eruption was caused by two recurrent two-sided loop jets in the filament channel and the mass drainage from the filament body, which resulted in the disequilibrium of the filament. The eruption of filaments is the result of multiple factors. The present study provides a possible explanation, we do not exclude other interpretations. Studying the dynamics of filaments formation, stability, and destabilization can shed light on the precursor of eruptive events. The results of the present study are highly encouraging for future investigations. More observational and simulation works are needed in the future to understand the filaments dynamics in detail.
Acknowledgements {#acknowledgements .unnumbered}
================
We thank the anonymous referee for his/her constructive comments which are valuable for improving the quality paper. We also thank the excellent observations provided by the [*SDO*]{} and [*SOHO*]{} teams. This work is supported by the Natural Science Foundation of China (11773068, 11633008,11403097, 11503084), the Yunnan Science Foundation (2017FB006, 2015FB191), the Specialized Research Fund for State Key Laboratories, the Youth Innovation Promotion Association (2014047) of Chinese Academy of Sciences, and the grant associated with the Project of the Group for Innovation of Yunnan Province.
natexlab\#1[\#1]{}
, D., & [Fletcher]{}, L. 1999, [[Solar Phys. ]{}]{}, 190, 167
, T., [Luciani]{}, J. F., [Mikic]{}, Z., & [Linker]{}, J. 2000, [ [Astrophys. J. Lett. ]{}]{}, 529, L49
, S. K., [Dahlburg]{}, R. B., & [Klimchuk]{}, J. A. 1994, [ [Astrophys. J. Lett. ]{}]{}, 420, L41
, S. K., [DeVore]{}, C. R., & [Klimchuk]{}, J. A. 1999, [ [Astrophys. J. ]{}]{}, 510, 485
, T. E., [Shine]{}, R. A., [Slater]{}, G. L., [et al.]{} 2008, [ [Astrophys. J. Lett. ]{}]{}, 676, L89
, T. E., [Slater]{}, G., [Hurlburt]{}, N., [et al.]{} 2010, [ [Astrophys. J. ]{}]{}, 716, 1288
, Y., [Jiang]{}, Y., [Yang]{}, J., [et al.]{} 2014, [ [Astrophys. J. ]{}]{}, 790, 100
—. 2015, [ [Astrophys. J. ]{}]{}, 805, 48
, Y., [Jiang]{}, Y. C., [Yang]{}, L. H., & [Zheng]{}, R. S. 2011, [ [New Astronomy ]{}]{}, 16, 276
, G. E., [Howard]{}, R. A., [Koomen]{}, M. J., [et al.]{} 1995, [[Solar Phys. ]{}]{}, 162, 357
, H., [Zhang]{}, J., [Li]{}, L., & [Ma]{}, S. 2016, [ [Astrophys. J. Lett. ]{}]{}, 818, L27
, Y., [Du]{}, G., [Zhao]{}, D., [et al.]{} 2016, [ [Astrophys. J. Lett. ]{}]{}, 820, L37
, Y. 2009, [ [Astrophys. J. ]{}]{}, 697, 1529
, T. G. 1990, [ [J. Geophys. Res. ]{}]{}, 95, 11919
, P. V. 2004, [Solar Astrophysics, 2nd, Revised Edition]{}, 480
, H. R., [Holzer]{}, T. E., & [Burkepile]{}, J. T. 2001, ApJ, 549, 1221
, J. P., [Keppens]{}, R., & [Poedts]{}, S. 2010, [Advanced Magnetohydrodynamics]{}
, J., [Liu]{}, Y., [Zhang]{}, H., [et al.]{} 2010, [ [Astrophys. J. ]{}]{}, 711, 1057
, T. 1985, [[Solar Phys. ]{}]{}, 100, 415
, J., [Jiang]{}, Y., [Yang]{}, J., [et al.]{} 2014, [ [Astrophys. J. ]{}]{}, 796, 73
, J., [Jiang]{}, Y., [Yang]{}, J., [Li]{}, H., & [Xu]{}, Z. 2017, [ [Astrophys. J. ]{}]{}, 835, 35
, H. S., [Wolfson]{}, C. J., & [Metcalf]{}, T. R. 2006, [[Solar Phys. ]{}]{}, 234, 79
Jenkins, J. M., Long, D. M., van Driel-Gesztelyi, L., & Carlyle, J. 2018, Solar Physics, 293, 7
, Y., [Shen]{}, Y., [Yi]{}, B., [Yang]{}, J., & [Wang]{}, J. 2008, [ [Astrophys. J. ]{}]{}, 677, 699
, B., & [T[ö]{}r[ö]{}k]{}, T. 2006, Physical Review Letters, 96, 255002
Kliem, B., Torok, T., Titov, V. S., [et al.]{} 2014, Astrophysical Journal, 792, 107
Lemen, J. R., Title, A. M., Akin, D. J., [et al.]{} 2012, Solar Physics, 275, 17
, H., [Yang]{}, J., [Jiang]{}, Y., [et al.]{} 2018, [ [Astrophys. Space Sci. ]{}]{}, 363, 26
, H., [Jiang]{}, Y., [Yang]{}, J., [et al.]{} 2017, [ [Astrophys. J. Lett. ]{}]{}, 842, L20
Lin, J. 2004, Solar Physics, 219, 169
, J., & [Forbes]{}, T. G. 2000, [ [J. Geophys. Res. ]{}]{}, 105, 2375
Lin, J., Soon, W., & Baliunas, S. L. 2003, New Astronomy Reviews, 47, 53
Liu, R., Kliem, B., Torok, T., [et al.]{} 2012, Astrophysical Journal, 756, 59
, Y., [Jiang]{}, Y., [Ji]{}, H., [Zhang]{}, H., & [Wang]{}, H. 2003, [ [Astrophys. J. Lett. ]{}]{}, 593, L137
Liu, Y., & Kurokawa, H. 2004, Astrophysical Journal, 610, 1136
, Y., [Kurokawa]{}, H., & [Shibata]{}, K. 2005, [ [Astrophys. J. Lett. ]{}]{}, 631, L93
Liu, Y., Su, J., Xu, Z., [et al.]{} 2009, The Astrophysical Journal Letters, 696, L70
, B. C. 2001, [ [J. Geophys. Res. ]{}]{}, 106, 25141
, B. C., [Fong]{}, B., & [Fan]{}, Y. 2003, [ [Astrophys. J. ]{}]{}, 594, 1060
, Z. X., [Keppens]{}, R., [Roussev]{}, I. I., & [Lin]{}, J. 2018, [ [Astron. Astrophys. ]{}]{}, 609, A2
, R. L., [Sterling]{}, A. C., [Hudson]{}, H. S., & [Lemen]{}, J. R. 2001, [ [Astrophys. J. ]{}]{}, 552, 833
, Z. 2016, [ [Astrophys. Space Sci. ]{}]{}, 361, 22
Pesnell, W. D., Thompson, B. J., & Chamberlin, P. C. 2012, Solar Physics, 275, 3
Qin, L., Na, D., Ju, J., & Haimin, W. 2017, The Astrophysical Journal, 841, 112
, J., [Scherrer]{}, P. H., [Bush]{}, R. I., [et al.]{} 2012, [[Solar Phys. ]{}]{}, 275, 229
, C. J., [Elmore]{}, C., [Kliem]{}, B., [T[ö]{}r[ö]{}k]{}, T., & [Title]{}, A. M. 2008, [ [Astrophys. J. ]{}]{}, 674, 586
, Y., & [Liu]{}, Y. 2012, [ [Astrophys. J. ]{}]{}, 754, 7
, Y., [Liu]{}, Y., [Liu]{}, Y. D., [et al.]{} 2015, [ [Astrophys. J. Lett. ]{}]{}, 814, L17
, Y., [Liu]{}, Y., & [Su]{}, J. 2012, [ [Astrophys. J. ]{}]{}, 750, 12
, Y., [Liu]{}, Y., [Su]{}, J., & [Deng]{}, Y. 2012, [ [Astrophys. J. ]{}]{}, 745, 164
, Y., [Liu]{}, Y., [Tian]{}, Z., & [Qu]{}, Z. 2017, [ [Astrophys. J. ]{}]{}, 851, 101
, Y., [Liu]{}, Y. D., [Su]{}, J., [Qu]{}, Z., & [Tian]{}, Z. 2017, [ [Astrophys. J. ]{}]{}, 851, 67
, Y.-D., [Liu]{}, Y., & [Liu]{}, R. 2011, Research in Astronomy and Astrophysics, 11, 594
, J., [Liu]{}, Y., [Kurokawa]{}, H., [et al.]{} 2007, [[Solar Phys. ]{}]{}, 242, 53
, Z., [Liu]{}, Y., [Shen]{}, Y., [et al.]{} 2017, [ [Astrophys. J. ]{}]{}, 845, 94
, T., [Kliem]{}, B., & [Titov]{}, V. S. 2004, [ [Astron. Astrophys. ]{}]{}, 413, L27
, A. A., & [Martens]{}, P. C. H. 1989, [ [Astrophys. J. ]{}]{}, 343, 971
Wang, Y.-M., & Muglach, K. 2007, The Astrophysical Journal, 666, 1284
—. 2013, The Astrophysical Journal, 763
, Z., [Yan]{}, X., [Qu]{}, Z., & [Zhao]{}, L. 2014, [ [New Astronomy ]{}]{}, 26, 23
, X. L., [Xue]{}, Z. K., [Pan]{}, G. M., [et al.]{} 2015, [ [Astrophys. Suppl. ]{}]{}, 219, 17
, B., [Jiang]{}, Y., [Yang]{}, J., [Bi]{}, Y., & [Li]{}, H. 2016, [ [Astrophys. J. ]{}]{}, 830, 16
, B., [Jiang]{}, Y., [Yang]{}, J., [Hong]{}, J., & [Xu]{}, Z. 2015, [ [Astrophys. J. ]{}]{}, 803, 86
, B., [Jiang]{}, Y., [Yang]{}, J., [Yu]{}, S., & [Xu]{}, Z. 2016, [ [Astrophys. J. ]{}]{}, 816, 41
, J., [Jiang]{}, Y., [Yang]{}, B., [et al.]{} 2012, [ [New Astronomy ]{}]{}, 17, 732
—. 2012, [[Solar Phys. ]{}]{}, 279, 115
, J., [Jiang]{}, Y., [Zheng]{}, R., [et al.]{} 2012, [ [Astrophys. J. ]{}]{}, 745, 9
—. 2011, [[Solar Phys. ]{}]{}, 270, 551
, L., [Yan]{}, X., [Li]{}, T., [Xue]{}, Z., & [Xiang]{}, Y. 2017, [ [Astrophys. J. ]{}]{}, 838, 131
, L., [Zhang]{}, J., [Liu]{}, W., [Li]{}, T., & [Shen]{}, Y. 2013, [ [Astrophys. J. ]{}]{}, 775, 39
, A. R., [Mackay]{}, D. H., & [van Ballegooijen]{}, A. A. 2008, [[Solar Phys. ]{}]{}, 247, 103
, T., & [Shibata]{}, K. 1995, [ [Nature ]{}]{}, 375, 42
, Q., [Liu]{}, R., [Wang]{}, Y., [et al.]{} 2014, [ [Astrophys. J. ]{}]{}, 789, 133
, Q. M., [Li]{}, D., & [Ning]{}, Z. J. 2017, [ [Astrophys. J. ]{}]{}, 851, 47
Zhu, C., & Alexander, D. 2014, Solar Physics, 289, 279
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
A discussion of elemental abundance ratios as tracers of stellar populations is presented. The emphasis is on F, G, and K stars, because they represent a wide range of ages and have atmospheres providing a ‘fossil’ record of the chemical evolution of the Galaxy.
Instrumentation and methods to determine chemical abundances in stellar atmospheres are discussed in Sect. \[sect:abundet\]. High-resolution ($R > 20\,000$) spectra are required to derive precise abundance ratios, but lower resolution spectra may be useful in connection with large statistical studies of populations. Most abundance analyses are based on homogeneous 1D model atmospheres and the assumption of local thermodynamic equilibrium (LTE), but recent works have shown that 3D non-LTE corrections can change the derived trends of abundance ratios as a function of stellar metallicity significantly. However, when comparing stars having similar effective temperatures, surface gravities and metallicities, 3D non-LTE corrections tend to cancel out. Such a differential approach is the best way to disentangle stellar populations on the basis of chemical abundances.
Abundance ratios particularly useful as population tracers are discussed in Sect. \[sect:tracers\], including C/O, Na/Fe, Ni/Fe, Ba/Y, Eu/Ba, and $\alpha$/Fe, where $\alpha$ is the average abundance of Mg, Si, Ca, and Ti. The nucleosynthesis of the elements involved occurs on different time-scales in stars and supernovae with different masses. This is the main reason that these abundance ratios can be used as population tracers.
The following sections deal with a discussion of populations in the Galactic disk, the bulge, and the halo. Based on abundance ratios, there is clear evidence for two main populations in the disk: an old, thick disk formed on a time-scale of $\sim \! 10^9$years, and a younger, thin disk formed over a more extended period. For the bulge, interesting new abundance results have been obtained in recent years, including data from microlensed dwarfs, but it is too early to draw any robust conclusions about how and when the bulge formed. For the halo, there is evidence for the existence of two discrete populations with low and high values of $\alpha$/Fe, respectively. The ‘low-$\alpha$’ population has probably been accreted from dwarf galaxies, whereas the ‘high-$\alpha$’ population may consist of ancient disk stars ‘heated’ to halo kinematics by merging satellite galaxies. Globular clusters stand out from the halo field stars by showing Na-O and Al-Mg anti-correlations; there is increasing evidence that they consist of multiple stellar populations.
author:
- Poul Erik Nissen
title: Chemical abundances as population tracers
---
[**Keywords.**]{} Techniques: spectroscopic – Stars: abundances – Stars: atmospheres – Galaxy: disk – Galaxy: bulge – Galaxy: halo – Globular clusters:general – Galaxies: dwarf
Introduction {#sect:intro}
============
A population consists of a group of stars with a common origin and history. Hence, it is of high importance for studies of the formation and evolution of the Galaxy to detect and describe existing Galactic populations. This may be done by analyzing distribution functions for stars in space, kinematics, age and chemical composition. In particular, it is important to know if the main Galactic components, the disk, the bulge, and the halo, each consists of a single stellar population or if multiple populations are needed to fit data for kinematics, ages and abundance ratios of stars belonging to these components.
Whereas the original spatial and kinematical distributions of stars in a population are modified during the dynamical evolution of the Galaxy, it is generally assumed that the chemical composition of a stellar atmosphere provides a ‘fossil’ record of the composition of the Galaxy at the time and the place for the formation of the star. In this connection, F and G main-sequence and subgiant stars are of particular interest, because they span an age range as long as the lifetime of the Galaxy. Furthermore, they have an upper convection zone that mixes matter in the atmosphere with deeper layers, which tends to reduce abundance changes induced by diffusion or accretion processes (see discussion in Sect. 2.5). On the other hand, the convection zone is not so deep that elements produced by nuclear reactions in the stellar interior are brought up to the stellar surface. Hence, chemical abundances of F and G main-sequence and subgiant stars are expected to be good tracers of stellar populations.
Stars with spectral types different from F and G are also of importance as tracers of Galactic populations. O, B, and A stars may be used to probe the present composition of the Galaxy, but in some cases the atmospheric composition is affected by diffusion or accretion processes. K giants are very useful as a supplement to the F and G main-sequence stars, because they can be observed to greater distances. Their space density is, however, smaller than that of F and G dwarfs, and care should be taken because the atmospheric abundances of some elements, e.g. C and N, may be affected by convective dredge-up of the products of nuclear processes in the stellar interior.
The present review deals with the use of stellar abundance ratios to disentangle the various stellar populations in the Galaxy. Methods to determine chemical abundances in stellar atmospheres are discussed in Sect. \[sect:abundet\] with emphasis on the high precision that may be obtained when analyzing stars in a limited region of the H-R diagram differentially. In Sect. \[sect:abundet\], it is also discussed if element abundances in F and G main-sequence stars are affected by diffusion or accretion processes. Sect. \[sect:tracers\] contains an inventory of abundance ratios that are particular useful as tracers of stellar populations, and a discussion of the nucleosynthesis of the involved elements. The following sections \[sect:disks\], \[sect:bulge\], and \[sect:halo\] deal with populations associated with the Galactic disk, the bulge and the halo including globular clusters and satellite galaxies. Relations between abundance ratios, kinematics and ages will be reviewed, and scenarios for the origin of the various populations will be discussed. Finally, Sect. \[sect:conclusions\] contains conclusions and some thoughts about future observing programmes related to chemical abundances as population tracers.
The present chapter focus on stars with metallicities in the range $-3.0 <$ \[Fe/H\] $< +0.4$, where \[Fe/H\] is a logarithmic measure of the ratio between the number of iron and hydrogen atoms in the star relative to the same ratio in the Sun [^1]. Extremely metal-poor stars with \[Fe/H\] $< -3.0$ are discussed by Frebel & Norris (this volume). To some extent their chemical abundances are related to single supernovae (SNe) events, whereas a mixture of SNe with a mass distribution determined by the initial mass function (IMF) have produced the elements in more metal-rich stars.
Determination of stellar abundance ratios {#sect:abundet}
=========================================
Observation and reduction of stellar spectra {#sect:obs}
--------------------------------------------
In order to derive precise abundance ratios, high-resolution ($R = \lambda / \Delta \! \lambda > 30\,000$) and high signal-to-noise ($S/N > 100$) spectra should ideally be obtained. For such spectra, it is possible to define a reliable continuum and measure equivalent widths of weak spectral lines that have high sensitivity to abundance changes and low sensitivity to broadening parameters as microturbulence and collisional damping. Thanks to the installation of efficient echelle spectrographs in connection with many large and medium-sized telescopes, a large number of high-quality optical ($3700 < \lambda < 9000$Å) spectra for F, G, and K stars have been obtained during the last couple of decades. The infrared spectral region is still lacking behind, but important abundance results for K giants in the Galactic bulge have been obtained with the Phoenix spectrograph on the Gemini South telescope (e.g. Meléndez et al. 2008) and with the ESO VLT cryogenic echelle spectrograph, CRIRES (Ryde et al. 2010).
Spectra with somewhat lower resolution ($R \sim 20\,000$) and $S/N \sim 50$ can also be used for determining abundance ratios, and may be obtained with multi-object spectrographs such as FLAMES at the ESO VLT. This has proven to be a very effective way of getting abundance data for stars in globular clusters (e.g. Carretta et al. 2009) and satellite galaxies (see review of Tolstoy et al. 2009). Furthermore, abundances of elements that are represented by many lines in stellar spectra, such as Fe and the $\alpha$-capture elements Mg, Si, Ca, and Ti, can be obtained from medium-resolution spectra ($R \sim 5\,000 - 10\,000$). A good example is the determination of stellar abundances for the Sculptor dwarf spheroidal (dSph) galaxy by Kirby et al. (2009) with the multi-object spectrograph, DEIMOS, at the Keck II telescope.
Even low-resolution ($R \sim 2\,000$) spectra are useful for statistical investigations of \[$\alpha$/Fe\][^2] in Galactic surveys such a the Sloan Digital Sky Survey (SDSS) (Lee et al. 2011). Another large survey, the Radial Velocity Experiment (RAVE), which will deliver medium resolution spectra ($R \simeq 7\,500$) of $\sim 10^6$ stars in the near-infrared Ca[ii]{}-triplet region (8410 - 8795Å), has also the potential of supplying \[$\alpha$/Fe\] with a decent precision (Boeche et al. 2008). In the future, the ESA GAIA mission will make it possible to determine \[$\alpha$/Fe\] values for a still larger sample of stars based also on spectra in the Ca[ii]{}-triplet region, but with a somewhat higher resolution, $R \simeq 11\,500$.
The reduction of raw spectral data should include background and sky subtraction, flat-field correction, extraction of spectra, and wavelength calibration. Standard IRAF[^3] tasks or special software can be used. Care should be taken to perform a good flatfielding, including removal of possible interference fringes such that a reliable continuum can be defined from wavelength regions free of spectral lines. After normalization of the spectra, equivalent widths (EWs) of weak spectral lines can be measured by Gaussian fitting of the line profiles. For spectral lines having significant line wings (typically $EW > 70$mÅ in F and G main-sequence stars) the fitting should be performed with a Voigt profile. The continuum setting and equivalent width measurements may be done interactively with the IRAF task [splot]{} or can be done automatically (e.g. Sousa et al. 2007).
![The ESO/VLT CRIRES spectrum of the turnoff halo star G29-23 (\[Fe/H\] = $-1.7$) around the near-IR, 1.046$\mu$m S[i]{} triplet (dots) compared with synthetic LTE model-atmosphere profiles for three sulphur abundances, corresponding to \[S/Fe\] = 0.0, 0.3 and 0.6, respectively. As seen, the \[S/Fe\] = 0.3 case provides a close fit to the observations. The average S abundance determined from the three lines corresponds to \[S/Fe\] = 0.27 (Nissen et al. 2007).[]{data-label="fig:syn1046.G29-23"}](nissen_fig1.eps)
Instead of using equivalent widths, abundances can be determined by fitting a synthetic spectral line profile, calculated for a model atmosphere, to the observed line profile using the abundance of the element as a free parameter of the fit (see Fig. \[fig:syn1046.G29-23\]). For a line significantly blended by other spectral lines, this is the only way to derive a reliable abundance. Other lines have to be used to determine element abundances for the blending lines and to estimate line broadening parameters associated with stellar rotation and macroturbulence. Hence, the fitting is an iterative process that involves several line regions. It can be done automatically and may include a determination of the basic atmospheric parameters, effective temperature $T_{\rm eff}$, and surface gravity $g$, from ratios between selected lines. A good example of an automatic method is presented by Barklem et al. (2005), who determine abundances of 22 elements from ESO VLT/UVES spectra by fitting hundreds of spectral windows containing suitable lines. The procedure includes the identification of continuum points in the windows, adjustments of line centers, rejection of lines disturbed by cosmic-ray hits and $\chi$-square minimization of the difference between the synthetic and the observed spectrum. For low- and medium-resolution spectra, for which individual lines are not resolved and the continuum is not reached, abundances can be determined by similar methods (e.g. Lee et al. 2011; Kirby et al. 2009), or may be based on line-indices that can be calibrated via high-resolution data or by model-atmosphere calculations.
Model atmospheres {#sect:models}
-----------------
Stellar abundances are normally based on a model-atmosphere analysis of the available spectra. In most cases, a plane-parallel, homogeneous (1D) model is adopted, and it is assumed that the distributions of atoms over the possible excitation and ionization states are given by Boltzmann’s and Saha’s equations. This condition is called ‘Local Thermodynamic Equilibrium’ (LTE). The temperature structure of the model is derived from the requirement that the total flux of energy as transported by radiation and convection should be constant throughout the atmosphere and given by $$\begin{aligned}
F = \sigma \, T_{\rm eff}^4, \end{aligned}$$ where $\sigma$ is the Stefan-Boltzmann constant, and $T_{\rm eff}$ is the effective temperature of the star. Furthermore, the atmosphere is assumed to be in hydrostatic equilibrium and the pressure $P$ as a function of optical depth $\tau$ is determined from the equation $$\begin{aligned}
\frac{dP}{d\tau} = \frac{g}{\kappa_{c}(T,P_e)},\end{aligned}$$ where $g$ is the gravity in the stellar atmosphere and $\kappa_{c}$ the continuous absorption coefficient as determined primarily by H$^-$ absorption in optical and infrared spectra of F, G, and K stars. For these cool stars, electrons in the stellar atmosphere mainly come from the ionization of elements like Mg, Si, and Fe and the relation between total pressure and electron pressure $P_e$ therefore depends on both metallicity and the $\alpha$/Fe ratio.
Details of the construction of 1D stellar models may be found in textbooks on stellar atmospheres. The most used grid of models are the ATLAS9 models of Kurucz (1993) and the Uppsala MARCS models (Gustafsson et al. 2008). In both sets of models, convection is treated in the classical mixing-length approximation.
As reviewed by Asplund (2005), 1D models give only a first approximation to the temperature structures of stellar atmospheres. The convection creates an inhomogeneous structure with hot rising granules and cool downflows. Inhomogeneous (3D) models can be constructed by solving the standard equations for conservation of mass, momentum and energy in connection with the radiative transfer equation for a representative volume of the stellar atmosphere. The mean temperature structure of such 3D models may differ significantly from that of 1D models especially in the case of metal-poor stars. Due to the expansion of rising granulation elements and the lack of radiative heating when the line absorption coefficient is small, the 3D models have much lower temperature and electron pressure in the upper layers than classical 1D models in radiative equilibrium.
Abundance analysis {#sect:abunanal}
------------------
For a given model atmosphere, the flux $F_{\lambda}$ in an absorption line can be calculated by solving the transfer equation. Integration over the line profile relative to the continuum flux $F_c$ then gives the equivalent width $$\begin{aligned}
EW = \int{\frac{F_c - F_{\lambda}}{F_c}} d\lambda .\end{aligned}$$
It is the ratio between the line and continuous absorption coefficients, $\kappa_l/\kappa_c$, that determines the line depth and hence the equivalent width. For a weak (un-saturated) line, the equivalent width is approximately proportional to the abundance ratio $N_{\rm X}/N_{\rm H}$, where X is the element corresponding to the line. For saturated lines, the equivalent width also depends on line broadening due to small scale turbulent gas motions. In 1D modelling this introduces an additional atmospheric parameter, the microturbulence, that can be determined from the requirement that the same Fe abundance should be derived from weak and medium-strong Fe[i]{} lines. Strong lines with damping wings are sensitive to the the value of the collisional damping constant. Clearly, the most accurate abundances are derived from weak lines if observed with high resolution and $S/N$.
The equivalent width of a line also depends on the oscillator strength and the populations of the energy levels corresponding to the line. In LTE, Boltzmann’s and Saha’s equations are used to determine the population numbers. This may, however, be a poor approximation as reviewed by Asplund (2005). Instead, one can use that a stellar atmosphere is in a steady state, i.e that the population $n_i$ of a level $i$ does not vary in time. This can be expressed as $$\begin{aligned}
n_i \sum_{j=1}^{N}{(R_{ij} + C_{ij})} = \sum_{j=1}^{N}{n_j (R_{ji} + C_{ji})},\end{aligned}$$ where $R$ and $C$ are the transition rates for radiative and collisional processes, respectively. The summation is extended over all $N$ levels with $j \neq i$. In such, so-called non-LTE calculations, the population numbers are found by solving $N$ equations of the same type as Eq. (4). In addition, the transfer equation must be solved, because the radiative transition rates depend on the mean intensity of the radiation.
Departures from LTE can be large and affect derived stellar abundances very significantly (Asplund 2005). However, in some cases collisional transition rates are not well known, and the calculated non-LTE populations become rather uncertain. In particular, this is the case for inelastic collisions with neutral hydrogen atoms. Often, the recipes of Drawin (1969) are adopted, but since these estimates are based on classical physics, they only provide an order-of-magnitude estimate of the collisional rates. Hence, a scaling factor $S_{\rm H}$ to the Drawin formula has to be introduced. It may be calibrated on the basis of solar spectra by requesting that lines with different excitation potential and from different ionization stages should provide the same abundance or one may vary $S_{\rm H}$ to investigate how the uncertainty of collisional rates affects the derived abundances.
Given that non-LTE calculations are sometimes uncertain and that a grid of 3D models is not yet available, a [*differential*]{} 1D LTE analysis is often applied to determine abundance ratios. For narrow ranges in the basic atmospheric parameters, say $\pm 400$K in $T_{\rm eff}$, $\pm 0.4$dex in log$g$, and $\pm 0.5$dex in \[Fe/H\], one may assume that non-LTE and 3D effects on the abundances are about the same for all stars. Hence, precise differential abundances with respect to a standard star can be derived in LTE.
For F and G stars with metallicities around \[Fe/H\] = 0, the Sun is an obvious choice as a standard, and logarithmic abundance ratios with respect to the Sun, like \[Mg/Fe\], can be derived from the same lines in the spectra of the stars and the Sun. At lower metallicities, bright stars with well known atmospheric parameters can be chosen as standards. This method has the additional advantage that the oscillator strength of a line cancels out so that its error plays no role. Such differential abundance ratios can be determined to a precision of about $\pm 0.03$dex (e.g. Neves et al. 2009; Nissen & Schuster 2010). When using chemical abundances to trace stellar population, it is just these very precise differential abundance ratios at a given metallicity that are needed. Trends of abundance ratios as a function of \[Fe/H\] derived under the LTE assumption are, on the other hand, less accurate, because non-LTE and 3D effects change with metallicity.
Determination of atmospheric parameters for F, G, and K stars {#sect:param}
-------------------------------------------------------------
In order to determine precise abundance ratios, reliable values of the stellar atmospheric parameters, $T_{\rm eff}$, $g$, and \[Fe/H\], must be determined. Some abundance ratios like \[Mg/Fe\] determined from neutral atomic lines are fairly insensitive to errors in the atmospheric parameters, but other ratios like \[O/Fe\] with the oxygen abundance determined from the high excitation O[i]{} triplet or from OH lines depend critically on the adopted values for $T_{\rm eff}$ and $g$.
The effective temperature of a late-type star can be determined from a colour index, e.g. $V\!-\!K$, calibrated in terms of $T_{\rm eff}$ by the infrared flux method. Two recent implementations of this method (González Hernández & Bonifacio 2009; Casagrande et al. 2010) give consistent calibrations of $V - K$. In the case of nearby stars for which colours are not affected by interstellar reddening, $T_{\rm eff}$ can be determined to an accuracy of the order of $\pm 50$K. For more distant stars, the reddening is, however, a problem and $T_{\rm eff}$ is better determined spectroscopically, e.g. from the wings of Balmer lines or from the requirement that \[Fe/H\] derived from Fe[i]{} lines should be independent of the excitation potential of the lines. In this way, differential values of $T_{\rm eff}$ can be determined to a precision of $\pm 25$K (Nissen 2008).
The best way to determine the stellar surface gravity $$\begin{aligned}
g = G \frac{\cal{M}}{R^2}\end{aligned}$$ is to estimate the mass $\cal{M}$ from stellar evolutionary tracks and the radius $R$ from the basic relation $L \propto R^{2} \, T_{\rm eff} ^{4}$, where $L$ is the luminosity of the star. This leads to the following expression for the gravity of a star relative to that of the Sun (log$g_{\odot} = 4.44$ in the cgs system)
$$\begin{aligned}
\log \frac{g}{g_{\odot}} = \log \frac{\cal{M}}{\cal{M}_{\odot}} +
4 \log \frac{T_{\rm eff}}{T_{\rm eff,\odot}} +
0.4 (M_{\rm bol} - M_{{\rm bol},\odot}),\end{aligned}$$
where $M_{bol}$ is the absolute bolometric magnitude, which can be determined from the apparent magnitude if the distance to the star is known.
This method of determining surface gravities works well for nearby stars for which distances are accurately known from Hipparcos parallaxes. For more distant late-type stars, the gravity can be determined spectroscopically from the difference in \[Fe/H\] derived from neutral and ionized iron lines. Fe[i]{} lines change very little with $g$, whereas Fe[ii]{} lines change significantly. Departures from LTE in the ionization equilibrium of Fe should, however, be taken into account. This may be done by requiring that the difference \[Fe/H\](Fe[ii]{})$-$\[Fe/H\](Fe[i]{}) has the same value as in the case of a standard star with a surface gravity that is accurately determined from Eq. (6). In this way differential values of log$g$ can be determined to a precision of about $\pm 0.05$dex (Nissen 2008).
Due to the non-LTE effects on the ionization balance of Fe, \[Fe/H\] should be determined from Fe[ii]{} lines, because they represent the dominating ionization stage of iron. If LTE is assumed, \[Fe/H\] derived from Fe[i]{} lines turns out to be 0.1 to 0.2 dex lower than \[Fe/H\] derived from Fe[ii]{} lines in the case of metal-poor F, G, and K stars. It is likely that this problem is due to a higher degree of Fe[i]{} ionization than predicted by the Saha equation (Mashonkina et al. 2011).
Diffusion and dust-gas separation of elements {#sect:diffusion}
---------------------------------------------
As mentioned in Sect. \[sect:intro\], it is generally assumed that the atmosphere of a late-type star with an upper convection zone has retained a ‘fossil’ record of the composition of the Galaxy at the time and the place for the formation of the star. A high-resolution study of the metal-poor (\[Fe/H\]$\sim -2$) globular cluster NGC6397 by Korn et al. (2007) indicates, however, that the abundances of Mg, Ca, Ti, and Fe in main-sequence turnoff stars are about 0.12dex (30%) lower than the abundances of these elements in K giants. This may be explained by downward diffusion of the elements at the bottom of the convection zone for turnoff stars. The elements are depleted by about the same factor, so the effect of diffusion on abundance ratios is less than 0.05dex.
For the solar atmosphere, the depletion by diffusion of elements heavier than boron is predicted to have been about 0.04dex (Turcotte & Wimmer-Schweingruber 2002) and the effect of diffusion on abundance ratios is negligible. This is confirmed by the good agreement of abundance ratios for non-volatile elements in the solar atmosphere and in the most primitive meteorites, the carbonaceous chondrites (Asplund et al. 2009). A very precise study of solar ‘twin’ stars (i.e. stars having nearly the same $T_{\rm eff}$, $g$, and \[Fe/H\] as the Sun) by Meléndez et al. (2009) shows, on the other hand, that the Sun has a higher abundance ratio of volatile elements (C, N, O, S, and Zn) with respect to Fe than the large majority of twin stars. The deviation is about 0.05dex. As suggested by the authors, this may be explained by selective accretion of refractory elements, including iron, on dust particles in the proto-solar disk. Thus, some fraction of the refractory elements may end up in terrestrial planets. If true, abundance ratios like \[O/Fe\] and \[S/Fe\] can deviate by $\sim \! 0.05$dex from the original ratio in the interstellar cloud that formed the star depending on whether the star is with or without terrestrial planets.
It is concluded that the effects of diffusion may change some abundance ratios by up to 0.05dex for metal-poor stars with relatively thin convection zones. In the case of disk stars, the abundances of volatile elements relative to iron could be affected by dust-gas separation in connection with star and planet formation by $\sim\!0.05$dex. For refractory elements there are no indications of differences in abundance rations between stars with and without detected planets (Neves et al. 2009).
Elements used as stellar population tracers {#sect:tracers}
===========================================
This section presents a discussion of some abundance ratios that have been used as tracers of stellar populations. In several cases, the nucleosynthesis and chemical evolution of the corresponding elements are not well understood; nevertheless, the abundance ratios have proven to be very useful in disentangling stellar populations. In addition, the observed differences and trends provide important constraints on SNe modeling and theories of Galactic chemical evolution.
Carbon and oxygen {#sect:CO}
-----------------
Abundances of C and O can be determined from spectral lines corresponding to forbidden transitions between low excitation states and allowed transitions between high-excitation states for neutral atoms. In addition, molecular CH and OH lines in the blue-UV and infrared spectral regions can be applied.
The most reliable C and O abundances are derived from the forbidden \[C[i]{}\] $\lambda 8727$ and \[O[i]{}\] $\lambda 6300$ lines, provided that these rather weak lines are measured with sufficient spectral resolution and $S/N$. Both lines are blended; \[O[i]{}\] by a Ni[i]{} line and \[C[i]{}\] by a weak Fe[i]{} line. These blends must be taken into account when deriving abundances. A strong collisional coupling with the ground states ensures that the \[C[i]{}\] and \[O[i]{}\] lines are formed in LTE. The correction for 3D effects is small for the Sun (Asplund 2005) but increases with decreasing metallicity and may reach as much as $-0.2$dex in turnoff stars with \[Fe/H\] = $-2$ (Nissen et al. 2002).
The \[C[i]{}\] line is too weak to be a useful abundance indicator for metal-poor (\[Fe/H\]$< -1$) dwarf and subgiant stars. In giants, carbon abundances are changed by dredge-up of gas affected by CN-cycle hydrogen burning. The \[O[i]{}\] line, on the other hand, can be used to derive oxygen abundances in dwarfs and subgiants down to \[Fe/H\]$\sim -2$ and in giants down to metallicities around $-3$. Alternatively, carbon and oxygen abundances in halo stars can be derived from high-excitation atomic lines (the C[i]{} lines around 9100Å and the O[i]{} triplet at 7774Å) but in both cases the non-LTE effects are uncertain (Fabbian et al. 2009). CH and OH lines can be used to derive carbon and oxygen abundances even at extremely low metallicities, but they are very sensitive to $T_{\rm eff}$ and large 3D corrections should be applied (Asplund 2005). Due to these problems, C and O abundances in halo stars are quite uncertain. Carbon seems to follow iron, i.e. \[C/Fe\] $\simeq 0$ from \[Fe/H\] = 0 to $-3$ (Bensby & Feltzing 2006; Fabbian et al. 2009). \[O/Fe\] raises from zero to about +0.5dex, when the metallicity of disk stars decreases from \[Fe/H\]=0 to $-1$ and then stays approximately constant at \[O/Fe\] $\simeq +0.5$ down to \[Fe/H\] $\simeq -2$ (Nissen et al. 2002). According to Cayrel et al. (2004), who derived oxygen abundances in giants from the \[O[i]{}\] line, the constant level of \[O/Fe\] continues all the way down to \[Fe/H\] $\simeq -3.5$, if one assumes that 3D corrections are the same as in metal-poor dwarf stars.
Although the trends of \[C/Fe\] and \[O/Fe\] as a function of \[Fe/H\] are somewhat uncertain due to non-LTE and 3D effects, the ratio between the abundances of C and O is more immune to these problems. This stems from the fact that the forbidden \[C[i]{}\] and \[O[i]{}\] lines have about the same dependence of temperature and pressure. Hence, the derived \[C/O\] is insensitive to 3D effects. The same is the case for \[C/O\] derived from high-excitation atomic lines. Furthermore, as shown by Fabbian et al. (2009), the non-LTE corrections of C and O abundances derived from C[i]{} and O[i]{} lines tend to cancel, so that the trend of \[C/O\] vs. \[O/H\] is fairly independent of the choice of the hydrogen collision parameter, $S_{\rm H}$ (see Sect. 2.3). Fig. \[fig:CO-O.all\] shows this trend for $S_{\rm H} = 1$, with the abundances determined from forbidden lines for disk stars and from high-excitation atomic lines for the halo stars.
![The carbon-to-oxygen ratio as a function of the oxygen abundance. Open circles refer to thin-disk and filled circles to thick-disk stars with data adopted from Bensby & Feltzing (2006). Crosses are halo stars from Fabbian et al. (2009), and (red) triangles show DLA data from Cooke et al. (2011).[]{data-label="fig:CO-O.all"}](nissen_fig2.update.eps)
Carbon is synthesized in stellar interiors by the triple-$\alpha$ process, but it is unclear which objects are the main contributors to the chemical evolution of carbon in the Galaxy. Type II SNe, Wolf-Rayet stars, intermediate- and low-mass stars in the planetary nebula phase, and stars at the end of the giant phase have been suggested. Oxygen, on the other hand, seem to be produced exclusively by $\alpha$-capture on C in short-lived massive stars and is dispersed to the interstellar medium by type II SNe. Hence, the C/O ratio has the potential of being a good tracer of stellar populations. The separation between thin- and thick-disk stars in Fig. \[fig:CO-O.all\] shows that this is indeed the case.
The approximately constant \[C/O\]$\simeq -0.5$ for halo stars with \[O/H\] between $-2.0$ and $-0.5$ probably corresponds to the C/O yield ratio for massive stars. The increase in \[C/O\] for the thick-disk stars may be due to metallicity dependent winds from Wolf-Rayet stars, whereas the delayed production of carbon by low- and intermediate-mass stars can explain the higher \[C/O\] in thin-disk stars. The upturn of \[C/O\] at the lowest values of \[O/H\], which is also found for distant damped Lyman-$\alpha$ (DLA) galaxies (Cooke et al. 2011), could be due to enhanced carbon production by massive first generation stars with extremely high rotation velocities (Chiappini et al. 2006).
Intermediate-mass elements {#sect:intermediate}
--------------------------
The even-$Z$ elements, Mg, Si, S, Ca, and Ti are mainly produced by successive capture of $\alpha$-particles in connection with carbon, oxygen and neon burning in massive stars and dispersed into the interstellar medium by type II supernovae explosions on a time-scale of $\sim 10^7$ years. Iron is also produced by SNe II, but the bulk of Fe comes from type Ia SNe on a much longer time-scale ($\sim 10^9$ years). Hence, the ratio between the abundance of an $\alpha$-capture element and iron in a star depends on how long the star-formation process had proceeded before the star was formed. In this way, \[$\alpha$/Fe\] becomes an important tracer of populations.
Mg, Si, Ca, and Ti abundances can be determined from several weak atomic lines in the optical spectra of late-type stars, whereas sulphur is more difficult as discussed below. Traditionally, \[$\alpha$/Fe\] is therefore defined as the average value of \[Mg/Fe\], \[Si/Fe\], \[Ca/Fe\], and \[Ti/Fe\] (see Sect. 2.1). As discussed by Asplund (2005), departures from LTE have some effects on the metallicity trends of \[$\alpha$/Fe\] but probably not more than 0.1dex, when weak lines are used. If the abundances are based on neutral lines, \[$\alpha$/Fe\] is quite insensitive to $T_{\rm eff}$ and 3D effects, although high excitation lines are to be preferred (Asplund 2005, Fig. 8)
The trend of \[$\alpha$/Fe\] with metallicity is characterized by an increase of about 0.3dex when \[Fe/H\] decreases from 0 to $-1$. Below \[Fe/H\] = $-1$, \[$\alpha$/Fe\] is distributed around a plateau at 0.3dex, but as discussed later, there are very significant differences in \[$\alpha$/Fe\] at a given metallicity related to stellar populations both for disk and halo stars.
Sulphur is an $\alpha$-capture element, and \[S/Fe\] is therefore expected to show a plateau-like behaviour for halo stars, but very high values \[S/Fe\] $\sim 0.8$ have been claimed at the lowest metallicities (Israelian & Rebolo 2001). These values may, however, be spurious due to the difficulty of measuring the very weak S[i]{} line at 8694.6Å. On the basis of stronger S[i]{} lines at 9212.9 and 9237.5Å measured with UVES and corrected for telluric absorption lines, Nissen et al. (2007) find a plateau-like behaviour of \[S/Fe\] as shown in Fig. 3. Non-LTE corrections from Takeda et al. (2005) corresponding to $S_{\rm H}$ = 1 were included for sulphur, and the iron abundances were derived from Fe[ii]{} lines with negligible non-LTE effects (Mashonkina et al. 2011) . This result is supported by CRIRES observations of the near-IR S[i]{} triplet (see Fig. \[fig:syn1046.G29-23\]). Further studies of sulphur in Galactic stars would be important, especially because S is a volatile element that is undepleted onto dust. As such it can be used to measure the $\alpha$-enhancement of DLA galaxies.
![\[S/Fe\] as a function of \[Fe/H\]. Plus signs refer to data for disk stars from Chen et al. (2002) and circles to halo stars from Nissen et al. (2007). Filled circles with error bars are based on S abundances derived from the $\lambda \lambda 9212.9, 9237.5$ S[i]{} lines, whereas open (red) circles show data determined from the weak $\lambda 8694.6$ S[i]{} line.[]{data-label="fig:SFe-Fe.NLTE"}](nissen_fig3.eps)
In addition to the $\alpha$-capture elements, Na is an interesting and sensitive tracer of stellar populations. It is thought to be made during carbon and neon burning in massive stars and is expelled by type II SNe together with the $\alpha$-elements. The amount of Na made is, however, controlled by the neutron excess, which depends on the initial heavy element abundance in the star (Arnett 1971). This is probably the explanation of the fact that \[Na/Mg\] in halo stars correlates with \[Mg/H\] with a slope of about 0.5 (Nissen & Schuster 1997; Gehren et al. 2004).
Na abundances are best determined from the relatively weak Na[i]{} doublets $\lambda \lambda 5682.6, 5688.2$ and $\lambda \lambda 6154.2, 6160.7$, for which non-LTE and 3D corrections are rather small (Asplund 2005). At low metallicities, \[Fe/H\] $< -2.0$, these lines are too faint to be measured, and Na abundances are derived from the Na[i]{}D $\lambda \lambda 5890.0, 5895.9$ resonance lines. For this doublet, the non-LTE correction is large and reaches $-0.4$dex for extremely metal-poor stars (Gehren et al. 2004). Furthermore, the use of these lines is sometimes complicated by overlapping interstellar Na[i]{}D lines and telluric H$_2$O lines.
In addition to Na production in massive stars, sodium can also be made in hydrogen-burning shells of intermediate- and low-mass stars via the CNO and Ne-Na cycles. This is probably the explanation of the Na-O anti-correlation in globular clusters as further discussed in Sect. \[sect:globclusters\].
The iron-peak elements {#sect:ironpeak}
----------------------
Among the iron-peak elements - Cr to Zn - the even-$Z$ elements Cr, Fe, and Ni are represented by many lines in the spectra of late-type stars, which makes it possible to determine very precise abundance ratios, \[Cr/Fe\] and \[Ni/Fe\].
The ratio between Cr and Fe abundances is found to be the same as in the Sun, i.e. \[Cr/Fe\] $\simeq 0$, for all Galactic populations with \[Fe/H\] $> -2$. Below this metallicity, McWilliam et al. (1995) and Cayrel et al. (2004) have found a smooth decrease of \[Cr/Fe\] to $\sim -0.5$ at \[Fe/H\] = $-4$ based on an LTE study of Cr[i]{} lines in very metal-poor giants. This is surprising, because Cr and Fe are predicted to be synthesized in constant ratios by explosive silicon burning in both type II and Ia SNe. The problem seems to have been solved by Bergemann & Cescutti (2010), who find that the derived decline of \[Cr/Fe\] is an artifact caused by the neglect of non-LTE effects. They obtain a very satisfactory agreement between non-LTE Cr abundances derived from Cr[i]{} and Cr[ii]{} lines when using surface gravities based on Hipparcos parallaxes.
For a long time, the Ni/Fe ratio was also thought to be solar in all stars. This is clearly the case for disk population stars (e.g. Chen et al. 2000), but Nissen & Schuster (1997, 2010) have shown that Ni is underabundant relative to Fe, i.e. \[Ni/Fe\] $\sim -0.1$ to $-0.2$, in some halo stars with \[Fe/H\] $> -1.4$. These stars have also low \[$\alpha$/Fe\] and \[Na/Fe\] values, and a tight correlation between \[Ni/Fe\] and \[Na/Fe\] is present (see Sect. \[sect:twohalopop\]). Similar stars are found in dwarf galaxies (Tolstoy et al. 2009). The reason for this correlation is probably that the supernovae yields of $^{23}$Na and $^{58}$Ni depend on the neutron excess (see the detailed discussion by Venn et al. 2004).
The determination of abundances of the odd-$Z$ elements Mn, Co, and Cu is more difficult than in the case of Cr, Fe, and Ni. There are fewer lines available and hyper-fine-structure (HFS) splitting has to be taken into account. Furthermore, non-LTE effects seem to be very significant. Positive corrections of both \[Mn/Fe\] (Bergemann & Gehren 2008) and \[Co/Fe\] (Bergemann et al. 2010) are obtained, although the size of these corrections depends somewhat on the adopted $S_{\rm H}$ parameter for hydrogen collisions. This means that the strong decline of \[Mn/Fe\] as a function of decreasing \[Fe/H\] found from an LTE analysis of Mn[i]{} lines by e.g. Reddy et al. (2006) and Neves et al. (2009) has to be modified to a more shallow trend. For Co the non-LTE study of Bergemann et al. (2010) leads to surprisingly large over-abundances of cobalt with respect to iron for halo stars, a result that is in disagreement with expectations from presently calculated supernovae yields.
![\[Cu/Fe\] as a function of \[Fe/H\]. Stars studied by Mishenina et al. (2002) are indicated by crosses. Open (blue) circles refer to high-$\alpha$ stars and filled (red) circles to low-$\alpha$ halo stars from Nissen & Schuster (2011). Filled (green) triangles are data for giant stars in the inner part of the Large Magellanic Cloud adopted from Pompéia et al. (2008). []{data-label="fig:Cu-Fe"}](nissen_fig4.update.eps)
Copper is a very interesting element as a tracer of metal-poor stellar populations. Abundances can be determined from the Cu[i]{} lines at 5105.5, 5218.2 and 5782.1Å. They are affected by HFS splitting to different degrees. Non-LTE calculations are not yet available, but LTE data show that \[Cu/Fe\] $\simeq 0$ for disk stars with no significant difference between the thin and the thick disk (Reddy et al. 2006). At \[Fe/H\] $\sim -1$, \[Cu/Fe\] starts to decline steeply and reaches a plateau of \[Cu/Fe\] $\sim -0.6$ below \[Fe/H\]$\simeq -1.6$ (Mishenina et al. 2002) as shown in Fig. \[fig:Cu-Fe\]. Low-$\alpha$ halo stars deviate, however, from this trend with½ a \[Cu/Fe\] deficiency of 0.2 to 0.5dex (Nissen & Schuster 2011). The same is the case for the more metal-rich part of stars in the globular cluster $\omega$ Centauri (Cunha et al. 2002), and stars belonging to the Sagittarius dSph galaxy (Sbordone et al. 2007) and the Large Magellanic Cloud (Pompéia et al. 2008) as shown in Fig. \[fig:Cu-Fe\]. These \[Cu/Fe\] data provide important constraints on the nucleosynthesis of Cu. Romano & Matteucci (2007) suggest that copper is initially made by explosive nucleosynthesis in type II SNe and later by a metallicity dependent neutron-capture process (the weak $s$-process) in massive stars.
Zinc abundances can be determined from the $\lambda \lambda 4722.2, 4810.5$ Zn[i]{} lines. Non-LTE and 3D corrections are modest, i.e. of the order of +0.1dex for metal-poor stars (Nissen et al. 2007). It is sometimes assumed that \[Zn/Fe\] = 0, which means that Zn can be used as a proxy of Fe in determining the metallicity of interstellar gas, e.g. in DLA systems; as a volatile element Zn is not depleted onto dust like Fe. Newer studies show, however, that \[Zn/Fe\] reaches +0.15dex in metal-poor disk stars with perhaps a small separation between thin- and thick-disk stars (Bensby et al. 2005). For halo stars, \[Zn/Fe\] increases from zero at \[Fe/H\] = $-1$ to \[Zn/Fe\] $\simeq 0.2$ at \[Fe/H\] = $-2.5$ (Nissen et al. 2007), and at still lower metallicities \[Zn/Fe\] increases steeply to \[Zn/Fe\] $\simeq 0.5$ at \[Fe/H\] = $-3.5$ (Cayrel et al. 2004, Nissen et al. 2007). Hence the behaviour of zinc is complicated, and the nucleosynthesis of this element is not well understood. Several ways of producing Zn have been suggested: the weak $s$-process in massive stars, explosive Si burning in type II and type Ia SNe, and the main $s$-process in low- and intermediate-mass stars.
The neutron capture elements {#sect:ncapture}
----------------------------
Among the heaviest elements, Y, Ba, and Eu are of particular interest as tracers of stellar populations. They are made by neutron capture processes, which are traditionally divided into the slow $s$-process and the rapid $r$-process. In the $s$-process, neutrons are added on a long time-scale compared to that of $\beta$ decays so that nuclides of the $\beta$-stability valley are built up. In the $r$-process, neutrons are added so fast that nuclides on the neutron-rich side of the stability valley are made. The $s$-process is divided into the main $s$-process that occurs in low- and intermediate-mass asymptotic giant branch (AGB) stars and the weak $s$-process occurring in massive stars. The $r$-process is not well understood, but it is thought to occur in connection with type II SNe explosions.
The relative contribution of the $s$- and the $r$-process to heavy elements in the solar system has been determined by Arlandini et al. (1999). Barium is called an $s$-process element, because 81% of the solar Ba is due to the $s$-process. Europium, on the other hand, is an $r$-process element, because 94% in the Sun originates from the $r$-process. In metal-poor stars, for which only massive stars have contributed to the nucleosynthesis of the elements, both Ba and Eu are, however, made by the $r$-process, provided that the contribution from the weak $s$-process is negligible. In such metal-poor stars, one would expect to find an $r$-process ratio $N_{\rm Eu}/N_{\rm Ba} \simeq 5$ between the abundances of europium and barium corresponding to \[Eu/Ba\] $\simeq 0.7$. These considerations suggest that \[Eu/Ba\] may be a useful tracer of stellar populations. As seen from Fig. \[fig:Eu-Ba\], this is confirmed by Mashonkina et al. (2003).
Ba abundances can be determined from subordinate Ba[ii]{} lines at 5853.7, 6141.7, and 6496.9Å. The stronger resonance line at 4554.0Å may also be used, but the analysis of this line is complicated by the presence of isotopic and HFS splitting. Europium abundances are primarily obtained from the Eu[ii]{} line at 4129.7Å, which has to be analyzed by spectrum synthesis techniques, because the line is strongly broadened by isotopic and HFS splitting.
![\[Eu/Ba\] as a function of \[Fe/H\] with data adopted from Mashonkina et al. (2003). Crosses refer to stars with halo kinematics, filled circles to thick-disk stars and open circles to thin-disk stars.[]{data-label="fig:Eu-Ba"}](nissen_fig5.eps)
The barium-to-yttrium ratio is another interesting tracer of stellar populations. Yttrium ($Z =39$) belongs to the first peak of $s$-process elements around the neutron magic number $N = 50$, whereas barium ($Z =56$) is at the second peak around $N = 82$. The ratio Ba/Y (sometimes called heavy-$s$ to light-$s$, hs/ls) depends on the the neutron flux per seed nuclei, and is predicted to be high for metal-poor, low-mass AGB stars (Busso et al. 1999). As pointed out by Venn et al. (2004), the Ba/Y ratio in stars belonging to dSph satellite galaxies is much higher than Ba/Y in Galactic halo stars with differences on the order of 0.6dex in the metallicity range $-2 <$ \[Fe/H\] $< -1$. Similar large offsets are found for stars with \[Fe/H\] around $-1$ in the globular cluster $\omega$Cen (Smith et al. 2000) and for stars in the LMC (Pompéia et al. 2008). According to Fenner et al. (2006), this indicates that the chemical evolution of these systems has been so slow that winds from low-mass AGB stars have started to enrich the interstellar medium with $s$-process elements at a metallicity around \[Fe/H\] $\sim -2$.
The Galactic disk {#sect:disks}
=================
The thick and the thin disk {#thinthick}
---------------------------
A long-standing problem in studies of Galactic structure and evolution has been the possible existence of a population of stars having kinematics, ages, and chemical abundances in between the characteristic values for the halo and the disk.
On the basis of a large program of $uvby$-$\beta$ photometry of F and early G type main-sequence stars within 100pc, Strömgren (1987) concluded that an [*intermediate Population II*]{} does exist. \[Fe/H\] was determined from a metallicity index $m_1$ = $(v-b) - (b-y)$, which is sensitive to the line blanketing in the $v$-band, and age was derived from the position of a star in the $c_1$ - $(b-y)$ diagram, where $c_1$ = $(u-v) - (v-b)$ is a measure of the Balmer discontinuity at 3650Å and hence of the surface gravity of the star. The colour index $(b-y)$ is used as a measure of $T_{\rm eff}$. After discussing the calibration of the $m_1$ index in terms of \[Fe/H\], and $c_1$ as a function of absolute magnitude, Strömgren found that intermediate Population II consists of 10-15 Gyr old stars having $-0.8 <$ \[Fe/H\] $< -0.4$ and velocity dispersions that are significantly greater than those of the younger, more metal-rich disk stars.
In a seminal paper, Gilmore & Reid (1983) showed that the distribution of stars in the direction of the Galactic South Pole cannot be fitted by a single exponential, but requires two disk components – a [*thin disk*]{} with a scale height of 300 pc and a [*thick disk*]{} with a scale height of about 1300 pc. They identified intermediate Population II with the sum of the metal-poor end of the thin disk and the thick disk. Following this work, it has been intensively discussed if the thin and thick disks are discrete components of our Galaxy or if there is a more continuous sequence of stellar populations connecting the Galactic halo and the thin disk.
In another important paper, Edvardsson et al. (1993) derived precise abundance ratios for 189 stars belonging to the disk. Stars in the temperature range $5600 < T_{\rm eff} < 7000$K and somewhat evolved away from the zero-age main-sequence were selected from $uvby$-$\beta$ photometry of stars in the solar neighbourhood (i.e. in a region around the Sun with a radius of $\sim$100pc) and divided into nine metallicity bins ranging from \[Fe/H\] $\sim -1.0$ to $\sim +0.3$. In each metallicity bin the $\sim \! 20$ brightest stars were observed. Hence, there is no kinematical bias in the selection of the stars.
The Edvardsson et al. (1993) survey provides clear evidence for a scatter in \[$\alpha$/Fe\] at a given metallicity for stars in the solar neighbourhood. This is shown in Fig. \[fig:Edv.alpha-Fe.Rm\]. As seen, \[$\alpha$/Fe\] for stars in the metallicity range $-0.8 <$ \[Fe/H\] $< -0.4$ is correlated with the mean galactocentric distance $R_m$ in the stellar orbit. Stars with $R_m > 9$ kpc tend to have lower \[$\alpha$/Fe\] than stars with $R_m < 7$ kpc, and stars belonging to the solar circle lie in between. Assuming that $R_m$ is a statistical measure of the distance from the Galactic center at which the star was born, Edvardsson et al. explained the \[$\alpha$/Fe\] variations as due to a star formation rate that declines with galactocentric distance. In other words, type Ia SNe start contributing iron at a higher \[Fe/H\] in the inner parts of the Galaxy than in the outer parts.
![\[$\alpha$/Fe\] as a function of \[Fe/H\] with data from Edvardsson et al. (1993). Stars shown with filled circles have a mean galactocentric distance in their orbits $R_m < 7$ kpc. Open circles refer to stars with $7 < R_m < 9$ kpc, and crosses refer to stars with $R_m > 9$ kpc.[]{data-label="fig:Edv.alpha-Fe.Rm"}](nissen_fig6.eps)
The group of stars in Edvardsson et al. (1993) with $R_m < 7$ kpc have kinematical properties similar to those of the thick disk, for which the dispersions of the Galactic velocity components, $U$, $V$, and $W$, with respect to the Local Standard of Rest (LSR) are determined to be ($\sigma_{U}, \sigma_{V}, \sigma_{W}) \simeq$ (65, 50, 40)kms$^{-1}$ , and for which the asymmetric drift with respect to the LSR is $V_{\rm ad} \simeq -50$kms$^{-1}$. In comparison, thin disk stars in the solar neighbourhood have ($\sigma_{U}, \sigma_{V}, \sigma_{W}) \simeq$ (40, 30, 20) kms$^{-1}$ and $V_{\rm ad} \simeq -10$kms$^{-1}$. Thus, thick-disk stars move on more eccentric orbits than the thin-disk stars, and due to the increasing density of stars towards the inner part of the Galaxy, thick disk stars presently situated in the solar neighbourhood tend to be close to the apo-galactic distance in their orbits. This means that they have smaller mean galactocentric distances than the thin-disk stars. The differences in \[$\alpha$/Fe\] shown in Fig. \[fig:Edv.alpha-Fe.Rm\] may, therefore, also be interpreted as due to a systematic difference in \[$\alpha$/Fe\] between the thin and the thick disk. Apparently, Gratton et al. (1996) were first to suggest this interpretation of the \[$\alpha$/Fe\] data.
A more clear chemical separation between thin- and thick-disk stars has been obtained by Fuhrmann (2004). For a sample of nearby stars with $5300 < T_{\rm eff} < 6600$K and $3.7 < \log\,g < 4.6$, he derives very precise Mg abundances from Mg[i]{} lines and Fe abundances from Fe[i]{} and Fe[ii]{} lines. In a \[Mg/Fe\] vs. \[Fe/H\] diagram, stars with thick disk kinematics have \[Mg/Fe\] $\simeq +0.4$ and \[Fe/H\] between $-1.0$ and $-0.3$. The thin disk stars show a well-defined sequence from \[Fe/H\] $\simeq -0.6$ to +0.4 with \[Mg/Fe\] decreasing from +0.2 to 0.0. Hence, there is a \[Mg/Fe\] separation between thick and thin disk stars in the overlap region $-0.6 <$ \[Fe/H\] $ < -0.3$ with only a few ‘transition’ stars. This is even more striking in a diagram, where \[Fe/Mg\] is plotted as a function of \[Mg/H\] (Fuhrmann 2004, Fig.34).
On the basis of stellar ages derived from evolutionary tracks in the $M_{\rm bol}$ - log$T_{\rm eff}$ diagram, Fuhrmann (2004) finds that the maximum age of thin disk stars is about 9 Gyr, whereas thick disk stars have ages around 13Gyr. This suggests that the systematic difference of \[Mg/Fe\] is connected to a hiatus in star formation between the thick and thin disk phases.
The \[$\alpha$/Fe\] distribution of disk stars {#sect:alphadistr}
----------------------------------------------
Two major studies of abundance ratios in thin- and thick-disk stars (Bensby et al. 2005; Reddy et al. 2003, 2006) are based on kinematically selected groups of stars in the solar neighbourhood. In these works, it is assumed that the kinematics of stars can be represented by Gaussian distribution functions for the velocity components, $U$, $V$, and $W$ with respect to the LSR. The kinematical probability that a star belongs to a given population: thin disk, thick disk or halo (i = 1, 2, 3) is then given by $$\begin{aligned}
P_i \propto k_i \, f_i {\rm exp}\, (- \frac{U^2}{2 \sigma_{U_i}^2}
- \frac{(V - V_{{\rm ad}_i})^2}{2 \sigma_{V_i}^2} - \frac{W^2}{2 \sigma_{W_i}^2}),\end{aligned}$$ where $k = (2 \pi)^{-3/2} (\sigma_{U} \sigma_{V} \sigma_{W})^{-1}$ is the standard normalization constant, and $f$ the relative number of stars in a given population. $\sigma_{U}, \sigma_{V}$, and $\sigma_{W}$ are the velocity dispersions in $U, V$, and $W$, respectively, and $V_{\rm ad}$ the asymmetric drift velocity for the population. As an example, the values used by Reddy et al. (2006) are given in Table 1. Some of these values have considerable uncertainties. Depending on how the populations is defined, the fraction of thick disk stars in the solar neighbourhood may be as high as 20% (Fuhrmann 2004), and the local fraction of halo stars is often estimated to be on the order of 0.001.
[lrrrrr]{} Population & $\sigma_{U}$ & $\sigma_{V}$ & $\sigma_{W}$ & $V_{\rm ad}$ & $f$\
& km$s^{-1}$ & km$s^{-1}$ & km$s^{-1}$ & km$s^{-1}$ &\
Thin disk & 43 & 28 & 17 & $-9$ & 0.93\
Thick disk & 67 & 51 & 42 & $-48$ & 0.07\
Halo & 131 & 106 & 85 & $-220$ & 0.006\
In the works of Bensby et al. (2005) and Reddy et al. (2003, 2006) precise abundance ratios for F and G dwarfs have been derived for samples of stars kinematically selected to have high probability of belonging to either the thin or the thick disk. Thus, in Reddy et al. (2006) the probability limit for each population is $P > 70$%. The resulting \[$\alpha$/Fe\] – \[Fe/H\] diagram is shown in Fig. \[fig:reddy.alpha-Fe.ThinThick\]. As seen, there is a gap in \[$\alpha$/Fe\] between thin- and thick-disk stars for the metallicity range $-0.7 <$ \[Fe/H\] $-0.4$. Around \[Fe/H\]$= -0.3$ a few stars seem to connect the two populations like in the corresponding diagram of Fuhrmann (2004).
![\[$\alpha$/Fe\] as a function of \[Fe/H\] according to Reddy et al. (2003, 2006). Stars indicated by open circles have a probability $P > 70$% of belonging to the thin disk, whereas stars represented by filled circles have $P > 70$% of belonging to the thick disk.[]{data-label="fig:reddy.alpha-Fe.ThinThick"}](nissen_fig7.eps)
The \[$\alpha$/Fe\] diagram of Bensby et al. (2005) looks similar to Fig. \[fig:reddy.alpha-Fe.ThinThick\] except that their thick-disk stars continue all the way up to solar metallicity with decreasing values of \[$\alpha$/Fe\]. Thus, Bensby et al. claim that star formation in the thick disk continued long enough to include chemical enrichment from type Ia SNe. This is, however, not so evident from the data of Reddy et al. (2006). In Fig. \[fig:reddy.alpha-Fe.ThinThick\], the thick disk terminates around \[Fe/H\] $\simeq -0.3$ with little or no decrease of \[$\alpha$/Fe\].
Given that the thin- and thick-disk stars of Bensby et al. (2005) and Reddy et al. (2003, 2006) have been kinematically selected, the question arises if there are stars with intermediate kinematics filling the gap in \[$\alpha$/Fe\] between the two populations. The problem has be addressed by Ramírez et al. (2007), who derived oxygen abundances for 523 nearby stars from a non-LTE analysis of the O[i]{} triplet at 7774Å. A similar splitting as in Fig. \[fig:reddy.alpha-Fe.ThinThick\] between stars with thin- and thick-disk kinematics is obtained, and stars with intermediate kinematics do not fill the gap in \[O/Fe\]; they have either high \[O/Fe\] or low \[O/Fe\].
Another set of abundance data that can be used to study the problem of a gap in \[$\alpha$/Fe\] between thin- and thick-disk stars has been obtained by Neves et al. (2009) from ESO/HARPS high-resolution spectra of 451 F, G, and K main-sequence stars in the solar neighbourhood. The main purpose of this project is to detect planets around stars by measuring radial velocity variations with a precision of 1ms$^{-1}$. As a by-product, stellar abundance ratios relative to those of the Sun have been derived.
Neves et al. (2009) show that trends of abundance ratios as a function of \[Fe/H\] are the same for stars with and without planets. For both groups there is a bifurcation of the abundances of $\alpha$-capture elements relative to iron. This is shown in Fig. \[fig:neves.alpha-Fe\], where only stars having $T_{\rm eff}$ within $\pm 300$K from the effective temperature of the Sun have been included in order to obtain a very high precision of \[$\alpha$/Fe\], i.e. $\sigma$\[$\alpha$/Fe\] $\simeq \pm 0.02$. At metallicities \[Fe/H\] $< -0.1$, there is a gap in \[$\alpha$/Fe\] between ‘high-$\alpha$’ and ‘low-$\alpha$’ stars. Hence, the data of Neves et al. (2009) confirm the dichotomy in \[$\alpha$/Fe\] found by Bensby et al. (2005) and Reddy et al. (2006). Furthermore, the data of Neves et al. support the claim of Bensby et al. (2005), that the thick disk have metallicities stretching up to solar metallicity.
The sample of stars from Neves et al. (2009) is [*volume limited*]{}; hence, the gap between high-$\alpha$ and low-$\alpha$ stars cannot be due to exclusion of stars with intermediate kinematics. Figure \[fig:neves.alpha-Fe\] shows the distribution of the two populations in a Toomre energy diagram. As seen, the majority of high-$\alpha$ stars have a total space velocity with respect to the LSR larger than 85km s$^{-1}$, which classify them as thick-disk stars, but several high-$\alpha$ stars are kinematically mixed with the low-$\alpha$ stars. This is an example of how an abundance ratio is a more clean separator of stellar populations than kinematics.
![The \[$\alpha$/Fe\] vs. \[Fe/H\] distribution for a [*volume limited*]{} sample of main-sequence stars from Neves et al. (2009). Only stars with $T_{\rm eff}$ within $\pm 300$K from the effective temperature of the Sun have been included.[]{data-label="fig:neves.alpha-Fe"}](nissen_fig8.eps)
![The Toomre diagram for stars from Fig. \[fig:neves.alpha-Fe\] with \[Fe/H\] $< -0.1$. High-$\alpha$ stars are shown with filled circles and low-$\alpha$ stars with open circles. The two circles delineate constant total space velocities with respect to the LSR, $V_{\rm tot} = 85$ and 180km s$^{-1}$, respectively.[]{data-label="fig:neves.toomre"}](nissen_fig9.eps)
Considering that the high-$\alpha$, thick-disk stars tend to be older than the oldest of the low-$\alpha$, thin-disk stars (Fuhrmann 2004; Reddy et al. 2006), the difference in \[$\alpha$/Fe\] is often explained in a scenario, where a period of rapid star formation in the early Galactic disk was interrupted by a merging satellite galaxy that ‘heated’ the already formed stars to thick-disk kinematics. This was followed by a hiatus in star formation, in which metal-poor gas was accreted and type Ia SNe caused \[$\alpha$/Fe\] to decrease. When star formation resumed, the first thin-disk stars were formed with low metallicity and low \[$\alpha$/Fe\]. This scenario also explains the systematic differences in \[C/O\] (Fig. \[fig:CO-O.all\]) and \[Eu/Ba\] (Fig. \[fig:Eu-Ba\]) between thin- and thick-disk stars.
Haywood (2008) has pointed out that the low-metallicity thin-disk stars in the solar neighbourhood tend to have positive values of the $V$ velocity component. Such stars have mean galactocentric distances larger than the distance of the Sun from the Galactic center, and hence they are likely to have been formed in the outer part of the Galactic disk. Thick-disk stars, on the other hand, have negative values of $V$ and tend to be formed in the inner Galactic disk. From these considerations, Haywood suggests that the bimodal distribution in \[$\alpha$/Fe\] may be due to radial mixing of stars in the disk.
This scenario has been further investigated by Schönrich & Binney (2009a, 2009b) in a model for the chemical evolution of the Galactic disk, for which the star formation rate is monotonically decreasing, and which includes radial migration of stars and gas flows. The model successfully fit the metallicity distribution and the large scatter in the age-metallicity relation for stars in the Geneva-Copenhagen survey (Nordström et al. 2004). The model also predicts a bimodal distribution of \[$\alpha$/Fe\] for stars in the solar neighbourhood with the high-$\alpha$ stars coming from the inner parts of the Galactic disk and the low-$\alpha$ stars from the outer part. It will, however, be interesting to see if the model can reproduce a gap in the \[$\alpha$/Fe\] distribution for disk stars as found from the data of Neves et al. (2009).
Abundance gradients in the disk {#sect:gradients}
-------------------------------
In models for the chemical evolution of the Galactic disk, observed abundance gradients provide important constraints. Gradients may be determined from B-type stars and HII regions, but the most precise results have been obtained from Cepheids. These variable stars are bright enough to be studied at large distances and accurate values of the distances can be obtained from the period-luminosity relation. Earlier results suggested a steeper metallicity gradient in the inner part of the galaxy as compared to the outer part with a break in the gradient occurring around 10kpc. According to newer work this is, however, not so obvious. For 54 Cepheids ranging in galactocentric distance $R_G$ from 4 to about 14kpc, Luck et al. (2006) find an overall gradient $d$\[Fe/H\]/$dR_G = -0.06$dex kpc$^{-1}$, but there seems to be a significant cosmic scatter around a linear fit to the data. A region at Galactic longitude $l \sim 120$ degrees and a distance of 3-4kpc from the Sun has enhanced metallicities with $\Delta$\[Fe/H\]$\simeq +0.2$ in the mean. Such spatial inhomogeneities could be due to recent SNe events.
Yong et al. (2006) have made an interesting study of 24 Cepheids in the outer Galactic disk based on high-resolution spectra. The distance range is 12$< \, R_G \,< 18$kpc. Most of the Cepheids continue the trend with galactocentric distance exhibited by the Luck et al. (2006) sample, but a minority of six Cepheids have \[Fe/H\] around $-0.8$dex and enhanced $\alpha$-element abundances, \[$\alpha$/Fe\]$\sim +0.3$. Thus, there is some evidence for two populations of Cepheids in the outer disk.
Cepheids can only provide information about ‘present-day’ gradients in the disk. Abundances of stars in open clusters may, however, be used to determine gradients at different ages. In this connection one benefits from the quite accurate ages and distances that can be determined from colour-magnitude diagrams of open clusters. For a review of results from clusters, the reader is referred to the chapter on open clusters by Friel (this volume).
The Galactic bulge {#sect:bulge}
==================
Due to a large distance and a high degree of interstellar absorption and reddening, the Galactic bulge is the least well known component of the Milky Way. It has been much discussed if the bulge contains only very old stars or if it also includes a younger population. This problem is related to two different scenarios for the formation of the bulge, a ‘classical’ bulge formed rapidly by the coalescence of star-forming clumps as suggested from the simulations of Elmegreen et al. (2008) or a ‘pseudobulge’ formed over a longer time through dynamical instabilities in the Galactic disk (see review by Kormendy & Kennicutt 2004). The measurement of abundance ratios in bulge stars may help to decide between these scenarios by providing information on the time-scale for the formation of the bulge.
The metallicity distribution of bulge stars has been debated for a long time. With the determination of \[Fe/H\] for 800 bulge giants based on VLT multi-fiber spectra with a resolution of $R \sim 20\,000$ (Zoccali et al. 2008) a robust result seems to have been obtained. Zoccali et al. observed stars in four fields having distances from the Galactic center ranging from 600 to 1600pc. The overall metallicity distribution is centered on solar metallicity and extends from \[Fe/H\]$\simeq -1.5$ to +0.5, but with few stars in the range $-1.5 <$ \[Fe/H\] $< -1.0$. A decrease of the mean metallicity along the bulge minor axis is suggested corresponding to a gradient of $\sim \! 0.6$dex per kpc.
A pioneering study of $\alpha$/Fe ratios in 12 bulge red giants was carried out by McWilliam & Rich (1994) based on 4-meter telescope optical echelle spectra with a resolution of $R \sim 20\,000$ and typical $S/N \sim 50$. Enhanced values of \[Mg/Fe\] and \[Ti/Fe\] ($\sim \! +0.3$dex) were found for all stars up to a metallicity of $\sim \! +0.4$dex suggesting a very rapid formation of the bulge. In contrast, \[Si/Fe\] and \[Ca/Fe\] showed solar values, which is difficult to understand. These data are now superseeded by higher resolution spectra obtained with 8-meter class telescopes both in the optical and the infrared spectral region.
Lecureur et al. (2007) used VLT/UVES spectra in the spectral region 4800 - 6800Å to derive O and Mg abundances relative to Fe for about 50 red bulge giants. They find that the \[O/Fe\] and \[Mg/Fe\] trends are more enhanced than the corresponding trends for thick-disk stars as determined by Bensby et al. (2005) for dwarf stars. The same conclusion is reached by Fulbright et al. (2007) from abundance ratios derived for 27 red giants observed at high resolution in the 5000 - 8000Å spectral region with the Keck/HIRES spectrograph.
![The \[$\alpha$/Fe\] - \[Fe/H\] relation for bulge K giant stars shown with filled (red) squares in comparison with K giants in the solar neighbourhood having either thick-disk kinematics (filled circles) or thin-disk kinematics (open circles). Data for \[Mg/Fe\], \[Si/Fe\], \[Ca/Fe\], and \[Ti/Fe\] are adopted from Alves-Brito et al. (2010).[]{data-label="fig:bulge.alpha-fe"}](nissen_fig10.eps)
In contrast to these results, Meléndez et al. (2008) find the same \[O/Fe\] - \[Fe/H\] trend for 19 red giants belonging to the bulge and 21 thick-disk giants in the solar neighbourhood. For both samples, the oxygen and iron abundances are based on spectrum synthesis of OH and Fe[i]{} lines in an infrared spectral region around 1.55$\mu$m as observed with Gemini/Phoenix. Hence, the comparison of the bulge and the thick disk is done in a differential way for the same type of stars. Meléndez et al. suggest that the comparisons of bulge giants with thick-disk dwarfs has led to a spurious offset of the two \[O/Fe\] - \[Fe/H\] trends due to systematic errors in the abundance ratios. When LTE is assumed and classical 1D model atmospheres are adopted, systematic errors could arise due to different non-LTE and 3D corrections for dwarf and giant stars. Recently, Ryde et al. (2010) have confirmed the results of Meléndez et al. (2008) from VLT/CRIRES high-resolution spectra of 11 red giants in the bulge.
Further evidence for an agreement between bulge and local thick-disk stars has been obtained by Alves-Brito et al. (2010) from optical high-resolution spectra. Their results are shown if Fig. \[fig:bulge.alpha-fe\]. As seen the trends for bulge and thick-disk stars agree well with no significant differences below solar metallicity. The thin-disk stars, on the other hand, fall below the bulge and the thick disk stars. Above solar metallicity, the bulge stars tend to have higher \[$\alpha$/Fe\] values than the disk stars but this needs to be confirmed for a larger sample.
Independent and very interesting data for abundances in the Galactic bulge have been obtained from spectra of microlensed main-sequence and subgiant stars. During a microlensing event, the flux from a star can be enhanced by a factor of 100 or even more; hence, the magnitude of a typical bulge turnoff star raises from $V \sim 18$ to $V \sim 13$. This makes it possible to obtain high-resolution spectra with good $S/N$. Bensby et al. (2011) have presented a homogeneous abundance study of 26 such microlensed stars using the same methods as applied for local thick-disk dwarf stars (Bensby et al. 2005). As seen from Fig. \[fig:all.alpha-fe\], there is a tendency that these microlensed bulge stars are separated into two regions in the \[$\alpha$/Fe\] – \[Fe/H\] plane: A metal-poor group with \[Fe/H\] $< -0.3$ and \[$\alpha$/Fe\] $\simeq +0.3$ like the thick-disk stars, and a metal-rich group with +0.1 $<$ \[Fe/H\] $<$ +0.6 for which the majority of stars have \[$\alpha$/Fe\] $\simeq 0$ like the thin-disk stars, but four stars have \[$\alpha$/Fe\] $\simeq +0.1$.
Another interesting aspect of the study of microlensed main-sequence and subgiant stars is the possibility to derive ages by comparing their position in a log$g$ - log$T_{\rm eff}$ diagram with isochrones computed from stellar models. According to Bensby et al. (2011) the metal-poor bulge stars have an average age of 11.2Gyr with a dispersion of $\pm 2.9$Gyr much of which may be due errors in the age determination. The metal-rich bulge stars are on average younger (7.6Gyr) and has a larger age dispersion ($\pm 3.9$Gyr).
On the basis of these new data, Bensby et al. (2011) suggest that the bulge consists of two stellar populations, i.e. a metal-poor population similar to the thick disk in terms of metallicity range, \[$\alpha$/Fe\], and age, and a metal-rich population, which could be related to the inner thin disk. Supporting evidence has recently been obtained by Hill et al. (2011) from a study of 219 red giants in Baade’s bulge window (situated about four degrees from the Galactic center). \[Fe/H\] and \[Mg/Fe\] are derived from ESO VLT FLAMES spectra with a resoltion of $R =$ 20000. The distribution of \[Fe/H\] is asymmetric and can be deconvolved into two Gaussian components: a metal-poor population centred at \[Fe/H\] = $-0.30$ having a dispersion of 0.25dex in \[Fe/H\], and a metal-rich population centred at \[Fe/H\] = $+0.32$ with a dispersion of 0.11dex only. The metal-poor population has \[Mg/Fe\] $\simeq 0.3$, whereas the stars in the metal-rich population distribute around the solar Mg/Fe ratio. These data agree well with those of Bensby et al. (2011). However, a larger set of data for microlensed bulge stars and information about chemical abundances of inner disk stars are needed before one can draw any robust conclusions about the existence of two bulge populations and the consequences this may have for models of the formation of the bulge.
The Galactic halo {#sect:halo}
=================
Evidence of two distinct halo populations {#sect:twohalopop}
-----------------------------------------
For a long time, it has been discussed if the Galactic halo consists of more than one population. The classical monolithic collapse model of Eggen, Lynden-Bell & Sandage (1962) corresponds to a single halo population, but from a study of globular clusters, Searle and Zinn (1978) suggested that the halo comprises two populations: $i)$ an inner, old, flattened population with a slight prograde rotation formed during a dissipative collapse, and $ii)$ an outer, younger, spherical population accreted from satellite systems. This dichotomy of the Galactic halo has found support in a study of $\sim\,$20000 stars from the Sloan Digital Sky Survey (SDSS) by Carollo et al. (2007). They find that the inner halo consists of stars with a peak metallicity at \[Fe/H\]$\simeq -1.6$, whereas the outer halo stars distribute around \[Fe/H\]$\simeq -2.2$.
Several studies suggest that there is a difference in \[$\alpha$/Fe\] between stars that can be associated with the inner and the outer halo, respectively. Fulbright (2002) shows that stars with large values of the total space velocity relative to the LSR, $V_{\rm tot} > 300$kms$^{-1}$, tend to have lower values of \[$\alpha$/Fe\] than stars with $150 < V_{\rm tot} < 300$kms$^{-1}$, and Stephens & Boesgaard (2002) find that \[$\alpha$/Fe\] is correlated with the apo-galactic orbital distance in the sense that the outermost stars have the lowest values of \[$\alpha$/Fe\]. Further support for such differences in \[$\alpha$/Fe\] comes from a study by Gratton et al. (2003), who divided stars in the solar neighbourhood into two populations according to their kinematics: $i)$ a ‘dissipative’ component, which includes thick-disk stars and prograde rotating halo stars, and $ii)$ an ‘accretion’ component that consists of retrograde rotating halo stars. The ‘accretion’ component has smaller values and a larger scatter for \[$\alpha$/Fe\] than the ‘dissipative’ component.
The differences in \[$\alpha$/Fe\] found in these works are not larger than about 0.1dex, and it is unclear, if the distribution of \[$\alpha$/Fe\] is continuous or bimodal. Nissen & Schuster (1997) found a more clear dichotomy in \[$\alpha$/Fe\] for 13 halo stars with $-1.3 <$ \[Fe/H\] $< -0.5$. Eight of these halo stars have \[$\alpha$/Fe\] in the range 0.1 to 0.2dex, whereas \[$\alpha$/Fe\] $\simeq 0.3$ for the other five halo stars. Interestingly, the low-$\alpha$ halo stars tend to have larger apo-galactic distances than the high-$\alpha$ stars.
A more extensive study of ‘metal-rich’ halo stars has been carried out by Nissen & Schuster (2010). Stars are selected from the Schuster et al. (2006) $uvby$-$\beta$ catalogue of high-velocity and metal-poor stars. To ensure that a star has a high probability of belonging to the halo population, the total space velocity with respect to the LSR is required to be larger than 180kms$^{-1}$. Furthermore, Strömgren photometric indices are used to select dwarfs and subgiants with $5200 < T_{\rm eff} < 6300$K and \[Fe/H\] $> -1.6$. High-resolution, high $S/N$ spectra were obtained with the ESO VLT/UVES and the Nordic Optical Telescope FIES spectrographs for 78 of these stars. The large majority of the stars are brighter than $V = 11.1$ and situated within a distance of 250pc. These spectra are used to derive high-precision LTE abundance ratios in a differential analysis that also includes 16 stars with thick-disk kinematics. The precision of the various abundance ratios range from 0.02 to 0.04dex.
![\[Mg/Fe\] and \[$\alpha$/Fe\] versus \[Fe/H\] based on data from Nissen & Schuster (2010). Stars with thick-disk kinematics are indicated by crosses and halo stars by circles. On the basis of the \[Mg/Fe\] distribution, the halo stars are divided into low-$\alpha$ stars shown by filled (red) circles, and high-$\alpha$ stars shown by open (blue) circles.[]{data-label="fig:nissen.mg.alpha-fe"}](nissen_fig11.eps)
Figure \[fig:nissen.mg.alpha-fe\] (top) shows the distribution of \[Mg/Fe\] as a function of \[Fe/H\] for stars in Nissen & Schuster (2010). As seen, the halo stars split into two distinct populations: ‘high-$\alpha$’ stars with a nearly constant \[Mg/Fe\] and ‘low-$\alpha$’ stars with declining values of \[Mg/Fe\] as a function of increasing metallicity. A classification into these two populations has been done on the basis of \[Mg/Fe\], but as seen from the bottom part of the figure, \[$\alpha$/Fe\] would have led to the same division of the halo stars except at the lowest metallicities, $-1.6 <$ \[Fe/H\] $< -1.4$, where the two populations tend to merge, and the classification is less clear.
As seen from Fig. \[fig:nissen.mg.alpha-fe\], the separation in \[Mg/Fe\] for the two halo populations is significantly larger than the separation in \[$\alpha$/Fe\]. At \[Fe/H\]$\simeq -0.8$ the mean difference in \[Mg/Fe\] is about 0.25dex, whereas it is only about 0.15dex in \[$\alpha$/Fe\]. This is probably caused by different degrees of SNe Ia contribution to the production of Mg, Si, Ca, and Ti. According to Tsujimoto et al. (1995, Table 3), the relative contribution of SNe Ia to the solar composition is negligible for Mg, 17% for Si and 25% for Ca (Ti was not included). For comparison, the SNe Ia contribution is 57% for Fe. Hence, \[Mg/Fe\] is a more sensitive measure of the ratio between type II and type Ia contributions to the chemical enrichment of matter than \[$\alpha$/Fe\]. On the other hand, it is possible to measure \[$\alpha$/Fe\] with a higher precision than \[Mg/Fe\], because many more spectral lines can be used to determine \[$\alpha$/Fe\].
![The relation between \[Ni/Fe\] and \[Na/Fe\]. Local main-sequence stars from Nissen & Schuster (2010) are indicated by the same symbols as in Fig. \[fig:nissen.mg.alpha-fe\], whereas the (green) triangles refer to K giants in dSph satellite galaxies from Venn et al. (2004). For both sets of data, the stars are confined to the metallicity range $-1.6 <$ \[Fe/H\] $< -0.4$.[]{data-label="fig:ni-na.all"}](nissen_fig12.eps)
The low-$\alpha$ halo stars also have low values of \[Na/Fe\] and \[Ni/Fe\] relative to the high-$\alpha$ stars, and as shown in Fig. \[fig:ni-na.all\], the two populations are well separated in a \[Ni/Fe\] - \[Na/Fe\] diagram. As seen, some stars in dSph galaxies are even more extreme in \[Na/Fe\] and \[Ni/Fe\] than the low-$\alpha$ halo stars.
Kinematics and origin of the two halo populations {#sect:kinorig}
-------------------------------------------------
The distribution of \[$\alpha$/Fe\] in Fig. \[fig:nissen.mg.alpha-fe\] can be explained, if the high-$\alpha$ stars have been formed in regions with such a high rate of chemical evolution that only type II SNe have contributed to the chemical enrichment up to \[Fe/H\]$\sim -0.4$. The low-$\alpha$ stars, on the other hand, originate from regions with a relatively slow chemical evolution so that type Ia SNe have started to contribute iron around \[Fe/H\]=$-1.6$ causing \[$\alpha$/Fe\] to decrease towards higher metallicities until \[Fe/H\]$\sim -0.8$.
![Toomre diagram for stars from Nissen & Schuster (2010) having \[Fe/H\]$> -1.4$. The same symbols for high-$\alpha$ halo, low-$\alpha$ halo and thick-disk stars as in Fig. \[fig:nissen.mg.alpha-fe\] are used. The short-dashed circle corresponds to $V_{\rm tot} = 180$km s$^{-1}$. The long-dashed line indicates zero rotation in the Galaxy and therefore separates retrograde moving stars from prograde moving.[]{data-label="fig:toomre.review"}](nissen_fig13.eps)
Further insight into the origin of the two halo populations can be obtained from kinematics. As seen from the Toomre energy diagram in Fig. \[fig:toomre.review\], the high-$\alpha$ stars show evidence for being more bound to the Galaxy and favoring prograde Galactic orbits, while the low-$\alpha$ stars are less bound with two-thirds of them being on retrograde orbits. This suggests that the high-$\alpha$ population is connected to a dissipative component of the Galaxy, while the low-$\alpha$ stars have been accreted from dwarf galaxies.
Several retrograde moving low-$\alpha$ stars have a Galactic $V$-velocity component similar to that of the $\omega$Cen globular cluster, i.e. $V \sim -260$km s$^{-1}$. As often discussed (e.g. Bekki & Freeman 2003), $\omega$Cen is probably the nucleus of a captured satellite galaxy with its own chemical enrichment history. Meza et al. (2005) have simulated the orbital characteristics of the tidal debris of such a satellite dragged into the Galactic plane by dynamical friction. The captured stars are predicted to have rather small $W$-velocities but a wide, double-peaked $U$-distribution. As shown in Fig. \[fig:W-U.review\], the $W$-$U$ distribution observed for the low-$\alpha$ halo corresponds quite well to that prediction. There are two groups of low-$\alpha$ stars with $U > 200$km s$^{-1}$ and $U < -200$km s$^{-1}$, respectively, corresponding to stars moving in and out of the solar neighbourhood on elongated radial orbits. Thus, a good fraction of the low-$\alpha$ stars, although not all, may well have originated in the $\omega$Cen progenitor galaxy. The high-$\alpha$ stars, on the other hand, are confined to a much smaller range in $U$, i.e. from about $-200$ to about +200km s$^{-1}$.
![The relation between the $U$ and $W$ velocity components for stars from Nissen & Schuster (2010) having \[Fe/H\]$> -1.4$.[]{data-label="fig:W-U.review"}](nissen_fig14.eps)
The \[$\alpha$/Fe\] vs. \[Fe/H\] trend for the low-$\alpha$ stars in Fig. \[fig:nissen.mg.alpha-fe\] and the Ni-Na trend in Fig. \[fig:ni-na.all\] resemble the corresponding trends for stars in dwarf galaxies. Stars in these systems tend, however, to have lower values of \[$\alpha$/Fe\], \[Na/Fe\] and \[Ni/Fe\] than low-$\alpha$ halo stars. This offset agrees with simulations of the chemical evolution of a hierarchically formed stellar halo in a $\Lambda$CDM Universe by Font et al. (2006, Fig. 9). The bulk of halo stars originate from early accreted, massive dwarf galaxies with efficient star formation, whereas surviving satellite galaxies in the outer halo on average have smaller masses and a slower chemical evolution with a larger contribution from Type Ia SNe at a given metallicity. The \[Mg/Fe\] vs. \[Fe/H\] trend for field stars, predicted by Font et al., agrees in fact remarkably well with the trend for the low-$\alpha$ halo stars.
The simulations of Font et al (2006) do not explain the existence of high-$\alpha$ halo stars. Two recent $\Lambda$CDM simulations suggest, however, a dual origin of stars in the inner Galactic halo. Purcell et al. (2010) propose that ancient stars formed in the Galactic disk can be ejected to the halo by merging satellite galaxies, and Zolotov et al. (2009, 2010) find that stars formed out of accreted gas in the inner 1kpc of the Galaxy can be displaced into the halo through a succession of mergers. Alternatively, the high-$\alpha$ population may simply belong to the high-velocity tail of a thick disk with a non-Gaussian velocity distribution.
Globular clusters and dwarf galaxies {#sect:globclusters}
------------------------------------
The Galactic halo contains more than 100 globular clusters and a number of dwarf spheroidal galaxies that are considered as Milky Way satellite systems. A dSph galaxy has a broad range in age and metallicity, whereas a globular cluster is a smaller system with a more limited range in these parameters . As often discussed, it is possible that some or all field halo stars come from dissolved globular clusters or have been accreted from previous generations of satellite galaxies. In this connection, it is is of great interest to compare the chemical abundance ratios in field stars with the corresponding ratios in still existing globular clusters and dwarf galaxies.
Globular cluster were for a long time considered as examples of systems containing a single population of stars with a well defined age and chemical composition. During the last decades it has, however, become more and more clear that many, if not all, globular clusters contain multiple stellar populations. Strong evidence comes from abundance ratios between elements from oxygen to aluminum. As reviewed by Gratton et al. (2004), high-resolution spectroscopy of red giants has revealed anti-correlations between \[Na/Fe\] and \[O/Fe\] and between \[Al/Fe\] and \[Mg/Fe\] in intermediate metallicity globular clusters. An example is shown in Fig. \[fig:NGC6752\] for K giants in NGC6752 (Yong et al. 2005).
![The Na-O and Al-Mg anti-correlations in the globular cluster NGC6752 based on data from Yong et al. (2005). Filled circles indicate stars near the bright end of the red-giant branch; open circles refer to less luminous stars around the red-giant bump.[]{data-label="fig:NGC6752"}](nissen_fig15.eps)
The extensive study of Carretta et al. (2009) based on high-resolution VLT/UVES spectra for giant stars in 19 globular clusters indicates the existence of a \[Na/Fe\] - \[O/Fe\] anti-correlation in all cases, but the amplitude of the variations is different from cluster to cluster. Variations in \[Al/Fe\] correlated with \[Na/Fe\] are seen in the majority of clusters and an anti-correlation between \[Al/Fe\] and \[Mg/Fe\] is detected in a few cases. In addition, Yong et al. (2003) have revealed a correlation between the heavy magnesium isotope $^{26}$Mg and Al in NGC6752 by determining Mg isotope ratios from the profiles of MgH lines near 5140Å.
All of these abundance variations are ascribed to hydrogen burning via the CNO-cycle and the NeNa and MgAl chains. Some years ago, it was much discussed if the variations were due to nuclear processes and non-standard mixing in the stars themselves or to an early generation of stars that have polluted the gas out of which the present low-mass stars in the clusters have formed. The fact that the abundance variations are not correlated with the luminosity of the giant stars speaks against the first possibility, and when Gratton et al. (2001) discovered that a Na-O anti-correlation is present in turnoff and subgiant stars in NGC6397 and NGC6752, the internal mixing case was ruled out. Hence, the abundance variations must be due to an early generation of stars. Candidates are intermediate-mass AGB stars undergoing hot bottom burning (Ventura et al. 2001) and massive rotating stars (Decressin et al. 2007).
In addition to the abundance variations of elements from O to Al, there is also increasing evidence for variations of Ca, Fe and $s$-process elements in several globular clusters. The classical example is $\omega$Cen for which giant stars are found to have metallicities from \[Fe/H\]$\simeq -1.9$ to $\simeq -0.5$ (see review by Gratton et al. 2004). Multiple sequences in colour-magnitude diagrams are present suggesting the existence of four or five discrete populations in $\omega$Cen with significant age and abundance differences. The more metal-rich stars in $\omega$Cen have unusually low values of \[Cu/Fe\] (Cunha et al. 2002) and very high values of the abundances of the second-peak $s$-process elements, i.e. \[Ba/Fe\] and \[La/Fe\] values around 1.0dex (Smith et al. 2000). These abundance anomalies point to a complicated chemical evolution history, in which $\omega$Cen was originally the nucleus of a much larger dwarf galaxy that merged with the early Galactic disk (Bekki & Freeman 2003).
Evidence for abundance variations of Ca, Fe, and $s$-process elements has been found for other clusters than $\omega$Cen. On the basis of high-resolution spectroscopy with UVES of eight giant stars in NGC1851, Yong & Grundahl (2008) found a range of 0.6dex in \[Zr/Fe\] and \[La/Fe\], and Marino et al. (2009) found similar large variations in \[Y/Fe\], \[Zr/Fe\] and \[Ba/Fe\] for 17 giant stars in M22. In addition, there is a correlation between these ratios and \[Fe/H\], which varies from $-1.9$ to $-1.5$ in M22. Evidence of variations in \[Ca/Fe\] is also found in other globular clusters by Lee et al. (2009) from photometric measurements of an index of the Ca[ii]{} H and K lines, but Carretta et al. (2010) do not confirm this on the basis of high-resolution UVES spectra. Instead, they suggest that the spread in the Ca[ii]{} H and K index may be related to variations in helium and nitrogen abundances.
Some authors, e.g. Lee et al. (2009), suggest that all globular clusters were once nuclei of dwarf galaxies that are now accreted and dissolved in the Milky Way. In this connection, one may wonder why practically no field halo stars share the Na-O abundance anomalies of the globular clusters. Perhaps the explanation is that the elements produced by AGB stars are confined to the potential wells of the clusters. According to the hydrodynamical simulations by D’Ercole et al. (2008), the gas ejected by AGB stars collects in cooling flows into the cores of globular clusters.
![\[$\alpha$/Fe\] vs. \[Fe/H\] for various stellar populations. Thin-disk stars from Reddy et al. (2003) are shown with plus symbols. Filled circles refer to thick-disk stars from Reddy et al. (2006) and Nissen & Schuster (2010). Filled (red) squares are microlensed bulge stars from Bensby et al. (2011). Open (blue) circles are high-$\alpha$ and filled (red) circles low-$\alpha$ halo stars from Nissen & Schuster (2010). Asterisks refer to stars in the Sagittarius dSph galaxy (Sbordone et al. 2007), and filled (green) triangles show data for stars in the Sculptor dSph galaxy from Kirby et al. (2009) for which the precision of \[$\alpha$/Fe\] is better than 0.15dex.[]{data-label="fig:all.alpha-fe"}](nissen_fig16.update.eps)
The chemical composition of giant stars in dSph galaxies is reviewed by Tolstoy et al. (2009). For the large majority of stars with \[Fe/H\]$> -2$, \[$\alpha$/Fe\] is significantly lower than in halo and disk stars belonging to the Milky Way. Two examples are shown in Fig. \[fig:all.alpha-fe\]. Stars in Sculptor (Kirby et al. 2009) distribute around \[$\alpha$/Fe\] = $-0.1$, and do not seem to have metallicities above \[Fe/H\]$\sim -0.9$. The other example is the Sagittarius dwarf galaxy, which was discovered by Ibata et al. (1994) to be merging with the Milky Way. The abundances derived by Sbordone et al. (2007) for giant stars show a declining trend of \[$\alpha$/Fe\] reaching \[$\alpha$/Fe\] as low as $-0.3$dex at solar metallicity.
Evidently, present day dSph galaxies have had a chemical evolution history different from that of the other main components of the Milky Way. In addition to the underabundance of \[$\alpha$/Fe\], the more metal-rich stars in dSph galaxies stand out by having low \[Na/Fe\] and \[Ni/Fe\] abundances (Fig. 11) as well as high \[Ba/Y\] ratios (Venn et al. 2004). At the lowest metallicities, \[Fe/H\]$< -2$, the abundance ratios are, however, similar to those in Galactic halo stars, which suggests that the abundance deviations of dSph stars should not be explained in terms of an anomalous IMF. The most obvious reason for the underabundances of \[$\alpha$/Fe\] is that the star formation rate in surviving dSph galaxies has been so slow that type Ia SNe started to contribute Fe at a metallicity of \[Fe/H\]$\sim -2$. Dwarf galaxies are, however, not ruled out as ‘building blocks’ of the Galactic halo, because early accreted dSph galaxies probably have had a somewhat faster chemical evolution than the less masive present-day dSph galaxies in the outer halo (Font et al. 2006).
Conclusions {#sect:conclusions}
===========
The last decade has seen great progress in determinations of abundance ratios that can be used as tracers of stellar populations. This has been possible due to the availability of high-resolution echelle spectrographs at large telescopes – in some cases with multiplex capabilities so that many stars, e.g. in clusters and dwarf galaxies, can be observed simultaneously. Determinations of \[$\alpha$/Fe\] from low- or medium-resolution spectroscopy, e.g. in the SDSS and RAVE surveys, are also important in connection with large statistical investigations of stellar populations. Similar measurements of \[$\alpha$/Fe\] will be obtained in connection with the GAIA mission. Furthermore, abundances of Galactic bulge stars are now been obtained from high-resolution infrared spectra, but the spectral coverage is small and there is no multiplex advantages. Hence, the number of bulge stars studied with infrared spectra is still limited. More efficient infrared echelle spectrographs with multiplex capabilities are needed to take advantage of the lower reddening in the infrared spectral region, when dealing with the bulge and the inner part of the disk.
Elemental abundance ratios are derived by the aid of a model atmosphere analysis of the available stellar spectra as discussed in Sect. \[sect:abundet\]. Most studies are still based on homogeneous model atmospheres and the assumption of LTE, but several works show that inhomogeneous models and deviations from LTE can change the derived abundances significantly. Such 3D and non-LTE modelling is difficult to carry out and in some cases the derived abundances are sensitive to uncertain hydrogen collision cross sections. Hence, the derived trends of abundance ratios as a function of metallicity can be quite uncertain. On the other hand, it is possible to determine precise differences of abundance ratios by a 1D LTE analysis of a sample of stars confined to small ranges in effective temperature, surface gravity and metallicity. In this way, abundance ratios can be used to disentangle stellar populations. Care should, however, be taken when comparing different spectral types and luminosity classes, such as F and G dwarfs with K giants, because systematic errors may be important.
As discussed in Sect. \[sect:tracers\], the abundance ratios \[$\alpha$/Fe\], \[C/O\], \[Na/Fe\], \[Ni/Fe\], \[Cu/Fe\], \[Eu/Ba\], and \[Ba/Y\] are of high interest as tracers of stellar populations. The usefulness of these ratios is related to the fact that the elements involved are produced in different types of stars. The nucleosynthesis is not well understood for all elements, but it seems that variations of the ratios from one population to the next can be explained in terms of different rates of star formation and chemical evolution. In general, one does not need to invoke variations in the initial mass function to explain the abundance ratios. A better understanding of the nucleosynthesis of the elements is, however, important in order to learn more about the formation and evolution of the various populations.
The power of abundance ratios as population tracers is evident from Fig. \[fig:all.alpha-fe\], where \[$\alpha$/Fe\] is plotted as a function of \[Fe/H\] for seven different populations. \[$\alpha$/Fe\] is defined as the average value of \[Mg/Fe\], \[Si/Fe\], \[Ca/Fe\] and \[Ti/Fe\], and may be used to estimate the time-scale for the chemical enrichment of the population, as explained in Sect. \[sect:intermediate\]. On the basis of this figure and the more detailed discussion in the previous sections, the following conclusions about stellar populations in the Galactic disk, the bulge and the halo can be made.
The disk consists of two main populations, the thin and the thick disk, which differ in \[$\alpha$/Fe\] as well as \[C/O\] and \[Eu/Ba\]. The thick disk is formed on a relatively short time-scale with enrichment from type II SNe only up to \[Fe/H\]$\simeq -0.4$. Thin-disk stars have lower values of \[$\alpha$/Fe\] due to type Ia SNe contributions to the chemical enrichment; hence the time-scale of evolution is longer than in the case of the thick disk. In the metallicity range $-0.7 <$ \[Fe/H\] $< -0.4$, a gap in \[$\alpha$/Fe\] is present between the two disks. Altogether, the \[$\alpha$/Fe\] distribution of disk stars is well explained by a scenario, for which a period of rapid star formation in the early Galactic disk was interrupted by a merging satellite galaxy that ‘heated’ the already formed stars to thick-disk kinematics. This was followed by a hiatus in star formation, in which metal-poor gas was accreted and type Ia SNe caused \[$\alpha$/Fe\] to decrease. When star formation resumed, the first thin disk stars formed with low metallicity and low \[$\alpha$/Fe\]. However, as discussed in Sect. \[sect:alphadistr\], an alternative model with a monotonically decreasing star formation rate and radial migration of stars and gas (Schönrich & Binney 2009a, 2009b) also predicts a bimodal distribution of \[$\alpha$/Fe\]. Very precise measurements of the distributions of \[$\alpha$/Fe\] as well as \[C/O\] and \[Eu/Ba\] for a large volume-limited sample of F and G main-sequence stars would be important to distinguish between the two competing models for disk formation.
For the bulge, there has been great progress in studies of abundance ratios in recent years. In some works, enhanced values of \[$\alpha$/Fe\]$\sim +0.3$ are found for stars as metal-rich as the Sun, but in the most precise differential studies (Alves-Brito et al. 2010), the trend of \[$\alpha$/Fe\] for the bulge is found to be similar to that of the thick disk (Fig. \[fig:bulge.alpha-fe\]). The abundances derived by Bensby et al. (2011) for microlensed main-sequence stars (Fig. \[fig:all.alpha-fe\]) suggest that the bulge may consist of two distinct populations: i.e. old metal-poor stars with enhanced \[$\alpha$/Fe\] ratios related to the thick disk, and younger very metal-rich stars with disk-like $\alpha$/Fe ratios. Still, one would like to see many more bulge stars observed before drawing conclusions concerning models for bulge formation from the abundance ratios.
For halo stars in the solar neighbourhood, there is evidence for the existence of two distinct populations clearly separated in \[$\alpha$/Fe\], \[Na/Fe\] and \[Ni/Fe\] as discussed in Sect. \[sect:twohalopop\]. The high-$\alpha$ stars have abundance ratios very similar to thick-disk stars. They may be ancient stars formed in the Galactic disk or bulge and ejected to the halo by merging satellite galaxies (Purcell et al. 2010; Zolotov et al. 2009, 2010), or they may simply belong to the high-velocity tail of a thick disk with a non-Gaussian velocity distribution. The low-$\alpha$ stars tend to have retrograde motions and many of them move on elongated radial orbits close to the Galactic plane as predicted for the stellar debris of the captured $\omega$Cen progenitor galaxy. It is likely that the low-alpha stars have been accreted from dSph galaxies with a relatively slow chemical evolution, for which type Ia SNe started to contribute iron at \[Fe/H\]$\sim -1.5$. Perhaps future precise studies of abundance ratios of larger samples of halo stars will reveal additional subpopulations.
Stars in surviving dSph galaxies have even lower values of \[$\alpha$/Fe\] than the low-$\alpha$ halo stars as seen from Fig. \[fig:all.alpha-fe\]. This is to be expected according to simulations of the chemical evolution of a hierarchically formed stellar halo in a $\Lambda$CDM Universe (Font et al. 2006). Present day dSph galaxies in the outer halo have experienced a slower chemical evolution than more massive satellite galaxies accreted in the early Galxy.
All globular clusters seem to consist of multiple stellar populations characterized by different values of \[O/Fe\], \[Na/Fe\], \[Al/Fe\], and \[Mg/Fe\]. This may be due to chemical enrichment from intermediate-mass AGB stars undergoing hot-bottom hydrogen burning. There is also evidence for variations of \[Fe/H\] and the abundance of $s$-process elements in several globular clusters, most notable in $\omega$Cen. In this connection, it has been suggested that globular clusters were once the nuclei of now dissolved dwarf galaxies.
Precise abundance ratios of field stars belonging to the Galactic halo have so far only been obtained in a small region around the Sun. It would be important to extend such studies to more distant halo regions. To do this in an efficient way one needs a fiber-coupled high-resolution spectrograph that would make it possible to observe many stars simultaneously over a relatively large field, say one degree in diameter. Such a spectrograph would also be very useful in exploring abundance gradients in the Galactic disk, both radially and in the direction towards the Galactic poles.
[99.]{}
Alves-Brito, A., Meléndez, J., Asplund, M., Ramírez, I., & Yong, D. 2010, A&A, 513, A35
Arlandini, C., K[ä]{}ppeler, F., Wisshak, K., et al. 1999, ApJ, 525, 886
Arnett, W. D. 1971, ApJ, 166, 153
Asplund, M. 2005, ARA&A, 43, 481
Asplund, M., Grevesse, N., Sauval, A.J., & Scott, P. 2009, ARA&A, 47, 481
Barklem, P. S., Christlieb, N., Beers, T. C., et al. 2005, A&A, 439, 129
Bekki, K., & Freeman, K. C. 2003, MNRAS, 346, L11
Bensby, T., & Feltzing, S. 2006, MNRAS, 367, 1181
Bensby, T., Ad[é]{}n, D., Mel[é]{}dez, J., et al. 2011, A&A, 533, A134
Bensby, T., Feltzing, S., Lundstr[ö]{}m, I., & Ilyin, I. 2005, A&A, 433, 185
Boeche, C., Siebert, A., & Steinmetz, M. 2008, AIP Conference Ser., 1082, 61
Bergemann, M., & Cescutti, G. 2010, A&A, 522, A9
Bergemann, M., & Gehren, T. 2008, A&A, 492, 823
Bergemann, M., Pickering, J. C., & Gehren, T. 2010, MNRAS, 401, 1334
Busso, M., Gallino, R., & Wasserburg, G. J. 1999, ARA&A, 37, 239
Carollo, D., Beers, T. C., Lee, Y. S., et al. 2007, Nature, 450, 1020
Carretta, E., Bragaglia, A., Gratton, R. G., & Lucatello, S. 2009, A&A, 505, 139
Carretta, E., Bragaglia, A., Gratton, R., et al. 2010, ApJ, 712, L21
Casagrande, L., Ramírez, I., Mel[é]{}ndez, J., Bessell, M., & Asplund, M. 2010, A&A, 512, A54
Cayrel, R., Depagne, E., Spite, M., et al. 2004, A&A, 416, 1117
Chen, Y. Q., Nissen, P. E., Zhao, G., & Asplund, A. 2002, A&A, 390, 225
Chen, Y. Q., Nissen, P. E., Zhao, G., Zhang, H. W., & Benoni, T. 2000, A&AS, 141, 491
Chiappini, C., Hirschi, R., Meynet, G., et al. 2006, A&A, 449, L27
Cooke, R., Pettini, M., Steidel, C.C., Rudie, G.C., & Nissen, P.E. 2011, MNRAS (in press), arXiv:1106.2805
Cunha, K., Smith, V. V., Suntzeff, N. B., et al. 2002, AJ, 124, 379
Decressin, T., Meynet, G., Charbonnel, C., Prantzos, N., & Ekstr[ö]{}m, S. 2007, A&A, 464, 1029
D’Ercole, A., Vesperini, E., D’Antona, F., et al. 2008, MNRAS, 391, 825
Drawin, H. W. 1969, Z. Phys., 225, 483
Edvardsson, B., Andersen, J., Gustafsson, B., et al. 1993, A&A, 275, 101
Eggen, O. J., Lynden-Bell, D., & Sandage, A. R. 1962, ApJ, 136, 748
Elmegreen, B. G., Bournaud, F., & Elmegreen, D. M. 2008, ApJ, 688, 67
Fabbian, D., Nissen, P. E., Asplund, M., Pettini, M., & Akerman, C. 2009, A&A, 500, 1143
Fenner, Y., Gibson, B. K., Gallino, R., & Lugaro, M. 2006, ApJ, 646, 184
Font, A. S., Johnston, K. V., Bullock, J. S., & Robertson, B. E. 2006, ApJ, 638, 585
Fuhrmann, K. 2004, AN, 325, 3
Fulbright, J. P. 2002, AJ, 123, 404
Fulbright, J. P., McWilliam, A., & Rich, R. M. 2007, ApJ, 661, 1152
Gehren, T., Liang, Y. C., Shi, J. R., Zhang, H. W., & Zhao, G. 2004, A&A, 413, 1045
Gilmore, G., & Reid, N. 1983, MNRAS, 202, 1025
Gonz[á]{}lez Hern[á]{}ndez, J. I., & Bonifacio, P. 2009, A&A, 497, 497
Gratton, R. G., Bonifacio, P., Bragaglia, A., et al. 2001, A&A, 369, 87
Gratton, R. G., Carretta, E., Desidera, S., et al. 2003, A&A, 406, 131
Gratton, R., Carretta, E., Matteucci, F., & Sneden, C. 1996, ASP Conf. Ser., 92, 307
Gratton, R. G., Sneden, S., & Carretta, E. 2004, ARA&A, 42, 385
Gustafsson, B., Edvardsson, B., Eriksson, K., et al. 2008, A&A, 486, 951
Haywood, M. 2008, MNRAS, 388, 1175
Hill, V., Lecureur, A., G[ó]{}mez, A., et al. 2011 A&A (in press), arXiv:1107.5199
Ibata, R. A., Gilmore, G.,& Irwin, M. J. 1994, Nature, 370, 194
Israelian, G., & Rebolo, R. 2001, ApJ, 557, L43
Kirby, E. N., Guhathakurta, P., Bolte, M., Sneden, C., & Geha, M. C. 2009, ApJ, 705, 328
Kormendy, J., & Kennicutt Jr., R. C. 2004, ARA&A, 42, 603
Korn, A. J., Grundahl, F., Richard, O., et al. 2007, ApJ, 671, 402
Kurucz, R. 1993, ATLAS9 Stellar Atmosphere Programs and 2 km/s grid. Kurucz CD-ROM No. 13. Cambridge, Mass., Smithsonian Astrophysical Observatory
Lecureur, A., Hill, V., Zoccali, M., et al. 2007, A&A, 465, 799
Lee, J-W., Kang, Y-W., Lee, J., & Lee, Y-W. 2009, Nature, 462, 480
Lee, Y.S., Beers, T.C., An, D., et al. 2011, ApJ, 738, 187
Luck, R. E., Kovtyukh, V. V., & Andrievsky, S. M. 2006, AJ, 132, 902
Marino, A. F., Milone, A. P., Piotto, G., et al. 2009, A&A, 505, 1099
Mashonkina, L., Gehren, T., Shi, J.-R., Korn, A.J., & Grupp, F. 2011, A&A, 528, A87
Mashonkina, L., Gehren, T., Travaglio, C., & Borkova, T. 2003, A&A, 397, 275
McWilliam, A., Preston, G. W., Sneden, C., & Searle, L. 1995, AJ, 109, 2757
McWilliam, A., & Rich, R. M. 1994, ApJS, 91, 749
Mel[é]{}ndez, J., Asplund, M., Alves-Brito, A., et al. 2008, A&A, 484, L21
Mel[é]{}ndez, J., Asplund, M., Gustafsson, B., & Yong, D. 2009, ApJ, 704, L66
Meza, A., Navarro, J. F., Abadi, M. G., & Steinmetz, M. 2005, MNRAS, 359, 93
Mishenina, T. V., Kovtyukh, V. V., Soubiran, C., Travaglio, C., & Busso, M. 2002, A&A, 396, 189
Neves, V., Santos, N. C., Sousa, S. G., Correia, A. C. M., & Israelian, G. 2009, A&A, 497, 563
Nissen, P. E. 2008, Physica Scripta, T133, 014022
Nissen, P. E., Akerman, C., Asplund, M., et al. 2007, A&A, 469, 319
Nissen, P. E., Primas, F., Asplund, M., & Lambert, D. L. 2002, A&A, 390, 235
Nissen, P. E., & Schuster, W. J. 1997, A&A, 326, 751
Nissen, P. E., & Schuster, W. J. 2010, A&A, 511, L10
Nissen, P. E., & Schuster, W. J. 2011, A&A, 530, A15
Nordstr[ö]{}m, B., Mayor, M., Andersen, J., et al. 2004, A&A, 418, 989
Pomp[é]{}ia, L., Hill, V., Spite, M., et al. 2008, A&A, 480, 379
Purcell, C. W., Bullock, J. S., & Kazantzidis, S. 2010, MNRAS, 404, 1711
Ram[í]{}rez, I., Allende Prieto, C., & Lambert, D. L. 2007, A&A, 465, 271
Reddy, B. E., Lambert, D. L., & Allende Prieto, C. 2006, MNRAS, 367, 1329
Reddy, B. E., Tomkin, J., Lambert, D. L., & Allende Prieto, C. 2003, MNRAS, 340, 304
Romano, D., & Matteucci, F. 2007, MNRAS, 378, L59
Ryde, N., Gustafsson, B., Edvardsson, B., et al. 2010, A&A, 509, A26
Sbordone, L., Bonifacio, P., Buonanno, R., et al. 2007, A&A, 465, 815
Schönrich, R., & Binney, J. 2009a, MNRAS, 396, 203
Schönrich, R., & Binney, J. 2009b, MNRAS, 399, 1145
Schuster, W. J., Moitinho, A., Márquez, A., Parrao, L., & Covarrubias, E. 2006, A&A, 445, 939
Searle, L., & Zinn, R. 1978, ApJ, 225, 357
Smith, V. V., Suntzeff, N. B., Cunha, K., et al. 2000, AJ, 119, 1239
Sousa, S. G., Santos, N. C., Israelian, G., Mayor, M., & Monteiro, M. J. P. F. G. 2007, A&A, 469, 783
Stephens, A., & Boesgaard, A. M. 2002, AJ, 123, 1647
Str[ö]{}mgren, B. 1987, In The Galaxy, ed. G. Gilmore & B. Carswell (Reidel, Dordrecht), 229
Takeda, Y., Hashimoto, O., & Taguchi, H. 2005, PASJ, 57, 751
Tolstoy, E., Hill, V., & Tosi, M. 2009, ARA&A, 47, 371
Tsujimoto, T., Nomoto, K., Yoshii, Y., et al. 1995, MNRAS, 277, 945
Turcotte, S., & Wimmer-Schweingruber, R. F. 2002, Journal of Geophysical Research, 107, 1442
Venn, K. A., Irwin, M., Shetrone, M. D., et al. 2004, AJ, 128, 1177
Ventura, P., D’Antona, F., Mazzitelli, I., & Gratton, R. 2001, ApJ, 550, L65
Yong, D., Carney, B. W., Teixera de Almeida, M. L., & Poha, B. L. 2006, AJ, 131, 2256
Yong, D., & Grundahl, F. 2008, ApJ, 672, L29
Yong, D., Grundahl, F., Lambert, D. L., Nissen, P. E., & Shetrone, M. D. 2003, A&A, 402, 985
Yong, D., Grundahl, F., Nissen, P. E., Jensen, H. R., & Lambert, D. L. 2005, A&A, 438, 875
Zoccali, M., Hill, V., Lecureur, A., et al. 2008, A&A, 486, 177
Zolotov, A., Willman, B., Brooks, A. M., et al. 2009, ApJ, 702, 1058
Zolotov, A., Willman, B., Brooks, A. M., et al. 2010, ApJ, 721, 738
[**List of abbreviations**]{}
AGB: Asymptotic Giant Branch\
CDM: Cold Dark Matter\
CRIRES: CRyogenic InfraRed Echelle Spectrograph\
DEIMOS: DEep Imaging Multi-Object Spectrograph\
DLA: Damped Lyman-Alpha\
dSph: dwarf Spheroidal\
ESA: European Space Agency\
ESO: European Southern Observatory\
EW: Equivalent Width\
FIES: FIbre fed Echelle Spectrograph\
FLAMES: Fibre Large Array Multi-Element Spectrograph\
GAIA: (ESA mission for exploring the Galaxy)\
HARPS: High Accuracy Radial velocity Planet Searcher\
HFS: Hyper-Fine Structure\
HIRES: HIgh Resolution Echelle Spectrometer\
H-R: Hertzsprung-Russell\
IMF: Initial Mass Function\
LMC: Large Magellanic Cloud\
LSR: Local Standard of Rest\
LTE: Local Thermodynamic Equilibrium\
NOT: Nordic Optical Telescope\
RAVE: Radial Velocity Experiment\
SDSS: Sloan Digital Sky Survey\
SNe: Supernovae\
UVES: Ultra-violet and Visible Echelle Spectrograph\
VLT: Very Large Telescope\
Chemical evolution\
Dwarf galaxies\
Echelle spectrographs\
Effective temperature\
Galactic halo\
Galactic disk\
Galactic bulge\
Globular clusters\
Nucleosynthesis\
Stellar atmospheres\
Stellar abundances\
Stellar metallicity\
Stellar populations\
[^1]: For two elements X and Y, \[X/Y\] = ${\rm log}(N_{\rm X}/N_{\rm Y})_{\rm star}\,\, - \,\,{\rm log}(N_{\rm X}/N_{\rm Y})_{\rm Sun}$, where $N_{\rm X}$ and $N_{\rm Y}$ are the number densities of the elements.
[^2]: Throughout this chapter, $\alpha$ refers to the average abundance of Mg, Si, Ca, and Ti, i.e. \[$\alpha$/Fe\] = $\frac{1}{4}$ (\[Mg/Fe\] + \[Si/Fe\] + \[Ca/Fe\] + \[Ti/Fe\])
[^3]: IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Two different approaches have recently been proposed for boundary handling in convolutional sparse representations, avoiding potential boundary artifacts arising from the circular boundary conditions implied by the use of frequency domain solution methods by introducing a spatial mask into the convolutional sparse coding problem. In the present paper we show that, under certain circumstances, these methods fail in their design goal of avoiding boundary artifacts. The reasons for this failure are discussed, a solution is proposed, and the practical implications are illustrated in an image deblurring problem.'
bibliography:
- 'paper.bib'
title: 'Convolutional Sparse Coding: Boundary Handling Revisited'
---
Convolutional Sparse Representations, Convolutional Sparse Coding, Boundary Effects, Deconvolution
Introduction {#sec:intro}
============
Convolutional sparse representations [@lewicki-1999-coding; @zeiler-2010-deconvolutional] provide a convenient alternative to the standard approach of block-based application of sparse representations to images. With the recent progress in the development of efficient algorithms for convolutional sparse coding (CSC) [@chalasani-2013-fast; @bristow-2013-fast; @wohlberg-2014-efficient; @wohlberg-2016-efficient], this form of sparse representations has become a practical approach for imaging inverse problems. A critical component of the current state of the art algorithms for CSC is the handling of the convolution in the frequency domain, which automatically implies circular boundary conditions. Recently, two different approaches have been proposed for avoiding boundary artifacts by performing a boundary extension and solving the CSC problem with a mask on the critical region in the data fidelity term. One of these proposes application of the mask decoupling (MD) technique [@almedia-2013-deconvolving] to the CSC problem [@heide-2015-fast]. The other, additive mask simulation (AMS) [@wohlberg-2016-boundary] takes advantage of the particular form of the CSC problem to represent the multiplicative mask by introducing an additional dictionary filter which is constrained in a way that simulates that mask.
Convolutional Sparse Coding {#sec:csc}
===========================
By far the most common form of CSC is $$\operatorname*{arg\,min}_{\{{\mathbf{x}}_m\}} \frac{1}{2} \normsz[\Big]{\sum_m {\mathbf{d}}_m \ast {\mathbf{x}}_m
- {\mathbf{s}}}_2^2 + \lambda \sum_m \alpha_m {\left\| {\mathbf{x}}_m \right\|}_1 \; ,
\label{eq:convbpdn}
\vspace{-1mm}$$ where $\{{\mathbf{d}}_m\}$ is a set of $M$ dictionary *filters*, $\ast$ denotes convolution, $\{{\mathbf{x}}_m\}$ is a set of coefficient maps, and the $\alpha_m$ allow distinct weighting of the $\ell_1$ term for each filter ${\mathbf{d}}_m$. Defining $D_m$ is a linear operator such that $D_m {\mathbf{x}}_m = {\mathbf{d}}_m \ast {\mathbf{x}}_m$, and defining block matrices and vectors
$$D = \left( \begin{array}{ccc}D_0 & D_1 &
\ldots \end{array} \right) \;\;
{\boldsymbol{\alpha}} = \left( \begin{array}{c} \alpha_0 I\\ \alpha_1 I\\
\vdots \end{array} \right)
\;\;
{\mathbf{x}} = \left( \begin{array}{c} {\mathbf{x}}_0\\ {\mathbf{x}}_1\\
\vdots \end{array} \right)
\vspace{-0.2mm}
\label{eq:dxcbpdn}$$
allows [(\[eq:convbpdn\])]{} to be written in the form $$\operatorname*{arg\,min}_{{\mathbf{x}}} \frac{1}{2} {\left\| D {\mathbf{x}} - {\mathbf{s}} \right\|}_2^2 + \lambda
{\left\| {\boldsymbol{\alpha}} \odot {\mathbf{x}} \right\|}_1 \;,
\label{eq:bpdn}
\vspace{-1mm}$$ where $\odot$ denotes the Hadamard product. This problem can be solved within the Alternating Direction Method of Multipliers (ADMM) [@boyd-2010-distributed] framework by an iterative scheme for the equivalent constrained problem $$\operatorname*{arg\,min}_{{\mathbf{x}},{\mathbf{y}}} \frac{1}{2} \normsz[\big]{
D {\mathbf{x}} - {\mathbf{s}}}_2^2 \!+\! \lambda
{\left\| {\boldsymbol{\alpha}} \odot {\mathbf{y}} \right\|}_1 \text{ s.t. } {\mathbf{x}} \! = \!
{\mathbf{y}} \; .
\label{eq:cbpdnsplit}
\vspace{-1mm}$$ The iterative scheme solves two subproblems, one associated with the data fidelity term, which is solved in the frequency domain [@wohlberg-2016-efficient], and the other involving the regularization term, which is solved via the closed-form expression for soft-thresholding.
Boundary masking involves introducing a diagonal matrix, $W$, implementing a spatial mask, into the data fidelity term, as in $$\operatorname*{arg\,min}_{{\mathbf{x}}} \frac{1}{2} \normsz[\big]{W D {\mathbf{x}}
- {\mathbf{s}}}_2^2 + \lambda {\left\| {\boldsymbol{\alpha}} \odot {\mathbf{x}} \right\|}_1 \; .
\label{eq:convbpdnws}
\vspace{-0.3mm}$$ The presence of the spatial mask prevents direct application of the frequency domain solution since, unlike the block components $D_m$ of $D$, $W$ is not diagonal in the frequency domain. The MD approach avoids this difficulty by replacing [(\[eq:bpdn\])]{} with a different constrained problem [@heide-2015-fast] This form decouples $W$ from the sum of convolutions $D {\mathbf{x}}$, facilitating the use of the same frequency domain solution used for [(\[eq:cbpdnsplit\])]{}.
The iterations of the ADMM algorithm for problem [(\[eq:cbpdnsplitmd\])]{} (see [@wohlberg-2016-boundary Sec. 4.2]) involve the following updates $$\begin{aligned}
(D^T D + I) {\mathbf{x}}^{(j+1)} &= D^T \big({\mathbf{y}}_1^{(j)} - {\mathbf{u}}_1^{(j)}\big) +
\big({\mathbf{y}}_0^{(j)} - {\mathbf{u}}_0^{(j)}\big) \label{eq:mdx} \\
{\mathbf{y}}_0^{(j+1)} &= \; {\mathcal{S}}_{\lambda {\boldsymbol{\alpha}} / \rho}\big( {\mathbf{x}}^{(j+1)} +
{\mathbf{u}}_0^{(j)} \big) \label{eq:mdy0} \\
(W^T W + \rho I) {\mathbf{y}}_1^{(j+1)} &= W^T {\mathbf{s}} + \rho \big(D
{\mathbf{x}}^{(j+1)} + {\mathbf{u}}_1^{(j)}\big) \label{eq:mdy1} \\
{\mathbf{u}}_0^{(j+1)} &= {\mathbf{u}}_0^{(j)} + {\mathbf{x}}^{(j+1)} - {\mathbf{y}}_0^{(j+1)}
\label{eq:mdu0}\\
{\mathbf{u}}_1^{(j+1)} &= {\mathbf{u}}_1^{(j)} + D {\mathbf{x}}^{(j+1)} -
{\mathbf{y}}_1^{(j+1)} \label{eq:mdu1}
\;,\end{aligned}$$ where the iteration index is indicated by a superscript in parentheses and $${\mathcal{S}}_{{\boldsymbol{\gamma}}}({\mathbf{u}}) = {\mathop{\mathrm{sign}}}({\mathbf{u}}) \odot \max(0, {\left| {\mathbf{u}} \right|}
- {\boldsymbol{\gamma}}) \;.
\label{eq:shrink}$$ Update [(\[eq:mdx\])]{} can be solved in the frequency domain as in [@wohlberg-2014-efficient], [(\[eq:mdy0\])]{} is solved via the closed-form expression for soft-thresholding, and [(\[eq:mdy1\])]{} is a computationally cheap linear problem since $W$ is diagonal.
The AMS method [@wohlberg-2016-boundary] takes a fundamentally different approach, retaining the original constrained form [(\[eq:cbpdnsplit\])]{}, but introducing into the representation an additive component that is constrained to be zero within the active part of the mask, and is unconstrained and un-penalized within the masked-out region. Due to length restrictions, this method is not discussed in further detail here, and all results presented in the following sections are computed using the MD method. It should be noted, though, that the AMS method is prone to the same issues, and amenable to similar solutions.
Boundary Masking Failure {#sec:mskfail}
========================
In principle these masking techniques provide for a complete decoupling between the active and masked-out regions of the solution, but, in practice, the effects of the presence of a boundary can propagate into the active region of the solution under certain circumstances.
Phenomenon
----------
{width="3cm"}
This effective failure of boundary masking is most easily illustrated with a simple synthetic test case[^1]. The test image was as in [Fig. \[fig:syntst\]]{}, and a corresponding mask matrix, $W$, was defined to represent a mask that is zero in the padded region and unity elsewhere. The dictionary consisted of three $16 \times 16$ sample filters, a Gaussian surface, ${\mathbf{d}}_0$, and horizontal and vertical lines, ${\mathbf{d}}_1$ and ${\mathbf{d}}_2$ respectively. The corresponding weights were set as $\alpha_0 = 0, \alpha_1 = \alpha_2 = 1$, [i.e.]{}no regularisation on the coefficient map for the Gaussian filter since its role was representing the smooth component of the image. This setup provides a simplified cartoon of convolutional decomposition of a natural image using a dictionary consisting of a single smooth filter for representing the smooth component of the image and a potentially large number of learned filters for representing the image texture and edges [@wohlberg-2016-convolutional2 Sec. 3].
Problem [(\[eq:cbpdnsplitmd\])]{} was solved via 500 iterations of the ADMM algorithm for the MD approach[^2]. In addition to the full signal reconstruction, $\sum_m {\mathbf{d}}_m \ast {\mathbf{x}}_m$, the individual contribution to this reconstruction from each filter, ${\mathbf{d}}_m \ast {\mathbf{x}}_m$, was also computed. A cross section of these results is displayed in [Fig. \[fig:mskfail\]]{}, and a zoom into the left boundary region is displayed in [Fig. \[fig:mskfailcrop\]]{}. The boundary masking failure is subtle in that it is not visible if we only look at the full signal reconstruction, $\sum_m {\mathbf{d}}_m \ast {\mathbf{x}}_m$, since the reconstruction matches the reference to very high accuracy all the way to the boundary of the masked region. The boundary artifacts only become visible if we examine the individual components, ${\mathbf{d}}_m \ast {\mathbf{x}}_m$: here we see that the smooth component starts to decay *before* the boundary is reached, with the accuracy of the final reconstruction being maintained by a compensating peak in the edge component, also *within* the active region of the mask. Since the overall sum is accurate, these artifacts in the individual components do not matter if the image is directly reconstructed from its sparse representation, but the sparse representation is usually computed as part of the solution of some inverse problem. In this scenario the coefficient maps are processed prior to reconstruction, so that canceling of the component artifacts to give an accurate reconstruction is no longer guaranteed, risking artifacts in the final output image.
Initialisation
--------------
This phenomenon is rather surprising since the functional we minimize is specifically designed to avoid boundary artifacts. To understand and address it, we need to consider the choice of initial values for variables ${\mathbf{y}}_0, {\mathbf{y}}_1, {\mathbf{u}}_0$, and ${\mathbf{u}}_1$ in the MD algorithm. This choice is not discussed at all in [@heide-2015-fast], but it is a reasonable assumption (supported by inspection of the corresponding publicly available software) that the authors follow the common approach for ADMM algorithms in setting them all to zero vectors. Although not discussed in [@wohlberg-2016-boundary] either, the actual experiments reported in that work (see [@wohlberg-2016-sporco]) made use of an alternative initialisation, setting ${\mathbf{y}}_1$ to the input signal ${\mathbf{s}}$. A third initialisation strategy is proposed here: setting ${\mathbf{y}}_1$ to a version of the signal that has been extended symmetric extension instead of zero-padding.
The convergence of the MD algorithm with these three initialization choices, for the synthetic test problem, is compared in [Fig. \[fig:fnccnv\]]{}. It can be seen that the two signal-based initialisations are initially quite similar, and substantially better than zero initialisation, but the symmetric extension exhibits better behaviour after 100 iterations. The corresponding evolution of the reconstructed edge component, $\sum_{m=1}^2 {\mathbf{d}}_m \ast {\mathbf{x}}_m$, at the location of the artifact (see [Figs. \[fig:mskfix\]]{} and [\[fig:mskfixcrop\]]{}) and at the signal centre are compared in [Fig. \[fig:smpcnv\]]{}. The cause of the apparent boundary masking failure is now clear: the reconstructed edge component converges very slowly to its “correct” value for both the zero and the zero padded initialisations. The symmetric-extension initialisation, in contrast, results in convergence at the boundary that is comparable to that at the signal centre, far removed from the boundary effects. With the use of this initialisation, the boundary artifact is no longer apparent in [Fig. \[fig:mskfix\]]{} and [Fig. \[fig:mskfixcrop\]]{} after only 100 iterations.
The effect of the initial choice of ${\mathbf{y}}_1$ can be understood by referring to the optimisation problem [(\[eq:convbpdnws\])]{} and the corresponding ADMM iterations [(\[eq:mdx\])]{}–[(\[eq:mdu1\])]{}. The coefficient map, ${\mathbf{x}}_0$, of the smooth dictionary filter, ${\mathbf{d}}_0$, has zero weight in the regularization term ($\alpha_0 = 0$) and its reconstruction, ${\mathbf{d}}_0 * {\mathbf{x}}_0$, has no influence on the data fidelity term outside of the active region of the mask matrix, $W$. The solution for the part of this coefficient map for which ${\mathbf{d}}_0 * {\mathbf{x}}_0$ lies outside of the active mask region is therefore completely undetermined, all possible choices having equal cost. It can be seen from [(\[eq:mdx\])]{} that the initial choice of ${\mathbf{y}}_1$ determines the initial solution ${\mathbf{x}}$ for the entire domain, [i.e.]{}with no application of the mask matrix $W$. The part of this initial solution that lies outside of the active region of the mask persists for the remainder of the iterations since [(\[eq:mdy1\])]{} does not modify ${\mathbf{y}}_1$ outside of the active region. As a result, the solution to which ADMM algorithm [(\[eq:mdx\])]{}–[(\[eq:mdu1\])]{} converges depends on the choice of the initial value for ${\mathbf{y}}_1$, as can be seen by comparing the “smooth component” reconstructions in [Figs. \[fig:mskfail\]]{} and [\[fig:mskfix\]]{}.
Gaussian Blur Deconvolution {#sec:rslt}
===========================
The impact of the slow boundary convergence is illustrated in a Gaussian blur deconvolution problem. The $512 \times 512$ pixel greyscale “Lena” image with pixels rescaled to the range $[0,1]$ was used as a reference image ${\mathbf{s}}$. A test image, ${\mathbf{s}}_{\mathrm{bn}}$, was constructed by convolving ${\mathbf{s}}$ with a $7 \times 7$ sample Gaussian filter ${\mathbf{h}}$ with unit standard deviation parameter, adding Gaussian white noise with standard deviation 0.01, and zero-padding by 39 pixels on all sides. The corresponding mask matrix, $W$, was defined to represent a spatial mask set to be zero in the padded region and unity elsewhere.
The CSC-based deconvolution method involved solving problem [(\[eq:cbpdnsplitmd\])]{} with the test image, ${\mathbf{s}}_{\mathrm{bn}}$, and a blurred dictionary, $\{{\mathbf{g}}_m\}$, and constructing the estimate of the deblurred image as the reconstruction of the resulting coefficient maps with respect to a corresponding unblurred dictionary, $\{{\mathbf{d}}_m\}$, [i.e.]{}$$\begin{aligned}
{\mathbf{x}}_{\mathrm{b}} &= \operatorname*{arg\,min}_{{\mathbf{x}}} \frac{1}{2} \normsz[\big]{W G {\mathbf{x}}
- {\mathbf{s}}_{\mathrm{bn}}}_2^2 + \lambda {\left\| {\boldsymbol{\alpha}} \odot
{\mathbf{x}} \right\|}_1 \\
\hat{{\mathbf{s}}} &= D {\mathbf{x}}_{\mathrm{b}} \;,\end{aligned}$$ where $\hat{{\mathbf{s}}}$ is the deconvolved estimate of ${\mathbf{s}}$, and $D$ and $G$ are the block-matrix forms, as introduced in [(\[eq:dxcbpdn\])]{}, of dictionaries $\{{\mathbf{d}}_m\}$ and $\{{\mathbf{g}}_m\}$ respectively. The reconstruction dictionary, $\{{\mathbf{d}}_m\}$, consisted of a smooth Gaussian filter of $64 \times 64$ samples (with standard deviation parameter 5.0) and a learned multiscale dictionary with 16 filters of $8 \times 8$ samples, 32 filters of $12 \times 12$ samples, and 48 filters of $16 \times 16$ samples. (As in the previous section, the component of ${\boldsymbol{\alpha}}$ corresponding to the Gaussian dictionary component was set to zero.) The corresponding blurred dictionary $\{{\mathbf{g}}_m\}$ was obtained by convolving each filter in $\{{\mathbf{d}}_m\}$ by the blurring kernel ${\mathbf{h}}$, [i.e.]{}${\mathbf{g}}_m = {\mathbf{h}} \ast {\mathbf{d}}_m$, with appropriate zero padding to avoid boundary effects on the blurred dictionary filters.
Two different deconvolved estimates were computed, one using 500 iterations of the MD algorithm with ${\mathbf{y}}_1$ initialised using the zero padded test image (CSC-zp), and the other initialised using a version of the test image extended via symmetric extension (CSC-se). The corresponding reconstruction PSNR values are presented in [Table \[tbl:deconv\]]{}, and the bottom right corners of the two images are displayed in [Fig. \[fig:dcnvcrp\]]{}. Note the very clear boundary artifacts in [Fig. \[fig:cscbflcrp\]]{}, and the substantial effect on the overall reconstruction PSNR in [Table \[tbl:deconv\]]{}, despite the occurrence of the boundary artifacts in a relatively small fraction of the entire image.
-- --
-- --
----------- ------- ------- ------- ------- -------
PSNR (dB) 26.81 29.90 30.41 30.17 30.60
----------- ------- ------- ------- ------- -------
: A comparison of PSNR values for the blurred and noisy test image and the deconvolved estimates obtained via the CSC MD algorithm with two different initialisations (CSC-zp and CSC-se) as well as via the TV [@dias-2006-total] and EPLL [@zoran-2011-learning] methods. []{data-label="tbl:deconv"}
Although the primary focus of these experiments is to demonstrate the impact of the boundary phenomenon discussed above, comparisons with the Total Variation (TV) [@dias-2006-total] and Expected Patch Log Likelihood (EPLL) [@zoran-2011-learning] methods are included in [Table \[tbl:deconv\]]{} as a performance reference. CSC-zp has the worst PSNR, while that of CSC-se is intermediate between those of TV and EPLL.
Conclusions {#sec:cnclsn}
===========
The mathematical formulation of boundary handling via a masked data fidelity term might lead one to conclude that the type of boundary extension is irrelevant to the solution since it lies outside of the masked region. The phenomenon of very slow convergence at the boundary when zero padding, however, contradicts this intuition and demonstrates that some care must be exercised in choosing the extension method. For the cases demonstrated here, initialisation of the MD algorithm via symmetric extension of the input signal completely suppresses the boundary artifacts that are observed when initialising to a zero vector or via zero padding of the input signal
[^1]: This is not a contrived example: it was constructed with the goal of understanding a phenomenon that was first observed while attempting to solve a practical inverse problem involving natural images.
[^2]: Computed using the Matlab version of SPORCO library [@wohlberg-2016-sporco].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We report an angle-resolved photoemission (ARPES) study of [$\beta$-YbAlB$_4$]{}, which is known to harbor unconventional quantum criticality (QC) without any tuning. We directly observe a quasiparticle peak (QP), emerging from hybridization, characterized by a binding energy and an onset of coherence both at about 4 meV. This value conforms with a previously observed reduced Kondo scale at about 40 K. Consistency with an earlier study of carriers in [$\beta$-YbAlB$_4$]{} via the Hall effect strongly suggests that this QP is responsible for the QC in [$\beta$-YbAlB$_4$]{}. A comparison with the sister polymorph [$\alpha$-YbAlB$_4$]{}, which is not quantum critical at ambient pressure, further supports this result. Indeed, within the limitation of our instrumental resolution, our ARPES measurements do not show tangible sign of hybridization in this locally isomorphic system, while the conduction band we observe is essentially the same as in [$\beta$-YbAlB$_4$]{}. We therefore claim that we identified by ARPES the carriers responsible for the QC in [$\beta$-YbAlB$_4$]{}. The observed dispersion and the underlying hybridization of this QP are discussed in the context of existing theoretical models.'
author:
- Cédric Bareille
- Shintaro Suzuki
- Mitsuhiro Nakayama
- Kenta Kuroda
- 'Andriy H. Nevidomskyy'
- Yosuke Matsumoto
- Satoru Nakatsuji
- Takeshi Kondo
- Shik Shin
title: 'Kondo hybridization and quantum criticality in $\beta$-YbAlB$_4$ by laser-ARPES'
---
Introduction
============
The formation of a non-magnetic Kondo-singlet ground-state commonly provides a description of heavy Fermi liquid (FL) in intermetallics rare-earth compounds. It emerges below the Kondo temperature ${T_{\mathrm{K}}}$, from the screening of the local moments of $f$ orbitals by conduction electrons, which accompanies the onset of coherence of a quasiparticle (QP) peak binding at $k_{\mathrm{B}}{T_{\mathrm{K}}}$: the Kondo resonance peak [@Hewson_Kondo2HF]. When external parameters are accurately tuned, numerous heavy fermion systems deviate from this normal FL. In the conventional scenario, spin-density-waves instabilities induce such non-FL behavior by driving the system to the critical regime of a magnetic quantum transition [@Lohneysen_RMP2007].
In [$\beta$-YbAlB$_4$]{}, quantum criticality (QC) has been observed below about 4 K at zero magnetic field and ambient pressure [@Matsumoto_Science2011], away from any magnetic order [@Tomita_Science2015]. It, therefor, can not be cast in this usual picture of magnetic instability. Moreover it shows an unexpected ($T/B$)-scaling [@Matsumoto_Science2011; @Matsumoto_JPSJ2015] and unusual critical exponents [@Nakatsuji_NatPhys2008], which challenge theoretical descriptions. Diverse works [@Misawa_JPSJ2009; @Ramires_PRL2012; @Pixley_PRL2012; @Hackl_PRB2011; @Watanabe_JPSJ2014; @Shaginyan_PRB2016] attempt to model this unconventional QC, without any possible consensus. Although the FL behavior is recovered with a moderate magnetic field, its description and formation in term of Kondo screening are still poorly understood. Indeed, despite a large Kondo scale ${T_{\mathrm{K}}}\approx 250$ K, incoherent skew scattering is observed down to ${T_{\mathrm{coh}}}\approx 40$ K [@OFarrell_PRL2012], and magnetic moments are still present as low as $T^* \approx 8$ K [@Matsumoto_JPSJ2015]. Kondo lattice behavior finally sets in only below $T^*$. Advanced analysis of Hall effect measurements shed some light by the identification of two different components of carriers with distinct Kondo scales [@OFarrell_PRL2012], which we call carriers $a$ and carriers $b$ for convenience. Carriers $a$ govern the longitudinal transport with ${T_{\mathrm{K}}}$ as the Kondo scale and a relatively high carrier density $n_a^\mathrm{Hall}$, while carriers $b$ have a low carrier density $n_b^\mathrm{Hall}$ with electronlike character and gains coherence below a second Kondo scale at ${T_{\mathrm{coh}}}$. The conclusion from Hall effect measurements further identifies carriers $b$ as responsible for both Kondo lattice and QC behaviors below $T^*$ [@OFarrell_PRL2012]. Nevertheless, the details about the emergence of these carriers and about the underlying Kondo physics are still lacking.
In the present article, we take advantage of the apparent location of the QC at zero magnetic field, without tuning of any control parameter [@Matsumoto_Science2011], to directly probe QP in [$\beta$-YbAlB$_4$]{} by high-resolution laser-ARPES. In order to highlight any relation between microscopic observations and the QC, we also perform ARPES measurements on the sister compound [$\alpha$-YbAlB$_4$]{}, whose structure is sketched alongside that of [$\beta$-YbAlB$_4$]{} in Fig. \[fig:kzDep\](a). This locally isomorphic compound exhibits very similar behaviors and temperature scales, however unlike [$\beta$-YbAlB$_4$]{}, it does not display QC, developing instead a heavy Fermi liquid [@Matsumoto_PRB2011]. Moreover, it has been shown that the magnetization in [$\beta$-YbAlB$_4$]{} can be separated into two components: the QC component and a heavy Fermi-liquid component similar to that in [$\alpha$-YbAlB$_4$]{} [@Matsumoto_JPSJ2015]. Thus, [$\alpha$-YbAlB$_4$]{} is a perfect control system to study the unconventional criticality in [$\beta$-YbAlB$_4$]{}.
Our measurements reveal the hybridization between the crystal-field split heavy Yb $f$-electron bands and a light conduction band in [$\beta$-YbAlB$_4$]{}. The resulting electronlike QP has a binding energy and an onset of coherence at about about 4 meV, consistent with carriers $b$ highlighted by Hall measurements as responsible for the QC. By contrast, we find almost no or very weak hybridization in the sister $\alpha$-phase, while the dispersion of the conduction band is very similar between the two phases. Our results strongly suggest the importance of the Kondo hybridization for the microscopic mechanism behind the QC in [$\beta$-YbAlB$_4$]{}.
Methods
=======
High-quality single crystals were grown by using the Al-flux method as described in the literature [@Macaluso_CoM2007]. The excess Al flux was etched by a water solution of sodium hydroxide. All ARPES measurements were performed with a SES R4000 analyzer, in a chamber shielded from magnetic field by mu-metal. Clean surfaces were obtained by cleaving crystals along the $(a,b)$ plane in ultra-high vacuum (below $5\times10^{-11}$ Torr). Measurements in the (100)-plane were carried out at the 1-cubed end-station at Helmholtz-Zentrum Berlin, by tuning the photon energy from 24 eV to 120 eV, thanks to the synchrotron radiation of the BESSY-II facility. The temperature was about 1 K, and the overall resolution was about 10 meV. Measurements in the (001) plane for both $\beta$ and $\alpha$ phases, were realized with a 6.994 eV laser at the ISSP, at a temperature of 5 K, with an overall resolution of about 3 meV. Data from [$\beta$-YbAlB$_4$]{} were measured with circular polarized light, while data from [$\alpha$-YbAlB$_4$]{} were measured with linear horizontal polarized light. Nevertheless, as Fig. S1 of the Supplemental Material shows [@SuppMat], measurements with linear horizontal polarized light give similar results to those with circular polarized light. Thus, this difference does not affect our observations. Treatments by 2D curvature of some spectra of this work are performed following the method described in Ref. .
Results
=======
The [$\beta$-YbAlB$_4$]{} crystal has an orthorhombic structure [@Macaluso_CoM2007]; Fig. \[fig:kzDep\](b) schematizes the corresponding Brillouin-zone (BZ). Samples were cleaved along the $(a,b)$ plane to perform photoemission. Fig. \[fig:kzDep\](c) shows the edges of the BZ overlaid on a mapping in the (100) plane, 50 meV below the Fermi level, while the dispersion along the normal emission, $\Gamma$-$Z$ line, is displayed on Fig. \[fig:kzDep\](d). Conversion into momentum space was made in the frame of the nearly-free-electron final-state approximation [@Damascelli_2004]. The inner potential was determined to be $V_0 \approx 22$ eV thanks to the point-group symmetry and periodicity, with, as a main indication, a hole-like band centered on the $Z$ point, depicted by the gray parabola in panel (d) and the gray ellipse in panel (c). Also visible in panel (d) is a nearly flat band bounded at about 100 meV below $E_\mathrm{F}$ – it is a part of the Yb$^{2+}$ $|J = 7/2\rangle$ multiplet, which has been previously observed by X-ray photoemission spectroscopy [@Okawa_PRL2010]. Complete investigation and discussion over the full three-dimensional BZ will be part of a future work. Here, we will focus instead on the results from laser-ARPES, which, while being more restricted in $\mathbf{k}$ space, offers very high resolution and intensity.
Given the laser photon energy $h\nu = 6.994$ eV, the available BZ cut is depicted by the lower purple circle in Fig. \[fig:kzDep\](c), which crosses the $\Gamma$-$Z$ line less than 0.2 Å$^{-1}$ below the Z point, named ${\widetilde{Z}}$ for convenience. Measurements with photon energy $h\nu = 54$ eV reach an equivalent position along the $\Gamma$-$Z$ line \[see the higher purple circle in Fig. \[fig:kzDep\](c)\]. The dispersive hole-like band with the top bound at about $-45$ meV, as well as the flat band at about $-90$ meV, were observed using the light source both with photon energy $h\nu = 54$ eV \[see panels (c) and (d)\] as well as with the laser in panel (e), thus providing a calibration of our laser-ARPES. In what follows, we focus on the bands above $E>-40$ meV, which are observable thanks to laser-ARPES, and we compare them to our measurements on [$\alpha$-YbAlB$_4$]{}. We investigate the vicinity of $E_\mathrm{F}$ with the stack of energy distribution curves (EDCs), Fig. \[fig:disp5K\](a), extracted along $k_\mathrm{y}$, i.e., along the analyzer slit, with $k_\mathrm{x} = 0$ \[full gray line $A$ on Fig. \[fig:disp5K\](b)\]. We observe three dispersions; they can be modeled by a light electronlike band with $m^\ast\approx 3~m_e$ and the band bottom of $\approx 20$ meV below $E_\mathrm{F}$, which hybridizes and forms anticrossings with two flat bands, a first one around $-4$ meV and a second one around $-13$ meV \[thin dashed lines in Figs. \[fig:disp5K\](c) and \[fig:disp5K\](d)\]. These two flat bands, separated by 9 meV are consistent with the crystal electric field of about 80 K, extracted from fitting the magnetic susceptibility [@Nevidomskyy_PRL2009]. From this fitting and local symmetry considerations detailed in Ref. , the two flat bands can then be identified as the ground-state $J_z=\pm5/2$ ([$\left|\pm5/2\right>$]{}), at $-4$ meV, and the first crystal-field excited state, dominated by $J_z=\pm3/2$ ([$\left|\pm3/2\right>$]{}), at $-13$ meV, of the J = 7/2 Yb multiplet. In Fig. \[fig:disp5K\](c), solid lines are the results of fitting to the hybridization model with a constant hybridization function $V \approx 4$ meV. The first anti-crossing at $-5$ meV is nicely fitted, and forms an electronlike QP band \[blue line in Fig. \[fig:disp5K\](c)\] bound at about 4 meV below $E_\mathrm{F}$. This energy scale, one order of magnitude smaller than ${T_{\mathrm{K}}}\approx 250$ K, is consistent with the expectation from previous experimental observations [@OFarrell_PRL2012; @Matsumoto_PRB2011] of a reduced Kondo scale at ${T_{\mathrm{coh}}}\approx 40$ K ($\approx 3.4$ meV).
The temperature dependence of EDCs in Fig. \[fig:Tdep\] confirms the relation between the QP peak and ${T_{\mathrm{coh}}}$. Panels (a) and (b) shows EDCs from 5 K to 60 K at the inner ($k_\mathrm{y} = 0$ Å$^{-1}$) and outer ($k_\mathrm{y} = 0.17$ Å$^{-1}$) sides of the anti-crossing respectively. The same EDCs divided by the Fermi-Dirac step are displayed at the bottom of these two panels. For both, the QP peak slowly disappears as the temperature goes up. It is clearer on the outer EDCs where the peak intensity suddenly drops from 30 K to 40 K. The peak is totally restored by cooling back down (light blue EDC), showing that our observations are not due to surface degradation. For both inner and outer peaks, the intensity integrated along the shaded area \[see panels (a) and (b), respectively\] of EDCs divided by the Fermi-Dirac step is displayed against the temperature in panel (c), relatively to the value at 60 K. Both peaks slowly drop when temperature crosses ${T_{\mathrm{coh}}}$. It is worth to emphasize that, to our knowledge, it is the first direct and fully resolved observation by ARPES of a Kondo-like hybridization with unified binding energy, onset of coherence, and Kondo scale. While a similar observation has been reported on YbRh$_2$Si$_2$, both the QP peak and the hybridization gap were not fully resolved [@Mo_PRB2012].
Nevertheless, here, the second anti-crossing at $E\approx -15$ meV is badly reproduced by the constant hybridization model \[pink line in Fig. \[fig:disp5K\](c)\]. By taking $\mathbf{k}$-dependent hybridization proposed theoretically [@Nevidomskyy_PRL2009; @Ramires_PRL2012], we obtain a more satisfying fit, as shown by the pink line in Fig. \[fig:disp5K\](d). The significance of $\mathbf{k}$-dependent hybridization is developed in the discussion section. The fitting parameters of both models are detailed in the Supplemental Material [@SuppMat]. Finally, the hybridization with the [$\left|\pm5/2\right>$]{} level at about $-4$ meV results in the small electronlike QP peak \[blue line in panels (c) and (d) of Fig. \[fig:disp5K\]\], which crosses the Fermi level at $k^{\beta}_{\mathrm{F1}} \approx 0.09~$Å$^{-1}$ and has a mass of about $3~m_e$ at $E_\mathrm{F}$. It forms quite an isotropic Fermi sheet, as can be seen via the in-plane mapping at $E_\mathrm{F}$ in Fig. \[fig:disp5K\](b)(blue circle). The size, mass, and isotropy of this small pocket are compatible with the observations by quantum oscillations on [$\beta$-YbAlB$_4$]{} [@OFarrell_PRL2009]. It should be noted that, along the $\left<010\right>$ direction, the [$\left|\pm5/2\right>$]{} level crosses $E_\mathrm{F}$ at about $k^{\beta}_{\mathrm{F2}} \approx 0.25$ Å$^{-1}$. This second Fermi-momentum is exhibited by red arrows on the Fermi-mapping, Fig. \[fig:disp5K\](b), as well as on the cut along the $\left<010\right>$ direction, Fig. \[fig:disp5K\](e). Blue, red and pink full lines are guides to the eyes for the same dispersions as Fig. \[fig:disp5K\](d). The observation of a second Fermi-momentum implies an overlapping of the blue and red hybridized bands, on the Fig. \[fig:disp5K\](e). In order to account for it, we need to introduce a small dispersion to the [$\left|\pm5/2\right>$]{} level in our model, corresponding to a mass $m_f^* \approx 70~m_e$ in the $\left<010\right>$ direction. This very large mass makes it difficult to be observed by quantum oscillations. Carrier density can be estimated at the $T=0$ limit assuming spherical Fermi sheets by $n=k_\mathrm{F}^3/3\pi^2$ [@AshcroftMermin]. Before hybridization, the little conduction band has a density of $n_c^\mathrm{ARPES} \approx 7.4\times10^{25}$ m$^{-3}$. It is of the same order as the value estimated from Hall measurements [@OFarrell_PRL2012] of carriers $b$ at $T = 140$ K (i.e., above the integration of the $f$ electron to the Fermi surface): $n_b^\mathrm{hall}(T = 140$ K$)\approx 16\times10^{25}$ m$^{-3}$. Below ${T_{\mathrm{coh}}}$, the carrier density could not be accurately extracted from the Hall measurements. Yet, estimation of the density of the two Fermi sheets from our measurements (i.e., including $f$ electrons), $n^\mathrm{ARPES} \approx 55\times10^{25}$ m$^{-3}$, is still one order of magnitude smaller than the density of carriers $a$ identified as the main contribution to the transport by Hall effect measurements, estimated at 50 K to $n_a^\mathrm{hall}(T = 50$ K$)\approx 416\times10^{25}$ m$^{-3}$. Therefore, carriers of [$\beta$-YbAlB$_4$]{} observed here, thanks to the high resolution of laser-ARPES, match carriers $b$ identified as being responsible for both Kondo lattice and QC behaviors below $T^*$ by Hall effect measurements. Specifically, they are of electronlike character, gain coherence below ${T_{\mathrm{coh}}}$, and have low carrier density. From these evidences, we argue that the observed Fermi sheets and the underlying Kondo hybridization are closely related to both Kondo lattice and QC behaviors below $T^*$. To verify our claim, we now look at the electronic structure of the sister compound [$\alpha$-YbAlB$_4$]{} in the vicinity of $E_\mathrm{F}$. Fig. \[fig:aYAB\](a) displays the laser-ARPES curvature spectra along the $\left<100\right>$ direction, where mainly two features are visible: an almost flat band and a light electronlike band dispersing all the way up to $E_\mathrm{F}$. The former is revealed by fitting the peaks in the EDCs (open circles) at about $-13$ meV below $E_\mathrm{F}$. The dispersive electronlike band is better seen by fitting the peaks from the momentum distribution curves (MDCs) \[full circles in Fig. \[fig:aYAB\](a) and \[fig:aYAB\](b)\] down to $-12$ meV; the two peaks are difficult to distinguish at higher binding energies. The peaks are directly visible on the stack of MDCs, Fig. \[fig:aYAB\](b). This light dispersion crosses $E_\mathrm{F}$ at $k^{\alpha}_{\mathrm{F}} \approx 0.13$ Å$^{-1}$.
In Fig. \[fig:bVSa\](a), we compare the two aforementioned bands with the ones observed in [$\beta$-YbAlB$_4$]{}. Intriguingly, they are identical to the dispersive electronlike band and the flat [$\left|\pm3/2\right>$]{} band derived from the $\mathbf{k}$-dependent hybridization model used for [$\beta$-YbAlB$_4$]{}, copied onto panel (a) with dashed lines. On the other hand, the [$\left|\pm5/2\right>$]{} level is seemingly not observed in [$\alpha$-YbAlB$_4$]{}. A peak is visible at $k=0$, however it is very limited in momentum space and may correspond to a different conduction band. We directly compare the EDCs from $\beta$- and [$\alpha$-YbAlB$_4$]{} in Fig. \[fig:bVSa\](b), in red and green lines, respectively. Red open arrows show the [$\left|\pm5/2\right>$]{} level as a peak on the [$\beta$-YbAlB$_4$]{} EDCs. On [$\alpha$-YbAlB$_4$]{}, only shoulders are visible. When divided by the Fermi-Dirac step, these shoulders disappear, and no peak is left, thus indicating the absence of the [$\left|\pm5/2\right>$]{} level in our data on [$\alpha$-YbAlB$_4$]{}. On the other hand, one would expect the crystal field splitting to be very similar between these two locally isomorphic phases, and the absence of the [$\left|\pm5/2\right>$]{} level in our data could be a result of a bad sample surface quality, which may especially affect the conclusions so close to $E_\mathrm{F}$. What can be said with certainty, however, is that we do not observe any sign of hybridization between either of the flat Yb $4f$ levels and the conduction electron bands in [$\alpha$-YbAlB$_4$]{}, within our instrumental resolution of $\sim 3$ meV. Indeed, the light electronlike band disperses straight through the $-4$ meV energy to cut through $E_\mathrm{F}$, while no gap is visible at about $-13$ meV where one would expect the second anti-crossing with the Yb [$\left|\pm3/2\right>$]{} level. This is evidenced by the EDCs in Fig. \[fig:bVSa\](b): on [$\alpha$-YbAlB$_4$]{}, the [$\left|\pm3/2\right>$]{} level (full green arrows) is at the same position on both $k = 0$ and $k = \pm0.15$ Å$^{-1}$, in clear contrast with EDCs on [$\beta$-YbAlB$_4$]{}. We conclude that the manifestation of the Kondo hybridization in the one-particle spectral function (see Ref. ) is much less pronounced in $\alpha$-YbAlB$_4$ compared to the $\beta$-phase, although within the experimental resolution of our laser-ARPES measurements, we cannot exclude manifestations of the hybridization smaller than about $\sim~3$ meV in the $\alpha$-phase.
Discussion
==========
The absence of tangible sign of Kondo hybridization in our ARPES data on [$\alpha$-YbAlB$_4$]{}, in contrast to [$\beta$-YbAlB$_4$]{}, further supports the idea that the hybridization and emerging QP peak play a crucial role in the unconventional quantum criticality observed in the latter compound. Moreover, our analysis shows that the heavy dispersions of [$\beta$-YbAlB$_4$]{} are best fitted with a $\mathbf{k}$-dependent hybridization function, as put forward by the theory of the critical nodal metal [@Ramires_PRL2012]. The key feature of this latter model is that the Kondo hybridization vanishes at the $\Gamma$ point quadratically in the in-plane momentum ($k_x$, $k_y$), stemming from the $|\pm5/2\rangle$ nature of the $f$-level. It implies the formation of a nodal line along the $c$ axis, which causes the non-FL behavior in $\beta$-YbAlB$_4$ [@Ramires_PRL2012]. It is important to note however that for this model to explain the observed $T/B$ scaling in [$\beta$-YbAlB$_4$]{}, the renormalized position $\tilde{E}_f$ of the ground state [$\left|\pm5/2\right>$]{} doublet must be pinned at the Fermi level. It is therefore difficult to unequivocally corroborate the theoretical model, in its original form, based on our ARPES data. It follows from our ARPES analysis that this doublet actually crosses the Fermi level at the wavevector $k^{\beta}_{\mathrm{F2}} \approx 0.25$ Å$^{-1}$ and has an intrinsic bandwidth, characterized by a heavy quasi-particle mass $m_f^\ast \sim 70~m_e$. Clearly, this situation is more complex than in the nodal metal theory of Ref. , where the [$\left|\pm5/2\right>$]{} doublet was assumed to be perfectly flat, in other words the Yb-Yb electron hopping was neglected. In this situation, it could be possible for a portion of the nodal line to be bound at the Fermi level. Since our present observations already support the $\mathbf{k}$-dependent hybridization, investigating the dispersion of the [$\left|\pm5/2\right>$]{} doublet along the $c$ axis would be desirable.
We note that the observation of the second Fermi-momenta $k^{\beta}_{\mathrm{F2}}$ may also be consistent with the model of deconfinement of the $f$ electrons [@Senthil_PhysicaB2005], whose application to [$\beta$-YbAlB$_4$]{} has been proposed by Hackl and Thomale [@Hackl_PRB2011]. In this model, $k^{\beta}_{\mathrm{F2}}$ hosts the critical fluctuations, while carriers at $k^{\beta}_{\mathrm{F1}}$, labeled ‘cold’, exhibit a FL behavior. Naively, this is compatible with the similar observation of the Fermi-momenta $k^{\alpha}_{\mathrm{F}}$ in the FL of [$\alpha$-YbAlB$_4$]{}. Nevertheless, we did not observe any direct sign of such deconfinement in our data; from the ARPES point of view, measurements at lower temperature are necessary to follow the coherence of the [$\left|\pm5/2\right>$]{} and to settle this question. In fact, previous electrical and thermal transport measurements [@Sutherland_PRB2015] discards this scenario by highlighting a discrepancy with the expectation advanced by Hackl and Thomale of a characteristic maximum in the Wiedemann-Franz ratio [@Hackl_PRB2011].
Finally, we note that the band structure observed in the present ARPES measurement is consistent with the theory of electron-spin resonance (ESR) in [$\beta$-YbAlB$_4$]{} by Ramires [*et al.*]{} [@Ramires_PRL2014]. Indeed, in addition to the unconventional QC, [$\beta$-YbAlB$_4$]{} shows a singular ESR signal, characterized by hyperfine satellites at low temperature, and by the constancy of the ESR signal intensity [@Holanda_PRL2011]. The work of Ramires [*et al.*]{} reproduces first the hyperfine satellites by including Yb atoms with nonzero nuclear spin into the Kondo screening, and then the constancy of the signal intensity by a crystal electric field comparable to the Kondo hybridization strength [@Ramires_PRL2014]. The band structure we observe in ARPES, with the hybridization of about 4 meV of the same order as the crystal electric field $\sim 9$ meV, directly confirms the theoretical assumption in Ref. . Additionally, the absence of tangible sign of the Kondo-like hybridization within the experimental resolution of our ARPES measurements of [$\alpha$-YbAlB$_4$]{}, along with the absence of hyperfine satellites in its ESR signal [@Holanda_JoP2015], tends to reinforce this model.
Conclusions
===========
In conclusion, we report an ARPES study of both $\beta$- and [$\alpha$-YbAlB$_4$]{}, with the key observation of a Kondo-like hybridization between the crystal-field split Yb $f$ levels and an electronlike conduction band in [$\beta$-YbAlB$_4$]{}. The crystal-field splitting inferred from our ARPES data has a magnitude of $\sim 9$ meV, in good agreement with previous estimation [@Nevidomskyy_PRL2009]. Based on this agreement, we are able to use the theoretical arguments to identify the low-lying heavy band at $-4$ meV below $E_\mathrm{F}$ as the ground state $|J_z = \pm5/2\rangle$ doublet, and the band at $-13$ meV as the first excited crystal-field state, dominated by $|J_z = \pm3/2\rangle$ of the Yb multiplet.
The observed hybridization gives rise to a small electronlike QP pocket whose size and effective mass are consistent with quantum oscillations measurements [@OFarrell_PRL2009], together with a heavier Fermi sheet as the hybridized bands overlap in energy. Importantly, this QP exhibits a binding energy and an onset of coherence which both conform with the earlier indications of a reduced Kondo coherence temperature ${T_{\mathrm{coh}}}\approx 40$ K [@OFarrell_PRL2012; @Matsumoto_PRB2011]. Thus, as evidenced by the Hall effect measurements, it is indeed these Fermi sheets that are responsible for both Kondo lattice and QC behaviors below $T^*$ [@OFarrell_PRL2012].
Measurements of [$\alpha$-YbAlB$_4$]{} further support our conclusion. Indeed, in [$\alpha$-YbAlB$_4$]{}, we similarly observe the electronlike conduction band and the flat band at $-13$ meV below $E_\mathrm{F}$, which we also identify as the [$\left|\pm3/2\right>$]{} state. However, where [$\beta$-YbAlB$_4$]{} shows two anti-crossings of Yb $4f$ levels with the dispersive conduction band, the bands in the $\alpha$ phase do not show, within our resolution, any apparent sign of hybridization. This dichotomy strongly suggests that this hybridization plays a crucial role in the unconventional quantum criticality in [$\beta$-YbAlB$_4$]{}, whereas [$\alpha$-YbAlB$_4$]{} is not critical at ambient conditions.
While our observations cannot sharply infer the microscopic mechanism behind the non-Fermi-liquid behavior in [$\beta$-YbAlB$_4$]{}, they provide an important insight for future experimental and theoretical investigations. As we already mentioned, further ARPES measurements at lower temperature, and in wider regions of the Brillouin-zone, will have the potential to shed more light on this intriguing quantum critical behavior.
We thank S. Burdin, S. Watanabe and D. Malterre for helpful discussions. We thank P. Zhang for sharing his procedure to perform 2D curvature. We thank D. Evtushinsky and E. Rienks for their precious support on the 1-cubed end-station. We thank HZB for the allocation of synchrotron radiation beamtime and thankfully acknowledge the financial support by HZB. This work was supported by CREST, Japan Science and Technology Agency, Grants-in-Aid for Scientific Research (Grant No. 16H02209, No. 25707030, and No. 26105002), by Grants-in-Aid for Scientific Research on Innovative Areas “J-Physics" (Grant No. 15H05882 and No. 15H05883) and Program for Advancing Strategic International Networks to Accelerate the Circulation of Talented Researchers (Grant No. R2604) from the Japanese Society for the Promotion of Science. A.H.N. was supported by the grant No. DMR-1350237 from the U.S. National Science Foundation.
[28]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [**]{}, edited by () [****, ()](\doibase
10.1103/RevModPhys.79.1015) [****, ()](\doibase 10.1126/science.1197531) [****, ()](\doibase
10.1126/science.1262054) [****, ()](\doibase
10.7566/JPSJ.84.024710) [****, ()](\doibase
10.1038/nphys1002) [****, ()](\doibase 10.1143/JPSJ.78.084707) [****, ()](\doibase 10.1103/PhysRevLett.109.176404) [****, ()](\doibase
10.1103/PhysRevLett.109.086403) [****, ()](\doibase 10.1103/PhysRevB.83.235107) [****, ()](\doibase 10.7566/JPSJ.83.103708) [****, ()](\doibase
10.1103/PhysRevB.93.205126) [****, ()](\doibase 10.1103/PhysRevLett.109.176405) [****, ()](\doibase
10.1103/PhysRevB.84.125126) [****, ()](\doibase 10.1021/cm062244+) @noop [****, ()](\doibase 10.1063/1.3585113) [****, ()](http://stacks.iop.org/1402-4896/2004/i=T109/a=005) [****, ()](\doibase 10.1103/PhysRevLett.104.247201) [****, ()](\doibase 10.1103/PhysRevLett.102.077202) [****, ()](\doibase 10.1103/PhysRevB.85.241103) [****, ()](\doibase 10.1103/PhysRevLett.102.216402) @noop [**]{}, (, , ) [****, ()](\doibase 10.1016/j.physb.2004.12.041), [****, ()](\doibase 10.1103/PhysRevB.92.041114) [****, ()](\doibase 10.1103/PhysRevLett.112.116405) [****, ()](\doibase 10.1103/PhysRevLett.107.026402) [****, ()](http://stacks.iop.org/0953-8984/27/i=25/a=255601)
![\[fig:kzDep\][ a. Representation from the $c$ axis of the crystal structure of [$\beta$-YbAlB$_4$]{}(left) and [$\alpha$-YbAlB$_4$]{} (right). b. Brillouin-zone of [$\beta$-YbAlB$_4$]{}. c. Mapping in the (100)-plane, 50 meV below $E_\mathrm{F}$. Full black lines are the edges of the BZ. Purple circles show the position reached with a light of $h\nu = 6.994$ eV and $54$ eV. d. Dispersion at normal emission, i.e., along the $\left<001\right>$ direction. e. Laser-ARPES spectra on [$\beta$-YbAlB$_4$]{}; $h\nu = 6.994$ eV. ]{}](bYAB_kzDepGoldV2.eps){width="16cm"}
![\[fig:disp5K\][ Laser-ARPES on [$\beta$-YbAlB$_4$]{}. a. Stack of EDCs next to ${\widetilde{Z}}$, along the full gray line $A$ on panel (b). Green markers are peak position from EDCs, fitted by Voigt with a quadratic background and cut by a Fermi-Dirac step. Detailed fitting for $k_{\mathrm{y}} = -0.15,~0$ and $0.15$ Å$^{-1}$ are illustrated with Fig. S2 of the Supplemental Material [@SuppMat]. b. Fermi-surface mapping performed by tuning the polar angle. Sketch on the right illustrates the geometry of the measurements; The analyzer slit was oriented along the line $A$. The blue circle shows the little electronlike Fermi sheet, while the two red arrows show an outer Fermi sheet. Black lines are edges of the BZ. c. Curvature of the ARPES spectra measured along the slit of the analyzer \[full gray line on panel (b)\]. Green markers are peak position as of panel (a). The blue, red and pink lines are the results of the modeling based on the constant hybridization of the original bands (represented by dashed black lines). d. Same as (c) with a $\mathbf{k}$-dependent nodal potential. See Supplemental Material for the raw data (Fig. S3) and details of the hybridization models [@SuppMat]. e. Curvature of the cut along the $\left<010\right>$ direction \[dashed gray line $B$ on panel (b)\]. Green markers are peak position from EDC fit. Full blue, red and pink lines are guides to the eye. See Fig. S3 of the Supplemental Material for the raw data [@SuppMat]. ]{}](YAB_disp5K_RevV2.eps){width="12.3cm"}
![\[fig:Tdep\][ Temperature dependence on [$\beta$-YbAlB$_4$]{}. a. EDCs at $k_y = 0$Å$^{-1}$ from $T=5$ K to $60$ K. b. EDCs at $k_y = 0.17$Å$^{-1}$ from $T=5$ K to $60$ K. \[The peak’s intensity difference with EDC in Fig. \[fig:disp5K\](a) is due to a slightly different orientation along the slit, as temperature dependence was performed on a different sample – see Fig. S1 of the Supplemental Material for the corresponding color scale [@SuppMat].\] c. Intensity, relative to the value at 60 K, of the inner (outer) peak from EDCs divided by the Fermi-Dirac step and integrated along the shade area of panel (a) \[(b)\]. ]{}](bYAB_TdepV2.eps){width="6cm"}
![\[fig:aYAB\][ Laser-ARPES on [$\alpha$-YbAlB$_4$]{}. a. Curvature of the spectra along the $\left<100\right>$ direction. Open and full circles are peak positions, extracted, respectively, from EDCs and MDCs. See Fig. S3 of the Supplemental Material for the raw data [@SuppMat]. b. MDC stack of previous panel. Black markers are peak positions for the electronlike band. They were obtained by fitting MDCs by Voigts with a constant background. ]{}](aYAB_Gold_Rev.eps){width="7.75cm"}
![\[fig:bVSa\][ a. MDC and EDC fits (open and full markers) from [$\alpha$-YbAlB$_4$]{} compared to EDC fit on [$\beta$-YbAlB$_4$]{} (blue, red and pink lines). The broken black lines denote the unhybridized electronlike conduction band and Yb $4f$ flat bands from the same model as in Fig. \[fig:disp5K\]. b. EDCs at $k = 0$; $\pm0.15$Å$^{-1}$ from [$\beta$-YbAlB$_4$]{} (red) and [$\alpha$-YbAlB$_4$]{} (green). In the left panel is the raw, while the right panel shows the EDCs divided by the effective Fermi-Dirac step. Open arrows indicate the [$\left|\pm5/2\right>$]{} level, while full arrows indicate the [$\left|\pm3/2\right>$]{} level. ]{}](YAB_bVSaGoldV2.eps){width="8cm"}
**SUPPLEMENTAL MATERIAL**
**Light polarization**
![\[fig:pola\][ Laser-ARPES spectra measured at normal emission, with linear polarized (left) and circular polarized (right) light. For both polarization, the 3 same heavy bands can be observed as in the main text, with an anticrossing giving rise to the little quasi-particle at about 4 meV. Difference in the momentum distribution of the intensity compared to the main text is due to a slightly different orientation of the sample. ]{}](bYAB_PolarizationSupp.eps){width="10cm"}
**EDCs fits**
![\[fig:voitFits\][ Fit of EDCs from laser-ARPES on [$\beta$-YbAlB$_4$]{} (see Fig.2 of the main text) at $k_\mathrm{y} = -0.15,~0$ and $0.15$ Å$^{-1}$, from left to right. The fit uses Voigt peaks with a quadratic background cut by a effective Fermi step. Resulted Voigts are visible at the bottom of each graph, while their contribution to the fit represented with colored area. Green areas depict contribution of the quadratic background. Top curves are residual from the fitting. ]{}](bYAB_EDCs_VoigtFit.eps){width="15.5cm"}
**Raw Spectra**
![\[fig:rawData\][ Raw spectra of the 2D curvature spectra shown in the main text. ]{}](YAB_rawData.eps){width="12cm"}
**Hybridization potential**\
Models in the main text use a simple hybridization Hamiltonian, which can be written as follow: $$\mathcal{H}_{\mathrm{hyb}}=
\left({ \begin{array}{ccc}
\tilde{\epsilon}_{f,5/2} & 0 & V_k \\
0 & \tilde{\epsilon}_{f,3/2} & V_k \\
V_k^* & V_k^* & \epsilon_{c,k}
\end{array} }\right)$$ With $\tilde{\epsilon}_{f,5/2}$ and $\tilde{\epsilon}_{f,3/2}$($=\tilde{\epsilon}_{f,5/2}-9$ meV), the renormalized energies of the Yb [$\left|\pm5/2\right>$]{} and [$\left|\pm3/2\right>$]{} $f$-levels respectively; $\epsilon_{c,k} = \epsilon_{c0}+\frac{\hbar^2}{m_c}\frac{1}{a^2}
\left(1-\cos(ak_\mathrm{y})\right)$, the conduction band dispersion, with $a = 7.308$ Å, the lattice parameter, $m_c$ its relative mass, and $\epsilon_{c0}$ the bottom of the band; and finally $V_k$, the hybridization potential.
Parameters values are given in this table for both the constant and quadratic hybridization toy models: $$\begin{array}{|c|c|c|}
\hline
~ & \mathrm{constant} & \mathrm{\mathbf{k}-dependent} \\
\hline
m_c & 3 m_e & 3 m_e \\
\epsilon_{c0} & -17~\mathrm{meV} & -20~\mathrm{meV} \\
\tilde{\epsilon}_{f,5/2} & -5.5~\mathrm{meV} & -4~\mathrm{meV} \\
\tilde{\epsilon}_{f,3/2} & -14.5~\mathrm{meV} & -13~\mathrm{meV} \\
V_k & 4~\mathrm{meV} & C\sin^2(ak_\mathrm{y}) \\
\hline
\end{array}$$ with $m_e$ the free electron mass, and $C = 16$ meV.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We investigate the reactions $p n \to d \omega$ and $p n\to d \phi$ close to the corresponding thresholds. The $S$-wave amplitudes are calculated within the framework of the two-step model which is described by a triangle graph with $\pi$, $\rho$ and $\omega$ mesons in the intermediate state. The cross sections of the reactions $p n
\to d \omega$ and $p n\to d \phi$ are predicted to be significantly larger than the cross sections of the corresponding reactions $p p
\to pp \omega$ and $p p\to pp\phi$ at the same values of the c.m. excess energy $Q$. The ratio of the yields of $\phi$ to $\omega$ is found to be $(30 \pm 7) \times 10^{-3}$.
[*PACS*]{} 25.10.+s; 13.75.-n
Meson production; Omega; Phi; OZI rule; pn.
address:
- 'Institute for Nuclear Research, 60th October Anniversary Prospect 7A, 117312 Moscow, Russia'
- 'Institute of Theoretical and Experimental Physics, B. Cheremushkinskaya 25, 117259 Moscow, Russia'
- 'Forschungszentrum Jülich, Institut für Kernphysik, 52425 Jülich, Germany'
author:
- 'V.Yu. Grishina'
- 'L.A. Kondratyuk'
- 'M. Büscher'
title: ' $\omega$- and $\phi$-meson production in $p n \to d M$ reactions near threshold and OZI-rule violation'
---
,
and
Introduction
============
It is well known (see e.g. [@Lipkin; @Ellis1; @Ellis2]) that the ratio of the $\phi /\omega$ yields $$R = \frac{\sigma_{A+B\to \phi X}}{\sigma_{A+B\to \omega X}}\ ,
\label{eq:phitoomega}$$ is a particularly sensitive probe of the OZI rule [@Okubo]. Using the standard value for the deviation $\delta = \theta -
\theta_{\mathrm{i}} = 3.7^{\circ}$ from the ideal SU(3)$_f$ mixing angle $\theta_{\mathrm{i}} = 35.3^{\circ}$ we have $R/f = 4.2\times
10^{-3}$ [@Ellis2], where $f$ is the ratio of the phase-space factors. However, experimental data show an apparent excess of $R/f$ above the standard value which varies from $(10-30)\times 10^{-3}$ in $\pi N$ and $NN$ collisions to $(100-250)\times 10^{-3}$ in $\bar{N}N$ annihilation at rest and in flight (see e.g. the discussion in [@Ellis2]). In Ref.[@Ellis2] the large excess of $R$ in $pp$ and $\bar{p}p$ collisions over the prediction by the OZI rule was treated in terms of “shake-out” and “rearrangement” of an intrinsic $\bar{s}s$ component in the nucleon wave function. On the other hand, in papers [@Locher; @Buzatu] the strong violation of the OZI rule in $\bar{p}p$ annihilation at rest was explained in terms of hadronic intermediate $K\bar{K}^{\ast}$ states which might create $\phi$ mesons.
Another argument in favor of a large admixture of hidden strangeness in nucleons was related to an apparently large contribution of the $\phi$-meson into the isoscalar spectral function which through the dispersion relation defines the isoscalar nucleon form factor (see Ref.[@Jaffe]). However, as it was shown later (see [@Meisner] and references therein), the main contribution to the isoscalar spectral function near 1 GeV stems from correlated $\pi \rho$ exchange which does not involve strange quarks.
Therefore, the question whether there is a large admixture of hidden strangeness in nucleons seems to be unclarified. Thus, it is important to investigate such reactions where uncertainties in the interpretation of $\omega$ and $\phi$ production in terms of intermediate hadronic states are comparably small. In this paper we argue that a good choice in this respect is the reaction $$pn\to dM\ .
\label{eq:Vd}$$ Here and below $M$ denotes the vector mesons $\omega$ and $\phi$.
We analyze contributions of hadronic intermediate states into the $S$-wave amplitudes of the reactions $p n \to d \phi$ and $p n \to
d\omega$ within the framework of the two-step model (TSM) described by triangle graphs with $\pi$-, $\rho$- and $\omega$-meson exchanges. Previously this model was applied to the description of the Pontecorvo reactions $\bar{p} d \to p M$ (see, e.g., Ref.[@kondrat1; @kondrat2]). In a recent paper (see Ref.[@Grishina]) it was demonstrated that the TSM can also describe the cross section of the reaction $p n \to d \eta$ near threshold with a reasonable choice of the coupling constants and cut-off parameters for $\pi$-, $\rho$- and $\omega$-meson exchanges. To predict the cross sections of the reactions $p n \to d\omega$ and $p n \to d \phi$ we use a similar approach and the same set of parameters for the $MNN$ coupling constants and cut-off parameters. Note that if the $\phi$ and $\omega$ yields will be measured in reaction (\[eq:Vd\]) near threshold (which e.g. can be done at COSY-Jülich), the results can be useful for a better understanding of the OZI-rule violation dynamics. For example, any significant deviation from the prediction of the two-step model could be an evidence for the above mentioned “shake out” or “rearrangement” of an intrinsic $\bar{s}s$ component in the nucleon wave function.
Note that recent measurements of the $\phi /\omega$ ratio in the reaction $pd\to ^3\!\! HeX$ (performed at SATURNE II [@Wurzinger1; @Wurzinger2]) yield $$R/f = \left(63 \pm 5\ ^{+27}_{-8} \right)\times 10^{-3}
\label{eq:wurzinger}$$ which is also clearly above the expectation 4.2$\times 10^{-3}$. However the dynamics of the reaction $pd\to ^3\!\! HeX$ is yet not well understood. According to [@Wilkin1] the two-step model underestimates the SATURNE data by a factor 2, while according to [@Uzikov] the discrepancy of the two-step model with the data might be even larger when spin-effects are taken into account.
Experiments on $\omega$ and $\phi$ production in the reaction $ p p
\to p p M$ close to threshold were performed by the SPES3 and DISTO collaborations at SATURNE [@SPES3; @DISTO] (see also the calculations of $\omega$ production in [@Speth]). According to the DISTO data the ratio of the $\phi/\omega$ production cross sections at 2.85 GeV is $\sigma_{\mathrm{tot}}(p p \to p p \phi)$/ $\sigma_{\mathrm{tot}} (p p \to pp\omega) = (3.7 \pm 1.3)\times
10^{-3}$. Introducing corrections for phase-space effects the authors of Ref.[@SPES3] found that in this case the $\phi/\omega$ ratio is $(49 \pm 26)\times 10^{-3}$. Note that near threshold the dynamics of the reactions $ p p \to p p M$, $ p n \to p n M$ and $ p n
\to d M$ are different because the first one is constrained by the Pauli principle and the two protons in the final state should be in a $^1 S_0$ state. In the third case the final $pn$ system is in the $^3
S_1$ state while in the second case it can be in both states. Therefore, a possible violation of the OZI rule is expected to be different in all those cases.
Finally, another interesting point is that within the framework of the line-reverse invariance (LRI) assumption the reaction $pn \to dM$ can be related to the Pontecorvo reaction $\bar{p}d \to MN$. The data from the OBELIX and Crystal-Barrel collaborations result in a $\phi /
\omega $ ratio of about $(230 \pm 60)\times 10^{-3}$ [@Sapozhnikov; @Crystal]. Therefore, if LRI is applicable we expect the violation of the OZI rule in the reaction $pn \to dM$ to be much larger than it is predicted by the two-step model, which assumes the dominance of the hadronic intermediate states.
The paper is organized as follows. In Sect. \[sec:twostep\] we derive the amplitudes of the reactions $pn\to d\phi$ and $pn\to
d\omega$ near threshold within the framework of the two-step model. In Sect. \[sec:results\] we discuss the choice of parameters and present the results of calculations. Sect. \[sec:conc\] contains our conclusions.
The non-relativistic two-step model for the reaction $pn\to d
M$ {#sec:twostep}
=============================================================
The triangle diagrams describing the TSM are shown in Fig. \[fig:tsm\]. Besides the $\pi$ exchange we also take into account $\rho$ and $\omega$ exchanges.
In the beginning let us consider the $\pi^0$-exchange term. In order to preserve the correct structure of the amplitude under permutations of the initial nucleons (which should be symmetric in the isoscalar state) the amplitude is written as the sum of the $t$- and $u$-channel contributions in the following form $$T_{pn \to dM}^{\pi}(s,t,u)= A_{pn \to dM}^{\pi}(s,t)+ A_{pn \to
dM}^{\pi}(s,u) \ ,
\label{Atu}$$ where $M$ is the vector meson $\omega$ or $\phi$. $s=(p_1+p_2)^2$, $t=(p_3-p_1)^2$, and $u=(p_3-p_2)^2$ where $p_1$, $p_2$, $p_3$, and $p_4$ are the 4-momenta of the proton, neutron, meson $M$ and deuteron, respectively. Since we are interested in the calculation of the cross section of reaction (\[eq:Vd\]) near threshold where the momenta of the deuteron and the meson are comparatively small, we can use a non-relativistic description of those particles by neglecting the 4th components of their polarization vectors. The relative motion of nucleons inside the deuteron is also treated non-relativistically. Then one can write the two terms on the right hand side of Eq. (\[Atu\]) as follows (see also [@Grishina]) $$\begin{aligned}
A_{pn\to dM}^{\pi} (s,t) &=& \frac{f_{\pi}}{m_{\pi}}
\varphi^{T}_{\lambda_{2}} ({\vec{p}}_2)\ (-i \sigma_2)
{\vec{\sigma}}\cdot{\vec{M}}^{\pi}({\vec p}_1)\ {\vec{\sigma}}\cdot
{\vec{\epsilon}} ^{\ast}_{d}\ {\vec{\sigma}}\cdot
{\vec{\epsilon}} ^{\ast}_M\ \varphi_{\lambda_1} ({\vec{p}}_1)
\ \times \nonumber \\
&& A_{\pi^0 N \to M N}(s_1,t) \ ,
\label{AVdt}\end{aligned}$$
$$\begin{aligned}
A_{pn\to d M}^{\pi} (s,u) &=& \frac{f_{\pi}}{m_{\pi}}
\varphi^{T}_{\lambda_{1}} ({\vec{p}}_1)\ (-i \sigma_2)
{\vec{\sigma}}\cdot{\vec{M}}^{\pi}(- {\vec p}_1 )\ {\vec{\sigma}}\cdot
{\vec{\epsilon}} ^{\ast}_{d}\ {\vec{\sigma}}\cdot
{\vec{\epsilon}} ^{\ast}_{M}\ \varphi_{\lambda_{2}} ({\vec{p}}_2)
\ \times \nonumber \\
&& A_{\pi^0 N \to M N}(s_1,u) \ ,
\label{AVdu}\end{aligned}$$
where $\vec{\epsilon}_d$ and $\vec{\epsilon}_M$ are the polarization vectors of the deuteron and the meson; $\varphi_{\lambda}$ are the spinors of the nucleons in the initial state, $m_{\pi}$ and $f_{\pi}$ are the pion mass and $\pi NN$ coupling constant. The vector function $\vec{M}^{\pi}({\vec p}_1 )$ is defined by the integral $$\begin{aligned}
&\vec{M}^{\pi}({\vec p}_1 ) &= \sqrt{2\,m} \int ({\vec{k}}+{\vec
p}_1)\, \Phi_{\pi}({\vec{k}},{\vec{p}_1})\Psi_d(\vec{k})\,
\frac{\mathrm{d}^3k}{(2\pi)^{3/2}} \ , \label{M} \\
&\Phi_{\pi}({\vec{k}},{\vec{p_1}}) &=
\frac{F_{\pi}(q^2)}{q^2-m_{\pi}^2} \ ,
\label{ff} \end{aligned}$$ which contains the deuteron wave function $\Psi_d(\vec{k})$ and the form factor at the $\pi NN$ vertex $F_{\pi}(q^2)$. Other kinematical quantities which are also dependent on the momenta ${\vec p_1}$ and ${\vec k}$ are defined as follows $$\begin{aligned}
&q^2 &= m_{\pi}^2 - \delta_0(\vec{k}^2+
\beta(\vec{p}_1)) - 2{\vec {p}}_1 {\vec k} , \
\vec{q} = \vec{k}+\vec{p}_1 \ , \nonumber \\
&\beta(\vec{p}_1) &=
({\vec{p}_1}^{\,2}+m_{\pi}^2-T_1^2)/\delta_0\ , \
\delta_0 = 1+T_1/m, \ T_1= \sqrt{{\vec{p}_1}^2 + m^2} - m \ .
\nonumber\end{aligned}$$ with $m$ being the nucleon mass.
Near threshold we take into account only the $S$-wave part of the amplitude of the elementary reaction $\pi N \to M N$. Deriving Eqs.(\[AVdt\],\[AVdu\]) we use the following spin structure of the $\pi^0 N \to M N$ amplitude $$\begin{aligned}
\lefteqn{ \langle p_3^{\prime}\,\lambda^{\prime}_{3}; p_4^{\prime}
\lambda_4^{\prime} | \hat{T}_{\pi N\to NM} |
p_1^{\prime};p_2^{\prime} \lambda_2^{\prime}\rangle =} \nonumber \\
&&\varphi_{\lambda^{\prime}_{4}}^{\ast}(\vec{p}_4^{\,\prime})\
\vec{\epsilon}_{\lambda^{\prime}_{3}}^{\ast \,(M)} \cdot
\vec{\sigma}\
\varphi_{\lambda^{\prime}_{2}}^{\ast}(\vec{p}_2^{\,\prime})\ A_{\pi
N\to NM}(s_1,t_1)\ ,
\label{eq:AVN}\end{aligned}$$ where $p_1^{\prime}$, $p_2^{\prime}$, $p_3^{\prime}$ and $p_4^{\prime}$ are the 4-momenta of the $\pi$ meson, the initial nucleon, the final nucleon and the vector meson, respectively. The $\lambda_i^{\prime}$ are the spin projections of the particles, $\vec{\epsilon}^{\,(V)}$ is the polarization vector of the vector meson and $s_1=(p_1^{\prime} +
p_2^{\prime})^2=(p_3^{\prime}+p_4^{\prime})^2$, $t_1=(p_1^{\prime} -
p_4^{\prime})^2=(p_2^{\prime}-p_3^{\prime})^2$.
The invariant amplitude is normalized to the total cross section as follows $$| A_{\pi^0 N \to M N}(s_1,t)|^2 = | A_{\pi^0 N \to M N}(s_1,u)|^2 =
{\frac8 3} \pi s_1
\frac{p_{\pi}^{\mathrm{cm}}}{p_M^{\mathrm{cm}}}\sigma_ {\pi^- p \to
M n}$$ where $s_1$ is the invariant mass squared of the $Mn$ system.
It was shown in Ref.[@Grishina] that apart from the $\pi$-exchange contributions heavier vector-meson exchanges — especially of $\rho$ mesons — are important for the case of the reactions $p n \to d \eta
$ and $p n \to d \eta^{\prime}$. In our case the amplitudes for the vector-meson exchanges can be written in the form $$\begin{aligned}
A^V_{pn\to dM}(s,t) &&= \frac{G_V}{2\, m}\varphi^T_{\lambda_2}
(\vec{p}_2) (-i\, \sigma_2) \cdot A_{V^0 N\to
MN}(s_1,t) \ \times \nonumber\\
&&\left\{ \vec{M}_1^V(\vec{p}_1)\cdot\vec{\epsilon}_M^{\ast}
\vec{\sigma} \cdot \vec{\epsilon}_d^{\ast} +
\vec{M}_2^V(\vec{p}_1)\cdot\vec{\epsilon}_d^{\ast} \vec{\sigma}
\cdot \vec{\epsilon}_M^{\ast} \ - \right. \nonumber\\
&&\left. \mbox{ }\vec{\sigma}\cdot \vec{M}_2^V(\vec{p}_1)\,
\vec{\epsilon}_d^{\ast} \cdot \vec{\epsilon}_M^{\ast} + i\left[
\vec{M}_2^V(\vec{p}_1)\times \vec{\epsilon}_d^{\ast}
\right]\cdot \vec{\epsilon}_M^{\ast}
\right\} \varphi_{\lambda_1}(\vec{p}_1) \label{AVt} \\
A^V_{pn\to dM}(s,u) &&= \frac{G_V}{2\, m}\varphi^T_{\lambda_1}
(\vec{p}_1) (-i\, \sigma_2) \cdot A_{V^0N\to
MN}(s_1,u) \ \times \nonumber\\
&&\left\{ \vec{M}_1^V(-\vec{p}_1)\cdot\vec{\epsilon}_M^{\ast}
\vec{\sigma} \cdot \vec{\epsilon}_d^{\ast} +
\vec{M}_2^V(-\vec{p}_1)\cdot\vec{\epsilon}_d^{\ast} \vec{\sigma}
\cdot
\vec{\epsilon}_M^{\ast} \ - \right. \nonumber\\
&&\left. \mbox{ } \vec{\sigma}\cdot \vec{M}_2^V(-\vec{p}_1)\,
\vec{\epsilon}_d^{\ast} \cdot \vec{\epsilon}_M^{\ast} + i\left[
\vec{M}_2^V(-\vec{p}_1)\times \vec{\epsilon}_d^{\ast}
\right]\cdot \vec{\epsilon}_M^{\ast} \right\}
\varphi_{\lambda_1}(\vec{p}_1) \ , \label{AVu}\end{aligned}$$ where $$\begin{aligned}
\vec{M}_1^V(\vec{p}_1)& =& \sqrt{2\, m}
\int [ (\vec{k}-\vec{p}_1) +
\frac{\vec{k}^2-\vec{p}^2_1}{m^2_V} (\vec{k}+\vec{p}_1)]
% + 2(1+\kappa_V)(\vec{k}+\vec{p}_1)] \ \times \nonumber \\&&
\, \Phi_V(\vec{k},\vec{p}_1)
\Psi_d^{\ast}(\vec{k}) \frac{\mathrm{d}^3k}{(2\pi)^{3/2}} \end{aligned}$$ and $$\begin{aligned}
\vec{M}_2^V(\vec{p}_1) &=& \sqrt{2\, m} \int
(1+\kappa_V)(\vec{k}+\vec{p}_1)\, \Phi_V(\vec{k},\vec{p}_1)
\Psi_d^{\ast}(\vec{k}) \frac{\mathrm{d}^3k}{(2\pi)^{3/2}} \ .\end{aligned}$$ The function $\Phi_V(\vec{k},\vec{p}_1)$ describes the product of the $V$-meson propagator $(q^2 - M_V^2)^{-1}$ and the form factor at the $VNN$ vertex $F_V(q^2)$. It is defined by Eq. (\[ff\]) where $m_{\pi}^2$ should be substituted by $m_V^2$. $G_V$ and $\kappa_V G_V$ are the vector and tensor coupling constants respectively.
The general spin structure of the $V N \to MN$ amplitude near threshold has the following form $$\begin{aligned}
\lefteqn{
\langle p_3^{\prime}\,\lambda^{\prime}_{3}; p_4^{\prime}
\lambda_4^{\prime} | \hat{T}_{V N\to NM} | p_1^{\prime}
\lambda^{\prime}_{1} ;
p_2^{\prime}\lambda_2^{\prime}\rangle =} \nonumber \\
&\varphi_{\lambda^{\prime}_{4}}^{\ast}(\vec{p}_4^{\,\prime})\
&\left( \vec{\epsilon}_{\lambda^{\prime}_{3}}^{\ast \,(M)} \cdot
\vec{\epsilon}_{\lambda^{\prime}_{1}}^{\,(V)} \
A_{V N\to NM}(s_1,t_1)\ + \right.\nonumber \\
&&\left.i \left[ \vec{\epsilon}_{\lambda^{\prime}_{3}}^{\ast \,(M)}
\times \vec{\epsilon}_{\lambda^{\prime}_{1}}^{ \,(V)}\right]
\cdot \vec{\sigma} \ B_{V N\to NM}(s_1,t_1)\right)
\varphi_{\lambda^{\prime}_{2}}^{\ast}(\vec{p}_2^{\,\prime})\ ,
\label{eq:AVM}\end{aligned}$$
where the notations are similar to the ones in Eq.(\[eq:AVN\]). Two invariant amplitudes $A_{V N\to NM}(s_1,t_1)$ and $B_{V N\to
NM}(s_1,t_1)$ are necessary to describe two possible transitions $\left(\frac{1}{2}\right)^- \to \left(\frac{1}{2}\right)^-$ and $\left(\frac{3}{2}\right)^- \to \left(\frac{3}{2}\right)^-$. It is known from the data on Compton scattering (see, e.g., [@Erbe]) that the spin-flip amplitude $B_{\gamma N\to \gamma N}(s_1,t_1)$ is small as compared with the non spin-flip amplitude $A_{\gamma N\to
\gamma N}(s_1,t_1)$ except in the $\Delta$-resonance region (see, e.g., [@Erbe]). Following the arguments of the Vector-Dominance Model (VDM) we assume that this amplitude is also small in our case and take into account only the first non spin-flip term in Eq.(\[eq:AVM\]).
Note that the amplitudes $A^{\pi}$ and $A^{\rho}$ correspond to the exchange of neutral $\pi$ and $\rho$ mesons only (see the left diagrams in Fig. \[fig:tsm\]). To take into account also the charged $\pi$ and $\rho$ exchanges we have to multiply amplitude (\[Atu\]) by a factor 3. Of course in the case of $\omega$ exchange such a factor is not necessary.
Therefore, the differential cross section of reaction (\[eq:Vd\]) can be written as $$\begin{aligned}
\lefteqn{ \frac{\d\sigma_{pn\to dM}}{\d t} = }
\nonumber \\
&& \frac{1}{64\,\pi s}\ \frac{1}{(p_{1}^{\mathrm{cm}})^2}\ F(I)\
\overline{|A_{pn \to dM}(s,t) + A_{pn \to dM}(s,u) |^2} \ .
\label{eq:sigmaVd}\end{aligned}$$ where the isospin factor $F(I)$ is equal to 9 for isovector exchanges ($\rho$ and $\pi$) and 1 for isoscalar exchange ($\omega$).
Choice of parameters and results of calculations {#sec:results}
================================================
We assume the form factors $F_{\pi}(q^2)$ and $F_{V}(q^2)$ to be of monopole type. Recent QCD lattice calculations [@QCD] suggest that the cut-off in the pion form factor should be quite soft $\Lambda_\pi \simeq 0.8 $ GeV/c (see also Refs. [@Coon; @Speth2]). Of course such a soft pion form factor suppresses pion exchange and contributions of heavier meson exchanges become more important. This for example was demonstrated in Ref. [@Grishina] where it was found that the $\rho$-exchange contribution in the reactions $p n \to
d \eta$ and $p n \to d \eta^{\prime}$ is significant. Here also $\Lambda_\pi = 0.8 $ GeV/c is used.
The coupling constants and vertex form factors for $\rho$ and $\omega$ mesons are taken from the full Bonn $NN$ potential [@Holinde]: $G_{\rho}^2/4\pi = 0.84$, $\kappa_\rho =
6.1$, $G_{\omega}^2/4\pi = 20$, $\kappa_{\omega} = 0$ and $\Lambda_\rho$ = 1.4 GeV/c, $\Lambda_\omega$ = 1.5 GeV/c.
For the deuteron wave function we take the parameterization from Ref.[@Lacomb] and neglect the $D$-wave part. As it was demonstrated in Ref.[@kondrat2] for the case of the reaction $\bar
p d \to M n$ (where the same structure integrals (\[M\]) for $\pi$, $\rho$ and $\omega$ exchanges occur) the $D$-wave term of the deuteron wave function gives a negligibly small contribution compared to the $S$-wave term.
To define the amplitudes $\pi N \to M N$ we use the following values of the $S$-wave cross sections (taken from Ref.[@Binnie]):\
$\sigma_{\pi^- p \to \omega n} = (8.3 \pm 0.07) p^M_{\mathrm{cm}}\
\mu$b and $\sigma_{\pi^- p \to \phi n} = (0.29 \pm 0.06)
p^M_{\mathrm{cm}}\ \mu$b$\\$ ($p^M_{\mathrm{cm}}$ in MeV/c). The experimental data show that the angular distribution in the reaction $\pi^- p \to n\omega$ is isotropic and the $S$-wave is dominant at least up to $k_V^{\mathrm{cm}}(s_1) = 260$ MeV/c (see the comment on p.2805 in [@Binnie]). We ignore an apparent suppression of the $S$-wave amplitude very close to threshold ($k_V^{\mathrm{cm}}(s_1)
\leq 80 - 100$ MeV/c), reported in Ref.[@Binnie], because according to Ref.[@Hanhart] this effect has a kinematical origin.
The contributions from the $\rho$ and $\omega$ exchanges are calculated using the vector-dominance model (VDM) prediction for the amplitude $\rho N\to \omega(\phi) N$ and assuming that for non-diagonal cases $A_{\omega N\to M N} \approx A_{\rho^0 N\to M N}$. We derive the $S$-wave $\gamma N\to \omega N$ amplitude from the ABBHHM data at $E_{\gamma} = 1.3$ GeV (see Ref.[@Erbe]) using a value of the cross section of the reaction $\gamma p\to \omega p$ equal to 5.6 – 7.8 $\mu$b. This would give the $\rho p\to \omega p$ cross section of about 2.7 $\pm$ 0.5 mb at low energies. The ratio of the $\gamma p\to \phi p$ and $\gamma p\to \omega p$ amplitudes squared was found from the data at $s$ = 5 – 6 GeV$^2$ to be 0.06 – 0.07. Then we assumed that it is the same for the the case of the reactions $\rho p\to \phi p$ and $\rho p\to \omega p$. For the elastic $\omega
N$ scattering cross section at low energies we took the value 15 mb which was evaluated in Ref.[@Lykasov] within the sigma-exchange model and is in agreement with previous estimations made using the Quark Model.
Since the relative phases of the different contributions are not known we calculate the cross section of the reaction $pn\to dM$ as the incoherent sum $$\sigma_{pn\to dM} = N [\sigma^{(\pi)}+\sigma^{(\rho)}+
\sigma^{(\omega)}] \ .$$
In Fig.\[fig:eta\] taken from Ref.[@Grishina] we show how the TSM (with the same coupling constants and cut-off parameters for $\pi$, $\rho$ and $\omega$ exchanges and the $S$-wave amplitudes $Vp \to \eta p$ and $ Vp \to \eta^{\prime}p$ estimated using VDM from the photo-production data) describes the experimental data on the reaction $p n \to d \eta$. The cross section of the reaction $pn \to d
\eta$ is presented as a function of the c.m. excess energy $Q$. The dashed curve shows the $\pi$-exchange contribution alone whereas the dash-dotted curve describes the sum of $\pi$, $\rho$, and $\omega$ exchanges. The solid curve includes all contributions ($\pi$, $\rho$, $\omega$) multiplied with a normalization factor $N =0.68$ in order to take into account effects from the initial state interaction (ISI). The data points for are taken from Refs.[@Cal1] (open circles) and [@Cal2] (filled circles). The reduction factor appeared to be not very different from the prediction of the ISI effect within a simple model which assumes the dominant contribution from the on-shell rescattering [@Nakayama] and gives $\lambda_{\mathrm{ISI}} \simeq 0.5$.
As we see from Fig.\[fig:eta\] pion exchange calculated with the soft cut-off parameter cannot describe the $\eta$-production data and the contribution from heavier meson exchanges (and especially of $\rho$ [@Grishina]) is quite important.
In Figs.\[fig:om\] and \[fig:phi\] we present the predictions of the TSM for the cross sections of the $\omega$ and $\phi$ production. The contribution of pion exchange is shown by the dashed curves. The lower and upper curves show the minimal and maximal values of the $\pi$-exchange contribution demonstrate which follow from the experimental errors of the elementary cross sections. The dash-dotted curves describe the sum of $\pi$-, $\rho$- and $\omega$-exchange contributions. The solid curves represent the results including all contributions ($\pi$, $\rho$, $\omega$) multiplied with the same normalization factor $N =0.68$ as in the case of $\eta$- production in order to take into account effects from ISI. It is clearly seen that similar to the case of $\eta$ production the $\rho$-exchange contribution to the cross sections of the reactions is very significant. The relative contribution of $\pi$ exchange is about 20 % in the case of $\omega$ production and is almost 2 times less in the case of $\phi$ production. The $\omega$ exchange is more important in the case of $\omega$ production where it gives about 20%; in the case of $\phi$ productions its relative contribution is about 5%.
The cross sections of the reactions $p n \to \omega d$ and $p n \to
\phi d$ can be parameterized as follows $$\label{pndM}
\sigma_{pn\to d M} \approx D_{M}\sqrt{Q}\:,$$ where $D_{\omega} =( 2.7\pm 0.3)$ $ \mu$b/MeV$^{1/2}$ and $D_{\phi} =
(0.09 \pm 0.02)$ $\mu$b/MeV$^{1/2}$ . At very low Q which are of the order of the resonance width each cross section might be a little larger because of the finite widths of the $\omega$ and $\phi$ [@SPES3].
In Fig.\[fig:om\] we show also experimental data on the near-threshold production of $\omega$ mesons in the $pp \to pp \omega$ reaction [@SPES3]. Near threshold the predicted cross section of $\omega$ production with the deuteron in the final state is much higher than that of the reaction the $pp \to pp \omega$. This is very similar to the case of $\eta$ production (see , e.g., [@Cal1; @Cal2]) and is related to isospin and phase-space factors (see, e.g., [@Wilkin98]).
Let us discuss the relation between $\sigma(pp\to pp \omega)$ and $\sigma(pn\to d \omega)$ near threshold in more detail. Fäldt and Wilkin [@Faldt] proposed the following parameterization of the cross section of the reaction $pp \to pp M$ near the threshold $$\label{FW}
\sigma_{pp\to pp M} = C_{M}\left(\frac{Q}{\epsilon}\right)^{\!{2}}
\left(1+\sqrt{1+Q/\epsilon}\,\right)^{\!-2}\:.$$ This formula takes into account the strong final state interaction of two protons including also Coulomb distortion with $\epsilon\approx
0.45$ MeV. For $\eta$ and $\omega$ production we have $C_{\eta} = (110
\pm 20)$ nb and $C_{\omega} = (37 \pm 8)$ nb [@SPES3]. At $Q$=15 MeV we have $\sigma(pp\to pp \eta) \approx 2.6\ \mu$b ($\sigma(pp\to
pp \omega) \approx 1\ \mu$b) which is 15(10) times less than the cross section of the reaction $pn\to d \eta$ ($pn\to d \omega$). Note that in line with suggestions by Wilkin (see, e.g., [@Wilkin98]) the ratios $\sigma(pn\to d \eta)/\sigma(pp\to pp \eta)$ and $\sigma(pn\to d \omega)/ \sigma(pp\to pp \omega)$ are, in fact, not very different.
The reaction $ p p \to p p \omega$ near the threshold was also analyzed within the framework of the meson-exchange model in Ref.[@Speth]. Adjusting the cut-off parameter of the form factor to the low energy data the authors of Ref.[@Speth] calculated the cross section of the reaction $ p p \to p p \omega$ for proton incident energies up to 2.2 GeV. This model predicts a cross section of about 15–20 $\mu$b at $Q \approx 100 $ MeV which is still not very different from parameterization (\[FW\]). If parameterizations (\[pndM\]) and (\[FW\]) would be valid up to $Q= 1$ GeV then the cross section of the reaction $\sigma(pp\to pp \omega)$ would reach the same value as the cross section of the reaction $pn\to d \omega$ only at 900 MeV. Of course those formulas can not be valid up to such large values of Q. Estimations within the framework of the Quark-Gluon String Model shows that the cross section of the reaction $pn\to d
\omega$ can reach maximum of about 30–50 $\mu$b at $Q$ = 100–200 MeV and then will start to fall (see [@proposal]). According to the parameterization of Ref.[@Sibirtsev] the cross section of the reaction $\sigma(pp\to pp \omega)$ reaches the value of $30\ \mu$b at $Q\approx 200$ MeV. Therefore we can expect that in a rather broad interval of $Q$ (at least up to about 100–150 MeV) the cross section of the reaction $pn\to d \omega$ will be larger than the cross section of the reaction $\sigma(pp\to pp \omega)$. This gives quite a good chance that the reaction $pn\to d \omega$ can be detected using missing mass method at COSY by measuring the forward deuteron and spectator proton in the reaction $p d \to d \omega p_{\mathrm{sp}}$.
For the case of $\phi$ production we also expect that near threshold the cross section of the reaction $pn\to d \phi$ will be larger than the cross section of the reaction $pp \to pp \phi$. The latter was estimated using DISTO data in Ref.[@Wilkin98] and found to be equal to 0.28$\pm$0.14 $\mu$b at $Q=82$ MeV. Though there are uncertainties in extrapolating the prediction of the TSM (Eq.(\[pndM\])) to such large $Q$ we would have $\sigma(pn\to d
\phi) \approx 0.6 -1\ \mu$b at this $Q$.
Let us discuss now the $\phi/\omega$ ratio. TSM predicts the following value $$\label{ratio}
R_{pn \to d M} =D_{\phi}/D_{\omega}=(30 \pm 7)\times 10^{-3}.$$ This is lower than the corresponding ratio in $pp$ collisions [@SPES3] $$\label{ratio2}
R_{pp \to ppM} =C_{\phi}/C_{\omega}=(49 \pm 26)\times 10^{-3}.$$ and in the reaction $pd \to ^{3}\!\!He\,M$ (see Eq.(3)). It is closer to the ratio of the $\phi$ to $\omega$ yields in $\pi^- p$ collisions (see, e.g., the discussion in Ref.[@Wilkin98]) $$\label{ratio3}
R_{\pi^-p \to nM} =(37 \pm 8)\times 10^{-3}.$$ Another estimate of $R$ can be found if we assume the line-reverse invariance of the amplitudes, which correspond to the diagrams presented in Fig.\[fig:tsm\]. In this case we have
$$\begin{aligned}
\overline{
|T^{\mathrm{LRI}}_{pn \rightarrow dM}(s,t)|^2} & = &
\overline{
|A^{\mathrm{LRI}}_{pn \rightarrow dM}(s,t) +
A^{\mathrm{LRI}}_{pn \rightarrow dM}(s,u)|^2}
\nonumber \\
& = &
\overline{
|A_{\bar{p}d \rightarrow nM}(s,t) +
A_{\bar{p}d \rightarrow nM}(s,u)|^2}\end{aligned}$$
and can define the ratio $$R_{\mathrm{LRI}}=|T^{\mathrm{LRI}}_{pn \to d \phi}|^2/
|T^{\mathrm{LRI}}_{pn \to d \omega}|^2=
|T_{\bar{p}d \to n \phi}|^2/
|T_{\bar{p}d \to n \omega}|^2.$$
Adopting the result of the OBELIX collaboration $ Y(\bar{p}d \to
n\phi)/ Y(\bar{p}d \to n\omega) = (230 \pm 60)\times 10^{-3}$ we get $$\begin{aligned}
R_{\mathrm{LRI}} & = &|T_{\bar{p}d\to n \phi}|^2/|T{\bar{p}d \to n
\omega}|^2
\nonumber \\
& \approx & (p_{\mathrm{cm}}^{\omega}/p_{\mathrm{cm}}^{\phi})
(Y(\bar{p}d \to n \phi)/ Y(\bar{p}d\to n\omega)) \simeq (250 \pm
60)\times 10^{-3},\end{aligned}$$ which is larger by an order of magnitude than the prediction of the TSM given by Eq.(\[ratio\]). If experimental studies will find an essential excess of $R(\phi /\omega)$ over the value predicted by the two-step model it might be interpreted as a possible contribution of the intrinsic $s\bar{s}$ component in the nucleon wave function.
Conclusions {#sec:conc}
===========
Using the two-step model which is described by triangle graphs with $\pi$-, $\rho$- and $\omega$-meson exchanges we calculated the cross sections of the reactions $pn\to dM$, where $M=\omega$ or $\phi$, close to threshold. The predicted cross section of the reaction $pn\to
d\omega$ is found to be significantly larger than the cross section of the reaction $pp\to pp\omega$. The same is expected to be the case for $\phi$ production. We find a $\phi/\omega$ ratio of $R_{pn \to
dM}=(30 \pm 7)\times 10^{-3}$. The measurement of the $\phi$ and $\omega$ yields in the reaction $pn\to dM $ at the same energy release $Q$ will be useful for a better understanding of the mechanism of the OZI-rule violation.
We are grateful to W. Cassing, Ye.S. Golubeva, M.G. Sapozhnikov and C. Wilkin for useful discussions.
[99]{} H. J. Lipkin, [*Phys. Lett. B*]{} [**60**]{} (1976) 371. J. Ellis, E. Gabathuler and M. Karliner, [*Phys. Lett. B*]{} [**217**]{} (1989) 173.
J. Ellis, M. Karliner, D. E. Kharzeev and M. G. Sapozhnikov, [*Phys. Lett. B*]{} [**353**]{} (1995) 319.
S. Okubo, [*Phys. Lett. B*]{} [**5**]{} (1963) 165;\
G. Zweig, [*CERN Report*]{} [**8419/TH 412**]{} (1964);\
I. Iizuka, [*Prog. Theor. Phys. Suppl.*]{} [**37–38**]{} (1966) 21.
M. P. Locher, Y. Lu and B. S. Zou, [*Z. Phys. A*]{} [**347**]{} (1994) 281.
D. Buzatu and F. Lev, [*Phys. Lett. B*]{} [**329**]{} (1994) 143.
R. L. Jaffe, [*Phys. Rev. Lett. B*]{} [**229**]{} (1989) 275.
U.-G. Meissner, V. Mull, J. Speth and J. W. Van Orden, [*Preprint KFA-IKP(TH)-1997-01*]{}, Forschungszentrum Jülich (1997). In [*Big Sky 1997, Intersections between particle and nuclear physics*]{}, 730–732.
L. A. Kondratyuk and M. G. Sapozhnikov, [ *Phys. Lett. B*]{} [**220**]{} (1989) 333;\
L. A. Kondratyuk and M. G. Sapozhnikov, [*Few Body Systems*]{}, [ **Suppl.5**]{} (1992) 201.
L. A. Kondratyuk, M. P. Bussa, Y. S. Golubeva, M. G. Sapozhnikov and L. Valacca, [*Yad. Fiz.*]{} [**61**]{} (1998) 1670.
V. Yu. Grishina et al. nucl-th/9905049; Submitted to Phys. Lett. B
R. Wurzinger et al., [*Phys. Rev. C*]{} [ **51**]{} (1995) R443. R. Wurzinger et al., [*Phys. Lett. B*]{} [ **374**]{} (1996) 283.
G. Fäldt and C. Wilkin, [*Nucl. Phys. A*]{} [ **587**]{} (1995) 769.
L. A. Kondratyuk and Y. N. Uzikov, [*Yad. Fiz.*]{} [**60**]{} (1997) 542.
F. Hibou et al., nucl-ex/9903003
F. Balestra et al., [*Phys. Rev. Lett.* ]{} [**81**]{} (1998) 4572.
K. Nakayama, A. Szczurek, C. Hanhart, J. Haidenbauer, and J. Speth, [*Phys. Rev. C*]{} [**57**]{} (1998) 1580.
OBELIX collaboration, [*Yad. Fiz.*]{} [**59**]{} (1996) 1511 ([*Phys. At. Nucl.*]{} [**59**]{} (1996) 1455).
CRYSTAL BARREL Collaboration, [*Z.Phys. A*]{} [ **351**]{} (1995) 325.
K. F. Liu et al., [*Phys. Rev. Lett.*]{} [**74**]{} (1995) 2172.
S. A. Coon and M. D. Scadron,[*Phys. Rev. C*]{} [**23**]{} (1981) 1150.
G. Janssen, K. Holinde, and J. Speth, [*Phys. Rev. C*]{} [**54**]{} (1996) 2218.
R. Machleidt, K. Holinde, and Ch. Elster, [*Phys. Rep.*]{} [**149**]{} (1987) 1.
M. Lacomb et al., [*Phys. Lett. B*]{} [**101**]{} (1981) 139.
D. M. Binnie et al., [*Phys. Rev. D*]{} [**8**]{} (1973) 2789.
C. Hanhart and A. Kudryavtsev, nucl-th/9812022
T. H. Bauer et al., [*Rev. Mod. Phys.*]{} [**50**]{} (1978) 261.
G. I. Lykasov , W. Cassing, A. Sibirtsev and, M. V. Rzjanin. Preprint UGI-98-37, Nov 1998; nucl-th/9811019 ( Submitted to Europhys. J.)
H. Calén et al., Phys. Rev. Lett. [**79**]{}, (1997) 2642.
H. Calén et al., Phys. Rev. Lett. [**80**]{}, (1998) 2069.
C. Hanhart and K. Nakayama, nucl-th/9809059. Phys. Lett. B, in print.
C. Wilkin, “Baryon98 conference” nucl-th/9810047.
G. Fäldt and C. Wilkin, [*Phys. Lett. B*]{} [**382**]{} (1996) 209.
M. Büscher et al., COSY proposal \#75 (1998).
A. Sibirtsev, [*Nucl. Phys. A*]{} [**604**]{} (1996) 455.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The observation of massive exoplanets at large separation ($\gtrsim
10$ AU) from their host star, like in the HR 8799 system, challenges theories of planet formation. A possible formation mechanism involves the fragmentation of massive self-gravitating discs into clumps. While the conditions for fragmentation have been extensively studied, little is known of the subsequent evolution of these giant planet embryos, in particular their expected orbital migration. Assuming a single planet has formed by fragmentation, we investigate its interaction with the gravitoturbulent disc it is embedded in. Two-dimensional hydrodynamical simulations are used with a simple prescription for the disc cooling. A steady gravitoturbulent disc is first set up, after which simulations are restarted including a planet with a range of masses approximately equal to the clump’s initial mass expected in fragmenting discs. Planets rapidly migrate inwards, despite the stochastic kicks due to the turbulent density fluctuations. We show that the migration timescale is essentially that of type I migration, with the planets having no time to open a gap. In discs with aspect ratio $\sim 0.1$ at their forming location, planets with a mass comparable to, or larger than Jupiter’s can migrate in as short as $10^4$ years, that is, about 10 orbits at 100 AU. Massive planets formed at large separation from their star by gravitational instability are thus unlikely to stay in place, and should rapidly migrate towards the inner parts of protoplanetary discs, regardless of the planet mass.
author:
- |
Cl[é]{}ment Baruteau$^{1,2}$[^1] , Farzana Meru$^{3,4}$ and Sijme-Jan Paardekooper$^{1}$\
$^{1}$DAMTP, University of Cambridge, Wilberforce Road, Cambridge CB3 0WA, UK\
$^{2}$Department of Astronomy and Astrophysics, University of California, Santa Cruz, CA 95064, USA\
$^{3}$School of Physics, University of Exeter, Stocker Road, Exeter, EX4 4QL, UK\
$^{4}$Institut f[ü]{}r Astronomie und Astrophysik, Universit[ä]{}t T[ü]{}bingen, Auf der Morgenstelle 10, 72076 T[ü]{}bingen, Germany
date: 'Accepted 2011 June 1. Received 2011 May 30; in original form 2011 March 25'
title: Rapid inward migration of planets formed by gravitational instability
---
\[firstpage\]
accretion, accretion discs — turbulence — methods: numerical — planetary systems: formation — planetary systems: protoplanetary discs
Introduction {#sec:intro}
============
The large majority of exoplanets have been discovered by the radial velocity and transiting techniques. Recent progress in the adaptive optics technique has revealed about 20 exoplanets by direct imaging. Since the latter requires a high brightness contrast between the planet and the parent star, massive planets have been observed at separations $\gtrsim 10$ AU [e.g., @Kalas08; @Marois08; @Lagrange10; @Lafreniere10; @Marois10]. One such example is the HR 8799 system [@Marois10], which comprises four planets with masses evaluated between 7 and 10 Jupiter-mass, and estimated separations of 14, 24, 38 and 68 AU, respectively.
How can the large orbital separations of such massive planets be explained? With the standard core accretion formation scenario [e.g., @Safronov69; @Mizuno80; @Pollack96; @IdaLin1], it is difficult to form Jupiter-like planets in isolation further than $\sim
10$ AU from a Sun-like star. Still, the formation of a first gap-opening planet close to the snow line [@IdaLin1] could pave the way for the formation of a second-generation planet at the outer edge of the first planet’s gap [@Bryden00; @Thommes05; @IdaLin4]. Provided that the second planet opens a gap, its outer edge could also trigger the formation of a third-generation planet. Subsequent planet formation could proceed as a sequence, as long as there are enough planetesimals and gas left.
The tidal interaction between a planet and its nascent protoplanetary disc alters the planet’s semi-major axis, causing its orbital migration [@gt80]. Assuming the central object is a Sun-like star, a planet typically more massive than Saturn opens a gap around its orbit and generally migrates inwards. It is thus unlikely that a massive planet formed through the core-accretion scenario within $\sim
10$ AU of its host star could migrate to a separation as large as 100 AU. A notable exception was proposed by [@Crida09], who showed that the tidal interaction between a pair of resonant massive planets embedded in a common gap, and their protoplanetary disc, could drive both planets to orbital separations much larger than that of their birth place, providing some specific conditions on the planets mass ratio and on the disc’s turbulent viscosity.
Another in-situ formation scenario is based on the gravitational instability [e.g., @Cameron78; @Boss97]. It involves early-stage protoplanetary discs, which are massive enough for its self-gravity to play a part in its evolution. The probability that discs form fragments is described by two dimensionless parameters. One is the Toomre Q-parameter, $Q = c_{\rm s} \kappa / \pi G \Sigma$, where $G$ is the gravitational constant, $c_{\rm s}$ and $\Sigma$ denote the disc’s sound speed and surface mass density, and $\kappa$ is the epicyclic frequency. For Keplerian discs, $\kappa$ approximately equals the angular frequency $\Omega$. [@toomre64] showed that, for an infinitesimally thin disc to fragment, $Q \lesssim
1$. The second parameter describing the stability of a self-gravitating disc is the disc’s cooling timescale in units of the orbital timescale, $\beta = t_{\rm cool}\,\Omega$, where $t_{\rm cool}
= e (de / dt)^{-1}$ is the cooling timescale and $e$ the disc’s thermal energy density [@Gammie01]. The critical value of the cooling parameter $\beta$, below which fragmentation will occur if $Q
\lesssim 1$, depends on the disc’s adiabatic index [@RLA05], and on the local disc to primary mass ratio [@MeruBate10b]. Its exact value is currently somewhat uncertain as [@MeruBate10c] have recently shown that previous simulations that determined this critical value were unresolved. However, current simulations show that the critical value for $\beta$ could be in the range $3 - 18$ [@Gammie01; @RLA05; @MeruBate10c; @MeruBate10b].
The vast majority of studies of planet formation by gravitational instability have focused on the conditions to fragment discs into bound objects. It is thought to operate at separations larger than $30-50$ AU from the central (Sun-like) star [e.g., @Rafikov05; @Matzner05; @Stamatellos08]. However, little is known of the evolution of these objects, in particular their expected migration resulting from the gravitational interaction with the gravitoturbulent disc they are embedded in. A few numerical studies have observed that clumps could drift inwards [e.g., @Mayer02; @boss2005; @cha10], and some of them have focused on the early phases of disc formation and evolution following the collapse from the prestellar core stage [@vb06; @vb10a; @vb10b; @Machida11]. As these giant planet embryos migrate inwards, they may become tidally disrupted (when their radius becomes comparable to the clump’s Hill radius), thereby delivering a variety of planets in the inner parts of protoplanetary discs [@Nayak10].
In this paper, we examine the orbital evolution of a single planet embedded in its nascent gravitoturbulent disc, following the assumption that the planet has formed by gravitational instability. We show that the planet migrates inwards in a timescale very similar to that of type I migration [@pp09a] in the absence of turbulence. Our physical and numerical models are described in Section \[sec:model\]. In Section \[sec:results\], we present our results of simulations. Conclusions and future directions are drawn in Section \[sec:conclusion\].
Model {#sec:model}
=====
We investigate the tidal interaction between a massive planet and the protoplanetary disc it formed in through the gravitational instability scenario. Once a gaseous clump has formed by fragmentation, it contracts until $H_2$ dissociation occurs, causing a rapid collapse of the clump down to planetary sizes. Our study does not address the formation of the planet, and we consider the clump and the planet as one object. In the following, our system of interest (clump and planet) is referred to as the planet. Our study focuses on its migration driven by the interaction with the disc, following the assumption that the planet has formed by gravitational instability. As described below, our approach is to first set up a steady state gravitoturbulent disc, then include a single planet and follow its orbital migration.
The disc’s properties after fragmentation are poorly constrained. They depend on how many clumps form, and how massive they are. In our study, we assume that the disc properties (density, temperature) are virtually unchanged by the fragmentation stage. In particular, we assume that the disc remains gravitoturbulent. This assumption is motivated by the findings of previous numerical studies on disc fragmentation [e.g., @MeruBate10b], where the images of the discs show similar structures after the formation of a clump.
{width="0.32\hsize"} {width="0.32\hsize"} {width="0.32\hsize"}
We carried out two-dimensional (2D) hydrodynamical simulations using the grid-based, staggered-mesh code FARGO[^2] [@fargo1], for which a self-gravity solver based on fast Fourier transforms was presented in [@bm08b]. When calculating the self-gravitating acceleration, all mass is assumed to be confined to the disc midplane (the razor-thin disc approximation). The thermal energy density $e$ satisfies the equation $$\frac{\partial e}{\partial t} + {\bf \nabla} \cdot (e {\bf v})
= -(\gamma-1) e {\bf \nabla} \cdot {\bf v}
+ Q^{+}_{\rm bulk}
- \frac{e}{\tau_{\rm cool}},
\label{eqnenergy}$$ where ${\bf v}$ denotes the gas velocity, and $\gamma$ is the adiabatic index [its value can be mapped to a three-dimensional adiabatic index, see @Gammie01 and references therein]. In Eq. (\[eqnenergy\]), the cooling timescale $\tau_{\rm cool} = \beta \Omega^{-1}$, with $\beta$ taken to be constant, and the heating source term $Q^{+}_{\rm bulk}$ is provided by artificial viscous heating at shocks arising from the gravitoturbulence. A von Neumann-Richtmyer artificial bulk viscosity is used, as described in [@zeus], where the coefficient $C_2$ is taken equal to 1.4 ($C_2$ measures the number of zones over which a shock is spread over by the artificial viscosity). Providing angular momentum transport driven by gravitoturbulence occurs locally, which is not necessarily the case [@bp99; @Cossins09], the alpha parameter associated with the angular momentum flux density is given by [@Gammie01] $$\alpha = \frac{4}{9} \frac{1}{\gamma(\gamma-1)\beta},
\label{alpha_gammie}$$ which indicates that varying $\beta$ is equivalent to varying the viscous stress in the disc. Preliminary calculations with our fiducial setup (initial profiles and grid’s resolution are described below) showed that the critical value of $\beta$ below which fragmentation occurs ranges from 10 to 15. Since we require the disc to be in a gravitoturbulent state, but we do not require further fragmentation so as to focus on the migration of a single planet, we simulated three disc models with $\beta = 15$, 20 and 30.
Each of our three models initially comprises a $0.4 M_{\odot}$ 2D disc around a star of fixed mass $M_{\star} = 1M_{\odot}$, and spanning a radial range $20 < r < 250$ AU. (As will be shown in Section \[sec:gts\], the disc mass rapidly decreases to about $0.2-0.25 M_{\odot}$ after 30 orbits at 100 AU). The initial surface mass density decreases as $r^{-2}$, and it equals $\approx 20$ g cm$^{-2}$ at 100 AU. The initial disc’s temperature is $\approx 15$ K at 100 AU, and it decreases as $r^{-1}$ (we assume a mean-molecular weight $\mu = 2.4$ and an adiabatic index $\gamma=5/3$). This corresponds to setting the initial disc aspect ratio $h$ (pressure scale height to radius ratio $H/r$) equal to $0.1$ at 100 AU. The disc’s initial Toomre Q-parameter is thus uniform, equal to 2. A small level ($\sim 0.1\%$) of white noise is added initially to break the disc’s axisymmetry. Here and in the following, disc and planet quantities are expressed in physical units. Since our simulations are scale-free, all the results presented throughout this paper can be easily rescaled by using different units of mass, length and temperature.
In all our simulations, the grid is covered with $512$ and $1536$ cells along the radial and azimuthal directions, respectively. A logarithmic spacing is used along the radial direction, as required by the self-gravity module in FARGO. Grid cells are approximately square, and the initial pressure scale height is resolved by about 20 cells along each direction. The radial and azimuthal components of the self-gravitating acceleration are smoothed over a softening length, $\varepsilon_{\rm sg}$. We take $\varepsilon_{\rm sg}=3\times
10^{-4}\,r$ throughout this study, which is a factor $3\times 10^{-3}$ times the initial pressure scale height. The influence of a large $\varepsilon_{\rm sg}$, which can be seen as a crude treatment of the effects of finite thickness, will be examined in Sect. \[sec:numerics\], where we find that the choice for $\varepsilon_{\rm sg}$ does not significantly alter the orbital evolution of a planet embedded in a gravitoturbulent disc. Outflow boundary conditions are used at the grid’s inner and outer edges. Hydrodynamical equations are solved in the frame centred onto the central star. Since a non-inertial frame of reference is adopted, the indirect terms in the expression for the disc’s gravitational potential are included.
Results of hydrodynamical simulations {#sec:results}
=====================================
Our numerical simulations were carried out in two steps. In a first step, a gravitoturbulent disc model was set up, whose properties are described in Section \[sec:gts\]. We then restarted the simulations, including planets of several masses, and following the time-evolution of their orbital separation. The results of the restart simulations are presented in Section \[sec:migration\]. Further investigation on the impact of the disc’s temperature follows in Section \[sec:temp\].
Gravitoturbulent stage {#sec:gts}
----------------------
We describe in this section the results of our three disc models with cooling parameters $\beta=15$, 20 and 30. The absence of heating source initially leads to decrease the disc’s temperature, and thus the Toomre Q-parameter, until gravitoturbulence sets in. As both the initial Q-profile and $\beta$ are taken to be uniform (the cooling timescale thus scales proportional to the orbital timescale), gravitoturbulence progressively develops through the disc, starting from the inner edge. After a few orbits at the disc’s outer edge, a thermal quasi-equilibrium is reached over the whole disc, where cooling is approximately balanced by shock heating due to gravitoturbulence.
The disc’s initial surface density that we assume does not correspond to a mechanical equilibrium with the gravitoturbulent stress. The disc therefore adjusts both its surface density and temperature profiles to satisfy a mechanical equilibrium along with a uniform Toomre-Q profile. This is illustrated in Fig. \[fig:profiles\], where the azimuthally-averaged surface mass density, temperature, and Toomre Q-parameter are depicted at 30 orbits at 100 AU. Their initial value is shown with a dashed curve. In this quasi-steady state, the surface density profile decreases approximately as $r^{-3/2}$, and the disc’s temperature is almost uniform, as is the Toomre Q-parameter. We comment that the high initial disc mass ($0.4 M_{\odot}$) promotes global modes that, along with the outflow boundary condition at the grid’s inner edge, lead to a rapid decrease in the disc mass. After 30 orbits at 100 AU, the latter ranges from $0.21M_{\odot}$ to $0.24M_{\odot}$ for $\beta$ increasing from 15 to 30. The rate at which the disc mass decreases is maximum at the onset of gravitoturbulence, and then declines (but does not vanish) as a quasi-steady state is attained.
![\[fig:alphas\]Gravitational, Reynolds and total alpha parameters, azimuthally- and time-averaged from 30 to 60 orbits at 100 AU (that is, over $\approx 8$ orbits at 250 AU, the location of the disc’s outer edge). Results are displayed for the disc model with cooling parameter $\beta=20$. The dashed curve shows the total alpha parameter expected from viscous transport theory, given by Eq. (\[alpha\_gammie\]).](f2.eps){width="\hsize"}
The alpha parameters associated with the Reynolds and gravitational turbulent stresses are denoted by $\alpha_{\rm Rey}$ and $\alpha_{\rm
sg}$, respectively. Details of their calculation in a 2D disc model are given in the Appendix. Fig. \[fig:alphas\] displays the azimuthally- and time-averaged radial profiles of $\alpha_{\rm Rey}$, $\alpha_{\rm sg}$ and of $\alpha_{\rm tot} = \alpha_{\rm sg} +
\alpha_{\rm Rey}$. Time average is done from 30 to 60 orbits at 100 AU (that is, over $\approx 8$ orbits at the grid’s outer edge). Results are shown for $\beta=20$. We note that $\alpha_{\rm sg}$ and $\alpha_{\rm Rey}$ take very similar values over the whole disc. The same behavior was obtained by [@Gammie01] with 2D local shearing-sheet calculations. The decrease in both $\alpha_{\rm sg}$ and $\alpha_{\rm Rey}$ at large ($\gtrsim 200$ AU) separation is most probably due to the decreasing grid’s resolution, as a logarithmic spacing is used along the radial direction. We comment that $\alpha_{\rm tot}$ is locally up to $50\%$ larger than the value expected from viscous (local) transport theory, given by Eq. (\[alpha\_gammie\]). The difference is probably due to our large disc mass ($\gtrsim 0.2 M_{\star}$) and large aspect ratio ($h \sim
0.1$), which tend to foster global transport[^3] of angular momentum [e.g., @Lodato04]. Although not shown here, very similar results were obtained for $\beta=15$ and $\beta=30$.
Finite resolution effects (numerical diffusion) leads to a negligible transport of angular momentum compared to gravitoturbulence [see e.g., @linpapa10 who estimated the numerical “alpha viscosity” in the FARGO code to be $\lesssim 10^{-5}$ for a number of grid cells similar to ours]. Artificial bulk viscosity most likely leads to a small, if not negligible transport of angular momentum compared to gravitoturbulence, but note that it is unclear how its contribution can be isolated in our simulations. We point out that if artificial bulk viscosity induced a net transport of angular momentum on average, it would lead to a discrepancy between the value of $\alpha_{\rm sg} +
\alpha_{\rm Rey}$ measured in local, steady-state disc models and the alpha value given in Eq. (\[alpha\_gammie\]). [@Gammie01] carried out 2D local shearing-sheet simulations of gravitoturbulent disc models using the ZEUS code, which is very similar to the FARGO code. In particular, both codes use the same prescription for the artificial bulk viscosity. He found that $\alpha_{\rm sg} + \alpha_{\rm Rey}$ agrees with the value given by the expression in Eq. (\[alpha\_gammie\]) to within $\sim 10\%$, which shows that artificial bulk viscosity should lead to a small transport of angular momentum compared to gravitoturbulence. Also, although not shown here, we have checked that increasing the amount of artificial bulk viscosity [the coefficient $C_2$ in @zeus] by a factor of 3 leads to a negligible change in the values of $\alpha_{\rm Rey}$ and $\alpha_{\rm sg}$.
Restart with an embedded planet {#sec:migration}
-------------------------------
We restarted the simulations of Section \[sec:gts\] at 30 orbits, including a single planet at $100$ AU. The disc’s temperature and surface mass density at this location are close to their initial value for our three values of $\beta$ (see Fig. \[fig:profiles\]). Note in particular that the disc aspect ratio at the planet’s initial location is $\approx 0.1$. Akin to the clump’s contraction timescale, the final mass of planets formed by gravitational instability is uncertain [e.g., @Boley10; @Helled11], and may substantially differ from the clump’s initial mass. The latter is estimated by [@Boley10] as $\sim h_p^3\,M_{\star}$ for $Q \sim 1.5$, with $h_p$ the disc aspect ratio at the clump’s forming location. This estimate of the clump’s initial mass, which is based on the fragmentation of spiral arms, is typically a factor $5-10$ smaller than the Toomre mass [see @Boley10 their figure 1, and references therein]. We considered three planet masses corresponding to a planet to primary mass ratio $q = M_p / M_{\star}$ around $h_p^3$: a Saturn-mass planet ($q=3\times 10^{-4}, q/h_p^3 \sim 0.3$), a Jupiter-mass planet ($q=10^{-3}, q/h_p^3 \sim 1$), and a 5 Jupiter-mass planet ($q=5\times
10^{-3}, q/h_p^3 \sim 5$). The planets gravitational potential is softened over a smoothing length $\varepsilon = 3\times 10^{-2}\,r_p$, with $r_p$ the planet’s orbital separation, and it takes its maximum value at the restart time (we do not gradually ramp up the planet’s mass). We will show in Section \[sec:numerics\] that our results are not altered by a more gentle introduction of the planet’s potential in the disc over a dynamical timescale, the typical formation timescale of a clump. We comment that in self-gravitating discs, the mass that is relevant for disc–planet interactions is the mass of the planet and of the gas envelope surrounding it. Although for simplicity the planet’s mass is fixed, gas can be accreted onto its envelope (circumplanetary disc), thereby contributing to increasing the effective planet’s mass. Note however that this accretion rate is probably not realistic due to our simple prescription for the disc cooling. Since the disc is fully self-gravitating, the calculation of the force exerted on the planet takes all the disc into account, it does not partly exclude the planet’s Hill radius [@cbkm09].
![\[fig:nicedens\]Contours of the disc’s surface density obtained in a restart simulation with $\beta = 30$, where a Jupiter-mass planet has been introduced at 100 AU. Results are displayed three orbits after the restart time. The planet is now located at about 60 AU from the central star (at $x\sim-60$ AU, $y\sim60$ AU).](f3.eps){width="\hsize"}
Fig. \[fig:nicedens\] displays the surface density obtained for the disc model with $\beta=30$, perturbed by a Jupiter-mass planet. Density contours are shown only three orbits after the restart, the planet is now located at $\approx 60$ AU from the central object. The density perturbation due to the planet’s potential is comparable to the turbulent density perturbations. This similarity stresses that the magnitude of the stochastic torque acting on the planet can be similar to that of the disc–planet tidal torque obtained in the absence of turbulence [e.g., @qmwmhd4; @bl10]. In such turbulent configurations, the migration timescale must be evaluated in a statistical way, by running several simulations. For each planet mass and each value of $\beta$, we performed a suite of eight simulations, varying the planet’s azimuth at restart as $\varphi_p = \pi/4 \times i, \,i \in [0-7]$ (that is, independently of the disc’s density structure at restart). The migration timescales obtained for these series of runs are presented in Section \[sec:timescale\]. The impact of stochastic density fluctuations is described in Section \[sec:kicks\]. A comparison with predictions of laminar disc models follows in Section \[sec:laminar\], and a brief discussion on the impact of numerics is given in Section \[sec:numerics\].
### Migration timescale {#sec:timescale}
The time evolution of the planets’ orbital separation is displayed in Fig. \[fig:azimuth\]. Planet mass increases from left to right, and the cooling parameter $\beta$ increases from top to bottom. In all panels, time is expressed in years (bottom x-axis), and in orbital periods at 100 AU (top x-axis). Results are shown for the eight evenly-spaced values of the planet’s azimuth $\varphi_p$ at restart (increases with increasing color’s wavelength).
{width="0.32\hsize"} {width="0.32\hsize"} {width="0.32\hsize"}
The net trend coming out of our results is that, in spite of the stochastic kicks triggered by turbulence, planets migrate inwards very rapidly. Averaging over $\varphi_p$ and $\beta$, the migration timescale (defined as the time to migrate from 100 AU to 20 AU, the location of the grid’s inner edge) is typically as short as 3 orbits for the 5 Jupiter-mass planet, 6 orbits for the Jupiter-mass planet, and about 8 orbits for the Saturn-mass planet. In addition, at fixed $\beta$, the amplitude of stochastic kicks increases with decreasing planet mass, and so does the spread in the migration timescale. Both results may be qualitatively interpreted as follows. The torque exerted by the disc on the planet may be decomposed as a background tidal torque, plus a stochastic torque due to turbulent fluctuations. In the mass regime that we consider ($q/h_p^3 \sim 1$), the amplitude of the tidal torque should increase with increasing planet mass. However, like in discs with turbulence driven by the magneto-rotational instability, it is unclear if the value of the tidal torque is similar to that predicted in laminar disc models, in the absence of turbulence [@qmwmhd4]. A detailed comparison with results of laminar disc models will be presented in Section \[sec:laminar\]. The standard deviation of the stochastic torque’s distribution is primarily controlled by $\alpha_{\rm tot}$, and therefore by the cooling parameter $\beta$ (see Eq. \[alpha\_gammie\]). For a given $\beta$, the smaller the planet mass, the smaller the tidal torque to stochastic torque ratio, the more sensitive the planet is to stochastic kicks, and therefore the larger the spread in the migration timescale. In the same vein, at fixed planet mass, the spread in the migration timescale should increase with decreasing $\beta$. This trend does not clearly show up in Fig. \[fig:azimuth\], most probably because our minimum and maximum values of $\beta$ only differ by a factor of 2. Some further insight into the effect of stochastic kicks is given in Section \[sec:kicks\]. We finally point out that the disc’s density and temperature profiles do not significantly decrease over the planets migration timescale (for the longest simulation with the Saturn-mass planet, the disc’s mass decreases by at most $10\%$).
### Stochastic kicks {#sec:kicks}
The results of Section \[sec:timescale\] show that planets formed by gravitational instability should rapidly migrate inwards. Even for the Saturn-mass planet, whose mass is smaller than the estimated initial clump’s mass (see first paragraph of Section \[sec:migration\]), inward migration occurs in typically less than 10 orbits. As illustrated in Fig. \[fig:azimuth\], planet migration in gravitoturbulent discs is not a smooth process. Planets may experience large kicks, either inwards or outwards, depending on the local density perturbations they encounter. An example of outward kick is illustrated in Fig. \[fig:kicks\], which displays the gas density obtained for the restart simulation with $\beta=30$ and the Saturn-mass planet that experiences a significant episode of outward migration in the bottom-left panel of Fig. \[fig:azimuth\] (red curve). Time increases by 0.1 orbital periods at 100 AU from left to right. In each panel, the planet’s location is indicated with a filled black circle.
We first comment that, because the planet’s mass is low enough ($q/h_p^3 \approx 0.3$), the wakes it generates are invisible, their density contrast being much weaker than that of the turbulent perturbations. In the case depicted in Fig. \[fig:kicks\], the planet finds itself with an underdense region behind it, and a (comparatively) overdense region in front of it. This density contrast yields a large, positive coorbital corotation torque that is responsible for the vigorous outward kick experienced by the planet. This kick increases the planet’s orbital separation from $\sim
75$ AU to $\sim 95$ AU in only $0.2$ orbital periods at 100 AU (that is, in 200 years). Similarly, inward kicks are obtained as the planet experiences a negative coorbital corotation torque induced by an underdense region ahead of it, and a (comparatively) overdense region behind it. As is shown for instance in the bottom-left panel of Fig. \[fig:azimuth\], kicks excite planets epicyclic motion, which is rapidly damped as the underlying disc–planet tidal torque takes over.
The stochastic kicks induced by gravitoturbulence are reminiscent of type III runaway migration. This migration regime relies on the formation of a coorbital mass deficit in the planet’s horseshoe region, which takes the form of an asymmetric density structure behind and in front of the planet [@mp03]. In massive ($Q \gtrsim 1$) laminar discs with moderate viscosity ($\alpha$ is a few $\times
10^{-3}$), runaway migration is particularly relevant for planets with $q \sim h_p^3$ [@mp03], like the planets considered in our study. However, in our model, the disc’s turbulent viscosity ($\alpha_{\rm tot}$ is a few percent) is too large to allow planets to open a partial dip around their orbit before they reach the disc’s inner edge. Thus, planets do not build up a coorbital mass deficit. But, through the density fluctuations it triggers, gravitoturbulence may provide an *effective* mass deficit on each side of the planet, and induce an effective type III migration. However, this effective mass deficit is not coorbital with the planet (as can be seen in Fig. \[fig:kicks\]), and therefore cannot lead to a runaway. Consequently, the relative fast inward migration that we obtain cannot be attributed to type III runaway migration. Also, although the planets we consider are massive, their short formation and migration timescales compared to their gap-opening timescale suggests they are subject to type I-like migration instead of type II, perturbed by stochastic kicks. This point will be clarified in Section \[sec:laminar\].
Stochastic kicks provide a source of very fast relative motion between the planet and the disc. Fig. \[fig:migrate\] displays the migration speed in units of the local sound speed for the three runs with $\beta=30$, and zero azimuth at restart. The migration rate can be a significant fraction of the local sound speed, and short episodes of supersonic motion are even obtained. We note that the maximum value for the migration speed to sound speed ratio tends to increase with decreasing planet mass, as less massive planets are more sensitive to stochastic kicks.
![\[fig:migrate\]Time variation of the (absolute value of the) migration speed, in units of the local sound speed, obtained for the restart simulations with $\beta = 30$ and zero azimuth at restart. They correspond to the results shown with black curves in the bottom row of Fig. \[fig:azimuth\]. The migration speed is measured as the time derivative of the planet’s orbital separation.](f6.eps){width="\hsize"}
### Comparison to laminar disc models {#sec:laminar}
Our results of simulations indicate that the migration of planets embedded in gravitoturbulent discs is driven by both a vigorous (negative) tidal torque, and a (positive or negative) stochastic torque. As the planet mass increases, the former prevails over the latter. In this section, we compare the results of our turbulent disc model with $\beta=30$ to those of an equivalent laminar disc model, including the viscous force in the momentum equation. A constant alpha viscosity is used, approximately equal to $1.3\%$. The comparison is done for the same three planet masses as previously considered. As the initial conditions for the laminar calculations, we take surface density and temperature profiles that fit those of the gravitoturbulent run with $\beta=30$ at 30 orbits: $\Sigma \approx
15\;{\rm g\;cm}^{-2}\times (r / 100\;{\rm AU})^{-3/2}$, and $T \approx
25$ K.
Two series of laminar calculations were carried out: (i) one with an isothermal equation of state, where the imposed (flat) temperature profile remains constant in time, and (ii) one with an energy equation with both viscous heating ($\alpha\approx1.3\%$) and $\beta-$cooling ($\beta=30$). For the latter series, the aforementioned[^4] value for $\alpha$ leads to an initial thermal balance between viscous heating and $\beta-$cooling, see Eq. (\[alpha\_gammie\]).
All laminar calculations include the axisymmetric component of the disc’s self-gravity. This is meant to avoid a large mismatch between the disc’s and the planet’s angular velocities due to the disc gravity, which would yield a spurious shift inwards of all Lindblad resonances, and therefore an artificial acceleration of the inward migration [@bm08b]. Also, in all laminar runs, the calculation of the force exerted by the disc on the planet excludes the planet’s circumplanetary disc [@cbkm09]. All other parameters are otherwise the same as in the gravitoturbulent calculations.
The results with the laminar disc model and an energy equation are depicted in Fig. \[fig:laminar\] as solid curves. It takes about 5 orbits for the 5 Jupiter-mass planet to migrate from 100 AU to 50 AU, and about 7 orbits for the Jupiter-mass planet. This corresponds to migration rates $\sim$ half those of the gravitoturbulent calculations (see bottom row of Fig. \[fig:azimuth\]). Interestingly, the difference in migration rates is larger for the Saturn-mass planet, which drifts from 100 to 60 AU in $20-25$ orbits in the laminar run, and (on average) in only $5-6$ orbits in the gravitoturbulent run. We note that the migration timescales of the Saturn- and Jupiter-mass planets differ by a factor approximately equal to their mass ratio, as expected in the type I migration regime, wherein the tidal torque encompasses the differential Lindblad torque and the horseshoe drag [@pp09a]. We comment that it is the large disc aspect ratio ($h\sim0.1$) that makes the Saturn-mass planet ($q/h^3\sim0.3$) and the Jupiter-mass planet ($q/h^3\sim1$) relevant to type I migration. The 5 Jupiter-mass planet ($q/h^3\sim5\gg1$) is, however, more prone to non-linear effects [e.g., @mak2006]. This is presumably why the migration timescales of the Jupiter- and the 5 Jupiter-mass planets do not differ by a factor of their mass ratio. It may seem surprising that the 5 Jupiter-mass planet does not open a gap and migrates in a very short timescale. The gap-opening criterion for a planet on a fixed circular orbit in a laminar viscous disc takes the form [@Crida06] $$1.1 \left( \frac{q}{h^3} \right)^{-1/3}
+ 50 \left( \frac{\alpha}{h} \right) \left( \frac{q}{h^3} \right)^{-1}
\lesssim 1,
\label{gapcriterion}$$ where $\alpha$ and $h$ are to be evaluated at the planet’s orbital separation. Although it is unclear to what extent the gap-opening criterion in Eq. (\[gapcriterion\]) may be applied to planets migrating in gravitoturbulent discs, it provides some insight into the fact the planets in our disc models are not expected to clear a gap. Notwithstanding that the large values of $h$ and $\alpha$ imply the 5 Jupiter-mass planet does not quite satisfy the above gap-opening criterion (the left-hand side of Eq. (\[gapcriterion\]) typically ranges from 2 to 4 in the disc inner parts), we stress that the vigorous tidal torque exerted by the disc, which is partly due to the large masses of the disc and the planet, conspires to make the planet migrate in a timescale shorter than the timescale required to open a gap (a few libration timescales, which would correspond here to a few tens of orbits). In other words, the 5 Jupiter-mass planet migrates very fast because it has no time to open a gap.
![\[fig:laminar\]Time evolution of the planets’ orbital separation obtained for the laminar disc model with initial profiles of surface density and temperature similar to those of the gravitoturbulent run with $\beta=30$ before restart. Solid curves show the results with inclusion of an energy equation (with viscous heating and $\beta$-cooling), and dashed curves those with an isothermal equation of state.](f7.eps){width="\hsize"}
The results of laminar simulations with an isothermal equation of state are overplotted as dashed curves in Fig. \[fig:laminar\]. The migration rates obtained in this series are smaller than with inclusion of an energy equation. The difference arises from a large, negative horseshoe drag obtained with an energy equation, and driven by a *positive* entropy gradient in our disc model [@bm08a; @pp08; @mc09; @pbck10 in our case, the radial profile of the disc entropy scales in proportion to the orbital separation].
Further insight into the impact of the horseshoe drag may be obtained by comparing the migration timescales of our laminar calculations to those predicted by type I migration theory. Note that the predicted migration timescales are inferred from the torque exerted on a planet held on a fixed circular orbit. In the isothermal series, where $\Sigma \propto r^{-3/2}$ and $T$ is uniform, the total torque reduces to the differential Lindblad torque. Using Eq. (14) of [@pbck10], we find migration timescales of about $75$ and $20$ orbits for the Saturn- and Jupiter-mass planets, respectively. This is in good agreement (to within $\sim 25\%$) with our findings (dashed curves in Fig. \[fig:laminar\]). In the series with the energy equation, the total torque comprises the differential Lindblad torque and the horseshoe drag. Despite the large value of $\alpha$ in our disc model, the horseshoe drag is found to be fully unsaturated, as the diffusion timescale across the planet’s horseshoe region remains larger than the U-turn timescale [@bm08a; @mc10; @pbk11]. Using Eqs. (14) and (45) of [@pbck10], we now find migration timescales of about $35$ and $10$ orbits for the Saturn- and Jupiter-mass planets, respectively. This is shorter than what we find (see solid curves in Fig. \[fig:laminar\]), although still in decent agreement. The difference presumably arises from the large velocity difference between the disc and the planet, which sets asymmetric horseshoe streamlines around the planet’s orbital radius [@masset02], an effect that is not taken into account in the horseshoe drag expression of [@pbck10].
![\[fig:mcpd\] Time variation of the mass enclosed in the planet’s Hill radius in units of the planet mass, obtained for the restart simulation with $\beta = 30$ and and zero azimuth at restart. The mass ratio is displayed before the planet crosses the inner edge of the computational domain.](f8.eps){width="\hsize"}
We further comment on having a positive entropy gradient in our gravitoturbulent disc model. Let us assume the disc profiles of surface density and temperature scale as $r^{-\sigma}$ and $r^{-\varpi}$, respectively. The disc entropy $s$, which we define as $s \equiv p \Sigma^{-\gamma}$, where the disc pressure $p$ verifies the ideal gas equation of state, scales as $r^{-\xi}$, with $\xi =
\varpi - (\gamma-1)\sigma$. As the disc evolves towards a steady state with approximately uniform Toomre-Q profile, $\xi \sim -3 +
\sigma(3-\gamma)$. Steady-state gravitoturbulent discs may thus have a globally decreasing profile of entropy only for surface density profiles steeper than $r^{-9/4}$, assuming $\gamma=5/3$. Furthermore, for $\gamma=5/3$ and our fiducial softening length $0.3H(r_p)$, the fully unsaturated horseshoe drag $\Gamma_{\rm hs}$ reads [@pbck10 their Eq. 45] $$\gamma \Gamma_{\rm hs} / \Gamma_0 \approx -20.5 + 8.6\sigma,
\label{eqhs}$$ where $\Gamma_0 = (q/h_p)^2 \Sigma_p r_p^4 \Omega_p^2$, and all quantities with the subscript $\emph{p}$ are to be evaluated at the planet’s location. Eq. (\[eqhs\]) shows that, in discs with a uniform Toomre-Q parameter, the horseshoe drag is negative whenever the surface density profile is shallower than $r^{-2.4}$, assuming $\gamma=5/3$. The differential Lindblad torque $\Gamma_{\rm L}$ can be similarly recast as [@pbck10 their Eq. 14] $$\gamma \Gamma_{\rm L} / \Gamma_0 \approx 3.2 - 4.0\sigma,$$ and the total torque $\Gamma_{\rm tot}$ as $$\gamma \Gamma_{\rm tot} / \Gamma_0 = -17.3 + 4.6\sigma.
\label{eqtotaltq}$$ Using Eq. (\[eqtotaltq\]), the type I migration timescale, $\tau_I =
r_p^2 \Omega_p M_p / 2 \Gamma_{\rm tot}$, reads $$\frac{\tau_I}{T_{\rm orb}} \approx 5.6 \times \left( 3.8 - \sigma\right)^{-1} \gamma Q_p \left( \frac{q}{h_p^3} \right)^{-1} \left( \frac{h_p}{0.1} \right)^{-2},
\label{eq:tauI}$$ where $T_{\rm orb}$ stands for the orbital period at the planet’s orbital radius. The estimate in Eq. (\[eq:tauI\]) applies to planets with $q \sim h_p^3$, and is approximate, since it is based on the torque expression for a planet on a fixed orbit. As we have already stressed above, the fast relative motion between the disc and planet is likely to affect this estimate. Nevertheless, it indicates that the type I migration of planets formed by fragmentation ($q \sim
h_p^3, Q_p \sim 1.5$) in discs with moderate aspect ratio ($h_p \sim
0.1$) should typically occur in a few tens of orbits at 100 AU, that is a few $10^4$ years.
{width="0.32\hsize"} {width="0.32\hsize"} {width="0.32\hsize"}
Still, the migration timescales in our gravitoturbulent simulations are a factor of a few shorter than obtained in laminar disc models. In addition to the stochastic kicks due to turbulence, the following two mechanisms may account for the larger migration rates in the gravitoturbulent disc model:
1. The non-axisymmetric component of the disc’s self-gravity shifts Lindblad resonances towards the planet’s orbital radius [@ph05]. For discs close to marginal stability, and for our softening parameter ($\varepsilon / H \sim 0.3$), this shift increases the differential Lindblad torque by typically a factor of 2 [@bm08b their Figure 10]. Note however that this mechanism has been investigated in laminar disc models. It is therefore not straightforward whether it should behave identically with turbulence. Still, numerical simulations indicate that, on average, the Lindblad torque behaves similarly in laminar and turbulent disc models [@bl10 where wave-like turbulence is generated by stochastic forcing of the disc].
2. In fully self-gravitating discs, the effective mass that dictates the strength of the tidal torque and thus the migration rate not only includes the planet mass, but also the mass of the gas surrounding the planet’s location. For massive planets ($q >
h_p^3$), the latter corresponds to the planet’s circumplanetary disc, which is a fraction of the planet’s Hill sphere [@cbkm09]. It is unclear what it corresponds to for smaller planets; we speculate that it might be the planet’s Bondi sphere. For illustration purposes, the time evolution of the mass inside the planet’s Hill radius is shown in Fig. \[fig:mcpd\] for three gravitoturbulent runs with $\beta=30$. It is comparable to, albeit somewhat larger than the planet mass. It progressively grows as the planet moves inwards, despite the decrease in the planet’s Hill radius with decreasing orbital separation. It is possible that the short cooling timescales used in our simple cooling prescription (a few dynamical timescales, depending on $\beta$) implies that the gas in the planet’s Hill radius can contract more rapidly than it would do with a more realistic cooling function. Again, notwithstanding the uncertain dynamical role of the planet’s Hill sphere in turbulent disc models, it is likely that the increased effective mass is responsible for an increase in the migration rate by a factor of a few compared to laminar calculations (which only include the axisymmetric component of the disc’s self-gravity). We comment that both the increase in the effective planet mass, and the decrease in the disc aspect ratio with radius (in discs with flat temperature profile, $h \propto r^{1/2}$), make planets less and less sensitive to stochastic kicks as they migrate inwards. This explains why in Fig. \[fig:kicks\] the amplitude of the stochastic kicks decreases with decreasing orbital separation. This also implies that, if a planet undergoes a vigorous inward kick, it will be increasingly harder to reverse the trend. However, should a planet experience a large outward kick, it would take longer for the tidal torque to take over and lead to inward migration again.
### Impact of numerics {#sec:numerics}
We briefly assess in this section the impact of numerics on our results of gravitoturbulent calculations. We have performed an additional suite of simulations with the Jupiter-mass planet and $\beta=30$, with the planet’s mass first smoothly ramped up over one orbit, while the planet is held on a Keplerian orbit. After this, the planet is released in the disc. This procedure is meant to mimic the clump formation, which typically occurs over a dynamical timescale, much shorter than the gap-opening timescale (see Section \[sec:laminar\]). The time evolution of the planet’s orbital separation is depicted in the left panel of Fig. \[fig:numerics\]. It shows the very same trend as obtained without smoothly increasing the planet’s mass (see the mid-bottom panel of Fig. \[fig:azimuth\]).
To assess the influence of resolution on our simulation results, we have varied the grid’s resolution, and the softening length $\varepsilon_{\rm sg}$ over which the self-gravitating acceleration is smoothed. We have run two additional disc models with $\beta=30$, one with an increased grid resolution ($1024\times 2048$), and one with a larger $\varepsilon_{\rm sg}$ ($3\times 10^{-2}\,r$). In the latter case, $\varepsilon_{\rm sg}$ is 0.3 times the initial pressure scale height, and it can be regarded as a crude treatment for the effects of the disc’s finite thickness. The value of $\varepsilon_{\rm sg}$ at $r=r_p$ equals the softening length of the planet’s gravitational potential. We have checked that before the introduction of the planet at 30 orbits the disc properties (density and temperature profiles) in both additional models are in close agreement with those of the corresponding run at our fiducial resolution and softening length value. We restarted these simulations with a Jupiter-mass planet. Again, in each case, a series of eight simulations was performed with evenly-spaced values for the planet’s azimuth at restart. Results with increased grid resolution, and increased $\varepsilon_{\rm sg}$ are depicted in the middle and right panels of Fig. \[fig:numerics\], respectively. Overall, the amplitude of the stochastic fluctuations is somewhat larger in the high-resolution simulations. Still, except in one run that features a rather long episode of outward migration, the migration timescale of the high-resolution simulations ranges from about 3 to 9 orbits, which is broadly consistent with the results at our fiducial resolution. Furthermore, increasing $\varepsilon_{\rm sg}$ from $3\times 10^{-4}\,r$ to $3\times 10^{-2}\,r$ does not significantly change the amplitude of the stochastic kicks experienced by the planet, and the migration timescales obtained with both values of $\varepsilon_{\rm sg}$ are in reasonable agreement. The results of both additional disc models indicate that the planet’s orbital evolution is essentially unaltered by density structures arising from gravitoturbulence with typical size $\lesssim 0.3H$.
{width="0.32\hsize"} {width="0.32\hsize"} {width="0.32\hsize"}
Results at lower disc temperature {#sec:temp}
---------------------------------
As shown in Section \[sec:migration\], the migration of planets formed by gravitational instability results from a competition between the background disc–planet tidal torque, and the stochastic torque due to turbulence. At least in the early stages of their evolution, such planets should have a planet-to-primary mass ratio $q \sim
h_p^3$, and be subject to type I-like migration (see Section \[sec:laminar\]). As far as the tidal torque is concerned, Eq. (\[eq:tauI\]) shows that the migration timescale is essentially controlled by three dimensionless parameters: $q/h_p^3$, which should be $\sim 1$ in the early stages of the planet evolution, $Q_p$, which is $\gtrsim 1.5$ in a steady state, and $h_p$. The disc model that we have considered so far has $h_p \sim 0.1$, which leads to rapidly formed Jupiter-mass planets. However, it is possible that the disc initially has a lower aspect ratio, thereby forming less massive planets. The latter may still become Jupiter-sized objects if mass growth is efficient.
To evaluate the impact of a lower disc temperature on our results, we restarted the simulation of Section \[sec:gts\] with $\beta=15$ at $85$ orbits. At this time, the disc aspect ratio at the planet’s restart location (100 AU) has decreased from 0.1 to about 0.06, and the disc’s mass from $0.2 M_{\star}$ to $0.15 M_{\star}$. We choose again three planet masses corresponding to a planet-to-primary ratio $q$ around $h_p^3$: a 20 Earth-mass planet ($q=6\times 10^{-5},
q/h_p^3 \sim 0.3$), a $\sim$ Saturn-mass planet ($q=2\times 10^{-4},
q/h_p^3 \sim 1$), and a Jupiter-mass planet ($q=10^{-3}, q/h_p^3 \sim
5$). As previously, we take eight evenly-spaced planet’s azimuths at restart. The time evolution of the planets’ orbital separation is shown in Fig. \[fig:lower\].
Compared to the results with $h_p = 0.1$, planets with the same $q/h_p^3$ are now more sensitive to stochastic fluctuations, the tidal torque becoming less vigorous. This can be seen from Eq. (\[eq:tauI\]), which indicates that at fixed values of $q/h_p^3$ and $Q_p$, the migration timescale due to the tidal torque scales with $h_p^{-2}$. The mean migration timescales that we obtain for $h_p
\sim 0.06$ (about 6, 12, and 50 orbits with decreasing $q/h_p^3$) are typically a factor of 3-5 longer than in the corresponding runs with $h_p \sim 0.1$, which is roughly consistent with the aforementioned scaling $\tau_{I} \propto h_p^{-2}$.
Concluding remarks {#sec:conclusion}
==================
Planets observed at large orbital separation from their host star (typically $\gtrsim 50$ AU) are thought to be potential candidates for the formation scenario based on the gravitational instability of massive discs. A number of studies have shown that the formation of clumps by fragmentation is possible at such separations [e.g., @Rafikov05; @Matzner05; @Stamatellos08]. The orbital evolution of clumps formed by gravitational instability has been investigated in a few studies [e.g., @Mayer02; @boss2005; @vb10a; @cha10; @Machida11]. Clumps are often observed to migrate inwards on short timescales, although in some disc models the formation of several clumps and/or the use of different cooling functions has a significant impact on migration, which sometimes appears to be suppressed [@boss2005].
In this paper, we have studied the interaction between a single planet and the gravitoturbulent disc it is embedded in, following the assumption that the planet has formed by gravitational instability. We have performed 2D hydrodynamical simulations with a simplified prescription for the disc’s cooling. Three disc models with different cooling timescales have been considered, giving rise to different levels of gravitoturbulence. After having set up a quasi-steady state gravitoturbulent disc, we have restarted our simulations including a single planet with a range of masses approximately equal to the expected initial mass of clumps formed by fragmentation [$\gtrsim h^3 M_{\star}$, with $h$ the disc aspect ratio at the fragmenting location; see @Boley10].
We find that a planet interacting with its nascent gravitoturbulent disc migrates inwards on very short timescales, despite the stochastic kicks due to turbulent density fluctuations. In a disc with aspect ratio $\sim 0.1$, it takes less than 10 orbits at 100 AU (that is, about $10^4$ years) for a Jupiter-mass planet to migrate from 100 AU to 20 AU (the inner edge of our computational grid). Planets less massive than the expected fragmenting mass are more sensitive to stochastic kicks, but also migrate inwards relatively fast. Their formation and migration timescales being shorter than their gap-opening timescale (partly because of the disc’s vigorous turbulence), planets formed by fragmentation do not open a dip or a gap around their orbit, and are therefore not subject to the type II and runaway type III migration regimes. Turbulent density fluctuations may provide an effective mass deficit in the planet’s horseshoe region that kicks the planet either inwards or outwards (see Section \[sec:kicks\]). These kicks can be seen as an effective, *temporary* type III migration coming on top of the inward migration due to the background disc–planet tidal torque.
The comparison of our results with those of laminar disc models shows that the averaged tidal torque driving the net inward migration in gravitoturbulent discs is essentially the one that would lead to type I migration in the absence of turbulence (see Section \[sec:laminar\]). Part of the reason as to why planets rapidly migrate inwards in our model is due to the set up of a positive entropy gradient in a steady state, which yields a large negative horseshoe drag, adding up to the negative differential Lindblad torque. We argue that in gravitoturbulent discs with uniform Toomre-Q profiles and cooling-to-orbital timescale ratios (that is, with a uniform alpha parameter), the horseshoe drag should be negative, unless the surface density decreases very steeply (see Eq. \[eqhs\]). Although this possibility cannot be excluded, we believe that fast inward migration should be a generic expectation for planets formed by gravitational instability.
The main conclusion of this present study is that a single planet formed by gravitational instability should migrate inwards on a timescale much shorter than the expected lifetime of protoplanetary discs. However, it is not possible at this stage to say that massive planets at large separation from their host star, such as the HR 8799 system, *could not* have formed by gravitational instability. The simulations presented here contain a number of simplifications that need to be addressed. We have considered a simplified cooling prescription, causing the entire disc to be gravitoturbulent. In reality, we would expect the inner parts of a disc to be too hot to be gravitationally unstable, and other sources of turbulence such as the magnetorotational instability could prevail [see e.g., @Zhu09; @RA09; @Clarke09 who studied the radial dependence of the alpha viscosity parameter in self-gravitating discs with realistic cooling]. It is thus possible that the rapid type I migration of planets formed by gravitational instability gets slowed down in the disc inner parts, resulting in gap formation. A more realistic cooling prescription will also impact the rate at which planets can contract, with various outcomes for the evolution of the planet mass over its orbital migration [e.g., @Nayak10]. Also, we have focused on the orbital evolution of a single planet assumed to have formed by fragmentation, whereas the formation of multiple clumps is a likely outcome of gravitational instability. It is fully expected that the interaction between clumps may significantly impact their migration. We will address these issues in future work.
Calculation of the alpha parameters in a two-dimensional disc {#sec:alpha}
=============================================================
When mass is confined to a razor-thin disc, the calculation of the alpha parameters associated with the Reynolds and gravitational stresses may be inferred from the general three-dimensional case by substituting the volume mass density $\rho$ with $\Sigma \delta(z)$, where $\Sigma$ is the surface mass density of the razor-thin disc, $\delta$ is Dirac’s delta function, and $z$ denotes the vertical coordinate. One obtains [@LBK72; @Gammie01] $$\alpha_{\rm Rey} = \frac{2}{3} \frac{\langle \Sigma \delta v_r \delta v_{\varphi} \rangle}{\langle \Sigma c_{\rm s}^2 \rangle},
\label{alpha_Rey}$$ where $\langle . \rangle$ stands for the azimuthal average, $\delta
v_{r} = v_{r} - \langle v_{r} \rangle$, $\delta
v_{\varphi} = v_{\varphi} - \langle v_{\varphi} \rangle$, and $$\alpha_{\rm sg} = \frac{2}{3} \frac{\langle \int_{-\infty}^{\infty} (4\pi G)^{-1} g_r g_{\varphi} dz \rangle}{\langle \Sigma c_{\rm s}^2 \rangle}.
\label{alpha_sg}$$ In both previous equations, $v_r$, $v_{\varphi}$ and $c_{\rm s}$ are to be evaluated in the razor-thin disc ($z \equiv 0$) and are therefore directly computed in a 2D simulation. However, the vertical dependence of $g_r$ and $g_{\varphi}$, the radial and azimuthal components of the self-gravitating acceleration, needs to be determined as the gravitational field outside the disc contributes to the shear stress despite mass being confined to the disc. The potential $\Phi$ from which $g_r$ and $g_{\varphi}$ can be derived is given by $$\Phi = -\int_{\cal{D}} \frac{G\Sigma(r^{\prime} ,\varphi^{\prime})r^{\prime}dr^{\prime}d\varphi^{\prime}}{\sqrt{r^2 + r^{\prime 2} - 2rr^{\prime} \cos(\varphi-\varphi^{\prime}) + z^2 + \varepsilon^2_{\rm sg}}}$$ where $\cal{D}$ is the domain where $\Sigma$ does not vanish, and $\varepsilon_{\rm sg}$ is a small softening length required to avoid numerical divergences when $z=0$. From the surface density of our 2D simulations, $g_r(r,\varphi,z)$ and $g_{\varphi}(r,\varphi,z)$ can be conveniently calculated by fast Fourier transforms provided that $z$ is taken proportional to $r$, condition required for $g_r$ and $g_{\varphi}$ to read as convolution products (another necessary condition is that a grid with logarithmic radial spacing is used). The same condition applies to $\varepsilon_{\rm sg}$ for the calculation of $g_r$ and $g_{\varphi}$ at $z=0$. Writing $\varepsilon_{\rm sg} = B
r$ and $z = \eta r$, with $B$ and $\eta$ constants, the expressions for $g_r(r,\varphi,z)$ and $g_{\varphi}(r,\varphi,z)$ are given by Eqs. (A1) and (A3) of [@bm08b] with $B^2 \rightarrow B^2 +
\eta^2$. The vertical integral in Eq. (\[alpha\_sg\]) is then numerically calculated from the values of $g_r$ and $g_{\varphi}$ obtained with a range of values for $\eta$.
Acknowledgments {#acknowledgments .unnumbered}
===============
CB is supported by a Herchel Smith Postdoctoral Fellowship. FM is grateful for funding from the EC Sixth Framework Programme and also acknowledges the support of the German Research Foundation (DFG) through grant KL 650/8-2 within the Collaborative Research Group FOR 759: [*The formation of Planets: The Critical First Growth Phase*]{}. SJP is supported by an STFC Postdoctoral Fellowship. Numerical simulations were performed on the Pleiades and Grape clusters at U.C. Santa Cruz. Part of this project was done during the 2010 edition of the ISIMA summer school, whose support is gratefully acknowledged. We thank John Papaloizou and Andrew Youdin for useful discussions, Min-Kai Lin for detailed comments on a first draft of this manuscript, and the referee for useful comments.
S. A., [Papaloizou]{} J. C. B., 1999, [ApJ]{}, 521, 650
C., [Lin]{} D. N. C., 2010, [ApJ]{}, 709, 759
C., [Masset]{} F., 2008a, [ApJ]{}, 672, 1054
C., [Masset]{} F., 2008b, [ApJ]{}, 678, 483
A. C., [Hayfield]{} T., [Mayer]{} L., [Durisen]{} R. H., 2010, [Icarus]{}, 207, 509
A. P., 1997, Science, 276, 1836
A. P., 2005, [ApJ]{}, 629, 535
G., [R[ó]{}[ż]{}yczka]{} M., [Lin]{} D. N. C., [Bodenheimer]{} P., 2000, [ApJ]{}, 540, 1091
A. G. W., 1978, Moon and Planets, 18, 5
S., [Nayakshin]{} S., 2010, ArXiv e-prints
C. J., 2009, [MNRAS]{}, 396, 1066
P., [Lodato]{} G., [Clarke]{} C. J., 2009, [MNRAS]{}, 393, 1157
A., [Baruteau]{} C., [Kley]{} W., [Masset]{} F., 2009, [A&A]{}, 502, 679
A., [Masset]{} F., [Morbidelli]{} A., 2009, [ApJL]{}, 705, L148
A., [Morbidelli]{} A., [Masset]{} F., 2006, Icarus, 181, 587
C. F., 2001, [ApJ]{}, 553, 174
P., [Tremaine]{} S., 1980, [ApJ]{}, 241, 425
R., [Bodenheimer]{} P., 2011, [Icarus]{}, 211, 939
S., [Lin]{} D. N. C., 2004, [ApJ]{}, 604, 388
S., [Lin]{} D. N. C., 2008, [ApJ]{}, 673, 487
P., [Graham]{} J. R., [Chiang]{} E., [Fitzgerald]{} M. P., [Clampin]{} M., [Kite]{} E. S., [Stapelfeldt]{} K., [Marois]{} C., [Krist]{} J., 2008, Science, 322, 1345
D., [Jayawardhana]{} R., [van Kerkwijk]{} M. H., 2010, [ApJ]{}, 719, 497
A., [Bonnefoy]{} M., [Chauvin]{} G., [Apai]{} D., [Ehrenreich]{} D., [Boccaletti]{} A., [Gratadour]{} D., [Rouan]{} D., [Mouillet]{} D., [Lacour]{} S., [Kasper]{} M., 2010, Science, 329, 57
M.-K., [Papaloizou]{} J. C. B., 2010, [MNRAS]{}, 405, 1473
G., [Rice]{} W. K. M., 2004, [MNRAS]{}, 351, 630
D., [Kalnajs]{} A. J., 1972, [MNRAS]{}, 157, 1
M. N., [Inutsuka]{} S., [Matsumoto]{} T., 2011, [ApJ]{}, 729, 42
C., [Macintosh]{} B., [Barman]{} T., [Zuckerman]{} B., [Song]{} I., [Patience]{} J., [Lafreni[è]{}re]{} D., [Doyon]{} R., 2008, Science, 322, 1348
C., [Zuckerman]{} B., [Konopacky]{} Q. M., [Macintosh]{} B., [Barman]{} T., 2010, [Nature]{}, 468, 1080
F., 2000, [A&AS]{}, 141, 165
F. S., 2002, [A&A]{}, 387, 605
F. S., [Casoli]{} J., 2009, [ApJ]{}, 703, 857
F. S., [Casoli]{} J., 2010, [ApJ]{}, 723, 1393
F. S., [D’Angelo]{} G., [Kley]{} W., 2006, [ApJ]{}, 652, 730
F. S., [Papaloizou]{} J. C. B., 2003, [ApJ]{}, 588, 494
C. D., [Levin]{} Y., 2005, [ApJ]{}, 628, 817
L., [Quinn]{} T., [Wadsley]{} J., [Stadel]{} J., 2002, Science, 298, 1756
F., [Bate]{} M. R., 2011a, [MNRAS]{}, 411, L1
F., [Bate]{} M. R., 2011b, [MNRAS]{}, 410, 559
H., 1980, Progress of Theoretical Physics, 64, 544
S., 2010, [MNRAS]{}, 408, L36
R. P., [Papaloizou]{} J. C. B., 2004, [MNRAS]{}, 350, 849
S., [Baruteau]{} C., [Crida]{} A., [Kley]{} W., 2010, [MNRAS]{}, 401, 1950
S., [Baruteau]{} C., [Kley]{} W., 2011, [MNRAS]{}, 410, 293
S.-J., [Baruteau]{} C., [Meru]{} F., 2011, [MNRAS]{}, submitted
S.-J., [Papaloizou]{} J. C. B., 2008, [A&A]{}, 485, 877
S.-J., [Papaloizou]{} J. C. B., 2009, [MNRAS]{}, 394, 2283
A., [Hur[é]{}]{} J.-M., 2005, [A&A]{}, 433, L37
J. B., [Hubickyj]{} O., [Bodenheimer]{} P., [Lissauer]{} J. J., [Podolak]{} M., [Greenzweig]{} Y., 1996, Icarus, 124, 62
R. R., 2005, [ApJL]{}, 621, L69
W. K. M., [Armitage]{} P. J., 2009, [MNRAS]{}, 396, 2228
W. K. M., [Lodato]{} G., [Armitage]{} P. J., 2005, [MNRAS]{}, 364, L56
V. S., 1969, [Evoliutsiia doplanetnogo oblaka.]{}. Moscow: Nakua
D., [Whitworth]{} A. P., 2008, [A&A]{}, 480, 879
J. M., [Norman]{} M. L., 1992, [ApJS]{}, 80, 753
E. W., 2005, [ApJ]{}, 626, 1033
A., 1964, [ApJ]{}, 139, 1217
E. I., [Basu]{} S., 2006, [ApJ]{}, 650, 956
E. I., [Basu]{} S., 2010a, [ApJL]{}, 714, L133
E. I., [Basu]{} S., 2010b, [ApJ]{}, 719, 1896
Z., [Hartmann]{} L., [Gammie]{} C., 2009, [ApJ]{}, 694, 1045
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: See `fargo.in2p3.fr`.
[^3]: Similar gravitoturbulent disc models, but with reduced aspect ratio in a steady state, have been simulated by [@pbm11] with the same hydrodynamical code and similar grid resolution. Using the same method to calculate the stresses as described in the appendix, they find an averaged value of $\alpha_{\rm tot}$ in very good agreement with the value given in Eq. (\[alpha\_gammie\]).
[^4]: As in the case with $\beta=20$, this value of $\alpha$, given by viscous transport theory, is slightly smaller than what we measured in the gravitoturbulent disc model ($\langle\alpha_{\rm tot}\rangle \approx
1.8\%$). We have checked that a higher value of $\alpha$ does not change our results of laminar calculations, presented below.
| {
"pile_set_name": "ArXiv"
} |
{width="2"}
[ [**Center for Sustainable Engineering of Geological and Infrastructure Materials (SEGIM)**]{}\
\[0.1in\] Department of Civil and Environmental Engineering\
\[0.1in\] McCormick School of Engineering and Applied Science\
\[0.1in\] Evanston, Illinois 60208, USA ]{}
\
0.5in
\
\
[**SEGIM INTERNAL REPORT No. 15-09/047A**]{}\
\
[ **Roozbeh Rezakhani [^1] and Gianluca Cusatis [^2]** ]{}
[[**Abstract**]{}: Discrete fine-scale models, in the form of either particle or lattice models, have been formulated successfully to simulate the behavior of quasi-brittle materials whose mechanical behavior is inherently connected to fracture processes occurring in the internal heterogeneous structure. These models tend to be intensive from the computational point of view as they adopt an “a priori” discretization anchored to the major material heterogeneities (e.g. grains in particulate materials and aggregate pieces in cementitious composites) and this hampers their use in the numerical simulations of large systems. In this work, this problem is addressed by formulating a general multiple scale computational framework based on classical asymptotic analysis and that (1) is applicable to any discrete model with rotational degrees of freedom; and (2) gives rise to an equivalent Cosserat continuum. The developed theory is applied to the upscaling of the Lattice Discrete Particle Model (LDPM), a recently formulated discrete model for concrete and other quasi-brittle materials, and the properties of the homogenized model are analyzed thoroughly in both the elastic and inelastic regime. The analysis shows that the homogenized micropolar elastic properties are size-dependent, and they are functions of the RVE size and the size of the material heterogeneity. Furthermore, the analysis of the homogenized inelastic behavior highlights issues associated with the homogenization of fine-scale models featuring strain-softening and the related damage localization. Finally, nonlinear simulations of the RVE behavior subject to curvature components causing bending and torsional effects demonstrates, contrarily to typical Cosserat formulations, a significant coupling between the homogenized stress-strain and couple-curvature constitutive equations.]{}
*Keywords:* Asymptotic Expansion Homogenization; Asymptotic Expansion; Discrete Models; Lattice Models; Cosserat Continuum; Strain Softening; Size Effect.\
Introduction
============
Discrete fine-scale models, in the form of either particle or lattice models, have been formulated successfully in the literature to simulate the behavior of a variety of different materials. Their use has become more and more popular in the last few decades due to a number of appealing properties that make them advantageous compared to continuum based formulations.
The geometry of discrete models is built with reference to the actual internal structure of the material of interest and it consists of “particles” connected through either “contact points” or “connecting struts” (also called “lattice elements”). This “a priori” discretization allows simulating material heterogeneity efficiently in the case of materials - such as concrete, rock, sea-ice, and toughened ceramics - characterized by hard and stiff inclusions embedded in a more compliant, weak, and brittle, matrix. In addition, the intrinsic particle/lattice spacing automatically provides the formulation with an internal characteristic length which can be made randomly variable if the discrete model is constructed according to the actual random distribution of material heterogeneity.
The degrees of freedom (displacements and rotations) are defined only at a finite number of points – referred also as “nodes” thereinafter – which, depending on the formulation, may or may not correspond to the partice center of mass or particle centroid. Strain and stress measures are defined at a finite number of points coinciding with the contact points or with some specified points along the connecting struts. The constitutive behavior is formulated through vectorial, as opposed to tensorial, stress versus strain relationships and stress tractions are supposed to be distributed over either a “contact area” or the cross sectional area of the connecting struts (in this paper, this area will be generically referred to as “facet”). Finally, the classical concepts of equilibrium and compatibility are formulated through algebraic equations, instead of partial differential equations typical of continuum mechanics. One of the main advantages of discrete models is that the discreteness of the formulation permits handling naturally displacement discontinuities arising during damage localization and fracture processes.
Rigid particle models, under the name of Discrete Element Method (DEM), were first formulated to simulate both natural materials, such as geomaterials [@Cundall-1; @Serrano-1; @Cundall-3; @Kawai-1], as well as man-made materials like concrete [@Zubelewicz-3; @Plesha-1; @Zubelewicz-4]. A somewhat similar model is the rigid-body-spring model (RBSM), which subdivides the material domain into rigid polyhedral elements interconnected by zero-size springs [@Kawai-1; @Bolander-1; @Bolander-2; @Bolander-3].
Lattice models, pioneered by Hrennikoff [@Hrennikoff-1] to solve elastic problems in the pre-computers era, were later developed by many authors to model fracture in quasi-brittle materials in both 2D [@Schlangen-1], and 3D [@Cusatis-1; @Cusatis-2; @Lilliu-1; @Berton-1].
More recently, various discrete models, in the form of either lattice or particle models, have been quite successful in simulating concrete materials [@Lilliu-1; @cusatis-ldpm-1; @cusatis-ldpm-2; @Leite-1; @Kim-1]. For an extensive review of the currently available models for concrete the reader is directed to a recent special issue [@CementComposite] collecting several papers covering a wide variety of concrete mechanics phenomena spanning several length scales, from the scale of cement particles to that of reinforced concrete structural members.
In most applications of interest in practice, fine-scale models lead to fairly large computational systems characterized by a huge computational cost making their practical use rather limited. For example, the full-scale computational analysis of an average concrete bridge would require millions of degrees of freedom or the simulation of a rock formation would require billions of degrees of freedom. The solution of such large problems, although possible in principle with large super computer clusters, is unimaginable in everyday engineering practice. For this reason, many studies have been devoted to finding optimal and rigorous approaches for multiscale computation.
Among different multiscale techniques available in the literature [@Galvanetto-1], the ones based on homogenization theory have been widely used over the past decades. The homogenization theory relies on two main assumptions. The first is the existence of a certain volume of material, the so called Representative Volume Element (RVE) or Unit Cell (UC), carrying a complete description of the internal material structure [@Gitman-1; @Kouznetsova-1]. The second is that the size of such a volume is much smaller than the size of the overall solid volume under consideration. The latter is also known as the “scale separation” assumption.
Hill [@Hill-1], Eshelby [@Eshelby-1], Hashin and Strikman [@Hashin-1] pioneered analytical homogenization techniques which were developed later by other authors [@Christensen-1; @Nemat-Nasser-1]. Analytical homogenization is able to reasonably approximate material properties when the exact solution of the boundary value problem associated with the RVE problem can be obtained. However, in this approach, elastic behavior, small strains, and relatively simple internal structure are the limiting assumptions typically adopted. When complicated heterogeneous structures are considered, or constitutive behavior of constituents are nonlinear, other homogenization techniques [@Lopez-Pamies-1; @Lopez-Pamies-2] needs to be considered.
To overcome these difficulties, computational homogenization is often used in the literature [@Smith-1; @Feye-1; @Kouznetsova-1; @Miehe-1]. In this approach, a single RVE is assigned to each calculation point (e.g. gauss point in a Finite Element mesh) in the macro domain and at each step of the nonlinear analysis, macro-strain increments are imposed as essential boundary conditions to the RVE. The solution of the RVE boundary value problem is then averaged for the calculation of the associated macroscopic stress tensor. Since no assumption is made for the macroscopic constitutive law, this method can be used for materials featuring extremely nonlinear behavior.
A somewhat similar but more mathematically rigorous homogenization technique is the so-called Asymptotic Expansion Homogenization (AEH) that uses the asymptotic expansion of the displacement field based on a length parameter representing the ratio between the length scale of material heterogeneity and the macroscopic length scale. Starting from this expansion hierarchical boundary value problems are obtained at different scales. This approach can easily handle problems with multiple (more than 2) scales in both space and time [@Fish-3]; it does not make assumptions on the character of the macroscopic constitutive equations; and its implementation in computer codes is relatively simple.
Within the extensive literature on AEH, remarkable is the work of the following authors. Hassani [@Hassani-1; @Hassani-2] investigated formulation of homogenization theory and topology optimization and its numerical application to materials with periodic microstructure. Chung [@Chung-1] presented detailed derivation of multiple scale formulation for elastic solids. Fish employed this approach to study elastic as well as elasto-plastic composites [@Fish-2]. Ghosh [@Ghosh-1] adopted MH along with Voronoi Cell Finite Element Method (VCFEM) to study the behavior of composites with random meso-structure [@Ghosh-2]. More recently, Fish [@Fish-3] introduced the Generalized Mathematical Homogenization (GMH) to derive continuum constitutive equations starting from Molecular Dynamics (MD).
All the aforementioned work is relevant to Cauchy continuum formulations. However, homogenization schemes were also used for the multiscale analysis of Cosserat continuum models, in which an independent rotation field appears in addition to the displacement field. Feyel [@Feye-1] built a homogenization scheme to couple a Cauchy continuum formulation at the micro-scale giving rise to a Cosserat continuum formulation at the macro-scale. Asymptotic homogenization technique was employed by Forest [@Forest-1] for upscaling elastic Cosserat solids. In this work, the author studied various types of asymptotic expansions for the displacement and rotation fields and investigated their effect on the resulting macroscopic continuum behavior. Results of this investigation, showed that the nature of the homogenized continuum depends on the ratio of the Cosserat characteristic length of constituents, size of heterogeneity and typical size of the structure.
Chan et al. [@Chan-1] derived the governing constitutive equations for strain gradient elasticity for both homogeneous and functionally graded materials using the strain energy density function and the related definitions of the stress fields. They showed that additional terms appear in the equations that are related to the strain gradient nonlocality and the interaction between material nonhomogeneity. Bardenhagen et al. [@Bardenhagen-1] obtained a nonlinear higher order gradient continuum representation of discrete periodic micro-structures by means of an energy approach. The developed model was then employed to investigate the existence and stability of localization bands and their relationship to the model loss of ellipticity. Finally, homogenization of discrete atomic models into equivalent continuum can be found in publications where the authors exploited asymptotic analysis techniques [@Caillerie-1] and the mathematical $\Gamma$-convergence method [@Braides-1].
The present study derives a general multiscale homogenization scheme suitable for upscaling materials whose fine-scale behavior can be successfully approximated through the use of discrete models featuring both translational and rotational degrees of freedom.
The Fine-Scale Problem
======================
With reference to Figure \[TwoScaleAnalysis\]a, let us consider the interaction of two adjacent particles, $I$ and $J$, sharing a generic facet. If one limits the analysis to the case of small strains and displacements – which is a reasonable assumption in absence of large plastic deformation prior to fracture as observed in brittle and quasi-brittle materials – meaningful measures of deformation [@cusatis-ldpm-1] can be defined as
$$\label{eps}
\epsilon^{IJ}_{\alpha}=\frac{1}{r} \left(\mathbf{U}^J + \mb \Theta^{J} \times \mathbf{c}^{J} - \mathbf{U}^I - \mb{\Theta}^I \times \mathbf{c}^I \right) \cdot \mathbf{e}^{IJ}_{\alpha}$$
and
$$\label{curvature}
\chi^{IJ}_{\alpha}=\frac{1}{r} \left( \mb{\Theta}^{J} - \mb{\Theta}^I \right) \cdot \mathbf{e}^{IJ}_{\alpha}$$
where $\epsilon^{IJ}_{\alpha}=$ facet strains; $\chi^{IJ}_{\alpha}=$ facet curvatures; $r=|\mathbf{x}^{IJ}|$; $\mathbf{x}^{IJ}=\mathbf{x}^J-\mathbf{x}^I $ is the vector connecting the particle nodes $P_I$ and $P_J$; $\mathbf{e}^{IJ}_{\alpha}$ ($\alpha=N,M,L$) are unit vectors defining a facet Cartesian system of reference such that $\mathbf{e}^{IJ}_{N}=$ is orthogonal to the facet and $\mathbf{e}^{IJ}_{N} \cdot \mathbf{x}^{IJ} >0$; $\mathbf{U}^I$, $\mathbf{U}^J$ = displacement vectors of node $P_I$ and $P_J$; $\mb{\Theta}^{I}$, $\mb{\Theta}^{J}$ = rotation vectors of node $P_I$ and $P_J$; and $\mathbf{c}^{I}$, $\mathbf{c}^{J}$ = vectors connecting nodes $P_I$ and $P_J$ to the facet centroid, see Fig. \[TwoScaleAnalysis\]a. It must be observed here that displacements and rotations are assumed to be independent variables.
For given strain and curvature vectors, a vectorial constitutive equation provide stress, $\mathbf{t}^{IJ}$, and couple, $\mathbf{m}^{IJ}$, tractions on each facet. Formally one can write $\mathbf{t}^{IJ} = t_{{\alpha}}(\epsilon_{N}, ...) \mathbf{e}^{IJ}_{\alpha}$ and $\mathbf{m}^{IJ} = m_{{\alpha}}(\chi_{N}, ...) \mathbf{e}^{IJ}_{\alpha}$ where, in general, summation rule applies over $\alpha$. As an example, the elastic behavior can be formulated through the following equations
$$\label{elastic}
t_{\alpha} = E_\alpha \epsilon_{\alpha}; \hspace{0.25 in} m_{\alpha} = W_{\alpha} \chi_{\alpha}= E_\alpha \ell_\alpha^2 \chi_{\alpha}; \hspace{0.25 in} (\alpha=N,M,L)$$
in which each traction component is proportional to the associated strain or curvature (summation rule does not apply); and $E_\alpha$, $W_\alpha$ are fine-scale elastic constants which are related by a characteristic length $\ell_\alpha$. An example of nonlinear facet constitutive equations is reported in Appendix \[LDPM\], Section \[LDPM-Constitutive\], with reference to the so-called Lattice Discrete Particle Model (LDPM) that will be considered in the numerical examples. Finally, the computational discrete fine-scale framework is completed by imposing the equilibrium of each single particle subject to the effect of all surrounding particles. Translational and rotational dynamic equilibrium equations read
$$\label{motion-1}
M_u^I\ddot{\mathbf U}^I +\mathbf{M}_{u \theta}^I\ddot{\mb \Theta}^I - V^I \mathbf{b}^0=\sum_{\mathcal{F}_I} A \mathbf{t}^{IJ}$$
and
$$\label{motion-2}
\mathbf{M}_{\theta}^I\ddot{\mb \Theta}^I = \sum_{\mathcal{F}_I} A ( \mathbf{w}^{IJ} +\mathbf{m}^{IJ})$$
where $\mathbf{w}^{IJ} = \mathbf c^I \times \mathbf {t}^{IJ}$ is the moment of the traction $\mathbf {t}^{IJ}$ with respect to the particle node $P_I$; $\mathcal{F}_I$ is the set of facets surrounding node P$_I$ and obtained by collecting all the facets associated with each node pair $(I,J)$; $A$ = facet area; superimposed dots represent time derivatives; $V^I$ is the particle volume; $\mathbf{b}^0$ is the body force vector; $M_u^I$ = mass of node $P_I$; and $\mathbf{M}_{u \theta}^I$, $\mathbf{M}_{\theta}^I$ = moment of inertia tensors. It is worth observing that $\mathbf{M}_{u \theta}^I = \mathbf{0}$ and $\mathbf{M}_{\theta}^I = M_{\theta}^I \mathbf{I}$ if the particle node is the particle center of mass; the axes of the system of reference are parallel to the particle principal axes of inertia; and the principal moments of inertia are the same in all directions. These conditions, although applicable only to a limited number of cases (e.g. spherical particles), do not reduce the conceptual generality of the derivation that will be presented in this paper and will be assumed thereinafter for simplicity.
Asymptotic Expansion Homogenization {#HomogTheory}
=====================================
In this section, the two-scale homogenization of the general fine-scale problem introduced in the previous section is pursued by means of the approach proposed in Ref. [@Fish-3]. In the original formulation only central forces were assumed to act on the particles and, consequently, the rotational equilibrium equation was not considered.
Two Scale Approximation and Asymptotic Expansions
-------------------------------------------------
In order to perform a two-scale asymptotic expansion homogenization, a periodic discrete system, composed by a number of adjacent RVEs, is considered in this section. In Figure \[TwoScaleAnalysis\]b, the generic macroscopic material domain and the corresponding global coordinate system $\bf X$ are shown. At any point in the macroscopic domain, two separate length scales and the corresponding local coordinate systems, $\bf x$ and $\bf y$, are introduced to represent (1) the macroscopic domain, in which the problem is defined as homogeneous continuum with no detail of material heterogeneity, and (2) the meso-scale domain, in which heterogeneity is modeled by the discrete meso-scale model. Vector ${\mathbf}X$, as shown in the figure, is the vector connecting the origin of the global macroscopic coordinate system to the mass center of a generic RVE. In Figure \[TwoScaleAnalysis\]c, a zoomed view of the macroscopic material point is shown in the local meso-scale coordinate system $\bf{y}$, in which a representative volume of heterogeneous material is depicted. One should consider that in Figure \[TwoScaleAnalysis\]a, particles $I$ and $J$ are shown in the local macroscopic coordinate system $\bf x$. Therefore, they should be plotted in smaller size compared to Figure \[TwoScaleAnalysis\]c, but this was not done for the sake of clarity. If the separation of scales exists, one can write the following relationship linking macro and meso local coordinate systems $$\begin{aligned}
\label{scale-link-1}
\mathbf{x}=\eta \mathbf{y}; \hspace{0.25 in} 0< \eta <<1\end{aligned}$$
where $\eta$ is a very small positive scalar. In addition, the displacement of a generic node P$_I$, $\mathbf{U}^I = \mathbf{u}(\mathbf{x}^I, \mathbf{y}^I)$, can be approximated by means of the following asymptotic expansion
$$\mathbf{u}(\mathbf x, \mathbf y) \approx \mathbf u^0(\mathbf x, \mathbf y)+\eta \mathbf u^1(\mathbf x, \mathbf y)
\label{disp-expansion}$$
where only terms up to order $\mathcal{O}(\eta)$ are considered. Functions $\mathbf u^0(\mathbf x, \mathbf y)$, and $\mathbf u^1(\mathbf x, \mathbf y)$ are continuous with respect to $\mathbf{x}$ and discrete (i.e. defined only at finite number of points) with respect to $\mathbf{y}$.
In order to define the asymptotic expansion for rotations, it is convenient first to postulate the existence of a continuous displacement-like field ${\mathbf}{d}^\eta (\mathbf x)$ such that $2 {\mathbf}\Theta^I= \mb \nabla \times {\mathbf}{d}^\eta|_{\mathbf x =\mathbf x^I}$. If ${\mathbf}{d}^\eta (\mathbf x)$ is replaced by a two-scale approximation similar to the one in Equation \[disp-expansion\], one can write, ${\boldsymbol \Theta}^I = \mb{\theta}(\mathbf{x}^I, \mathbf{y}^I)$, and
$$\mb \theta (\mathbf x, \mathbf y) \approx \eta^{-1} \mb \omega^{0}(\mathbf x, \mathbf y) + \mb \varphi^0(\mathbf x, \mathbf y)+ \mb \omega^{1}(\mathbf x,\mathbf y)+\eta \mb\varphi^{1}(\mathbf x, \mathbf y)
\label{rot-expansion}$$
where $2 \boldsymbol\omega^{0} = \boldsymbol \nabla_y \times {\mathbf}{d}^0$; $2 \boldsymbol\varphi^{0} = \boldsymbol \nabla_x \times {\mathbf}{d}^0$; $2 \boldsymbol \omega^{1} = \boldsymbol \nabla_y \times {\mathbf}{d}^1$; $2 \boldsymbol \varphi^{1} = \boldsymbol \nabla_x \times {\mathbf}{d}^1$; and subscripts $x$ and $y$ identify the nabla operator in the coarse- and fine-scale, respectively. Thus, $\mb\omega^{0}$, $\mb\omega^{1}$ should be interpreted as rotations in the fine-scale whereas $\mb\varphi^{0}$, $\mb\varphi^{1}$ as the corresponding coarse-scale rotations. It is worth observing here that, contrarily to the expansion of displacements, the asymptotic expansion for rotations features a term of order $\mathcal{O}(\eta^{-1})$ and two distinct terms of order $\mathcal{O}(1)$.
In the macroscopic coordinate $\mathbf{x}$, the difference in position between nodes $P_I$ and $P_J$ can be considered as infinitesimal. Hence, in order to obtain the asymptotic expansion of strains and curvatures, it is convenient first to obtain the Taylor series expansion of displacement and rotation at nodes $P_J$ around point $P_I$ of coordinate $\mathbf{x}^I$ in the local coordinate system $\bf x$. By assuming that the displacement and rotation fields in Equations \[disp-expansion\] and \[rot-expansion\], are continuous and differentiable with respect to $\mathbf{x}$, one can write
$$\begin{aligned}
\label{taylor-1-J}
\begin{aligned}
U_i^J=u_i(\mathbf{x}^J,\mathbf{y}^J) = u_i^J+u^J_{i,j} \,x^{IJ}_j + \frac{1}{2}u^J_{i,jk}\, x^{IJ}_j x^{IJ}_k +\cdots
\end{aligned}\end{aligned}$$
$$\begin{aligned}
\label{taylor-2-J}
\begin{aligned}
\Theta_i^J=\theta_i(\mathbf{x}^J,\mathbf{y}^J)= \theta_i^J+\theta_{i,j}^J x^{IJ}_j +\frac{1}{2} \theta^J_{i,jk}\, x^{IJ}_j x^{IJ}_k +\cdots
\end{aligned}\end{aligned}$$
where $u_i^J=u_i(\mathbf{x}^I,\mathbf{y}^J)$; $ u^J_{i,j} = \partial u_i/\partial x_j (\mathbf{x}^I,\mathbf{y}^J)$; $u^J_{i,jk} =\partial^2 u_i/\partial x_j \partial x_k (\mathbf{x}^I,\mathbf{y}^J)$; $\theta_i^J= \theta_i(\mathbf{x}^I,\mathbf{y}^J)$; $ \theta^J_{i,j} = \partial \theta_i/\partial x_j (\mathbf{x}^I,\mathbf{y}^J)$; $\theta^J_{i,jk} =\partial^2 \theta_i/\partial x_j \partial x_k (\mathbf{x}^I,\mathbf{y}^J)$; $x^{IJ}_j$ is a vector connecting node $P_I$ to node $P_J$ in the $\mathbf{x}$ space. By substituting Equations \[disp-expansion\] and \[rot-expansion\] into Equation \[eps\], and using the Taylor expansion of displacement and rotation of node $P_J$ around node $P_I$ (Equations \[taylor-1-J\] and \[taylor-2-J\]) one obtains the multiple scale definition of facet strains (see Appendix \[exp-strains-details\] for details)
$$\label{eps-expansion}
\epsilon_{\alpha}=\eta^{-1} \epsilon_{\alpha}^{-1} + \epsilon_{\alpha}^0 + \eta \epsilon_{\alpha}^1$$
where
$$\label{eps-expansion-minus}
\epsilon_{\alpha}^{-1} = \bar{r}^{-1} \bigg[ u_i^{0J} - u^{0I}_i + \varepsilon_{ijk} \omega_j^{0J} \bar c_{k}^{J} - \varepsilon_{ijk} \omega_j^{0I} \bar c_{k}^{I} \bigg]e^{IJ}_{\alpha i}$$
$$\label{eps-expansion-zero}
\epsilon_{\alpha}^0 =\bar{r}^{-1} \bigg[ u_i^{1J} + u^{0J}_{i,j} y^{IJ}_j - u^{1I}_i + \varepsilon_{ijk} \bigg( \varphi_j^{0J} + \omega_j^{1J} + \omega_{j,m}^{0J} y^{IJ}_m \bigg) \bar c_{k}^{J} - \varepsilon_{ijk} \bigg( \varphi_j^{0I} + \omega_j^{1I} \bigg) \bar c_{k}^{I} \bigg] e^{IJ}_{\alpha i}$$
$$\label{eps-expansion-plus}
\begin{split}
\epsilon_{\alpha}^1 = \bar{r}^{-1} \bigg[ u^{1J}_{i,j} y^{IJ}_j + \frac{1}{2}u^{0J}_{i,jk} y^{IJ}_j y^{IJ}_k + \varepsilon_{ijk} \bigg( \varphi_j^{1J} + \varphi_{j,m}^{0J} y^{IJ}_m + \omega_{j,m}^{1J} y^{IJ}_m + \frac{1}{2} \omega^{0J}_{j,mn} y^{IJ}_m y^{IJ}_n \bigg) \bar c_{k}^{J} - \varepsilon_{ijk} \varphi_j^{1I} \bar c_{k}^{I} \bigg] e^{IJ}_{\alpha i}
\end{split}$$
In the previous equations, $\varepsilon_{ijk}$ is the Levi-Civita permutation symbol and length type variables have been changed into their $\mathcal{O}(1)$ counterparts by using Equation \[scale-link-1\]: $r=\eta \bar{r}$, $c_{k}^I=\eta \bar{c}_{k}^I$, $c_{k}^J=\eta \bar{c}_{k}^J$.
Similarly, multiple scale definition of facet curvature can be calculated as (see Appendix \[exp-strains-details\] for details)
$$\label{curv-expansion}
\eta \chi_{\alpha}=\eta^{-1} \psi_{\alpha}^{-1} + \psi_{\alpha}^{0} + \eta \psi_{\alpha}^1$$
where
$$\label{curv-expansion-minus2}
\psi_{\alpha}^{-1} =\bar{r}^{-1} \bigg[ \omega_i^{0J} - \omega_i^{0I} \bigg] {e}^{IJ}_{\alpha i}$$
$$\label{curv-expansion-minus1}
\psi_{\alpha}^{0} = \bar{r}^{-1} \bigg[ \varphi_i^{0J}+ \omega_i^{1J} + \omega_{i,j}^{0J} y^{IJ}_j - \varphi_i^{0I} - \omega_i^{1I} \bigg] {e}^{IJ}_{\alpha i}$$
$$\label{curv-expansion-zero}
\psi_{\alpha}^1 = \bar{r}^{-1} \bigg[ \varphi_i^{1J} + \omega_{i,j}^{1J} y^{IJ}_j + \varphi_{i,j}^{0J} y^{IJ}_j + \frac{1}{2} \omega_{i,jk}^{0J} y^{IJ}_j y^{IJ}_k - \varphi_i^{1I} \bigg] {e}^{IJ}_{\alpha i}$$
It is worth noting that in this section as well as in the rest of the paper superscript $IJ$ has been dropped when the permutation of $I$ and $J$ is not associated with a sign change.
Multiple-Scale Equilibrium Equations
------------------------------------
In order to obtain the correct scale separation of the governing equations, a rescaling of the discrete equilibrium equations needs to be performed. For the sake of simplicity, and since only quasi-static problems are concerned in the current research, it is assumed $\mathbf M^I_{u \theta} = 0$ and $\mathbf{M}_{\theta}^I = M_{\theta}^I \mathbf{I}$ on the left hand side of Equation \[motion-1\]. Rescaling is pursued by assuming that the material density, mass per unit volume, is of order zero: $\rho \sim \mathcal{O}(1)$, which along with the displacement asymptotic expansion implies that the left-hand-side of Equation \[motion-1\] is $\sim \mathcal{O}(\eta^3)$. By dividing both sides of Equation \[motion-1\] by $\eta^3$, and considering that all length variables should be considered $\sim \mathcal{O}(\eta^1)$, one obtains
$$\label{motion-1-rescaled}
\bar{M}_u^I\ddot { {\mathbf u}}^I - \bar{V}^I \mathbf{b}^0=\eta^{-1}\sum_{\mathcal{F}_I}{\bar{A}\, t_{\alpha} {\mathbf e}_\alpha^{IJ}}$$
where $\bar{M}_u^I = M_u^I/\eta^3$, $\bar{V}^I=V^I/\eta^3$ , $\bar{A} = A/\eta^2$ are all quantities $\sim \mathcal{O}(1)$. For reason of dimensionality, body forces $b^{0}_i$ can be always assumed to be proportional to gravity $\rho g$ and, consequently, they can be considered $\mathcal{O}(1)$ quantities as well.
One can rescale the rotational equation in a similar fashion by recognizing that, according to the previous discussion, the rotational moment of inertia is $\sim \mathcal{O}(\eta^5)$. Dividing both sides of Equation \[motion-2\] by $\eta^4$ one obtains
$$\label{motion-2-rescaled}
\eta \bar {M_\theta^I}\ddot{\mb \theta}^I=\eta^{-1} \sum_{\mathcal{F}_I} \bar{A}\, (\eta^{-1} w_{\alpha}{\mathbf e}_\alpha^{IJ} + \eta^{-1} m_{\alpha}{\mathbf e}_\alpha^{IJ})$$
where $\bar{M}_\theta^I = M_\theta^I/\eta^5$ is $\sim \mathcal{O}(1)$.
In the elastic regime one can write: $t_{\alpha} = \eta^{-1} t^{-1}_{\alpha} + t^{0}_{\alpha} + \eta t^{1}_{\alpha}$; where $t^{(\cdot)}_{\alpha} = E_\alpha \epsilon^{(\cdot)}_\alpha$, and $E_\alpha$ is assumed to be $\sim \mathcal{O}(1)$. In addition, $q_\alpha = \eta^{-1} m_{\alpha} = \eta^{-1} q^{-1}_{\alpha} + q^{0}_{\alpha} + \eta q^{1}_{\alpha}$ in which $q^{(\cdot)}_{\alpha} =\bar{W}_\alpha \psi^{(\cdot)}_\alpha$; $\bar{W}_\alpha=E_\alpha {\bar{\ell}}^{2}_\alpha$; and $\bar{\ell} = \ell / \eta$. Finally, $p_{\alpha} = \eta^{-1}w_{\alpha} = \eta^{-1} p^{-1}_{\alpha} + p^{0}_{\alpha} + \eta p^{1}_{\alpha}$ where $p^{(\cdot)}_{\alpha} {\mathbf e}_\alpha^{IJ} = \bar{\mathbf{c}}^I \times t^{(\cdot)}_{\alpha} {\mathbf e}_\alpha^{IJ}$. Since $w_\alpha$ and $m_\alpha$ are moments, it is reasonable that the asymptotic expansion of those variables divided by $\eta$ is similar to the one for tractions $t_\alpha$, considering that length type variables are considered to be $\sim \mathcal{O}(\eta)$.
Introducing these traction expressions along with the asymptotic definition of displacement and rotation fields Equations \[disp-expansion\] and \[rot-expansion\] (which also imply $\eta \ddot{\mb \theta}^{0I} =\ddot{\mb \omega}^{0I}+\mathcal{O}(\eta)$), into the rescaled equilibrium equations leads to
$$\label{motion-1-sep}
\eta^{-2} \sum_{\mathcal{F}_I}{\bar{A}\, t^{-1}_{\alpha} {\mathbf e}_\alpha^{IJ}}+ \eta^{-1} \sum_{\mathcal{F}_I}{\bar{A}\, t^{0}_{\alpha} {\mathbf e}_\alpha^{IJ}} + \sum_{\mathcal{F}_I}{\bar{A}\, t^1_{\alpha} {\mathbf e}_\alpha^{IJ}} - \bar{M}_u^I\ddot { {\mathbf u}}^{0I} + \bar{V}^I \mathbf{b}^0+\mathcal{O}(\eta) = \mathbf{0}$$
and
$$\begin{aligned}
\label{motion-2-sep}
\begin{aligned}
\eta^{-2}\sum_{\mathcal{F}_I} \bar{A}\, (p^{-1}_{\alpha}{\mathbf e}_\alpha^{IJ} + q^{-1}_{\alpha}{\mathbf e}_\alpha^{IJ}) + \eta^{-1}\sum_{\mathcal{F}_I} \bar{A}\, (p^{0}_{\alpha}{\mathbf e}_\alpha^{IJ} + q^{0}_{\alpha}{\mathbf e}_\alpha^{IJ}) \\ - \bar {M_\theta^I}\ddot{\mb \omega}^{0I} + \sum_{\mathcal{F}_I} \bar{A}\, (p^{1}_{\alpha}{\mathbf e}_\alpha^{IJ} + q^{1}_{\alpha}{\mathbf e}_\alpha^{IJ}) + \mathcal{O}(\eta) = \mathbf{0}
\end{aligned}\end{aligned}$$
in which terms of different orders are gathered together. The multiple scale equations reported above can also be used for nonlinear constitutive equations provided that facet tractions and facet moments can be expressed through the multiple scale decomposition exploited above. It will be shown later in the paper that this can be indeed achieved under some reasonable assumptions.
The RVE Problem {#RVEproblem}
---------------
Let’s first consider the equilibrium equations at the $\mathcal{O}(\eta^{-2})$ scale. From Equations \[motion-1-sep\] and \[motion-2-sep\], it is evident that the $\mathcal{O}(\eta^{-2})$ equilibrium equations represent the equilibrium of all particles in the RVE subjected to the stress tractions $t_\alpha^{-1}$ and the moment tractions $q_\alpha^{-1}$ and without any applied external load. Consequently, solution of the $\mathcal{O}(\eta^{-2})$ problem implies $t_\alpha^{-1}=0$ and $q_\alpha^{-1}=0$, which in turn, leads to $\epsilon_{\alpha}^{-1}=0$ and $\psi_{\alpha}^{-1}=0$. By taking into consideration the definitions of $\epsilon^{-1}_{\alpha}$ and $\psi^{-1}_{\alpha}$ (Equations \[eps-expansion-minus\] and \[curv-expansion-minus2\]) such result indicates that the $\mathcal{O}(\eta^{-2})$ problem represents a rigid body rototranslation of the RVE. This can be expressed as
$$\label{U0}
u_i^0(\mathbf{X}, \mathbf{y}) = v_i^0(\mathbf{X}) + \varepsilon_{ijk} y_k \omega_j^{0}(\mathbf{X})$$
in which the fields $\mathbf{v}^0$ and $\mb \omega^{0}$ are only dependent on macroscopic coordinate system $\bf X$, i.e. these quantities varies smoothly in the macro-scale material domain; they do not change within the RVE domain; and they can be calculated when kinematic boundary conditions are specified for the $\mathcal{O}(\eta^{-2})$ problem. These boundary conditions must describe the physical fact that the RVE is attached to a point in the macroscopic continuum. Hence, $\mathbf{v}^0$ must correspond to the macroscopic displacement field, and $\mb \omega^{0}$ must be equal to the macroscopic rotation field: $\mb \varphi^{0} = \mb \omega^{0}$. Since $\mb \omega^{0}$ is constant over the RVE, then $\mb \varphi^{0}$ is also constant in the RVE.
On the basis of Equation \[U0\] and the discussion above, one can rewrite the $\mathcal{O}(1)$ strains and curvatures as (See Appendix \[Revised-strain-curvature\] for details) $$\label{eps-expansion-zero'}
\epsilon_{\alpha}^0 =\bar{r}^{-1} \left( u_i^{1J} - u^{1I}_i + \varepsilon_{ijk} \omega_j^{1J} \bar c_{k}^{J} - \varepsilon_{ijk} \omega_j^{1I} \bar c_{k}^{I}\right) {e}^{IJ}_{\alpha i} + P^\alpha_{ij} \left(\gamma_{ij}
+ \varepsilon_{jmn} \kappa_{im} y_{n}^{c} \right) $$ $$\label{curv-expansion-minus1'}
\psi_{\alpha}^{0} = \bar{r}^{-1} \left( \omega_i^{1J}- \omega_i^{1I} \right) {e}^{IJ}_{\alpha i}+ P^\alpha_{ij} \kappa_{ij}$$
where $\gamma_{ij} = v^{0}_{j,i} - \varepsilon_{ijk} \omega_k^{0}$, $\kappa_{ij}=\omega_{j,i}^{0}$ are the macroscopic Cosserat strain and curvature tensors, respectively. The vector ${\mathbf}{y}^c$ is the position vector of the centroid of the common facet between particle $I$ and $J$ and $P^\alpha_{ij} = n^{IJ}_i e^{IJ}_{\alpha j}$ is a projection operator. Comparing the first term of Equation \[eps-expansion-zero’\] with Equation \[eps\], it can be concluded that this term is the lower scale definition of the three components of the facet strains (one normal and two tangential) written in terms of fine-scale displacements and rotations $u^1$ and $\omega^1$. The second term of Equation \[eps-expansion-zero’\], $P^\alpha_{ij} \left(\gamma_{ij} + \varepsilon_{jmn} \kappa_{im} y_{n}^{c} \right)$, is the projection of macroscopic Cosserat strain and curvature tensors on each facet. Similarly, Equation \[curv-expansion-minus1’\] shows that the $\mathcal{O}(1)$ curvature includes a fine-scale term (see Equation \[curvature\]), which depends on fine-scale rotation term $\omega^1$, and a coarse-scale term corresponding to the projection of macroscopic curvature tensor on each facet. Therefore, Equations \[eps-expansion-zero’\] and \[curv-expansion-minus1’\] express the $\mathcal{O}(1)$ facet strains and curvatures as the sum of their fine-scale counterparts and the projection of macroscopic strain and curvature tensors onto the facet level. It is worth nothing that the projection operator $P^\alpha_{ij}$ corresponds exactly to the one used in the microplane model [@Bazant-2; @xinwei-1] if $ e^{IJ}_{N i}\equiv n^{IJ}_i $, i.e. the discrete model is formulated in such a way the facets are orthogonal to the associated lattice struts. In addition, it must be noted that the term $\varepsilon_{jmn} \kappa_{im} y_{n}^{c}$ transforms the macroscopic curvature tensor, which is constant over the RVE, to different strain values at different positions $y^c_n$ inside the RVE, which is then projected on the facets through the operator $P^\alpha_{ij}$. Expanding this term for different components of curvature tensor, it can be shown that it perfectly corresponds to the strain field generated by curvatures in classical beam theories.
Strains and curvatures of order $\mathcal{O}(\eta)$ can also be rewritten by taking into account Equation \[U0\]. One gets
$$\label{eps-expansion-plus'}
\begin{aligned}
\epsilon_{\alpha}^1 = \bar{r}^{-1} \bigg[& u^{1J}_{i,j} y^{IJ}_j + \varepsilon_{ijk} \varphi_j^{1J} \bar c_{k}^{J} + \varepsilon_{ijk} \omega_{j,m}^{1J} y^{IJ}_m \bar c_{k}^{J} - \varepsilon_{ijk} \varphi_j^{1I} \bar c_{k}^{I} \\
&+ \frac{1}{2}v^{0}_{i,jk} y^{IJ}_j y^{IJ}_k + \frac{1}{2} \varepsilon_{ijk} \omega^{0}_{j,mn} y^{IJ}_m y^{IJ}_n y_{k}^{c} + \varepsilon_{ijk} \omega_{j,m}^{0} y^{IJ}_m \bar c_{k}^{J} \bigg] e^{IJ}_{\alpha i}
\end{aligned}$$
$$\label{curv-expansion-zero'}
\psi_{\alpha}^1 = \bar{r}^{-1} \left[ \varphi_i^{1J} + \omega_{i,j}^{1J} y^{IJ}_j - \varphi_i^{1I} + \omega_{i,j}^{0} y^{IJ}_j + \frac{1}{2} \omega_{i,jk}^{0J} y^{IJ}_j y^{IJ}_k \right] {e}^{IJ}_{\alpha i}$$
Detailed mathematical derivation of Equations \[eps-expansion-zero’\] through \[curv-expansion-zero’\] is provided in Appendix \[Revised-strain-curvature\].
In the previous derivation, where linear elastic behavior was assumed, the equilibrium equations at the $\mathcal{O}(\eta^{-2})$ scale were shown to represent the rigid body motion conditions for the RVE and, consequently, they led to zero strains, $\epsilon_{\alpha}^{-1}$, curvatures, $\psi_{\alpha}^{-1}$, tractions, $t_\alpha^{-1}$, and moments, $p_\alpha^{-1}$, and $q_\alpha^{-1}$, at the $\mathcal{O}(\eta^{-1})$ scale. These conditions can be reasonably assumed *a priori* in the case of nonlinear material behavior. In this case case one may write $t_\alpha = t_\alpha(\epsilon_{\beta}^{0}+\eta \epsilon_{\beta}^{1})$; $p_\alpha = p_\alpha(\epsilon_{\beta}^{0}+\eta \epsilon_{\beta}^{1})$, and $q_\alpha = q_\alpha(\eta^{-1}\psi^0_\beta+\psi^1_\beta)$ in which $\alpha,~\beta = N,M,L$. Since $\eta$ is a small quantity, one can also write the Taylor expansion of $t_\alpha$ and $p_\alpha$ around the $\mathcal{O}(1)$ component of strain and the Taylor expansion of $q_\alpha$ around the $\mathcal{O}(\eta^{-1})$ component of curvature:
$$\label{TaylorExpansions}
\begin{split}
& t_\alpha = t_\alpha(\epsilon_{\beta}^{0}+\eta \epsilon_{\beta}^{1}) = t_\alpha(\epsilon_{\beta}^{0}) + \eta \frac{\partial t_\alpha(\epsilon_{\beta}^{0})}{ \partial \epsilon^0_\gamma } \epsilon_\gamma^1 \\
& p_\alpha = p_\alpha(\epsilon_{\beta}^{0}+\eta \epsilon_{\beta}^{1}) = p_\alpha(\epsilon_{\beta}^{0}) + \eta \frac{\partial p_\alpha(\epsilon_{\beta}^{0})}{ \partial \epsilon^0_\gamma } \epsilon_\gamma^1 \\
& q_\alpha = q_\alpha(\eta^{-1}\psi^0_\beta+\psi^1_\beta) = q_\alpha(\eta^{-1}\psi^0_\beta) + \eta \frac{\partial q_\alpha(\eta^{-1}\psi^0_\beta)}{ \partial \psi^0_\gamma}\psi_\gamma^1
\end{split}$$
which can be rewritten as $t_{\alpha} = t^{0}_{\alpha} + \eta t^{1}_{\alpha}$; $p_{\alpha} = p^{0}_{\alpha} + \eta p^{1}_{\alpha}$; $q_{\alpha} = q^{0}_{\alpha} + \eta q^{1}_{\alpha}$, with the following conditions
$$\label{ZeroOne-Terms-Def}
\begin{gathered}
t_\alpha^0=t_\alpha(\epsilon^0_\beta); ~~~ p_\alpha^0=p_\alpha(\epsilon^0_\beta); ~~~ q_\alpha^0=q_\alpha(\eta^{-1} \psi^0_\beta); \\
t_\alpha^1=\frac{\partial t^0_\alpha}{ \partial \epsilon^0_\gamma } \epsilon_\gamma^1; ~~~ p_\alpha^1=\frac{\partial p^0_\alpha}{ \partial \epsilon^0_\gamma } \epsilon_\gamma^1; ~~~ q_\alpha^1=\frac{\partial q^0_\alpha}{ \partial \psi^0_\gamma }\psi_\gamma^1
\end{gathered}$$
This demonstrates that Equations \[motion-1-sep\], and \[motion-2-sep\] are valid also in the case of nonlinear material behavior under the assumption that traction and moments at the $\mathcal{O}(\eta^{-1})$ scale are zero as required, in the linear case, by the rigid body motion of the RVE.
The RVE problem is governed by the $\mathcal{O}(\eta^{-1})$ terms in Equations \[motion-1-sep\] and \[motion-2-sep\]. Considering those terms and scaling back all the variables, one can write the $\mathcal{O}(\eta^{-1})$ equations as
$$\label{RVE-1}
\sum_{\mathcal{F}_I}{{A}\, t^{0}_{\alpha} {{\mathbf}e}_\alpha^{IJ}} = 0; \hspace{0.25 in} \sum_{\mathcal{F}_I} {A}\, ({\mathbf}{c}^I \times t^{0}_{\alpha} {\mathbf}{e}_\alpha^{IJ} + m^{0}_{\alpha}{{\mathbf}e}_\alpha^{IJ}) = 0$$
Equations \[RVE-1\] are force and moment equilibrium equations of each single particle inside the RVE subjected to $\mathcal{O}(1)$ facet traction $t_\alpha^0$ and moment $m_\alpha^0$ vectors, which, in turn, are functions of $\epsilon^0_\alpha$ and $\psi^0_\alpha$, consisting of a coarse-scale and a fine-scale term (see Equations \[eps-expansion-zero’\] and \[curv-expansion-minus1’\]). In other words, Equations \[RVE-1\] state that the macroscopic strain, $\gamma_{ij} = v^{0}_{j,i} - \varepsilon_{ijk} \omega_k^{0}$, and curvature, $\kappa_{ij}=\omega_{j,i}^{0}$, tensors should be applied on all RVE facets as negative eigenstrains, and the fine-scale solution, in terms of displacements $u^1_i$ and rotations $\omega^1_i$ of each particle, must be calculated satisfying its force and moment equilibrium equations, while periodic boundary conditions are enforced on the RVE. The solution of the equilibrium equations also provides facet traction $t_\alpha^0$ and moment $m_\alpha^0$ vectors that are later used to compute the macroscopic stress and couple tensors.
The Macroscopic Problem {#macro-derivation}
-----------------------
Finally, let us consider the $\mathcal{O}(1)$ equilibrium equations in Equations \[motion-1-sep\] and \[motion-2-sep\]. The $\mathcal{O}(1)$ translational equilibrium equation for each particle in the RVE reads
$$\label{macro-1}
{M}_u^I \ddot {u}_i^{0I} = \eta \sum_{\mathcal{F}_I}{{A} \frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + {V}^I {b}^{0}_i$$
where all the variables have been scaled back in the original system of reference, and $t_i^{IJ}=t^0_\beta e_{\beta i}^{IJ}$. By using Equation \[U0\] and by averaging the contribution of all particles in the RVE, one can write (see Appendix \[MacroEquil-Derivation\] for details)
$$\label{macro-1-1-averaged}
\rho_u \ddot {v}_i^{0} = \frac{1}{V_0}\sum_I \sum_{\mathcal{F}_I}{\eta A \frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + b_i$$
where $V_0$ is the volume of the RVE; $\rho_u=\sum_I {M}^I_u/V_0$ is the mass density of the macroscopic continuum; $b_i=b^0_i (1 - \phi)$; and $\phi= 1 -\sum_I {V}^I/V_0$ is the porosity of the macroscopic continuum. Equation \[macro-1-1-averaged\] was derived under the assumption that $\sum_I {M}_u^I y^{I}_i=0$, which corresponds to the assumption that the local system of reference is the mass centroid of the particle system within the RVE.
Before proceeding with the derivations, let’s take a closer look at the definition of $\epsilon^{1}_{\alpha}$ and the term $(\partial {t}^{IJ}_{i}/\partial \epsilon^0_{\alpha}) \epsilon^1_{\alpha}$ on the RHS of Equation \[macro-1-1-averaged\]. Each facet in the material domain is shared between two particles, say $I$ and $J$. Therefore, by summing up the contributions of two adjacent particles, one obtains
$$\begin{aligned}
\label{higher-strain-1}
\begin{aligned}
\frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha} + \frac {\partial {t}^{JI}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha} = & \frac{1}{\bar{r}} \frac {\partial {t}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \bigg[ \bigg( u^{1J}_{n,j} y^{IJ}_j + \varepsilon_{njk} \varphi_j^{1J} \bar c_{k}^{J} + \varepsilon_{njk} \omega_{j,m}^{1J} y^{IJ}_m \bar c_{k}^{J} - \varepsilon_{njk} \varphi_j^{1I} \bar c_{k}^{I} \\
& \hspace{0.5 in} + \frac{1}{2} v^{0}_{n,jk} y^{IJ}_j y^{IJ}_k + \frac{1}{2} \varepsilon_{njk} \omega^{0}_{j,mo} y^{IJ}_m y^{IJ}_o y_{k}^{c} + \varepsilon_{njk} \omega_{j,m}^{0} y^{IJ}_m \bar c_{k}^{J} \bigg) e^{IJ}_{\alpha n} \bigg] \\
& + \frac{1}{\bar{r}} \frac {\partial {t}_{i}^{JI}}{\partial \epsilon^0_{\alpha}} \bigg[ \bigg( u^{1I}_{n,j} y^{JI}_j + \varepsilon_{njk} \varphi_j^{1I} \bar c_{k}^{I} + \varepsilon_{njk} \omega_{j,m}^{1I} y^{JI}_m \bar c_{k}^{I} - \varepsilon_{njk} \varphi_j^{1J} \bar c_{k}^{J} \\
& \hspace{0.5 in} + \frac{1}{2} v^{0}_{n,jk} y^{JI}_j y^{JI}_k + \frac{1}{2} \varepsilon_{njk} \omega^{0}_{j,mo} y^{JI}_m y^{JI}_o y_{k}^{c} + \varepsilon_{njk} \omega_{j,m}^{0} y^{JI}_m \bar c_{k}^{I} \bigg) e^{JI}_{\alpha n} \bigg]
\end{aligned}\end{aligned}$$
Considering the definition of the vector $\mathbf{y} ^{IJ} = \mathbf{y} ^{J} - \mathbf{y} ^{I}$, one can write $y_m^{IJ} = - y_m^{JI}$ and $\bar c^I_k-\bar c^J_k=y_k^{IJ}$. In addition, $e^{IJ}_{\alpha i} = - e^{JI}_{\alpha i}$ and $t_{i}^{IJ} = -t_{i}^{JI}$ hold for each facet. Finally, the sign of $\epsilon^{0}_\alpha$ does not change by interchanging $I$ and $J$ in its definition. This leads to $\partial t_{i}^{IJ}/\partial \epsilon^{0}_\alpha = - \partial t_{i}^{JI}/\partial \epsilon^{0}_\alpha$. Taking all above facts into account, Equation \[higher-strain-1\] can be written as
$$\begin{aligned}
\label{higher-strain-2}
\begin{aligned}
\frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha} + \frac {\partial {t}^{JI}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha} = \frac{1}{\bar{r}}\frac {\partial {t}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \bigg[ y^{IJ}_m \bigg( & u^{1J}_{n,m} - u^{1I}_{n,m} + \varepsilon_{njk} \omega_{j,m}^{1J} \bar c_{k}^{J} - \varepsilon_{njk} \omega_{j,m}^{1I} \bar c_{k}^{I} \\
& + v^{0}_{n,jm} y^{IJ}_j - \varepsilon_{njk} \omega_{j,m}^{0} + \varepsilon_{njk} \omega^{0}_{j,mo} y^{IJ}_o y_{k}^{c} \bigg) e^{IJ}_{\alpha n} \bigg] \\
\end{aligned}\end{aligned}$$
Comparing the expression inside the bracket on the RHS of Equation \[higher-strain-2\] to the definition of $\epsilon^0_{\alpha}$ in Equation \[eps-expansion-zero’\], it can be concluded that
$$\begin{aligned}
\label{higher-strain-3}
\begin{aligned}
\frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha} + \frac {\partial {t}^{JI}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha} = \frac {\partial {t}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \frac {\partial \epsilon^0_{\alpha}}{\partial x_m} y_m^{IJ} = \frac {\partial {t}_{i}^{IJ}}{\partial x_m} y_m^{IJ}
\end{aligned}\end{aligned}$$
Therefore, one can average the term $(\partial {t}^{IJ}_{i}/\partial \epsilon^0_{\alpha}) \epsilon^1_{\alpha}$ on each facet and replace it with $1/2(\partial {t}_{i}^{IJ}/\partial x_m) y_m^{IJ}$ in the equilibrium Equations \[macro-1-1-averaged\], which can be rewritten as
$$\label{macro-1-2-averaged}
\rho_u \ddot {v}_i^{0} = \frac{1}{2V_0} \sum_I \sum_{\mathcal{F}_I}{A} r \frac{\partial {t}^{IJ}_{i}}{\partial x_j} n_j^{IJ} + b_i$$
Finally, by considering that (1) $\partial (t^{IJ}_i n^{IJ}_j )/ \partial x_j = \partial t^{IJ}_i/ \partial x_j n^{IJ}_j + t^{IJ}_i \partial n^{IJ}_j/ \partial x_j$ and (2) $\partial n^{IJ}_j/ \partial x_j=0$ for the periodicity of the problem; and by recalling that $t_i^{IJ}=t^0_\alpha e_{\alpha i}^{IJ}$, one obtains
$$\label{macro-eq-cont}
\rho_u \ddot{v}^0_i = \sigma^0_{ji,j} + b_i$$
and
$$\label{macro-stress-formula}
\sigma^0_{ij} = \frac{1}{2V_0} \sum_I \sum_{\mathcal{F}_I}{A} r t^0_\alpha P_{i j}^{\alpha}$$
Equation \[macro-eq-cont\] is the classical partial differential equation governing the equilibrium of continua whereas Equation \[macro-stress-formula\] provides the macroscopic stress tensor by averaging the solution of the RVE problem. It is worth mentioning that Equation \[macro-stress-formula\] coincides with the virial stress formula for atomistic systems derived in Ref. [@Fish-3], but it is also equivalent to the averaging formula used in the classical microplane model [@Bazant-2] formulation and derived through an energetic equivalence.
The $\mathcal{O}(1)$ moment equilibrium equation is considered next. Since the purpose in this section is to average the equation of motion of all particles inside the RVE and derive the macroscopic equilibrium equation governing the entire RVE, to have a consistent formulation for all particles and RVEs, one must consider the moment of all forces with respect to a fixed point in space.
For the generic particle $I$, by taking the moment of all forces with respect to the origin of a global macroscopic coordinate system as shown in Figure \[TwoScaleAnalysis\]b and by considering the results of the $\mathcal{O}(\eta^{-2})$ problem, one can write (see Appendix \[MacroEquil-Derivation\] for details)
$$\label{macro-2}
{M}_u^I \varepsilon_{ijk} X^{I}_j \left( \ddot {v}_k^{0} + \varepsilon_{kmn} \eta^{-1} \ddot {\omega}_m^{0} x^{I}_n \right) + \eta^{-1} {M_\theta^I} \ddot{\omega}_i^{0} = \eta \sum_{\mathcal{F}_I} A \left( {\frac {\partial {w}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + \frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} \right) + {V}^I \varepsilon_{ijk} X^{I}_j {b}_k^{0}$$
where $X^I_j$ is the position vector of particle $I$ in global coordinate system; $w^{IJ}_i = \varepsilon_{ijk} X^C_j t^{IJ}_k$ is the moment of facet traction with respect to the point $O$; $X^C_j$ is the position vector of the contact point $C$ between the particles $I$ and $J$ in the global coordinate system, and $m_i^{IJ}=m^0_\beta e_{\beta i}^{IJ}$. Also, $x_j^I$ and $x_j^C$ are the position vectors of the particle $I$ and the contact point $C$ with respect to the mass center of the RVE, respectively.
By summing up the moment equilibrium equations of all particles inside the RVE and dividing by the volume of the RVE, and considering that $X^I_j = X_j + x^I_j$, one obtains (see Appendix \[MacroEquil-Derivation\] for details)
$$\label{macro-2-averaged-init}
\frac{1}{V_0} \sum_I {M}_u^I \varepsilon_{ijk} X_j \ddot {v}_k^{0} + \rho_{im}^{\theta} (\eta^{-1} \ddot{\omega}_m^{0}) = \frac{\eta}{V_0} \sum_I \sum_{\mathcal{F}_I} A \left( {\frac {\partial {w}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + \frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} \right) + \frac{1}{V_0} \sum_I {V}^I \varepsilon_{ijk} X_j {b}_k^{0}$$
where $\rho_{im}^{\theta} =\sum_I \left[ M_\theta^I \delta_{im} + M_u^I \varepsilon_{ijk} \varepsilon_{kmn} x_j^I x_n^I \right]/V_0$ is the inertia tensor of the RVE. In deriving Equation \[macro-2-averaged-init\], the particle density $M^I/V^I$ was assumed to be constant for all particles; and the local system of reference at the center of the RVE was chosen such that $\sum_I M_u^I x^I_i x^I_j =0$ for any $i \neq j$, i.e. as mentioned earlier in this paper, the axes of the system of reference are principal axes of inertia for the system of particles within the RVE.
Before moving forward with the derivation, let’s first consider the second term on the RHS of Equation \[macro-2-averaged-init\]. For a facet in the material domain which is shared between particles $I$ and $J$, by summing the contribution of two particles $I$ and $J$ on the term $(\partial {m}_{i}^{IJ}/\partial \psi^{0}_{\alpha}) \psi^1_{\alpha}$ and by considering the definition of $\psi^1_{\alpha}$ (see Equation \[curv-expansion-zero’\]), one gets
$$\begin{aligned}
\label{higher-curv-1}
\begin{aligned}
\frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} + \frac {\partial {m}_{i}^{JI}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} = & \frac{1}{\bar{r}} \frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \left[ \varphi_i^{1J} + \omega_{i,j}^{1J} y^{IJ}_j - \varphi_i^{1I} + \omega_{i,j}^{0} y^{IJ}_j + \frac{1}{2} \omega_{i,jk}^{0J} y^{IJ}_j y^{IJ}_k \right] {e}^{IJ}_{\alpha i} \\
& + \frac{1}{\bar{r}} \frac {\partial {m}_{i}^{JI}}{\partial \psi^{0}_{\alpha}} \left[ \varphi_i^{1I} + \omega_{i,j}^{1I} y^{JI}_j - \varphi_i^{1J} + \omega_{i,j}^{0} y^{JI}_j + \frac{1}{2} \omega_{i,jk}^{0I} y^{JI}_j y^{JI}_k \right] {e}^{JI}_{\alpha i}
\end{aligned}\end{aligned}$$
Since the moment stress vector applied on a single facet belonging to two particles $I$ and $J$ are the same in magnitude but opposite in direction, one can write $m_{i}^{IJ} = -m_{i}^{JI}$; and consequently, $\partial m_{i}^{IJ}/\psi^{0}_\alpha = - \partial m_{i}^{JI}/\psi^{0}_\alpha$. In addition, the sign of $\psi^{0}_\alpha$ does not change by interchanging $I$ and $J$ in its definition, and that $y_m^{IJ} = - y_m^{JI}$, $e^{IJ}_{\alpha i} = - e^{JI}_{\alpha i}$, Equation \[higher-curv-1\] can be written as
$$\begin{aligned}
\label{higher-curv-2}
\begin{aligned}
\frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} + \frac {\partial {m}_{i}^{JI}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} = & \frac{1}{\bar{r}} \frac{\partial {m}_{i}^{IJ}}{\partial \psi^{0}_\alpha} \bigg[ y^{IJ}_j \left(\omega_{n,j}^{1J} + \omega_{n,jk}^{0} y^{IJ}_k - \omega_{n,j}^{1I} \right) \bigg] e^{IJ}_{\alpha n}
\end{aligned}\end{aligned}$$
If one compares the definition of $\psi^0_{\alpha}$ (see Equation \[curv-expansion-minus1’\]) to the expression in the bracket on the RHS of Equation \[higher-curv-2\], it yields
$$\begin{aligned}
\label{higher-curv-3}
\begin{aligned}
\frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} + \frac {\partial {m}_{i}^{JI}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} = \frac{\partial {m}_{i}^{IJ}}{\partial \psi^{0}_\alpha} \frac{\partial \psi^{0}_\alpha}{\partial x_j} y^{IJ}_j = \frac{\partial {m}_{i}^{IJ}}{\partial x_j} y^{IJ}_j
\end{aligned}\end{aligned}$$
As a result, one can replace the term $(\partial {m}_{i}^{IJ}/\partial \psi^{0}_{\alpha}) \psi^1_{\alpha}$ in Equation \[macro-2-averaged-init\], with the averaged expression derived in the above Equation \[higher-curv-3\]. Similarly to the derivation relevant to the translational equation of motion, one can replace the term $(\partial {w}_{i}^{IJ}/{\partial \epsilon^0_{\alpha}}) \epsilon^1_{\alpha} $ on the RHS of Equation \[macro-2-averaged-init\], by the average value $1/2(\partial {w}_{i}^{IJ}/{\partial x_m}) y_m^{IJ}$ for each facet. Equation \[macro-2-averaged-init\] can be then rewritten as
$$\begin{aligned}
\label{macro-2-3-averaged}
\begin{aligned}
\rho_{im}^{\theta} (\eta^{-1} \ddot{\omega}_m^{0}) = \frac{\eta}{2V_0}\sum_I \sum_{\mathcal{F}_I} {{A} \bigg( \frac{\partial w^{IJ}_i}{\partial x_j} y^{IJ}_j} + \frac{\partial m^{IJ}_i}{\partial x_j} y^{IJ}_j \bigg) + \frac{1}{V_0} \sum_I \bigg( {V}^I \varepsilon_{ijk} X_j {b}_k^{0} - {M}_u^I \varepsilon_{ijk} X_j \ddot {v}_k^{0} \bigg)
\end{aligned}\end{aligned}$$
Using ${X}^C_j = {X}_j + {x}^C_j$ in the definition of $w^{IJ}_i$ along with the identity equations $\partial (m^{IJ}_i n^{IJ}_j )/ \partial x_j = \partial m^{IJ}_i/ \partial x_j n^{IJ}_j$ and $\partial (w^{IJ}_i n^{IJ}_j )/ \partial x_j = \partial w^{IJ}_i/ \partial x_j n^{IJ}_j$, Equation \[macro-2-3-averaged\] can be written as
$$\begin{aligned}
\label{macro-2-4-averaged}
\begin{aligned}
\rho_{im}^{\theta} (\eta^{-1} \ddot{\omega}_m^{0}) = & \frac{\eta}{2V_0}\sum_I \sum_{\mathcal{F}_I} {A} (y^{IJ}_j \varepsilon_{imk} X_m t^{IJ}_k)_{,j} + \frac{\eta}{2V_0}\sum_I \sum_{\mathcal{F}_I} {A} (y^{IJ}_j \varepsilon_{imk} x^C_m t^{IJ}_k + y^{IJ}_j m^{IJ}_i)_{,j} \\
& + ( \varepsilon_{ijk} X_j {b}_k - \rho_u \varepsilon_{ijk} X_j \ddot {v}_k^{0} )
\end{aligned}\end{aligned}$$
The last term on the RHS of Equation \[macro-2-4-averaged\] is written considering the fact that $b_k^{0}$ and $\ddot {v}_k^{0}$ are equal for all particles inside the RVE. Furthermore, the first term on the RHS of Equation \[macro-2-4-averaged\] can be expanded as $(y^{IJ}_j \varepsilon_{imk} X_m t^{IJ}_k)_{,j} = \varepsilon_{ijk} y^{IJ}_j t^{IJ}_k + \varepsilon_{imk} X_m (y^{IJ}_j t^{IJ}_{k})_{,j}$, in which $\partial y^{IJ}_j/ \partial x_j=0$ is used. Therefore, Equation \[macro-2-4-averaged\] becomes
$$\begin{aligned}
\label{macro-2-5-averaged}
\begin{aligned}
\rho_{im}^{\theta} (\eta^{-1} \ddot{\omega}_m^{0}) = & \frac{1}{2V_0}\sum_I \sum_{\mathcal{F}_I} {A} \varepsilon_{ijk} x^{IJ}_j t^{IJ}_k+ \frac{1}{2V_0}\sum_I \sum_{\mathcal{F}_I} {A} (x^{IJ}_j \varepsilon_{imk} x^C_m t^{IJ}_k + x^{IJ}_j m^{IJ}_i)_{,j} \\
& + \varepsilon_{ijk} X_j \bigg( \frac{1}{2V_0}\sum_I \sum_{\mathcal{F}_I} A (x^{IJ}_m t^{IJ}_{k})_{,m} + {b}_k^{0} - \rho_u \ddot {v}_k^{0} \bigg)
\end{aligned}\end{aligned}$$
The last term on the RHS of Equation \[macro-2-5-averaged\] is the moment of the translational equilibrium equation of the RVE (see Equation \[macro-1-2-averaged\]) around the origin of the macroscopic global coordinate system; therefore, it is equal to zero. Comparing the first term on RHS of Equation \[macro-2-5-averaged\] with definition of macroscopic stress tensor of the RVE in Equation \[macro-stress-formula\], one can replace it with $\varepsilon_{ijk} \sigma^0_{jk}$. The second term on the RHS of Equation \[macro-2-5-averaged\] is the divergence of the averaged moment stress tensor of the RVE. The macro-scale rotational equation of motion can be then written as follows
$$\label{macro-rotational-final}
\begin{gathered}
\rho_{\theta ij} (\eta^{-1} \ddot {\omega}_j ^{0}) = {\varepsilon}_{ijk} {\sigma}_{ij}^0 + \frac{\partial \mu^0_{ji}}{\partial x_j}
\end{gathered}$$
where
$$\label{macro-momentstress-formula}
\begin{gathered}
\mu^0_{ij} = \frac{1}{2V_0}\sum_I \sum_{\mathcal{F}_I} {A}r (m_{\alpha}^0 P_{ij}^{\alpha} + t_{\alpha}^0 Q_{ij}^{\alpha})
\end{gathered}$$
and the matrix $Q_{ij}^{\alpha}$ is defined as $Q_{ij}^{\alpha} = n_i^{IJ} \varepsilon_{jkl} x^C_k e_{\alpha l}^{IJ}$. $\mu^0_{ij}$ is the macroscopic moment stress tensor calculated using the results of RVE analysis, and Equation \[macro-rotational-final\] corresponds to the classical rotational equilibrium equation of Cosserat continua [@Chan-1; @Cosserat-1]. According to Equation \[macro-momentstress-formula\] for macroscopic moment stress tensor and considering that ${x}^C_k = {x}^{I}_k + {c}^I_k$, one can conclude that $\mu^0_{ij}$ consists of three terms: (1) the effect of the facet couple traction ${\mathbf}{m}$; (2) the effect of the moment of the facet stress traction ${\mathbf}{t}$ around the particle node which the facet belongs to, and (3) the effect of the moment of the facet stress traction ${\mathbf}{t}$, transferred to the particle node, around the centroid of the RVE. As result, the moment stress tensor is characterized by three length scales: (1) the facet size, associated to ${\mathbf}{m}$; (2) the particle size or facet spacing; and (3) the size of the RVE.
Numerical Results {#NumRes}
=================
The homogenization theory formulated and discussed in the previous sections was implemented in the MARS computational software [@mars-1] with the objective of upscaling the Lattice Discrete Particle Model (LDPM). LDPM, formulated, calibrated, and validated by Cusatis and coworkers [@cusatis-ldpm-1; @cusatis-ldpm-2], is a meso-scale discrete model which simulates the mechanical interaction of concrete coarse aggregate pieces. LDPM has shown superior capabilities in modeling concrete behavior under dynamic loading [@cusatis-Jovanca; @cusatis-alonerate], Alkali Silica Reaction (ASR) deterioration [@cusatis-mohammed], as well as failure and fracture of fiber-reinforced concrete [@cusatis-Ed1; @cusatis-Ed2].
The complete LDPM formulation is summarized in Appendix \[LDPM\]. It is worth mentioning here that the LDPM computational units are polyhedral cells whose construction is anchored to the Delaunay triangulation of the simulated concrete aggregate pieces that are assumed to be spherical and size-graded according to the Fuller size distribution. In the LDPM formulation, each polyhedral cell represents one concrete spherical aggregate piece embedded in the surrounding mortar and the interfaces among the cells represent potential mortar cracks. Figure \[PeriodicLDPM\]a shows a typical LDPM system of polyhedral cells and Figure \[PeriodicLDPM\]b its periodic approximation.
[0.45]{} ![Polyhedral particle distribution in a LDPM prism: (a) generic LDPM system, (b) Periodic LDPM system.[]{data-label="PeriodicLDPM"}](CellPrism.pdf "fig:")
[0.45]{} ![Polyhedral particle distribution in a LDPM prism: (a) generic LDPM system, (b) Periodic LDPM system.[]{data-label="PeriodicLDPM"}](CellPrismPeriodic.pdf "fig:")
The generic RVE shown in Figure \[PeriodicLDPM\]b is constructed as follows. Eight nodes are created at the vertexes of a cube (Figure \[RVEgen\]a). Then nodes are randomly placed on a RVE edge parallel to $x$ axis, see node $a$ in Figure \[RVEgen\]b. Then, these nodes are duplicated on the other three parallel edges along the $x$ axis, see nodes $b, c$, and $d$ in Figure \[RVEgen\]b. Similar procedure is carried out over the edges parallel to $y$ and $z$ axes. Next, the node generation on the RVE surfaces is performed by randomly placing nodes on a cube face with $z$ axis as normal vector, see node $e$ in Figure \[RVEgen\]c. The same nodes are then duplicated on the opposite RVE faces, see node $f$ in Figure \[RVEgen\]c. Nodes on parallel cube faces with $x$ and $y$ axes as normal vectors are constructed with the same algorithm. Finally, nodes are placed inside the RVE based on the general LDPM procedure (see Appendix \[LDPM\] and relevant publications [@cusatis-ldpm-1] for details).
As mentioned earlier in this paper the RVE analysis is conducted by imposing periodic boundary conditions. This is obtained by setting the displacements and rotations of the RVE vertexes to be zero and by imposing, through a master-slave constraint, that the periodic edge nodes and face nodes have the same rotations and displacements.
The overall multiscale numerical procedure adopted in this paper can be summarized as follows.
- The finite element method is employed to solve the macro-scale homogeneous problem in which external loads and essential BCs are applied incrementally. During each numerical step, strain increments $\Delta{\gamma}_{ij} = \Delta{v}^{0}_{j,i} - \varepsilon_{ijk} \Delta{\varphi}_k^{0}$ and curvature increments $\Delta{\kappa}_{ij}=\Delta{\omega}_{j,i}^{0}$ tensors are calculated at each integration point based on the nodal displacement and rotation increments of the corresponding finite element.
- The macroscopic strain and curvature increments are projected into the RVE facets through the proper projection operators: $\Delta{\epsilon}^c_{\alpha} = P^\alpha_{ij} \left(\Delta{\gamma}_{ij} + \varepsilon_{jmn} \Delta{\kappa}_{im} y_{n}^{c} \right)$ and $\Delta{\psi}^c_{\alpha} = P^\alpha_{ij} \Delta{\kappa_{ij}}$. These projected strains and curvatures are imposed, upon sign change, as eigen-strains and eigen-curvatures, $\Delta{\epsilon}^0_{\alpha} =\Delta{\epsilon}^c_{\alpha}+\Delta{\epsilon}^f_{\alpha}=\Delta{\epsilon}^c_{\alpha}- (-\Delta{\epsilon}^f_{\alpha})$ and $\Delta{\psi}^0_{\alpha} =\Delta{\psi}^c_{\alpha}+\Delta{\psi}^f_{\alpha}=\Delta{\psi}^c_{\alpha}- (-\Delta{\psi}^f_{\alpha})$ (See section \[RVEproblem\]), to the RVE allowing the calculation of the fine-scale solution governed by the fine-scale constitutive equations.
- Finally, the fine-scale facet tractions and moments are used to compute, through Equations \[macro-stress-formula\] and \[macro-momentstress-formula\], the macroscopic stresses, $\sigma^0_{ij}$, and couple stresses, $\mu_{ij}^0$, for each Gauss point in the FE mesh.
Elastic RVE Analysis {#Elastic Analysis}
--------------------
This section presents the analysis of the elastic macroscopic behavior of one LDPM RVE. The macroscopic homogenized behavior is analyzed with reference to the classical constitutive equation for Cosserat elasticity, which, in non-dimensional variables, can be written as:
$$\label{Cauchy-Constitutive}
\hat{\sigma}_{ij} = p_0\hat{\gamma}_{kk} \delta_{ij} + p_1\hat{\gamma}_{(ij)} + p_2 \hat{\gamma}_{[ij]}~; ~~~\hat{\mu}_{ij} = q_0 \hat{\kappa}_{kk} \delta_{ij} + q_1 \hat{\kappa}_{(ij)} + q_2 \hat{\kappa}_{[ij]}$$
where $\hat{\sigma}_{ij} = \sigma_{ij} / (2 \mu + \chi)$ and $\hat{\mu}_{ij}= L\mu_{ij} /[(2 \mu+\chi)D^2]$ are the normalized stress and couple tensors, $L$ = characteristic size of the structure of interest, $D$ = size of the RVE; $\hat{\gamma}_{ij}=\gamma_{ij}$, $\hat{\kappa}_{ij} =L \kappa_{ij}$ normalized strains and curvatures; $p_0=\lambda/ (2 \mu + \chi) $, $p_1=1$; $p_2 = \chi / (2 \mu+\chi)$; $q_0=\pi_1/[(2 \mu + \chi)D^2]$, $q_1=(\pi_2+\pi_3)/[(2 \mu + \chi)D^2]$; $q_2=(\pi_2-\pi_3)/[(2 \mu + \chi)D^2]$; $\delta_{ij}$ = kronecker delta; $\mu$, $\lambda$, $\chi$, $\pi_1$, $\pi_2$, and $\pi_3$ are the elastic constants; and the subscript parentheses and brackets represent extraction of the symmetric and antisymmetric, respectively, part of the tensors.
In this section, eight different LDPM RVE sizes $D$= 15, 20, 25, 35, 50, 75, 100, and 150 mm are considered and 5 RVEs, characterized by different placement of the aggregate pieces, is studied for each case. It is worth mentioning that, in LDPM, different spherical aggregate placement inside the RVE yield to different RVE polyhedral particle configurations. The numerical calculations were performed by assuming the concrete mix design and model parameters reported in Appendix \[LDPM\]. Figure \[young-poisson-norm\]a shows the homogenized values of $p_0$, $p_2$, and of the normalized Young’s modulus defined as $e=E/(2\mu +\chi)=(3 \lambda +2\mu +\chi)/(2 \lambda +2\mu +\chi)$, as function of the RVE size normalized by the maximum spherical aggregate size, $d=D/d_a$. The error bars represent the scatter in the results obtained by simulating 5 different RVEs of the same size but with different realization of spherical aggregate positions inside the RVE. As one can see, the calculated values of the parameters tend to converge to a constant value as the size of the RVE increases and, at the same time, the results become independent of the spherical aggregate distribution inside the RVE. The value of $p_2$ is very close to zero for all RVE sizes and decreases rapidly with respect to the RVE size; this suggests that, for the analyzed fine-scale model, the homogenized stress tensor is symmetric. This result is due to the fact that in the LDPM formulation facet moments are zero, and this leads to facet traction distributions around each particle that have zero moment resultant around the particle node. In Figure \[young-poisson-norm\]b the homogenized Poisson’s ratio is reported based on the equation $\nu = \lambda / (2 \lambda +2\mu +\chi)$ and the calculated asymptotic value, 0.18, corresponds well with the value of 0.175 calculated by exploiting the equivalence between particle models and microplane models [@cusatis-ldpm-1]. Finally, Figure \[young-poisson-norm\]c shows the homogenized parameters, $q_0$, $q_1$, and $q_2$, as a function of the RVE size. These quantities also converge to an asymptotic value and become independent of the RVE spherical aggregate distribution for large enough value of $D/d_a$. By virtue of these results and by recalling the definitions of $q_0$, $q_1$, and $q_2$, it is interesting to note that the macroscopic Cosserat elastic parameters of the homogenized continuum depend quadratically on the RVE size.
[0.32]{} ![Variation of elastic normalized effective material properties: (a) $p_0$, $p_2$ and normalized Young modulus $E$. (b) $\nu$ Poisson’s ratio. (c) $q_0$, $q_1$ and $q_2$, with respect to the ratio of RVE size to maximum spherical aggregate size.[]{data-label="young-poisson-norm"}](p-Constants.pdf "fig:"){width="\textwidth"}
[0.32]{} ![Variation of elastic normalized effective material properties: (a) $p_0$, $p_2$ and normalized Young modulus $E$. (b) $\nu$ Poisson’s ratio. (c) $q_0$, $q_1$ and $q_2$, with respect to the ratio of RVE size to maximum spherical aggregate size.[]{data-label="young-poisson-norm"}](Poisson-Norm.pdf "fig:"){width="\textwidth"}
[0.32]{} ![Variation of elastic normalized effective material properties: (a) $p_0$, $p_2$ and normalized Young modulus $E$. (b) $\nu$ Poisson’s ratio. (c) $q_0$, $q_1$ and $q_2$, with respect to the ratio of RVE size to maximum spherical aggregate size.[]{data-label="young-poisson-norm"}](q-constants.pdf "fig:"){width="\textwidth"}
Nonlinear RVE Analysis
----------------------
In this section, the nonlinear response of the RVE is investigated under different strain and curvature loading conditions. Three different RVE sizes, $D=$25, 50, and 100 mm, and 7 different spherical aggregate placement inside the RVE are considered for each case. Typical polyhedral particle systems and geometry of each RVE size are shown in Figure \[RVEgeom\]. The nonlinear homogenized behavior of the RVE is studied under the effect of uniaxial strain tension and compression, hydrostatic compression, bending and torsional curvatures. In the following numerical examples, concrete mix design and model parameters are the same as the ones used in the elastic analysis.
[0.1]{} ![RVE geometry and polyhedral particle distribution: (a) 25 mm (b) 50 mm (c) 100 mm[]{data-label="RVEgeom"}](UC-Cell-25.pdf "fig:"){width="90.00000%"}
[0.3]{} ![RVE geometry and polyhedral particle distribution: (a) 25 mm (b) 50 mm (c) 100 mm[]{data-label="RVEgeom"}](UC-Cell-50.pdf "fig:"){width="55.00000%"}
[0.3]{} ![RVE geometry and polyhedral particle distribution: (a) 25 mm (b) 50 mm (c) 100 mm[]{data-label="RVEgeom"}](UC-Cell-100.pdf "fig:"){width="100.00000%"}
### Nonlinear Analysis of RVE subject to components of the strain tensor {#non-ten}
Figure \[stress-strainNonlinear-tens\] shows the homogenized stress-strain curves for different RVE sizes and polyhedral particle realizations relevant to RVEs subjected to uniaxial tensile strain. The results illustrate that the different polyhedral particle realizations do not affect the linear elastic and nonlinear pre-peak responses, but on the other hand, it clearly influences on the post-peak softening response. One can notice that the post-peak response of smaller RVE sizes is more scattered, while fine-scale randomness effect on the homogenized response diminishes for the larger RVEs [@Gitman-1; @Nguyen-1]. Therefore, one can conclude that the mesh realization is a more influential factor on the post-peak softening response of the RVEs of smaller sizes. Average of peak stress and strain values of different mesh realizations are calculated for each RVE size, and its variation with respect to the RVE size is plotted in Figures \[PeakStress\] and \[PeakStrain\]. As one can see these quantities as well as mesh realization effect decrease as the size of the RVE increases.
[0.3]{} ![Macroscopic stress-strain curve for three different RVE size: 25mm, 50mm, 100mm under uni-axial tension[]{data-label="stress-strainNonlinear-tens"}](25_sig_str.pdf "fig:"){width="\textwidth"}
[0.3]{} ![Macroscopic stress-strain curve for three different RVE size: 25mm, 50mm, 100mm under uni-axial tension[]{data-label="stress-strainNonlinear-tens"}](50_sig_str.pdf "fig:"){width="\textwidth"}
[0.3]{} ![Macroscopic stress-strain curve for three different RVE size: 25mm, 50mm, 100mm under uni-axial tension[]{data-label="stress-strainNonlinear-tens"}](100_sig_str.pdf "fig:"){width="\textwidth"}
[0.3]{} ![Variation of (a) average peak stress and (b) average peak strain, with respect to the RVE size.[]{data-label="PeakStat"}](sig_peak.pdf "fig:"){width="\textwidth"}
[0.3]{} ![Variation of (a) average peak stress and (b) average peak strain, with respect to the RVE size.[]{data-label="PeakStat"}](str_peak.pdf "fig:"){width="\textwidth"}
Furthermore, the average stress-strain curves of different polyhedral particle configurations for each RVE size are calculated and plotted in Figure \[AverageSigmaTens\]. As one can see clearly, increasing size of the RVE affects the post-peak behavior and increases the brittleness of the response. This is consistent with the well-known size effect associated to damage localization in quasi-brittle materials [@Bazant-Book].
[0.3]{} ![(a) Average tensile stress-strain curves for three different RVE sizes. (b) Coarse- and fine-scale strain energy density for different RVE sizes.[]{data-label="TenSigEneAvg"}](Avg_sig_str.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Average tensile stress-strain curves for three different RVE sizes. (b) Coarse- and fine-scale strain energy density for different RVE sizes.[]{data-label="TenSigEneAvg"}](TensileEne.pdf "fig:"){width="\textwidth"}
This phenomenon is depicted in Figure \[rve-damaged-tens\], which shows damaged RVEs of different sizes at the end of the tensile loading process. The contour plots present meso-scale crack opening distributions corresponding to macroscopic imposed uniaxial strain equal to $10^{-3}$. One can easily notice that the damaged area does not scale with the RVE size leading to the post peak size dependency on the RVE size.
Evolution of damage for a 100 mm RVE is also shown in Figure \[Curve100\] at five different macroscopic strain levels. Strain levels (1) and (2) are in pre-peak regime, in which damage is distributed throughout the RVE, which corresponds to the fact that homogenized response is not size dependent in the pre-peak regime. At strain level (3) which corresponds to the peak of the stress-strain curve, damage is still distributed over the RVE; However, as the material undergoes softening, damage localization initiates. Strain levels (4) and (5) are relevant to the softening branch of the response, in which damage localization is clearly visible. The size dependence of the homogenized softening RVE response leads to mesh-dependence of the macroscopic response. This issue has been investigated by some authors [@Gitman-1; @Nguyen-1; @Gitman-2] with reference to continuum-based fine scale models. The complete analysis of this aspect with reference to the current LDPM-based homogenization scheme will be pursed in future work by the writers.
Finally, in Figure \[TensileEne\], the Hill-Mandel condition is verified by comparing the RVE strain energy density calculated through fine-scale and macroscopic quantities.
Next, the nonlinear homogenized behavior of the RVE is studied under confined (uniaxial strain) and hydrostatic compression. For the confined compression test, a strain tensor with a longitudinal component up to -0.03 is considered, whereas for the hydrostatic compression case, all normal components of the strain tensor are set equal and with value up to -0.03. Figure \[stress-strainNonlinear-comp\] shows the nonlinear response of RVEs of different sizes and 7 different polyhedral particle configurations.
[0.3]{} ![Volumetric stress-strain curve for three different RVE sizes under confined compression and hydrostatic compression[]{data-label="stress-strainNonlinear-comp"}](25NonComp.pdf "fig:"){width="\textwidth"}
[0.3]{} ![Volumetric stress-strain curve for three different RVE sizes under confined compression and hydrostatic compression[]{data-label="stress-strainNonlinear-comp"}](50NonComp.pdf "fig:"){width="\textwidth"}
[0.3]{} ![Volumetric stress-strain curve for three different RVE sizes under confined compression and hydrostatic compression[]{data-label="stress-strainNonlinear-comp"}](100NonComp.pdf "fig:"){width="\textwidth"}
In this case, due to the confinement, the stress-strain response is strain-hardening, and as one can see the different polyhedral particle realizations do not affect significantly the homogenized response in both the elastic and inelastic regime. In addition, the average of different mesh realization stress-strain responses is calculated and plotted for each RVE size in Figure \[AverageSigmaComp\]. The nonlinear compressive response does not depend on the RVE size, which is consistent with the fact that plastic deformations are distributed through out the specimen, and strain localization does not take place. Finally, the Hill-Mandel condition is verified with reference to the confined compression test, and the fine- and coarse-scale strain energy density of different RVE sizes are plotted in Figure \[compEne\].
[0.3]{} ![(a) Average compressive volumetric stress-strain curves for three different RVE sizes. (b) Coarse- and fine-scale strain energy density for different RVE sizes.[]{data-label="AverageSigmaComp-compEne"}](AverageCompNon.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Average compressive volumetric stress-strain curves for three different RVE sizes. (b) Coarse- and fine-scale strain energy density for different RVE sizes.[]{data-label="AverageSigmaComp-compEne"}](CompEne.pdf "fig:"){width="\textwidth"}
### Nonlinear Analysis of RVE subject to components of the curvature tensor {#non-cur}
In this section, the nonlinear homogenized behavior of RVEs of 3 different sizes, 50, 75, and 100 mm and 5 five different mesh configurations for each size, is studied under the effect of components of macroscopic curvature tensor. Bending and torsional behavior of the RVEs are investigated by applying macroscopic curvature tensors with the only non-zero components of $\kappa_{12} = 1$ and $\kappa_{11} = 1$, respectively. Figure \[rve-damaged-curv\] shows crack opening contour of damaged RVEs at the macroscopic curvature for $\kappa_{12} = 0.5$. The resulting crack pattern conforms with the fracture mode that one may expect from bending theories. Multiple crack lines are generated in the tensile strain domain, which is the top half of the RVEs, while half bottom part in under compression. As more strain is applied in compressive part, splitting cracks take place in the latter region due to transverse tensile stress. Typical crack pattern of RVEs under torsion are plotted in Figure \[rve-damaged-tor\] for $\kappa_{11} = 0.5$. Crack opening contours show that the amount of damage close to the RVE center is negligible, while it increases as the facets are placed at further positions. This corresponds to the deformation mechanism and strain distribution in solids subject to torsion.
[0.3]{} ![(a) Homogenized couple stresses $\mu_{11}$ and $\mu_{12}$ versus curvature $\kappa_{11}$ and $\kappa_{12}$ for five different polyhedral particle configurations for each RVE size. (b) Average of the homogenized couple stress of different polyhedral particle configurations for each RVE size for $\kappa_{11}$ and $\kappa_{12}$ cases. (c) Scaled couple stress versus curvature curves. (d) Macro and Fine-Scale strain energy density evolution. (e) Scaled strain energy density evolution for the case $\kappa_{12}$. (f) Trace of stress tensor due to elastic and nonlinear analysis of RVE under macroscopic $\kappa_{12}$.[]{data-label="NC"}](Sig_12_all_n.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Homogenized couple stresses $\mu_{11}$ and $\mu_{12}$ versus curvature $\kappa_{11}$ and $\kappa_{12}$ for five different polyhedral particle configurations for each RVE size. (b) Average of the homogenized couple stress of different polyhedral particle configurations for each RVE size for $\kappa_{11}$ and $\kappa_{12}$ cases. (c) Scaled couple stress versus curvature curves. (d) Macro and Fine-Scale strain energy density evolution. (e) Scaled strain energy density evolution for the case $\kappa_{12}$. (f) Trace of stress tensor due to elastic and nonlinear analysis of RVE under macroscopic $\kappa_{12}$.[]{data-label="NC"}](Sig_all_avg_n.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Homogenized couple stresses $\mu_{11}$ and $\mu_{12}$ versus curvature $\kappa_{11}$ and $\kappa_{12}$ for five different polyhedral particle configurations for each RVE size. (b) Average of the homogenized couple stress of different polyhedral particle configurations for each RVE size for $\kappa_{11}$ and $\kappa_{12}$ cases. (c) Scaled couple stress versus curvature curves. (d) Macro and Fine-Scale strain energy density evolution. (e) Scaled strain energy density evolution for the case $\kappa_{12}$. (f) Trace of stress tensor due to elastic and nonlinear analysis of RVE under macroscopic $\kappa_{12}$.[]{data-label="NC"}](Sig_scaled_11-12.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Homogenized couple stresses $\mu_{11}$ and $\mu_{12}$ versus curvature $\kappa_{11}$ and $\kappa_{12}$ for five different polyhedral particle configurations for each RVE size. (b) Average of the homogenized couple stress of different polyhedral particle configurations for each RVE size for $\kappa_{11}$ and $\kappa_{12}$ cases. (c) Scaled couple stress versus curvature curves. (d) Macro and Fine-Scale strain energy density evolution. (e) Scaled strain energy density evolution for the case $\kappa_{12}$. (f) Trace of stress tensor due to elastic and nonlinear analysis of RVE under macroscopic $\kappa_{12}$.[]{data-label="NC"}](Ene_all_n.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Homogenized couple stresses $\mu_{11}$ and $\mu_{12}$ versus curvature $\kappa_{11}$ and $\kappa_{12}$ for five different polyhedral particle configurations for each RVE size. (b) Average of the homogenized couple stress of different polyhedral particle configurations for each RVE size for $\kappa_{11}$ and $\kappa_{12}$ cases. (c) Scaled couple stress versus curvature curves. (d) Macro and Fine-Scale strain energy density evolution. (e) Scaled strain energy density evolution for the case $\kappa_{12}$. (f) Trace of stress tensor due to elastic and nonlinear analysis of RVE under macroscopic $\kappa_{12}$.[]{data-label="NC"}](Ene_12_scaled.pdf "fig:"){width="\textwidth"}
[0.3]{} ![(a) Homogenized couple stresses $\mu_{11}$ and $\mu_{12}$ versus curvature $\kappa_{11}$ and $\kappa_{12}$ for five different polyhedral particle configurations for each RVE size. (b) Average of the homogenized couple stress of different polyhedral particle configurations for each RVE size for $\kappa_{11}$ and $\kappa_{12}$ cases. (c) Scaled couple stress versus curvature curves. (d) Macro and Fine-Scale strain energy density evolution. (e) Scaled strain energy density evolution for the case $\kappa_{12}$. (f) Trace of stress tensor due to elastic and nonlinear analysis of RVE under macroscopic $\kappa_{12}$.[]{data-label="NC"}](Trace_n.pdf "fig:"){width="\textwidth"}
Homogenized moment stress components $\mu_{12}$ and $\mu_{11}$ versus macroscopic curvature tensor components $\kappa_{12}$ and $\kappa_{11}$ of RVEs of different sizes and polyhedral particle configurations are plotted in Figure \[mu\_all\]. One can see that effect of different polyhedral particle realizations on the homogenized response is negligible, which is due to the occurrence of distributed damage inside the RVE. Homogenized response of RVEs with different polyhedral particle realizations are averaged for each size and plotted in Figure \[mu\_all\_avg\]. It can be seen that the homogenized response consists of an initial elastic part and a hardening branch, which is related to the confinement due to the fact that all components of the strain tensor are zero, and the RVE cannot expand laterally. It is illustrated that, at any level of macroscopic curvature, magnitude of the moment stress for larger RVE sizes is bigger compared to the smaller ones. Size dependency of moment stress was discussed in Section \[Elastic Analysis\], and it was shown that the elastic Cosserat coefficients are proportional to the RVE size squared. In order to study size dependency in the nonlinear regime, the normalized quantities $\hat{\mu}_{ij} = {\mu}_{ij}/D$ and $\hat{k}_{ij} = k_{ij} \times D$ are plotted in Figure \[mu\_avg\_scaled\]. One can see that the normalized curves of three different RVE sizes are unique for both bending and torsion. This implies that the proportionality of the homogenized micropolar properties to the RVE size squared is still valid in the nonlinear regime. In Figure \[Ene\_all\], the Hill-Mandel condition is verified, and coarse- and fine-scale strain energy density are plotted for each RVE size for both the aforementioned cases.
Finally, the existence of coupling effect characterized by the dependency of the homogenized stress tensor on the curvature tensor is investigated in both elastic and nonlinear regimes. The macroscopic curvature $\kappa_{12}=1$ is applied on the RVEs, and the trace of the homogenized stress tensor is calculated and plotted in Figure \[trace\] for different RVE sizes. One can observe that the trace of the stress tensor for the case of elastic RVE behavior is zero throughout the analysis. On the other hand, for the case of nonlinear behavior, it increases monotonically with the curvature. This implies that in elastic regime, stresses and strains are totally uncoupled from couple stresses and curvatures, whereas these quantities are strongly coupled in the nonlinear case. This aspect has been investigated very little in the literature where fully uncoupled behavior has been always postulated.
Tension Test on a Concrete Prism with Parallel Elastic Bars {#prismExample}
-----------------------------------------------------------
In this section, the behavior of a reinforced concrete prism under tension is studied in a full fine-scale simulation, and the obtained results are compared to the solution of the same problem through a two-scale homogenization algorithm, in which the concrete prism is modeled as a homogeneous continuum with a meso-scale material RVE assigned to every macroscopic integration point. Figure \[fullldpm-Cbar\] shows the concrete prism and the two elastic bars attached to it, which are simulated by LDPM and solid finite elements, respectively. The same specimen in a two-scale homogenization problem is depicted in Figure \[homogenization-Cbar\], in which concrete prism is modeled by tetrahedral finite elements. Cross section of the concrete prism is 100 mm $\times$ 100 mm, and its height is 500 mm. Two rigid loading plates are attached at the top and the bottom of the whole specimen cross section to apply the boundary condition. Young modulus and Poisson’s ratio of the elastic bars are 28 GPa and 0.18, respectively. The same LDPM parameters used in the previous sections are adopted here. The specimen is pulled in the longitudinal direction up to a displacement equal to 0.7 mm. The RVE size is chosen to be 30 mm which approximately corresponds to the volume of each tetrahedral FE in the coarse mesh. This is done to mitigate the mesh-dependence due to the softening behavior of the RVE. The concrete prism and the elastic bars are connected through a master-slave algorithm. The numerical simulations of the coarse scale are performed by neglecting the couple stresses which are expected to be negligible for this particular application.
[0.2]{} 
[0.3]{} 
[0.4]{} 
The global force-displacement response of the full fine-scale and homogenization problems are plotted in Figure \[stress-strain-curve\]. Since concrete prism and elastic bars are tied and deform together during the loading process, distributed damage takes place through the whole specimen during the initial stages of the loading process, see Figure \[prism\]a. This damage state represents the linear elastic and the first hardening segment of the stress-strain response of the structure. The same damage state is captured through the homogenization procedure. Figure \[prism\]e shows finite elements normal strain distribution along the loading direction through the specimen. One can see that the strain values are all in the same range, and no localization has occurred. The response of the full fine-scale and homogenization problems show excellent agreement in the elastic and the first hardening segment. As further deformation is applied on the structure, damage localizes in one section of the concrete bar and this causes a sudden drop in the global force-displacement curve. Subsequently, since the elastic bars and the concrete prism are forced to deform in parallel, the overall system can carry more load leading to a rehardening of the global response. Analysis of Figure \[stress-strain-curve\] shows that five damage localization events occur during the deformation process which corresponds to five sudden drops in the load-displacement curve. Crack pattern of the specimen is plotted after the formation of two, four, and five damage localization in Figure \[prism\]b, c, and d. It is interesting to show that the homogenization framework is able to generate the same damage distribution pattern. Figures \[prism\]f, g, and h show that two, four and five strain localization band appear in the specimen, which corresponds to the damage configuration obtained from full fine-scale problem. The global load-displacement curve of the homogenization problem also shows five sudden drops which conforms to the full fine-scale response, see Figure \[stress-strain-curve\]. The homogenized response captures well the displacement at which the first three localization events occur, while it underestimates its value for later events. This is likely due to the relatively coarse mesh adopted at the macroscopic scale.
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](cntr1.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](cntr2.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](cntr3.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](cntr4.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](Homog1.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](Homog2.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](Homog3.pdf "fig:")
[0.2]{} ![(Top row) Crack opening contour at different loading from full fine-scale simulation (Bottom row) Strain distribution contour at different loading steps from homogenization algorithm[]{data-label="prism"}](Homog4.pdf "fig:")
Conclusions
===========
This paper presents the asymptotic expansion homogenization of fine-scale periodic discrete systems featuring independent translational and rotational degrees of freedom. Employing consistent asymptotic expansion of displacement and rotation fields, a rigorous analytical derivation was performed for elastic behavior, and it was extended to the nonlinear case upon making reasonable assumptions on the rigid body motions of a RVE. Based on this work, the following general conclusions can be drawn.
- The equivalent homogenized continuum is of Cosserat-type characterized by nonsymmetric stress and couple tensors energetically conjugate to nonsymmetric strain and curvature tensors, respectively. The classical linear and rotational momentum balance equations can be derived from the homogenization of the fine-scale equilibrium equations.
- The fine-scale kinematic quantities, namely facet strains and curvatures, are demonstrated to be related to the projection of the coarse-scale strains and curvatures into the local facet system of reference. This allows a straightforward implementation of the RVE problem into any computational framework.
- Similarly to previous research, the derived formula linking the fine-scale response to the coarse-scale stress tensor corresponds to the virial stress formulation commonly used for atomistic systems.
- The derived formula linking the fine-scale response to the coarse-scale couple tensor is shown to consist of three terms with clear physical meaning. The first term is associated with the fine-scale couple tractions and it can be related to the facet size, which, in turn can be associated with the size of weak spots in the material internal structure. The second term arises from the moment of the fine-scale stress tractions with respect to the particle node. As such, it depends on the size of the fine-scale particles and it can be related to the spacing or characteristic distance of the weak spots in the material internal structure. Finally, the third term is the effect of the moment of fine-scale stress tractions with respect to the center of the RVE and, consequently, it depends on the RVE size.
The developed framework was then implemented in a computational software and applied to the upscaling of LDPM. Specific to this fine-scale model, the numerical results demonstrate the following interesting features of the equivalent homogenized continuum.
- The macro-scale elastic parameters relating the stress tensor to the strain tensor become independent on RVE size and on the random position of the polyhedral particles inside the RVE for RVE sizes larger than about 5 times the maximum spherical aggregate size. On the contrary, the macro-scale parameters relevant to the relationship between curvature and couple tensors are shown to depend on the RVE size squared and they become independent on the random position of the polyhedral particles inside the RVE for RVE sizes larger than about 5 times the maximum spherical aggregate size.
- The non-symmetric part of the macro-scale stress tensor is negligible since the relevant parameter is at least one order of magnitude smaller than the one governing the symmetric part. As a consequence, the linear and rotational momentum balance equations are decoupled.
- In the elastic regime the stress-strain and couple-curvature constitutive equations are completely uncoupled.
- In the non linear regime, for tensile loading and because the fine-scale behavior is strain-softening, the response is RVE-size-dependent. This is an expected result, although very often not acknowledged by most authors in the literature, associated with strain localization induced by softening. On the contrary, such dependence is not observed for compressive dominated loading conditions because the LDPM fine-scale behavior in compression is strain-hardening.
- The coarse-scale couple-curvature constitutive equations scale with the square of the RVE size in the nonlinear range also but, contrarily to the elastic case, they show a strong coupling with the stress-strain constitutive equations. Such coupling, never considered in the current literature of Cosserat media, will be studied in future work by the authors.
ACKNOWLEDGEMENTS
================
This material is based upon work supported by the National Science Foundation under grant no. CMMI-1201087.
[99]{}
P. A. Cundall. A computer model for simulating progressive large-scale movements in block rock mechanics. Proc. Symp. Int. Soc. Rock Mech. 1971; Nancy, p. 2.
A. A. Serrano, J. M. Rodriguez-Ortiz. A contribution to the mechanics of heterogeneous granular media. Proc. Symp. on the Role of Plasticity in Soil Mechanics 1973; Cambridge, England.
P.A. Cundall, O.D.L Strack. A discrete element model for granular assemblies. Geotechnique 1979; 29(1): 47-65.
T. Kawai. New discrete models and their application to seismic response analysis of structures. Nucl. Eng. Des. 1978; 48: 207-229.
A. Zubelewicz, Z. Mroz. Numerical simulation of rockburst processes treated as problems of dynamic instability. Rock Mech. Rock Eng. 1983; 16: 253-274.
M. E. Plesha, E. C. Aifantis. On the modeling of rocks with microstructure. Proc., 24th U.S. Symp. Rock Mech. 1983; Texas A & M Univ., College Station, Tex., 27-39.
A. Zubelewicz, Z. P. [[Bažant]{}]{}. Interface element modeling of fracture in aggregate composites. J. Eng. Mech. 1987; 113(11): 1619-1630.
J. E. Bolander, S. Saito, Fracture analysis using spring network with random geometry. Eng. Fract. Mech. 1998; 61(5-6): 569-591.
J. E. Bolander, K. Yoshitake, J. Thomure. Stress analysis using elastically uniform rigid-body-spring networks. J. Struct. Mech. Earthquake Eng. JSCE 1999; 633(I-49): 25-32.
J. E. Bolander, G. S. Hong, K. Yoshitake. Structural concrete analysis using rigid-body-spring networks. J. Comput.-Aided Civil Infrastruct. Eng. 2000; 15: 120-133.
A. Hrennikoff. Solution of problems of elasticity by the framework method. J Appl Mech 1941; 12: 169-75.
E. Schlangen, J.G.M. van Mier. Experimental and numerical analysis of micromechanisms of fracture of cement-based composites. Cement Concrete Composite 1992; 14:105-118.
G. Cusatis, Z. P. [[Bažant]{}]{}, L. Cedolin. Confinement-shear lattice model for concrete damage in tension and compression. I. theory. J Eng Mech – ASCE 2003; 129(12): 1439-1448.
G. Cusatis, Z. P. [[Bažant]{}]{}, L. Cedolin. Confinement-shear lattice model for concrete damage in tension and compression. II. computation and validation. J Eng Mech – ASCE 2003; 129(12): 1449-1458.
G. Lilliu, J.G.M. van Mier. 3D lattice type fracture model for concrete. Engineering Fracture Mechanics 2003; 70: 927-941.
S. Berton, J. E. Bolander. Crack band model of fracture in irregular lattices. Comput. Methods Appl. Mech. Engrg. 2006; 195: 7172-7181.
G. Cusatis, D. Pelessone, A. Mencarelli. Lattice Discrete Particle Model (LDPM) for failure behavior of concrete. I: Theory. Cement and Concrete Composites 2011; 33: 881-890.
G. Cusatis, A. Mencarelli, D. Pelessone, J. Baylot. Lattice Discrete Particle Model (LDPM) for failure behavior of concrete. II: Calibration and validation. Cement and Concrete Composites 2011; 33: 891-905.
J.P.B. Leite, V. Slowik, H. Mihashi. Computer simulation of fracture processes of concrete using mesolevel models of lattice structures. Cement and Concrete Research 2004; 34: 1025-1033.
H. Kim, M. P. Wagoner, W. G. Buttlar. Simulation of Fracture Behavior in Asphalt Concrete Using a Heterogeneous Cohesive Zone Discrete Element Model. J. Mater. Civ. Eng. 2008; 20(8): 552-563.
G. Cusatis, H. Nakamura. Discrete modeling of concrete materials. Preface to the Special Issue on discrete models, Cement and Concrete Composites. 2011; 33(9): 865-992.
U. Galvanetto, M. H. Ferri Aliabadi. Multiscale modeling in solid mechanics: computational approaches. 2009; Vol. 3. Imperial College Press.
I.M. Gitman, H. Askes, L.J. Sluys. Representative volume: Existence and size determination. Engineering Fracture Mechanics 2007; 74: 2518-–2534.
V. Kouznetsova, M.G.D. Geers, W.A.M. Brekelmans. Size of a Representative Volume Element in a Second-Order Computational Homogenization Framework. International Journal for Multiscale Computational Engineering 2004; 2(4): 575-598.
R. Hill. Elastic properties of reinforced solids: some theoreticel principles. J Mech Phys Solids 1963; 11: 357–-372.
J.D. Eshelby. The determination of the elastic field of an ellipsoidal inclusion and related problems. Proc. R. Soc. London 1958; 241A: 376-396.
Z. Hashin, S. Strikman. A variational approach to the theory of the elastic behavior of multiphase materials.J. Mech. Phys. Solids 1963; 11: 127-140.
R.M. Christensen, K.H. Lo. Solutions for effective shear properties in three phase sphere and cylinder models.J. Mech. Phys. Solids 1979; 27: 315-330.
S. Nemat-Nasser, N. Yu, M. Hori. Bounds and estimates of overall moduli of composites with periodic microstructure. Mech. Mater. 1993; 15: 163–181.
O. Lopez-Pamies, T. Goudarzi, T. Nakamura. The nonlinear elastic response of suspensions of rigid inclusions in rubber: I - An exact result for dilute suspensions. J. Mech. Phys. Solids 2013; 61(1): 1–18.
O. Lopez-Pamies, T. Goudarzi, K. Danas. The nonlinear elastic response of suspensions of rigid inclusions in rubber: II - A simple explicit approximation for finite-concentration suspensions. J. Mech. Phys. Solids 2013; 61(1): 19–37.
R.J.M. Smit, W.A.M. Brekelmans, H.E.H. Meijer. Prediction of the mechanical behavior of nonlinear heterogeneous systems by multi-level finite element modeling. Computer Methods in Applied Mechanics and Engineering 1998; 155: 181–192.
F. Feyel. A multilevel finite element method (fe2) to describe the response of highly non-linear structures using generalized continua. Comput. Methods Appl. Mech. Engrg 2003; 192: 3233-–3244.
C. Miehe, J. Schröder, J. Schotte. Computational homogenization analysis in finite plasticity Simulation of texture development in polycrystalline materials. Comput. Methods Appl. Mech. Engrg. 1999; 171: 387–418.
J. Fish, W. Chen, R. Li. Generalized mathematical homogenization of atomistic media at finite temperatures in three dimensions. Comput. Methods Appl. Mech. Engrg. 2007; 196: 908–922.
B. Hassani, E. Hinton. A review of homogenization and topology optimization I: homogenization theory for media with periodic structure. Computers and Structures 1998; 69: 707–717.
B. Hassani, E. Hinton. A review of homogenization and topology opimization II: analytical and numerical solution of homogenization equations. Computers and Structures 1998; 69: 719–738.
P.W. Chung, K.K. Tammaa, R.R. Namburub. Asymptotic expansion homogenization for heterogeneous media: computational issues and applications. Composites Part A 2001; 32: 1291–-1301.
J. Fish, K. Shek, M. Pandheeradi, M.S. Shephard .Computational plasticity for composite structures based on mathematical homogenization: Theory and practice. Comput. Methods Appl. Mech. Engrg. 1997; 15: 53-73.
S. Ghosh, K. Lee, S. Moorthy. Multiple scale analysis of heterogeneous elastic structures using homogenization theory and Voronoi cell finite element method. Int. J. Solids and Structures 1995; 32: 27–62.
S. Ghosh, K. Lee, S. Moorthy. Two scale analysis of heterogeneous elastic-plastic materials with asymptotic homogenization and Voronoi cell finite element model. Comput. Methods Appl. Mech. Engrg. 1996; 132: 63–116.
S. Forest, F. Pradel, K. Sab. Asymptotic analysis of heterogeneous Cosserat media. Int. J. Solids and Structures 2001; 38: 4585-4608.
Y. S. Chan, G. H. Paulino, A. C. Fannjiang. Change of constitutive relations due to interaction between strain-gradient effect and material gradation. Journal of Applied Mechanics. 2006; 73(5): 871–875.
S. Bardenhagen, N. Triantafyllidis. Derivation of higher order gradient continuum theories in 2, 3-D non-linear elasticity from periodic lattice models. J. Mech. Phys. Solids. 1994; 42(1): 111–139.
D. Caillerie, A. Mourad, A. Raoult. Discrete homogenization in graphene sheet modeling. Journal of Elasticity. 2006; 84(1): 33–68.
A. Braides, A. J. Lew, M. Ortiz. Effective cohesive behavior of layers of interatomic planes. Archive for Rational Mechanics and Analysis. 2006; 180(2): 151–182.
Z.P. [[Bažant]{}]{}, F.C. Caner, I. Carol, M.D. Adley, S.A. Akers. Microplane model M4 for concrete I: formulation with work-conjugate deviatoric stress. Journal of Engineering Mechanics 2000; 126(9): 944–953.
G. Cusatis, X. Zhou. High-Order Microplane Theory for Quasi-Brittle Materials with Multiple Characteristic Lengths. Journal of Engineering Mechanics 2013.
E. Cosserat, F. Cosserat. Théorie des Corps déformables. Paris: A, Hermann et Fils. 1909.
MARS: Modeling and Analysis of the Response of Structures - User’s Manual, D. Pelessone, ES3, Solana Beach (CA), USA. http://www.es3inc.com/mechanics/ MARS/Online/MarsManual.htm, 2009.
J. Smith, G. Cusatis, D. Pelessone, E. Landis, J. O’Daniel, J. Baylot. Discrete modeling of ultra-high-performance concrete with application to projectile penetration. International Journal of Impact Engineering 2014; 65: 13–32.
G. Cusatis. Strain-rate effects on concrete behavior. International Journal of Impact Engineering 2011; 38: 162–170.
M. Alnaggar, G. Cusatis, G. Di Luzio. Lattice Discrete Particle Modeling (LDPM) of Alkali Silica Reaction (ASR) deterioration of concrete structures. Cement & Concrete Composites 2013; 41: 45-–59.
E. A. Schauffert, G. Cusatis. Lattice discrete particle model for fiber-reinforced concrete. I: Theory. Journal of Engineering Mechanics 2011; 138.7: 826–833.
E. A. Schauffert, G. Cusatis, D. Pelessone, J. L. O’Daniel, J. T. Baylot. Lattice discrete particle model for fiber-reinforced concrete. II: Tensile fracture and multiaxial loading behavior. Journal of Engineering Mechanics 2011; 138.7: 834–841.
P. Stroeven. A stereological approach to roughness of fracture surfaces and tortuosity of transport paths in concrete. Cem Conc Compos 2000; 22(5): 331–41.
V. P. Nguyen, O. L. Valls, M. Stroeven, L. J. Sluys. On the existence of representative volumes for softening quasi-brittle materials – A failure zone averaging scheme. Comput. Methods Appl. Mech. Engrg. 2010; 199: 3028–3038.
Z.P. [[Bažant]{}]{}, J. Planas. Fracture and size effect in concrete and other quasibrittle materials. Boca Raton, London: CRC Press; 1998.
I.M. Gitman a, H. Askes, L.J. Sluys. Coupled-volume multi-scale modelling of quasi-brittle material. European Journal of Mechanics A/Solids. 2008; 27: 302-–327
Short Review of the Lattice Discrete Particle Model (LDPM) Geometrical Construction and Constitutive Equations {#LDPM}
==============================================================================================================
LDPM model generation procedure and governing constitutive equations are explained in the following two sections.
LDPM model construction {#LDPM-Construction}
-----------------------
Concrete meso-scale structure is modeled by LDPM through the following steps:
- Spherical aggregate generation is the first step which is carried out assuming that each aggregate piece can be approximated as a sphere. Under this assumption, the following spherical aggregate size distribution function proposed by Stroeven [@Stroeven-1] is considered
$$\label{psd}
f(d) = \frac{qd_0^q}{[1-(d_0/d_a)^q]d^{q+1}}$$
in which $d_a$ and $d_0$ are the maximum and minimum spherical aggregate size, respectively, and $q$ is a material parameter. It can be shown [@Stroeven-1] that Equation \[psd\] is associated with a sieve curve (percentage of spherical aggregate by weight retained by a sieve of characteristic size $d$) in the form
$$\label{sieve}
f(d) = \bigg(\frac{d}{d_a}\bigg)^{n_f}$$
where $n_f = 3-q$. For $n_f=0.5$ Equation \[sieve\] corresponds to the classical Fuller curve which for its optimal packing properties, is extensively used in concrete technology. Considering concrete cement content $c$, water-to-cement ratio $w/c$, specimen volume, maximum $d_a$ and minimum $d_0$ spherical aggregate size along with the considered distribution function Equation \[sieve\], the spherical aggregate system can be generated using a random number generator.
- By using a try-and-error random procedure, spherical aggregate pieces are introduced into the concrete volume from the largest to the smallest size. Figure \[DogbonePRTC\] shows the spherical aggregate system generated for a typical dogbone specimen.
- Delaunay tetrahedralization of the spherical aggregate piece centers is employed to define the interactions of the spherical aggregate system (Figure \[PolyCellsGeom\]).
- Finally, a three-dimensional domain tessellation anchored to the Delaunay tetrahedralization is carried out to create a system of polyhedral particles interacting through triangular facets, and a lattice system composed of the line segments connecting the spherical aggregate centers. Figure \[DogboneCells\] shows the final polyhedral particle discretization of a typical dogbone specimen.
[0.3]{} ![(a) Spherical aggregate system for a typical dogbone specimen. (b) LDPM polyhedral particles for two adjacent spherical aggregate particle. (c) LDPM cell distretization for a typical dogbone specimen.[]{data-label="LDPMFigures"}](DogbonePRTC.pdf "fig:")
[0.3]{} ![(a) Spherical aggregate system for a typical dogbone specimen. (b) LDPM polyhedral particles for two adjacent spherical aggregate particle. (c) LDPM cell distretization for a typical dogbone specimen.[]{data-label="LDPMFigures"}](PolyCellsParticle.pdf "fig:")
[0.3]{} ![(a) Spherical aggregate system for a typical dogbone specimen. (b) LDPM polyhedral particles for two adjacent spherical aggregate particle. (c) LDPM cell distretization for a typical dogbone specimen.[]{data-label="LDPMFigures"}](Dogbone-Cell.pdf "fig:")
LDPM Kinematics {#LDPM-Kinematics}
---------------
The triangular facets forming the rigid polyhedral particles are assumed to be the potential material failure locations. Each facet is shared between two polyhedral particle and is characterized by a unit normal vector $\bf{n}$ and two tangential vectors $\bf{m}$ and $\bf{l}$. Accordingly, three strain components are defined on each triangular facet using Equations \[eps\] and \[curvature\], which for LDPM gives
$$\label{LDPMstr}
\epsilon_{N} = \frac{\mathbf{n}^T \llbracket {\mathbf{u}_{C}} \rrbracket}{r}; \hspace{0.25 in} \epsilon_{M} = \frac{\mathbf{m}^T \llbracket {\mathbf{u}_{C}} \rrbracket}{r}; \hspace{0.25 in} \epsilon_{L} = \frac{\mathbf{l}^T \llbracket {\mathbf{u}_{C}} \rrbracket}{r}$$
where $\llbracket {\mathbf{u}_{C}} \rrbracket$ is the displacement jump vector calculated at the facet centroid. One should consider that the LDPM constitutive equations explained in the next section are independent of facet curvatures.
LDPM constitutive equations {#LDPM-Constitutive}
---------------------------
This section reviews the specific constitutive equations governing the response of LDPM. First of all, it must be mentioned that LDPM assumes zero couple stresses at the meso-scale in both elastic and inelastic regime. This implies $m_\alpha=0$ for $\alpha=N,M,L$
In the elastic regime, the normal and shear stresses are proportional to the corresponding strains: $t_{N}= E_N \epsilon_{N};~ t_{M}= E_T \epsilon_{M};~ t_{L}= E_T \epsilon_{L}$, where $E_N=E_0$, $E_T=\alpha E_0$, $E_0=$ effective normal modulus, and $\alpha=$ shear-normal coupling parameter. Beyond the elastic regime, the vectorial constitutive relations are meant to reproduce three distinct sources of nonlinearity as described below.
### Fracture and cohesion due to tension and tension-shear {#LDPM-tens}
For tensile loading ($\epsilon_N>0$), fracturing and cohesive behavior due to tension and tension-shear are formulated through an effective strain, $\epsilon = \sqrt{\epsilon _{N}^{2}+\alpha (\epsilon _{M}^{2} + \epsilon _{L}^{2})}$, and stress, $t = \sqrt{{ t _{N}^2+ (t_{M}+t_{L})^2 / \alpha}}$, which define the normal and shear stresses as ; ; . The effective stress $t$ is incrementally elastic ($\dot{t}=E_0\dot{\epsilon}$) and must satisfy the inequality $0\leq t \leq \sigma _{bt} (\epsilon, \omega) $ where $\sigma_{bt} = \sigma_0(\omega) \exp \left[-H_0(\omega) \langle \epsilon-\epsilon_0(\omega) \rangle / \sigma_0(\omega)\right]$, $\langle x \rangle=\max \{x,0\}$, and $\tan(\omega) =\epsilon _N / \sqrt{\alpha} \epsilon _{T}$ = $t_N \sqrt{\alpha} / t_{T}$. The post peak softening modulus is defined as $H_{0}(\omega)=H_{t}(2\omega/\pi)^{n_{t}}$, where $n_t$ is the softening exponent, $H_{t}$ is the softening modulus in pure tension ($\omega=\pi/2$) expressed as $H_{t}=2E_0/\left(l_t/l_e-1\right)$; $l_t=2E_0G_t/\sigma_t^2$; $l_e$ is the length of the tetrahedron edge; and $G_t$ is the mesoscale fracture energy. LDPM provides a smooth transition between pure tension and pure shear ($\omega=0$) with parabolic variation for strength given by $\sigma_{0}(\omega )=\sigma _{t}r_{st}^2\Big(-\sin(\omega)+ \sqrt{\sin^2(\omega)+4 \alpha \cos^2(\omega) / r_{st}^2}\Big)/ [2 \alpha \cos^2(\omega)]$, where $r_{st} = \sigma_s/\sigma_t$ is the ratio of shear strength to tensile strength.
### Compaction and pore collapse from compression {#LDPM-comp}
For compressive loading ($\epsilon_N<0$), the normal stress evolves incrementally elastically and is subjected to the inequality $-\sigma_{bc}(\epsilon_D, \epsilon_V)\leq t_N \leq 0$ where $\sigma_{bc}$ is a strain-dependent boundary function of the volumetric strain, $\epsilon_V$, and the deviatoric strain, $\epsilon_D$. The function expressing $\sigma_{bc}$ models pore collapse for $-\epsilon_V \leq\epsilon_{c1} = \kappa_{c0} \epsilon_{c0}=\kappa_{c0} \sigma_{c0}/E_0$, and it is formulated as $\sigma_{bc} = \sigma_{c0} + \langle-\epsilon_V-\epsilon_{c0}\rangle H_c(r_{DV})$ where $H_c(r_{DV})=H_{c0}/(1 + \kappa_{c2} \left\langle r_{DV} - \kappa_{c1} \right\rangle)$, $r_{DV}=\epsilon_D/\epsilon_V$, $\sigma_{c0}$ is the mesoscale compressive yield stress; and $\kappa_{c0}$, $\kappa_{c1}$, $\kappa_{c2}$ and $H_{c0}$ are material parameters. Compaction and rehardening occur beyond pore collapse for $-\epsilon_V \geq \epsilon_{c1}$. In this case one has $\sigma_{bc} = \sigma_{c1}(r_{DV})$ $\exp \left[( -\epsilon_{V}-\epsilon_{c1} ) H_c(r_{DV})/\sigma_{c1}(r_{DV}) \right]$ and $\sigma_{c1}(r_{DV}) = \sigma_{c0} + (\epsilon_{c1}-\epsilon_{c0}) H_c(r_{DV})$.
### Friction due to compression-shear {#LDPM-shear}
The evolution of shear stresses simulate frictional behavior due to compression-shear. The incremental shear stresses are computed as $\dot{t}_M=E_T(\dot{\epsilon}_M-\dot{\epsilon}_M^p)$ and , where , , and $\lambda$ is the plastic multiplier with loading-unloading conditions $\varphi \dot{\lambda} \leq 0$ and $\dot{\lambda} \geq 0$. The plastic potential is defined as , where the nonlinear frictional law for the shear strength is assumed to be $\sigma_{bs} = \sigma_s + (\mu_0 - \mu_\infty)\sigma_{N0}[1 - \exp(t_N / \sigma_{N0})] - \mu_\infty t_N$; $\sigma_{N0}$ is the transitional normal stress; $\mu_0$ and $\mu_\infty$ are the initial and final internal friction coefficients. Detailed description of model behavior in the nonlinear range can be found in Ref. [@cusatis-ldpm-1].
Concrete Mix-Design and Model Parameters Used in the Numerical Simulations
--------------------------------------------------------------------------
Minimum and maximum spherical aggregate size are $d_0=$ 4 mm and $d_a=$ 8 mm, respectively; cement content c = 612 $\text{kg/m}^\text{3}$; water to cement ratio w/c = 0.4; aggregate to cement ratio a/c = 2.4; Fuller curve coefficient $n_f$ = 0.42.
The following LDPM parameters are used: $E_N = 60$ GPa, $\sigma_t = 3.45$ MPa, $\sigma_{c0} = 150$ MPa, $\alpha = 0.25$, $n_t=0.4$, $l_t=500$ mm, $r_{st}=2.6$, $H_{c0}/E_0 = 0.4$, $\mu_0 = 0.4$, $\mu_\infty = 0$, $k_{c1} = 1$, $k_{c2}=5$, $\sigma_{N0} = 600$ MPa, $\alpha=E_T/E_N=0.25$.
Asymptotic Expansion of Strains and Curvatures {#exp-strains-details}
==============================================
In order to obtain multiple scale definition of facet strain vector, one should first plug macroscopic Taylor series expansion of displacement and rotation of particle $J$ around particle $I$, Eqs. \[taylor-1-J\] and \[taylor-2-J\], into facet strain definition, Equation \[eps\]. In addition, equation ${\mathbf}{x} = \eta{\mathbf}{y}$ is used to change the length type variables to fine-scale quantities; $\eta y^{IJ}_j = x^{IJ}_j$, $\eta \bar{c}^{I}_k = c^{I}_k$ and $\eta \bar{c}^{J}_k = c^{J}_k$. Equation \[eps\] writes
$$\begin{aligned}
\label{eps-expansion-1}
\begin{aligned}
\epsilon_{\alpha}=\eta^{-1} \bar{r}^{-1} \bigg[& u_i^J+ \eta u^J_{i,j} y^{IJ}_j + \eta^2 \frac{1}{2}u^J_{i,jk} y^{IJ}_j y^{IJ}_k - u_i^I \\
& +\eta \varepsilon_{ijk} \bigg( \theta_j^{J}+\eta \theta_{j,m}^J y^{IJ}_m +\frac{1}{2} \eta^2 \theta^J_{j,mn} y^{IJ}_m y^{IJ}_n \bigg) \bar{c}_{k}^{J} -\eta \varepsilon_{ijk} \theta_j^I \bar{c}_{k}^{I} \bigg] e^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Spatial derivatives of displacement and rotation in equation above are partial derivative with respect to $x$. So, first and second order partial derivative of displacement and rotation asymptotic expansions, Eqs. \[disp-expansion\] and \[rot-expansion\], with respect to $x$ are as follows
$$u_{i,j} \approx u^0_{i,j} +\eta u^1_{i,j} \hspace{0.25in} u_{i,jk} \approx u^0_{i,jk} +\eta u^1_{i,jk}
\label{disp-asymp-der}$$
$$\theta_{i,j} \approx \eta^{-1} \omega^0_{i,j} + \varphi^0_{i,j} + \omega^1_{i,j} + \eta \varphi^1_{i,j} \hspace{0.25in} \theta_{i,jk} \approx \eta^{-1} \omega^0_{i,jk} + \varphi^0_{i,jk} + \omega^1_{i,jk} + \eta \varphi^1_{i,jk}
\label{rot-asymp-der}$$
Using asymptotic expansion of displacement and rotation of a particle, Eqs. \[disp-expansion\] and \[rot-expansion\], along with their macro-scale derivatives, Eqs. \[disp-asymp-der\], \[rot-asymp-der\], and replacing them into Equation \[eps-expansion-1\], one obtains
$$\begin{aligned}
\label{eps-expansion-2}
\begin{aligned}
\epsilon_{\alpha} = \eta^{-1} \bar{r}^{-1} \bigg[& u_i^{0J} + \eta u_i^{1J} + \eta u^{0J}_{i,j} y^{IJ}_j + \eta^2 u^{1J}_{i,j} y^{IJ}_j + \eta^2 \frac{1}{2}u^{0J}_{i,jk} y^{IJ}_j y^{IJ}_k + \eta^3 \frac{1}{2}u^{1J}_{i,jk} y^{IJ}_j y^{IJ}_k - u^{0I}_i - \eta u^{1I}_i \\
& + \eta \varepsilon_{ijk} \bigg( \eta^{-1} \omega_j^{0J} + \varphi_j^{0J} + \omega_j^{1J} + \eta \varphi_j^{1J} + \omega_{j,m}^{0J} y^{IJ}_m + \eta \varphi_{j,m}^{0J} y^{IJ}_m + \eta \omega_{j,m}^{1J} y^{IJ}_m + \eta^2 \varphi_{j,m}^{1J} y^{IJ}_m \\
& ~~~~~~~~~~~ + \eta \frac{1}{2} \omega^{0J}_{j,mn} y^{IJ}_m y^{IJ}_n + \eta^2 \frac{1}{2} \varphi^{0J}_{j,mn} y^{IJ}_m y^{IJ}_n + \eta^2 \frac{1}{2} \omega^{1J}_{j,mn} y^{IJ}_m y^{IJ}_n + \eta^3 \frac{1}{2} \varphi^{1J}_{j,mn} y^{IJ}_m y^{IJ}_n\bigg) \bar c_{k}^{J} \\
& - \eta \varepsilon_{ijk} \bigg( \eta^{-1} \omega_j^{0I} + \varphi_j^{0I} + \omega_j^{1I} + \eta \varphi_j^{1I} \bigg) \bar c_{k}^{I} \bigg) \bigg] e^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Regrouping terms of the same order in above equation, one would get multiple scale definition of facet strain $$\begin{aligned}
\label{eps-expansion-3}
\begin{aligned}
\epsilon_{\alpha} = \bar{r}^{-1} \bigg[ & \eta^{-1} \bigg( u_i^{0J} - u^{0I}_i + \varepsilon_{ijk} \omega_j^{0J} \bar c_{k}^{J} - \varepsilon_{ijk} \omega_j^{0I} \bar c_{k}^{I} \bigg) \\
& + \eta^0 \bigg( u_i^{1J} + u^{0J}_{i,j} y^{IJ}_j - u^{1I}_i + \varepsilon_{ijk} \bigg( \varphi_j^{0J} + \omega_j^{1J} + \omega_{j,m}^{0J} y^{IJ}_m \bigg) \bar c_{k}^{J} - \varepsilon_{ijk} \bigg( \varphi_j^{0I} + \omega_j^{1I} \bigg) \bar c_{k}^{I} \bigg) \\
& + \eta \bigg( u^{1J}_{i,j} y^{IJ}_j + \frac{1}{2}u^{0J}_{i,jk} y^{IJ}_j y^{IJ}_k + \varepsilon_{ijk} \bigg( \varphi_j^{1J} + \varphi_{j,m}^{0J} y^{IJ}_m + \omega_{j,m}^{1J} y^{IJ}_m + \frac{1}{2} \omega^{0J}_{j,mn} y^{IJ}_m y^{IJ}_n \bigg) \bar c_{k}^{J} \\
& ~~~~~~ - \varepsilon_{ijk} \varphi_j^{1I} \bar c_{k}^{I} \bigg)
\bigg] e^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
In equation above, terms of order two and higher are neglected. Multiple scale definition of facet strain is derived, which consists of three classes of terms of $\mathcal{O}(-1)$, $\mathcal{O}(0)$, and $\mathcal{O}(1)$. Multiple scale definition of facet curvature vector will be obtained subsequently. Taylor series definition of rotation of particle $J$ with respect to particle $I$ in macro coordinate system, Equation \[taylor-2-J\], should be inserted into definition of facet curvature, Equation \[curvature\]
$$\label{cur-exp-1}
\chi_{\alpha}=\eta^{-1} \bar{r}^{-1} \left[ \theta_i^J+\eta \theta_{i,j}^J y^{IJ}_j + \frac{1}{2} \eta^2 \theta_{i,jk}^J y^{IJ}_j y^{IJ}_k - \theta^I_i \right] {e}^{IJ}_{\alpha i}$$
Asymptotic expansion of rotation, Equation \[rot-expansion\], along with its macroscopic first and second order derivatives, Equation \[rot-asymp-der\], are inserted into Equation \[cur-exp-1\]
$$\begin{aligned}
\label{cur-exp-2}
\begin{aligned}
\chi_{\alpha}= \eta^{-1} \bar{r}^{-1} \bigg[ & \eta^{-1} \omega_i^{0J} + \varphi_i^{0J} + \omega_i^{1J} + \eta \varphi_i^{1J} + \omega_{i,m}^{0J} y^{IJ}_m + \eta \varphi_{i,m}^{0J} y^{IJ}_m + \eta \omega_{i,m}^{1J} y^{IJ}_m + \eta^2 \varphi_{i,m}^{1J} y^{IJ}_m \\
& \eta \frac{1}{2} \omega^{0J}_{i,mn} y^{IJ}_m y^{IJ}_n + \eta^2 \frac{1}{2} \varphi^{0J}_{i,mn} y^{IJ}_m y^{IJ}_n + \eta^2 \frac{1}{2} \omega^{1J}_{i,mn} y^{IJ}_m y^{IJ}_n + \eta^3 \frac{1}{2} \omega^{1J}_{i,mn} y^{IJ}_m y^{IJ}_n \\
& - \eta^{-1} \omega_i^{0J} - \varphi_i^{0J} - \omega_i^{1J} - \eta \varphi_i^{1J}
\bigg] {e}^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Collecting the terms of the same order and neglecting the ones of order more than zero, one can restate above equation as
$$\begin{aligned}
\label{cur-exp-3}
\begin{aligned}
\chi_{\alpha}=\bar{r}^{-1} \bigg[ & \eta^{-2} \bigg( \omega_i^{0J} - \omega_i^{0I} \bigg) +\\
& \eta^{-1} \bigg( \varphi_i^{0J}+ \omega_i^{1J} + \omega_{i,j}^{0J} y^{IJ}_j - \varphi_i^{0I} - \omega_i^{1I} \bigg) \\
& + \eta^0 \bigg(\varphi_i^{1J} + \omega_{i,j}^{1J} y^{IJ}_j + \varphi_{i,j}^{0J} y^{IJ}_j + \frac{1}{2} \omega_{i,jk}^{0J} y^{IJ}_j y^{IJ}_k - \varphi_i^{1I} \bigg)
\bigg] {e}^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Equation \[cur-exp-3\] is the multiple scale definition of facet curvature vector, which consists of terms of $\mathcal{O}(-2)$, $\mathcal{O}(-1)$, and $\mathcal{O}(0)$.
Asymptotic Expansion of Facet Strain and Curvature using definition of rigid body motion of RVE {#Revised-strain-curvature}
===============================================================================================
Multiple scale definition of facet strain, Equation \[eps-expansion-3\], and facet curvature, Equation \[cur-exp-3\], can be rewritten regarding the definition of ${\mathbf}{u}^0$, Equation \[U0\]. One can calculate first and second partial derivative of $\mathbf{u}^0$ with respect to $\mathbf{x}$ as follows
$$\label{UJ0-der}
u^{0J}_{i,j} = v^{0J}_{i,j} +\varepsilon_{imn} \omega ^{0J}_{m,j} y^J_n \hspace{0.5 in} u^{0J}_{i,jk} = v^{0J}_{i,jk} +\varepsilon_{imn} \omega ^{0J}_{m,jk} y^J_n$$
Using Eqs. \[U0\], \[UJ0-der\] along with the fact that ${\mathbf}{v}^0$, $\mb{\omega}^0$ and $\mb{\varphi}^0$ are constant over the RVE: $\mathbf{v}^{0I} = \mathbf{v}^{0J} = \mathbf{v}^{0}$, $\mb {\omega}^{0I} = \mb{\omega}^{0J} = \mb{\omega}^{0}$ and $\mb {\varphi}^{0} = \mb{\omega}^{0}$, one can revise Equation \[eps-expansion-3\]
$$\begin{aligned}
\label{eps-expansion-4}
\begin{aligned}
\epsilon_{\alpha}=\bar{r}^{-1} \bigg[ & \eta^{-1} \bigg( v_i^{0} - v^{0}_i + \varepsilon_{ijk} \omega_j^{0} ( \bar y_{k}^{J} - \bar y_{k}^{I} ) + \varepsilon_{ijk} \omega_j^{0} ( \bar c_{k}^{J} - \bar c_{k}^{I} ) \bigg) \\
&+ \eta^0 \bigg( u_i^{1J} - u^{1I}_i + \varepsilon_{ijk} \omega_j^{1J} \bar c_{k}^{J} - \varepsilon_{ijk} \omega_j^{1I} \bar c_{k}^{I} \\
& ~~~~~~~ + v^{0}_{i,j} y^{IJ}_j + \varepsilon_{ijk} \omega_j^{0} ( \bar c_{k}^{J} - \bar c_{k}^{I} ) + \varepsilon_{imn} \omega ^{0}_{m,j} y^{IJ}_j y^J_n + \varepsilon_{ijk} \omega_{j,m}^{0} y^{IJ}_m \bar c_{k}^{J} \bigg) \\
& + \eta \bigg( u^{1J}_{i,j} y^{IJ}_j + \frac{1}{2}v^{0}_{i,jk} y^{IJ}_j y^{IJ}_k + \frac{1}{2} \varepsilon_{imn} \omega^{0}_{m,jk} y^{IJ}_j y^{IJ}_k y^{J}_n + \frac{1}{2} \varepsilon_{ijk} \omega^{0J}_{j,mn} y^{IJ}_m y^{IJ}_n \bar c_{k}^{J} \\
& ~~~~~~ + \varepsilon_{ijk} \bigg( \varphi_j^{1J} + \omega_{j,m}^{0} y^{IJ}_m + \omega_{j,m}^{1J} y^{IJ}_m \bigg) \bar c_{k}^{J}
- \varepsilon_{ijk} \varphi_j^{1I} \bar c_{k}^{I} \bigg) \bigg] e^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Using $\mathbf{y}^{IJ} = \mathbf{y}^{J} - \mathbf{y}^{I}$ and $\mathbf{y}^{IJ} = \mathbf{\bar c}^{I} - \mathbf{\bar c}^{J}$ in above equation along with ${\mathbf}{y}^J+{\mathbf}{\bar c}^J={\mathbf}{y}^c$ , one would get
$$\begin{aligned}
\label{eps-expansion-5}
\begin{aligned}
\epsilon_{\alpha}=\bar{r}^{-1} \bigg[ & \eta^0 \bigg( u_i^{1J} - u^{1I}_i + \varepsilon_{ijk} \omega_j^{1J} \bar c_{k}^{J} - \varepsilon_{ijk} \omega_j^{1I} \bar c_{k}^{I} + v^{0}_{i,j} y^{IJ}_j - \varepsilon_{ijk} \omega_j^{0} y^{IJ}_k + \varepsilon_{ijk} \omega_{j,m}^{0} y^{IJ}_m y_{k}^{c} \bigg) \\
& + \eta \bigg( u^{1J}_{i,j} y^{IJ}_j + \varepsilon_{ijk} \varphi_j^{1J} \bar c_{k}^{J} + \varepsilon_{ijk} \omega_{j,m}^{1J} y^{IJ}_m \bar c_{k}^{J}
- \varepsilon_{ijk} \varphi_j^{1I} \bar c_{k}^{I} \\
& ~~~~~~ + \frac{1}{2}v^{0}_{i,jk} y^{IJ}_j y^{IJ}_k + \frac{1}{2} \varepsilon_{ijk} \omega^{0}_{j,mn} y^{IJ}_m y^{IJ}_n y_{k}^{c} + \varepsilon_{ijk} \omega_{j,m}^{0} y^{IJ}_m \bar c_{k}^{J} \bigg) \bigg] e^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Multiple scale definition of facet curvature can also be revised by using $\mb {\omega}^{0I} = \mb{\omega}^{0J} = \mb{\omega}^{0}$ and $\mb {\varphi}^{0} = \mb{\omega}^{0} $ along with $\mathbf{y}^{IJ} = \mathbf{y}^{J} - \mathbf{y}^{I}$ , one can rewrite Equation \[cur-exp-3\]
$$\begin{aligned}
\label{cur-exp-4}
\begin{aligned}
\chi_{\alpha}=\bar{r}^{-1} \bigg[ & \eta^{-1} \bigg( \omega_i^{1J} + \omega_{i,j}^{0J} y^{IJ}_j - \omega_i^{1I} \bigg) \\
& + \eta^0 \bigg(\varphi_i^{1J} + \omega_{i,j}^{1J} y^{IJ}_j - \varphi_i^{1I} + \omega_{i,j}^{0} y^{IJ}_j + \frac{1}{2} \omega_{i,jk}^{0J} y^{IJ}_j y^{IJ}_k \bigg)
\bigg] {e}^{IJ}_{\alpha i}
\end{aligned}\end{aligned}$$
Macroscopic Translational and Rotational Equilibruim Equations {#MacroEquil-Derivation}
==============================================================
In order to derive macroscopic RVE translational equation of motion, one should consider the terms of $\mathcal{O}(1)$ in Equation \[motion-1-sep\]
$$\label{macro-trans-derv-1}
\bar{M}_u^I\ddot {u}_i^{0I} = \sum_{\mathcal{F}_I}{\bar{A}\, t^1_{\alpha} {e}_{\alpha i}^{IJ}} + \bar{V}^I b_i^0$$
Scaling back Equation \[macro-trans-derv-1\] by multiplying both sides of the equation by $\eta^3$ and using the definition of $t^1_{\alpha}$ presented in Equation \[ZeroOne-Terms-Def\], one can get
$$\label{macro-trans-derv-2}
{M}_u^I \ddot {u}_i^{0I} = \eta \sum_{\mathcal{F}_I}{{A} \frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + {V}^I {b}^{0}_i$$
where $t_i^{IJ}=t^0_\beta e_{\beta i}^{IJ}$. Equation \[macro-trans-derv-2\] represents the $\mathcal{O}(1)$ translational equilibrium equation for each particle inside the RVE. One can derive the RVE macroscopic translational equilibrium equation by summing up Equation \[macro-trans-derv-2\] over all RVE particles and dividing by the RVE volume $V_0$
$$\label{macro-trans-derv-3}
\frac{1}{V_0}\sum_I {M}_u^I (\ddot {v}_i^{0I} + \varepsilon_{imn} \ddot {\omega} ^{0I}_{m} y^I_n ) = \frac{1}{V_0}\sum_I \sum_{\mathcal{F}_I}{\eta A \frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + \frac{1}{V_0}\sum_I {V}^I {b}^{0}_i$$
In above equation ${u}_i^{0I}$ is replaced by its definition, Equation \[U0\]. Considering the fact that ${v}_i^{0I}$ and ${\omega}_m^{0I}$ are equal for all RVE particles and the body force $b_i^0$ is considered to be constant over the RVE, Equation \[macro-trans-derv-3\] can be written as
$$\label{macro-trans-derv-4}
\ddot {v}_i^{0} \bigg(\frac{1}{V_0}\sum_I {M}_u^I \bigg) + \varepsilon_{imn} \ddot {\omega} ^{0}_{m} \bigg(\frac{1}{V_0}\sum_I {M}_u^I y^I_n \bigg) = \frac{1}{V_0}\sum_I \sum_{\mathcal{F}_I}{\eta A \frac {\partial {t}^{IJ}_{i}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + {b}^{0}_i \bigg(\frac{1}{V_0}\sum_I {V}^I \bigg)$$
Second term on the left hand side of the Equation \[macro-trans-derv-4\] is equal to zero considering the assumption that the local system of reference is the mass center of the particle system within the RVE; $\sum_I {M}_u^I y^{I}_i=0$. Final form of the Equation \[macro-trans-derv-4\] is presented in Equation \[macro-1-1-averaged\] in Section \[macro-derivation\].
Macroscopic RVE rotational equation of motion can be derived by considering the terms of $\mathcal{O}(1)$ in Equation \[motion-2-sep\]. To have a consistent formulation for all particles and RVEs, one should consider the moment of all forces with respect to a fixed point in space, say the origin of a global coordinate system as shown in Figure \[TwoScaleAnalysis\]b, which implies that the moment of Equation \[macro-trans-derv-1\] should be taken into account. Therefore, one can write $\mathcal{O}(1)$ moment equilibrium equation of particle $I$ as
$$\label{macro-rot-derv-1}
\bar{M}_u^I \varepsilon_{ijk} Y^{I}_j \ddot {u}_k^{0I} + \bar {M_\theta^I}\ddot{\omega}_i^{0I} = \sum_{\mathcal{F}_I} \bar{A}\, (p^{1}_{\alpha}{e}_{\alpha i}^{IJ} + q^{1}_{\alpha}{e}_{\alpha i}^{IJ}) + \bar{V}^I \varepsilon_{ijk} Y^{I}_j {b}_k^{0}$$
where $Y^I_j$ is the position vector of particle $I$ in the fine-scale global coordinate system $\mathbf{Y}=\mathbf{X}/\eta$; $p^{1}_{\alpha} {\mathbf e}_\alpha^{IJ} = {\mathbf{Y}}^C \times t^{1}_{\alpha} {\mathbf e}_\alpha^{IJ}$ is the moment of the facet traction with respect to the origin of the fine-scale global coordinate system, in which ${\mathbf{Y}}^C = {\mathbf{X}}^C/\eta$ is the position vector of the contact point $C$ between particles $I$ and $J$ in the global coordinate system. Scaling back Equation \[macro-rot-derv-1\] by multiplying both sides of the equation by $\eta^4$ and using the definition of $p^1_{\alpha}$ and $q^1_{\alpha}$ presented in Equation \[ZeroOne-Terms-Def\], one can get
$$\label{macro-rot-derv-2}
{M}_u^I \varepsilon_{ijk} X^{I}_j \ddot {u}_k^{0I} + \eta^{-1} {M_\theta^I} \ddot{\omega}_i^{0I} = \eta \sum_{\mathcal{F}_I} A \left( {\frac {\partial {w}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + \frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} \right) + {V}^I \varepsilon_{ijk} X^{I}_j {b}_k^{0}$$
Equation \[macro-rot-derv-2\] represents the $\mathcal{O}(1)$ rotational equilibrium equation for each particle inside the RVE. RVE macroscopic rotational equilibrium equation can be obtained by summing up Equation \[macro-rot-derv-2\] over all RVE particles and dividing by the RVE volume $V_0$
$$\label{macro-rot-derv-3}
\begin{aligned}
\frac{1}{V_0}\sum_I {M}_u^I \varepsilon_{ijk} X^{I}_j (\ddot {v}_k^{0I} + & \varepsilon_{kmn} \eta^{-1} \ddot {\omega} ^{0I}_{m} x^I_n ) + \frac{1}{V_0}\sum_I \eta^{-1} {M_\theta^I} \ddot{\omega}_i^{0I} = \\
& \frac{\eta }{V_0}\sum_I \sum_{\mathcal{F}_I} A \left( {\frac {\partial {w}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + \frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} \right) + \frac{1}{V_0}\sum_I {V}^I \varepsilon_{ijk} X^{I}_j {b}_k^{0}
\end{aligned}$$
In above equation ${u}_i^{0I}$ is replaced by its definition, Equation \[U0\]. Considering equality of ${v}_i^{0I}$ and ${\omega}_m^{0I}$ for all RVE particles along with $X^I_j = X_j + x^I_j$, Equation \[macro-rot-derv-3\] can be written as
$$\label{macro-rot-derv-4}
\begin{aligned}
& \frac{1}{V_0}\sum_I {M}_u^I \varepsilon_{ijk} X_j \ddot {v}_k^{0} + \frac{1}{V_0}\sum_I \left( {M_\theta^I} \delta_{im}+ {M}_u^I \varepsilon_{ijk} \varepsilon_{kmn} x^I_j x^I_n\right) \eta^{-1} \ddot{\omega}_m^{0} \\
& + \frac{1}{V_0}\sum_I {M}_u^I \varepsilon_{ijk} x^I_j \ddot {v}_k^{0} + \frac{1}{V_0}\sum_I {M}_u^I \varepsilon_{ijk} \varepsilon_{kmn} X_j x^I_n \eta^{-1} \ddot {\omega} ^{0}_{m} = \\
& \frac{\eta }{V_0}\sum_I \sum_{\mathcal{F}_I} A \left( {\frac {\partial {w}_{i}^{IJ}}{\partial \epsilon^0_{\alpha}} \epsilon^1_{\alpha}} + \frac {\partial {m}_{i}^{IJ}}{\partial \psi^{0}_{\alpha}} \psi^1_{\alpha} \right) + \frac{1}{V_0}\sum_I {V}^I \varepsilon_{ijk} X_j {b}_k^{0} + \frac{1}{V_0}\sum_I {V}^I \varepsilon_{ijk} x^I_j {b}_k^{0}
\end{aligned}$$
Considering $\sum_I {M}_u^I x^{I}_i=0$ along with the equality of ${v}_i^{0}$, ${\omega}_m^{0}$, $X_j$ for all RVE particles, one can conclude that the third and the forth terms on the left hand side and the last term on the right hand side of the Equation \[macro-rot-derv-4\] is equal to zero. Final form of the Equation \[macro-rot-derv-4\] is presented in Equation \[macro-2-averaged-init\] in Section \[macro-derivation\].
[^1]: Research Assistant, Northwestern University, CEE Department, 2145 N Sheridan Rd Evanston, IL 60208, USA.
[^2]: Corresponding author. Email: [email protected]. Associate Professor, Northwestern University, CEE Department, 2145 N Sheridan Rd Evanston, IL 60208, USA.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A three-dimensional continuum dislocation theory for single crystals containing curved dislocations is proposed. A set of governing equations and boundary conditions is derived for the true placement, plastic slips, and loop functions in equilibrium that minimize the free energy of crystal among all admissible functions, provided the resistance to the dislocation motion is negligible. For the non-vanishing resistance to dislocation motion the governing equations are derived from the variational equation that includes the dissipation function. A simplified theory for small strains is also provided. An asymptotic solution is found for the two-dimensional problem of a single crystal beam deforming in single slip and simple shear.'
address: |
Lehrstuhl für Mechanik - Materialtheorie, Ruhr-Universität Bochum,\
D-44780 Bochum, Germany
author:
- 'K. C. Le[^1]'
title: 'Three-dimensional continuum dislocation theory'
---
dislocations (A) ,crystal plasticity (B) ,finite strain (B) ,variational calculus (C) .
Introduction {#sec:introduction}
============
In view of a huge number of dislocations appearing in plastically deformed crystals (which typically lies in the range $10^8\div 10^{15}$ dislocations per square meter) the necessity of developing a physically meaningful continuum dislocation theory (CDT) to describe the evolution of dislocation network and predict the formation of microstructure in terms of mechanical and thermal loading conditions becomes clear to all researchers in crystal plasticity. One of the main guiding principles in seeking such a continuum dislocation theory has first been proposed by @Hansen1986 in form of the so-called LEDS-hypothesis: the true dislocation structure in the final state of deformation minimizes the energy of crystal among all admissible dislocation configurations. In view of numerous experimental evidences supporting this hypothesis (see, i.e., [@Hughes1997; @Kuhlmann1989; @Kuhlmann2001; @Laird1986]), its use in constructing the continuum dislocation theory seems to be quite reasonable and appealing. For the practical realization one needs to i) specify the whole set of unknown functions and state variables of the continuum dislocation theory, and ii) lay down the free energy of crystals as their functional to be minimized. Such program has been implemented by @Berdichevsky06a in the linear, and by @Le2014nonlinear in the nonlinear setting of CDT for networks of dislocations, whose lines are straight and remain so during the whole deformation process (see also [@Ortiz99; @Ortiz00]). The developed CDT has been successfully applied to various two-dimensional problems of dislocation pileups, bending, torsion, as well as formation of dislocation patterns in single crystals (see [@Berdichevsky-Le07; @Kaluza2011torsion; @Kochmann08b; @Kochmann08a; @Kochmann09; @Koster2015; @Le08a; @Le08b; @Le2009plane; @Le2012polygonization; @Le2013on]). Let us mention here the similar approaches suggested in [@Acharya2000; @Acharya2001; @Engels2012; @Gurtin2002; @Gurtin2007; @Mayeur2014; @Oztop2013] which do not use the LEDS-hypothesis explicitly but employ instead the extended principle of virtual work for the gradient plasticity. However, as experiences and experiments show, dislocation lines are in general loops that, as a rule, can change their directions and curvatures depending on the loading condition and crystal’s geometry. Therefore the extension of CDT to networks of dislocations whose lines are curves in the slip planes is inevitable. To the best of author’s knowledge, such three-dimensional continuum dislocation theory based on the LEDS-hypothesis for curved dislocations has not been developed until now. It became also clear to him that the latter’s absence was due to the missing scalar dislocation densities for the network of curved dislocations.
The first attempt at constructing a continuum theory that can predict in principle not only the dislocation densities but also the direction and curvature of the dislocation lines has been made by @Hochrainer2007 in form of the so-called continuum dislocation dynamics. Their theory starts with the definition of the dislocation density that contains also the information about the orientation and curvature of the dislocation lines. Then the set of kinematic equations is derived for the dislocation density and curvature that requires the knowledge about the dislocation velocity. The relation between the dislocation density and the macroscopic plastic slip rate via the dislocation velocity is postulated in form of Orowan’s equation. The couple system of crystal plasticity and continuum dislocation dynamics becomes closed by the constitutive equation of a flow rule type (see [@Hochrainer2014; @Sandfeld2011; @Sandfeld2015; @Wulfinghoff2015]). In addition to the heavy computational cost of such theory, the relation to thermodynamics of crystal plasticity and to the LEDS-hypothesis is completely lost: the equilibrium solution found in this theory may not minimize the energy of crystal among all admissible dislocation configurations. Let us mention also a continuum approach proposed recently by @Zhu2013 [@Zhu2014] in which the three-dimensional dislocation structure is characterized by two families of disregistry functions that may take only integer values. The dislocation density can then be expressed in their terms. The coupled system of equations is derived from the underlying discrete dislocation dynamics for the displacement and disregistry functions. This approach is subject to the same critics as that proposed in [@Hochrainer2007].
The aim of this paper is to extend the nonlinear continuum dislocation theory (CDT) developed recently by @Le2014nonlinear to the case of crystals containing curved dislocations. Provided the dislocation network is regular in the sense that nearby dislocations have nearly the same direction and orientation, we introduce a loop function whose level curves coincide with the dislocation lines. Taking an infinitesimal area perpendicular to the dislocation line at some point of the crystal, we express the densities of edge and screw dislocations at that point through the resultant Burgers vectors of dislocations whose lines cross this area at right angle. Such scalar densities contain not only the information about the number of dislocations, but also the information about the orientation and curvature of the dislocation lines. In case of dislocation motion we introduce the vector of normal velocity of dislocation line through the time derivative of the loop function. Following @Kroener92 and [@Berdichevsky06] we require that the free energy density of crystal depends only on the elastic strain tensor and on the above scalar densities of dislocations. Then we formulate a new variational principle of CDT according to which the placement, the plastic slip, and the loop function in the final state of equilibrium minimize the free energy functional among all admissible functions. We derive from this variational principle a new set of equilibrium equations, boundary conditions, and constitutive equations for these unknown functions. In case the resistance to dislocation motion is significant, the variational principle must be replaced by the variational equation that takes the dissipation into account. The constructed theory is generalized for single crystals having a finite number of active slip systems. We provide also the simplifications of the theory for small strains. As compared to the continuum dislocation dynamics proposed in [@Hochrainer2007; @Zhu2014] our theory is advantageous not only in the computational cost due to its simplicity, but also in its full consistency with the LEDS-hypothesis. In the problem of single crystal beam having only one active slip system and deforming in simple shear, the energy minimization problem reduces to the two-dimensional variational problem. We solve this problem analytically for the circular cross section and asymptotically for the rectangular cross section. We will show that this solution reduces to that found in [@Berdichevsky-Le07] for the crystals with thin and long cross-section.
The paper is organized as follows. After this short introduction we present in Section 2 the three-dimensional kinematics for single crystals deforming in single slip. Section 3 formulates the variational principles of the three-dimensional CDT and derives its governing equations. Section 4 extends this nonlinear theory to the case of single crystals with $n$ active slip system. Section 5 studies the three-dimensional small strain CDT. Section 6 is devoted to the analytical and asymptotic solutions of the two-dimensional energy minimization problem of a single crystal beam deforming in simple shear. Finally, Section 7 concludes the paper.
3-D kinematics for single crystals deforming in single slip {#sec:ncdt}
===========================================================
Nonlinear CDT starts from the basic kinematic resolution of the deformation gradient $\mathbf{F}=\partial \mathbf{y}/\partial \mathbf{x}$ into elastic and plastic parts [@Bilby57] $$\label{eq:2.1}
\mathbf{F}=\mathbf{F}^e\cdot \mathbf{F}^p.$$ We attribute an active role to the plastic deformation: $\mathbf{F}^p$ is the deformation [*creating*]{} dislocations (either inside or at the boundary of the volume element) or [*changing*]{} their positions in the crystal without distorting the lattice parallelism (see Fig. \[fig:Nhslip\]). On the contrary, the elastic deformation $\mathbf{F}^e$ deforms the crystal lattice having [*frozen*]{} dislocations [@Le2014nonlinear]. Note that the lattice vectors remain unchanged when the plastic deformation is applied, while they change together with the shape vectors by the elastic deformation.
![Multiplicative decomposition[]{data-label="fig:Nhslip"}](Fp){height="6.5cm"}
We consider first a single crystal deforming in single slip. In this case let us denote the right-handed triad of unit lattice vectors of the active slip system by $\mathbf{s}$, $\mathbf{p}$, and $\mathbf{m}$, where $\mathbf{s}$ points to the slip direction, $\mathbf{p}$ lies in the slip plane and is perpendicular to $\mathbf{s}$, and $\mathbf{m}$ is normal to the slip plane. Without restricting generality we may choose the rectangular cartesian coordinate system $(x_1,x_2,x_3)$ in the reference configuration such that its basis vectors coincide with these lattice vectors (see Fig. \[fig:loop\]) $$\mathbf{e}_1=\mathbf{s},\quad \mathbf{e}_2=\mathbf{p}, \quad \mathbf{e}_3=\mathbf{m}.$$ The plastic deformation is then given by $$\label{eq:2.0}
\mathbf{F}^p=\mathbf{I}+\beta (\mathbf{x})\mathbf{s}\otimes \mathbf{m}=\mathbf{I}+\beta (\mathbf{x})\mathbf{e}_1\otimes \mathbf{e}_3,$$ with $\beta $ being the plastic slip. We assume that all dislocations causing this plastic deformation lie completely in the slip planes and the dislocation network is regular in the sense that nearby dislocations have nearly the same direction and orientation. This enables one to introduce a scalar function $l(x_1,x_2,x_3)$ (called a loop function) such that its level curves $$\label{eq:2.a}
l(x_1,x_2,c_3)=c,$$ with $c_3$ and $c$ being constants, coincide with the dislocation lines. Thus, in this three-dimensional kinematics we admit, according to equation , only the conservative motion of dislocations and exclude from consideration the dislocation climb which is an important mechanism of temperature-dependent creep. We denote by ${\boldsymbol{\nu}}$ and ${\boldsymbol{\tau}}$ the plane unit vectors normal and tangential to the dislocation line. From equation follow $${\boldsymbol{\nu}}=\frac{1}{\sqrt{l_{,1}^2+l_{,2}^2}}(l_{,1}\mathbf{e}_1+l_{,2}\mathbf{e}_2), \quad {\boldsymbol{\tau}}=\frac{1}{\sqrt{l_{,1}^2+l_{,2}^2}}(-l_{,2}\mathbf{e}_1+l_{,1}\mathbf{e}_2),$$ where the comma before an index denotes the partial derivative with respect to the corresponding coordinate. Note that ${\boldsymbol{\nu}}$, ${\boldsymbol{\tau}}$, $\mathbf{m}$ form a right-handed basis vectors of the three-dimensional space (see Fig. \[fig:loop\]).
![A dislocation loop in the chosen coordinate system[]{data-label="fig:loop"}](loop){height="6.5cm"}
@Ortiz99 introduced the resultant Burgers vector of excess dislocations, whose lines cross the area $\mathcal{A}$ in the reference configuration, in the following way $$\label{eq:2.2}
\mathbf{b}_r=\oint_{\mathcal{C}} \mathbf{F}^p\cdot d\mathbf{x} ,$$ where $\mathcal{C}$ is the close contour surrounding $\mathcal{A}$. @Le2014nonlinear have shown that, in the continuum limit, when the atomic distance goes to zero at the fixed sizes of the representative volume element and the fixed density of dislocations per area of unit cell, integral gives the total closure failure induced by $\mathbf{F}^p$ which must be equal to the resultant Burgers vector. It is natural to assume $\mathbf{F}^p$ continuously differentiable in this continuum limit, so, applying Stoke’s theorem we get from $$\mathbf{b}_r=- \int _{\mathcal{A}}(\mathbf{F}^p\times \nabla ) \cdot \mathbf{n}da,$$ where $\times $ denotes the vector product, $da$ the surface element, and $\mathbf{n}$ the unit vector normal to $\mathcal{A}$. This legitimates the introduction of the dislocation density tensor $$\mathbf{T}=-\mathbf{F}^p\times \nabla .$$ For the plastic deformation taken from $$\mathbf{T}=-\mathbf{F}^p\times \nabla =\mathbf{s}\otimes (\nabla \beta \times \mathbf{m}).$$ If we choose now an infinitesimal area $da$ with the unit normal vector ${\boldsymbol{\tau}}$, then the resultant Burgers vector of all excess dislocations, whose dislocation lines cross this area at right angle is given by $$\mathbf{b}_r=\mathbf{T}\cdot {\boldsymbol{\tau}}\, da=-\mathbf{s}\, (\nabla \beta \cdot {\boldsymbol{\nu}})da=-\mathbf{s}\, \partial _\nu \beta \, da.$$ This resultant Burgers vector can be decomposed into the sum of two vectors $$\mathbf{b}_r=\mathbf{b}_{r\perp}+\mathbf{b}_{r\parallel}=-({\boldsymbol{\nu}}s_\nu +{\boldsymbol{\tau}}s_\tau )\partial _\nu \beta\, da,$$ where $s_\nu =\mathbf{s}\cdot {\boldsymbol{\nu}}=\nu _1$ and $s_\tau =\mathbf{s}\cdot {\boldsymbol{\tau}}=\tau _1$ are the projections of the slip vector onto the normal and tangential direction to the dislocation line, respectively. This allows us to define two scalar densities (or the numbers of excess dislocations per unit area) of edge and screw dislocations $$\label{eq:2.3a}
\begin{split}
\rho_\perp =\frac{|\mathbf{b}_{r\perp}|}{b}=\frac{1}{b}|s_\nu \partial _\nu \beta| =\frac{1}{b}\left| \frac{l_{,1}(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{l_{,1}^2+l_{,2}^2}\right| ,
\\
\rho_\parallel =\frac{|\mathbf{b}_{r\parallel}|}{b}=\frac{1}{b}|s_\tau \partial _\nu \beta|=\frac{1}{b}\left| \frac{l_{,2}(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{l_{,1}^2+l_{,2}^2}\right| ,
\end{split}$$ with $b$ the magnitude of Burgers vector. We see that the three-dimensional dislocation densities $\rho_\perp $ and $\rho_\parallel $ depend on both the gradient of the plastic slip and the gradient of the loop function $l(\mathbf{x})$ through the vectors ${\boldsymbol{\nu}}$ and ${\boldsymbol{\tau}}$.
Consider now the case of motion of dislocation loops in the slip plane. In this case we allow the loop function to depend explicitly on time $t$ such that equation $$\label{eq:2.4}
l(x_1,x_2,c_3,t)=c$$ with fixed constants $c_3$ and $c$ describes one and the same dislocation line during its motion in the slip plane. Letting $\zeta $ be the variable along the dislocation line, we may represent the level curve defined by in the parametric form $$x_1=x_1(\zeta ,t),\quad x_2=x_2(\zeta ,t).$$ Fixing $\zeta $ and taking the differential of we obtain $$l_{,1}dx_1+l_{,2}dx_2+l_{,t}dt=0$$ that yields $$l_{,1}\frac{dx_1}{dt}+l_{,2}\frac{dx_2}{dt}=-l_{,t}.$$ Since $\mathbf{v}=\frac{dx_1}{dt}\mathbf{e}_1+\frac{dx_2}{dt}\mathbf{e}_2$ is the velocity of the fixed point on the dislocation line with coordinate $\zeta $, we define the normal velocity of the dislocation line as follows $$\label{eq:2.5}
v_\nu =\mathbf{v}\cdot {\boldsymbol{\nu}}=\frac{1}{\sqrt{l_{,1}^2+l_{,2}^2}}(l_{,1}\frac{dx_1}{dt}+l_{,2}\frac{dx_2}{dt})=-\frac{\dot{l}}{\sqrt{l_{,1}^2+l_{,2}^2}},$$ with $\dot{l}=l_{,t}$. This kinematic quantity will be used in the case of non-vanishing resistance to dislocation motion.
Governing equations for single crystals deforming in single slip
================================================================
According to @Kroener92, the elastic deformation $\mathbf{F}^e$ and the dislocation densities $\rho_\perp $ and $\rho_\parallel $ characterize the current state of the crystal, so these quantities are the state variables of the continuum dislocation theory. The reason why the plastic deformation $\mathbf{F}^p$ cannot be qualified for the state variable is that it depends on the cut surfaces and consequently on the whole history of creating dislocations. Likewise, the gradient of plastic strain tensor $\mathbf{C}^p$ cannot be used as the state variable by the same reason. In contrary, the dislocation densities depend only on the characteristics of dislocations in the current state (Burgers vector and positions of dislocation lines) and not on how they are created, so $\rho_\perp $ and $\rho_\parallel $, in addition to $\mathbf{F}^e$, are the proper state variables. Thus, if we consider isothermal processes of deformation, then the free energy per unit volume of crystal (assumed as macroscopically homogeneous) must be a function of $\mathbf{F}^e$, $\rho_\perp $, and $\rho_\parallel $ $$\psi = \psi (\mathbf{F}^e,\rho_\perp ,\rho_\parallel ).$$ Now, if we superimpose an elastic rotation $\mathbf{R}$ onto the actual deformation of the body, then the total and elastic deformation change according to $$\mathbf{F}^{\ast}=\mathbf{R}\cdot \mathbf{F},\quad \mathbf{F}^{e\ast}=\mathbf{R}\cdot \mathbf{F}^e.$$ At the same time, the dislocation densities $\rho_\perp $ and $\rho_\parallel $ remain unchanged. As such superimposed elastic rotation does not change the elastic strain and the dislocation densities, we expect that the energy remains unchanged. The standard argument (see, e.g., [@Gurtin1981]) leads then to $$\psi = \psi (\mathbf{C}^e,\rho_\perp ,\rho_\parallel ),$$ where $\mathbf{C}^e$ is the elastic strain defined by $$\mathbf{C}^e=\mathbf{F}^{eT}\cdot \mathbf{F}^e.$$
Let the undeformed single crystal occupy some region $\mathcal{V}$ of the three-dimensional space. The boundary of this region, $\partial \mathcal{V}$, is assumed to be the closure of union of two non-intersecting surfaces, $\partial _k$ and $\partial _s$. Let the placement be a given smooth function of coordinates $$\label{eq:2.6}
\mathbf{y}(\mathbf{x})=\mathbf{x}+\mathbf{u}_0(\mathbf{x}) \quad \text{at $\partial _k$},$$ where $\mathbf{u}_0(\mathbf{x})=\mathbf{y}(\mathbf{x})-\mathbf{x}$ is the given displacement vector. Such condition does not admit dislocations to reach this part $\partial _k$ of the boundary, so we set $$\label{eq:2.6a}
\beta (\mathbf{x})=0, \quad l(\mathbf{x})=0 \quad \text{at $\partial _k$}.$$ At the remaining part $\partial _s$ the “dead” load (traction) $\mathbf{t}$ is specified. Note that, in case the whole boundary is free, we do not have any kinematic constraint at $\partial \mathcal{V}$. However, as the energy density is invariant with respect to the shift of the loop function $l(\mathbf{x})$ on an arbitrary constant which does not change the dislocation densities, we can impose on this scalar function the following constraint $$\int_{\mathcal{V}}l(\mathbf{x})\, dx=0,$$ where $dx=dx_1dx_2dx_3$ denotes the volume element. If no body force acts on this crystal, then its energy functional is defined as $$\label{eq:2.7}
I[\mathbf{y}(\mathbf{x}),\beta (\mathbf{x}),l(\mathbf{x})]=\int_{\mathcal{V}}w(\mathbf{F},\beta ,\nabla \beta ,\nabla l)\, dx-\int_{\partial _s} \mathbf{t} \cdot \mathbf{y}\, da,$$ where $$\label{eq:2.8}
w(\mathbf{F},\beta ,\nabla \beta ,\nabla l)=\psi (\mathbf{C}^e,\rho_\perp ,\rho_\parallel ),$$ Provided the resistance to the dislocation motion is negligibly small and no surfaces of discontinuity occur inside crystals, then the following variational principle is valid for single crystals with one active slip system: the true placement vector $\check{\mathbf{y}}(\mathbf{x})$, the true plastic slip $\check{\beta }(\mathbf{x})$, and the true loop function $\check{l}(\mathbf{x})$ in the [*final*]{} equilibrium state of deformation minimize energy functional among all continuously differentiable fields $\mathbf{y}(\mathbf{x})$, $\beta (\mathbf{x})$, and $l(\mathbf{x})$ satisfying constraints and .
Let us derive the equilibrium equations from this variational principle. We compute the first variation of functional $$\delta I= \int_{\mathcal{V}}
\left( \mathbf{P}\mathbf{:}\delta \mathbf{y}\nabla +\frac{\partial w}{\partial \beta }\delta \beta +\frac{\partial w}{\partial \nabla \beta }\cdot \nabla \delta \beta +\frac{\partial w}{\nabla l}\cdot \nabla \delta l \right) dx
- \int_{\partial _s} \mathbf{t} \cdot \delta \mathbf{y}\, da ,$$ where $\mathbf{P}=\partial w /\partial \mathbf{F}$. Integrating the first, third, and fourth term by parts with the help of Gauss’ theorem and taking the conditions and into account, we obtain $$\begin{gathered}
\label{eq:2.9}
\delta I = \int_{\mathcal{V}}[ -\delta \mathbf{y}\cdot (\mathbf{P}\cdot \nabla ) +(w_\beta -\nabla \cdot w_{\nabla \beta })\delta \beta - (\nabla \cdot w_{\nabla l})\delta l ] \, dx
\\
+ \int_{\partial _s} [(\mathbf{P}\cdot \mathbf{n}-\mathbf{t} )\cdot \delta \mathbf{y} +w_{\nabla \beta }\cdot \mathbf{n}\, \delta \beta]\, da + \int_{\partial \mathcal{V}}w_{\nabla l }\cdot \mathbf{n}\, \delta l \, da=0.\end{gathered}$$ Equation implies that the minimizer must satisfy in $\mathcal{V}$ the equilibrium equations $$\label{eq:2.10}
\mathbf{P}\cdot \nabla =0,
\quad
-w_\beta +\nabla \cdot w_{\nabla \beta }=0, \quad \nabla \cdot w_{\nabla l} =0,$$ subjected to the kinematic boundary conditions and at $\partial _k$, and the following natural boundary conditions $$\label{eq:2.11}
\mathbf{P}\cdot \mathbf{n}=\mathbf{t},
\quad
w_{\nabla \beta }\cdot \mathbf{n}=0 , \quad w_{\nabla l }\cdot \mathbf{n}=0 \quad \text{at $\partial _s$}.$$ We call $\mathbf{P}$ the first Piola-Kirchhoff stress tensor, $\tau _r=-w_\beta $ the resolved shear stress (or Schmid stress), and $\varsigma =-\nabla \cdot w_{\nabla \beta }$ the back stress. The first equation of is nothing else but the equilibrium of macro-forces acting on the crystal, the second equation represents the equilibrium of micro-forces acting on dislocations, while the last one expresses the equilibrium condition for the curved dislocation lines.
The constitutive equations for $\mathbf{P}=w_{\mathbf{F}}$, $-w_\beta $, $w_{\nabla \beta }$, and $w_{\nabla l}$ can easily be obtained from the free energy density . First, we express $\mathbf{F}^e$ in terms of $\mathbf{F}$ and $\beta $ with the use of and $$\mathbf{F}^e=\mathbf{F}\cdot \mathbf{F}^{p-1}=\mathbf{F}\cdot (\mathbf{I}-\beta \mathbf{s}\otimes \mathbf{m}).$$ Now, the standard differentiation using the chain rule and the above relation yields the first Piola-Kirchhoff stress tensor $$\label{eq:2.12}
\mathbf{P}=w_{\mathbf{F}}=2\mathbf{F}^e\cdot \psi _{\mathbf{C}^e} \cdot \mathbf{F}^{p-T}.$$ For the resolved shear stress (Schmid stress) we get $$\label{eq:2.13}
\tau _r=-w_\beta =2\mathbf{s}\cdot \mathbf{F}^{p-T}\cdot \mathbf{C}^e \cdot \psi _{\mathbf{C}^e}\cdot \mathbf{m}.$$ Likewise, from follows $$\label{eq:2.14}
w_{\nabla \beta }=\frac{1}{b}[\psi _{\rho_\perp }{\text{sign}}(s_\nu \partial _\nu \beta )s_\nu +\psi _{\rho_\parallel }{\text{sign}}(s_\tau \partial _\nu \beta )s_\tau ] {\boldsymbol{\nu}}.$$ Thus, the vector $w_{\nabla \beta }$ is two-dimensional. Finally, we compute $w_{\nabla l }$ directly in components using formulas . Since $\rho _\perp $ and $\rho _\parallel $ do not depend on $l_{,3}$, so $w_{l_{,3}}=0$, and the vector $w_{\nabla l}$ is also two-dimensional. For its first two components we have $$\label{eq:2.15}
\begin{split}
w_{l_{,1}}=\frac{1}{b}\left[ \psi _{\rho _\perp }{\text{sign}}(s_\nu \partial _\nu \beta )\left( -\frac{2l_{,1}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{(l_{,1}^2+l_{,2}^2)^2}+\frac{2\beta _{,1}l_{,1}+\beta _{,2}l_{,2}}{l_{,1}^2+l_{,2}^2} \right) \right.
\\
+\left. \psi _{\rho _\parallel }{\text{sign}}(s_\tau \partial _\nu \beta )\left( -\frac{2l_{,1}l_{,2}(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{(l_{,1}^2+l_{,2}^2)^2}+\frac{\beta _{,1}l_{,2}}{l_{,1}^2+l_{,2}^2} \right) \right] ,
\\
w_{l_{,2}}=\frac{1}{b}\left[ \psi _{\rho _\perp }{\text{sign}}(s_\nu \partial _\nu \beta )\left( -\frac{2l_{,1}l_{,2}(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{(l_{,1}^2+l_{,2}^2)^2}+\frac{\beta _{,2}l_{,1}}{l_{,1}^2+l_{,2}^2} \right) \right.
\\
+\left. \psi _{\rho _\parallel }{\text{sign}}(s_\tau \partial _\nu \beta )\left( -\frac{2l_{,2}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{(l_{,1}^2+l_{,2}^2)^2}+\frac{\beta _{,1}l_{,1}+2\beta _{,2}l_{,2}}{l_{,1}^2+l_{,2}^2} \right) \right] .
\end{split}$$ Substituting the constitutive equations - into - we get the completely new system of equations and boundary conditions which, together with and , enable one to determine $\check{\mathbf{y}}(\mathbf{x})$, $\check{\beta }(\mathbf{x})$, and $\check{l}(\mathbf{x})$. Note that equations $_1$ and $_2$ are coupled via the first Piola-Kirchhoff stress tensor and the Schmid stress containing both $\mathbf{F}$ and $\beta $, while equations $_2$ and $_3$ are coupled because both contain the gradients of $\beta $ and $l$. All equations are strongly nonlinear partial differential equations.
The above theory has been developed for the case of negligibly small resistance to dislocation motion and plastic slip. In real crystals there is however always the resistance to the dislocation motion and plastic slip causing the energy dissipation that changes the above variational principle as well as the equilibrium conditions. We assume that the dissipation function depends on the plastic slip rate $\dot{\beta }$ and on the normal velocity of the dislocation loop $v_\nu $ given by (or, equivalently, on $\dot{l}$). Thus, $$D=D(\dot{\beta },\dot{l}),$$ When the dissipation is taken into account, the above formulated variational principle must be modified. Following [@Sedov65; @Berdichevsky1967] we require that the true placement $\check{\mathbf{y}}(\mathbf{x},t)$, the true plastic slips $\check{\beta }(\mathbf{x},t)$, and the true loop function $\check{l}(\mathbf{x},t)$ obey the variational equation $$\label{eq:2.15b}
\delta I+\int_{\mathcal{V}} (\frac{\partial D}{\partial \dot{\beta }}\delta \beta +\frac{\partial D}{\partial \dot{l}}\delta l)\, dx=0$$ for all variations of admissible fields $\mathbf{y}(\mathbf{x},t)$, $\beta (\mathbf{x},t)$, and $l(\mathbf{x},t)$ satisfying the constraints and . Together with the above formula for $\delta I$ and the arbitrariness of $\delta \mathbf{y}$, $\delta \beta $, and $\delta l$ in $\mathcal{V}$ as well as at $\partial _s$, equation yields $$\label{eq:2.15c}
\mathbf{P}\cdot \nabla =0,
\quad
-w_\beta +\nabla \cdot w_{\nabla \beta }=\frac{\partial D}{\partial \dot{\beta }}, \quad \nabla \cdot w_{\nabla l}=\frac{\partial D}{\partial \dot{l}},$$ which are subjected to the kinematic boundary conditions and , and the natural boundary conditions . The constitutive equations remain exactly the same as -. For the rate-independent theory the dissipation function can be assumed in a simple form $$D=K_1|\dot{\beta }|+K_2|\dot{l}|,$$ with $K_1$ and $K_2$ being positive constants. We call $K_1$ the critical resolved shear stress and $K_2$ the Peierls threshold. In this case equations $_{2,3}$ become $$-w_\beta +\nabla \cdot w_{\nabla \beta }=K_1\, {\text{sign}}\dot{\beta }, \quad \nabla \cdot w_{\nabla l}=K_2\, {\text{sign}}\dot{l}$$ for non-vanishing $\dot{\beta}$ and $\dot{l}$. These are the yield conditions for $\beta $ and $l$: $\dot{\beta}$ and $\dot{l}$ are non-zero if and only if $$|-w_\beta +\nabla \cdot w_{\nabla \beta }|=K_1, \quad |\nabla \cdot w_{\nabla l}|=K_2.$$ On the contrary, if the expressions on the left-hand sides are less than $K_1$ and $K_2$, the plastic slip cannot evolve and the dislocation lines cannot move: $\dot{\beta}=0$ and $\dot{l}=0$. Thus, they are frozen in the crystal.
Extension to multiple slip
==========================
The extension to the case of single crystals having $n$ active slip systems can be done straightforwardly under the assumption[^2] $$\label{eq:2.16}
\mathbf{F}^p=\mathbf{I}+\sum_{\mathfrak{a}=1}^n \beta ^\mathfrak{a}(\mathbf{x}) \mathbf{s}^\mathfrak{a}\otimes \mathbf{m}^\mathfrak{a} ,$$ with $\beta ^\mathfrak{a}$ being the plastic slip, where the pair of constant and mutually orthogonal unit vectors $\mathbf{s}^\mathfrak{a}$ and $\mathbf{m}^\mathfrak{a}$ is used to denote the slip direction and the normal to the slip planes of the corresponding $\mathfrak{a}$-th slip system, respectively. Here and later, the Gothic upper index $\mathfrak{a}$ running from 1 to $n$ numerates the slip systems, so one could clearly distinguish $\beta ^{\mathfrak{a}}$ from the power function. We denote by $\mathbf{p}^{\mathfrak{a}}$ the unit vector lying in the slip plane such that $\mathbf{s}^{\mathfrak{a}}$, $\mathbf{p}^{\mathfrak{a}}$, and $\mathbf{m}^{\mathfrak{a}}$ form a right-handed basis vectors. For each slip system we can introduce the coordinates associated with these basis vectors $$\label{eq:2.17}
\xi ^{\mathfrak{a}}_1=\mathbf{s}^{\mathfrak{a}}\cdot \mathbf{x}, \quad \xi ^{\mathfrak{a}}_2=\mathbf{p}^{\mathfrak{a}}\cdot \mathbf{x},\quad \xi ^{\mathfrak{a}}_3=\mathbf{m}^{\mathfrak{a}}\cdot \mathbf{x}.$$ Equations can be regarded as the one-to-one linear transformation relating ${\boldsymbol{\xi}}^{\mathfrak{a}}$ and $\mathbf{x}$ according to $${\boldsymbol{\xi}}^{\mathfrak{a}}=\mathbf{M}^{\mathfrak{a}}\cdot \mathbf{x}, \quad \mathbf{x}=\mathbf{M}^{{\mathfrak{a}}-1}{\boldsymbol{\xi}}^{\mathfrak{a}},$$ where $\mathbf{M}^{\mathfrak{a}}$ is the $3\times 3$ matrix whose rows are basis vectors $\mathbf{s}^{\mathfrak{a}}$, $\mathbf{p}^{\mathfrak{a}}$, and $\mathbf{m}^{\mathfrak{a}}$. Thus, any function of $\mathbf{x}$ can be expressed as function of ${\boldsymbol{\xi}}^{\mathfrak{a}}$ and [*vice versa*]{}. For the plastic slip $\beta ^\mathfrak{a}$ caused by dislocations of the slip system ${\mathfrak{a}}$ we assume that their lines lie completely in the slip planes parallel to the $(\xi^{\mathfrak{a}}_1,\xi^{\mathfrak{a}}_2)$-plane. To describe the latter we introduce the loop function $l^{\mathfrak{a}}(\xi^{\mathfrak{a}}_1,\xi^{\mathfrak{a}}_2, \xi^{\mathfrak{a}}_3)$ such that its level curves $$\label{eq:2.18}
l^{\mathfrak{a}}(\xi ^{\mathfrak{a}}_1,\xi ^{\mathfrak{a}}_2,c _3)=c,$$ where $c_3$ and $c$ are constants, coincide with the dislocation lines. We denote by ${\boldsymbol{\nu}}^{\mathfrak{a}}$ and ${\boldsymbol{\tau}}^{\mathfrak{a}}$ the plane unit vectors normal and tangential to the dislocation line. From equation follow $${\boldsymbol{\nu}}^{\mathfrak{a}}=\frac{1}{\sqrt{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2}}(l^{\mathfrak{a}}_{;1}\mathbf{s}^{\mathfrak{a}}+l^{\mathfrak{a}}_{;2}\mathbf{p}^{\mathfrak{a}}), \quad {\boldsymbol{\tau}}^{\mathfrak{a}}=\frac{1}{\sqrt{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2}}(-l^{\mathfrak{a}}_{;2}\mathbf{s}^{\mathfrak{a}}+l^{\mathfrak{a}}_{;1}\mathbf{p}^{\mathfrak{a}}),$$ where the semicolon in indices denotes the partial derivatives of the loop function with respect to $\xi ^{\mathfrak{a}}_1$, $\xi ^{\mathfrak{a}}_2$, so these vectors lie in the slip planes parallel to the $(\xi _1,\xi _2)$-plane as expected.
For the plastic deformation the dislocation density tensor becomes $$\label{eq:2.20}
\mathbf{T}=-\mathbf{F}^p\times \nabla =\sum_{\mathfrak{a}=1}^n \mathbf{s}^{\mathfrak{a}}\otimes (\nabla \beta ^{\mathfrak{a}}\times \mathbf{m}^{\mathfrak{a}}).$$ To characterize the geometrically necessary dislocations belonging to one slip system we consider one term $\mathbf{T}^{\mathfrak{a}}=\mathbf{s}^{\mathfrak{a}}\otimes (\nabla \beta ^{\mathfrak{a}}\times \mathbf{m}^{\mathfrak{a}})$ in the sum . Let us choose an infinitesimal area $da$ with the unit normal vector ${\boldsymbol{\tau}}^{\mathfrak{a}}$ and compute the resultant Burgers vector of all excess dislocations of the system ${\mathfrak{a}}$, whose dislocation lines cross this area at right angle $$\mathbf{b}^{\mathfrak{a}}_r=\mathbf{T}^{\mathfrak{a}}\cdot {\boldsymbol{\tau}}^{\mathfrak{a}}\, da=-\mathbf{s}^{\mathfrak{a}}\, (\nabla \beta ^{\mathfrak{a}}\cdot {\boldsymbol{\nu}}^{\mathfrak{a}})da=-\mathbf{s}^{\mathfrak{a}}\, \partial _{\nu ^{\mathfrak{a}}}\beta ^{\mathfrak{a}}\, da.$$ This resultant Burgers vector can be decomposed into the sum of two vectors $$\mathbf{b}^{\mathfrak{a}}_r=\mathbf{b}^{\mathfrak{a}}_{r\perp}+\mathbf{b}^{\mathfrak{a}}_{r\parallel}=-({\boldsymbol{\nu}}^{\mathfrak{a}}s^{\mathfrak{a}}_\nu +{\boldsymbol{\tau}}^{\mathfrak{a}}s^{\mathfrak{a}}_\tau )\partial _{\nu ^{\mathfrak{a}}}\beta^{\mathfrak{a}}\, da,$$ where $s^{\mathfrak{a}}_\nu =\mathbf{s}^{\mathfrak{a}}\cdot {\boldsymbol{\nu}}^{\mathfrak{a}}$ and $s^{\mathfrak{a}}_\tau =\mathbf{s}^{\mathfrak{a}}\cdot {\boldsymbol{\tau}}^{\mathfrak{a}}$ are the projections of the slip vector onto the normal and tangential direction to the dislocation line, respectively. This allows us to define two scalar densities of edge and screw dislocations of the corresponding slip system $$\label{eq:2.21}
\begin{split}
\rho^{\mathfrak{a}}_\perp =\frac{|\mathbf{b}^{\mathfrak{a}}_{r\perp}|}{b} =\frac{1}{b}|s^{\mathfrak{a}}_\nu \partial _{\nu ^{\mathfrak{a}}}\beta^{\mathfrak{a}}| =\frac{1}{b}\left| \frac{l^{\mathfrak{a}}_{;1}(\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2})}{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2}\right| ,
\\
\rho^{\mathfrak{a}}_\parallel =\frac{|\mathbf{b}^{\mathfrak{a}}_{r\parallel }|}{b} =\frac{1}{b}|s^{\mathfrak{a}}_\tau \partial _{\nu ^{\mathfrak{a}}}\beta^{\mathfrak{a}}|=\frac{1}{b}\left| \frac{l^{\mathfrak{a}}_{;2}(\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2})}{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2}\right| ,
\end{split}$$ We see that the dislocation densities $\rho^{\mathfrak{a}}_\perp $ and $\rho^{\mathfrak{a}}_\parallel $ depend only on the partial derivatives $\beta ^{\mathfrak{a}}_{;\alpha }$ and $l^{\mathfrak{a}}_{;\alpha }$, $\alpha =1,2$. For the moving dislocations we allow the loop functions to depend on time $t$ such that the level curves $$l^{\mathfrak{a}}(\xi ^{\mathfrak{a}}_1,\xi ^{\mathfrak{a}}_2,c _3,t)=c,$$ with $c_3$ and $c$ being constants, coincide with the dislocation lines during their motion. Similar to the single slip we introduce the normal velocities of dislocation lines as follows $$v^{\mathfrak{a}}_\nu =\mathbf{v}^{\mathfrak{a}}\cdot {\boldsymbol{\nu}}^{\mathfrak{a}}=-\frac{\dot{l}^{\mathfrak{a}}}{\sqrt{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2}},$$ with $\dot{l}=l_{,t}$. These kinematic quantities will be used in the model with dissipation.
From the above discussion of kinematics we see that a single crystal with $n$ active slip systems is a generalized continuum with $3+2n$ degrees of freedom at each point: $\mathbf{y}(\mathbf{x})$, $\beta ^{\mathfrak{a}}(\mathbf{x})$, and $l^{\mathfrak{a}}({\boldsymbol{\xi}}^{\mathfrak{a}}(\mathbf{x}))$, ${\mathfrak{a}}=1,\ldots ,n$. We require that the free energy per unit volume of crystal (assumed as macroscopically homogeneous) must be a function of $\mathbf{C}^e=\mathbf{F}^{eT}\cdot \mathbf{F}^e$ (where $\mathbf{F}^e=\mathbf{F}\cdot \mathbf{F}^{p-1}$), $\rho^{\mathfrak{a}}_\perp $, and $\rho^{\mathfrak{a}}_\parallel $ $$\psi = \psi (\mathbf{C}^e,\rho^{\mathfrak{a}}_\perp ,\rho^{\mathfrak{a}}_\parallel ).$$ Under the same loading condition as for the crystal with single slip we write down the energy functional $$\label{eq:2.22}
I[\mathbf{y}(\mathbf{x}),\beta ^{\mathfrak{a}}(\mathbf{x}),l^{\mathfrak{a}}({\boldsymbol{\xi}}^{\mathfrak{a}}(\mathbf{x}))]=\int_{\mathcal{V}}w(\mathbf{F},\beta ^{\mathfrak{a}},\nabla \beta ^{\mathfrak{a}},\nabla l^{\mathfrak{a}})\, dx-\int_{\partial _s} \mathbf{t} \cdot \mathbf{y}\, da,$$ where $$w(\mathbf{F},\beta ^{\mathfrak{a}},\nabla \beta ^{\mathfrak{a}},\nabla l^{\mathfrak{a}})=\psi (\mathbf{C}^e,\rho^{\mathfrak{a}}_\perp ,\rho^{\mathfrak{a}}_\parallel ).$$ Provided the resistance to the dislocation motion is negligibly small, we formulate the following variational principle for single crystals with $n$ active slip systems: the true placement vector $\check{\mathbf{y}}(\mathbf{x})$, the true plastic slips $\check{\beta }^{\mathfrak{a}}(\mathbf{x})$, and the true loop functions $\check{l}^{\mathfrak{a}}({\boldsymbol{\xi}}^{\mathfrak{a}}(\mathbf{x}))$ in the [*final*]{} equilibrium state of deformation minimize energy functional among all continuously differentiable fields $\mathbf{y}(\mathbf{x})$, $\beta ^{\mathfrak{a}}(\mathbf{x})$, and $l^{\mathfrak{a}}({\boldsymbol{\xi}}^{\mathfrak{a}}(\mathbf{x}))$ satisfying the constraints $$\label{eq:2.23}
\mathbf{y}(\mathbf{x})=\mathbf{x}+\mathbf{u}_0(\mathbf{x}),\quad \beta ^{\mathfrak{a}}(\mathbf{x})=0, \quad l^{\mathfrak{a}}(\mathbf{x})=0 \quad \text{at $\partial _k$},$$
Applying the same calculus of variation and taking into account the arbitrariness of the variations of $\mathbf{y}(\mathbf{x})$, $\beta (\mathbf{x})$, and $l({\boldsymbol{\xi}}^{\mathfrak{a}}(\mathbf{x}))$ in $\mathcal{V}$ as well as at $\partial _s$, one can show that the minimizer must satisfy in $\mathcal{V}$ the equilibrium equations $$\label{eq:2.25}
\mathbf{P}\cdot \nabla =0,
\quad
-w_\beta ^{\mathfrak{a}}+\nabla \cdot w_{\nabla \beta ^{\mathfrak{a}}}=0, \quad \nabla \cdot w_{\nabla l^{\mathfrak{a}}} =0,$$ subjected to the kinematic boundary conditions at $\partial _k$ and the following natural boundary conditions $$\label{eq:2.26}
\mathbf{P}\cdot \mathbf{n}=\mathbf{t},
\quad
w_{\nabla \beta ^{\mathfrak{a}}}\cdot \mathbf{n}=0, \quad w_{\nabla l^{\mathfrak{a}}}\cdot \mathbf{n}=0 \quad \text{at $\partial _s$}.$$ The constitutive equations for $\mathbf{P}=w_{\mathbf{F}}$, $-w_{\beta ^{\mathfrak{a}}}$, $w_{\nabla \beta ^{\mathfrak{a}}}$, and $w_{\nabla l^{\mathfrak{a}}}$ can easily be obtained from the above free energy density by standard differentiation. For the first Piola-Kirchhoff stress tensor and the Schmid stresses we have $$\label{eq:2.27}
\mathbf{P}=w_{\mathbf{F}}=2\mathbf{F}^e\cdot \psi _{\mathbf{C}^e} \cdot \mathbf{F}^{p-T}.$$ $$\label{eq:2.28}
\tau ^{\mathfrak{a}}_r=-w_{\beta ^{\mathfrak{a}}}=2\mathbf{s}^{\mathfrak{a}}\cdot \mathbf{F}^{p-T}\cdot \mathbf{C}^e \cdot \psi _{\mathbf{C}^e}\cdot \mathbf{m}^{\mathfrak{a}}.$$ Likewise, from follows $$\label{eq:2.29}
w_{\nabla \beta ^{\mathfrak{a}}}=\frac{1}{b}[\psi _{\rho^{\mathfrak{a}}_\perp }{\text{sign}}(s^{\mathfrak{a}}_\nu \partial _{\nu ^{\mathfrak{a}}} \beta ^{\mathfrak{a}})s^{\mathfrak{a}}_\nu +\psi _{\rho^{\mathfrak{a}}_\parallel }{\text{sign}}(s^{\mathfrak{a}}_\tau \partial _{\nu ^{\mathfrak{a}}} \beta ^{\mathfrak{a}})s^{\mathfrak{a}}_\tau ] {\boldsymbol{\nu}}^{\mathfrak{a}}.$$ Thus, the vectors $w_{\nabla \beta ^{\mathfrak{a}}}$ are two-dimensional. Finally, for $w_{\nabla l^{\mathfrak{a}}}$ we have $$\label{eq:2.30}
w_{\nabla l^{\mathfrak{a}}}=\left( \psi _{\rho _\perp ^{\mathfrak{a}}}\frac{\partial \rho _\perp ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;1}}+\psi _{\rho _\parallel ^{\mathfrak{a}}}\frac{\partial \rho _\parallel ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;1}}\right)\mathbf{s}^{\mathfrak{a}}+\left( \psi _{\rho _\perp ^{\mathfrak{a}}}\frac{\partial \rho _\perp ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;2}}+\psi _{\rho _\parallel ^{\mathfrak{a}}}\frac{\partial \rho _\parallel ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;2}}\right)\mathbf{p}^{\mathfrak{a}}.$$ Differentiating formulas for the dislocation densities $\rho^{\mathfrak{a}}_\perp $ and $\rho^{\mathfrak{a}}_\parallel $ with respect to $l^{\mathfrak{a}}_{;1}$ and $l^{\mathfrak{a}}_{;2}$, we get $$\begin{aligned}
\frac{\partial \rho _\perp ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;1}}&=\frac{1}{b} {\text{sign}}(s^{\mathfrak{a}}_\nu \partial _{\nu ^{\mathfrak{a}}}\beta ^{\mathfrak{a}})\left( -\frac{2(l^{\mathfrak{a}}_{;1})^2(\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2})}{((l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2)^2}+\frac{2\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2}}{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2} \right) ,
\\
\frac{\partial \rho _\parallel ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;1}}&=\frac{1}{b}{\text{sign}}(s^{\mathfrak{a}}_\tau \partial _{\nu ^{\mathfrak{a}}}\beta ^{\mathfrak{a}})\left( -\frac{2l^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;2}(\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2})}{((l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2)^2}+\frac{\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;2}}{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2} \right) ,
\\
\frac{\partial \rho _\perp ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;2}}&=\frac{1}{b}{\text{sign}}(s^{\mathfrak{a}}_\nu \partial _{\nu ^{\mathfrak{a}}}\beta ^{\mathfrak{a}})\left( -\frac{2l^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;2}(\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2})}{((l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2)^2}+\frac{\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;1}}{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2} \right) ,
\\
\frac{\partial \rho _\parallel ^{\mathfrak{a}}}{\partial l^{\mathfrak{a}}_{;1}}&=\frac{1}{b} {\text{sign}}(s^{\mathfrak{a}}_\tau \partial _{\nu ^{\mathfrak{a}}}\beta ^{\mathfrak{a}})\left( -\frac{2(l^{\mathfrak{a}}_{;2})^2(\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2})}{((l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2)^2}+\frac{\beta ^{\mathfrak{a}}_{;1}l^{\mathfrak{a}}_{;1}+2\beta ^{\mathfrak{a}}_{;2}l^{\mathfrak{a}}_{;2}}{(l^{\mathfrak{a}}_{;1})^2+(l^{\mathfrak{a}}_{;2})^2} \right) .\end{aligned}$$ Substituting the constitutive equations - into - we get the completely new system of equations and boundary conditions which, together with , enables one to determine $\check{\mathbf{y}}(\mathbf{x})$, $\check{\beta }^{\mathfrak{a}}(\mathbf{x})$, and $\check{l}^{\mathfrak{a}}(\mathbf{x})$.
For the case of non-zero resistance to dislocation motion leading to the energy dissipation we take the dissipation function in the form $$D=D(\dot{\beta }^{\mathfrak{a}},\dot{l}^{\mathfrak{a}}),$$ We require that the true placement $\check{\mathbf{y}}(\mathbf{x},t)$, the true plastic slips $\check{\beta }^{\mathfrak{a}}(\mathbf{x},t)$, and the true loop functions $\check{l}^{\mathfrak{a}}(\mathbf{x},t)$ obey the variational equation $$\label{eq:2.32}
\delta I+\int_{\mathcal{V}} \sum_{{\mathfrak{a}}=1}^n (\frac{\partial D}{\partial \dot{\beta }^{\mathfrak{a}}}\delta \beta ^{\mathfrak{a}}+\frac{\partial D}{\partial \dot{l}^{\mathfrak{a}}}\delta l^{\mathfrak{a}})\, dx=0$$ for all variations of admissible fields $\mathbf{y}(\mathbf{x},t)$, $\beta ^{\mathfrak{a}}(\mathbf{x},t)$, and $l^{\mathfrak{a}}(\mathbf{x},t)$ satisfying the constraints . It is then easy to show by exactly the same arguments like those used at the end of the previous Section that equation yields $$\mathbf{P}\cdot \nabla =0,
\quad
-w_{\beta ^{\mathfrak{a}}}+\nabla \cdot w_{\nabla \beta ^{\mathfrak{a}}}=\frac{\partial D}{\partial \dot{\beta }^{\mathfrak{a}}}, \quad \nabla \cdot w_{\nabla l^{\mathfrak{a}}} =\frac{\partial D}{\partial \dot{l}^{\mathfrak{a}}},$$ which are subjected to the boundary conditions and . The constitutive equations remain exactly the same as -.
Small strain theory
===================
Let us simplify the above theory for small strains. In this case, instead of the placement $\mathbf{y}(\mathbf{x})$ we regards the displacement $\mathbf{u}(\mathbf{x})$ that is related to the former by $$\mathbf{u}(\mathbf{x})=\mathbf{y}(\mathbf{x})-\mathbf{x}$$ as the unknown function. Thus, the total compatible deformation is $$\mathbf{F}=\frac{\partial \mathbf{y}}{\partial \mathbf{x}} =\mathbf{I}+\mathbf{u}\nabla .$$ We assume that the displacement gradient $\mathbf{u}\nabla $ (called distortion) is small compared with $\mathbf{I}$. Concerning the plastic deformation given by we also assume that the plastic slips $\beta ^{\mathfrak{a}}$ are much smaller than 1. Using the the multiplicative resolution to express $\mathbf{F}^e$ through $\mathbf{F}$ and $\mathbf{F}^{p-1}$ and neglecting the small nonlinear terms in it, we get $$\mathbf{F}^e=\mathbf{F}\cdot \mathbf{F}^{p-1}=\mathbf{I}+{\boldsymbol{\beta}}^e$$ where $${\boldsymbol{\beta}}^e =\mathbf{u}\nabla -\sum_{{\mathfrak{a}}=1}^n \beta ^{\mathfrak{a}}\mathbf{s}^{\mathfrak{a}}\otimes \mathbf{m}^{\mathfrak{a}}.$$ The last equation can be interpreted as the additive resolution of the total distortion into the plastic and elastic parts. The total compatible strain tensor field can be obtained from the displacement field according to $${\boldsymbol{\varepsilon}}=\frac{1}{2}(\mathbf{u}\nabla +\nabla \mathbf{u}).$$ The incompatible plastic strain tensor field is the symmetric part of the plastic distortion field $${\boldsymbol{\varepsilon}}^p =\frac{1}{2}({\boldsymbol{\beta}}+{\boldsymbol{\beta}}^T)=\frac{1}{2}\sum_{{\mathfrak{a}}=1}^n \beta ^{\mathfrak{a}}(\mathbf{s}^{\mathfrak{a}}\otimes \mathbf{m}^{\mathfrak{a}}+\mathbf{m}^{\mathfrak{a}}\otimes \mathbf{s}^{\mathfrak{a}}).$$ Accordingly, the elastic strain tensor field is equal to $${\boldsymbol{\varepsilon}}^e={\boldsymbol{\varepsilon}}-{\boldsymbol{\varepsilon}}^p.$$
The dislocation densities remain exactly the same as in the finite strain theory. They are given by the formulas . Concerning the free energy density we will assume that it depends on the elastic strain ${\boldsymbol{\varepsilon}}^e$ and on the dislocation densities $\rho^{\mathfrak{a}}_\perp $ and $\rho^{\mathfrak{a}}_\parallel $ $$\psi = \psi ({\boldsymbol{\varepsilon}}^e,\rho^{\mathfrak{a}}_\perp ,\rho^{\mathfrak{a}}_\parallel ).$$ The energy of crystal containing dislocations reads $$\label{eq:3.5}
I[\mathbf{u}(\mathbf{x}),\beta (\mathbf{x}),l(\mathbf{x})]=\int_{\mathcal{V}}w(\mathbf{u}\nabla ,\beta ^{\mathfrak{a}},\nabla \beta ^{\mathfrak{a}},\nabla l^{\mathfrak{a}})\, dx-\int_{\partial _s} \mathbf{t}\cdot \mathbf{u}\, da,$$ where $w(\mathbf{u}\nabla ,\beta ^{\mathfrak{a}},\nabla \beta ^{\mathfrak{a}},\nabla l^{\mathfrak{a}})=\psi ({\boldsymbol{\varepsilon}}^e,\rho_\perp ,\rho_\parallel )$. Provided the resistance to the dislocation motion can be neglected, then the following variational principle is valid for single crystals: the true displacement field $\check{\mathbf{u}}(\mathbf{x})$, the true plastic slips $\check{\beta }^{\mathfrak{a}}(\mathbf{x})$, and the true loop functions $\check{l}^{\mathfrak{a}}(\mathbf{x})$ in the [*final*]{} state of deformation in equilibrium minimize energy functional among all admissible fields satisfying the constraints $$\label{eq:3.6}
\mathbf{u}(\mathbf{x})=\mathbf{u}_0(\mathbf{x}),\quad \beta ^{\mathfrak{a}}(\mathbf{x})=0,\quad l^{\mathfrak{a}}(\mathbf{x})=0 \quad \text{at $\partial _k$}.$$
The standard calculus of variation similar to the previous case leads to the equilibrium equations $$\label{eq:3.7}
{\boldsymbol{\sigma}}\cdot \nabla =0, \quad w_\beta ^{\mathfrak{a}}-\nabla \cdot w_{\nabla \beta ^{\mathfrak{a}}}=0, \quad \nabla \cdot w_{\nabla l^{\mathfrak{a}}} =0,$$ subjected to the kinematic boundary conditions at $\partial _k$ and the following natural boundary conditions $$\label{eq:3.8}
{\boldsymbol{\sigma}}\cdot \mathbf{n}=\mathbf{t},\quad w_{\nabla \beta ^{\mathfrak{a}}}\cdot \mathbf{n}=0, \quad w_{\nabla l^{\mathfrak{a}}}\cdot \mathbf{n}=0 \quad \text{at $\partial _s$}.$$ The constitutive equation for the Cauchy stress tensor becomes $$\label{eq:3.9}
{\boldsymbol{\sigma}}=\frac{\partial \psi }{\partial {\boldsymbol{\varepsilon}}^e}.$$ For the Schmid stress we obtain $$\label{eq:3.10}
\tau ^{\mathfrak{a}}_r=-w_{\beta ^{\mathfrak{a}}}=\mathbf{s}^{\mathfrak{a}}\cdot {\boldsymbol{\sigma}}\cdot \mathbf{m}^{\mathfrak{a}}.$$ The constitutive equations for $w_{\nabla \beta ^{\mathfrak{a}}}$ and $w_{\nabla l^{\mathfrak{a}}}$ remain unchanged as compared with and . The new set of governing equations and boundary conditions are obtained by substituting , , , and into the equilibrium equations and boundary conditions . Note that even for small strain the system of governing equations remain as a whole nonlinear.
It is a simple matter to modify the theory for the case of non-zero resistance to dislocation motion and plastic slip leading to the energy dissipation.
Simple shear deformation of a single crystal beam {#sec:pldeformation}
=================================================
![Simple shear deformation of a beam of rectangular cross section[]{data-label="fig:plane"}](planestrain){width="8cm"}
Let us consider now the simple shear deformation of a single crystal beam having only one active slip system. The crystal occupies in its initial configuration a long cylinder of an arbitrary cross section such that $(x_1,x_2)\in \mathcal{A}$ and $0\le x_3\le L$ (see Fig. \[fig:plane\] for the beam of rectangular cross section). As before, the slip system is chosen such that the vectors $\mathbf{s}$, $\mathbf{p}$, and $\mathbf{m}$ coincide with $\mathbf{e}_1$, $\mathbf{e}_2$, and $\mathbf{e}_3$, respectively. We realize the simple shear deformation by placing this crystal beam in a “hard” device with the prescribed displacements at the boundary of the crystal such that $$\label{eq:5.1}
y_1=x_1+\gamma x_3,\quad y_2=x_2,\quad y_3=x_3.$$ We assume that the length of the crystal $L$ is large enough compared with the sizes of the cross section to guarantee the uniform simple shear deformation state for which equation is valid everywhere in the beam. If the overall shear $\gamma $ is sufficiently small, then it is natural to expect that the crystal deforms elastically and the plastic slip as well as the loop function vanish. If this parameter exceeds some critical threshold, then dislocation loops may appear (see one dislocation loop in Fig. \[fig:plane\]). Due to the almost translational invariance in $x_3$-direction we may assume that $\beta (\mathbf{x})$ and $l(\mathbf{x})$ depend only on $x_1$ and $x_2$ (except perhaps the neighborhoods of $x_3=0$ and $x_3=L$).
For simplicity let us consider the small strain theory. Then the only non-zero components of the total strain tensor, under the condition that is valid everywhere, are $$\varepsilon _{13}=\varepsilon _{31}=\frac{1}{2}\gamma .$$ Since the plastic distortion tensor ${\boldsymbol{\beta}}$ has only one non-zero component $\beta _{13}=\beta (x_1,x_2)$, the non-zero components of the elastic strain tensor read $$\label{eq:5.2}
\varepsilon ^e_{13}=\varepsilon ^e_{31}=\frac{1}{2}(\gamma -\beta (x_1,x_2)).$$ With the loop function being $l(x_1,x_2)$ the dislocation densities are given by $$\label{eq:5.3}
\begin{split}
\rho_\perp =\frac{1}{b}\left| \frac{l_{,1}(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{l_{,1}^2+l_{,2}^2}\right| ,
\\
\rho_\parallel =\frac{1}{b}\left| \frac{l_{,2}(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})}{l_{,1}^2+l_{,2}^2}\right| .
\end{split}$$
For the small strain theory we propose the free energy per unit volume of the undeformed crystal in the most simple form $$\label{eq:5.4}
\psi ({\boldsymbol{\varepsilon}}^e,\rho _\perp, \rho _\parallel ) = \frac{1}{2}\lambda ({\text{tr}}{\boldsymbol{\varepsilon}}^e)^2 + \mu {\text{tr}}({\boldsymbol{\varepsilon}}^e \cdot {\boldsymbol{\varepsilon}}^e) + \frac{1}{2}\mu k_1 \frac{\rho_\perp ^2}{\rho _s^2}+ \frac{1}{2}\mu k_2 \frac{\rho_\parallel ^2}{\rho _s^2} .$$ Here ${\boldsymbol{\varepsilon}}^e$ is the elastic strain tensor, $\lambda $ and $\mu $ are the Lamé constants, $k_1$ and $k_2$ are material constants, while $\rho _s$ can be interpreted as the saturated dislocation density. The first two terms in represent the free energy of the crystal due to the macroscopic elastic strain, where we assume that the crystal is elastically isotropic. The last two terms in correspond to the energy of the dislocation network for moderate dislocation densities [@Gurtin2002; @Gurtin2007]. Note that, for the small or extremely large dislocation densities close to the saturated value, the logarithmic energy proposed by @Berdichevsky06 [@Berdichevsky06a] turns out to be more appropriate. Substituting formulas and into we obtain $$\begin{gathered}
w(\beta ,\nabla \beta ,\nabla l)=\frac{1}{2}\mu (\gamma -\beta )^2+\frac{1}{2}\frac{\mu }{b^2\rho _s^2}\left[ k_1\frac{l_{,1}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})^2}{(l_{,1}^2+l_{,2}^2)^2} \right.
\\
\left. +k_2 \frac{l_{,2}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})^2}{(l_{,1}^2+l_{,2}^2)^2} \right] .\end{gathered}$$ Since in this case the side boundary does not allow dislocations to reach it, we can pose on both functions $\beta (x_1,x_2)$ and $l(x_1,x_2)$ the Dirichlet boundary conditions $$\label{eq:5.6}
\beta (x_1,x_2)=0,\quad l(x_1,x_2)=0 \quad \text{for $(x_1,x_2)\in \partial \mathcal{A}$},$$ The variational problem reduces to minimizing the two-dimensional functional $$\label{eq:5.7}
I[\beta (x_1,x_2),l(x_1,x_2)]=L\int_{\mathcal{A}} w(\beta ,\nabla \beta ,\nabla l)dx_1dx_2$$ among all admissible functions $\beta (x_1,x_2)$ and $l(x_1,x_2)$ satisfying the boundary conditions .
It is convenient to simplify the functional and minimize it in the dimensionless form. Introducing the dimensionless variables and quantity $$\bar{x}_1=b\rho _sx_1, \quad \bar{x}_2=b\rho _sx_2,\quad (\bar{x}_1,\bar{x}_2)\in \bar{\mathcal{A}},\quad \bar{I}=\frac{Ib^2\rho _s^2}{\mu L},$$ we can write functional in the form $$\label{eq:5.8}
I=\frac{1}{2} \int_{\mathcal{A}}\left[ (\gamma -\beta )^2+ k_1 \frac{l_{,1}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})^2}{(l_{,1}^2+l_{,2}^2)^2} + k_2 \frac{l_{,2}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})^2}{(l_{,1}^2+l_{,2}^2)^2} \right] dx_1dx_2,$$ where the bar over the quantities are dropped for short. The problem is to minimize functional among all admissible functions $\beta (x_1,x_2)$ and $l(x_1,x_2)$ satisfying Dirichlet boundary conditions . Since the integrand of is positive definite, the existence of the minimizer in this variational problem is guaranteed.
In one special case the problem degenerates and admits an analytical solution. Indeed, if we choose $k_1=k_2=k$ and take $\mathcal{A}$ to be a circular cross section whose boundary is given in the polar coordinates by $r=R$, then due to the symmetry we may assume that both $\beta $ and $l$ are functions of $r=\sqrt{x_1^2+x_2^2}$ only. It is now a simple matter to show that $$\frac{l_{,1}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})^2}{(l_{,1}^2+l_{,2}^2)^2} + \frac{l_{,2}^2(\beta _{,1}l_{,1}+\beta _{,2}l_{,2})^2}{(l_{,1}^2+l_{,2}^2)^2}=\beta _{,r}^2.$$ Thus, functional (normalized by $2\pi $) does not depend on $l(r)$ and takes the form $$I=\int_0^R [ \frac{1}{2} (\gamma -\beta )^2+\frac{1}{2} k \beta _{,r}^2 ] r dr,$$ which leads to the Euler equation $$(\gamma -\beta )+\frac{k}{r}(\beta _{,r}r)_{,r}=0.$$ This is nothing else but the inhomogeneous modified Bessel equation that yields the following solution (regular at $r=0$) $$\beta (r)=\gamma +C I_0(r/\sqrt{k}),$$ with $I_0(x)$ being the modified Bessel function of the first kind. The coefficient $C$ must be found from the boundary condition $\beta (R)=0$ giving $$C=-\gamma /I_0(R/\sqrt{k}).$$ Thus, $$\label{eq:5.9}
\beta (r)=\gamma [1-I_0(r/\sqrt{k})/I_0(R/\sqrt{k})].$$ The plot of solution for different values of $\gamma $ is shown in Fig. \[fig:sbeta\], where we choose the radius such that $\bar{R}=b\rho _sR=1$ and the parameter $k=10^{-4}$. As $\gamma $ increases, the plastic slip increases too. It is seen from this Figure that the plastic slip is nearly constant in the middle of the cross section and changes strongly only in the thin layer in form of ring near the boundary. Since $l$ is a function of $r$, dislocations in form of circles pile up against the boundary of the cross section, leaving the middle of the cross section almost dislocation-free. Since the dislocation loops are circles, they have the purely edge character at $\varphi =0$ and $\varphi =\pi $ and the purely screw character at $\varphi =\pm \pi /2$. For all other angles the dislocation loops have the mixed character. Note that if $k_1\ne k_2$, the strictly axi-symmetric solution is no longer valid because the contributions of the edge and screw components to the energy of dislocation network are not equal. Note also that, if the logarithmic energy is used instead of the quadratic energy, $\beta $ becomes non-zero only if $\gamma >\gamma _c$, so there is a threshold stress for the dislocation nucleation (see [@Berdichevsky-Le07]). Besides, the existence of the dislocation-free zone in the middle of the cross-section can be established.
![The plastic slip $\beta (r)$: i) $\gamma =0.001$, ii) $\gamma =0.005$, iii) $\gamma =0.01$.[]{data-label="fig:sbeta"}](Sbeta){width="8cm"}
For an arbitrary cross section and for $k_1\ne k_2$ the problem does not admit exact analytical solution. However, based on the character of solution that changes strongly only in the normal direction to the boundary observed in the previous case, we may use the asymptotic method to find the solution in the thin boundary layer. Take for example the rectangular cross-section $(x_1,x_2)\in (0,W)\times (0,H)$, with $W$ and $H$ being its width and height. In this case let us assume that there are two boundary layers near the left and right boundaries $x_1=0$ and $x_1=W$, and two other boundary layers near the bottom and top boundaries $x_2=0$ and $x_2=H$. Near the boundaries parallel to the $x_2$ axis the derivative with respect to $x_2$ can be neglect as compared to the derivative with respect to $x_1$, while the opposite is true near the boundaries parallel to $x_1$-axis. In the middle of the cross section the plastic slip remains constant. Functional reduces to the sum of four integrals, and all of them do not depend on $l$. Consider for instance the integral along the left boundary layer $$I_1[\beta (x_1)]=\int_0^\lambda [ \frac{1}{2} (\gamma -\beta )^2+\frac{1}{2}k_1\beta _{,1}^2]dx_1,$$ with $\lambda $ being still an unknown length. We omit here the integration over $x_2$ (because in this boundary layer it plays just the role of a parameter) and try to find the minimum among $\beta $. The standard variational calculus leads to the Euler equation $$\label{eq:5.11}
\gamma -\beta +k_1\beta _{,11}=0$$ which must be subjected to the boundary conditions $$\label{eq:5.12}
\beta (0)=0.$$ The solution of and that does not grow exponentially as $x_1\to \infty $ reads $$\beta (x_1)=\gamma (1-e^{-x_1/\sqrt{k_1}}).$$ Due to the mirror symmetry of the problem, the plastic slip in the boundary layer near $x_1=W$ must be $$\beta (x)=\gamma (1-e^{-(W-x_1)/\sqrt{k_1}}).$$ Similarly, the solution in the boundary layers parallel to the $x_1$ axis equals $$\label{eq:5.14}
\beta (x_2)=\begin{cases}
\gamma (1-e^{-x_2/\sqrt{k_2}}) & \text{near $x_2=0$}, \\
\gamma (1-e^{-(H-x_2)/\sqrt{k_2}}) & \text{near $x_2=H$}.
\end{cases}$$ Thus, the width of the boundary layers parallel to the $x_1$ axis must be of the order $\sqrt{k_2}$, while that of the boundary layers parallel to the $x_2$-axis must be of the order $\sqrt{k_1}$. Since $l$ depends only on the normal coordinate to the boundary, the dislocation lines must be parallel to the boundary of the cross section except at four corners where the asymptotic solution becomes no longer valid (the numerical solution should lead to a smoothing of the corners of dislocation loops). We see that near the vertical boundaries the dislocation loops have the edge character, while near the horizontal boundaries they have the screw character. This agrees well with the widths of the boundary layers determined by the corresponding contributions of the edge and screw components to the energy of the dislocation network.
For very thin rectangular cross section with $H\ll W$ we may neglect the influence of the edges near $x_1= 0$ and $x_1=W$ by considering the dislocation network in the central part of the beam. In this case we will have only screw dislocations which pile up against two obstacle at $x_2=0$ and $x_2=H$. The solution reduces to that found in [@Berdichevsky-Le07].
Conclusion
==========
In this paper we have developed the nonlinear CDT for crystals containing curved dislocations based on the LEDS-hypothesis. The completely new set of equilibrium equations, boundary conditions and constitutive equations have been derived from the principle of minimum free energy. As the outcome, we have obtained the system of strongly nonlinear partial differential equations for the placement, the plastic slip, and the loop function. In the case of non-vanishing resistance to dislocation motion we have derived the governing equations from the variational equation that takes the dissipation into account. We have extended the theory to the case of multiple slip and simplified it for small strains. The application of the theory has been illustrated on the problem of single crystal beam having one slip system and deforming in simple shear. Under the simplified assumption $k_1=k_2$ the analytical solution of this problem has been found for the circular cross section. For arbitrary cross sections the problem has been solved by the asymptotic method. We have shown that the asymptotic solution found for the rectangular cross section reduces to the well-known solution in [@Berdichevsky-Le07] if it is thin and long.
[*Acknowledgments*]{}
The financial support by the German Science Foundation (DFG) through the research projects LE 1216/4-2 and GP01-G within the Collaborative Research Center 692 (SFB692) is gratefully acknowledged.
[42]{} Acharya, A., 2001. A model of crystal plasticity based on the theory of continuously distributed dislocations. J. Mech. Phys. Solids 49, 761-784. Acharya, A., Bassani, J.L., 2000. Lattice incompatibility and a gradient theory of crystal plasticity. J. Mech. Phys. Solids 48, 1565-1595. Berdichevsky, V.L., 2006a. Continuum theory of dislocations revisited. Continuum Mech. Therm. 18, 195-222. Berdichevsky, V.L., 2006b. On thermodynamics of crystal plasticity. Scripta Materialia 54, 711-716. Berdichevsky, V.L., Le, K.C., 2007. Dislocation nucleation and work hardening in anti-planed constrained shear. Continuum Mech. Therm. 18, 455-467. Berdichevsky, V.L., Sedov, L.I., 1967. Dynamic theory of continuously distributed dislocations. Its relation to plasticity theory. Appl. Math. Mech. (PMM) 31, 989-1006. Bilby, B.A., Gardner, L.R.T., Stroh, A.N., 1957. Continuous distributions of dislocations and the theory of plasticity, in: Extrait des actes du $IX^e$ congrès international de mécanique appliquée, pp. 35-44. Engels, P., Ma, A., Hartmaier, A., 2012. Continuum simulation of the evolution of dislocation densities during nanoindentation. Int. J. Plasticity 38, 159-169. Gurtin, M.E., 1981. An Introduction to Continuum Mechanics. Academic Press, New York. Gurtin, M. E., 2002. A gradient theory of single-crystal viscoplasticity that accounts for geometrically necessary dislocations. J. Mech. Phys. Solids 50, 5-32. Gurtin, M.E., Anand, L., Lele, S.P., 2007. Gradient single-crystal plasticity with free energy dependent on dislocation densities. J. Mech. Phys. Solids 55, 1853-1878. Hansen, N., Kuhlmann-Wilsdorf, D., 1986. Low energy dislocation structures due to unidirectional deformation at low temperatures. Mater. Sci. Eng. 81, 141-161. Hochrainer, T., Sandfeld, S., Zaiser, M., Gumbsch, P., 2014. Continuum dislocation dynamics: towards a physical theory of crystal plasticity. J. Mech. Phys. Solids 63, 167-178. Hochrainer, T., Zaiser, M., Gumbsch, P., 2007. A three-dimensional continuum theory of dislocation systems: kinematics and mean-field formulation. Phil. Mag. 87, 1261-1282. Hughes, D.A., Hansen, N., 1997. High angle boundaries formed by grain subdivision mechanisms. Acta Mater. 45, 3871-3886. Kaluza, M., Le, K.C., 2011. On torsion of a single crystal rod. Int. J. Plasticity 27, 460-469. Kochmann, D.M., Le, K.C., 2008a. Dislocation pile-ups in bicrystals within continuum dislocation theory. Int. J. Plasticity 24, 2125-2147. Kochmann, D.M., Le, K.C., 2008b. Plastic deformation of bicrystals within continuum dislocation theory. Math. Mech. Solids 14, 540-563. Kochmann, D.M., Le, K.C., 2009. A continuum model for initiation and evolution of deformation twinning. J. Mech. Phys. Solids 57, 987-1002. Koster, M., Le, K.C., Nguyen, B.D., 2015. Formation of grain boundaries in ductile single crystals at finite plastic deformations. Int. J. Plasticity 69, 134-151. Kr[ö]{}ner, E., 1992. Mikrostrukturmechanik. GAMM-Mitteilungen 15, 104-119. Kuhlmann-Wilsdorf, D., 1989. Theory of plastic deformation: - properties of low energy dislocation structures. Mater. Sci. Eng. A 113, 1-41. Kuhlmann-Wilsdorf, D., 2001. Q: Dislocations structures - how far from equilibrium? A: Very close indeed. Mater. Sci. Eng. A 315, 211-216. Kuhlmann-Wilsdorf, D., Hansen, N., 1991. Geometrically necessary, incidental and subgrain boundaries. Scripta Metall. Mater. 25, 1557-1562. Laird, C., Charsley, P., Mughrabi, H., 1986. Low energy dislocation structures produced by cyclic deformation. Mater. Sci. Eng. 81, 433-450. Le, K.C., 2010. Introduction to Micromechanics. Nova Science, New York. Le, K.C., G[ü]{}nther, C., 2014. Nonlinear continuum dislocation theory revisited. Int. J. Plasticity 53, 164-178. Le, K.C., Nguyen, B.D., 2012. Polygonization: Theory and comparison with experiments. Int. J. Eng. Sci. 59, 211-218. Le, K.C., Nguyen, B.D., 2013. On bending of single crystal beam with continuously distributed dislocations. Int. J. Plasticity 48, 152-167. Le, K.C., Sembiring, P., 2008a. Analytical solution of plane constrained shear problem for single crystals within continuum dislocation theory. Arch. Appl. Mech. 78, 587–597. Le, K.C., Sembiring, P., 2008b. Plane-constrained shear of a single crystal strip with two active slip-systems. J. Mech. Phys. Solids 56, 2541-2554. Le, K.C., Sembiring, P., 2009. Plane constrained uniaxial extension of a single crystal strip. Int. J. Plasticity 25, 1950-1969. Mayeur, J.R., McDowell, D.L., 2014. A comparison of Gurtin type and micropolar theories of generalized single crystal plasticity. Int. J. Plasticity 57, 29-51. ztop, M.S., Niordson, C.F., Kysar, J.W., 2013. Length-scale effect due to periodic variation of geometrically necessary dislocation densities. Int. J. Plasticity 41, 189-201. Ortiz, M., Repetto, E.A., 1999. Nonconvex energy minimization and dislocation structures in ductile single crystals. J. Mech. Phys. Solids 47, 397-462. Ortiz, M., Repetto, E.A., Stainier, L., 2000. A theory of subgrain dislocation structures. J. Mech. Phys. Solids 48, 2077-2114. Sandfeld, S., Hochrainer, T., Zaiser, M., Gumbsch, P., 2011. Continuum modeling of dislocation plasticity: Theory, numerical implementation, and validation by discrete dislocation simulations. Journal of Materials Research 26, 623-632. Sandfeld, S., Thawinan, E., Wieners, C., 2015. A link between microstructure evolution and macroscopic response in elasto-plasticity: formulation and numerical approximation of the higher-dimensional continuum dislocation dynamics theory. Int. J. Plasticity 72, 1-20. Sedov, L.I., 1965. Mathematical methods of constructing models of continuum media. Usp. Matem. Nauk. 20, 123-182. Wulfinghoff, S., B[ö]{}hlke, T., 2015. Gradient crystal plasticity including dislocation-based work-hardening and dislocation transport. Int. J. Plasticity 69, 152-169. Zhu, X., Dai, S., Xiang, Y., 2013. Numerical simulation of dynamics of dislocation arrays and long-range stress fields of nonplanar dislocation arrays. Int. J. Plasticity 43, 85-100. Zhu, X., Xiang, Y., 2014. Continuum framework for dislocation structure, energy and dynamics of dislocation arrays and low angle grain boundaries. J. Mech. Phys. Solids 69, 175-194.
[^1]: phone: +49 234 32-26033, email: [email protected]. The paper is dedicated to the 70th birthday of my teacher V. Berdichevsky.
[^2]: In conventional crystal plasticity the kinematic equation for $\mathbf{F}^p$ is usually formulated in rate form that does not always reduces to [@Ortiz99; @Ortiz00].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This Resource Letter is designed to guide students, educators, and researchers through (some of) the literature on black holes. We discuss both the physics and astrophysics of black holes. We emphasize breadth over depth, and review articles over primary sources. We include resources ranging from nontechnical discussions appropriate for broad audiences to technical reviews of current research. Topics addressed include classification of stationary solutions, perturbations and stability of black holes, numerical simulations, collisions, the production of gravity waves, black-hole thermodynamics and Hawking radiation, quantum treatments of black holes, black holes in both higher and lower dimensions, and connections to nuclear and condensed-matter physics. On the astronomical end, we also cover the physics of gas accretion onto black holes, relativistic jets, gravitationally red-shifted emission lines, evidence for stellar-mass black holes in binary systems and supermassive black holes at the centers of galaxies, the quest for intermediate-mass black holes, the assembly and merging history of supermassive black holes through cosmic time, and their affects on the evolution of galaxies.'
author:
- Elena Gallo
- Donald Marolf
title: 'Resource Letter BH-2: Black Holes'
---
\#1\#2\#3\#4[[*\#1* ]{} [**\#2:**]{} \#3 (\#4)]{} \#1[$\langle$\#1]{}
Introduction
============
The status of black-hole physics has changed dramatically since the publication of the first black-hole Resource Letter, ref. below. The observational evidence for black holes is now plentiful and strong. Moreover, black holes have progressed from a curiosity of mathematical physics to a [*tool*]{} for building or studying models in other branches of physics. Examples range from feedback mechanisms that may control the formation and evolution of galaxies (see refs. \[dimatteo\], \[feedback\]) to string-theoretic connections to nuclear and condensed matter physics (see refs. \[Nature\], \[RHIC\], \[CM\]). Nevertheless, black holes remain fascinating objects in their own right that may yet divulge deep clues to further revolutions in fundamental physics.
We have endeavored to provide a guide to this extraordinarily diverse literature, and to include resources at a variety of levels. Some advanced resources will require additional specialized knowledge of general relativity, astrophysics, or quantum-field theory. We hope that bringing together many levels of material will allow students of black-hole physics to find the right entry point for their background and teachers of black-hole physics to extract useful material for their courses.
Given the breadth of our subject, it is impossible to be complete. We have by no means attempted to do so. Experts will find many of their favorite resources, and perhaps even their favorite topics to be missing entirely. We have concentrated on a “few” review articles and seminal papers, though we have tried to include sufficient commentary to help the user locate the most useful resource. Where possible, we have chosen review articles over original works and modern treatments over classic presentations. We made this decision for purely practical reasons, and we encourage the reader to consult Steve Detweiler’s original black-hole resource letter (ref. below) for an excellent guide to the historical literature. Indeed, we view our work as a supplement to his, not as a replacement.
1. [ “Resource Letter BH-1: Black Holes,”]{} S. Detweiler, Am. J. Phys. [**49**]{} 394-400 (1981).
Resources in the following sections are organized first by level: popular material appears in section \[ele\] while most U.S. undergraduate material appears in section \[ug\]. Graduate general-relativity textbooks are found in section \[gen\] while more specialized material is presented in sections \[ABH\]-\[last\]. However, some of the material in these latter sections is also of intermediate level, accessible to advanced undergraduate students. We have then attempted to sort the more specialized papers by topic. This task is far from straightforward, as many excellent reviews overlap several obvious categories. The result is that many resources are cross-referenced, typically by adding a note to “see also ref. nnn." The reader is well-advised to do so.
We have arranged the specialized material as follows: Section \[ABH\] contains resources dealing with black holes in astrophysics. This includes observations of black holes, models of astrophysical processes involving black holes, and the implications of these models.
Section \[SDT\] then addresses well-established mathematical physics describing the statics, dynamics, and thermodynamics of black holes. The focus here is on classical black-hole physics such as the treatment and classification of black holes in general relativity and related theories (in a variety of dimensions), their perturbations and stability properties, their formation and interaction through collisions, the production of gravitational radiation and the relation to gravity-wave detectors, and black-hole thermodynamics. In addition, however, section \[thermo\] includes broadly accepted material on quantum field theory in curved spacetime and the associated Hawking radiation. Such subjects are deeply connected to the thermodynamics of black holes and are addressed in much of the same literature. Discussions of the possible production of black holes in particle collisions has also been included in section \[dnot4\].
In contrast, those references that concern the deep quantum microphysics of black holes have been placed in section \[micro\]. Such resources are more speculative, at least in the sense that they typically make assumptions about the quantum nature of gravity that are beyond the reach of current experiments. We therefore separate this material from that of section \[thermo\], even though Hawking radiation is again a common theme. Finally, we have collected connections to nuclear and condensed-matter physics in the short section \[last\]. Such connections include the use of established fluid dynamics and condensed-matter physics to describe certain analogues of black holes, as well as current attempts to use black-hole physics (and string theory) to describe as yet poorly understood effects in both nuclear and condensed-matter physics. Here the emphasis is again on classical black-hole physics but in novel, and sometimes somewhat speculative contexts.
Basic Resources and Notation
----------------------------
Perhaps the most important information we can convey is how the reader might find and access useful works that will appear only [*after*]{} the publication of this Resource Letter. Web-based search engines continue to make this more and more possible, but it is worthwhile to point out a few places to start. When entering a new subject, it is natural to begin with a review article. Several journals are dedicated entirely to such reviews, and one may begin by searching their websites. Certain online tools are also of particular help in locating reviews. Some suggestions are:
1. [*Living Reviews in Relativity*]{}. This online journal publishes only review articles on gravitational physics. As you might guess, a large fraction of its contents deal with black holes, making it a prime resource. Indeed, many of the articles described below are from this journal.
2. [*Reviews of Modern Physics*]{}. A review journal published by the American Physical Society. Topics covered span all of physics.
3. [*Physics Reports.*]{} A review journal by Elsevier. Topics covered span all of physics.
4. [*HEP Reviews at SPIRES*]{}, [[www.slac.stanford.edu/spires/reviews/](www.slac.stanford.edu/spires/reviews/)]{}. An ongoing project to index online reviews in high-energy and theoretical physics. Black holes are currently category Id2. This website also explains how to perform a more general search of the SPIRES database for relevant review articles.
5. [*The SAO/NASA Astrophysics Data System Abstract Service (ADS)*]{}, [[adsabs.harvard.edu/abstract\_service.html/](adsabs.harvard.edu/abstract_service.html/)]{}. Hosted by the Smithsonian Astrophysical Observatory for NASA, this website is the most powerful browser for astrophysics-related material. It allows to search for, [*e.g.*]{}: specific authors (or list thereof), refereed vs. non-refereeed publications, works mentioning your favorite astronomical object, as well as title keywords, publication year, number of citations, etc. It covers a wide spectrum of levels, from scholarly research papers, to Ph.D. thesis, to conference proceedings. A full description of the search criteria is given at: [[doc.adsabs.harvard.edu/abs\_doc/help\_pages/basicsearch.html\#Basic\_Query\_Form](doc.adsabs.harvard.edu/abs_doc/help_pages/basicsearch.html#Basic_Query_Form)]{}. ADS has become an irreplaceable instrument in the workday of every astrophysicist.
6. [*Annual Reviews of Astronomy and Astrophysics*]{}, [[arjournals.annualreviews.org/loi/astro/](arjournals.annualreviews.org/loi/astro/)]{}. ARAA publishes comprehensive reviews written by well-known authors. They are an excellent starting point for the interested reader who wishes to approach a new astrophysical subject at the intermediate-advanced level.
The resources that we list below should not be hard to locate. Most appear in journals that can be found in a typical university library or online. In addition, most also include an “Arxiv identifier" either of the form \[arXiv:astro-ph/yymmnnn\] or \[arXiv:yymm.nnnn \[astro-ph\]\], with astro-ph perhaps replaced by qr-qc, hep-th, or hep-ph. These reference numbers refer to the online physics archive currently maintained by the Cornell University Library: [[arxiv.org/](arxiv.org/)]{}. Many papers can be found here by using the search feature. To find the precise desired work, simply search by “identifier" and enter it into the search box: [*e.g.*]{}, astro-ph/yymmnnn in the first (older style) numbering system or simply yymm.nnnn in the second system. Here yy denotes the year, mm the month, and nnn or nnnn the number assigned to the work when it was uploaded to arxiv.org.
As a final remark, the reader should be aware that two acronyms are frequently used in the text below: General Relativity is abbreviated “GR," and “AGN" refers to Active Galactic Nuclei. We have otherwise attempted to make the comments following each resource as approachable as possible by the general reader.
Popular Material {#ele}
================
Material in this section should be accessible to persons with a high school or college-level education without specialized training in physics. The goal of these works is typically to convey the general idea of the subject, without going into technical details. Certain resources are nevertheless somewhat demanding of the reader.
Popular Books
-------------
1. [**Black Holes and Time Warps: Einstein’s Outrageous Legacy**]{}, K. S. Thorne (W.W. Norton, New York, 1994). Although it is now 14 years old, this book remains the most popular one on this subject. First and foremost, it is an accessible introduction to basic issues: What is a black hole? How are they made? What happens when you fall in? The text begins by explaining the essential ideas in relativity and proceeds through the development of black-hole physics itself. The approach is historical, and the author’s personal involvement in this history cements a gripping story line. But it is the style of the text that captivates many readers. The author’s charming personality shines through. The final chapter on wormholes should be regarded as more speculative material, but the rest of the book focuses on well-established physics. (E)
2. [**Black Hole Physics: Basic Concepts and New Developments**]{}, V. P. Frolov and I. D. Novikov (Springer, New York, 1998). An encyclopedic, self-contained text on the physics of black holes, with a special focus on black holes in astrophysics (I).
3. [**Gravity’s Fatal Attraction: Black Holes in the Universe**]{}, \[BR\] M. Begelman and M. Rees (Scientific American Library, New York, 1995). Do “spacetime singularities” actually exist in nature? Have we found them? Begelman and Rees tell us the captivating story of the scientific endeavors –and related anecdotes– that led to the discovery of black holes. Powerful, inspiring, and fun to read. (E)
4. [**General Relativity from A to B**]{}, R. Geroch (U. Chicago Press, Chicago, 1978). Developed as lectures to nonscience majors at the University of Chicago, Geroch’s book gives a solid conceptual explanation of special and GR. Some math is used. This book requires more effort than those above, but delivers a deeper understanding of relativity. (E/I)
5. [**Was Einstein Right? Putting General Relativity to the Test**]{}, C. M. Will (BasicBooks, New York, 1993). Most popular GR books emphasize the theory and conceptual underpinnings of the theory. In contrast, Will’s book focusses on the many empirical tests of this theory, including Hulse and Taylor’s Nobel-prize winning verification of the decay of neutron star orbits via gravitational radiation. (E/I)
6. [**Relativity**]{}, A. Einstein, (Three Rivers Press, New York, 1961). Most later expositions of GR are based at least in part on this book, Einstein’s attempt to describe GR to anyone with a standard education. In his own words, “despite the shortness of the book, a fair amount of patience and force of will" is required of the reader. Nevertheless, the book is an invaluable opportunity to learn at least a bit of relativity from Einstein himself. (I)
Websites
--------
We highly recommend the websites below, all at the elementary level. Of course, websites are in constant flux. Check your favorite search engine.
1. \[wiki\] [[en.wikipedia.org/wiki/Black\_hole/](en.wikipedia.org/wiki/Black_hole/)]{} , Wikipedia. As usual, this Wikipedia entry is surprisingly broad. The reference list is excellent. (E)
2. \[hubblesite\] [[hubblesite.org/explore\_astronomy/black\_holes/](hubblesite.org/explore_astronomy/black_holes/)]{} , Hubble Site. Packed with graphics, simulated “experiments," and animations, this professional site from NASA and the Space Telescope Science Institute provides an exciting tour of black-hole physics. (E)
3. \[Jillian\] [[www.gothosenterprises.com/black\_holes/](www.gothosenterprises.com/black_holes/)]{} , J. Bornak. A charming, informal, and plain-speaking introduction to black holes and GR. Written by a black-hole fan as an undergraduate project. Nonprofessional in a delightful way and a marvelous introduction to the basic physics. (E)
4. \[space\] [[www.space.com/blackholes/](www.space.com/blackholes/)]{} . Lovely videos and simulations, as well as black-hole news items. (E)
5. \[RN\] [[antwrp.gsfc.nasa.gov/htmltest/rjn\_bht.html/](antwrp.gsfc.nasa.gov/htmltest/rjn_bht.html/)]{}, R. Nemiroff. The way black holes curve spacetime makes light do crazy things. What would you see if you approached a black hole? Check this classic website to find out. (E)
6. \[Hamilton\] [[casa.colorado.edu/\~ajsh/schw.shtml/](casa.colorado.edu/~ajsh/schw.shtml/)]{}, A. Hamilton. More modern simulations of trips to and inside black holes, but with a somewhat confusing portrayal of the horizon. (E)
7. \[stellar\] [[www-xray.ast.cam.ac.uk/xray\_introduction/Blackholebinary.html/](www-xray.ast.cam.ac.uk/xray_introduction/Blackholebinary.html/)]{} . Black holes and astrophysics from the Institute of Astronomy, Cambridge. (E)
8. library.thinkquest.org/25715/index.htm/ . Black-hole basics with great simulations. (E)
9. \[xrays\] [[www.astro.umd.edu/\~chris/Research/X-rays\_and\_Black\_holes/x-rays\_and\_black\_holes.html/](www.astro.umd.edu/~chris/Research/X-rays_and_Black_holes/x-rays_and_black_holes.html/)]{} , C. Reynolds, University of Maryland. A brief, informal and yet accurate description of the status of research on astrophysical black holes, explained for the nonexperts by a forefront leader in the field. (E)
10. [[www.youtube.com/watch?v=hoLvOvGW3Tk](www.youtube.com/watch?v=hoLvOvGW3Tk)]{} , YouTube. Simulations of black holes feeding and being born through collisions, all set to an exciting sound track. (E)
11. \[NoEsc\] [[amazing-space.stsci.edu/resources/explorations/blackholes/](amazing-space.stsci.edu/resources/explorations/blackholes/)]{} , Space Telescope Science Institute. This website provides black-hole resources for teachers and students. (E)
Popular Articles
----------------
The reader can find a number of interesting black-hole articles in popular-science publications such as [*Scientific American*]{}, [*Physics World*]{}, [*Physics Today*]{}, [*New Scientist*]{} and, at a slightly more advanced level, [*Science*]{} and [*Nature*]{}. Some fairly recent articles are listed below, but there is no doubt that more will appear in the near future. We have sorted these articles roughly into those that emphasize the astrophysics of black holes and those emphasizing fundamental aspects of theoretical physics.
[** Astrophysics:** ]{}
1. \[BHB\] [ “Black Hole Blowback,”]{} W. Tucker, H. Tananbaum, A. Fabian, [Scientific American]{}, [**296**]{}, 42-49 (Feb. 2007). Black holes and active galactic nuclei as possible drivers of feedback in galaxy cluster evolution. (E)
2. [“Unmasking Black Holes,”]{} J. Lasota, Scientific American, [**280**]{}, 30-37 (May 1999). On how we might directly detect black-hole horizons. (E)
3. [“The Galactic Odd Couple,”]{} K. Weaver, Scientific American, [**289**]{}, 34-41 (July 2003). Is there a connection between black holes and bursts of start formation? (E)
4. [“The Midlife Crisis of the Cosmos,”]{} A. J. Barger, Scientific American, [**292**]{} 32-39 (Jan. 2005). The authors describe the present and past of star and black-hole formation in our universe. (E)
5. [“The Brightest Explosions in the Universe,”]{} N. Gehrels, L. Piro and P. J. T. Leonard, Scientific American, [**287**]{} 84-91(Dec. 2002). Connecting the birth of black holes to enormous explosions called gamma-ray bursts. (E)\
[**Fundamental Physics:**]{}
6. \[QBH\] [“Quantum Black Holes,”]{} B. J. Carr and S. B. Giddings, Scientific American, [**292**]{}, 48-55 (May 2006). Might we soon be able to make black holes in the laboratory? (E)
7. [“The Reluctant Father of Black Holes,”]{} J. Bernstein, Scientific American, [**274**]{}, 80-85 (June 1996). The early history of our understanding of black holes and of Einstein’s resistance to accepting them. (E)
8. [“An Echo of Black Holes,”]{} T. A. Jacobson and R. Parentani, Scientific American, [**293**]{} 68-75 (May 2006). Flowing fluids in the laboratory can act much like black holes, and even have a form of Hawking radiation. (E)
9. [“Black Hole Computers,”]{} S. Lloyd and Y. J. Ng, Scientific American, [**291**]{}, 52-61 (Nov. 2004). Black Holes and Information Theory. (E)
10. [ “Information in the Holographic Universe,”]{} J. D. Bekenstein, Scientific American, [**289**]{}, 58-65 (Aug 2003). How black holes may be related to fundamental bounds on the information that objects can hold. (E)
11. \[IoG\] [“The Illusion of Gravity,”]{} J. Maldacena, Scientific American, [**293**]{}, 57-63 (Nov. 2005). String theory has holographic properties that seem deeply linked to black hole physics. (E)
12. [“Reality-Bending Black Holes,”]{} Scientific American, Special Edition, March 2007. This volume collects references \[BHB\]-\[IoG\] above, along with other articles on related topics. (E)
Undergraduate Material
======================
This section describes resources at the advanced (U.S.) undergraduate level. Some entries should be accessible to any student who has completed a one-year calculus-based physics course, while others require background in intermediate classical mechanics, quantum mechanics, and electromagnetism. The articles in section \[ugart\] address special topics not covered by standard textbooks, but which are nevertheless approachable by readers without detailed knowledge of GR. Students who wish to establish a working understanding of black holes should consult one of the undergraduate textbooks in section \[ugbook\].
\[ug\]
Articles {#ugart}
--------
1. [“Black holes: the inside story,”]{} S. Droz, W. Israel and S. M. Morsink, Physics World, [**9**]{}, 34-37 (1996). One of a very few articles with a good description of black-hole interiors. A key point is the instability of the so-called inner horizon of charged and rotating black holes. This instability forbids the use of black holes as passageways to other universes. The article is also posted online at [[www.phys.ualberta.ca/\~morsink/blackhole/](www.phys.ualberta.ca/~morsink/blackhole/)]{} in both English and French translations. (E)
2. \[BHD1\] [ “Spacetime embedding diagrams for black holes,”]{} D. Marolf, Gen. Rel. Grav. [**31**]{}, 919-944 (1999). \[arXiv:gr-qc/9806123\]. The curved space[*time*]{} of a Schwarzschild black hole can be visualized by drawing the so-called $rt$-plane as a curved surface inside flat 2+1 Minkowski space. The main text is accessible to those with some training in special relativity, though some knowledge of GR is required to follow the construction (in the appendix) of the diagrams from the black hole metric. (I)
3. [ “Gravity and the quantum,”]{} A. Ashtekar, New J. Phys. [**7**]{}, 198-198 (2005) \[arXiv:gr-qc/0410054\]. Ashtekar gives a broad overview of loop quantum gravity for the nonexpert. After introducing the basic ideas, section IIIB addresses black hole entropy in loop quantum gravity. The article should be accessible to students familiar with undergraduate quantum mechanics, the description of electromagnetism via 4-vector potentials, and some knowledge of GR. (I)
4. \[EW\] [ “Black Holes and Quark Confinement,”]{} E. Witten, Current Science [**81**]{}, 1576-1581 (2001). Witten takes the reader on a tour of ideas ranging from strings to quarks to black holes and the AdS/CFT duality. The climax is the fusion of these ideas into a stringy derivation of quark confinement in theories mathematically similar to, but still different than the QCD that describes the strong interactions of our universe. (I)
5. \[SW\] [ “A Microscopic Theory of Black Holes in String Theory,”]{} S. R. Wadia, Current Science [**81**]{} (12), 1591-1597 (2001). After providing an undergraduate-level description of black-hole thermodynamics and Hawking radiation, Wadia describes the extent to which these phenomena are currently described by string theory. The concluding section includes a useful set of open questions. (I)
6. \[Nature\] [ “Theoretical physics: A black hole full of answers,”]{} J. Zaanen, Nat. [**448**]{}, 1000-1001 ( 2007). Zaanen briefly describes current attempts to use black holes and string theory as mathematical tools to explain the physics of both high-temperature superconductors and quark-gluon plasmas. (I)
General Relativity Textbooks
----------------------------
A number of excellent GR textbooks are now available at the U.S. undergraduate level, each with a good treatment of black-hole physics. The texts differ in terms of both perspective and level of sophistication, but each one provides what is in principle a self-contained introduction. We have ordered the books below roughly in terms of level of presentation.
\[ugbook\]
1. \[GroundUp\] [**Gravity from the ground up,**]{} B. Schutz (Cambridge University Press, Cambridge, 2003). This excellent book emphasizes [*gravity*]{}, and its interactions with astrophysics. Assuming only high-school math and physics, Schutz addresses the solar system, the birth and life cycle of stars, relativity, cosmology, gravity waves, and gravitational lenses. The chapter on black holes summarizes a wide range of physics from accretion and black hole growth to Hawking radiation and black-hole entropy. (E/I)
2. \[Notes\] [**Notes on Relativity and Cosmology**]{}, D. Marolf (2003). Available on-line at [[www.physics.ucsb.edu/Smarolf/MasterNotes.pdf](www.physics.ucsb.edu/Smarolf/MasterNotes.pdf)]{}. Designed to convey understanding of relativity using only elementary calculus, these notes first thoroughly treat accelerated reference frames in special relativity. Spacetime diagrams are favored over calculations. GR is introduced by piecing together such frames using the equivalence principle. The chapter on black holes emphasizes similarities between horizons and accelerated frames in flat spacetime, and also summarizes ref. \[BHD1\] above. (E/I)
3. \[Taylor\] [**Exploring Black Holes: Introduction to General Relativity**]{}, E. F. Taylor and J. A. Wheeler (Addison Wesley Longman, San Francisco, 2000). Taylor and Wheeler focus on a nuts-and-bolts exploration of black-hole solutions as an introduction to GR. The text requires algebra and basic calculus but does not assume prior knowledge of physics or relativity. The emphasis is on the experience of an observer near a black hole. (I)
4. \[EllisWilliams\] [**Flat and Curved Space-times, 2nd Edition**]{}, by G. F. R. Ellis and R. Williams (New York, Oxford University Press, 2000). This excellent book on special and general relativity includes a survey of basic black-hole results including no-hair theorems, energy extraction, and thermodynamics. Ellis and Williams assume knowledge of basic calculus and introduce other mathematics as needed. (I)
5. \[hartle\] [**Gravity, An Introduction to Einstein’s General Relativity**]{}, J. B. Hartle (Addison Wesley, San Francisco, 2003). We recommend Hartle’s advanced undergraduate text to beginners and experts alike. It provides a thorough introduction to GR with a delightful emphasis on both physical concepts and experimental tests. Hartle treats black holes in some depth, discussing orbits, energy extraction, and the detailed solutions. (I)
Advanced Material: General {#gen}
==========================
Good introductions to black-hole physics can be found in many general-relativity textbooks, of which a large number are available at the U.S. graduate-student level. Some of our favorites are described below. Unless otherwise noted, these books require a thorough understanding of undergraduate-level classical mechanics but do not presume knowledge of more advanced material. Sections \[ABH\]-\[last\] provide a guide to particular topics in black-hole physics and include a few more specialized texts.
1. \[SchutzGreen\] [**A First Course in General Relativity**]{}, B. F. Schutz (Cambridge University Press, Cambridge, 1985). A text at the advanced undergraduate or early graduate-level with an excellent treatment of stress-energy tensors and a brief introduction to black-hole physics. (I/A)
2. \[Carroll\] [**Spacetime and Geometry: An Introduction to General Relativity,**]{} S. Carroll (Benjamin Cummings, San Francisco, 2003). A modern beginning graduate-level text that emphasizes links to field theory. Carroll’s two chapters on black-hole physics provide a readable introduction to the subject. (A)
3. \[WaldBook\] [**General Relativity**]{}, R. M. Wald (U. Chicago Press, Chicago, 1984). Wald’s book is the standard text for a mathematical-physics-based advanced graduate course on general relativity. (A)
4. \[MTW\] [**Gravitation**]{}, C. Misner, K. S. Thorne, J. A. Wheeler (W.H. Freeman and Company, New York, 1973). Misner, Thorne, and Wheeler provide an encyclopedic reference for GR as of the early 1970s. (A)
5. \[Weinberg\] [**Gravitation and Cosmology: Principles and Applications of the General Theory of Relativity**]{}, S. Weinberg (John Wiley, New York, 1972). A good text for those who wish to do calculations. Weinberg’s book emphasizes equations over geometric intuition and provides a solid connection to cosmology. (A)
6. \[HE\] [**The large scale structure of space-time**]{}, S. W. Hawking and G. F. R. Ellis (Cambridge University Press, Cambridge, 1973). The classic reference on causal structures, singularity theorems, and mathematical physics aspects of general relativity. (A)
Astrophysical Black Holes {#ABH}
=========================
It is now thought that almost all galaxies contain “supermassive" black holes at their centers, millions or even billions of times more massive than the Sun. Some of these objects power the most energetic (nonexplosive) phenomena in the universe, such as active galactic nuclei (AGN) and quasars. Others, like the black hole at the center of the Milky Way, are much more quiet. Galaxies are also thought to contain many examples of stellar-mass black holes, with masses a few times greater than that of the Sun, which are the evolutionary endpoint of very massive stars.
While we cannot observe black holes directly, we can observe and measure their effects on the surroundings. In fact, black holes are extremely efficient at converting the gravitational potential energy of in-falling gas into radiation. Except under special circumstances, the gas that falls into a black hole (be it interstellar/intergalactic gas, or supplied by a companion star) does not plunge in directly; instead it forms what is known as an accretion disk. The inner disks of supermassive black holes reach thousands of degrees Kelvin, while stellar-mass black holes can heat their disks to millions of degrees, where they emit in the X-ray part of the spectrum.
After listing a few basic texts in section \[Atexts\], in sections \[stellarMBH\]-\[IMBH\] we organize other resources by whether the relevant black holes are stellar mass, supermassive, or of intermediate mass scale(between the former two). Material concerned with specific astrophysical sources, most notably the supermassive black hole in the center of our own galaxy, appears separately in section \[BHCS\].
Astrophysics: Basic Textbooks and Reviews {#Atexts}
-----------------------------------------
See ref. \[BR\] from section \[ele\] A.
1. [ “Probes and Tests of Strong-Field Gravity with Observations in the Electromagnetic Spectrum,”]{} D. Psaltis, Living Reviews in Relativity (2008), \[arXiv:0806.1531\]. This review serves as a nexus between the previous sections of this Resource Letter and the following ones, which cover the observable aspects of black holes. Psaltis summarizes the current prospects for using astrophysical sources hosting black holes as tools for testing the predictions of GR in the strong-field regime, and does so in a concise and clear manner. (I)
2. [**Accretion Power in Astrophysics**]{}, J. Frank, A. King, and D. Raine (Cambridge University Press, Cambridge, 2002). Assuming a basic knowledge of physics, the authors examine the process of accretion, [*i.e.*]{} the extraction of gravitational potential energy from matter falling onto a compact object. A fundamental textbook for astronomy graduate students. (I/A)
3. \[super\] [**The Galactic Supermassive Black Hole**]{}, F. Melia (Princeton University Press, Princeton, 2007). This book is an excellent starting point for astrophysics graduate students interested in the black hole associated with galactic center source Sagittarius A\*. The history of observations is described along with the observational signatures themselves. These observations range from radio to X-ray wavelengths. Nearby stellar orbits, accretion, flares, and lensing are discussed. While primarily written from the astrophysical point of view, the text devotes one chapter to explaining elements of GR that are key to understanding black holes in general. (A)
4. [ “Fluorescent iron lines as a probe of astrophysical black hole systems,”]{} C. Reynolds and M. Nowak, Physics Reports, [**377**]{}, 389-466 (2003) \[arXiv:astro-ph/0212065\]. As gas accretes onto a black hole, its velocity approaches the speed of light. This should give rise to observable relativistic effects, such as gravitationally red-shifted emission lines. In this extensive review, Reynolds and Nowak describe relativistic iron lines as probe of the innermost regions around accreting black holes. It also features a broad, pedagogical introduction. (I)
5. [ “Black Holes and Relativistic Jets,”]{} R. D. Blandford, Progress of Theor. Phys. Suppl. [**143**]{}, 182-201 (2001) \[arXiv:astro-ph/0110394\]. Accreting black holes of all mass scales seem to be able to form narrow, bipolar streams of outflowing matter/energy, as known as jets. Blandford discusses a variety of viable mechanisms through which mass, angular momentum, and energy can escape the pull of gravity and be ejected outwards at relativistic velocities. Particular attention is paid to the connection between the jet, the accretion disk, and the black hole itself. Written by a world leader, as well as perhaps the most cited author(ity), in the field. (I/A)
stellar-mass black holes {#stellarMBH}
------------------------
1. [**Black Holes, White Dwarfs and Neutron Stars: The Physics of Compact Objects**]{}, S. L. Shapiro and A. Teukolsky (John Wiley & Sons Inc., New York, 1983). This excellent textbook brings together different branches of physics – from nuclear physics, to hydrodynamics, to special relativity and GR – to understand the birth, evolution and death of a massive star under the special circumstances that lead to the formation of a binary star system hosting a relativistic compact object. (A)
2. [ “Black holes in binary systems. Observational appearance,”]{} N. I. Shakura and R. A. Sunyaev, Astron. Astrophys., [**24**]{}, 337-355 (1973). In this seminal paper, Shakura and Sunyaev explain how, in a binary system made of an evolved star that transfers part of its envelope mass to a black hole, the outward transfer of angular momentum leads to the formation of a disk around the black hole. (A)
3. [ “Image of a spherical black hole with thin accretion disk,”]{} J.-P. Luminet, Astron. Astrophys., [**75**]{}, 228-235 (1979). The first investigation of the actual optical appearance of a spherical black hole surrounded by thin accretion disk; the spectral shifts arising from gravitational and Doppler shifts result in strong asymmetry in the flux distribution. This calculation can be generalized to black holes of any mass. (A)
4. [ “Black hole binaries,”]{} \[mccr\] J. E. McClintock and R. A. Remillard, in [**“Compact stellar X-ray sources,”**]{} edited by W. Lewin and M. van der Klis (Cambridge University Press, Cambridge, pp. 157-214, 2006) \[arXiv:astro-ph/0306213\]. The reader is introduced to the 40+ known systems hosting candidate stellar-mass black holes in the Milky Way. About half of them are known to contain an object whose mass exceeds the maximum theoretical mass for a neutron star, and are referred to as “dynamically confirmed” black hole accretors. (I)
5. [ “Relativistic Jets from X-ray binaries,”]{} R. P. Fender, in [**“Compact stellar X-ray sources,”**]{} edited by W. Lewin and M. van der Klis (Cambridge University Press, Cambridge, pp. 381-420, 2006) \[arXiv:astro-ph/0303339\]. Relativistic jets are among the most spectacular objects in the universe, and are increasingly being found throughout the Milky Way. Despite being discovered in the radio band, they have been proven to emit across the entire electromagnetic spectrum. This review deals with the variety of jets from galactic stellar-mass black holes (as well as neutron stars), and their connection with the accretion flow. (I)
6. [ “Advection-dominated Accretion and Black Hole Event Horizons,”]{} R. Narayan and J. E. McClintock, to appear in New Astronomy Reviews \[arXiv:astro-ph/0803.0322\]. This work summarizes the observational evidence for the presence of an event horizon in stellar-mass black holes by comparing the luminosities of black holes and neutron stars accreting mass at low rates. If the accretion flow has a low enough density, the gravitational potential energy can remain trapped in the flow. In the case of a neutron star, half of that energy will be released as X-rays upon impact on the star surface. Instead, the energy will be advected across the event horizon in the case of a black hole. Indeed, this luminosity gap has been observed in X-rays. (A)
7. [ “Gamma-ray bursts from stellar mass accretion disks around black holes,”]{} S. Woosley, Astrophys. J. [**405**]{}, 273-277 (1993). Gamma-ray bursts are the most energetic, explosive astrophysical events we know of. In a flash lasting few ticks of a clock, more energy is released than over the entire Sun lifetime. Woosley’s model for the production of (long) gamma-ray bursts involves the formation of a stellar-mass black hole as the evolutionary endpoint of Wolf-Rayets, i.e. a particular class of massive stars. (A)
supermassive black holes {#stellarSMBH}
------------------------
1. [**Active galactic nuclei: from the central black hole to the galactic environment**]{}, J. Krolik (Princeton University Press, Princeton, 1998). A comprehensive description of the observational properties of supermassive black holes that are actively accreting at the center of AGN. A must-read resource for astronomers, physicists interested in applications of the theory of gravitation, and graduate students. (E/I)
2. [ “Black Hole Models for Active Galactic Nuclei,”]{} M. J. Rees, Annu. Rev. Astron. Astroph., [**22**]{}, 471-506 (1984). An inspired – and inspiring – work that reviews the body of evidence for supermassive black holes as the power engine for AGN: quasars, radio galaxies, blazars, and related objects. (I)
3. [ “Variability of Active Galactic Nuclei,”]{} B. M. Peterson, in [**Advanced Lectures on the Starburst-AGN Connection**]{}, edited by I. Aretxaga, D. Kunth and R. M[ú]{}jica (World Scientific, Singapore, 2001) pp. 3-68. A comprehensive description of ‘reverberation mapping’ techniques, which allow one to infer the masses of black holes in active galaxies at high redshifts. This method is based on measuring the time delay between variations in the line and continuum emission from distant AGN. The delay provides an indirect measure of the distance between the supermassive black holes and the regions where the emission lines are being excited. This characteristic scale size constrains the size of the black hole, and thus its mass. (I)
4. [ “Quasars and galaxy formation,”]{} J. Silk and M. J. Rees, Astron. and Astrophys. [**331**]{}, L1-L4 (1998) \[arXiv:astro-ph/9801013\]. A visionary paper on the interplay between the formation of black-hole seeds in the early universe and the evolution of (proto-)galaxies. (I/A)
5. [ “Supermassive Black Holes in Galactic Nuclei: Past, Present and Future Research,”]{} L. Ferrarese and F. Holland, Space Sciences Rev., [**116**]{}, 523-624, (2005) \[arXiv:astro-ph/0411247\]. This review discusses the current status of supermassive black-hole research from an observational perspective. Recent observations have unveiled empirical scaling relations between the mass of the central black hole and the global properties of the host galaxy (see refs. \[scaling1\],\[scaling2\]), suggesting that black holes influence the very formation and evolution of cosmic structures since the dawn of time. (I)
6. [ “Massive black holes: formation and evolution”]{} M. J. Rees and M. Volonteri, in [**Black Holes from Stars to Galaxies – Across the Range of Masses,**]{} edited by V. Karas and G. Matt (Cambridge University Press, Cambridge, 2007) \[arXiv:astro-ph/0701512\]. Rees and Volonteri summarize the progresses and open questions about the formation and growth of supermassive black holes, as well as the prospects for using them as laboratories for testing the strong gravity regime. This work includes a list of the most important (albeit advanced) bibliographic references in the field. (I/A)
7. \[dimatteo\] [ Energy input from quasars regulates the growth and activity of black holes and their host galaxies]{}, T. Di Matteo, V. Springel, L. Hernquist, Nat. [**433**]{}, 604-607 (2005) \[arXiv:astro-ph/0502199\]. The evolution of galaxies along with their nuclear supermassive black holes has been followed by means of detailed numerical simulations. It is found that a galaxy-galaxy merger leads to a massive enhancement of the inner accretion rate, enough to switch on the nuclear black hole as a quasar. In turn, the energy release by the quasar is able to quench star formation and halt further growth. This self-regulatory mechanism sets the quasars’ duty cycle and may be able to explain the empirical scaling relations refs. \[scaling1\] and \[scaling2\]. (I)
8. \[feedback\] [ “The many lives of active galactic nuclei: cooling flows, black holes and the luminosities and colors of galaxies,”]{} D. J. Croton [*et al.*]{}, Mon. Not. R. Astron. Soc., [**365**]{}, 11-28 (2006) \[arXiv:astro-ph/0508046\]. Energy “feedback” from supermassive black holes could solve a number of problems faced by the hierarchical paradigm for the formation and evolution of cosmic structures at galactic and sub-galactic scales, such as the observed red colors of massive spheroidal galaxies. Croton et al. couple the output from state-of-the-art dark-matter simulations to semianalytical modeling to show that nuclear supermassive black holes may be able to quench the formation of young blue stars by injecting heat and mechanical power into the surroundings. In order for black-hole feedback to be effective, low levels of prolonged activity are needed, as opposed to sudden bursts of power. These effects should be observable in the cores of nearby, formally inactive galaxies. (A)
9. [ “Massive Black Hole Binary Evolution,”]{} D. Merritt and M. Milosavljević, Living Reviews in Relativity (2005), [<http://relativity.livingreviews.org/Articles/lrr-2005-8/>]{}. This article reviews the observational evidence (or lack thereof) for binary supermassive black holes and discusses their evolution and final fate. Although a compelling case for a bound binary system is yet to be discovered, a number of sources are believed to host two supermassive black-hole binaries at projected separations of less than 3000 light-years. (A)
10. [ “LISA observations of rapidly spinning massive black hole binary systems,”]{} A. Vecchio, Phys. Rev. D, [**70**]{} 042001, 17pp. (2004) \[arXiv:astro-ph/0304051\]. The Laser Interferometer Space Antenna (LISA: [[lisa.nasa.gov/](lisa.nasa.gov/)]{}) is a (planned) space-based gravitational-wave observatory. Detecting gravitational waves from supermassive black-hole binaries will have tremendous impact on our understanding of gravity in the highly nonlinear relativistic regime, as well as the formation and evolution of galaxies in a cosmological context. This review focuses on the effects of black-hole spin on the expected LISA gravitational-wave signal, and describes how accounting for these effects can substantially reduce the uncertainties on other parameters, such as the reduced mass. (A)
Intermediate-Mass Black Holes {#IMBH}
-----------------------------
1. [ “Ultraluminous X-Ray Sources in External Galaxies,”]{} A. R. King et al., Astrophys. J. [**552**]{}, L109-L112 (2001) \[arXiv:astro-ph/0104333\]. With the launch of sensitive X-ray observatories, a new population of compact, luminous, off-nuclear X-ray sources has been unveiled in external galaxies. While these so called “ultraluminous X-ray sources” are natural intermediate-mass black-hole candidates, several –less exotic– explanations are offered in this paper. (I/A)
2. [ “A Comparison of Intermediate-Mass Black Hole Candidate Ultraluminous X-Ray Sources and Stellar-Mass Black Holes,”]{} J. M. Miller, A. C. Fabian and M. C. Miller, Astrophys. J., [**614**]{}, L117-L120 (2004) \[arXiv:astro-ph/0406656\]. On the same topic as the previous work, this paper focusses instead on the observational evidence for intermediate mass black holes in a small number of ultraluminous X-ray sources, namely those where the X-ray continuum spectra appear too cold to be due to accreting stellar-mass black holes. (I/A)
3. [ “X-ray observations of ultraluminous X-ray sources,”]{} T. P. Roberts, Astrophys. and Space Science, [**311**]{}, 203 (2007) \[arXiv:astro-ph/0706.2562\]. A brief, critical review on the X-ray properties of ultraluminous X-ray sources, and their possible physical nature. (I/A)
Black Holes: Case Studies {#BHCS}
-------------------------
Throughout this section, we list a number of journal papers that report on observational milestones on specific black-hole sources, or an ensemble of them. The two last entries are devoted to the case for a 3-4 million solar-mass black hole at the center of our own Milky Way. This list is not meant to be complete, but simply to offer a glance at groundbreaking astrophysical discoveries.
1. [ “Observation of a Correlated X-Ray Transition in Cygnus X-1,”]{} H. Tananbaum et al., Astrophys. J. [**177**]{}, L5-L5 (1972). X-ray astronomy began only in the early 1960’s. The first rocket flight to detect a cosmic X-ray source was launched in 1962 by a group at American Science and Engineering. In the 1970s, the dedicated X-ray satellite “Uhuru” repeatedly observed a newly discovered X-ray source in the Cygnus constellation over a period of 16 months. Its X-ray flux was found to vary significantly, and in correlation with its radio counterpart. The first “black-hole state transition” (see ref. \[mccr\]) had just been discovered. (I)
2. [ “X-ray fluorescence from the inner disc in Cygnus X-1,”]{} A. C. Fabian [*et al.*]{}, Mon. Not. R. Astron. Soc. [**238**]{}, 729-736 (1989). It is argued that the broad emission line in the X-ray spectrum of the 10-solar-mass black-hole candidate in Cygnus X-1 is well accounted for by a model of fluorescent emission from the inner parts of an inclined accretion disk. (I/A)
3. [ “A Superluminal Source in the Galaxy,”]{} I. F. Mirabel and L. F. Rodriguez, Nat., [**371**]{}, 46-48 (1994). A seminal paper reporting on the discovery of radio-emitting plasmons traveling with superluminal velocities (apparent velocities greater than the speed of light) within the Milky Way. The accepted explanation is that the plasmons are ejected in opposite directions from the central stellar-mass black hole at speeds higher than 0.7$c$. If the ejection direction is close to the observer’s line of sight, this can lead to apparent superluminal velocities. (E/I)
4. [ “Gravitationally Redshifted Emission Implying an Accretion Disk and Massive Black-Hole in the Active Galaxy MCG 6-30-15,”]{} Y. Tanaka [*et al.*]{}, Nat., [**375**]{}, 659-661 (1995). Tanaka and collaborators report on the discovery of relativistic effects in an X-ray emission line from ionized iron in the galaxy MCG 6-30-15. The line is extremely broad and asymmetric, with most of the line flux being red-shifted. (I/A)
5. [ “Formation of the radio jet in M87 at 100 Schwarzschild radii from the central black hole” ]{} W. Junor, J. A. Biretta, and M. Livio, Nat., [**401**]{}, 891-892 (1999). This work reports on high-resolution radio observations of the inner regions of the nearby active galaxy M87, which powers a highly collimated jet. The jet radio maps are consistent with the hypothesis that jets are formed by an accretion disk around the central black hole, which is threaded by a magnetic field. (I)
6. [ “Inward Bound—The Search For Supermassive Black Holes In Galactic Nuclei”]{} \[scaling1\] J. Kormendy and D. Richstone, Annu. Rev. Astron. Astroph. [**33**]{}, 581-581 (1995). The discovery of an empirical scaling relation between the mass of nuclear supermassive black holes and the mass of the host galaxy’s bulge. (A)
7. [ “A Fundamental Relation between Supermassive Black Holes and Their Host Galaxies,”]{} \[scaling2\] L. Ferrarese and D. Merritt, Astrophys. J. [**539**]{}, L9-L12, (2000) \[arXiv:astro-ph/0006053\] & [ “A Relationship between Nuclear Black Hole Mass and Galaxy Velocity Dispersion,”]{} K. Gebhardt et al., Astrophys. J. [**539**]{}, L13-L16 (2000) \[arXiv:astro-ph/0006289\]. The discovery (by two different groups) of an empirical scaling relation between the mass of nuclear supermassive black holes and the velocity dispersion of stars orbiting much farther than the black-hole sphere of influence. (A)
8. [ “A 20,000 Solar Mass Black Hole in the Stellar Cluster G1,”]{} K. Gebhardt, R. M. Rich and L. C. Ho, Astrophys. J., [**578**]{}, L41-L46 (2002) \[arXiv:astro-ph/0209313\]. At the center of a globular cluster, this object represents the strongest intermediate-mass black-hole candidate to date. Unlike in the examples listed in section \[IMBH\], its presence is inferred from dynamics rather than from electromagnetic signals. (A)
9. \[sgr1\][ “The nucleus of our Galaxy,”]{} R. Genzel, D. Hollenbach and C. H. Townes, Reports on Progress in Physics, [**57**]{}, 417-479 (1994). The subject of this review is the central 100 parsecs of our galaxy, with a strong focus on the central few parsecs. The authors discuss the stellar and interstellar components, the importance of magnetic and gravitational forces, the evidence for stellar formation and a central massive black hole, and the origin and nature of ionization, outflows and interstellar-gas dynamics. (I/A)
10. \[sgr2\] [ “High Proper-Motion Stars in the Vicinity of Sagittarius A\*: Evidence for a Supermassive Black Hole at the Center of Our Galaxy,”]{} A. Ghez, B. L. Klein, M. Morris, E. E. Becklin, Astrophys. J., [**509**]{}, 678-686 (1998) \[arXiv:astro-ph/9807210\]. The description of a groundbreaking astrophysics experiment: The motion of 90 faint stars in the vicinity of the galactic center has been followed over a period of two years with the Keck 10m telescope. Their motion allows for the determination of the enclosed mass through Kepler’s laws. A central dark mass of about 3 million solar masses is confined to an area about 10 times smaller than Earth’s orbit around the Sun. The inferred density leads us to the conclusion that our galaxy harbors a massive central black hole. (A)
11. [“ Evidence for a Black Hole from High Rotation Velocities in a Sub-Parsec Region of NGC4258 ,”]{} M. Miyoshi, J. Moran, J. Herrnstein, L. Greenhill, N. Nakai, P. Diamond, and M. Inoue, Nat., [**373**]{}, 127-129 (1995). Along with the nucleus of the Milky Way (refs. \[sgr1\],\[sgr2\]), this galaxy presents the best known case for a super-massive black hole. Here the central dark mass is weighed by reconstructing the line-of-sight velocity profile of water-maser-emitting gas orbiting around it. Since water-maser emission is a narrow feature, the line-of-sight velocity can be determined from the Doppler shift variations of the observed maser frequency. The emission regions trace Keplerian orbits which allow one to estimate the enclosed mass as a function of radius, and hence the dark mass density. NGC4258 is thought to host a 40 million solar mass black hole. (I)
Image Galleries: Black Holes from Space
---------------------------------------
This is all well and fine, one may say. But what is it that we actually [*see*]{} when we go up there, outside of Earth’s atmosphere? Browsing the following websites, the reader will be taken through a journey of the universe as seen by NASA’s three “Great Observatories.”
1. [hubblesite.org/gallery/album/exotic\_collection/black\_hole/](hubblesite.org/gallery/album/exotic_collection/black_hole/),. An amazing gallery of AGN pictures, taken by one the most powerful optical eyes ever built. NASA’s crown jewel, the Hubble Space Telescope, has provided us with a window to the unexplored universe since 1990. (E)
2. [[chandra.harvard.edu/photo/category/blackholes.html](chandra.harvard.edu/photo/category/blackholes.html)]{},. Black holes shine in the X-ray band because of the high temperatures attained by the in-falling gas; however, X-rays from space are absorbed by the Earth atmosphere. Since its launch in 1999, the Chandra X-ray Observatory has contributed substantially to our understanding of accreting black holes in the Milky Way as well as external galaxies. This friendly site is a collage of Chandra images, all equipped with accessible explanations. (E)
3. [[gallery.spitzer.caltech.edu/Imagegallery/subcat.php?cat=Astronomical\_Images&subcat=Galaxies\_and\_the\_Universe](gallery.spitzer.caltech.edu/Imagegallery/subcat.php?cat=Astronomical_Images&subcat=Galaxies_and_the_Universe)]{},. The last of NASA’s Great Observatories to fly, Spitzer is a flying infrared telescope: UV/X-rays from accreting supermassive black holes can be partly absorbed by surrounding dust and re-emitted at lower frequencies. Here is how AGN would look if human eyes were sensitive to the same frequencies as microwaves. (E)
Statics, Dynamics, and Thermodynamics of Black Holes {#SDT}
====================================================
This section addresses the well-established mathematical physics of black holes. We have attempted to impose some structure on this highly interconnected body of knowledge by dividing the material as follows: Section \[statics\] deals with the statics of black holes, including the construction of stationary solutions, no-hair and uniqueness theorems, some work on hairy black holes, and definitions of horizons. Perturbations, quasi-normal modes, and stability of stationary solutions is addressed in section \[pert\]. We have chosen to include the important subject of extreme-mass in-spirals and the calculation of gravity-wave production by post-Newtonian expansions in this section. Section \[nonlin\] then considers the strongly nonlinear dynamics of black holes associated with critical phenomena or with black-hole collisions. Most of our discussion of numerical GR can be found here, as is the associated material on gravity waves. Section \[thermo\] contains resources on black-hole thermodynamics, quantum-field theory in curved spacetime, and the associated description of Hawking radiation. More microscopic treatments of Hawking radiation and black-hole entropy are postponed to section \[micro\]. Sections \[statics\]-\[thermo\] focus primarily on black holes in $d=4$ spacetime dimensions; resources on black holes in other spacetime dimensions have been collected in section \[dnot4\]. We have chosen to include discussions of possible black-hole production at particle colliders in this latter section.
The reader should be aware of a few peculiarities that result from this organization. Resources on gravity-wave production by black holes appear in both sections \[pert\] and \[nonlin\]. Resources on isolated and dynamical horizons appear in sections \[statics\], \[nonlin\], \[thermo\], and also in section \[loop\]. There is also much overlap between section \[dnot4\] (black holes in $d \neq 4$ dimensions) and sections \[thermo\], \[strings\], and \[last\], and between sections \[thermo\] (thermodynamics) and \[other\] (where we have chosen to locate discussions of entropy bounds and the proposed holographic principle, as well as Jacobson’s work (ref. \[EEOS\]) on the Einstein equation of state). We apologize to the reader for any confusion this may cause.
Statics
-------
1. [ ’No-Hair’ Theorems: Folklore, Conjectures, Results]{}, Chrusciel, P.T., in [Differential Geometry and Mathematical Physics]{}, J Beem, and K. L. Duggal., eds., vol 170, pp. 23-49 (Am. Math. Soc., Providence, 1994) \[arXiv:gr-qc/9402032\] . [ “Uniqueness of Stationary, Electro-Vacuum Black Holes Revisited,”]{} P.T. Chrusciel, Helv. Phys. Acta, 69, 529-552 (1996). For a related online version see: [ “Uniqueness of Stationary, Electro-Vacuum Black Holes Revisited,”]{} P.T. Chrusciel (October, 1996) \[arXiv:gr-qc/9610010 \]. An important property of black holes is that, when left to themselves, they settle down to solutions described by only a few parameters, most importantly their mass $M$ and angular momentum $J$. In other words, the surface of a black hole [*must*]{} be very smooth; even in principle, the surface of a black hole cannot support mountains, valleys, or other complicated features. One says that black holes “have no hair.” These papers review and complete classic work on the no-hair theorems. (A)
2. [ “Stationary Black Holes: Uniqueness and Beyond,”]{} Markus Heusler, Living Reviews in Relativity (1998), [[relativity.livingreviews.org/Articles/lrr-1998-6/index.html](relativity.livingreviews.org/Articles/lrr-1998-6/index.html)]{}. In sufficiently complicated settings, the number of parameters needed to describe black holes increases and black holes can support a limited amount of “hair." This review describes the original no-hair theorems as well as some of the more complicated hairy black holes. See also ref. \[NA2\] for more on non-abelian black holes. (A)
3. \[IB\] [ “Black hole boundaries,”]{} I. Booth, Can. J. Phys. [**83**]{}, 1073-1099 (2005) \[arXiv:gr-qc/0508107\]. Black holes are traditionally defined as regions of space-time from which in principle signals cannot propagate to infinity. Such definitions clearly break down in cosmological settings, where appropriate asymptotic regions may not exist. Booth reviews recent work on alternative “quasi-local” definitions of black-hole horizons. See also refs. ref[IH1]{},\[AK\],\[Ann\]. (A)
Perturbative Black-Hole Dynamics {#pert}
--------------------------------
1. \[QN1\] [ “Quasi-Normal Modes of Stars and Black Holes,”]{} K. D. Kokkotas and B. Schmidt, Living Reviews in Relativity (1999), [<http://relativity.livingreviews.org/Articles/lrr-1999-2/>]{}. When a black hole is perturbed, it “rings” with a characteristic set of frequencies. However, the amplitude at each frequency decays as energy falls through the horizon and radiates to infinity. As a result, the solutions associated with this ringing are called “quasi-normal” modes. Kokkotas and Schmidt review such perturbations for both black holes and relativistic stars, and in the process review the stability of black holes. See also ref. \[QN2\]. (A)
2. \[QN2\] [ “Quasi-Normal Modes and Gravitational Wave Astronomy,”]{} V. Ferrari and L. Gualtieri, Gen. Rel. Grav. [**40**]{}, 945-970 (2008) \[arXiv:0709.0657 \[gr-qc\]\]. A review of quasi-normal modes with emphasis on possible detection of the associated gravitational waves. See also ref. \[QN1\]. (A)
3. \[EMRI1\] [ “Extreme Mass Ratio Inspirals: LISA’s unique probe of black hole gravity,”]{} K. Glampedakis, Class. Quant. Grav. [**22**]{}, S605-S659 (2005) \[arXiv:gr-qc/0509024\]. The proposed LISA gravitational-radiation detector is expected to be sensitive to gravity waves produced when stars or stellar-mass black holes spiral into supermassive black holes. In such cases the extreme ratio between the two black-hole masses provides a small parameter that allows controlled calculations of both these gravitational waves and the back-reaction on the small body’s orbit. This review begins with free-particle motion on the black-hole background and builds up toward more complicated calculations. (A)
4. \[EMRI2\] [ Analytic Black Hole Perturbation Approach to Gravitational Radiation,”]{} M. Sasaki and H. Tagoshi, Living Reviews in Relativity (2003), [[relativity.livingreviews.org/Articles/lrr-2003-6/index.html](relativity.livingreviews.org/Articles/lrr-2003-6/index.html)]{}. A slightly more advanced review of extreme-mass in-spirals describing both the relevant perturbative and post-Newtonian techniques. (A)
5. [ “Resource Letter GrW-1: Gravitational waves,”]{} J. M. Centrella, Am. J. Phys. [**71**]{}, 520-525 (2003) \[arXiv:gr-qc/0211084\]. Centrella’s Resource Letter is an excellent guide to the pre-2003 literature on the production of gravity waves by events involving black holes. (E/I/A)
6. [**Black Holes: The Membrane Paradigm**]{}, by K. S. Thorne, D. A. MacDonald, and R. H. Price (Yale University Press, 1986). This compilation of classic papers shows that black-hole horizons respond to outside perturbations like a two-dimensional viscous fluid or membrane. The horizon conducts electricity but (at least classically) does not conduct heat. As a result, horizons are described by both the Navier-Stokes equations and Ohm’s law. (A)
7. [ “The internal structure of black holes]{}, W. Israel, in [Black holes and relativistic stars]{} edited by R. Wald (U. of Chicago Press, Chicago, 1998) 137-151. Analytic solutions for charged and rotating black holes contain a so-called Cauchy horizon at which predictability appears to break down. However, this Cauchy horizon is unstable to the formation of a singularity (where quantum effects clearly become relevant). This concise summary of black-hole interiors provides a good introduction and guide to the pre-1998 literature on Cauchy-horizon instabilities and the resulting phenomenon of mass inflation. See also [ “The interior of charged black holes and the problem of uniqueness in general relativity,”]{} M. Dafermos, Communications on Pure and Applied Mathematics, [**58**]{} 445-504, (2005) \[gr-qc/0307013\]. (A)
8. \[NA2\] [ Internal Structure of Black Holes and Spacetime Singularities: Proceedings]{}, Annals of the Israel Physical Society vol. 13, edited by L. M. Burko and A. Ori (Institute of Physics Publishing, London, 1998). This volume contains a number of excellent articles on subjects ranging from Cauchy-horizon instabilities (in diverse contexts) to the internal structure of non-abelian black holes. (A)
Nonlinear Black-Hole Dynamics: Numerical GR and Black-Hole Collisions {#nonlin}
---------------------------------------------------------------------
1. [ “Critical Phenomena in Gravitational Collapse,”]{} C. Gundlach, Living Reviews in Relativity (1999), [[relativity.livingreviews.org/Articles/lrr-1999-4/index.html](relativity.livingreviews.org/Articles/lrr-1999-4/index.html)]{}. In 1993, Choptuik discovered a kind of critical phenomenon associated with black-hole formation in certain contexts. For spherically symmetric scalar fields, he considered generic 1-parameter families of initial data that interpolate between weakly gravitating data (which does not form a black hole) and strongly gravitating data (which does). At the threshold of black-hole formation one finds a complicated, singular, scale-periodic solution while black holes of arbitrarily small mass arise close to this threshold. Related phenomena were later found in many systems. Gundlach reviews the modern understanding of this field. (A)
2. [ “Binary Black Hole Coalescence,”]{} F. Pretorius, arXiv:0710.1338 \[gr-qc\]. The coalescence of two black holes of comparable mass is a difficult problem, best treated by numerical techniques. (See refs. \[EMRI1\], \[EMRI2\] for the case of extreme-mass ratios.) After many years of effort, such methods have recently become quite successful at calculating the properties of the final black hole and producing detailed waveforms for gravitational radiation. Pretorius reviews these techniques, their results, and the implications for astrophysics and other fields. (A)
3. [ “Initial Data for Numerical Relativity,”]{} Greg Cook, Living Reviews in Relativity, lrr-2000-5, relativity.livingreviews.org/Articles/lrr-2000-5/index.html Extracting useful information from numerical simulations requires starting with appropriate initial data. Cook describes the current methods and their limitations, with an eye toward future progress. (A)
4. [ “Introduction to Isolated Horizons in Numerical Relativity,”]{} O. Dreyer, B. Krishnan, D. Shoemaker and E. Schnetter, Phys. Rev. D [**67**]{}, 024018, 14pp. (2003) \[arXiv:gr-qc/0206008\]. Dreyer [*et al*]{} describe methods for using isolated-horizon techniques (see refs. \[IB\],\[IH1\], \[AK\], \[Ann\]) to compute the mass $M$ and angular momentum $J$ of black holes in numerical simulations. (A)
Black-Hole Thermodynamics and Hawking Radiation {#thermo}
-----------------------------------------------
See also refs. \[2d\], \[TeV1\]-\[TeV3\].
1. \[FirstLaw\] [ “The Four Laws of Black Hole Mechanics,”]{} J. M. Bardeen, B. Carter, and S. W. Hawking, Commun. Math. Phys., [**31**]{}, 161-170, (1973). This seminal paper derives a “first law of black-hole mechanics” with striking similarities to the first law of thermodynamics. The authors argue that black holes also satisfy analogues of the 0th, 2nd, and 3rd laws of thermodynamics. The idea that black holes are thermodynamic systems is now widely regarded as a fundamental clue to the nature of quantum gravity. (A)
2. \[entropy\] [ “Black Holes and Entropy,”]{} J. D. Bekenstein, Phys. Rev. D, [**7**]{}, 2333-2346 (1973). [ “Generalized Second Law of Thermodynamics in Black-Hole Physics,”]{} J.D. Bekenstein, Phys. Rev. D, [**9**]{}, 3292-3300 (1974). These foundational works suggest that the area of a black hole measures its entropy and formulate the generalized second law of thermodynamics: that the summed entropy of matter plus black holes should not decrease. (A)
3. \[Wald\] [ “The thermodynamics of black holes,”]{} R. Wald, Living Rev. Rel. [**4**]{}, 6 pp. 46 (2001) \[arXiv:gr-qc/9912119\]. relativity.livingreviews.org/Articles/lrr-2001-6/index.html . Wald provides a comprehensive survey of black-hole thermodynamics and the generalized second law. A detailed guide to the literature is presented, to which the reader is referred for derivations. Hawking radiation and associated issues are also discussed. (I/A)
4. \[IH1\] [“Black Hole Mechanics]{},” A. Ashtekar, in [**Encyclopedia of Mathematical Physics**]{}, (Elsevier, New York, 2006) 300-305 . Also available at cgpg.gravity.psu.edu/people/Ashtekar/articles/bhm.pdf Ashtekar briefly reviews the standard formulation of black-hole mechanics and then describes the more local isolated- and dynamical-horizon formulations. See also refs. \[IB\],\[AK\],\[Ann\]. (A)
5. \[AK\] [ “Isolated and Dynamical Horizons and Their Applications,”]{} A. Ashtekar and B. Krishnan, Living Reviews in Relativity, lrr-2004-10. relativity.livingreviews.org/Articles/lrr-2004-10/ This more thorough review begins with the basics of the isolated and dynamical horizon quasi-local formalism for black-hole thermodynamics, but proceeds to address applications in numerical GR, mathematical physics, and quantum gravity. (A)
6. \[HawkRad\] [ “Particle Creation by Black Holes,”]{} S. W. Hawking, Commun. Math. Phys., [**43**]{}, 199-220 (1975). Hawking shows that, due to quantum effects, black holes emit thermal radiation at (very low) temperatures set by their so-called surface gravity. This is the original paper on such Hawking radiation. The calculation identifies the black-hole entropy with $A/4\ell_p^2$ where $A$ is the event-horizon area and $\ell_p$ is the so-called Planck length. (A)
7. \[TJRev\] [ “Introduction to quantum fields in curved spacetime and the Hawking effect,”]{} T. Jacobson in [**Lectures on Quantum Gravity**]{}, edited by A. Gomberoff and D. Marolf (Springer, New York, 2005). \[arXiv:gr-qc/0308048.\] Based on lectures given at a summer school, Jacobson’s notes provide a very physical introduction to quantum fields in curved spacetime and the Hawking effect. This is an excellent reference for students new to the subject. Cosmological particle creation, transplanckian questions, and other issues are also addressed. The focus is on Lorentzian techniques. (A)
8. \[SRRev\] [ “Black hole thermodynamics,”]{} S. F. Ross, arXiv:hep-th/0502195. Also based on lectures at a summer school, Ross’s notes are a good complement to ref. \[TJRev\] above. This work includes a brief introduction to black holes and a review of classical black-hole thermodynamics. Ross also discusses quantum-field theory in curved spacetime, but emphasizes Euclidean techniques. (A)
9. \[BD\] [**Quantum Fields in curved space**]{}, N. D. Birrell and P. C. W. Davies (Cambridge University Press, Cambridge, 1982). A classic textbook on quantum-field theory in curved spacetime. This text begins with the basics and works up through the derivation of Hawking radiation. (A)
10. \[WaldQFT\] [**Quantum Field Theory in Curved Spacetime and Black Hole Thermodynamics**]{}, R. M. Wald (U. Chicago Press, Chicago, 1994). This brief text quickly summarizes these fields as of 1994. The approach is somewhat more mathematical than refs. \[TJRev\]-\[BD\]. (A)
11. [ “Path Integral Derivation Of Black Hole Radiance,”]{} J. B. Hartle and S. W. Hawking, Phys. Rev. D [**13**]{}, 2188-2203 (1976); [ “Action Integrals and Partition Functions in Quantum Gravity,”]{} G. Gibbons, and S. W. Hawking, Phys. Rev. D, [**15**]{}, 2752-2756 (1977). Together, these two papers introduce Euclidean techniques into black-hole thermodynamics. Hartle and Hawking use analytic continuation to derive Hawking radiation, while Gibbons and Hawking introduce the use of Euclidean solutions to approximate the gravitational partition functions. (A)
12. \[modify\] [ “On the origin of the particles in black hole evaporation,”]{} R. Schutzhold and W. G. Unruh, arXiv:0804.1686 \[gr-qc\]; The above references derive Hawking radiation within the framework of locally Lorentz-invariant quantum-field theory on a fixed background spacetime. This is certainly not the complete description of our universe, and some modification of Hawking radiation is expected. The effect of breaking the local Lorentz symmetry has been explored in some detail, and Schutzhold and Unruh give the best understanding to date. The present evidence (as does ref. \[TJRev\] above) is that Hawking radiation is essentially a low-energy phenomenon, and should not receive significant corrections. (A)
13. \[Helfer\] [ “Do black holes radiate?,”]{} A. D. Helfer, Rept. Prog. Phys. [**66**]{}, 943-1008, (2003) \[arXiv:gr-qc/0304042\]. As a counterpoint to ref. \[modify\], we include Helfer’s review of arguments that Hawking radiation might receive large corrections from quantum-gravity effects. The reader should note, however, that certain arguments in this work are out of date in view of ref. \[TJRev\] and ref. \[modify\]. (A)
14. \[higher\] [ “Black hole entropy is the Noether charge,”]{} R. M. Wald, Phys. Rev. D [**48**]{}, 3427-3431 (1993) \[arXiv:gr-qc/9307038\]. [ “Some properties of Noether charge and a proposal for dynamical black hole entropy,”]{} V. Iyer and R. M. Wald, Phys. Rev. D [**50**]{}, 846-864 (1994) \[arXiv:gr-qc/9403028\]. [ “A Comparison of Noether charge and Euclidean methods for computing the entropy of stationary black holes,]{} Phys. Rev. D [**52**]{}, 4430-4439 (1995) \[arXiv:gr-qc/9503052\]. These works generalize the first law of black-hole thermodynamics to theories more complicated than just Einstein gravity plus matter. Higher derivative theories are included. (A)
Black Holes in Dimensions $d \neq 4$ {#dnot4}
------------------------------------
1. \[BTZ\] [ “The Black hole in three-dimensional space-time,”]{} M. Banados, C. Teitelboim and J. Zanelli, Phys. Rev. Lett. [**69**]{}, 1849-1851 (1992) \[arXiv:hep-th/9204099\]; [ “Geometry of the (2+1) black hole,”]{} M. Banados, M. Henneaux, C. Teitelboim and J. Zanelli, Phys. Rev. D [**48**]{}, 1506-1525 (1993) \[arXiv:gr-qc/9302012\]. In three dimensions, black holes arise only in the presence of a negative cosmological constant ($\Lambda < 0$). These so-called BTZ black holes are particularly simple, as they are quotients of 2+1 Anti-de Sitter space. The above works introduce such solutions and study their properties. (I/A)
2. \[2d\] [ “Dilaton gravity in two dimensions,”]{} D. Grumiller, W. Kummer and D. V. Vassilevich, Phys. Rept. [**369**]{}, 327-430 (2002) \[arXiv:hep-th/0204253\]. Many calculations simplify in 1+1 dimensions. While there is no meaningful theory of Einstein gravity in 1+1 dimensions, the so-called dilaton gravity theories capture some of the same physics and can be related to Einstein gravity in higher dimensions. This work reviews such models, with emphasis on black-hole solutions, Hawking radiation, and quantum effects. (A)
3. \[ER1\] [ “Black Holes in Higher Dimensions,”]{} R. Emparan and H. S. Reall, arXiv:0801.3471 \[hep-th\]. To appear in Living Reviews in Relativity. The space of black-hole solutions and behaviors becomes much richer in $d > 4$ dimensions where one finds black strings, black branes, and black rings in addition to the natural generalizations of 3+1 black holes. New behaviors arise as well, as some black objects are unstable and appear to lead to violations of certain cosmic censorship conjectures. Emparan and Reall review these solutions, the techniques used to construct them, their thermodynamics, and what is known about their dynamics. See also ref. \[instab\]. The section on open problems may be of particular interest for the motivated student. (A)
4. \[ERRings\] [ “Black rings,”]{} R. Emparan and H. S. Reall, Class. Quant. Grav. [**23**]{}, R169-R197 (2006) \[arXiv:hep-th/0608012\]. This more-focused review addresses both the classical physics and microscopic string theory of black rings. (A)
5. \[instab\] [ “Instabilities of black strings and branes,"]{} T. Harmark, V. Niarchos and N. A. Obers, Class. Quant. Grav. [**24**]{}, R1-R90 (2007) \[arXiv:hep-th/0701022\]. While black holes in 3+1 dimensions are typically stable, black objects in higher dimensions exhibit interesting instabilities that may lead to violations of at least certain forms of the Cosmic Censorship Hypothesis. This work reviews the classic Gregory-Laflamme instability of 4+1 black strings as well as many generalizations. The correlated stability conjecture and its limitations are discussed. As in ref. \[ER1\], the section on open problems may be of particular interest for the motivated student. (A)
6. \[KolReview\] [ “The phase transition between caged black holes and black strings: A review,”]{} B. Kol, Phys. Rept. [**422**]{}, 119-165 (2006) \[arXiv:hep-th/0411240\]. In $d> 4$ dimensions, which sort of black-hole solution is most stable can depend on the size of the box or “cage” in which it is placed. Large boxes favor black holes, while small boxes favor black strings or branes. Kol reviews the complex structure associated with the associated phase transitions. (A)
7. [ “Black holes and solitons in string theory,”]{} D. Youm, Phys. Rept, [**316**]{}, 1-232 (1999) \[arXiv:hep-th/9710046\]. An encyclopedic review of black-hole and black-brane solutions in string theory as of 1997. (A)
8. \[stelle\] [ “BPS branes in supergravity,”]{} K. S. Stelle, arXiv:hep-th/9803116, in [**1997 Summer School in High Energy Physics and Cosmology**]{}, edited by E. Gava [*et al.*]{} (World Scientific, New Jersey, 1998), pp. 29-127. Stelle introduces brane solutions in supergravity, discussing super-symmetry, Kaluza-Klein reduction, and low-velocity scattering of branes. (A)
9. \[marolf\] [ “String/M-branes for relativists,”]{} D. Marolf, arXiv:gr-qc/9908045. A brief introduction to stringy branes, intended to convey some basic aspects of brane physics and perspectives on string theory to those trained in GR. (E/I/A)
10. \[TeV1\] [ “Black holes at future colliders and beyond: A review,”]{} G. L. Landsberg, arXiv:hep-ph/0211043. If our universe has more than four dimensions, then it may be possible to produce black holes at particle colliders, or they may be produced naturally in our atmosphere owing to the impacts of cosmic rays. Landsberg reviews the status of these ideas as of 2002, including the signatures of such events that could be used to detect them. (A)
11. \[TeV2\] [ “High-energy black hole production,”]{} S. B. Giddings, AIP Conf. Proc. [**957**]{}, 69-78 (2007) \[arXiv:0709.1107 \[hep-ph\]\]. A brief but more recent review of the possible production and detection of TeV-scale black holes. Includes a discussion of open problems. (A)
12. \[TeV3\] [ “Black holes in theories with large extra dimensions: A review,”]{} P. Kanti, Int. J. Mod. Phys. A [**19**]{}, 4899-4951 (2004) \[arXiv:hep-ph/0402168\]. A more thorough review of mini black holes in models with large extra dimensions, focussing on their decay processes. (A)
13. [ “Braneworld black holes in cosmology and astrophysics,” ]{} A. S. Majumdar and N. Mukherjee, Int. J. Mod. Phys. D [**14**]{}, 1095-1129 (2005) \[arXiv:astro-ph/0503473\]. A review of larger black holes in models with large extra dimensions and their astrophysical implications. (A)
Black-Hole Microphysics {#micro}
=======================
The resources below address the deep microphysics of black holes. The reader should be aware that most of these works make certain assumptions about the quantum nature of gravity and spacetime that are beyond the reach of current experiments. We have attempted to sort this material on the basis of these assumptions into string theory (section \[strings\]), loop quantum gravity (section \[loop\]), and other (section \[other\]). Section \[other\] includes not only other approaches to quantum gravity, but also discussions of entropy bounds, the proposed holographic principle, black-hole complementarity, and Jacobson’s work on the Einstein equation of state ref. \[EEOS\].
Certain overlaps and ambiguities are of course unavoidable. The string-theory material is closely related to that of sections \[dnot4\] (higher dimensions) and \[last\] and has at least historical connections to some material in section \[other\], while the loop gravity work below is intimately tied to the treatments of isolated and dynamical horizons refs. \[IB\],\[IH1\], and \[AK\] (which we have chosen to list again as ref. \[AK2\]). There is also much overlap between sections \[other\] and \[thermo\].
Black-Hole Microphysics in String Theory
----------------------------------------
1. [**String Theory, Vols. 1 & 2**]{}, J. Polchinski (Cambridge University Press,Cambridge, 1998). [**D-branes**]{}, C. Johnson (Cambridge University Press, Cambridge, 2003). [**String Theory and M-Theory: A Modern Introduction**]{}, K. Becker, M. Becker, and J. Schwarz (Cambridge University Press, Cambridge, 2007). [**String theory in a Nutshell**]{}, E. Kiritsis (Princeton University Press, Princeton, 2007). These standard string-theory textbooks each include at least a chapter on black holes in string theory. They are designed for readers with a working knowledge of quantum-field theory and gauge theories in addition to the usual undergraduate physics material. (A)
2. [ “Resource letter: The nature and status of string theory,”]{} D. Marolf, Am. J. Phys. [**72**]{}, 730-741 (2004) \[arXiv:hep-th/0311044\]. This Resource Letter contains a more comprehensive guide to the string theory literature, including the pre-2004 literature on stringy black holes. (A)
3. [ “TASI lectures on black holes in string theory,”]{} A. W. Peet, arXiv:hep-th/0008241, in [**Strings,Branes, and Gravity: TASI 99: Boulder, Colorado, 31 May - 25 June 1999**]{}, edited by J. Harvey, S. Kachru, and E. Silverstein (World Scientific, Singapore, 2001), pp. 353-433. Peet’s thorough introduction to black holes and branes takes the reader up though a stringy treatment of black-hole entropy and Hawking radiation. (A)
4. [ “The Quantum Physics Of Black Holes: Results From String Theory,”]{} S. R. Das and S. D. Mathur, Ann. Rev. Nucl. Part. Sci. [**50**]{}, 153-206 (2000) \[arXiv:gr-qc/0105063\]. This review takes the reader as directly as possible to the stringy description of black-hole entropy and Hawking radiation. (A)
5. [ “Microscopic formulation of black holes in string theory,”]{} J. R. David, G. Mandal, and S. R. Wadia, Phys. Rept, [**369**]{}, 549-686 (2002) \[arXiv:hep-th/0203048\]. The authors review the calculation of black-hole properties from string theory for the case of the so-called nearly-extreme D1-D5 black hole. Many important details of the relevant gauge and conformal field theories are discussed, leading to black-hole thermodynamics and Hawking radiation. (A)
6. [ “Black holes in string theory,”]{} J. M. Maldacena, arXiv:hep-th/9607235. Maldacena’s Ph.D. thesis was not written as an introduction for outsiders, but does contain detailed treatments of gauge-theory aspects relevant to the stringy counting of black-hole entropy that are hard to find in other sources. Other useful aspects of brane and black-hole physics are also described. (A)
7. [ “The fuzzball proposal for black holes: An elementary review,”]{} S. D. Mathur, Fortsch. Phys. [**53**]{}, 793-837 (2005) \[arXiv:hep-th/0502050\]. Mathur reviews a particular set of speculations about black-hole microphysics in string theory. This proposal is by no means established, but has attracted significant interest. (A)
Black-Hole Microphysics in Loop Quantum Gravity {#loop}
-----------------------------------------------
1. \[Ann\] [ “Interface of general relativity, quantum physics and statistical mechanics: Some recent developments,”]{} A. Ashtekar, Annalen Phys. [**9**]{}, 178-198 (2000) \[arXiv:gr-qc/9910101\]. Ashtekar provides an introductory overview of both isolated and dynamical horizons and black-hole entropy in loop quantum gravity. This work attempts to be as nontechnical as possible, and makes a good first introduction to the subject. (I/A)
2. \[LBH\] [ “Quantum geometry and black-hole entropy,”]{} A. Ashtekar, J. Baez, A. Corichi and K. Krasnov, Phys. Rev. Lett. [**80**]{}, 904-907 (1998) \[arXiv:gr-qc/9710007\]. Microstates of black holes are counted in loop quantum gravity. Their number agrees with the Bekenstein-Hawking entropy up to issues associated with the so-called Immirzi parameter. (A)
3. [ “Quantum horizons and black-hole entropy: Inclusion of distortion and rotation,”]{} A. Ashtekar, J. Engle and C. Van Den Broeck, Class. Quant. Grav. [**22**]{}, L27-L34 (2005) \[arXiv:gr-qc/0412003\]. The state counting of ref. \[LBH\] is generalized to include distorted and rotating black holes. (A)
4. \[AK2\] [ “Isolated and Dynamical Horizons and Their Applications,”]{} A. Ashtekar and B. Krishnan, Living Reviews in Relativity, lrr-2004-10. relativity.livingreviews.org/Articles/lrr-2004-10/ Ashtekar and Krishnan review the loop quantum gravity description of black holes and provide a thorough discussion of background material. (A)
Other Microphysics and Black Hole Entropy {#other}
-----------------------------------------
1. \[carlip\] [ “Black Hole Thermodynamics from Euclidean Horizon Constraints” ]{}, S. Carlip, Phys. Rev. Lett. [**99**]{}, 021301 4pp. (2007) \[arXiv:gr-qc/0702107\]. Carlip investigates whether the Bekenstein-Hawking entropy $A/4\ell_p^2$ might be determined by subtle symmetries of black holes, independent of the detailed microscopic theory of quantum gravity. (A)
2. [ “Black Hole Entropy as Causal Links,”]{} D. Dou and R. D. Sorkin, Found. Phys. [**33**]{}, 279-296 (2003) \[arXiv:gr-qc/0302009\]. Within the causal-set approach to quantum gravity, Dou and Sorkin show that the number of causal links crossing the horizon is proportional to the black-hole area. (A)
3. \[entangle\] [ “The Entropy Of The Vacuum Outside A Horizon,”]{} R. D. Sorkin, Gen. Rel. Grav., Proceedings of the GR10 Conference, Padova 1983, edited by B. Bertotti, F. de Felice, A. Pascolini (Consiglio Nazionale della Ricerche, Roma, 1983) Vol. 2; [ “A Quantum Source Of Entropy For Black Holes,”]{} L. Bombelli, R. K. Koul, J. H. Lee and R. D. Sorkin, Phys. Rev. D [**34,**]{} 373-383 (1986); [ “Entropy and area,”]{} M. Srednicki, Phys. Rev. Lett. [**71,**]{} 666-669 (1993) \[arXiv:hep-th/9303048\]. These works introduce the idea that black-hole entropy might fundamentally describe correlations in the vacuum across the horizon of the black hole. The corresponding entropy diverges in quantum field theory but can be made to agree with the Bekenstein-Hawking entropy $A/4 \ell_p^2$ by imposing a cut-off at the Planck length. (A)
4. \[atmosphere\] [ “On The Quantum Structure Of A Black Hole,”]{} G. ’t Hooft, Nucl. Phys. B [**256**]{}, 727-745 (1985). A calculation much like that of ref. \[entangle\] is presented, but from a different perspective. In this so-called brick-wall model, the entropy of a black hole is associated with states of quantum fields outside of, but close to the horizon. (A)
5. \[Centangle\] [ “Black hole horizon fluctuations,”]{} A. Casher, F. Englert, N. Itzhaki, S. Massar and R. Parentani, Nucl. Phys. B [**484**]{}, 419-434 (1997) \[arXiv:hep-th/9606106\]. [ “How wrinkled is the surface of a black hole?,”]{} R. D. Sorkin, in [**Proceedings of the First Australasian Conference on General Relativity and Gravitation**]{}, edited by D. Wiltshire, (University of Adelaide, Adelaide, 1996) pp. 163-174. arXiv:gr-qc/9701056. [ “On the quantum width of a black hole horizon,” ]{} D. Marolf, Springer Proc. Phys. [**98**]{}, 99-112 (2005) \[arXiv:hep-th/0312059\]. We include these works as a counterpoint to \# \[entangle\], \#\[atmosphere\]. They argue that a natural cut-off occurs at a larger length scale, so that the vacuum entanglement entropy is negligible compared with $A/4\ell_p^2$. (A)
6. \[BS\] [ “TASI lectures on the holographic principle,”]{} D. Bigatti and L. Susskind, arXiv:hep-th/0002044, in [**Strings,Branes, and Gravity: TASI 99: Boulder, Colorado, 31 May - 25 June 1999**]{}, edited by J. Harvey, S. Kachru, and E. Silverstein (World Scientific, Singapore, 2001), pp. 883-933. This surprisingly nontechnical work reviews a number of interesting but controversial ideas about black-hole microphysics ranging from black-hole complementarity to entropy bounds and the holographic principle: the assertion that the number of states of a region of space is determined by the area of its boundary. The more established anti-de Sitter/conformal field theory correspondence is discussed as an example of these ideas. (I)
7. \[RB\] [ “The holographic principle,”]{} R. Bousso, Rev. Mod. Phys. [**74**]{}, 825-874 (2002) \[arXiv:hep-th/0203101\]. Bousso provides a thorough review of entropy bounds, and related issues. Applications and examples are presented and some relevant black-hole physics is reviewed. Much of this discussion is accessible to nonexperts. (I)
8. \[CHolog\] [ “Acceleration Radiation And Generalized Second Law Of Thermodynamics,”]{} W. G. Unruh and R. M. Wald, Phys. Rev. D [**25**]{}, 942-958 (1982). [ “Entropy Bounds, Acceleration Radiation, And The Generalized Second Law,”]{} W. G. Unruh and R. M. Wald, Phys. Rev. D [**27**]{}, 2271-2276 (1983). [ “On the status of highly entropic objects,”]{} D. Marolf and R. D. Sorkin, Phys. Rev. D [**69**]{}, 024014 5pp. (2004) \[arXiv:hep-th/0309218\]. [ “Notes on spacetime thermodynamics and the observer-dependence of entropy,”]{} D. Marolf, D. Minic and S. F. Ross, Phys. Rev. D [**69**]{}, 064006 (2004) \[arXiv:hep-th/0310022\]. We include these works as a counterpoint to some of the arguments in refs. \[BS\] and \[RB\]. See also ref. \[Wald\]. (A)
9. [ “Ultimate physical limits to computation,”]{} S. Lloyd, Nature [**406**]{} 1047-1054 (2000) \[arXiv:quant-ph/9908043\]; [ “Black hole entropy and quantum information,”]{} M. J. Duff and S. Ferrara, Ê arXiv:hep-th/0612036. Ê These works explore connections between black-hole physics and quantum information theory. (A)
10. \[EEOS\] [ “Thermodynamics of space-time: The Einstein equation of state,”]{} T. Jacobson, Phys. Rev. Lett. [**75**]{}, 1260-1263 (1995) \[arXiv:gr-qc/9504004\]. All horizons in general relativity share the basic features of black-hole thermodynamics. This intriguing paper shows that the usual derivations can be reversed: by assuming that the first law holds for all horizons in its standard form, one can derive the full dynamics of Einstein gravity. (A)
Connections to Nuclear and Condensed Matter Physics {#last}
===================================================
This brief section addresses connections between black-hole physics and that of the at first sight unrelated fields of nuclear and condensed-matter physics. The resources below are of two types. First, ref. \[analogue\] reviews the use of established hydrodynamics and condensed matter physics to design analogues of black holes in certain fluid systems and Bose-Einstein condensates. In contrast, the remaining works attempt to use black holes and a bit of string theory as mathematical tools to understand quark-gluon plasmas (ref. \[RHIC\]) and condensed-matter physics (ref. \[CM\]). It will be very interesting to see how these ideas develop in the near future.
1. \[analogue\] [ “Analogue gravity,”]{} C. Barcelo, S. Liberati and M. Visser, Living Rev. Rel. [**8**]{}, 12, 151pp. (2005) \[arXiv:gr-qc/0505065\]. The authors review the vast literature on analogues of black holes in fluid and condensed matter systems. Both classical properties of black holes and Hawking radiation are discussed. There are interesting connections to ref. \[modify\]. (A)
2. \[RHIC\] [ “String Theory and Quantum Chromodynamics,”]{} D. Mateos, Class. Quant. Grav. [**24**]{}, S713-S739 (2007) \[arXiv:0709.1523 \[hep-th\]\]. The so-called gauge/gravity dualities of string theory imply that certain theories of gravity can be used to calculate properties of what might appear to be completely unconnected quantum-field theories. The best-studied example of this is the anti-de Sitter/conformal field theory correspondence (AdS/CFT). Finite temperature effects in quantum-field theory are associated with black holes in the gravitational theory. Mateos reviews attempts over the last few years to use this correspondence to understand the physics of quark-gluon plasmas such as those currently being generated at the Relativistic Heavy Ion Collider (RHIC) in Brookhaven, New York. Some background in gauge theory and string theory is required. (A)
3. \[CM\] [ “Quantum critical transport, duality, and M-theory,”]{} C. P. Herzog, P. Kovtun, S. Sachdev and D. T. Son, Phys. Rev. D [**75**]{}, 085020, 21pp. (2007) \[arXiv:hep-th/0701036\]; [ “Theory of the Nernst effect near quantum phase transitions in condensed matter, and in dyonic black holes,”]{} S. A. Hartnoll, P. K. Kovtun, M. Muller and S. Sachdev, Phys. Rev. B [**76**]{}, 144502, 17pp. (2007) \[arXiv:0706.3215 \[cond-mat.str-el\]\]. Strongly coupled quantum effects are believed to play an important role in certain condensed-matter systems, such as high-temperature superconductors. Near a phase transition these effects should be described by a strongly coupled conformal field theory. As such, they may be amenable to study via the AdS/CFT correspondence. These papers begin what will surely be a long process of attempting to do so. (A)
$~$\
The authors thank Amitabh Virmani, Steve Giddings and Marta Volonteri for help in locating certain references. We also thank Jessica Wirts for assistance in tracking down certain bibliographic information. E.G. is funded by NASA through a Hubble Fellowship grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Incorporated, under NASA contract NAS5-26555. D.M. was supported in part by the US National Science Foundation under Grant No. PHY05-55669, and by funds from the University of California.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Taro <span style="font-variant:small-caps;">Kimura</span>$^{1}$and Tatsuhiro <span style="font-variant:small-caps;">Misumi</span>$^{2}$'
title:
- Characters of Lattice Fermions Based on the Hyperdiamond Lattice
- Characters of Lattice Fermions Based on the Hyperdiamond Lattice
---
Introduction
============
Recently, Creutz[@Creutz:2007af] and Boriçi[@Borici:2007kz] have proposed a two-parameter class of fermion actions called “Creutz fermion”, inspired by the relativistic condensed matter system, graphene[@neto:109]. This fermion is defined on the hyperdiamond lattice distorted by two parameters ($B, C$) and includes the non-nearest hopping terms. What is notable about “Creutz fermion” is that it has desirable properties for the lattice simulation such as locality, chiral symmetry and the minimal number of fermion doubling. Among them, the minimal fermion doubling is the most outstanding characteristic of this fermion. As is well-known, although there exist only two (or three) light quarks in QCD, Nielsen-Ninomiya’s no-go theorem[@Nielsen:1980rz; @Nielsen:1981xu; @Nielsen:1981hk] states that the lattice fermion with chiral symmetry and other common features inevitably yields degrees of freedom of multiple number of two in a continuum limit. On the other hand, the lattice fermions which bypass the no-go theorem such as domain-wall fermion[@Kaplan:1992bt; @Furman:1994ky] and overlap fermion[@Ginsparg:1981bj; @Neuberger:1998wv] demand an expensive numerical task. Therefore the chiral-symmetric fermion including only the minimal number of doubling, such as Creutz fermion, will be much faster and more useful in the simulation since the two fermion degrees of freedom can be directly interpreted as the two light quarks in lattice QCD simulation.
However it was pointed out [@Bedaque:2008xs] that Creutz action lacks sufficient discrete symmetry to prohibit relevant and marginal operators to be generated through the loop corrections which are serious obstacles for a good continuum limit in the lattice simulation [@Capitani:2009yn; @Capitani:2010nn]. In particular, such a problem for the lattice action, which is equivalent to Boriçi action, has already investigated[@Alonso:1987bb]. In Ref. the authors show that if the non-nearest hopping terms are dropped with the parameters chosen to $B=1/\sqrt{5}$, $C=1$ in Creutz action, the requisite discrete symmetry of cyclic group $\mathbb{Z}_{5}$ recovers, although the modified action yields an unphysical excitation from the pole of propagator (or mutilated pole [@Celmaster:1982ht; @Celmaster:1983jq; @Drouffe:1983kq]). They also construct a simple fermion action on a hyperdiamond lattice including only the nearest-neighbor hoppings. However, it is argued in that although this fermion action has the sufficient discrete symmetry of alternating group $\mathfrak{A}_{5}\supset \mathbb{Z}_{5}$, it yields more than minimal number of doublers.
In this paper we investigate fermion actions on hyperdiamond and deformed-hyperdiamond lattices, with emphasis on the real-space construction of them and Lorentz-covariant excitations from poles of propagators, then obtain a generalized class of Creutz-type minimal-doubling actions on a deformed hypercubic lattice. Firstly we propose the spatial construction of Creutz fermion action on a deformed hyperdiamond lattice, which is an improved version of that in . Secondly we study conditions for a hyperdiamond-lattice action to produce Lorentz-covariant excitations from fermion poles. It is pointed out that the non-nearest neighbor hoppings in terms of sites are essential for the correct excitations. Then we propose a class of minimal-doubling fermion actions defined on a deformed hypercubic (rhombus) lattice as a generalization of Creutz-type actions, where the link variables are easily introduced. We also introduce a two-parameter class of Wilczek-type minimal-doubling actions, which will be the simplest form of Creutz-type lattice action.
In Sec. \[sec:Creutz\] we briefly review the hyperdiamond lattice and Creutz fermion. In Sec. \[sec:hyperdiamond\] we discuss the spatial construction of Creutz action and investigate the conditions for the correct fermionic excitations. In Sec. \[sec:examples\] we study examples of hyperdiamond-lattice fermions and propose a related action “Appended Creutz action”. In Sec. \[sec:rhombus\] we generalize Creutz-type and Wilczek-type actions to classes of actions on deformed hypercubic lattices. Section \[sec:summary\] is devoted to a summary and discussion.
Creutz fermion {#sec:Creutz}
==============
We now consider the minimal-doubling action proposed by Creutz, which is called Creutz action[@Creutz:2007af; @Borici:2007kz] and also Boriçi-Creutz action in . The action is related to the hyperdiamond lattice, which is the higher dimensional generalization of graphene system. General dimensional aspects of the hyperdiamond lattice and lattice fermions defined on them are discussed in .
The four-dimensional hyperdiamond lattice is constructed with five bond vectors, $$\begin{array}{c}
\bm{e}^1=\frac{1}{4}(\sqrt{5},\sqrt{5},\sqrt{5},1), \quad
\bm{e}^2=\frac{1}{4}(\sqrt{5},-\sqrt{5},-\sqrt{5},1), \\
\bm{e}^3=\frac{1}{4}(-\sqrt{5},-\sqrt{5},\sqrt{5},1), \quad
\bm{e}^4=\frac{1}{4}(-\sqrt{5},\sqrt{5},-\sqrt{5},1), \\
\bm{e}^5=(0,0,0,-1), \label{hyperdiamond_bond}
\end{array}$$ satisfying $$\bm{e}^\mu \cdot \bm{e}^\nu = \left\{ \begin{array}{ccc}
1 & \mbox{for} & \mu =
\nu \\
\cos \theta = -1/4 & \mbox{for} & \mu \not= \nu \\
\end{array} \right..$$ The spatial translation symmetry of the hyperdiamond lattice is characterized by primitive vectors $\bm{d}_\mu\,\,(\mu=1,2,3,4)$ defined as $$\bm{d}_\mu = \bm{e}^\mu - \bm{e}^5 \quad \mbox{for} \quad \mu = 1,
\cdots, 4. \label{primitive01}$$ Then an angle $\eta$ between them is given by $$\cos \eta =
\frac{\bm{d}_\mu\cdot\bm{d}_\nu}{\left|\bm{d}_\mu\right|\left|\bm{d}_\nu\right|}
= \frac{1}{2}. \label{diamond_angular}$$ This is a common property for any dimensional hyperdiamond lattices[@KM:2009lf].
An important property of the hyperdiamond lattice is its sublattice structure, such that it consists of two kinds of sites. These sublattices are called $L$-node and $R$-node[^1], whose positions are labeled by $x_L = \sum_{\mu=1}^4 x_\mu \bm{d}_\mu$ and $x_R
= \sum_{\mu=1}^4 x_\mu \bm{d}_\mu + \bm{e}^5$, but this node-index is often omitted in the following discussion because only positions of unit cells are important for a lattice action. This sublattice structure corresponds to chirality of fermions, and is in common with the honeycomb lattice $(d=2)$ and the diamond lattice $(d=3)$. That means, if poles of the Dirac operator are arranged to construct the hyperdiamond lattice in momentum space, the number of doublers is only two. This is a strategy to obtain the hyperdiamond-type minimal-doubling fermion, and actually done in Creutz’s original paper[@Creutz:2007af].
Then let us show how four dimensional generalization of the graphene system is considered. To investigate the graphene system, we often use the tight-binding model on a honeycomb lattice for $\pi$-electrons of carbon atoms. The Hamiltonian in momentum space is given by $$H(p) = K \left(\begin{array}{cc}
0 & z(p) \\ z^*(p) & 0
\end{array}\right)$$ where $K$ is a hopping amplitude and an off-diagonal component is defined as $z(p)=1+e^{ip_1}+e^{ip_2}$. This Hamiltonian is associated with a conventional anti-hermitian Dirac operator in lattice field theory as $H=\sigma_3 D$, and we now use non-orthogonal coordinates defined by primitive vectors, $p_\mu =
\bm{d}_\mu\cdot p$.
To consider four-component Dirac spinor in four dimensions, a complex number $z(p) \in \mathbb{C}$ is generalized to a quaternion $z = c_0+i\vec{c}\cdot\vec\sigma \in \mathbb{H}$. In the original paper[@Creutz:2007af], the associated Dirac operator is given by $$\begin{aligned}
D(p)\ =
&& (\sin p_1 + \sin p_2 - \sin p_3 - \sin p_4) i \gamma_1 \nonumber \\
&+& (\sin p_1 - \sin p_2 - \sin p_3 + \sin p_4) i \gamma_2 \nonumber \\
&+& (\sin p_1 - \sin p_2 + \sin p_3 - \sin p_4) i \gamma_3 \nonumber \\
&+& B (4C - \cos p_1 - \cos p_2 - \cos p_3 - \cos p_4) i \gamma_4.
\label{Creutz_Dirac}\end{aligned}$$ Gamma matrices are defined as $\gamma_i = \tau_1 \otimes
\sigma_i$, $\gamma_4 = \tau_2 \otimes {{\mathbbm 1}}$ and $\gamma_5 = \tau_3
\otimes {{\mathbbm 1}}$ where $\tau$’s and $\sigma$’s are both Pauli matrices acting on the sublattice and internal spinor structure, respectively. Here, as the graphene system, we use non-orthogonal coordinates, $p_\mu = \bm{a}_\mu\cdot p$, where $\bm{a}_\mu$ are primitive vectors of the lattice for Creutz action. However in Sec. \[sec:hyperdiamond\] we will show they are different from those defined in (\[primitive01\]).
This Dirac operator possesses two poles at $p=\pm(\tilde p, \tilde p, \tilde
p, \tilde p)$ with $\cos \tilde p =C$, if and only if the lattice parameter $C$ satisfying $1/2 < C < 1$, to suppress extra poles such as $(\tilde p, \tilde p, \tilde p, \pi - \tilde p)$. Although these two poles induce physical Dirac fermions, in the case $C=1$ they are reduced to only one cut rather than a pole, and thus it turns out to be unphysical. To show this, we expand the Dirac operator (\[Creutz\_Dirac\]) around the pole as $p_\mu = \tilde p + q_\mu$, $$\begin{aligned}
D(p)\ = && C (q_1 + q_2 - q_3 - q_4) i \gamma_1 \nonumber \\
&+& C (q_1 - q_2 - q_3 + q_4) i \gamma_2 \nonumber \\
&+& C (q_1 - q_2 + q_3 - q_4) i \gamma_3 \nonumber \\
&+& BS (q_1 + q_2 + q_3 + q_4) i \gamma_4 + {\cal O}(q^2)
\label{Creutz_Dirac02}\end{aligned}$$ with $S = \sin \tilde p$. This Dirac operator behaves as $i \vec{\gamma} \cdot \vec{k}$ around $p_\mu = 0$ with $S=0\ (C=1)$, and thus it has been shown that covariance of this fermion is broken. This unphysical fermion is known as a mutilated fermion, often found in some attempts on nonhypercubic lattices[@Celmaster:1982ht; @Celmaster:1983jq; @Drouffe:1983kq].
Since gamma matrices satisfy anti-commutation relations $\{\gamma_\mu,
\gamma_\nu\}=2\delta_{\mu\nu}$, coefficients of gamma matrices are interpreted as those of a momentum represented by Euclidean coordinates. Thus reciprocal vectors $\{\bm{b}^\mu\}$ giving momentum space basis are obtained by (\[Creutz\_Dirac02\]), $$\begin{array}{c}
\bm{b}^1=(C,C,C,BS), \quad
\bm{b}^2=(C,-C,-C,BS), \\
\bm{b}^3=(-C,-C,C,BS), \quad
\bm{b}^4=(-C,C,-C,BS).
\end{array} \label{reciprocal01}$$ These vectors characterize the translation structure in momentum space, thus they can be interpreted as “primitive vectors” of momentum space.
To consider a situation such that poles construct the exact hyperdiamond lattice, as primitive vectors (\[diamond\_angular\]), it is imposed in Ref. that an angle $\xi$ between reciprocal vectors satisfies $\cos \xi = 1/2$. Here it is given by $$\cos \xi =
\frac{\bm{b}^\mu\cdot\bm{b}^\nu}{|\bm{b}^\mu||\bm{b}^\nu|}
= \frac{B^2S^2 - C^2}{B^2S^2 + 3 C^2}.
\label{reciprocal_angle}$$ With adjusting lengths of reciprocal vectors, Creutz chose two parameters, $C=\cos (\pi/5)$ and $B=\sqrt{5}\cot (\pi/5)$. On the other hand, the orthogonal condition $\bm{b}^\mu\cdot\bm{b}^\nu=0$ gives $BS=C$ applied in Ref. .
We now remark an important property of reciprocal vectors to consider the lattice structure in real space. Because an arbitrary momentum is represented by $p = \sum_{\mu=1}^4
\left(p\cdot\bm{a}_\mu\right) \bm{b}^\mu$, associated primitive vectors are determined by a relation $\bm{a}_\mu\cdot\bm{b}^\nu=\delta_\mu^{\nu}$. We will discuss real-spatial construction of Creutz fermion with this relation in Sec. \[sec:hyperdiamond\].
Hyperdiamond lattice fermion {#sec:hyperdiamond}
============================
As discussed in the previous section, Creutz fermion was directly constructed in momentum space. Then we should consider its lattice construction in real space to reproduce Creutz’s Dirac operator (\[Creutz\_Dirac\]). The translation structure in momentum space is given by the expanded Dirac operator (\[Creutz\_Dirac02\]), and provides the associated real-spatial lattice structure.
In this section we introduce a real-spatial construction of Creutz fermion on the deformed hyperdiamond lattice. Then we show that the bond vectors of the lattice and the hopping vectors of the fermion fields should be determined by the reciprocal vectors Eq. (\[reciprocal01\]) consistently. In this sense our construction is more reasonable than that in although they seem similar to each other. Based on this construction, we will give a parameter condition for Creutz fermion to be defined on the exact hyperdiamond lattice in real space. We will also discuss conditions for hyperdiamond-lattice fermions to yield only physical or Lorentz-covariant excitations.
Creutz fermion on hyperdiamond lattice {#sec:hyperdiamond_Creutz}
--------------------------------------
To obtain a four-component Dirac fermion on the four dimensional hyperdiamond lattice, we consider two-component chiral spinors, left-handed $\phi$ and right-handed $\bar{\phi}$ located on $L$-node, and right-handed $\chi$ and left-handed $\bar{\chi}$ located on $R$-node. Here we note that $\phi$ and $\bar{\phi}$ are not hermite conjugate, but independent degrees of freedom. This configuration of the lattice fermions will play an important role on the discussion in section \[sec:features\].
Here we define “spinor vectors”, which appear as the coefficient vectors of the gamma matrices in the action. The spinor vectors with parameters $B$ and $C$ are given by $$\begin{array}{c}
\bm{s}^1 = (1,1,1,B) , \quad \bm{s}^2 = (1,-1,-1,B) , \\
\bm{s}^3 = (-1,-1,1,B) , \quad \bm{s}^4 = (-1,1,-1,B) , \\
\bm{s}^5 = (0,0,0,-4BC). \label{twisted_bond}
\end{array}$$ The action is given by $$\begin{aligned}
S_{\mathrm{C}}
& = & \frac{1}{2} \sum_x \Big[ \sum_{\mu=1}^4 \left(
\bar{\phi}_{x-\bm{a}_\mu} \Sigma \cdot \bm{s}^\mu \chi_x
- \bar{\chi}_{x+\bm{a}_\mu} \Sigma \cdot \bm{s}^\mu \phi_x
\right) \nonumber \\
&& + \bar{\phi}_x \Sigma \cdot \bm{s}^5 \chi_x
- \bar{\chi}_x \Sigma \cdot \bm{s}^5 \phi_x \nonumber \\
&& + \sum_{\mu=1}^4 \left(
\bar{\chi}_{x-\bm{a}_\mu} \bar \Sigma \cdot \bm{s}^\mu \phi_x
- \bar{\phi}_{x+\bm{a}_\mu} \bar \Sigma \cdot \bm{s}^\mu \chi_x
\right) \nonumber \\
&& + \bar{\chi}_x \bar\Sigma \cdot \bm{s}^5 \phi_x
- \bar{\phi}_x \bar\Sigma \cdot \bm{s}^5 \chi_x \Big]. \label{Creutz_action}\end{aligned}$$ where the hopping vectors $\bm{a}_{\mu}$ (or primitive vectors) will be determined to be consistent with the reciprocal vectors in the momentum space afterward. Here the fourth components of spinor matrices are twisted as $\Sigma=(\vec{\sigma},-1)$, $\bar{\Sigma}=(\vec{\sigma},1)$. It is notable that hopping terms of $\phi_x \to \bar\chi_{x-\bm{a}_\mu}$ and $\chi_x \to \bar\phi_{x+\bm{a}_\mu}$ in (\[Creutz\_action\]) represent non-nearest site interactions.
Thus we obtain Creutz’s Dirac operator (\[Creutz\_Dirac\]) by considering Fourier transformation of the lattice action (\[Creutz\_action\]) as $$S_{\mathrm{C}} = \int \frac{d^4 p}{(2\pi)^4}\ \bar{\psi}_p D(p) \psi_p$$ with a Dirac spinor $$\psi_p = \left( \begin{array}{c} \phi_p \\ \chi_p \end{array} \right), \quad
\bar{\psi}_p = \left( \bar{\phi}_p \quad \bar{\chi}_p \right).$$
In this construction the reciprocal vectors are obtained as (\[reciprocal01\]). Since primitive and reciprocal vectors satisfy $\bm{a}_\mu\cdot\bm{b}^\nu=\delta_\mu^{\nu}$, the associated primitive vectors, or the hopping vectors of the fermion fields, for Creutz fermion are given by $$\begin{array}{c}
\bm{a}_1=\frac{1}{4C}(1,1,1,\frac{C}{BS}), \quad
\bm{a}_2=\frac{1}{4C}(1,-1,-1,\frac{C}{BS}), \\
\bm{a}_3=\frac{1}{4C}(-1,-1,1,\frac{C}{BS}), \quad
\bm{a}_4=\frac{1}{4C}(-1,1,-1,\frac{C}{BS}).
\end{array} \label{primitive02}$$ Since primitive vectors of the (distorted) hyperdiamond lattice are given by (\[primitive01\]), bond vectors $\{\bm{e}^\mu\}$ are obtained by introducing another free parameter $f$ ($f>0$), $$\begin{aligned}
\bm{e}^\mu &=& \bm{a}_\mu + \bm{e}^5 \quad \mbox{for} \quad \mu=1,2,3,4,
\nonumber \\
\bm{e}^5 &=& (0,0,0,-f).
\label{twisted_bond02}\end{aligned}$$ These vectors are identified with the bond vectors of the hyperdiamond lattice on which the action is defined, while the spinor vectors $\{\bm{s}^\mu\}$ in the action (\[Creutz\_action\]) are just related to the lattice structure indirectly.
![Two dimensional analogue of the lattice deformation: (a) the two dimensional regular diamond (honeycomb) lattice and (b) the distorted lattice with $\bm{e}^{3}$ direction specified $(\theta''>\theta>\theta')$.[]{data-label="honeycomb_lattice"}](honey_reg.eps){height="2.7cm"}
![Two dimensional analogue of the lattice deformation: (a) the two dimensional regular diamond (honeycomb) lattice and (b) the distorted lattice with $\bm{e}^{3}$ direction specified $(\theta''>\theta>\theta')$.[]{data-label="honeycomb_lattice"}](honey_dis.eps){height="2.7cm"}
For general values of the parameters, the hyperdiamond lattice is deformed such that it is elongated in $\bm{e}^5$ direction as shown in Fig. \[honeycomb\_lattice\]. As the case of the exact hyperdiamond lattice (\[diamond\_angular\]), an angle between the primitive vectors (\[primitive02\]) is given by $$\cos \eta = \frac{C^2 - B^2 S^2}{C^2 + 3B^2S^2}.$$ We note that it is related to (\[reciprocal\_angle\]) by exchanging $C
\leftrightarrow BS$. Then if we choose deformation parameters as $f = 1/(\sqrt{5}C)$ and $BS=C/\sqrt{5}$, the angles become $\cos \eta = 1/2$ and $\cos \theta =
\bm{e}^\mu \cdot \bm{e}^\nu/(|\bm{e}^\mu||\bm{e}^\nu|)= -1/4$, and thus Creutz fermion is defined on the exact hyperdiamond lattice. We will call this condition “hyperdiamond condition”. In this case, the angle between reciprocal vectors becomes $\cos
\xi = -1/4$. Furthermore, the Creutz and Boriçi’s conditions such that in momentum space poles are located on the exact hyperdiamond lattice and on the orthogonal lattice imply the associated angles become $$\cos \eta = \left\{ \begin{array}{cl}
-1/4 & \mbox{(Creutz)} \\ 0 & \mbox{(Bori\c{c}i)}
\end{array}\right. .$$ This means, in the sense of real and momentum spaces, the hyperdiamond condition and the Creutz condition are dual, and the Boriçi condition is self-dual.
Here we remark discrete symmetry of Creutz fermion. With general parameters, this distorted lattice and also the action on this have only $\mathfrak{S}_{4} \not\supset \mathbb{Z}_{5}$ symmetry which is not the sufficient discrete symmetry for a good continuum limit [@Bedaque:2008jm]. In the case of the hyperdiamond condition, this distorted lattice becomes the regular $\mathfrak{S}_{5} \supset \mathbb{Z}_{5}$ symmetric hyperdiamond lattice. However, even if the hyperdiamond lattice becomes exact, the non-nearest hopping terms reduce the discrete symmetry of the action to $\mathfrak{S}_{4}$. Thus the physical Creutz fermion cannot have the requisite discrete symmetry to prohibit the redundant operators. It indicates a general property that the minimal-doubling lattice fermion lacks the sufficient discrete symmetry for a continuum limit on hyperdiamond lattices.
Conditions for physical hyperdiamond lattice fermions {#sec:features}
-----------------------------------------------------
We have seen that physical fermionic excitations are obtained from minimal-doubling poles of Creutz’s Dirac operator, although the action lacks sufficient discrete symmetry. Here we investigate conditions for a hyperdiamond-lattice action to produce Lorentz-covariant excitations from poles of fermion propagators, which actually Creutz fermion satisfies. We then argue “minimal-doubling” on the hyperdiamond lattice is incompatible with the sufficient discrete symmetry of the action for a good continuum limit since the above conditions, which we will call “physicality conditions”, cannot be satisfied without lowering the symmetry of the action.
[*Non-nearest neighbor hoppings.*]{} We have shown that Creutz action can be regarded as being defined on the hyperdiamond lattice. From the viewpoint of this real-spatial interpretation, the action (\[Creutz\_action\]) contains the non-nearest neighbor hoppings which sometimes lead to breaking of locality. However in this case, non-nearest neighbor hoppings are actually [*non-nearest*]{} interactions in the sense of sites, but [*nearest*]{} for unit cells. The hoppings based on the vectors $\bm{a}_{\mu}$ in Eq. (\[primitive02\]) stand for nearest neighbor hoppings between the unit cells, not the sites. Thus the locality is not broken in the continuum limit of Creutz action as seen in . Here we denote the nearest (non-nearest) hoppings in the sense of sites as “nearest-site (non-nearest-site) hoppings”. As follows, we will show the non-nearest-site hoppings are required for Lorentz-covariant or physical excitations of fermions on hyperdiamond lattices although the nearest-site hoppings are enough for a correct form of a Euclidean Dirac operator, namely anti-hermitian Dirac operator.
![$L \to R$ hoppings to nearest unit cells[@KM:2009lf]: (a) forward (nearest neighbor site) hoppings, (b) backward (non-nearest neighbor site) hoppings, and (c) the remaining nearest unit cell hoppings which are not included by Creutz action. $L$-node and $R$-node are denoted by shaded and open circles, respectively. Unit cells are encircled.[]{data-label="honeycomb_nonnearest"}](honey_for.eps "fig:"){height="2.7cm"} ![$L \to R$ hoppings to nearest unit cells[@KM:2009lf]: (a) forward (nearest neighbor site) hoppings, (b) backward (non-nearest neighbor site) hoppings, and (c) the remaining nearest unit cell hoppings which are not included by Creutz action. $L$-node and $R$-node are denoted by shaded and open circles, respectively. Unit cells are encircled.[]{data-label="honeycomb_nonnearest"}](honey_back.eps "fig:"){height="2.7cm"} ![$L \to R$ hoppings to nearest unit cells[@KM:2009lf]: (a) forward (nearest neighbor site) hoppings, (b) backward (non-nearest neighbor site) hoppings, and (c) the remaining nearest unit cell hoppings which are not included by Creutz action. $L$-node and $R$-node are denoted by shaded and open circles, respectively. Unit cells are encircled.[]{data-label="honeycomb_nonnearest"}](honey_near.eps "fig:"){height="2.7cm"}
For nearest-site $L \to R$ hoppings on the hyperdiamond lattice, a forward hopping as $x \to
x+\bm{a}_\mu$ is allowed but a backward hopping $x \to x - \bm{a}_\mu$ is not as shown in Fig. \[honeycomb\_nonnearest\], and they are inverted in the case of $R \to L$ hoppings. Now let us consider a hyperdiamond-lattice action only with nearest-site hoppings as discussed in . Although either of $L \to R$ or $R \to L$ hopping corresponds to a non-hermitian operator as $i(\partial_\mu - 1)$ or $i(1 - \partial_\mu^\dagger)$, we can make the Euclidean Dirac operator anti-hermitian (namely a correct form) by including both of $L \to R$ and $R \to L$ nearest-site hopping terms. However in this case, only either of $e^{ip_\mu}$ or $e^{-ip_\mu}$ appears in the coefficients of spinor matrices such as $\Sigma$ and $\bar{\Sigma}$ in the momentum space. Since these coefficients are complex numbers, we should introduce two anti-hermitian basis, $i\gamma$ and $\gamma\gamma_5$, in order to expand this kind of the operator by gamma matrices. (Here the chirality of the Dirac operator is not broken since both of $i\gamma$ and $\gamma\gamma_5$ anticommute with $\gamma_5$.) Then this kind of the Dirac operator is given by $$D(p)\,=\, \sum_{\mu}i\gamma_{\mu}F_{\mu}(p)+\sum_{\mu}\gamma_{\mu}\gamma_{5}G_{\mu}(p)
\label{generalD}$$ where $F_{\mu}(p)$ and $G_{\mu}(p)$ are independent [*real*]{} functions in general. This type of Dirac operator is actually proposed in , as the generally chiral symmetric operator. In this operator $i\gamma$-terms and $\gamma\gamma_5$-terms are regarded as “vector” and “axial-vector” functions, respectively, and thus Nielsen-Ninomiya’s no-go theorem cannot be applied to this kind of Dirac operator because it is based on Poincaré-Hopf theorem for “either” of vector or axial-vector functions, not for both of them. Therefore, there is no guarantee that the Dirac operator yields physical poles of fermion propagator and the number of poles becomes even. Thus it indicates that the index of the Dirac operator including both of $i\gamma$ and $\gamma\gamma_5$-terms is ill-defined.
Actually the operator including both of $i\gamma$-terms and $\gamma\gamma_5$-terms yields unphysical fermion doublers in general. (Even on hypercubic lattices, the difference operator including only either of forward or backward hoppings induces unphysical poles [@Aoki:2004].) On the other hand, as seen in Creutz action, we can expand the Dirac operator by only $i\gamma$-terms if we introduce hoppings to non-nearest neighbor sites but nearest unit cells. Thus non-nearest-site interactions are required for constructing a physical Dirac operator, which produces only physical degrees of freedom on the hyperdiamond lattice.
Here let us note that hermiticity of the hyperdiamond-lattice Dirac operator without non-nearest hoppings, which is one of the necessary conditions for Nielsen-Ninomiya’s theorem, does not imply the operator can be expanded by only either of “vector” or “axial-vector” functions, as is different from the hypercubic case. Thus we may be able to claim that it is, in a sense, a counter example of Nielsen-Ninomiya’s theorem since the associated Dirac operator includes both of $i\gamma$ and $\gamma\gamma_5$ terms although all the conditions for the theorem are satisfied.
As shown above, non-nearest neighbor interactions are necessary for a physical fermionic mode on hyperdiamond lattices. However this kind of the non-nearest hopping terms lower the discrete symmetry of the action [@Bedaque:2008jm]. Therefore it seems impossible to construct a physical fermion action with the sufficient discrete symmetry on the hyperdiamond lattice as far as it is based on the initial setting of the field configuration as discussed in section \[sec:Creutz\], namely, two kinds of chiral fermions on $L$-nodes and $R$-nodes, respectively. It also means that the requisite discrete symmetry for a good continuum limit is incommensurate with physical minimal-doubling actions on hyperdiamond lattices in this configuration of fermion fields. This no-go property is first conjectured in , and we discuss it systematically as shown above.
[*Twisting spinor structure*]{}. To obtain a physical mode, Dirac operator must be expanded by either of $i\gamma$ or $\gamma\gamma_5$-terms. The most important condition is including non-nearest-site hoppings, but we now show some subsidiary conditions are required for constructing a physical hyperdiamond lattice fermion. To construct a physical Dirac operator, we should multiply some spinor matrices depending on hopping directions. The naive choice is $\sigma =(\vec{\sigma}, i)$ as proposed in . On the other hand, in the case of Creutz action the fourth component of spinor structure is twisted as $\Sigma = (\vec{\sigma}, -1)$, and the same coefficients are applied to non-nearest-site hoppings. Due to this twist, $\sin p$ is converted to $\cos p$ in the coefficient of the corresponding gamma matrix, and thus Creutz’s Dirac operator becomes anti-hermitian by twisting the spinor component.
Although the spinor structure of Creutz action is apparently unnatural, it should be determined in order to construct an anti-hermitian Dirac operator. It is expected that a nontrivial spinor structure is related to an action based on a non-Bravais lattice which possesses sublattice structure, e.g. staggered fermion[@Susskind:1976jm].
[*Lattice deformation*]{}. Besides the above prescriptions to obtain physical modes, we need deformed “spinor vectors” by elongating in one specific direction. As discussed in Sec. \[sec:hyperdiamond\_Creutz\], these spinor vectors cannot be interpreted as bond vectors of the lattice. But we have shown they are related as follows: spinor vectors $\{\bm{s}^\mu\}$ give the Dirac operator and also reciprocal vectors $\{\bm{b}^\mu\}$ characterizing the translation symmetry in momentum space, then primitive vectors $\{\bm{a}_\mu\}$ are derived from the condition $\bm{a}_\mu\cdot\bm{b}^\nu=\delta_\mu^{\nu}$. Thus two-parameter deformation of spinor vectors leads to the distorted hyperdiamond lattice. Although we have also shown that Creutz fermion can be constructed on the exact hyperdiamond lattice, the action yields a cut on $(p_{1}, p_{2},
p_{3})=(0,0,0)$, not a pole when we apply the regular hyperdiamond lattice bond vectors with the condition $C=1$ to spinor vectors. In this sense, the lattice deformation is also necessary for physical poles although the discrete symmetry of the hyperdiamond lattice is broken.
Examples {#sec:examples}
========
According to the three conditions discussed in section \[sec:features\], (i) Non-nearest-site hopping, (ii) Spinor twist, and (iii) Distortion of the lattice, we can consider eight kinds of fermions. In this section we try to complete the classification of lattice fermions based on the hyperdiamond lattice in the aspects of the three features.
BBTW fermion and Dropped Creutz fermion {#sec:BBTW}
---------------------------------------
Firstly let us consider hyperdiamond-lattice actions with the sufficient discrete symmetry for a good continuum limit. One of these fermion actions is proposed in , which we will call BBTW action in this paper. It is constructed with exact hyperdiamond spinor vectors and includes only nearest-site hoppings with untwisted spinor structure vectors $\sigma=(\vec{\sigma}, i)$ and $\bar{\sigma}=(\vec{\sigma}, -i)$ as $$S_{\mathrm{BBTW}} = \sum_x \Big[ \sum_{\mu=1}^4 \left( \bar{\phi}_{x-\bm{a}_\mu} \sigma \cdot \bm{s}^\mu \chi_x - \bar{\chi}_{x+\bm{a}_\mu} \bar{\sigma} \cdot \bm{s}^\mu \phi_x \right) + \bar{\phi}_x \sigma \cdot \bm{s}^5 \chi_x - \bar{\chi}_x \bar{\sigma} \cdot \bm{s}^5 \phi_x \Big],\label{BBTW_action}$$ where spinor vectors $\bm{s}^\mu$ are those in (\[twisted\_bond\]) with $B=1/\sqrt{5}$, $C=1$. The hopping vectors $\bm{a}_{\mu}$ should be obtained from the reciprocal vectors in the momentum space as discussed in Sec.\[sec:hyperdiamond\_Creutz\]. Taking Fourier transformation of the action, we obtain the associated Dirac operator represented as $$D(p) = i \sum_\mu^4 \left( \bm{s}^\mu \cdot \gamma \right) \sin p_\mu - \left( \sum_{\mu=1}^4 \bm{s}^\mu \cos p_\mu + \bm{s}^5 \right) \cdot \gamma \gamma_5 \label{BBTW_Dirac}$$ The Dirac operator of BBTW action has at least seven spectral zeros at $p_\mu = 0$ and $p_1 = - p_2 = - p_3 = p_4 = \cos^{-1}(-2/3)$, etc. Denoting the poles as $p_\mu = \hat{p}_\mu$, the operator is expanded around the pole with momentum $p_\mu = \hat{p}_\mu + q_\mu$, $$D(p) = \sum_{\mu=1}^4 \Big[ i \left( \bm{s}^\mu \cdot \gamma \right) q_\mu \cos \hat{p}_\mu - \left( \bm{s}^\mu \cdot \gamma \gamma_5 \right) q_\mu \sin \hat{p}_\mu \Big] + {\cal O}(q^2). \label{BBTW_Dirac_exp}$$ For the pole at $\hat{p}_\mu = 0$, the Dirac operator (\[BBTW\_Dirac\_exp\]) becomes $$D(p) = i \left( \bm{s}^\mu \cdot \gamma \right) q_\mu + {\cal O}(q^2).$$ Thus this lattice action is reduced to the covariant Dirac form $i \left( \bm{s}^\mu \cdot \gamma \right) p_\mu \equiv i \slash{k}$, in the low energy region where we identify $k_\mu$ as the Cartesian momentum. In the cases of other poles $\hat{p}_\mu \not=0$, however, the Dirac operator includes not only $i\gamma$-terms but $\gamma \gamma_5$-terms as seen in Eq.(\[BBTW\_Dirac\_exp\]). As a result, one cannot obtain a covariant form of excitation from these poles, but unphysical fermions.
Another lattice action with the sufficient discrete symmetry is obtained by modifying Creutz fermion. To recover the sufficient discrete symmetry $\mathbb{Z}_5$ of Creutz action to prohibit redundant operators, it was suggested in that one drop non-nearest hopping terms and choose $B=1/\sqrt{5}$, $C=1$. Then we obtain another type of Creutz action, $$S_{\mathrm{dC}} = \sum_x \Big[ \sum_{\mu=1}^4 \left( \bar{\phi}_{x-\bm{a}_\mu} \Sigma \cdot \bm{s}^\mu \chi_x - \bar{\chi}_{x+\bm{a}_\mu} {\Sigma} \cdot \bm{s}^\mu \phi_x \right) + \bar{\phi}_x \Sigma \cdot \bm{s}^5 \chi_x - \bar{\chi}_x {\Sigma} \cdot \bm{s}^5 \phi_x \Big], \label{dCreutz_action}$$ and then we call this action Dropped Creutz action.
In this case, the Dirac operator is almost the same as (\[BBTW\_Dirac\]) but the definition of $i \gamma_4$ is modified as $\gamma_5 \gamma_4$, which is also anti-hermitian. As the case of BBTW fermion, both $\gamma$-terms and $\gamma\gamma_5$-terms are included, and thus Nielsen-Ninomiya’s theorem cannot be applied to this lattice action. Indeed the lattice fermion around $p_\mu=0$ is not written as a covariant form, $i \vec{\gamma} \cdot
\vec{k} + \gamma_5 \gamma_4 k_4$.
In BBTW action (\[BBTW\_action\]) any of the three conditions are not satisfied, while two of them (i) Non-nearest-site hopping and (iii) Lattice deformation are not satisfied in Dropped Creutz action (\[dCreutz\_action\]). The point is that these actions do not satisfy (i), namely they contain none of non-nearest-site hoppings necessary for Lorentz covariant excitations of fermions as discussed in Sec. \[sec:features\]. As a consequence, although they preserve the discrete symmetry for a continuum limit, they inevitably produce unphysical fermions as seen above. These results are consistent with the argument discussed in Sec. \[sec:features\] that the requisite discrete symmetry for a good continuum limit is incompatible with physical minimal-doubling actions on hyperdiamond lattices.
Other fermions
--------------
We then consider the lattice action based on the hyperdiamond lattice with (i)non-nearest hoppings and (iii)lattice deformation while the spinor structure is not twisted. We call this Untwisted Creutz action. The associated Dirac operator is given by $i \gamma_4 \to \gamma_4$ in Eq. (\[Creutz\_Dirac\]) as $$\begin{aligned}
D(p)\ =
&& (\sin p_1 + \sin p_2 - \sin p_3 - \sin p_4) i \gamma_1 \nonumber \\
&+& (\sin p_1 - \sin p_2 - \sin p_3 + \sin p_4) i \gamma_2 \nonumber \\
&+& (\sin p_1 - \sin p_2 + \sin p_3 - \sin p_4) i \gamma_3 \nonumber \\
&+& B (4C - \cos p_1 - \cos p_2 - \cos p_3 - \cos p_4) \gamma_4.
\label{utCreutz_Dirac}\end{aligned}$$ Thus we obtain minimal-doubling fermions for $1/2<C<1$ as Creutz action. However, since $\gamma_4$ is not anti-hermitian, covariance of them is broken as $i\vec{\gamma}\cdot\vec{k}+\gamma_4 k_4$. As the case of Creutz action discussed in section \[sec:Creutz\], Untwisted Creutz action with the exact hyperdiamond spinor vectors is given by the condition $C=1$. Then the same unphysical fermion $i \vec{\gamma} \cdot \vec{k}$ is obtained at $p=0$ as Creutz fermion with $C=1$.
The remaining lattice fermions are BBTW fermion and Dropped Creutz fermion with the lattice deformation. Since they include only nearest neighbor hoppings, both of $i\gamma$-terms and $\gamma\gamma_5$-terms arise in the action, and thus Nielsen-Ninomiya’s no-go theorem cannot be applied to them as discussed before. All these fermions implies the necessity of the three conditions for physical fermions on the hyperdiamond lattice.
Appended Creutz fermion {#sec:appended}
-----------------------
In this section we propose a new lattice action, which also possesses physical minimal-doubling fermions as the case of the original Creutz action. It is pointed out in section \[sec:features\] that Creutz action includes non-nearest-site hopping terms but nearest unit cell hoppings. However, as shown in Fig. \[honeycomb\_nonnearest\], all of nearest unit cell hoppings are not included by Creutz action, and we now consider a new lattice action including all of them, $$\begin{aligned}
S_{\mathrm{aC}} & = & S_{\mathrm{C}} + \frac{1}{2} \sum_x \sum_{\mu<\nu} \Big[ \bar{\phi}_{x-\bm{a}_\mu+\bm{a}_\nu} \Sigma \cdot \left( \bm{s}^\mu - \bm{s}^\nu \right) \chi_x - \bar{\chi}_{x+\bm{a}_\mu-\bm{a}_\nu} \Sigma \cdot \left( \bm{s}^\mu - \bm{s}^\nu \right) \phi_x \nonumber \\
&& \hspace{6.5em} - \bar{\phi}_{x+\bm{a}_\mu-\bm{a}_\nu} \bar{\Sigma} \cdot \left( \bm{s}^\mu - \bm{s}^\nu \right) \chi_x + \bar{\chi}_{x-\bm{a}_\mu+\bm{a}_\nu} \bar{\Sigma} \cdot \left( \bm{s}^\mu - \bm{s}^\nu \right) \phi_x \Big]. \label{appended_action}\end{aligned}$$ We will call this new action Appended Creutz action. Here the hopping vectors $\bm{a}_{\mu}$ are obtained from the reciprocal vectors in the momentum space, which differ slightly from those of the original Creutz action.
The additive contribution to the Dirac operator is given by $$\begin{aligned}
D'(p)\ = && 2 ( \sin p_{13} + \sin p_{14} + \sin p_{23} + \sin p_{24} ) i\gamma_1 \nonumber \\
& + & 2 ( \sin p_{12} + \sin p_{13} - \sin p_{24} - \sin p_{34} ) i\gamma_2 \nonumber \\
& + & 2 ( \sin p_{12} + \sin p_{14} - \sin p_{23} + \sin p_{34} )
i\gamma_3
\label{appended_Dirac}\end{aligned}$$ where we define $p_{12} = p_1 - p_2$, etc. The coefficient of $i\gamma_4$ is constantly zero and the total Dirac operator is obtained as the sum of (\[Creutz\_Dirac\]) and (\[appended\_Dirac\]). Thus it gives the same minimal-doubling poles at $p = \pm (\tilde p, \tilde p, \tilde p, \tilde p)$ with $\cos \tilde p = C$ and the minimal-doubling condition $1/2 < C < 1$ as the case of Creutz action. Because the Dirac operator (\[appended\_Dirac\]) is written by only $i\gamma$-terms, the physical minimal-doubling fermions are obtained from (\[appended\_action\]).
This action satisfies all of three conditions discussed in section \[sec:features\], the non-nearest-site hoppings, the spinor twist and the lattice deformation. As a result, it lacks the sufficient discrete symmetry to suppress the redundant operators generated by the loop corrections[@Capitani:2009yn]. Actually although this Dirac operator (\[appended\_Dirac\]) includes only $\gamma_1$, $\gamma_2$ and $\gamma_3$ terms, fermionic excitations from the pole of this operator are the same as those of the original one (\[Creutz\_Dirac02\]) up to a factor. After all, Appended Creutz action is quite similar to the original Creutz action, and thus it suggests stability of the minimal-doubling poles to some kinds of perturbations.
-- ------------ ------------ ------------ --
$\bigcirc$ $\bigcirc$ $\bigcirc$
$\bigcirc$ $\bigcirc$ $\bigcirc$
$\bigcirc$ $\bigcirc$ $\times$
$\bigcirc$ $\times$ $\bigcirc$
$\bigcirc$ $\times$ $\times$
$\times$ $\bigcirc$ $\bigcirc$
$\times$ $\bigcirc$ $\times$
$\times$ $\times$ $\bigcirc$
$\times$ $\times$ $\times$
-- ------------ ------------ ------------ --
: List of lattice fermions based on the hyperdiamond lattice. The lattice actions including only physical modes are Creutz and Appended Creutz action, which are the minimal-doubling action. BBTW action has a physical and some unphysical poles. The others do not possess any physical poles. The fermions with the sufficient discrete symmetry are BBTW fermion and Dropped Creutz fermion.[]{data-label="3features"}
At last, we have investigated all of lattice fermions in the context of the three features as listed in Table \[3features\]. Furthermore, we have proposed a new lattice action, called Appended Creutz action, which gives the physical minimal-doubling fermions. After all, the lattice actions giving only physical fermions are the original Creutz action and Appended Creutz action. Other actions have some problems such as including both of $i\gamma$ and $\gamma\gamma_5$ or breaking covariance.
While we have discussed free lattice fermions, we now remark effects of gauge interaction. As discussed in preceding studies, since one of the spacetime direction is specified in lattice actions we have considered, we have to renormalize the light of speed to resolve the anisotropy generated by interactions. Thus it is expected that the Lorentz covariance is just modified, but not broken through gauge interactions and quantum corrections because physical fermions discussed here actually satisfy all of the conditions for Nielsen-Ninomiya’s theorem. On the other hand, it is slightly meaningless to consider quantum corrections for the lattice actions including unphysical fermionic modes.
We then comment on chiral charge of the lattice fermions. As the case of Creutz fermion, the physical minimal-doubling fermions we have discussed possess the sublattice structure of the hyperdiamond lattice, and thus their total chiral charge becomes zero. However the index of the Dirac operator, which is interpreted as that of the real vector field in the context of Poincaré-Hopf theorem, is ill-defined if both of $i\gamma$ and $\gamma\gamma_5$ are included.
A novel construction of minimal-doubling fermion {#sec:rhombus}
================================================
As discussed in Sec. \[sec:hyperdiamond\_Creutz\], we can construct Creutz fermion on the distorted hyperdiamond lattice. However in this real-space construction, the interactions based on the vectors $\bm{a}_{\mu}$ (primitive vectors) stand for hoppings from one unit cell to another unit cell, not between hyperdiamond-lattice sites. In addition there is no nontrivial hopping between two sites in the same unit cell except for on-site terms. As seen from these reasons, the real-space construction of Creutz fermion on the hyperdiamond lattice is somewhat misleading although it gives intuitive understanding of the discrete symmetry of the action, and it is much more natural that the two sites in one unit cell is identified as a single site as seen in Fig. \[honeycomb\_rhombus\]. Based on this argument, in this section we explicitly show alternative and more reasonable spatial construction of Creutz-type minimal-doubling action on a deformed hypercubic [*rhombus*]{} lattice, which is, in a sense, a generalized version of Creutz action.
Comparing it with another well-known minimal-doubling fermion on hypercubic lattice, called Wilczek fermion[@Wilczek:1987kw], both of them include additive terms proportional to $\gamma_4$. However the location of poles of Creutz’s Dirac operator is adjustable while that of Wilczek fermion is not. In this section we also consider a modification of Wilczek action, and construct an orthogonal lattice action with adjustable poles.
Creutz-type lattice action
--------------------------
We especially remark the translation symmetry of the real-spatial lattice on which the action is defined, and show a novel minimal-doubling fermion including Creutz fermion can be constructed on a deformed hypercubic (rhombus) lattice only with nearest neighbor hoppings.
![(a) Translation symmetry of the honeycomb lattice. Unit cells consisting of $L$-node and $R$-node are encircled. (b) A deformed square (rhombus-like) lattice with the translation symmetry equivalent to that of the honeycomb lattice.[]{data-label="honeycomb_rhombus"}](honey_prim.eps "fig:"){height="10em"} ![(a) Translation symmetry of the honeycomb lattice. Unit cells consisting of $L$-node and $R$-node are encircled. (b) A deformed square (rhombus-like) lattice with the translation symmetry equivalent to that of the honeycomb lattice.[]{data-label="honeycomb_rhombus"}](honey_rhom.eps "fig:"){height="10em"}
As shown in Fig. \[honeycomb\_rhombus\], we can consider rhombus-like lattice which possesses the translation symmetry equivalent to that of the hyperdiamond lattice. Its spatial primitive vectors $\bm{a}_\mu$ will be found later, but we anyway introduce a lattice action defined on the rhombus-like lattice, $$S = \frac{1}{2} \sum_{x} \Bigg[\sum_{\mu=1}^4 \left(
\bar\psi_{x}\Gamma\cdot\bm{s}^\mu\psi_{x+\bm{a}_\mu}
-
\bar\psi_{x+\bm{a}_\mu}\bar\Gamma\cdot\bm{s}^\mu\psi_x
\right) +2i\,t\, \bar\psi_x \gamma_4 \psi_x \Bigg] \label{rhom_action01}$$ with $\Gamma=(\vec\gamma, -i\gamma_4)$ and $\bar\Gamma=(\vec\gamma,
i\gamma_4)$. Here $t$ stands for a free parameter to be fixed for minimal number of doublers. This expression is similar to what is presented in , but in this paper we explicitly show alternative spatial construction. There are five spinor vectors in the previous lattice action (\[Creutz\_action\]), but in this action only four vectors and an on-site parameter instead of fifth vector. This on-site term is a Wilson-like term with one specific direction[@Karsten:1981gd; @Wilczek:1987kw].
If we apply (\[twisted\_bond\]) to spinor vectors $\{\bm{s}^\mu\}$, this lattice action is reduced to Creutz action. To show this, we investigate its momentum space structure of the action (\[rhom\_action01\]). Taking its Fourier transformation, the associated Dirac operator is obtained as $$D(p) = i \sum_{\mu=1}^4 \zeta_\mu(p) \gamma_\mu \label{Creutz_Dirac03}$$ where the coefficients of gamma matrices are $$\zeta_\mu(p) = \left\{ \begin{array}{ll}
\sum_{\nu=1}^4 \left(\bm{s}^\nu\right)_\mu \sin p_\nu &
( \mu=1,2,3 )\\
t - \sum_{\nu=1}^4 \left(\bm{s}^\nu\right)_\mu \cos
p_\nu & (\mu = 4)
\end{array} \right. ,$$ and $\left(\bm{s}^\nu\right)_\mu$ represents $\mu$-th components of $\bm{s}^\nu$. In order that this Dirac operator has minimal-doubling poles at $p =
\pm(\tilde p, \tilde p, \tilde p, \tilde p)$ with $\tilde p>0$, spinor vectors should satisfy $$\sum_{\nu=1}^4 \left(\bm{s}^\nu\right)_\mu = \left\{\begin{array}{cl}
0 & (\mu=1,2,3) \\
t / \tilde C & (\mu=4)
\end{array}\right. , \quad
\tilde C = \cos \tilde p.$$ It is easy to show that (\[twisted\_bond\]) actually satisfy this condition, and to relate this Dirac operator (\[Creutz\_Dirac03\]) to Creutz’s Dirac operator (\[Creutz\_Dirac\]) we choose $\tilde C = C$ and $t=4BC$.
To study fermionic excitations around poles, we expand the coefficients of gamma matrices around poles with $p_\mu=\tilde p + q_\mu$, $$\zeta_\mu(p) = \left\{ \begin{array}{ll}
\tilde C \sum_{\nu=1}^4 \left(\bm{s}^\nu\right)_\mu
q_\nu + \mathcal{O}(q^2) &
(\mu=1,2,3) \\
\tilde S \sum_{\nu=1}^4 \left(\bm{s}^\nu\right)_\mu q_\nu + \mathcal{O}(q^2) & (\mu = 4)
\end{array} \right.$$ with $\tilde S=\sin \tilde p$. Thus reciprocal vectors are given by $$\left(\bm{b}^\nu\right)_\mu = \left\{ \begin{array}{ll}
\tilde C \left(\bm{s}^\nu\right)_\mu &
(\mu=1,2,3) \\
\tilde S \left(\bm{s}^\nu\right)_\mu & (\mu = 4)
\end{array} \right. .$$ At this stage, we obtain primitive (hopping) vectors $\{\bm{a}_\mu\}$ of the lattice by the condition $\bm{a}_\mu\cdot\bm{b}^\nu=\delta_\mu^{\nu}$. In general, the lattice becomes non-orthogonal, deformed hypercubic lattice as shown in Fig. \[honeycomb\_rhombus\].
An advantage of this action is of course that we can easily introduce gauge fields by link variables on bonds of the rhombus-like lattice. The fermionic part of the lattice action with gauge fields and a mass-term is given by $$S = \frac{1}{2} \sum_{x,\mu} \left[
\bar\psi_{x}\Gamma\cdot\bm{s}^\mu
U_{x,\mu}\psi_{x+\bm{a}_\mu}
-
\bar\psi_{x+\bm{a}_\mu}\bar\Gamma\cdot\bm{s}^\mu
U^\dag_{x,\mu}\psi_x
\right] + \sum_x \left[ M\bar\psi_x\psi_x +it \bar\psi_x \gamma_4
\psi_x \right].$$ Note that link variables are also represented in non-orthogonal coordinates. Then we can introduce the action of gauge fields as the plaquette action on the deformed hypercubic lattice.
Orthogonal lattice action
-------------------------
We have discussed much simpler and generalized expression of Creutz-type minimal-doubling lattice fermion. Then we claim minimal-doubling poles are due to the modification of the lattice action by introducing on-site term which is proportional to $\gamma_4$. In this sense, its minimal-doubling mechanism is similar to Wilczek action. However in the case of Creutz action, the location of poles is adjustable while that of Wilczek action is not. We then construct an orthogonal lattice action whose poles are adjustable. It will be the simplest form of Creutz-type lattice action, and useful for us to understand its structure.
We now construct a lattice action on an orthogonal lattice, $$S = \frac{1}{2} \sum_{x,\mu}
\left[ \bar\psi_x \gamma_\mu \psi_{x+\bm{a}_\mu}
- \bar\psi_{x+\bm{a}_\mu}\gamma_\mu \psi_x
\right]
+ \frac{i}{2}r \sum_x \left[
2(3+t)\bar\psi_x\gamma_4\psi_x - \sum_{\mu=1}^4
\left(
\bar\psi_x \gamma_4 \psi_{x+\bm{a}_\mu}
+ \bar\psi_{x+\bm{a}_\mu}\gamma_4 \psi_x
\right) \right],$$ where $t$ and $r$ are free positive parameters to be fixed for minimal number of doublers. The explicit expression of primitive vectors will be shown later. The associated Dirac operator is given by $$D(p) = i \sum_{\mu=1}^4 \gamma_\mu \sin p_\mu + i r \gamma_4
\left[\sum_{j=1}^3\left(1-\cos p_j\right) + \left(t-\cos
p_4\right)\right]. \label{Dirac_orth}$$ To obtain poles of this Dirac operator, we take coefficients of $\gamma$’s to be zero. In the cases of $j=1,2,3$, we obtain $$\sin p_j = 0,\ \pi \quad (j=1,2,3).$$ The coefficient of $\gamma^4$ reads $$\begin{aligned}
&& \sin p_4 + r \left[\sum_{j=1}^3(1-\cos p_j) + (t-\cos p_4)\right]
\nonumber \\
& = & \sqrt{1+r^2} \sin(p_4-\alpha) + r (t+2N_\pi)\end{aligned}$$ with $N_\pi=\#\{p_j=\pi,\ j=1,2,3\}$, $\cos \alpha = 1/\sqrt{1+r^2}$, $\sin \alpha = r/\sqrt{1+r^2}$. Thus the condition such that this Dirac operator induces only minimal-doubling two poles is given by $$\left|\frac{r}{\sqrt{1+r^2}}\right|t < 1, \quad
\left|\frac{r}{\sqrt{1+r^2}}\right|(t+2) > 1,$$ and then minimal-doubling poles become $p = (0,0,0,p^{(\pm)}+\alpha)$ with $\sin p^{(\pm)} = -rt/\sqrt{1+r^2}$, $\cos p^{(\pm)} = \pm
\sqrt{(1+r^2(1-t^2))/(1+r^2)}$.
To obtain translation symmetry of the lattice action, we expand the Dirac operator around a pole with momentum $(q_1,q_2,q_3,q_4+p^{(+)}+\alpha)$ as $$D(p) = i\sum_{j=1}^3 \gamma_j q_j + i \sqrt{1+r^2(1-t^2)}\gamma_4 q_4 + \mathcal{O}(q^2).$$ Therefore reciprocal vectors for this lattice action are represented by $$\left(\bm{b}^\nu\right)_\mu = \left\{ \begin{array}{cl}
\delta^\nu_\mu &
(\mu=1,2,3) \\
\delta^\nu_\mu \sqrt{1+r^2(1-t^2)} & (\mu = 4)
\end{array} \right. ,$$ and by the condition $\bm{a}_\mu\cdot\bm{b}^\nu=\delta_\mu^{\nu}$, primitive vectors are obtained, $$\left(\bm{a}_\nu\right)_\mu = \left\{ \begin{array}{cl}
\delta_{\nu\mu} &
(\mu=1,2,3) \\
\delta_{\nu\mu}/\sqrt{1+r^2(1-t^2)} & (\mu = 4)
\end{array} \right. .$$ It indicates the lattice action is constructed on orthogonal lattice with one direction specified. We now remark although both this Wilczek-type action and Boriçi action are defined on an orthogonal lattice, they are not equivalent.
As the case of Creutz action, minimal-doubling poles induce two Dirac fermions. Although the Dirac operator (\[Dirac\_orth\]) has parity symmetry, the action lacks sufficient discrete symmetry to remove redundant operators generated by loop corrections. Besides, when the fourth component of the reciprocal vector is zero, covariance of the associated fermion is broken and it becomes unphysical.
Summary {#sec:summary}
=======
In this paper we investigate minimal-doubling fermion actions on deformed-hyperdiamond lattices, with emphasis on the real-space construction of them and the correct excitations from poles of propagators, then generalize them to an action on a rhombus lattice.
In Sec. \[sec:hyperdiamond\_Creutz\] we propose the spatial construction of Creutz fermion action on a deformed hyperdiamond lattice, where the hopping vectors (or the primitive vectors) are consistently determined by the reciprocal vectors in the momentum space. It means that the spatial lattice structure, on which fermions live, depends not only on the spinor vector $\bm{s}^{\mu}$ but also on the form of the action itself. Based on this construction we give a condition for the action to be defined on the exact hyperdiamond lattice in the real space while a condition for the poles of propagators to be located on the hyperdiamond-lattice sites is proposed in .
In Sec. \[sec:features\] we investigate the conditions for a hyperdiamond-lattice action to produce physical or Lorentz-covariant excitations from poles of fermion propagators, which actually Creutz fermion satisfies. Then it is pointed out that the non-nearest-site (but nearest-unit-cell) hoppings are essential for the correct excitations from the poles of doublers: If the action on the hyperdiamond lattice does not contain any non-nearest-site hopping, it shall yield fermions with unphysical excitations since the associated Dirac operator inevitably includes both of $i\gamma$- and $\gamma \gamma_{5}$-terms while Nielsen-Ninomiya’s no-go theorem assumes either of vector or axial-vector functions. This fact implies that the requisite discrete symmetry of the action for a good continuum limit is incompatible with “minimal-doubling” with correct excitations on hyperdiamond lattices as firstly discussed in . All the fermion actions we have discussed in Sec. \[sec:examples\] back up this incompatibility. In the section we also study a related minimal-doubling fermion called “Appended Creutz fermion”, which contains all the nearest neighbor interactions in terms of unit cells.
As discussed in Sec. \[sec:rhombus\], we can construct Creutz-type minimal-doubling actions more naturally on a deformed hypercubic lattice, instead of a hyperdiamond lattice. We propose a class of minimal-doubling fermion actions defined on a rhombus lattice. In a sense, this class of the actions is a generalization of Creutz-type actions since it reduces to the original Creutz action by choosing the parameters appropriately. Based on this alternative and reasonable spatial construction of minimal-doubling actions, the link variables are easily introduced. In the section we also introduce a two-parameter class of Wilczek-type minimal-doubling actions, which is the most simple form of Creutz-type action.
As a future work we will search for a general and unified form of minimal-doubling actions, which can reduce to all the known minimal-doubling actions including Wilczek fermion and Creutz fermion. This kind of the actions, if exists, will reveal more on the incompatibility between “minimal-doubling” and the requisite discrete symmetry for a good continuum limit.
Acknowledgments {#acknowledgments .unnumbered}
===============
We would like to thank T. Onogi for reading the manuscript and useful discussions. We also thank S. Aoki, M. Creutz and Y. Kikukawa for valuable comments. TM is supported by Grant-in-Aid for the Japan Society for Promotion of Science (JSPS) Research Fellows.
[^1]: In the case of the graphene system, these sublattice structure is often represented by $A$ and $B$ sites.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Contrary to commonly held belief, we show that one can obtain a low value for $M_R$, the $SU(2)_R$ breaking scale, in grand unification theories based on $SO(10)$. This possibility emerges in the supersymmetric version of $SO(10)$ with a judicious choice of Higgs content. The unification scale is found to be consistent with the constraint from proton decay. This result is first explicitly demonstrated using the one loop renormalization group equations, and then a full two loop analysis is carried out.'
---
-0.5in 6.5in 9.00in 4.0=0.5in -0.5in
in \#1\#2
plus0pt minus0pt \#1\#2[0.7ex]{} \#1[$\underline{\smash{\vphantom{y}\hbox{#1}}}$]{}
\#1[$^{#1}$ ]{} \#1
\#1
\#1[\[\#1\]]{}
[**SO(10) Grand Unification with a Low-Energy $M_R$**]{}
N.G Deshpande, E. Keith, and T.G. Rizzo\
Institute of Theoretical Science\
University of Oregon\
Eugene, OR 97403\
It is commonly believed[@one; @two] that the group $SU(2)_R$ has to be broken at a large energy scale $M_R \sim 10^{10}\gev$ if it is to emerge from a grand unified symmetry such as $SO(10)$. This is also assumed to be true for the supersymmetric version of $SO(10)$[@one; @two]. Consequently, additional gauge bosons that could possibly be produced at supercollider energies are thought to originate only from additional $U(1)$ factors which lead to $Z'$ bosons[@three]. The phenomenology of new charged $W'$ bosons at supercolliders is therefore less frequently investigated[@four]. We will show in this Letter that although the above result is true for the simplest Higgs structure, if this sector is suitably enlarged, the scale for the right handed gauge bosons, $M_R$, could be made arbitrarily low. We will consider only the supersymmetric version in detail, with some brief remarks on the non-supersymmetric case given at the end of our discussion.
We investigate the breaking chain $$SO(10)\, (SUSY)
\chain{M_U}{} 2_L 2_R 1_{B-L} 3_C\, (SUSY) \chain{M_R}{} 2_L 1_Y 3_C\,
(SUSY) \chain{M_Z} {} 3_C1_Q\, .\label{+}$$ where, as an example of our notation, $2_L$ represents $SU(2)_L$. Here, we have set the ‘effective’ supersymmetry breaking scale to be $M_Z$, and will comment on this later.
In Ref. 5, it was shown that if both $2_L 2_R 1_{B-L} 3_C$ singlets of the [**210**]{} representation, together with the $2_L 2_R
1_{B-L}3_C$ singlet of the [**45**]{} representation, acquire vacuum expectation values (VEVs) then this is sufficient to break supersymmetric $SO(10)$ down to supersymmetric $2_L 2_R 1_{B-L} 3_C$ without D-parity. The breaking at $M_R$ can be performed either by the Higgs fields in the ${\bf 126} \oplus \overline{\bf 126}$ representation, or in the ${\bf 16} \oplus \overline{\bf 16}$ representation and we consider both these possibilities in our discussion below. We further assume that ordinary electroweak breaking at the $Z$ scale is achieved as usual by a complex [**10**]{} representation. For the purpose of generating fermion masses, we assume that the entire bi-doublet of the [**10**]{} representation has a mass at the scale of $M_Z$. (We remind the reader that a bi-doublet corresponds to the $(2,2,0,1)$ representation of $2_L2_R1_{B-L}3_C$.) Also, we assume that the $SU(2)_R$ triplets of the [**126**]{} and $\overline{\bf 126}$ representations and the $SU(2)_R$ doublets of [**16**]{} and $\overline{\bf 16}$ representations have masses at the scale $M_R$. All other Higgs multiplets are given masses of order $M_U$ as follows from the survival hypothesis. We make the important observation that in this symmetry breaking pattern pseudo-Goldstone bosons do not appear[@five].
First let us examine the one loop equations: $$\begin{aligned}
\alpha^{-1}_{1Y}(M_Z) &=& \alpha^{-1}_{U}(M_U)+{b_{1Y}\over 2\pi}R
+{1\over 2\pi}\left( {3b_{2R}\over 5} +{2b_{B-L}\over 5}\right) (U-R)\, ,
\nonumber\\
\alpha^{-1}_{2L}(M_Z) &=&
\alpha^{-1}_{U}(M_U)+{b_{2L}\over 2\pi}U\, ,\label{++} \\
\alpha^{-1}_{3C}(M_Z) &=& \alpha^{-1}_{U}(M_U)+{b_{3C}\over 2\pi}U\,
,\nonumber\end{aligned}$$ where $$\begin{aligned}
R=\ln{M_R\over M_Z}\, ,\nonumber\\
U=\ln{M_U \over M_Z}\, .\end{aligned}$$ The $b_i$’s are the one loop beta functions, which for the supersymmetric case are given by $$b_{N}^{SUSY}=2 n_g-3N+T(S_N)\, ,$$ for $n_g$ generations, the gauge group $SU(N)$, and the complex Higgs fields contribution which is parameterized by $T(S_N)$. For $U(1)$ gauge groups, $N=0$ in the above equation and the gauge couplings are normalized as usual. Explicitly we find the Higgs contributions to be given by $$\begin{aligned}
T_{1Y} &=& {3\over5}n_{10}\; , \; T_{2L}=n_{10}\; , \; T_{3c}=0\, ,\nonumber\\
T_{2R} &=& n_{10}+n_{16}+4n_{126}\; , \; T_{1X}={3\over 2}n_{16}+ 9n_{126}\,
,\label{+++}\end{aligned}$$ where the subscripts on the $T$’s refer to the relevant gauge group. In the above, $n_{10}$ is the number of complex [**10**]{} Higgs bi-doublets at the scale $M_Z$, and $n_{16}$ and $n_{126}$ are the number of ${\bf 16} \oplus \overline{\bf 16}$ and ${\bf 126} \oplus \overline{\bf
126}$ Higgs pairs, respectively, which are used to break the intermediate gauge symmetry. Using Eqs. (\[++\]) and (\[+++\]) together with the definitions $$\begin{aligned}
\alpha_{1Y}^{-1}(M_Z) &=& {3\over 5}{1-\tilde{x} \over \alpha(M_Z)}\,,
\nonumber \\
\alpha_{2L}^{-1}(M_Z) &=& {\tilde{x} \over \alpha (M_Z)}\, ,\end{aligned}$$ gives the relations $$\begin{aligned}
{2\pi\over \alpha (M_Z)}\left(1-{8\alpha (M_Z)\over 3 \alpha_{3C}(M_Z)}
\right) &=& \left( C_1-C_2\right) U+C_2 R\, ,\nonumber\\
{2\pi\over \alpha(M_Z)}\left( 1-{8\over 3}\tilde{x} \right) &=& \left(
{5\over 3}C_3-C_2\right) U+C_2 R\, , \label{++++}\end{aligned}$$ where the abbreviation $\tilde{x} \equiv \sin^2 \theta_W (\overline{MS})$ is used, and the $C_i$ are given by $$\begin{aligned}
C_1 &\equiv &b_{2L}+{5\over 3}b_{1Y}-{8\over 3}b_{3C}=18+2n_{10}\, ,\nonumber\\
C_2 &\equiv &{5\over 3}b_{1Y}-b_{2R}-{2\over 3}b_{1X}=6-2n_{16} -10n_{126}\,
,\label{XX}\\
C_3 &\equiv & b_{1Y}-b_{2L}=6-{2\over 5}n_{10}\, .\nonumber\end{aligned}$$ We make the observation that if $C_2=0$ then the scale $M_R$ is completely undetermined at the one loop level. This gives us hope that when $C_2=0$, a solution with a low energy $M_R$ will exist. We can have $C_2=0$ only when $n_{16}=3$ and $n_{126}=0$. We then need only to require that the two equations in (7) give agreeing values of $M_U$ within the level of accuracy of the one loop approximation. The latest values[@six] of the input parameters that we use in this analysis are $$\begin{aligned}
\alpha^{-1}(M_Z) &=& 127.9\pm 0.1\, ,\nonumber\\
\alpha_{3C} (M_Z) &=& 0.118\pm 0.007\, ,\label{X} \\
\tilde{x} (M_Z) &=& 0.2326\pm 0.0011\, ,\nonumber\\
M_Z &=& 91.187\pm 0.007\, GeV\, .\nonumber\end{aligned}$$ Consistency of the two equations in (7) with $n_{16}=3$, $n_{126}=0$ and $n_{10} = 1$, and taking the central values of $\alpha^{-1} (M_Z)$ and $\tilde{x} (M_Z)$ from above implies $\alpha_{3C} (M_Z) = 0.112$, which is within the experimentally allowed range given above. Of course we will have to perform a full two loop analysis to insure a solution exists with a low energy $M_R$ with $n_{16}=3$, $n_{126}=0$ and $n_{10}=1$. Note that for this particular chioce of Higgs representations the combination ${3\over 5}\alpha^{-1}_{2R} +{2\over 5}\alpha^{-1}_{1X}$ runs identically at the one loop level as $\alpha^{-1}_{1Y}$ in the minimal supersymmetric standard model (MSSM). Since supersymmetric $SU(5)$ grand unification is consistent with $M_S \simeq M_Z$[@one; @two; @seven], one would naively expect that a two loop analysis will show the $SO(10)$ scenario to be equally consistent and we will now show this explitcitly.
We numerically integrate the two loop equations $$\mu{\partial \alpha_i (\mu)\over \partial \mu}={1\over 2\pi} \left(
b_i+{b_{ij}\over 4\pi}\alpha_j (\mu) \right) \alpha_i^2 (\mu) \, .\label{+++++}$$ assuming the breaking pattern as given in Eq. (1) (with $n_{10}=1,\
n_{16}=3$ and $n_{126}=0$) and we use the approximation that all superparticles have mass $M_S=M_Z$. Actually, the ‘average’ sparticle masses could be somewhat different than this effective value[@one; @two; @seven]. We will also assume that the top quark and right handed neutrino have masses $\simeq M_Z$. We use the appropriate two loop matching conditions[@eight] at $M_U$ that follows from dimensional reduction: $$\alpha^{-1}_U
(M_U)-{C_U \over 12\pi}=\alpha^{-1}_i (M_U)-{C_i\over 12\pi} \, ,$$ where $i$ represents the intermediate gauge groups $2_L$, $2_R$, $1_{B-L}$ or $3_C$ and $C_G$ is the quadratic Casmir invariant for group $G$. Similarly at $M_R$ we have: $$\alpha^{-1}_{1Y}(M_R)={3\over 5}\left( \alpha^{-1}_{2R} (M_R)
-{C_2 \over 12\pi}\right) +{2\over 5}\alpha^{-1}_{1B-L} (M_R)\, .$$ $C_G=N$ for $SU(N)$ and $C_G=0$ for $U(1)$. In Eq.(\[+++++\]), we assume the MSSM below the scale so that, with $i=1_Y \, ,\, 2_L \, ,\, 3_C$, respectively, we use the following two loop beta functions[@nine] $$\begin{aligned}
b^{MSSM}_{ij}=\left( \matrix{ 0&0&0\cr 0&-24&0\cr
0&0&-54\cr} \right) +n_g \left( \matrix{ {38\over 15}&{6\over
5}&{88\over 15}\cr {2\over 5}&{14}&{8}\cr {11\over 15}&{3}&{68\over
3}\cr} \right) +n_{10} \left( \matrix{ {9\over 25}&{9\over 5}&{0}\cr
{3\over 5}&{7}&{0}\cr {0}&{0}&{0}\cr} \right) \, .\end{aligned}$$ For the intermediate symmetry SUSY $2_L 2_R 1_{B-L} 3_C$, we derive from the generic two loop expression in reference[@nine] $$\begin{aligned}
b^{int.}_{ij} &=& \left( \matrix{ {-24}&{0}&{0}&{0}\cr
{0}&{-24}&{0}&{0}\cr {0}&{0}&{0}&{0}\cr {0}&{0}&{0}&{-54}\cr } \right)
+n_g \left( \matrix{ {14}&{0}&{1}&{8}\cr {0}&{14}&{1}&{8}\cr
{3}&{3}&{7\over 3}&{8\over 3}\cr {3}&{3}&{1\over 3}&{68\over 3}\cr }
\right) \nonumber\\ &+& n_{10} \left( \matrix{ {7}&{3}&{0}&{0}\cr
{3}&{7}&{0}&{0}\cr {0}&{0}&{0}&{0}\cr {0}&{0}&{0}&{0}\cr } \right)
+n_{16} \left( \matrix{ {0}&{0}&{0}&{0}\cr {0}&{7}&{3\over 2}&{0}\cr
{0}&{9\over 2}&{9\over 4}&{0}\cr {0}&{0}&{0}&{0}\cr } \right) \, ,\end{aligned}$$ where $i\, ,\, j=2_L\, ,\, 2_R\, ,\, 1_{B-L}\, ,\, 3_C$ respectively. In evaluating the above two equations for $b_{ij}$, we of course assume $n_{g}=3$, $n_{10}=1$ and $n_{16}=3$. We treat as a free parameter. From the one loop calculation, we expect that the two loop analysis will yield solutions for arbitrary values of $M_R$ between $M_Z$ and $M_U$. We have explored the possibility that can take on a wide range of potential values and find the expectation above to be fulfilled. As an example of this and in particular to show that $M_R$ can be low, we display in Fig. 1 the case $M_R = 1\tev$. We find that $\alpha^{-1}_{U} (M_U)=23.4\pm 0.5$ and $M_U =10^{16.2 \pm .4}\gev$, which is sufficiently large to be consistent with the non-observation of proton decay. We also note that at the scale $M_R$, $\alpha_{2L} / \alpha_{2R}
\simeq 1.5$, thus implying the bound $M_{W_R} > 380\gev$ from muon decay[@ten], and 480 GeV from direct collider searches[@eleven; @twelve]. We have investigated the influence of heavy top quark Yukawa couplings on the two loop evolution[@seven], and find the effect to be negligible. [*Thus, at least for this choice of symmetry breaking, can indeed be sufficiently low as to be of consequence for existing and future colliders.*]{}
We have now described the conditions under which $M_R$ may be associated with a low scale in supersymmetric $SO(10)$ grand unification. From the previous discussion, we can see that we need $M_S \simeq M_Z$ for our scenario to be realized as in the case of SUSY $SU(5)$. What about non-supersymmetric $SO(10)$ grand unification? In this case, the equations analagous to (7) and (8) imply that for $M_R$ to drop out of the one loop equations we require that $5n_{126}+n_{16}=22$, where $n_{126}$ and $n_{16}$ refer to the number of [**126**]{} and [**16**]{} dimensional representations, respectively, used to break the intermediate gauge symmetry $2_L 2_R 1_{B-L} 3_C$ to the SM. Demanding consistency of the two one loop equations analagous to Eq. (8) and using the values of the low energy parameters as given above further implies $n_{10} = 4$, where $n_{10}$ is the number of scalar bi-doublets at the scale $M_Z$. (The entire Higgs multiplet $(2,2,0,1)$, within the [**10**]{} representation must have mass less than $M_R$, otherwise the one loop equations will still depend on $M_R$.) This illustrates the result that in order to achieve grand unification in this non-supersymmetric $SO(10)$ case, as in the conventional $SU(5)$ model[@seven], we would have to employ many Higgs doublets. In the $SO(10)$ case, we then obtain unification with $M_U\simeq 10^{13.6}\gev$, a value which is clearly inconsistent with limits on the proton lifetime.
In summary, we have shown for the first time that a low $SU(2)_R$ breaking scale is compatible with $SO(10)$ grand unification. The symmetry breaking at the scale $M_R$ is accomplished by three generations of Higgs in the [**16**]{} representation, not unlike the three families of quarks and leptons. Further consequenses of this symmetry breaking scenario for neutrino masses will be the subject of a future investigation.
.25in
ACKNOWLEDGEMENTS
TGR would like to thank the Institute of Theoretical Science for its hospitality and JoAnne L. Hewett for discussions related to this work. This work was supported in part by the U.S. Department of Energy grant number DE-FG06-85ER-40224.
\#1 \#2 \#3 [Mod. Phys. Lett. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Nucl. Phys. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Phys. Lett. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Phys. Rep. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Phys. Rev. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Phys. Rev. Lett. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Rev. Mod. Phys. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Z. Phys. [**\#1**]{}, \#2 (\#3)]{} \#1 \#2 \#3 [Int. J. Mod. Phys. [**\#1**]{}, \#2 (\#3)]{}
[99]{} N.G. Deshpande , D46 2261 1992 and University of Oregon report OITS-488 (1992); D. Chang , D31 1718 1985 ; J.M. Gibson and R.E. Marshak, D31 1705 1985 ; Y. Tosa, D28 1731 1983 ; P. Langacker and M. Luo, D44 817 1991 . For a review, see P. Langacker, 72 185 1981 ; see also, G. Senjanovic and T.G. Rizzo, 46 1315 1981 and D24 704 1981 ; M.K. Parida and C.C. Hazra, D40 3074 1989 ; D. Chang , B142 55 1984 ; R.N. Mohapatra and M.K. Parida, University of Maryland reports UMD-PP92-170 and UMD-PP92-179 (1992). See, for example, J.L. Hewett and T.G. Rizzo, in [*Proceedings of the 1988 Snowmass Summer Study on High Energy Physics in the 1990’s*]{}, Snowmass, CO 1988, ed. S. Jensen; V. Barger , D35 166 1987 ; L.S. Durkin and P. Langacker, B166 436 1986 ; F. Del Aguila, M. Quiros, and F. Zwirner, B287 419 1987 , [**B284**]{}, 530 (1987) ; P. Chiappetta , in the [*Proceedings of the Large Hadron Collider Workshop*]{}, Aachen, Germany, 1990; see also F. del Aguila and J. Vidal, A4 4097 1989 ; J.L. Hewett and T.G. Rizzo, D45 161 1992 ; Argonne National Laboratory reports ANL-HEP-PR-92-33 (1992) and ANL-HEP-PR-92-33 (1992); M. Cvetic and P. Langacker, D46 14 1992 and M. Cvetic, P. Langacker, and B. Kayser, 68 2871 1992 ; A. Fiandrino and P. Taxil, D44 3490 1991 and Marseille report CPT-92/P.2717 (1992). See, however, J.L. Hewett and T.G. Rizzo, 183 193 1989 and references therein. X.-G. He and S. Melijanic, D41 1620 1990 . See talks by S.C.C. Ting, R.K. Ellis, and W.J. Marciano at the [*1992 Division of Particles and Fields Meeting*]{}, Fermilab, November 1992. For a complete survey of the situation in SUSY $SU(5)$, see P. Langacker and N. Polonsky, University of Pennsylvania report UPR-0513T (1992). I. Atoniadis , B119 377 1982 ; D.M. Capper , B167 479 1980 . D.R.T. Jones D25 581 1982 ; M.B. Einhorn and D.R.T. Jones, B196 475 1982 ; M.E. Machacek and M.T. Vaughn, B222 83 1983 ; L.J. Hall, B178 75 1981 . P. Langacker and S. Uma Sankar, D40 1569 1989 and references therein. F. Abe , CDF Collaboration, 68 1463 1992 . T.G. Rizzo, talk given at the [*Workshop on Extended Gauge Model Phenomenology*]{}, Gyongyostarjan, Hungary, September 1992.
[**Figure Caption**]{}
- [Evolution of coupling constants for the symmtery breaking chain given in Eq. (1). We have used $n_{10}=1$, $n_{16}=3$, $n_{126}=0$ and $M_R=1\tev$, where these quantities are defined in the text. The error bars we show arise from uncertainties in the low energy parameters in Eq. (9). In the figure, $\alpha^{-1}_{i}$ is calculated via the dimensional reduction scheme at two loop order.\[fig1\]]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Cauchy problem of a multi-dimensional ($d{\geqslant}2$) compressible viscous liquid-gas two-phase flow model is concerned in this paper. We investigate the global existence and uniqueness of the strong solution for the initial data close to a stable equilibrium and the local in time existence and uniqueness of the solution with general initial data in the framework of Besov spaces. A continuation criterion is also obtained for the local solution.'
address:
- |
Institute of Mathematics, Academy of Mathematics & Systems Science\
and Hua Loo-Keng Key Laboratory of Mathematics, Chinese Academy of Sciences\
Beijing 100190, China
- |
Department of Mathematics, Capital Normal University\
Beijing 100048, China
author:
- Chengchun Hao
- 'Hai-Liang Li'
title: 'Well-posedness for a multi-dimensional viscous liquid-gas two-phase flow model'
---
Introduction
============
The models of two-phase or multi-phase flows have a very broad applications of hydrodynamics in industry, engineering, biomedicine and so on, where the fluids under investigation contain more than one component. Indeed, it has been estimated that over half of anything which is produced in a modern industrial society depends, to some extent, on a multi-phase flow process for their optimum design and safe operations. In nature, there is a variety of different multi-phase flow phenomena, such as sediment transport, geysers, volcanic eruptions, clouds and rain. In addition, the models of multi-phase flows also naturally appear in many contexts within biology, ranging from tumor biology and anticancer therapies, development biology and plant physiology, etc. The principles of single-phase flow fluid dynamics and heat transfer are relatively well understood, however, the thermo-fluid dynamics of two-phase flows is an order of magnitude more complicated subject than that of the single-phase flow due to the existence of moving and deformable interface and its interactions with two phases [@Brennen; @IshiiHibiki; @Kolev1].
We consider the drift-flux model of two-phase flows in the present paper, which is principally developed by Zuber and Findlay (1965), Wallis (1969) and Ishii (1977). The basic idea about drift-flux models is that both phases are well mixed, but the relative motion between the phases is governed by a particular subset of the flow parameters. In the case of liquid-gas fluids, it relates the liquid-gas velocity difference to the drift-flux (or “drift velocity”) of the vapor relative to the liquid due to buoyancy effects. In general, the drift-flux models consist of two mass conservation equations corresponding to each of the two phases, and one equation for the conservation of the mixture momentum, and are particularly useful in the analysis of sedimentation, fluidization (batch, cocurrent and countercurrent), and so on ([@Ishii; @Wallis; @Zuber; @ZuberFindlay]).
The Cauchy problem to a simplified version of the viscous compressible liquid-gas two-phase flow model of drift-flux type in ${\Bbb{R}}^d$ ($d{\geqslant}2$), where the gas phase has not been taken into account in the momentum equation except the pressure term and the equal velocity of the liquid and gas flows has been assumed, reads $$\begin{aligned}
\label{eq.system}
\left\{\begin{aligned}
&\tilde{m}_t+{\mathrm{div}}(\tilde{m}{\mathbf{u}})=0,\\
&\tilde{n}_t+{\mathrm{div}}(\tilde{n}{\mathbf{u}})=0,\\
&(\tilde{m}{\mathbf{u}})_t+{\mathrm{div}}(\tilde{m}{\mathbf{u}}\otimes {\mathbf{u}})
+\nabla P(\tilde{m},\tilde{n})
=\tilde{\mu}\Delta {\mathbf{u}}+(\tilde{\mu}+\tilde{\lambda})\nabla {\mathrm{div}}{\mathbf{u}},
\end{aligned}
\right.\end{aligned}$$ with the initial data $$\begin{aligned}
\label{eq.data}
(\tilde{m},\tilde{n},{\mathbf{u}})|_{t=0}=(\tilde{m}_0,\tilde{n}_0,{\mathbf{u}}_0)(x), \quad \text{in }
{\Bbb{R}}^d,\end{aligned}$$ where $\tilde{m}=\alpha_l\rho_l$ and $\tilde{n}=\alpha_g\rho_g$ denote the liquid mass and the gas mass, respectively. The unknowns $\alpha_l$, $\alpha_g\in [0,1]$ denote the liquid and gas volume fractions, satisfying the fundamental relation: $\alpha_l+\alpha_g=1$. The unknown variables $\rho_l$ and $\rho_g$ denote the liquid and gas densities, satisfying the equations of states $\rho_l=\rho_{l,0}+(P-P_{l,0})/a_l^2$, $\rho_g=P/a_g^2$, where $a_l$ and $a_g$ denote the sonic speeds of the liquid and the gas, respectively, and $P_{l,0}$ and $\rho_{l,0}$ are the reference pressure and density given as constants. ${\mathbf{u}}$ denotes the mixed velocity of the liquid and the gas, and $P$ is the common pressure for both phases, which satisfies $$\begin{aligned}
\label{eq.pressure}
P(\tilde{m},\tilde{n})=C_0\left(-b(\tilde{m},\tilde{n})+\sqrt{b^2(\tilde{m},\tilde{n}) +c(\tilde{m},\tilde{n})}\right),\end{aligned}$$ with $C_0=a_l^2/2$, $k_0=\rho_{l,0}-P_{l,0}/a_l^2>0$, $a_0=a_g^2/a_l^2$ and $$\begin{aligned}
b(\tilde{m},\tilde{n})=k_0-\tilde{m}-a_0\tilde{n}, \quad c(\tilde{m},\tilde{n})=4k_0a_0\tilde{n}.\end{aligned}$$ $\tilde{\mu}$ and $\tilde{\lambda}$ are the viscosity constants, satisfying $$\begin{aligned}
\label{eq.mu.1}
\tilde{\mu}>0,\quad 2\tilde{\mu}+d\tilde{\lambda}{\geqslant}0.\end{aligned}$$
For the one-dimensional case, the existence and/or uniqueness of the global weak solution to the free boundary value problem was studied in [@EFF09; @EK09; @YZ09; @YZ10] where the liquid is incompressible and the gas is polytropic, and in [@EK08] where both of two fluids are compressible. However, there are few results for multi-dimensional cases except for some computational results [@PT07]. As a generalization of the results in [@EK08], the existence of the global solution to the 2D model was obtained in [@YZZ10] for small initial energies. In [@YZZ11], a blow-up criterion for the 2D model was proved in terms of the upper bound of the liquid mass for the strong solution in a smooth bounded domain.
One of the main results of the present paper is the existence and uniqueness of the global strong solution to the Cauchy problem – under the framework of Besov spaces, for all multi-dimensions $d{\geqslant}2$, provided that the initial data are close to a constant equilibrium state. The other result is the local well-posedness and the continuation criterion to the Cauchy problem with general initial data. Because of the similarity of the viscous liquid-gas two-phase flow model to the compressible Navier-Stokes equations, we can apply some ideas adopted in the proof of well-posedness for the compressible Navier-Stokes equations to deal with the two-phase flow model. It is Danchin who first makes important progress in applying the Littlewood-Paley theory and Besov spaces to sovle the existence and uniqueness for the compressible Navier-Stokes equations or barotropic viscous fluid in [@Danchin00; @Danchin07] and for the flows of compressible viscous and heat-conductive gases in [@Danchin01; @Danchin01a]. However, it is non-trivial to apply directly the ideas used in single-phase models into the two-phase models because the momentum equation is given only for the mixture and that the pressure involves the masses of two phases in a nonlinear way, which makes it rather difficult to obtain the estimates of the masses and the mixed velocity $(\tilde{m},\tilde{n},{\mathbf{u}})$ in Lebesgue spaces $L^p$ with respect to the time. In addition, it seems impossible to get the estimates of $\tilde{m}$ and $\tilde{n}$ from the system simultaneously due to the strong coupling among the corresponding terms, even if we change the variables $(\tilde{m},\tilde{n})$ linearly.
To overcome these difficulties in global well-posedness theory, we make use of a nonlinear variable transform so as to separate the two mass variables from each other, which enable us to decompose the original system into a transport equation and a coupled hyperbolic-parabolic system. To be more precisely, we first divide the momentum equation by $\tilde{m}$ (which supplies additional information) and take a new variable $n=a_0(\tilde{n}/\tilde{m}-{\bar{n}}/{\bar{m}})$ for some constants ${\bar{n}}$ and ${\bar{m}}$. This makes the resulted equation for $n$ a homogeneous transport equation with the velocity ${\mathbf{u}}$, and the expected estimates of the new variable depend only on the mixed velocity. Then, we remove the linear terms involving $n$ from the momentum equation so as to separate linearly the equation about $n$ from the others, which can be done by virtue of the variables changes with a careful choice of coefficient factors. Finally, to establish the *a priori* estimates for the global existence theory, we deal with the linearized system directly instead of separating the velocity into the compressible and incompressible parts.
As for the local well-posedness theory for general data, we need to reformulate the original system and deal with the resulted nonlinear system directly, and in terms of the improvement of the *a priori* estimates on the densities, we can generalize the local well-posedness result in [@BCDbook; @Danchin07] to the two-phase flow model with the specified pressure function.
Before stating the main results, we introduce some notations. Throughout the paper, $C$ (or $c$) stands for a harmless constant, and we sometimes use $A\lesssim B$ to stand for $A{\leqslant}CB$. $B^s$ and $B^{s,t}$ denote usual homogeneous Besov spaces and hybrid Besov spaces, respectively; ${\tilde{L}^\infty}(B^{s,t})$ and $\tilde{\mathcal{C}}(B^{s,t})$ are mixed time-spatial spaces, see the appendix for details. Let us now introduce the functional spaces which appear in the theorems.
For $T>0$ and $s\in{\Bbb{R}}$, we denote $$\begin{aligned}
E_T^s=\big\{(m,n,{\mathbf{u}}):\; &n\in \mathcal{C}([0,T];B^{s-1,s}({\Bbb{R}}^d))\\
&m\in\mathcal{C}([0,T];B^{s-1,s}({\Bbb{R}}^d))\cap L^1([0,T];B^{s+1,s}({\Bbb{R}}^d))\\
&{\mathbf{u}}\in\big(\mathcal{C}([0,T];B^{s-1}({\Bbb{R}}^d))\cap L^1([0,T];B^{s+1}({\Bbb{R}}^d))\big)^d\big\},\end{aligned}$$ and $$\begin{aligned}
{\|(m,n,{\mathbf{u}})\|}_{E_T^s}={\|n\|}_{{\tilde{L}^\infty}([0,T];B^{s-1,s})} &+{\|m\|}_{{\tilde{L}^\infty}([0,T];B^{s-1,s})} +{\|{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T];B^{s-1})} \\ &+{\|m\|}_{L^1([0,T];B^{s+1,s})}+{\|{\mathbf{u}}\|}_{L^1([0,T];B^{s+1})}.\end{aligned}$$
We use the notation $E^s$ if $T=+\infty$, changing $[0,T]$ into $[0,\infty)$ in the definition above.
Let $\alpha\in[0,1]$ and $T>0$, denote $$\begin{aligned}
F_T^\alpha:= &(\tilde{\mathcal{C}}([0,T];B^{d/2,d/2+\alpha}))^{1+1}\\
&\qquad\times(\tilde{\mathcal{C}}([0,T]; B^{d/2-1,d/2-1+\alpha})\cap L^1([0,T];B^{d/2+1,d/2+1+\alpha}))^d.\end{aligned}$$
Now, we state the global well-posedness results briefly as follows. For more information about the solution, one can see Theorem \[thm.2\] in the second section.
\[thm.1\] Let $d{\geqslant}2$, ${\bar{n}}{\geqslant}0$, ${\bar{m}}>(1-\mathrm{sgn}{\bar{n}})k_0$, $\tilde{\mu}>0$ and $2\tilde{\mu}+d\tilde{\lambda}{\geqslant}0$, in addition, $\tilde{\mu}+\tilde{\lambda}>0$ if $d=2$. There exist two positive constants $\sigma$ and $Q$ such that if $\tilde{m}_0-{\bar{m}}$, $\tilde{n}_0-{\bar{n}}\in
B^{d/2-1,d/2}$ and ${\mathbf{u}}_0\in B^{d/2-1}$ satisfying $$\begin{aligned}
\label{eq.thm.1}
{\|\tilde{m}_0-{\bar{m}}\|}_{B^{d/2-1,d/2}} +{\|\tilde{n}_0-{\bar{n}}\|}_{B^{d/2-1,d/2}}+{\|{\mathbf{u}}_0\|}_{B^{d/2-1}}{\leqslant}\sigma,
\end{aligned}$$ then the following results hold
[(i)]{} Existence: The system has a solution $(\tilde{m},\tilde{n},{\mathbf{u}})$ satisfying $$\tilde{m}-{\bar{m}},\; \tilde{n}-{\bar{n}}\in \mathcal{C}\left({\Bbb{R}}^+;B^{d/2-1,d/2}\right), \quad {\mathbf{u}}\in \mathcal{C}\left({\Bbb{R}}^+;B^{d/2-1}\right),$$ and moreover, $$\begin{aligned}
&{\|(a(\tilde{m}-{\bar{m}})+ba_0 (\tilde{n}/\tilde{m}-{\bar{n}}/{\bar{m}}),\tilde{n}/\tilde{m}-{\bar{n}}/{\bar{m}},{\mathbf{u}})\|}_{E^{d/2}}\\
{\leqslant}& Q\big({\|\tilde{m}_0-{\bar{m}}\|}_{B^{d/2-1,d/2}}+{\|\tilde{n}_0-{\bar{n}}\|}_{B^{d/2-1,d/2}} +{\|{\mathbf{u}}_0\|}_{B^{d/2-1}}\big),
\end{aligned}$$ where the constants $a$ and $b$ are defined by $$\begin{aligned}
\label{eq.ab}
\begin{aligned}
a= & \frac{1}{{\bar{m}}^2}\left(a_0{\bar{n}}+{\bar{m}}+\frac{({\bar{m}}-a_0{\bar{n}})({\bar{m}}-a_0{\bar{n}}-k_0)}{ \sqrt{({\bar{m}}+a_0{\bar{n}}-k_0)^2+4k_0a_0{\bar{n}}}}\right)>0, \\ b= & 1+\frac{({\bar{m}}+a_0{\bar{n}}+k_0)}{\sqrt{({\bar{m}}+a_0{\bar{n}}-k_0)^2+4k_0a_0{\bar{n}}}}>0.
\end{aligned}\end{aligned}$$
[(ii)]{} Uniqueness: Uniqueness holds in $\mathcal{C}\left({\Bbb{R}}^+; (B^{d/2-1,d/2})^{1+1}\times (B^{d/2})^d\right)$ if $d{\geqslant}3$. If $d=2$, one should also suppose that $\tilde{m}_0-{\bar{m}}$, $\tilde{n}_0-{\bar{n}}\in B^{{\varepsilon},1+{\varepsilon}}$ and ${\mathbf{u}}_0\in B^{{\varepsilon}}$ for a ${\varepsilon}\in(0,1)$, to get uniqueness in $\mathcal{C}({\Bbb{R}}^+; (B^{0,1})^{1+1}\times (B^1)^d)$.
For the general data bounded away from the infinity and the vacuum, we have the following local well-posedness theory (one can refer to Theorem \[thm.4\] for the corresponding statement in terms of new variables).
\[thm.3\] Let $d{\geqslant}2$, $\tilde{\mu}>0$, $2\tilde{\mu}+d\tilde{\lambda}{\geqslant}0$, the constants ${\bar{m}}>0$ and ${\bar{n}}{\geqslant}0$. Assume that $\tilde{m}_0^{-1}-{\bar{m}}^{-1}\in B^{d/2,d/2+1}$, $\tilde{n}_0-{\bar{n}}\in B^{d/2,d/2+1}$ and ${\mathbf{u}}_0\in B^{d/2-1,d/2}$. In addition, $\sup_{x\in{\Bbb{R}}^d} \tilde{m}_0(x)<\infty$. Then there exists a positive time $T$ such that the system has a unique solution $(\tilde{m},\tilde{n},{\mathbf{u}})$ on $[0,T]\times{\Bbb{R}}^d$ and that $(\tilde{m}^{-1}-{\bar{m}}^{-1},\tilde{n}-{\bar{n}},{\mathbf{u}})$ belongs to $F_T^1$ and satisfies $\sup_{(t,x)\in[0,T]\times{\Bbb{R}}^d} \tilde{m}(t,x)<\infty$.
We also have the following continuation criterion for the local existence of the solution (see also Proposition \[prop.cc\]).
Under the hypotheses of Theorem \[thm.3\], assume that the system has a solution $(\tilde{m},\tilde{n},{\mathbf{u}})$ on $[0,T)\times{\Bbb{R}}^d$ such that $(\tilde{m}^{-1}-{\bar{m}}^{-1},\tilde{n}-{\bar{n}},{\mathbf{u}})$ belongs to $F_{T'}^1$ for all $T'<T$ and satisfies $$\begin{aligned}
& \tilde{m}^{-1}-{\bar{m}}^{-1},\tilde{n}-{\bar{n}}\in L^\infty([0,T);B^{d/2,d/2+1}),\\
& \sup_{(t,x)\in [0,T)\times{\Bbb{R}}^d} \tilde{m}(t,x)<\infty, \quad \int_0^T {\|\nabla{\mathbf{u}}\|}_{\infty}dt<\infty.
\end{aligned}$$ Then, there exists some $T^*>T$ such that $(\tilde{m},\tilde{n},{\mathbf{u}})$ may be continued on $[0,T^*]\times{\Bbb{R}}^d$ to a solution of such that $(\tilde{m}^{-1}-{\bar{m}}^{-1},\tilde{n}-{\bar{n}},{\mathbf{u}})$ belongs to $F_{T^*}^1$.
The results of the present paper are independent of the special structure of the nonlinear pressure term $P$. Indeed, the similar results hold true as long as the term $\nabla P/\tilde{m}$ can be decomposed into a linear term involving the modified mass and some nonlinear terms, similarly as in the next section.
The rest of this paper is organized as follows. In Section \[sec.gwp\], we investigate the global well-posedness of the Cauchy problem. We first reformulate the system through changing variables in order to obtain *a priori* estimates in the subsection \[subsec.gwp1\]. In the subsection \[subsec.gwp2\], we are devoted to deriving *a priori* estimates for the transport equation and the linear coupled parabolic-hyperbolic system with convection terms. The subsection \[subsec.gwp3\] involves the proof of the existence and uniqueness of the solution. In Section \[sec.lwp\], we prove the local well-posedness of the problem through some subsections similar to the global ones. An appendix is devoted to recalling some properties of the Littlewood-Paley decomposition and Besov spaces which we have used in this paper.
Global well-posedness for small data {#sec.gwp}
====================================
Reformulation of the system {#subsec.gwp1}
---------------------------
Let ${\bar{n}}{\geqslant}0$ and ${\bar{m}}>(1-\mathrm{sgn}{\bar{n}})k_0$, we introduce new variables $n=a_0(\tilde{n}/\tilde{m}-{\bar{n}}/{\bar{m}})$ and $m=a(\tilde{m}-{\bar{m}})+b n$, i.e. $\tilde{m}={\bar{m}}+(m-b n)/a$ in order to cancel the linear terms involving one modified mass from the momentum equation, where $a$ and $b$ are positive constants defined in . We also denote $n_0=a_0(\tilde{n}_0/\tilde{m}_0-{\bar{n}}/{\bar{m}})$ and $m_0=a(\tilde{m}_0-{\bar{m}})+bn_0$ throughout the sections for the global well-posedness theory. Then, we have $$\begin{aligned}
\frac{P}{C_0}=\left(1+\frac{a_0{\bar{n}}}{{\bar{m}}}+n\right)\tilde{m}-k_0
+\sqrt{\left(\left(1+\frac{a_0{\bar{n}}}{{\bar{m}}}+n\right) \tilde{m}-k_0\right)^2 +4k_0\left(n+\frac{a_0{\bar{n}}}{{\bar{m}}}\right)\tilde{m}}.\end{aligned}$$ Taking the gradient of both sides, we get $$\begin{aligned}
\label{eq.2.1}
\frac{\nabla P}{C_0\tilde{m}}=&\nabla m+{\mathbf{H}}(m,n),\end{aligned}$$ where the nonlinear term is $$\begin{aligned}
{\mathbf{H}}(m,n):=&\frac{\nabla m-b\nabla n}{a^2{\bar{m}}^2\tilde{m}}\Big(-(a_0{\bar{n}}+{\bar{m}})m+(a{\bar{m}}^2+b(a_0{\bar{n}}+{\bar{m}}))n\Big)\\
&+(K(m,n)-K(0,0))\Big\{(n+\frac{a_0{\bar{n}}}{{\bar{m}}}+1)\tilde{m}\nabla n+(n+\frac{a_0{\bar{n}}}{{\bar{m}}}+1)^2\nabla \tilde{m}\\
&\qquad\qquad\qquad\qquad\qquad+k_0(n+\frac{a_0{\bar{n}}}{{\bar{m}}}-1)\frac{\nabla m-b\nabla n}{a\tilde{m}}+k_0\nabla n\Big\}\\
&+K(0,0)\left\{(n+\frac{a_0{\bar{n}}}{{\bar{m}}}+1)\frac{\nabla n(m-bn)}{a}+{\bar{m}}n\nabla n\right.\\
&\qquad\qquad\qquad+[n^2+2n(\frac{a_0{\bar{n}}}{{\bar{m}}}+1)]\frac{\nabla m-b\nabla n}{a}+\frac{k_0}{a\tilde{m}}n(\nabla m-b\nabla n)\\
&\qquad\qquad\qquad\left.-k_0(\frac{a_0{\bar{n}}}{{\bar{m}}}-1)\frac{(m-bn)(\nabla m-b\nabla n)}{a^2{\bar{m}}\tilde{m}}\right\}.\end{aligned}$$ Here, $$\begin{aligned}
K(m,n)=\frac{1}{\sqrt{\left[\left({\bar{m}}+\frac{m-bn}{a}\right)\left(n+\frac{a_0{\bar{n}}}{{\bar{m}}}+1\right) -k_0\right]^2 +4 k_0\left(n+\frac{a_0{\bar{n}}}{{\bar{m}}}\right)\left({\bar{m}}+\frac{m-bn}{a}\right)}},\end{aligned}$$ and $K(0,0)=1/\sqrt{({\bar{m}}+a_0{\bar{n}}-k_0)^2+4k_0a_0{\bar{n}}}>0$.
Therefore, with the new unknowns, we can rewrite the Cauchy problem of the system as follows $$\begin{aligned}
\label{eq.reform.system}
\left\{\begin{aligned}
&n_t+{\mathbf{u}}\cdot\nabla n=0,\\
&m_t+{\mathbf{u}}\cdot\nabla m+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}=F(m,n,{\mathbf{u}}),\\
&{\mathbf{u}}_t+{\mathbf{u}}\cdot\nabla{\mathbf{u}}-\mu\Delta{\mathbf{u}}-(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}+C_0\nabla m={\mathbf{G}}(m,n,{\mathbf{u}}),\\
&(m,n,{\mathbf{u}})|_{t=0}=(m_0,n_0,{\mathbf{u}}_0),
\end{aligned}\right.\end{aligned}$$ where $$\begin{gathered}
\mu= \tilde{\mu}/{\bar{m}},\quad \lambda=\tilde{\lambda}/{\bar{m}}, \quad F(m,n,{\mathbf{u}})=-(m-bn){\mathrm{div}}{\mathbf{u}},\\
{\mathbf{G}}(m,n,{\mathbf{u}})= -C_0{\mathbf{H}}(m,n) -\frac{m-bn}{a\tilde{m}}(\mu\Delta{\mathbf{u}}+(\mu+\lambda) \nabla{\mathrm{div}}{\mathbf{u}}).\end{gathered}$$ Note here that the first equation in is a homogeneous transport equation, the estimates of $n$ depend only on those of the velocity ${\mathbf{u}}$. The second and the third ones in consist of a coupled parabolic-hyperbolic system with the modified mass $m$ and mixed velocity ${\mathbf{u}}$ involved. Thus, with the help of the decomposition , the original system is decoupled into a transport equation for the modified gas flow and a coupled system for the motion of the modified liquid fluid.
We can get the following result for the reformulated system.
\[thm.2\] Let $d{\geqslant}2$, ${\bar{n}}{\geqslant}0$, ${\bar{m}}>(1-\mathrm{sgn}{\bar{n}})k_0$, $\mu>0$ and $2\mu+d\lambda{\geqslant}0$, in addition, $\mu+\lambda>0$ if $d=2$. There exist two positive constants $\eta$ and $Q$ such that if $m_0$, $n_0\in B^{d/2-1,d/2}$ and ${\mathbf{u}}_0\in B^{d/2-1}$ satisfying $$\label{thm.2.a}
{\|m_0\|}_{B^{d/2-1,d/2}}+{\|n_0\|}_{B^{d/2-1,d/2}}+{\|{\mathbf{u}}_0\|}_{B^{d/2-1}}{\leqslant}\eta,$$ then the following results hold:
[(i)]{} Existence: The system has a solution $(m,n,{\mathbf{u}})$ in $E^{d/2}$ which satisfies $${\|(m,n,{\mathbf{u}})\|}_{E^{d/2}}{\leqslant}Q\big({\|m_0\|}_{B^{d/2-1,d/2}}+{\|n_0\|}_{B^{d/2-1,d/2}}+{\|{\mathbf{u}}_0\|}_{B^{d/2-1}}\big).$$ It also belongs to the affine space $$(m_L,n_0,{\mathbf{u}}_L)+(\mathcal{C}^{1/2}({\Bbb{R}}^+;B^{d/2-1}))^{1+1}\times (\mathcal{C}^{1/8}({\Bbb{R}}^+; B^{d/2-5/4}))^d,$$ where $(m_L,{\mathbf{u}}_L)$ is the solution of the linear system $$\begin{aligned}
\label{eq.system.app3}
\left\{\begin{aligned}
&\partial_t m_L+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}_L=0,\\
&\partial_t{\mathbf{u}}_L-\mu\Delta{\mathbf{u}}_L -(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_L +C_0\nabla m_L=0,\\
&(m_L,{\mathbf{u}}_L)|_{t=0}=(m_0,{\mathbf{u}}_0).
\end{aligned}\right.
\end{aligned}$$
[(ii)]{} Uniqueness: Uniqueness holds in $E^{d/2}$ if $d{\geqslant}3$. If $d=2$, one should also suppose that $n_0,m_0\in B^{{\varepsilon},1+{\varepsilon}}$ and ${\mathbf{u}}_0\in B^{{\varepsilon}}$ for a ${\varepsilon}\in(0,1)$, to get uniqueness in $E^1$.
With the help of Theorem \[thm.2\], we can prove Theorem \[thm.1\] as follows.
[*Proof of Theorem \[thm.1\]*]{}. From the conditions, we have $n_0\in B^{d/2-1,d/2}$. In addition, from $\tilde{m}_0-{\bar{m}}\in
B^{d/2-1,d/2}$, we can derive $m_0=a(\tilde{m}_0-{\bar{m}})/{\bar{m}}+b
n_0\in B^{d/2-1,d/2}$. Since implies , the conclusion of Theorem \[thm.2\] follows for $(m,n,{\mathbf{u}})$. Changing back to the original variables $(\tilde{m},\tilde{n},{\mathbf{u}})$, it leads to Theorem \[thm.1\]. By Lemma \[lem.fgbesov\], it is easy to see that $\tilde{m}-{\bar{m}}$ and $\tilde{n}-{\bar{n}}$ also belong to $\mathcal{C}({\Bbb{R}}^+;B^{d/2-1}\cap B^{d/2})$.
[A priori]{} estimates for linear system with convection terms {#subsec.gwp2}
--------------------------------------------------------------
We first investigate some *a priori* estimates for the linear system with convection terms $$\begin{aligned}
\label{eq.reform.linear}
\left\{\begin{aligned}
&n_t+{\mathbf{v}}\cdot\nabla n=0,\\
&m_t+{\mathbf{v}}\cdot\nabla m+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}=F,\\
&{\mathbf{u}}_t+{\mathbf{v}}\cdot\nabla{\mathbf{u}}-\mu\Delta{\mathbf{u}}-(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}+C_0\nabla m={\mathbf{G}},\\
&(m,n,{\mathbf{u}})|_{t=0}=(m_0,n_0,{\mathbf{u}}_0).
\end{aligned}\right.\end{aligned}$$ We do not need to separate the velocity into the compressible and incompressible parts. In fact, we can prove the following proposition.
\[proposition\] Let $a>0$, ${\bar{m}}>0$, $s\in(1-d/2,d/2+1]$ and $s_1$, $s_2\in(-d/2,d/2+1]$ be constants. Assume ${\mathbf{v}}\in L^1([0,T];B^{d/2+1})$ and denote $V(t)=\int_0^t{\|{\mathbf{v}}(\tau)\|}_{B^{d/2+1}}d\tau$. Let $(m,n,{\mathbf{u}})$ be a solution of on $[0,T]$, then the following estimates hold: $$\begin{aligned}
\label{eq.prop.1}
{\|n\|}_{{\tilde{L}^\infty}([0,T]; B^{s_1,s_2})}{\leqslant}e^{CV(T)}{\|n_0\|}_{B^{s_1,s_2}},
\end{aligned}$$ and $$\begin{aligned}
\label{eq.prop.2}
\begin{aligned}
&{\|m\|}_{{\tilde{L}^\infty}([0,T]; B^{s-1,s})}+{\|{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T]; B^{s-1})}\\
&\qquad\qquad +{\|m\|}_{L^1([0,T]; B^{s+1,s})} +{\|{\mathbf{u}}\|}_{L^1([0,T]; B^{s+1})}\\
\lesssim &e^{CV(T)} \left({\|m_0\|}_{B^{s-1,s}} +{\|{\mathbf{u}}_0\|}_{B^{s-1}}+ {\|F\|}_{L^1([0,T]; B^{s-1,s})} +{\|{\mathbf{G}}\|}_{L^1([0,T]; B^{s-1})}\right).
\end{aligned}
\end{aligned}$$
*Step 1: Estimates for the homogeneous transport equation.* We derive the estimates for the first equation of in Besov spaces.
Applying the Littlewood-Paley operator ${\triangle_k }$ to the $\eqref{eq.reform.linear}_1$, it yields $$\begin{aligned}
\label{eq.transport}
\left\{\begin{aligned}
&\partial_t{\triangle_k n}+{\triangle_k }({\mathbf{v}}\cdot\nabla n)=0,\\
&{\triangle_k n}|_{t=0}={\triangle_k }n_0.
\end{aligned}\right.\end{aligned}$$
Taking the inner product of with ${\triangle_k n}$, we get for any $s_1,s_2\in (-d/2,1+d/2]$ $$\begin{aligned}
\frac{1}{2}\frac{d}{dt}{\|{\triangle_k n}\|}_2^2=-({\triangle_k }({\mathbf{v}}\cdot\nabla n),{\triangle_k n})
\lesssim\gamma_k2^{-k\varphi^{s_1,s_2}(k)}{\|{\mathbf{v}}\|}_{B^{d/2+1}} {\|n\|}_{B^{s_1,s_2}}{\|{\triangle_k n}\|}_2.\end{aligned}$$ It follows $$\begin{aligned}
\sum_k 2^{k\varphi^{s_1,s_2}(k)}{\|{\triangle_k n}\|}_2{\leqslant}{\|n_0\|}_{B^{s_1,s_2}}+C\int_0^t {\|{\mathbf{v}}\|}_{B^{d/2+1}}{\|n\|}_{B^{s_1,s_2}},\end{aligned}$$ which implies the desired estimate with the help of the Gronwall inequality.
*Step 2: Estimates for $(m,{\mathbf{u}})$.* Applying the Littlewood-Paley operator ${\triangle_k }$ to $\eqref{eq.reform.linear}_2$ and $\eqref{eq.reform.linear}_3$, we have $$\begin{aligned}
\label{eq.ns}
\left\{\begin{aligned}
&\partial_t{\triangle_k m}+{\triangle_k }({\mathbf{v}}\cdot\nabla m)+a{\bar{m}}{\mathrm{div}}{\triangle_k {\mathbf{u}}}={\triangle_k F},\\
&\partial_t{\triangle_k {\mathbf{u}}}+{\triangle_k }({\mathbf{v}}\cdot\nabla{\mathbf{u}})-\mu\Delta{\triangle_k {\mathbf{u}}} -(\mu+\lambda)\nabla{\mathrm{div}}{\triangle_k {\mathbf{u}}} +C_0\nabla {\triangle_k m}={\triangle_k {\mathbf{G}}}.
\end{aligned}\right.\end{aligned}$$
Taking the inner product of $\eqref{eq.ns}_1$ with ${\triangle_k m}$ and $-\Delta{\triangle_k m}$, and $\eqref{eq.ns}_2$ with ${\triangle_k {\mathbf{u}}}$, we obtain $$\begin{aligned}
\label{eq.ns.1}
\frac{1}{2}\frac{d}{dt}{\|{\triangle_k m}\|}_2^2&+({\triangle_k }({\mathbf{v}}\cdot\nabla m),{\triangle_k m})+a{\bar{m}}({\mathrm{div}}{\triangle_k {\mathbf{u}}},{\triangle_k m})=({\triangle_k F},{\triangle_k m}),\\
\frac{1}{2}\frac{d}{dt}{\|\nabla{\triangle_k m}\|}_2^2&+({\triangle_k }({\mathbf{v}}\cdot\nabla m),-\Delta{\triangle_k m})-a{\bar{m}}({\mathrm{div}}{\triangle_k {\mathbf{u}}},\Delta{\triangle_k m})\label{eq.ns.2}\\
&\qquad=-({\triangle_k F},\Delta{\triangle_k m}),\nonumber\\
\frac{1}{2}\frac{d}{dt}{\|{\triangle_k {\mathbf{u}}}\|}_2^2&+({\triangle_k }({\mathbf{v}}\cdot\nabla{\mathbf{u}}),{\triangle_k {\mathbf{u}}}) +\mu{\|\nabla{\triangle_k {\mathbf{u}}}\|}_2^2+(\mu+\lambda){\|{\mathrm{div}}{\triangle_k {\mathbf{u}}}\|}_2^2 \label{eq.ns.3}\\
&+C_0(\nabla {\triangle_k m},{\triangle_k {\mathbf{u}}})=({\triangle_k {\mathbf{G}}},{\triangle_k {\mathbf{u}}}).\nonumber\end{aligned}$$ For the intersected term, we have $$\begin{aligned}
\label{eq.ns.4}
\begin{aligned}
&\frac{d}{dt}({\triangle_k {\mathbf{u}}},\nabla{\triangle_k m})+({\triangle_k }({\mathbf{v}}\cdot\nabla{\mathbf{u}}),\nabla{\triangle_k m}) -({\triangle_k }({\mathbf{v}}\cdot\nabla m),{\mathrm{div}}{\triangle_k {\mathbf{u}}})\\
&\qquad-a{\bar{m}}{\|{\mathrm{div}}{\triangle_k {\mathbf{u}}}\|}_2^2+C_0{\|\nabla {\triangle_k m}\|}_2^2+(2\mu+\lambda)({\mathrm{div}}{\triangle_k {\mathbf{u}}},\Delta{\triangle_k m})\\
=&-({\triangle_k F},{\mathrm{div}}{\triangle_k {\mathbf{u}}})+({\triangle_k {\mathbf{G}}},\nabla{\triangle_k m}).
\end{aligned}\end{aligned}$$ Let $$\begin{aligned}
\alpha_k^2:=\frac{C_0}{a{\bar{m}}}{\|{\triangle_k m}\|}_2^2
+{\|{\triangle_k {\mathbf{u}}}\|}_2^2 +\frac{(2\mu+\lambda)A}{a{\bar{m}}}{\|\nabla {\triangle_k m}\|}_2^2+2A({\triangle_k {\mathbf{u}}},\nabla{\triangle_k m}).\end{aligned}$$ For $A=(\mu+\lambda)/(2a{\bar{m}})>0$, there exist two positive constants $c_1$ and $c_2$ such that $$\begin{aligned}
c_1\alpha_k^2{\leqslant}{\|{\triangle_k m}\|}_2^2 +{\|{\triangle_k {\mathbf{u}}}\|}_2^2+{\|\nabla {\triangle_k m}\|}_2^2{\leqslant}c_2\alpha_k^2,\end{aligned}$$ since we have, for $M\in(a{\bar{m}}/(2\mu+\lambda),2a{\bar{m}}/(\mu+\lambda))$, that $$\begin{aligned}
{|2({\triangle_k {\mathbf{u}}},\nabla{\triangle_k m})|}{\leqslant}M{\|{\triangle_k {\mathbf{u}}}\|}_2^2+{\|\nabla{\triangle_k m}\|}_2^2/M.\end{aligned}$$
Combining -, it yields, with the help of Lemma \[lem.innner\], that $$\begin{aligned}
&\frac{1}{2}\frac{d}{dt}\alpha_k^2 +\mu{\|\nabla {\triangle_k {\mathbf{u}}}\|}_2^2+(\mu+\lambda-a{\bar{m}}A){\|{\mathrm{div}}{\triangle_k {\mathbf{u}}}\|}_2^2
+C_0A{\|\nabla{\triangle_k m}\|}_2^2\nonumber\\
=&\frac{C_0}{a{\bar{m}}}({\triangle_k F},{\triangle_k m}) +({\triangle_k {\mathbf{G}}},{\triangle_k {\mathbf{u}}})-\frac{(2\mu+\lambda)A}{a{\bar{m}}}({\triangle_k F},\Delta{\triangle_k m}) -A({\triangle_k F},{\mathrm{div}}{\triangle_k {\mathbf{u}}})\nonumber\\
&+A({\triangle_k {\mathbf{G}}},\nabla{\triangle_k m}) -\frac{C_0}{a{\bar{m}}}({\triangle_k }({\mathbf{v}}\cdot\nabla m),{\triangle_k m}) -({\triangle_k }({\mathbf{v}}\cdot\nabla{\mathbf{u}}),{\triangle_k {\mathbf{u}}}) \nonumber\\
&+\frac{(2\mu+\lambda)A}{a{\bar{m}}}({\triangle_k }({\mathbf{v}}\cdot\nabla m),\Delta{\triangle_k m})+A({\triangle_k }({\mathbf{v}}\cdot\nabla{\mathbf{u}}),\nabla{\triangle_k m})\\
&+A(\nabla{\triangle_k }({\mathbf{v}}\cdot\nabla m),{\triangle_k {\mathbf{u}}})
\nonumber\\
\lesssim &({\|{\triangle_k F}\|}_2+{\|\nabla{\triangle_k F}\|}_2+{\|{\triangle_k {\mathbf{G}}}\|}_2) ({\|{\triangle_k m}\|}_2 +{\|{\triangle_k {\mathbf{u}}}\|}_2+{\|\nabla {\triangle_k m}\|}_2)\\
&+\gamma_k2^{-k\varphi^{s-1,s}(k)}{\|{\mathbf{v}}\|}_{B^{d/2+1}} {\|m\|}_{B^{s-1,s}}{\|{\triangle_k m}\|}_2\\
&+\gamma_k2^{-k(s-1)}{\|{\mathbf{v}}\|}_{B^{d/2+1}} {\|{\mathbf{u}}\|}_{B^{s-1}}{\|{\triangle_k {\mathbf{u}}}\|}_2\\
&+\gamma_k2^{-k(\varphi^{s-1,s}(k)-1)}{\|{\mathbf{v}}\|}_{B^{d/2+1}} {\|m\|}_{B^{s-1,s}}{\|\nabla{\triangle_k m}\|}_2\\
&+\gamma_k{\|{\mathbf{v}}\|}_{B^{d/2+1}}\left(2^{-k(s-1)}{\|\nabla{\triangle_k m}\|}_2
{\|{\mathbf{u}}\|}_{B^{s-1}}+2^{-k(\varphi^{s-1,s}(k)-1)} {\|m\|}_{B^{s-1,s}} {\|{\triangle_k {\mathbf{u}}}\|}_2\right)\\
\lesssim &\Big({\|{\triangle_k F}\|}_2+{\|\nabla{\triangle_k F}\|}_2+{\|{\triangle_k {\mathbf{G}}}\|}_2 +\gamma_k2^{-k(s-1)}{\|{\mathbf{v}}\|}_{B^{d/2+1}}( {\|m\|}_{B^{s-1,s}}+{\|{\mathbf{u}}\|}_{B^{s-1}})\Big) \\
&\quad\times({\|{\triangle_k m}\|}_2 +{\|{\triangle_k {\mathbf{u}}}\|}_2+{\|\nabla {\triangle_k m}\|}_2).\end{aligned}$$ Thus, it follows $$\begin{aligned}
&\frac{1}{2}\frac{d}{dt}\alpha_k^2+c_0\min(2^{2k},1)\alpha_k^2\\
\lesssim& \gamma_k2^{-k(s-1)}\Big[{\|F\|}_{B^{s-1,s}} +{\|{\mathbf{G}}\|}_{B^{s-1}}+{\|{\mathbf{v}}\|}_{B^{d/2+1}}\Big( {\|m\|}_{B^{s-1,s}}+{\|{\mathbf{u}}\|}_{B^{s-1}}\Big)\Big]\alpha_k,\end{aligned}$$ which implies $$\begin{aligned}
&2^{k(s-1)}\alpha_k+c_0\int_0^t\min(2^{2k},1)2^{k(s-1)}\alpha_k(\tau)d\tau\\
{\leqslant}&2^{k(s-1)}\alpha_k(0)+C\gamma_k\int_0^t\left[{\|F\|}_{B^{s-1,s}} +{\|{\mathbf{G}}\|}_{B^{s-1}}+{\|{\mathbf{v}}\|}_{B^{d/2+1}} \sum_k2^{k(s-1)}\alpha_k\right].\end{aligned}$$ Thus, by the Gronwall inequality, we have $$\begin{aligned}
\label{eq.ns.5}
\begin{aligned}
&{\|m\|}_{{\tilde{L}^\infty}([0,T]; B^{s-1,s})}+{\|{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T]; B^{s-1})}\\
&\qquad\qquad+{\|m\|}_{L^1([0,T]; B^{s+1,s})} +{\|{\mathbf{u}}\|}_{L^1([0,T]; B^{s+1,s-1})}\\
\lesssim &e^{CV(T)} \left({\|m_0\|}_{B^{s-1,s}} +{\|{\mathbf{u}}_0\|}_{B^{s-1}}+ {\|F\|}_{L^1([0,T]; B^{s-1,s})} +{\|{\mathbf{G}}\|}_{L^1([0,T]; B^{s-1})}\right).
\end{aligned}\end{aligned}$$
*Step 3: The smoothing effect for ${\mathbf{u}}$.* By , we have $$\begin{aligned}
&\frac{1}{2}\frac{d}{dt}{\|{\triangle_k {\mathbf{u}}}\|}_2^2+C2^{2k}{\|{\triangle_k {\mathbf{u}}}\|}_2^2\\
\lesssim& {\|{\triangle_k {\mathbf{u}}}\|}_2(2^k{\|{\triangle_k m}\|}_2+{\|{\triangle_k {\mathbf{G}}}\|}_2+\gamma_k 2^{-k(s-1)} {\|{\mathbf{v}}\|}_{B^{d/2+1}}{\|{\mathbf{u}}\|}_{B^{s-1}}).\end{aligned}$$ It follows that $$\begin{aligned}
&\frac{d}{dt}\sum_{k{\geqslant}0}2^{k(s-1)}{\|{\triangle_k {\mathbf{u}}}\|}_2 +C\sum_{k{\geqslant}0}2^{k(s+1)}{\|{\triangle_k {\mathbf{u}}}\|}_2\\
\lesssim & \sum_{k{\geqslant}0}2^{k(s-1)}\left[2^k{\|{\triangle_k m}\|}_2+{\|{\triangle_k {\mathbf{G}}}\|}_2+\gamma_k 2^{-k(s-1)} {\|{\mathbf{u}}\|}_{B^{d/2+1}}{\|{\mathbf{u}}\|}_{B^{s-1}}\right]\\
\lesssim & \sum_{k{\geqslant}0}2^{ks}{\|{\triangle_k m}\|}_2 +{\|{\mathbf{G}}\|}_{B^{s-1}} +{\|{\mathbf{v}}\|}_{B^{d/2+1}}{\|{\mathbf{u}}\|}_{B^{s-1}},\end{aligned}$$ which implies, with the help of , that $$\begin{aligned}
&\int_0^t\sum_{k{\geqslant}0}2^{k(s+1)}{\|{\triangle_k {\mathbf{u}}}(\tau)\|}_2\\
\lesssim &{\|{\mathbf{u}}_0\|}_{B^{s-1}} +\int_0^t\sum_{k{\geqslant}0}2^{ks}{\|{\triangle_k m}(\tau)\|}_2d\tau +\int_0^t{\|{\mathbf{G}}(\tau)\|}_{B^{s-1}} d\tau\\
&+\sup_{\tau\in[0,t]} {\|{\mathbf{u}}(\tau)\|}_{B^{s-1}}\int_0^t{\|{\mathbf{v}}(\tau)\|}_{B^{d/2+1}}d\tau\\
\lesssim &e^{CV(t)} \left({\|m_0\|}_{B^{s-1,s}}+{\|{\mathbf{u}}_0\|}_{B^{s-1}}\int_0^t \left[{\|F(\tau)\|}_{B^{s-1,s}} +{\|{\mathbf{G}}(\tau)\|}_{B^{s-1}}\right]d\tau\right).\end{aligned}$$ Combining with , we get .
From the proof of Proposition \[proposition\], we immediately have
\[coro\] If a bounded operator $\mathcal{B}$ acts on the convection terms in , then the same estimates hold for the refined system $$\begin{aligned}
\label{eq.reform.linearB}
\left\{\begin{aligned}
&n_t+\mathcal{B}({\mathbf{v}}\cdot\nabla n)=0,\\
&m_t+\mathcal{B}({\mathbf{v}}\cdot\nabla m)+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}=F,\\
&{\mathbf{u}}_t+\mathcal{B}({\mathbf{v}}\cdot\nabla{\mathbf{u}})-\mu\Delta{\mathbf{u}}-(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}+C_0\nabla m={\mathbf{G}}.
\end{aligned}\right.\end{aligned}$$
Global existence and uniqueness of the solution {#subsec.gwp3}
-----------------------------------------------
*Step 1: Friedrich’s approximation.* Let $L_\ell^2$ be the set of $L^2$ functions spectrally supported in the annulus $\mathcal{C}_\ell:=\{\xi\in{\Bbb{R}}^d:\, 1/\ell{\leqslant}{|\xi|}{\leqslant}\ell\}$ endowed with the standard $L^2$ topology. In order to construct the classical Friedrichs approximation, we first define the Friedrichs projectors $({\mathbb{F}_\ell})_{\ell\in{\Bbb{N}}}$ by $$\begin{aligned}
{\mathbb{F}_\ell}f:={\mathscr{F}}^{-1}\mathbf{1}_{\mathcal{C}_\ell}(\xi){\mathscr{F}}f,\end{aligned}$$ for any $f\in L^2({\Bbb{R}}^d)$ where $\mathbf{1}_{\mathcal{C}_\ell}(\xi)$ denotes the characteristic function on the annulus $\mathcal{C}_\ell$. Then, we can define the following approximate system $$\begin{aligned}
\label{eq.system.app}
\left\{\begin{aligned}
&n_t^\ell+{\mathbb{F}_\ell}({\mathbb{F}_\ell}{\mathbf{u}}^\ell\cdot\nabla {\mathbb{F}_\ell}n^\ell)=0,\\
&m_t^\ell+{\mathbb{F}_\ell}({\mathbb{F}_\ell}{\mathbf{u}}^\ell\cdot\nabla{\mathbb{F}_\ell}m^\ell)+a{\bar{m}}{\mathrm{div}}{\mathbb{F}_\ell}{\mathbf{u}}^\ell=F^\ell,\\
&{\mathbf{u}}_t^\ell+{\mathbb{F}_\ell}({\mathbb{F}_\ell}{\mathbf{u}}^\ell\cdot\nabla{\mathbb{F}_\ell}{\mathbf{u}}^\ell)-\mu\Delta{\mathbb{F}_\ell}{\mathbf{u}}^\ell -(\mu+\lambda)\nabla{\mathrm{div}}{\mathbb{F}_\ell}{\mathbf{u}}^\ell+C_0\nabla {\mathbb{F}_\ell}m^\ell={\mathbf{G}}^\ell,\\
&(m^\ell,n^\ell,{\mathbf{u}}^\ell)|_{t=0}=(m_\ell,n_\ell,{\mathbf{u}}_\ell),
\end{aligned}\right.\end{aligned}$$ where $$\begin{aligned}
m_\ell=&{\mathbb{F}_\ell}m_0, \quad n_\ell={\mathbb{F}_\ell}n_0, \quad {\mathbf{u}}_\ell={\mathbb{F}_\ell}{\mathbf{u}}_0, \\
F^\ell=&{\mathbb{F}_\ell}F({\mathbb{F}_\ell}m^\ell,{\mathbb{F}_\ell}n^\ell,{\mathbb{F}_\ell}{\mathbf{u}}^\ell),\quad
{\mathbf{G}}^\ell={\mathbb{F}_\ell}{\mathbf{G}}({\mathbb{F}_\ell}m^\ell,{\mathbb{F}_\ell}n^\ell,{\mathbb{F}_\ell}{\mathbf{u}}^\ell).\end{aligned}$$
It is easy to check that it is an ordinary differential equation in $L_\ell^2\times L_\ell^2 \times (L_\ell^2)^d$ for every $\ell\in{\Bbb{N}}$. By the usual Cauchy-Lipschitz theorem, there is a strictly positive maximal time $T_\ell^*$ such that a unique solution $(m^\ell,n^\ell,{\mathbf{u}}^\ell)$ exists in $[0,T_\ell^*)$ which is continuous in time with value in $L_\ell^2\times L_\ell^2 \times (L_\ell^2)^d$, i.e. $(m^\ell,n^\ell,{\mathbf{u}}^\ell)\in \mathcal{C}([0,T_\ell^*); L_\ell^2\times L_\ell^2 \times (L_\ell^2)^d)$. As ${\mathbb{F}_\ell}^2={\mathbb{F}_\ell}$, we see that ${\mathbb{F}_\ell}(m^\ell,n^\ell,{\mathbf{u}}^\ell)$ is also a solution, so the uniqueness implies that ${\mathbb{F}_\ell}(m^\ell,n^\ell,{\mathbf{u}}^\ell)=(m^\ell,n^\ell,{\mathbf{u}}^\ell)$. Thus, this system can be rewritten as the following system $$\begin{aligned}
\label{eq.system.app1}
\left\{\begin{aligned}
&n_t^\ell+{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla n^\ell)=0,\\
&m_t^\ell+{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla m^\ell)+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}^\ell=F_1^\ell,\\
&{\mathbf{u}}_t^\ell+{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla{\mathbf{u}}^\ell)-\mu\Delta{\mathbf{u}}^\ell -(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}^\ell +C_0\nabla m^\ell={\mathbf{G}}_1^\ell,\\
&(m^\ell,n^\ell,{\mathbf{u}}^\ell)|_{t=0}=(m_\ell,n_\ell,{\mathbf{u}}_\ell),
\end{aligned}\right.\end{aligned}$$ where $$\begin{aligned}
F_1^\ell={\mathbb{F}_\ell}F( m^\ell, n^\ell,{\mathbf{u}}^\ell), \text{ and }
{\mathbf{G}}_1^\ell={\mathbb{F}_\ell}{\mathbf{G}}( m^\ell,n^\ell,{\mathbf{u}}^\ell).\end{aligned}$$
*Step 2: Uniform estimates.* Denote $$\begin{aligned}
E_0={\|m_0\|}_{B^{d/2-1,d/2}}+{\|n_0\|}_{B^{d/2}}+{\|{\mathbf{u}}_0\|}_{B^{d/2-1}},\end{aligned}$$ and $$\begin{aligned}
T_\ell:=\sup\{&T\in[0,T_\ell^*):
{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}{\leqslant}A \bar{C}E_0\},\end{aligned}$$ where $\bar{C}$ corresponds to the constant in Proposition \[proposition\] and $A>\max(2,1/\bar{C})$ is a constant. Thus, by the continuity, we have $T_\ell>0$.
Let $M_0$ be the continuity modulus of the embedding relation $B^{d/2}({\Bbb{R}}^d) \hookrightarrow L^\infty({\Bbb{R}}^d)$. We make the assumption $$\begin{aligned}
2(1+b)A\bar{C}M_0E_0{\leqslant}a{\bar{m}}.\end{aligned}$$ Then, it implies $$\begin{aligned}
{\|m^\ell\|}_{L^\infty([0,T]\times{\Bbb{R}}^d)}{\leqslant}& M_0{\|m^\ell\|}_{L^\infty([0,T];B^{d/2})}
{\leqslant}M_0{\|m^\ell\|}_{L^\infty([0,T];B^{d/2-1,d/2})}\\
{\leqslant}&A\bar{C}M_0E_0
{\leqslant}\frac{a{\bar{m}}}{2(1+b)}.\end{aligned}$$ Similarly, we have $$\begin{aligned}
{\|n^\ell\|}_{L^\infty([0,T]\times{\Bbb{R}}^d)}{\leqslant}\frac{a{\bar{m}}}{2(1+b)}.\end{aligned}$$ Then, $$\begin{aligned}
\tilde{m}(m^\ell,n^\ell)={\bar{m}}+\frac{m-bn}{a} \in\left[\frac{{\bar{m}}}{2},\frac{3{\bar{m}}}{2}\right].\end{aligned}$$
By Proposition \[proposition\], we have $$\begin{aligned}
&{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{S_T}\\
\lesssim &e^{C{\|{\mathbf{u}}^\ell\|}_{L^1([0,T];B^{d/2+1})}} \Big(E_0+{\|F_1^\ell\|}_{L^1([0,T];B^{d/2-1,d/2})} +{\|{\mathbf{G}}_1^\ell\|}_{L^1([0,T]; B^{d/2-1})}\Big).\end{aligned}$$
From Lemmas \[lem.comp\] and \[lem.fgbesov\], we get $$\begin{aligned}
&{\|F_1^\ell\|}_{L^1([0,T];B^{d/2-1,d/2})}\\
\lesssim &{\|m^\ell{\mathrm{div}}{\mathbf{u}}^\ell\|}_{L^1([0,T];B^{d/2-1,d/2})} +{\|n^\ell{\mathrm{div}}{\mathbf{u}}^\ell\|}_{L^1([0,T];B^{d/2-1,d/2})}\\
\lesssim &({\|m^\ell\|}_{L^\infty([0,T]; B^{d/2-1,d/2})} +{\|n^\ell\|}_{L^\infty([0,T]; B^{d/2-1,d/2})}) {\|{\mathrm{div}}{\mathbf{u}}^\ell\|}_{L^1([0,T];B^{d/2})}\\
\lesssim &({\|m^\ell\|}_{L^\infty([0,T]; B^{d/2-1,d/2})} +{\|n^\ell\|}_{L^\infty([0,T]; B^{d/2-1,d/2})}) {\|{\mathbf{u}}^\ell\|}_{L^1([0,T];B^{d/2+1})}\\
\lesssim &{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}^2.\end{aligned}$$ Thus, $$\begin{aligned}
&{\|\frac{m^\ell-bn^\ell}{a\tilde{m}(m^\ell,n^\ell)} (\mu\Delta{\mathbf{u}}^\ell+(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}^\ell) \|}_{L^1([0,T];B^{d/2-1})}\\
\lesssim &({\|n^\ell\|}_{L^\infty([0,T];B^{d/2})}+{\|m^\ell\|}_{L^\infty([0,T];B^{d/2})}){\|{\mathbf{u}}^\ell\|}_{L^1([0,T];B^{d/2+1})}\\
\lesssim &{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{S_T}^2.\end{aligned}$$ Similarly, we can get $$\begin{aligned}
{\|{\mathbf{H}}(m^\ell,n^\ell)\|}_{L^1([0,T];B^{d/2-1})}\lesssim & ({\|n^\ell\|}_{L^\infty([0,T];B^{d/2})}+{\|m^\ell\|}_{L^\infty([0,T];B^{d/2})})^2\\
\lesssim&{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}^2.\end{aligned}$$ Hence $$\begin{aligned}
{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}{\leqslant}&\bar{C}e^{\bar{C}{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}} \Big(E_0+C{\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}^2\Big)\\
{\leqslant}&\bar{C}e^{\bar{C}^2AE_0} (1+CA^2\bar{C}^2 E_0)E_0.\end{aligned}$$ Thus, we can choose $E_0$ so small that $$\begin{aligned}
\label{eq.small}
1+CA^2\bar{C}^2E_0{\leqslant}\frac{A^2}{A+2}, \quad e^{\bar{C}^2AE_0}{\leqslant}\frac{A+1}{A}\quad \text{and } 2(1+b)A\bar{C}M_0E_0{\leqslant}a{\bar{m}},\end{aligned}$$ which yields ${\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}{\leqslant}\frac{A+1}{A+2}A\bar{C}E_0$ for any $T<T_\ell$.
We claim that $T_\ell=T_\ell^*$. Indeed, if $T_\ell<T_\ell^*$, we have seen that ${\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_T^{d/2}}{\leqslant}\frac{A+1}{A+2}A\bar{C}E_0$. So by the continuity, for a sufficiently small constant $s>0$, we can obtain ${\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_{(T+s)}^{d/2}}{\leqslant}A\bar{C}E_0$ which contradicts with the definition of $T_\ell$.
Now, we show the approximate solution is a global one, i.e. $T_\ell^*=\infty$. We assume $T_\ell^*<\infty$, then we have shown ${\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{E_{T}^{d/2}}{\leqslant}A\bar{C}E_0$. As $$m^\ell\in L^\infty([0,T_\ell^*);B^{d/2-1,d/2}), n^\ell\in L^\infty([0,T_\ell^*);B^{d/2}) \text{ and } {\mathbf{u}}^\ell\in L^\infty([0,T_\ell^*);B^{d/2-1}),$$ it implies that $${\|(m^\ell,n^\ell,{\mathbf{u}}^\ell)\|}_{L^\infty([0,T_\ell^*);L_\ell^2)}<\infty.$$ Thus, we may extend the solution continuously beyond the time $T_\ell^*$ by the Cauchy-Lipschitz theorem. This contradicts the definition of $T_\ell^*$. Therefore, the solution $(m^\ell,n^\ell,{\mathbf{u}}^\ell)_{\ell\in{\Bbb{N}}}$ exists global in time.
*Step 3: Time derivatives.* For convenience, we split the approximate solution $(m^\ell,n^\ell,{\mathbf{u}}^\ell)$ into a solution of the linear system with initial data $(m_\ell,n_\ell,{\mathbf{u}}_\ell)$, and the discrepancy to that linear solution. More precisely, we denote by $(m_L^\ell,{\mathbf{u}}_L^\ell)$ the solution to the linear system $$\begin{aligned}
\label{eq.system.app2}
\left\{\begin{aligned}
&m_t^\ell+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}^\ell=0,\\
&{\mathbf{u}}_t^\ell-\mu\Delta{\mathbf{u}}^\ell -(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}^\ell +C_0\nabla m^\ell=0,\\
&(m^\ell,{\mathbf{u}}^\ell)|_{t=0}=(m_\ell,{\mathbf{u}}_\ell),
\end{aligned}\right.\end{aligned}$$ and $(m_D^\ell,n_D^\ell,{\mathbf{u}}_D^\ell)=(m^\ell-m_L^\ell,n^\ell-n_\ell,{\mathbf{u}}^\ell-{\mathbf{u}}_L^\ell)$.
It is clear that the definition of $(m_\ell,n_\ell,{\mathbf{u}}_\ell)$ implies $$\begin{aligned}
m_\ell\to m_0 \text{ in } B^{d/2-1,d/2}, \quad n_\ell\to n_0 \text{ in } B^{d/2-1,d/2}, \quad {\mathbf{u}}_\ell\to {\mathbf{u}}_0 \text{ in } B^{d/2-1}.\end{aligned}$$ From Corollary \[coro\], we have $$\begin{aligned}
(m_L^\ell,n_\ell,{\mathbf{u}}_L^\ell)\to ( m_L,n_0,{\mathbf{u}}_L) \text{ in } E^{d/2},\end{aligned}$$ where $m_L$ and ${\mathbf{u}}_L$ satisfy the linear system .
Now, we derive the uniform boundedness of the time derivatives of the discrepancy $(m_D^\ell,n_D^\ell,{\mathbf{u}}_D^\ell)$.
\[lem.time\] $((m_D^\ell,n_D^\ell,{\mathbf{u}}_D^\ell))_{\ell\in{\Bbb{N}}}$ is uniformly bounded in $$(\mathcal{C}^{1/2}({\Bbb{R}}^+;B^{d/2-1}))^{1+1}\times (\mathcal{C}^{1/8}({\Bbb{R}}^+; B^{d/2-5/4}))^d.$$
Since $$\begin{aligned}
\partial_t n_D^\ell=-{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla n^\ell),
\end{aligned}$$ we have $\partial_t n_D^\ell\in L^2({\Bbb{R}}^+;B^{d/2-1})$ since $n^\ell\in L^\infty({\Bbb{R}}^+;B^{d/2})$ and ${\mathbf{u}}^\ell\in L^2({\Bbb{R}}^+;B^{d/2})$ with the help of the interpolation theorem.
From the equation $$\begin{aligned}
\partial_t m_D^\ell=-{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla m^\ell)-a{\bar{m}}{\mathrm{div}}{\mathbf{u}}^\ell+a{\bar{m}}{\mathrm{div}}{\mathbf{u}}_L^\ell- {\mathbb{F}_\ell}((m^\ell-bn^\ell){\mathrm{div}}{\mathbf{u}}^\ell),\end{aligned}$$ it follows that $\partial_t m_D^\ell\in L^2({\Bbb{R}}^+;B^{d/2-1})$.
Recall that $$\begin{aligned}
\partial_t {\mathbf{u}}_D^\ell=&-{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla{\mathbf{u}}^\ell)+\mu\Delta{\mathbf{u}}^\ell+\mu\Delta{\mathbf{u}}_L^\ell +(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}^\ell-(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_L^\ell\\
&-C_0\nabla m^\ell-C_0\nabla m_L^\ell+{\mathbf{G}}_1^\ell,\end{aligned}$$ we can obtain $\partial_t {\mathbf{u}}_D^\ell\in (L^\infty+L^{8/3}+L^{8/7})({\Bbb{R}}^+;B^{d/2-5/4})$ through easy but tedious computations with the help of Lemmas \[lem.comp\], \[lem.fgbesov\] and \[lem.inter\].
Applying the Morrey embedding relation $W^{1,p}({\Bbb{R}})\subset C^{1-1/p}({\Bbb{R}})$ to the time variable for $1<p{\leqslant}\infty$, we obtain the desired result.
*Step 4: Compactness and convergence.* The proof of the existence of a solution is now standard. Indeed, we can use Arzelà-Ascoli theorem to get strong convergence of the approximate solutions. We need to localize the spatial space in order to utilize some compactness results of local Besov spaces (see [@BCDbook Chapter 2]). Let $(\chi_p)_{p\in{\Bbb{N}}}$ be a sequence of $\mathscr{D}({\Bbb{R}}^d)$ cut-off functions supported in the ball $B(0,p+1)$ of ${\Bbb{R}}^d$ and equal to $1$ in a neighborhood of $B(0,p)$. In view of Lemma \[lem.time\] and uniform estimates obtained in Step 2, we see that $((\chi_pm_D^\ell,\chi_pn_D^\ell,\chi_p{\mathbf{u}}_D^\ell))_{\ell\in{\Bbb{N}}}$ is bounded in $E^{d/2}$ and uniformly equi-continuous in $$\mathcal{C}\left([0,T]; (B^{d/2-1})^{1+1}\times (B^{d/2-5/4})^d\right)$$ for any $p\in{\Bbb{N}}$ and $T>0$. Moreover, the mapping $f\mapsto \chi_p f$ is compact from $B^{d/2-1,d/2}$ into $B^{d/2-1}$ and from $B^{d/2-1}$ into $B^{d/2-5/4}$.
Applying the Arzelà-Ascoli theorem to the family $((\chi_pm_D^\ell,\chi_pn_D^\ell,\chi_p{\mathbf{u}}_D^\ell))_{\ell\in{\Bbb{N}}}$ on the time interval $[0,p]$, then we use the Cantor diagonal process. This finally provides us with a distribution $(m_D,n_D,{\mathbf{u}}_D)$ continuous in time with values in $(B^{d/2-1})^{1+1}\times
(B^{d/2-5/4})^d$ and a subsequence (which we still denote by the same notation) such that we have for all $p\in{\Bbb{N}}$ $$(\chi_pm_D^\ell,\chi_pn_D^\ell,\chi_p{\mathbf{u}}_D^\ell)
\to (\chi_pm_D,\chi_pn_D,\chi_p{\mathbf{u}}_D), \text{ as } \ell\to\infty$$ in $\mathcal{C}([0,p]; (B^{d/2-1})^{1+1}\times (B^{d/2-5/4})^d)$. This obviously implies that $(m_D^\ell,n_D^\ell,{\mathbf{u}}_D^\ell)$ tends to $(m_D,n_D,{\mathbf{u}}_D)$ in $\mathscr{D}'({\Bbb{R}}^+\times{\Bbb{R}}^d)$.
Coming back to the uniform estimates and Lemma \[lem.time\], we further obtain that $(m_D,n_D,{\mathbf{u}}_D)$ belongs to $E^{d/2}$ and to $(\mathcal{C}^{1/2}({\Bbb{R}}^+;B^{d/2-1}))^{1+1}\times
(\mathcal{C}^{1/8}({\Bbb{R}}^+; B^{d/2-5/4}))^d$. The convergence results stemming from this last result and the interpolation argument enable us to pass to the limit in $\mathscr{D}'({\Bbb{R}}^+\times{\Bbb{R}}^d)$ in the system and to prove that $(m,n,{\mathbf{u}}):=(m_L,n_L,{\mathbf{u}}_L)+(m_D,n_D,{\mathbf{u}}_D)$ is indeed a solution of with the initial data. Since it is just a matter of doing tedious verifications, we omit the details.
*Step 5: Continuities in time.* The continuity of ${\mathbf{u}}$ is straightforward. Indeed, from the third equation of , we have ${\mathbf{u}}_t\in
(L^1+L^2)({\Bbb{R}}^+;(B^{d/2-1})^d)$ which implies ${\mathbf{u}}\in\mathcal{C}({\Bbb{R}}^+;(B^{d/2-1})^d)$ in view of the Morrey embedding and the embedding relation $W^{1,1}({\Bbb{R}})\subset
\mathcal{C}({\Bbb{R}})$. Consequently, the continuity of $n$ in time is obtained from . For $m$, it is easily to see that $m_t\in L^2({\Bbb{R}}^+;B^{d/2-1})\cap L^1({\Bbb{R}}^+;B^{d/2})$ from the second equation of which yields $m\in
\mathcal{C}({\Bbb{R}}^+;B^{d/2-1,d/2})$ by the embeddings mentioned above.
*Step 6: Uniqueness.* Next, we prove the uniqueness of solutions. Let $(m_1,n_1,{\mathbf{u}}_1)$ and $(m_2,n_2,{\mathbf{u}}_2)$ be two solutions of in $E_T^{d/2}$ with the same initial data. Denote $(\delta m,\delta
n,\delta{\mathbf{u}})=(m_2-m_1,n_2-n_1,{\mathbf{u}}_2-{\mathbf{u}}_1)$. Then they satisfy $$\begin{aligned}
\left\{\begin{aligned}
&\partial_t\delta n+{\mathbf{u}}_2\cdot\nabla \delta n=-\delta{\mathbf{u}}\cdot\nabla n_1,\\
&\partial_t\delta m+{\mathbf{u}}_2\cdot\nabla \delta m+a{\bar{m}}{\mathrm{div}}\delta{\mathbf{u}}=-\delta{\mathbf{u}}\cdot\nabla m_1+\delta F,\\
&\partial_t\delta{\mathbf{u}}+{\mathbf{u}}_2\cdot\nabla\delta{\mathbf{u}}-\mu\Delta\delta{\mathbf{u}}-(\mu+\lambda)\nabla{\mathrm{div}}\delta{\mathbf{u}}+C_0\nabla\delta m=-\delta{\mathbf{u}}\cdot\nabla {\mathbf{u}}_1+\delta{\mathbf{G}},\\
&(\delta m,\delta n,\delta{\mathbf{u}})|_{t=0}=(0,0,\mathbf{0}),
\end{aligned}\right.\end{aligned}$$ where $\delta F=F(m_2,n_2,{\mathbf{u}}_2)-F(m_1,n_1,{\mathbf{u}}_1)$ and $\delta {\mathbf{G}}={\mathbf{G}}(m_2,n_2,{\mathbf{u}}_2)-{\mathbf{G}}(m_1,n_1,{\mathbf{u}}_1)$.
We first consider the case $d{\geqslant}3$. Similar to the derivation of , we can get for $t\in[0,T]$ with the help of $$\begin{aligned}
{\|\delta n\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-2,d/2-1})}{\leqslant}& e^{C\int_0^T{\|{\mathbf{u}}_2\|}_{B^{d/2+1}}d\tau}{\|\delta{\mathbf{u}}\|}_{L^1([0,T];B^{d/2})} {\|n_1\|}_{L^\infty([0,T];B^{d/2-1,d/2})}\nonumber\\
\lesssim& e^{C\int_0^T{\|({\mathbf{u}}_1,{\mathbf{u}}_2)\|}_{B^{d/2+1}}d\tau}{\|n_0\|}_{B^{d/2-1,d/2}} {\|\delta{\mathbf{u}}\|}_{L^1([0,T];B^{d/2})} .\end{aligned}$$ By Lemma \[lem.comp\] and \[lem.fgbesov\], we have $$\begin{aligned}
&{\|\delta m\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-2,d/2-1})}+{\|\delta{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T];B^{d/2-2})}\\
&\qquad\qquad+{\|\delta m\|}_{L^1([0,T]; B^{d/2,d/2-1})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T];B^{d/2})}\nonumber\\
\lesssim &e^{C\int_0^T{\|{\mathbf{u}}_2\|}_{B^{d/2+1}}d\tau} \big[{\|\delta{\mathbf{u}}\cdot\nabla m_1\|}_{L^1([0,T]; B^{d/2-2,d/2-1})} +{\|\delta F(\tau)\|}_{L^1([0,T]; B^{d/2-2,d/2-1})}\\
&\qquad\qquad+{\|\delta{\mathbf{u}}\cdot\nabla{\mathbf{u}}_1\|}_{L^1([0,T]; B^{d/2-2})} +{\|\delta{\mathbf{G}}(\tau)\|}_{L^1([0,T]; B^{d/2-2})}\big]\\
\lesssim & e^{C\int_0^T{\|{\mathbf{u}}_2\|}_{B^{d/2+1}}d\tau}({\|\delta{\mathbf{u}}\|}_{L^1([0,T];B^{d/2})} {\|m_1\|}_{{\tilde{L}^\infty}([0,T];B^{d/2-1,d/2})}\\
&+{\|(\delta m,\delta n)\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-2,d/2-1})} {\|{\mathbf{u}}_2\|}_{L^1([0,T]; B^{d/2+1})}\\
&+{\|(m_1,n_1)\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-1,d/2})}{\|\delta{\mathbf{u}}\|}_{L^1([0,T]; B^{d/2})}\\
&+{\|\delta{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-2})}{\|{\mathbf{u}}_1\|}_{L^1([0,T]; B^{d/2+1})}+{\|\delta{\mathbf{G}}(\tau)\|}_{L^1([0,T]; B^{d/2-2})}\\
\lesssim & e^{C{\|{\mathbf{u}}_2\|}_{L^1([0,T]; B^{d/2+1})}} \Big(\big(1+{\|(m_2,n_2)\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-1,d/2})}\big)\\
&\qquad\qquad\qquad\qquad \times{\|(m_1,n_1)\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-1,d/2})}+Z(T)\Big){\|(\delta m,\delta n,\delta{\mathbf{u}})\|}_{S_T^{d/2-1}},\end{aligned}$$ where $\limsup_{T\to 0^+}Z(T)=0$. Thus, $$\begin{aligned}
&{\|(\delta m,\delta n,\delta{\mathbf{u}})\|}_{E_T^{d/2-1}}\\
{\leqslant}& C e^{C{\|({\mathbf{u}}_1,{\mathbf{u}}_2)\|}_{L^1([0,T]; B^{d/2+1})}} \Big((1+{\|(m_2,n_2)\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-1,d/2})})\\
&\times{\|(m_1,n_1)\|}_{{\tilde{L}^\infty}([0,T]; B^{d/2-1,d/2})}+E_0+Z(T)\Big){\|(\delta m,\delta n,\delta{\mathbf{u}})\|}_{E_T^{d/2-1}}.\end{aligned}$$ We take $E_0$ small enough such that it satisfies the condition $2C(1+CA\bar{C}E_0)A\bar{C}E_0+E_0<1/4$ and , and choose $T>0$ so small that $C{\|({\mathbf{u}}_1,{\mathbf{u}}_2)\|}_{L^1([0,T]; B^{d/2+1})}{\leqslant}\ln 2$ and $Z(T)<1/2$, then it follows ${\|(\delta m,\delta n,\delta{\mathbf{u}})\|}_{E_T^{d/2-1}}\equiv 0$. Hence, $(m_1,n_1,{\mathbf{u}}_1)(t)=(m_2,n_2,{\mathbf{u}}_2)(t)$ on $[0,T]$. By a standard argument (e.g. [@Danchin00]), we can conclude that $(m_1,n_1,{\mathbf{u}}_1)(t)=(m_2,n_2,{\mathbf{u}}_2)(t)$ on ${\Bbb{R}}^+$.
For the case $d=2$, we have to raise the regularity of the spaces. Thus, we also suppose that $m_0,n_0\in B^{{\varepsilon},1+{\varepsilon}}$ and ${\mathbf{u}}_0\in B^{{\varepsilon}}$ for a ${\varepsilon}\in(0,1)$. By the same process, we can prove the existence of solution $(m,n,{\mathbf{u}})$ in the space $E^{1+{\varepsilon}}$ provided the norms of initial data is sufficiently small. Then, in the same way as in the case $d{\geqslant}3$, we may prove the uniqueness of solutions in the space $E^{{\varepsilon}}$ (of course, holds in $E^1$). We omit the details.
Local well-posedness for large data {#sec.lwp}
===================================
Reformulation of the system {#reformulation-of-the-system}
---------------------------
We change variables to $\rho={\bar{m}}(\tilde{m}^{-1}-{\bar{m}}^{-1})$ and $g=\tilde{n}-{\bar{n}}$. Then we can reformulate the system - as $$\begin{aligned}
\label{eq.system.2}
\left\{\begin{aligned}
&\rho_t+{\mathbf{u}}\cdot\nabla\rho=(\rho+1){\mathrm{div}}{\mathbf{u}},\\
&g_t+{\mathbf{u}}\cdot\nabla g=-(g+{\bar{n}}){\mathrm{div}}{\mathbf{u}},\\
&{\mathbf{u}}_t+{\mathbf{u}}\cdot\nabla{\mathbf{u}}-(1+\rho)(\mu\Delta{\mathbf{u}}+(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}) +Q(\rho,g)=0,\\
&(\rho,g,{\mathbf{u}})|_{t=0}=(\rho_0,g_0,{\mathbf{u}}_0),
\end{aligned}\right.\end{aligned}$$ where $\rho_0={\bar{m}}(\tilde{m}_0^{-1}-{\bar{m}}^{-1})$, $g_0=\tilde{n}_0-{\bar{n}}$ and $$\begin{aligned}
Q(\rho,g):=&{\bar{m}}^{-1}(1+\rho)\nabla P({\bar{m}}/(1+\rho),g+{\bar{n}})\\
=&\frac{\rho\nabla \rho}{\rho+1}-\nabla\rho+\frac{a_0}{{\bar{m}}}(\rho+1)\nabla g+B(\rho,g)\Big[-\frac{{\bar{m}}}{(\rho+1)^2}\nabla\rho +\frac{k_0-a_0{\bar{n}}}{\rho+1}\nabla\rho\\
&+\frac{a_0g\nabla\rho}{\rho+1} +\frac{a_0^2}{{\bar{m}}}(g+g\rho)\nabla g
+\frac{a_0(k_0+{\bar{m}}+a_0{\bar{n}})}{{\bar{m}}}\nabla g+\frac{a_0k_0+a_0^2{\bar{n}}}{{\bar{m}}}\rho\nabla g\Big],\end{aligned}$$ with $$\begin{aligned}
B(\rho,g):=\Big[\big(\frac{{\bar{m}}}{\rho+1}+a_0(g+{\bar{n}})-k_0\big)^2+4k_0a_0(g+{\bar{n}})\Big]^{-1/2}.\end{aligned}$$
We now state the result for the local theory for general data bounded away from the vacuum as follows.
\[thm.4\] Let $d{\geqslant}2$, $\mu>0$, $2\mu+d\lambda{\geqslant}0$, the constants ${\bar{m}}>0$ and ${\bar{n}}{\geqslant}0$. Assume that $\rho_0\in B^{d/2,d/2+1}$, $g_0\in B^{d/2,d/2+1}$ and ${\mathbf{u}}_0\in B^{d/2-1,d/2}$. In addition, $\inf\limits_{x\in{\Bbb{R}}^d} \rho_0(x)>-1$. Then there exists a positive time $T$ such that the system has a unique solution $(\rho,g,{\mathbf{u}})$ on $[0,T]\times{\Bbb{R}}^d$ which belongs to $$(\tilde{\mathcal{C}}([0,T];B^{d/2,d/2+1}))^{1+1}\times(\tilde{\mathcal{C}}([0,T]; B^{d/2-1,d/2})\cap L^1([0,T];B^{d/2+1,d/2+2}))^d,$$ and satisfies $\inf\limits_{(t,x)\in[0,T]\times{\Bbb{R}}^d} \rho(t,x)>-1$.
A priori Estimates
------------------
Now, let us recall some estimates for the following parabolic system which is obtained by linearizing the momentum equation $$\begin{aligned}
\left\{\begin{aligned}
&{\mathbf{u}}_t+{\mathbf{v}}\cdot\nabla{\mathbf{u}}+{\mathbf{u}}\cdot\nabla{\mathbf{w}}-b(t,x)(\mu\Delta{\mathbf{u}}+(\mu+\lambda) \nabla{\mathrm{div}}{\mathbf{u}}) =f,\\
&{\mathbf{u}}|_{t=0}={\mathbf{u}}_0,
\end{aligned}\right.\end{aligned}$$ which had been studied in [@BCDbook; @Danchin07]. Precisely, we have the following lemma (cf. [@BCDbook Proposition 10.12]).
\[lem.moment\] Let $\alpha\in(0,1]$, $s\in(-d/2,d/2]$, $\underline{\nu}=\min(\mu,\lambda+2\mu)$ and $\bar{\nu}=\mu+{|\mu+\lambda|}$. Assume that $b=1+\rho$ with $\rho\in L^\infty([0,T]; B^{d/2+\alpha})$ and that $$\begin{aligned}
b_*:=\inf_{(t,x)\in[0,T]\times{\Bbb{R}}^d} b(t,x)>0.
\end{aligned}$$ There exist a universal constant $\kappa$, and a constant $C$ depending only on $d$, $\alpha$ and $s$, such that for all $t\in[0,T]$, $$\begin{aligned}
&{\|{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,t];B^{s})}+\kappa b_*\underline{\nu}{\|{\mathbf{u}}\|}_{L^1([0,t];B^{s+2})}\\
{\leqslant}&\left({\|{\mathbf{u}}_0\|}_{B^{s}} +{\|f\|}_{L^1([0,t];B^{s})}\right)\\
&\times\exp\left(C\int_0^t\Big({\|{\mathbf{v}}\|}_{B^{d/2+1}}+{\|{\mathbf{w}}\|}_{B^{d/2+1}} +(b_*\underline{\nu})^{1-2/\alpha}\bar{\nu}^{2/\alpha} {\|\rho\|}_{B^{d/2+\alpha}}^{2/\alpha}\Big)d\tau\right).
\end{aligned}$$ If ${\mathbf{v}}$ and ${\mathbf{w}}$ depend linearly on ${\mathbf{u}}$, then the above inequality is true for all $s\in(0,d/2+\alpha]$, and the argument of the exponential term may be replaced with $$\begin{aligned}
C\int_0^t\left({\|\nabla{\mathbf{u}}\|}_\infty+(b_*\underline{\nu})^{1-2/\alpha} \bar{\nu}^{2/\alpha}{\|\rho\|}_{B^{d/2+\alpha}}^{2/\alpha}\right)d\tau.
\end{aligned}$$
For the mass equations, we only need to study the following equation with two constants $\theta\in{\Bbb{R}}$ and $\beta>0$ $$\begin{aligned}
\label{eq.trans}
\left\{\begin{aligned}
&h_t+{\mathbf{v}}\cdot\nabla h=\theta(h+\beta){\mathrm{div}}{\mathbf{v}},\\
&h|_{t=0}=h_0.
\end{aligned}\right.\end{aligned}$$
\[prop.2\] Let $s\in(-d/2,d/2+1]$, $T>0$, $\theta\in{\Bbb{R}}$ and $\beta{\geqslant}0$ be constants. Assume that $h_0\in B^{d/2}$, ${\mathbf{v}}\in L^1([0,T); B^{d/2+1})$ and $a$ satisfies . There exists a constant $C$ depending only on $d$ such that for all $t\in[0,T]$, we have $$\begin{aligned}
\label{eq.mass.1}
{\|h\|}_{{\tilde{L}^\infty}([0,t];B^{d/2})}{\leqslant}e^{C(1+2{|\theta|})\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau} \left({\|h_0\|}_{B^{d/2}}+\frac{\beta}{1+2{|\theta|}}\right) -\frac{\beta}{1+2{|\theta|}},\end{aligned}$$ and $$\begin{aligned}
\label{eq.mass.1a}
\begin{aligned}
&{\|h\|}_{{\tilde{L}^\infty}([0,t];B^{s})}
{\leqslant}e^{C(1+{|\theta|})\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau}\Big({\|h_0\|}_{B^{s}}
\\ &+C{|\theta|}\Big[e^{C(1+2{|\theta|})\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau} \Big({\|h_0\|}_{B^{d/2}}+\frac{\beta}{1+2{|\theta|}}\Big) +\frac{2{|\theta|}\beta}{1+2{|\theta|}}\Big]\int_0^t{\|{\mathbf{v}}\|}_{B^{s+1}}d\tau\Big).
\end{aligned}\end{aligned}$$
Applying the operator ${\triangle_k }$ to yields $$\begin{aligned}
\partial_t{\triangle_k h}+{\triangle_k }({\mathbf{v}}\cdot\nabla h)=\theta{\triangle_k }((h+\beta){\mathrm{div}}{\mathbf{v}}).
\end{aligned}$$ Taking the $L^2$ inner product with ${\triangle_k h}$, we get, with the help of Lemmas \[lem.fgbesov\] and \[lem.innner\], that $$\begin{aligned}
&\frac{1}{2}\frac{d}{dt}{\|{\triangle_k h}\|}_2^2\\
=&-({\triangle_k }({\mathbf{v}}\cdot\nabla a),{\triangle_k h})+\theta({\triangle_k }((h+\beta){\mathrm{div}}{\mathbf{v}}),{\triangle_k h})\\
\lesssim&\gamma_k 2^{-ks}{\|{\mathbf{v}}\|}_{B^{d/2+1}}{\|h\|}_{B^s}{\|{\triangle_k h}\|}_2 +{|\theta|}\gamma_k 2^{-ks}({\|h{\mathrm{div}}{\mathbf{v}}\|}_{B^{s}}+\beta{\|{\mathrm{div}}{\mathbf{v}}\|}_{B^{s}}){\|{\triangle_k h}\|}_2\\
\lesssim& \gamma_k 2^{-ks}((1+{|\theta|}){\|h\|}_{B^{s}}{\|{\mathbf{v}}\|}_{B^{d/2+1}} +({\|h\|}_{B^{d/2}}+\beta){|\theta|}{\|{\mathbf{v}}\|}_{B^{s+1}}){\|{\triangle_k h}\|}_2.
\end{aligned}$$ Eliminating the factor ${\|{\triangle_k h}\|}_2$ from both sides and integrating in the time, we have $$\begin{aligned}
{\|{\triangle_k h}\|}_2{\leqslant}& {\|{\triangle_k h_0}\|}_2\\
&+C\gamma_k\int_0^t 2^{-ks}((1+{|\theta|}){\|h\|}_{B^{s}}{\|{\mathbf{v}}\|}_{B^{d/2+1}} +({\|h\|}_{B^{d/2}}+\beta){|\theta|}{\|{\mathbf{v}}\|}_{B^{s+1}})d\tau.\end{aligned}$$ It follows, for any $k\in{\mathbb Z}$ and any $t\in[0,T]$, that $$\begin{aligned}
\label{eq.mass.4}
\begin{aligned}
&2^{ks}{\|{\triangle_k h}\|}_2{\leqslant}2^{ks}{\|{\triangle_k h_0}\|}_2\\
&\qquad\qquad+C\gamma_k\int_0^t( (1+{|\theta|}){\|h\|}_{B^{s}}{\|{\mathbf{v}}\|}_{B^{d/2+1}} +({\|h\|}_{B^{d/2}}+\beta){|\theta|}{\|{\mathbf{v}}\|}_{B^{s+1}})d\tau.
\end{aligned}\end{aligned}$$ Summing up on $k\in {\mathbb Z}$, it yields $$\begin{aligned}
{\|h\|}_{{\tilde{L}^\infty}([0,T];B^{s})}{\leqslant}&{\|h_0\|}_{B^{s}}\\
&+\int_0^t C[(1+{|\theta|}){\|{\mathbf{v}}\|}_{B^{d/2+1}}{\|h\|}_{B^{s}} +({\|h\|}_{B^{d/2}}+\beta){|\theta|}{\|{\mathbf{v}}\|}_{B^{s+1}}]d\tau.\end{aligned}$$
By the Gronwall inequality, we have for $s=d/2$ and then for any $s\in(-d/2,d/2+1]$ $$\begin{aligned}
&{\|h\|}_{{\tilde{L}^\infty}([0,t];B^{s})}\\
{\leqslant}& e^{C(1+{|\theta|})\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau} \left({\|h_0\|}_{B^{s}}+C\int_0^t{|\theta|}({\|h\|}_{B^{d/2}}+\beta) {\|{\mathbf{v}}\|}_{B^{s+1}}d\tau\right)\\
{\leqslant}&e^{C(1+{|\theta|})\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau}\Big({\|h_0\|}_{B^{s}} \\ &+C{|\theta|}\left[e^{C(1+2{|\theta|})\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau} \Big({\|h_0\|}_{B^{d/2}}+\frac{\beta}{1+2{|\theta|}}\Big) +\frac{2{|\theta|}\beta}{1+2{|\theta|}}\right]\int_0^t{\|{\mathbf{v}}\|}_{B^{s+1}}d\tau\Big).\end{aligned}$$ This completes the proofs.
In general, for the transport equation $$\begin{aligned}
\left\{\begin{aligned}
&h_t+{\mathbf{v}}\cdot\nabla h=f,\\
&h(0)=h_0,
\end{aligned}\right.\end{aligned}$$ we can get, in a similar way with Proposition \[prop.2\], that
\[prop.3\] Let $s_1,s_2\in(-d/2,d/2+1]$ and $T>0$. Then it holds for $t\in[0,T]$ $$\begin{aligned}
{\|h\|}_{{\tilde{L}^\infty}([0,t]; B^{s_1,s_2})}{\leqslant}e^{C\int_0^t{\|{\mathbf{v}}\|}_{B^{d/2+1}}d\tau}\left( {\|h_0\|}_{B^{s_1,s_2}}+\int_0^t{\|f\|}_{B^{s_1,s_2}}d\tau\right).
\end{aligned}$$
Existence of local solution
---------------------------
*Step 1: The Friedrich’s approximation.* For convenience, we introduce the solution ${\mathbf{u}}_{\mathrm{ls}}$ to the linear system $$\begin{aligned}
\label{eq.ls}
\partial_t{\mathbf{u}}_{\mathrm{ls}}-\mu\Delta{\mathbf{u}}_{\mathrm{ls}}-(\mu+\lambda) \nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}=0, \quad {\mathbf{u}}_{\mathrm{ls}}(0)={\mathbf{u}}_0.\end{aligned}$$ Denote ${\mathbf{u}}_{\mathrm{ls}}^\ell:={\mathbb{F}_\ell}{\mathbf{u}}_{\mathrm{ls}}$ and $\tilde{{\mathbf{u}}}^\ell:={\mathbf{u}}^\ell-{\mathbf{u}}_{\mathrm{ls}}^\ell$. Then we can construct the following approximation $(\rho^\ell,g^\ell,\tilde{{\mathbf{u}}}^\ell)$ satisfying $$\begin{aligned}
\label{eq.system.app4}
\left\{\begin{aligned}
&\rho_t^\ell+{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla\rho^\ell)={\mathbb{F}_\ell}((\rho^\ell+1){\mathrm{div}}{\mathbf{u}}^\ell),\\
&g_t^\ell+{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla g^\ell)=-{\mathbb{F}_\ell}((g^\ell+{\bar{n}}){\mathrm{div}}{\mathbf{u}}^\ell),\\
&\partial_t\tilde{{\mathbf{u}}}^\ell+{\mathbb{F}_\ell}({\mathbf{u}}_{\mathrm{ls}}^\ell\cdot\nabla \tilde{{\mathbf{u}}}^\ell)+{\mathbb{F}_\ell}(\tilde{{\mathbf{u}}}^\ell\cdot\nabla {\mathbf{u}}^\ell)-{\mathbb{F}_\ell}[(1+\rho^\ell)(\mu\Delta\tilde{{\mathbf{u}}}^\ell +(\mu+\lambda)\nabla {\mathrm{div}}\tilde{{\mathbf{u}}}^\ell)]\\
&\qquad={\mathbb{F}_\ell}[\rho^\ell(\mu\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell +(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}^\ell)] -{\mathbb{F}_\ell}({\mathbf{u}}_{\mathrm{ls}}^\ell\cdot\nabla {\mathbf{u}}_{\mathrm{ls}}^\ell)-{\mathbb{F}_\ell}Q(\rho^\ell,g^\ell),\\
&(\rho^\ell,g^\ell,\tilde{{\mathbf{u}}}^\ell)|_{t=0}=(\rho_0^\ell,g_0^\ell,\mathbf{0}),
\end{aligned}\right.\end{aligned}$$ where $\rho_0^\ell:={\mathbb{F}_\ell}\rho_0$, $g_0^\ell:={\mathbb{F}_\ell}g_0$ and ${\mathbf{u}}^\ell:={\mathbf{u}}_{\mathrm{ls}}^\ell+\tilde{{\mathbf{u}}}^\ell$.
Note that if $1+\rho_0$ is bounded away from zero, then so is $1+{\mathbb{F}_\ell}\rho_0$ for sufficiently large $\ell$. It is easy to check that is an ordinary differential equation in $L_\ell^2\times L_\ell^2 \times (L_\ell^2)^d$ for every $\ell\in{\Bbb{N}}$. By the usual Cauchy-Lipschitz theorem, there is a strictly positive maximal time $T_\ell^*$ such that a unique solution $(\rho^\ell,g^\ell,\tilde{{\mathbf{u}}}^\ell)$ exists in $[0,T_\ell^*)$ which is continuous in time with value in $L_\ell^2\times L_\ell^2 \times
(L_\ell^2)^d$, i.e. $(\rho^\ell,g^\ell,\tilde{{\mathbf{u}}}^\ell)\in
\mathcal{C}([0,T_\ell^*); L_\ell^2\times L_\ell^2 \times
(L_\ell^2)^d)$, and $1+\rho^\ell$ is bounded away from zero.
*Step 2: Lower bound for lifespan and uniform estimates.* We introduce the following notations $$\begin{aligned}
M_0:=&{\|\rho_0\|}_{B^{d/2,d/2+\alpha}},\quad M^\ell(t):={\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2,d/2+\alpha})},\\
N_0:=&{\|g_0\|}_{B^{d/2,d/2+\alpha}},\quad N^\ell(t):={\|g^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2,d/2+\alpha})},\\
U_0:=&{\|{\mathbf{u}}_0\|}_{B^{d/2-1,d/2-1+\alpha}},\quad U_{\mathrm{ls}}^\ell(t):={\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{L^1([0,t]; B^{d/2+1,d/2+1+\alpha})},\\
\tilde{U}^\ell(t):=&{\|\tilde{{\mathbf{u}}}^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2-1,d/2-1+\alpha})} +b_*\underline{\nu}{\|\tilde{{\mathbf{u}}}^\ell\|}_{L^1([0,t];B^{d/2+1,d/2+1+\alpha})}.\end{aligned}$$
In view of Lemma \[lem.moment\], we take ${\mathbf{v}}={\mathbf{w}}=f=\mathbf{0}$ and $\rho=0$ there to get $$\begin{aligned}
{\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2-1,d/2-1+\alpha})}\lesssim U_0.\end{aligned}$$
Note that for all $k\in{\mathbb Z}$, we have $$\begin{aligned}
{\|{\triangle_k }\rho_0^\ell\|}_2{\leqslant}{\|{\triangle_k }\rho_0\|}_2,\quad {\|\rho_0^\ell\|}_{B^{s}}{\leqslant}{\|\rho_0\|}_{B^{s}}\end{aligned}$$ and similar properties for $g_0^\ell$ because of the boundedness of the operators ${\triangle_k }$.
From and , we get $$\begin{aligned}
\label{eq.mass.5}
{\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2})}{\leqslant}e^{3C\int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}d\tau} \left({\|\rho_0\|}_{B^{d/2}}+\frac{1}{3}\right) -\frac{1}{3},\end{aligned}$$ $$\begin{aligned}
\label{eq.mass.6}
&{\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2+\alpha})}
{\leqslant}e^{2C\int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}d\tau} \Big({\|\rho_0\|}_{B^{d/2+\alpha}}\nonumber \\ &+C\left[e^{3C\int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}d\tau} \Big({\|\rho_0\|}_{B^{d/2}}+\frac{1}{3}\Big) +\frac{2}{3}\right] \int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1+\alpha}}d\tau\Big),\end{aligned}$$ $$\begin{aligned}
\label{eq.mass.7}
{\|g^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2})}{\leqslant}e^{3C\int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}d\tau} \left({\|g_0\|}_{B^{d/2}}+\frac{{\bar{n}}}{3}\right) -\frac{{\bar{n}}}{3},\end{aligned}$$ and $$\begin{aligned}
\label{eq.mass.8}
&{\|g^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2+\alpha})}
{\leqslant}e^{2C\int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}d\tau} \Big({\|g_0\|}_{B^{d/2+\alpha}}\nonumber \\ &+C\left[e^{3C\int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}d\tau} \Big({\|g_0\|}_{B^{d/2}}+\frac{{\bar{n}}}{3}\Big) +\frac{2{\bar{n}}}{3}\right] \int_0^t{\|{\mathbf{u}}^\ell\|}_{B^{d/2+1+\alpha}}d\tau\Big).\end{aligned}$$
Let $b(t,x)=1+\rho^\ell$. From Step 1, we know $b_*>0$. Thus, by Lemma \[lem.moment\], \[lem.comp\] and \[lem.fgbesov\], we have $$\begin{aligned}
\label{eq.ns.9}
\begin{aligned}
&{\|\tilde{{\mathbf{u}}}^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2-1})}+\kappa b_*\underline{\nu} {\|\tilde{{\mathbf{u}}}^\ell\|}_{L^1([0,t];B^{d/2+1})}\\
{\leqslant}&\int_0^t\Big({\|\rho^\ell(\mu\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell +(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}^\ell)\|}_{B^{d/2-1}} \\ &\qquad\qquad+{\|{\mathbf{u}}_{\mathrm{ls}}^\ell\cdot\nabla {\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1}} +{\|Q(\rho^\ell,g^\ell)\|}_{B^{d/2-1}}\Big)d\tau \\ &
\times\exp\left(C\int_0^t\Big({\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}+ {\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2+1}} +(b_*\underline{\nu})^{1-2/\alpha}\bar{\nu}^{2/\alpha} {\|\rho^\ell\|}_{B^{d/2+\alpha}}^{2/\alpha}\Big)d\tau\right),
\end{aligned}\end{aligned}$$ and similarly $$\begin{aligned}
\label{eq.ns.10}
\begin{aligned}
&{\|\tilde{{\mathbf{u}}}^\ell\|}_{{\tilde{L}^\infty}([0,t];B^{d/2-1+\alpha})}+\kappa b_*\underline{\nu} {\|\tilde{{\mathbf{u}}}^\ell\|}_{L^1([0,t];B^{d/2+1+\alpha})}\\
{\leqslant}&\int_0^t\Big({\|\rho^\ell(\mu\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell +(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}^\ell)\|}_{B^{d/2-1+\alpha}} \\ &\qquad\qquad+{\|{\mathbf{u}}_{\mathrm{ls}}^\ell\cdot\nabla {\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1+\alpha}} +{\|Q(\rho^\ell,g^\ell)\|}_{B^{d/2-1+\alpha}}\Big)d\tau \\ &
\times\exp\left(C\int_0^t\Big({\|{\mathbf{u}}^\ell\|}_{B^{d/2+1}}+ {\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2+1}} +(b_*\underline{\nu})^{1-2/\alpha}\bar{\nu}^{2/\alpha} {\|\rho^\ell\|}_{B^{d/2+\alpha}}^{2/\alpha}\Big)d\tau\right).
\end{aligned}\end{aligned}$$
By Lemma \[lem.fgbesov\], we get, for all $\sigma\in\{0,\alpha\}$, that $$\begin{aligned}
&{\|\rho^\ell\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1+\sigma}}\lesssim {\|\rho^\ell\|}_{B^{d/2}}{\|\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1+\sigma}}
\lesssim {\|\rho^\ell\|}_{B^{d/2}}{\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2+1+\sigma}},\\
&{\|\rho^\ell\nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1+\sigma}}\lesssim {\|\rho^\ell\|}_{B^{d/2}}{\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2+1+\sigma}},\\
&{\|{\mathbf{u}}_{\mathrm{ls}}^\ell\cdot\nabla{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1+\sigma}}
\lesssim {\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2}} {\|\nabla{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2-1+\sigma}} \lesssim {\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2}} {\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2+\sigma}}.\end{aligned}$$ Recall that $$\begin{aligned}
Q(\rho^\ell,g^\ell)
=&\frac{\rho^\ell\nabla \rho^\ell}{\rho^\ell+1}-\nabla\rho^\ell+\frac{a_0}{{\bar{m}}}(\rho^\ell+1)\nabla g^\ell+B(\rho^\ell,g^\ell)\Big[-\frac{{\bar{m}}}{(\rho^\ell+1)^2}\nabla\rho^\ell\\
&+\frac{k_0-a_0{\bar{n}}}{\rho^\ell+1}\nabla\rho^\ell+\frac{a_0g^\ell\nabla\rho^\ell}{\rho^\ell+1} +\frac{a_0^2}{{\bar{m}}}(g^\ell+g^\ell\rho^\ell)\nabla g^\ell+\frac{a_0(k_0+{\bar{m}}+a_0{\bar{n}})}{{\bar{m}}}\nabla g^\ell\\
&+\frac{a_0k_0+a_0^2{\bar{n}}}{{\bar{m}}}\rho^\ell\nabla g^\ell\Big],\end{aligned}$$ where $$\begin{aligned}
B(\rho^\ell,g^\ell):=\Big[\Big(\frac{{\bar{m}}}{\rho^\ell+1}+a_0(g^\ell+{\bar{n}}) -k_0\Big)^2+4k_0a_0(g^\ell+{\bar{n}})\Big]^{-1/2}.\end{aligned}$$ Similarly, for the third term of $Q$ $$\begin{aligned}
{\|(\rho^\ell+1)\nabla g^\ell\|}_{B^{d/2-1+\sigma}}\lesssim ({\|\rho^\ell\|}_{B^{d/2}}+1){\|\nabla g^\ell\|}_{B^{d/2-1+\sigma}}
\lesssim ({\|\rho^\ell\|}_{B^{d/2}}+1){\|g^\ell\|}_{B^{d/2+\sigma}}.\end{aligned}$$
By Lemmas \[lem.fgbesov\] and \[lem.comp\], we get for the first two terms of $Q$ $$\begin{aligned}
{\left\Vert\frac{\rho^\ell\nabla \rho^\ell}{\rho^\ell+1}-\nabla\rho^\ell\right\Vert}_{B^{d/2-1+\sigma}} \lesssim& \left({\left\Vert\frac{\rho^\ell}{\rho^\ell+1}\right\Vert}_{B^{d/2}}+1\right){\|\nabla \rho^\ell\|}_{B^{d/2-1+\sigma}}\\
\lesssim& ({\|\rho^\ell\|}_{B^{d/2}}+1){\|\rho^\ell\|}_{B^{d/2+\sigma}},\end{aligned}$$ and $$\begin{aligned}
&{\left\VertB(\rho^\ell,g^\ell)\frac{\nabla\rho^\ell}{(\rho^\ell+1)^2}\right\Vert}_{B^{d/2-1+\sigma}}\\
\lesssim& {\left\Vert[B(\rho^\ell,g^\ell)-B(0,0)]\frac{\nabla\rho^\ell}{ (\rho^\ell+1)^2}\right\Vert}_{B^{d/2-1+\sigma}}+B(0,0){\left\Vert\frac{\nabla\rho^\ell}{ (\rho^\ell+1)^2}\right\Vert}_{B^{d/2-1+\sigma}}\\
\lesssim&\left({\|B(\rho^\ell,g^\ell)-B(0,0)\|}_{B^{d/2}}+1\right) \left({\left\Vert\frac{\rho^\ell\rho^\ell+2\rho^\ell}{(\rho^\ell+1)^2}\right\Vert}_{B^{d/2}}+1\right) {\|\nabla\rho^\ell\|}_{B^{d/2-1+\sigma}}\\
\lesssim& \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right)^2 {\|\rho^\ell\|}_{B^{d/2+\sigma}}.\end{aligned}$$ Similarly, $$\begin{aligned}
&{\left\VertB(\rho^\ell,g^\ell)\frac{\nabla \rho^\ell}{\rho^\ell+1}\right\Vert}_{B^{d/2-1+\sigma}}
\lesssim \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right)^2 {\|\rho^\ell\|}_{B^{d/2+\sigma}},\\
&{\left\VertB(\rho^\ell,g^\ell)\frac{g^\ell\nabla \rho^\ell}{\rho^\ell+1}\right\Vert}_{B^{d/2-1+\sigma}}
\lesssim \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right)^2 {\|g^\ell\|}_{B^{d/2}}{\|\rho^\ell\|}_{B^{d/2+\sigma}},\\
&{\|B(\rho^\ell,g^\ell)(1+\rho^\ell)g^\ell\nabla g^\ell\|}_{B^{d/2-1+\sigma}}
\lesssim \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right)^2 {\|g^\ell\|}_{B^{d/2}}{\|g^\ell\|}_{B^{d/2+\sigma}},\\
&{\|B(\rho^\ell,g^\ell)\nabla g^\ell\|}_{B^{d/2-1+\sigma}}
\lesssim \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right) {\|g^\ell\|}_{B^{d/2+\sigma}},\\
&{\|B(\rho^\ell,g^\ell)\rho^\ell\nabla g^\ell\|}_{B^{d/2-1+\sigma}}
\lesssim \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right)
{\|\rho^\ell\|}_{B^{d/2}}{\|g^\ell\|}_{B^{d/2+\sigma}}.\end{aligned}$$ Thus, we get $$\begin{aligned}
{\|Q(\rho^\ell,g^\ell)\|}_{B^{d/2-1+\sigma}}
\lesssim \left({\|\rho^\ell\|}_{B^{d/2}}+{\|g^\ell\|}_{B^{d/2}}+1\right)^3
\left({\|\rho^\ell\|}_{B^{d/2+\sigma}}+{\|g^\ell\|}_{B^{d/2+\sigma}}\right).\end{aligned}$$
Therefore, from -, we conclude that $$\begin{aligned}
M^\ell(T)\lesssim & e^{C(U_{\mathrm{ls}}^\ell(T)+\tilde{U}^\ell(T)/(b_*\underline{\nu}))}(M_0+\frac{1}{3})\\ &\qquad+e^{C(U_{\mathrm{ls}}^\ell(T) +\tilde{U}^\ell(T)/(b_*\underline{\nu}))}(M_0+\frac{2}{3}) (U_{\mathrm{ls}}^\ell(T)+\tilde{U}^\ell(T)/(b_*\underline{\nu}))-\frac{1}{3},\\
N^\ell(T)\lesssim & e^{C(U_{\mathrm{ls}}^\ell(T)+\tilde{U}^\ell(T)/(b_*\underline{\nu}))}(N_0+\frac{{\bar{n}}}{3})\\ &\qquad+e^{C(U_{\mathrm{ls}}^\ell(T)+\tilde{U}^\ell(T)/(b_*\underline{\nu}))}(N_0+\frac{2{\bar{n}}}{3}) (U_{\mathrm{ls}}^\ell(T)+\tilde{U}^\ell(T)/(b_*\underline{\nu}))-\frac{{\bar{n}}}{3},\\
\tilde{U}^\ell(T)\lesssim &\left((\bar{\nu}M^\ell(T)+U_0)U_{\mathrm{ls}}^\ell(T)+(M^\ell(T)+N^\ell(T)+1)^3(M^\ell(T) +N^\ell(T))T\right)\\
&\qquad\times e^{C[U_{\mathrm{ls}}^\ell(T)+\tilde{U}^\ell(T)/(b_*\underline{\nu}) +(b_*\underline{\nu})^{1-2/\alpha}\bar{\nu}^{2/\alpha}(M^\ell(T))^{2/\alpha}T]}.\end{aligned}$$ Now, if we take $T$ so small that $$\begin{aligned}
\exp\left(CU_{\mathrm{ls}}^\ell(T)\right){\leqslant}\sqrt{2}, \quad \exp\left(\frac{C\tilde{U}^\ell(T)}{b_*\underline{\nu}}\right){\leqslant}\sqrt{2},
\end{aligned}$$ and $$\begin{aligned}
\exp\left(C(b_*\underline{\nu})^{1-2/\alpha}\bar{\nu}^{2/\alpha} (M^\ell(T))^{2/\alpha}T\right){\leqslant}2,\end{aligned}$$ then we have $$\begin{aligned}
\label{eq.uni.1}
\left\{\begin{aligned}
&M^\ell(T){\leqslant}4M_0+\frac{5}{3}, \quad N^\ell(T){\leqslant}4N_0+\frac{5{\bar{n}}}{3}, \\
&\tilde{U}^\ell(T){\leqslant}C\left((M_0+N_0+1)^4(T+\bar{\nu}U_{\mathrm{ls}}^\ell(T))+U_0 U_{\mathrm{ls}}^\ell(T)\right).
\end{aligned}\right.\end{aligned}$$
Noticing that $({\mathbb{F}_\ell}\rho^\ell,{\mathbb{F}_\ell}g^\ell,{\mathbb{F}_\ell}\tilde{{\mathbf{u}}}^\ell)=(\rho^\ell, g^\ell,\tilde{{\mathbf{u}}}^\ell)$ by the construction of the approximated system. Thus, we have $$\begin{aligned}
\partial_t(1+\rho^\ell)^{\pm 1}+{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla(1+\rho^\ell)^{\pm 1})\pm{\mathbb{F}_\ell}((1+\rho^\ell)^{\pm } {\mathrm{div}}{\mathbf{u}}^\ell)=0.\end{aligned}$$ It follows, by noticing that ${|\partial_t{|f|}|}={|\partial_t f|}$, that $$\begin{aligned}
{\|(1+\rho^\ell)^{\pm 1}(t)\|}_\infty {\leqslant}&{\|(1+\rho_0^\ell)^{\pm 1}\|}_\infty\\
&+\int_0^t [{\|{\mathbf{u}}^\ell\cdot\nabla(1+\rho^\ell)^{\pm 1}\|}_\infty+{\|(1+\rho^\ell)^{\pm } {\mathrm{div}}{\mathbf{u}}^\ell\|}_\infty] d\tau,\end{aligned}$$ which yields, by the Gronwall inequality, that $$\begin{aligned}
&{\|(1+\rho^\ell)^{\pm 1}(t)\|}_\infty
{\leqslant}e^{\int_0^t{\|{\mathrm{div}}{\mathbf{u}}^\ell\|}_\infty d\tau}\left({\|(1+\rho_0^\ell)^{\pm 1}\|}_\infty+\int_0^t{\|{\mathbf{u}}^\ell\|}_\infty{\|\nabla \rho^\ell\|}_\infty d\tau\right)\\
{\leqslant}&e^{\int_0^t{\|{\mathrm{div}}{\mathbf{u}}^\ell\|}_\infty d\tau}\left({\|(1+\rho_0^\ell)^{\pm 1}\|}_\infty+C\int_0^t({\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{B^{d/2}}+{\|\tilde{{\mathbf{u}}}^\ell\|}_{B^{d/2}}) {\|\rho^\ell\|}_{B^{d/2+1}}d\tau\right)\\
{\leqslant}&e^{\int_0^t{\|{\mathrm{div}}{\mathbf{u}}^\ell\|}_\infty d\tau}\left({\|(1+\rho_0^\ell)^{\pm 1}\|}_\infty+CT(U_0+\tilde{U}^\ell(T))M^\ell(T)\right),\end{aligned}$$ where we have to choose $\alpha=1$ in the previous estimates. Hence, if we assume that there exist two positive constants $b_*$ and $b^*$ such that $$\begin{aligned}
b_*{\leqslant}1+\rho_0{\leqslant}b^*,\end{aligned}$$ then we can take $T$ small enough such that $$\begin{aligned}
\int_0^T{\|{\mathrm{div}}{\mathbf{u}}^\ell\|}_\infty d\tau{\leqslant}\ln 2, \text{ and } CT(U_0+\tilde{U}^\ell(T))M^\ell(T){\leqslant}1,\end{aligned}$$ and so $$\begin{aligned}
\label{eq.uni.2}
\frac{b_*}{2(1+b_*)}{\leqslant}1+\rho^\ell{\leqslant}2(b^*+1).\end{aligned}$$ Now, by means of a bootstrap argument, we can get that there exist two constants $\eta$ and $C$ depending only on $d$ such that if $$\begin{aligned}
\label{eq.uni.3}
\left\{\begin{aligned}
&(b_*\underline{\nu})^{1-2/\alpha}\bar{\nu}^{2/\alpha} (M^\ell(T))^{2/\alpha}T{\leqslant}\eta,\\
&(M_0+N_0+1)^4(T+\bar{\nu}U_{\mathrm{ls}}^\ell(T))+U_0 U_{\mathrm{ls}}^\ell(T){\leqslant}\eta b_*\underline{\nu},
\end{aligned}\right.\end{aligned}$$ then we have and .
Therefore, $T_\ell^*$ may be bounded from below by any time $T$ satisfying , and the inequalities and are satisfied by $(\rho^\ell,g^\ell,{\mathbf{u}}^\ell)$. In particular, $(\rho^\ell,g^\ell,{\mathbf{u}}^\ell)_{\ell\in{\Bbb{N}}}$ is bounded in $F_T^1$.
*Step 3: Time derivatives.* In order to pass to the limit in the approximated system, we first give the following lemma.
\[lem.timelocal\] Let $\tilde{\rho}^\ell:=\rho^\ell-{\mathbb{F}_\ell}\rho_0$, $\tilde{g}^\ell:=g^\ell-{\mathbb{F}_\ell}g_0$. Then the sequences $(\tilde{\rho}^\ell)_{\ell\in{\Bbb{N}}}$ and $(\tilde{g}^\ell)_{\ell\in{\Bbb{N}}}$ are uniformly bounded in $$\mathcal{C}([0,T]; B^{d/2,d/2+1})\cap \mathcal{C}^{1/2}([0,T];B^{d/2-1,d/2}),$$ and the sequence $(\tilde{{\mathbf{u}}}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in $$(\mathcal{C}([0,T]; B^{d/2-1,d/2})\cap \mathcal{C}^{1/4}([0,T];B^{d/2-1,d/2}+B^{d/2-3/2,d/2-1/2}))^d.$$
From $\partial_t\tilde{\rho}^\ell=-{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla \rho^\ell)+{\mathbb{F}_\ell}((\rho^\ell+1){\mathrm{div}}{\mathbf{u}}^\ell)$, we have $$\begin{aligned}
&{\|\partial_t\tilde{\rho}^\ell\|}_{L^2([0,T];B^{d/2-1,d/2})} {\leqslant}{\|\partial_t\tilde{\rho}^\ell\|}_{{\tilde{L}^2}([0,T];B^{d/2-1,d/2})}\\
{\leqslant}& {\|{\mathbf{u}}^\ell\cdot\nabla \rho^\ell\|}_{{\tilde{L}^2}([0,T];B^{d/2-1,d/2})} +{\|(\rho^\ell+1){\mathrm{div}}{\mathbf{u}}^\ell\|}_{{\tilde{L}^2}([0,T];B^{d/2-1,d/2})}\\
\lesssim &{\|{\mathbf{u}}^\ell\|}_{{\tilde{L}^2}([0,T];B^{d/2})}{\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2,d/2+1})} \\ &+({\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2})}+1){\|{\mathbf{u}}^\ell\|}_{{\tilde{L}^2}([0,T];B^{d/2,d/2+1})}.\end{aligned}$$ Since $({\mathbf{u}}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in $${\tilde{L}^\infty}([0,T];B^{d/2-1,d/2})\cap L^1([0,T];B^{d/2+1,d/2+2}),$$ it is also bounded in ${\tilde{L}^2}([0,T];B^{d/2,d/2+1})$ by Lemma \[lem.inter\]. Recall that $(\rho^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in ${\tilde{L}^\infty}([0,T];B^{d/2,d/2+1})$, then $(\partial_t\tilde{\rho}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in $$L^2([0,T];B^{d/2-1,d/2}),$$ and so $(\tilde{\rho}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in $$\mathcal{C}^{1/2}([0,T];B^{d/2-1,d/2}) \text{ and in } \mathcal{C}([0,T]; B^{d/2,d/2-1}).$$ Similarly, we have the same arguments for $\tilde{g}^\ell$.
Recall that $$\begin{aligned}
\partial_t\tilde{{\mathbf{u}}}^\ell=&-{\mathbb{F}_\ell}({\mathbf{u}}^\ell\cdot\nabla {\mathbf{u}}^\ell)+{\mathbb{F}_\ell}[(1+\rho^\ell)(\mu\Delta\tilde{{\mathbf{u}}}^\ell +(\mu+\lambda)\nabla {\mathrm{div}}\tilde{{\mathbf{u}}}^\ell)]\\
&\qquad+{\mathbb{F}_\ell}[\rho^\ell(\mu\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell +(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}^\ell)]-{\mathbb{F}_\ell}Q(\rho^\ell,g^\ell).\end{aligned}$$ Since $$\begin{aligned}
&{\|{\mathbf{u}}^\ell\cdot\nabla{\mathbf{u}}^\ell\|}_{L^2([0,T];B^{d/2-1,d/2})}
\lesssim {\|{\mathbf{u}}^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2-1,d/2})}{\|{\mathbf{u}}^\ell\|}_{L^2([0,T];B^{d/2+1})},\\
&{\|(1+\rho^\ell)(\mu\Delta\tilde{{\mathbf{u}}}^\ell +(\mu+\lambda)\nabla {\mathrm{div}}\tilde{{\mathbf{u}}}^\ell)\|}_{L^{4/3}([0,T];B^{d/2-3/2,d/2-1/2})}\\
\lesssim& (1+{\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2})} ) {\|\tilde{{\mathbf{u}}}^\ell\|}_{L^{4/3}([0,T];B^{d/2+1/2,d/2+3/2})},\\
&{\|\rho^\ell(\mu\Delta{\mathbf{u}}_{\mathrm{ls}}^\ell +(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_{\mathrm{ls}}^\ell)\|}_{L^{4/3}([0,T];B^{d/2-3/2,d/2-1/2})} \\ \lesssim& {\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2})} {\|{\mathbf{u}}_{\mathrm{ls}}^\ell\|}_{L^{4/3}([0,T];B^{d/2+1/2,d/2+3/2})},\\
&{\|Q(\rho^\ell,g^\ell)\|}_{{\tilde{L}^\infty}([0,T];B^{d/2-1,d/2})}\\
\lesssim & \left({\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2}}+{\|g^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2})}+1\right)^3
\left({\|\rho^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2,d/2+1})} \right.\\ &\left.\qquad+{\|g^\ell\|}_{{\tilde{L}^\infty}([0,T];B^{d/2,d/2+1})}\right),\end{aligned}$$ by Lemma \[lem.inter\], we can conclude that $(\tilde{{\mathbf{u}}}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in $$\mathcal{C}^{1/4}([0,T];B^{d/2-1,d/2}+B^{d/2-3/2,d/2-1/2}) \text{ and in } \mathcal{C}([0,T]; B^{d/2-1,d/2}).$$ This completes the proof of the lemma.
*Step 4: Compactness and convergence.* The proof is based on the Arzelà-Ascoli theorem and compact embeddings for Besov spaces. Since it is similar to the arguments for global well-posedness, we only give the outlines of the proof.
From Lemma \[lem.timelocal\], $(\tilde{\rho}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in the space $${\tilde{L}^\infty}([0,T];B^{d/2,d/2+1})$$ and equicontinuous on $[0,T]$ with values in $B^{d/2-1,d/2}$. Since the embedding $$B^{d/2-1,d/2}\hookrightarrow B^{d/2-1}$$ is (locally) compact, and $(\rho_0^\ell)_{\ell\in{\Bbb{N}}}$ tends to $\rho_0$ in $B^{d/2,d/2+1}$, we conclude that $(\rho^\ell)_{\ell\in{\Bbb{N}}}$ tends (up to an extraction) to some distribution $\rho$. Given that $(\rho^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in ${\tilde{L}^\infty}([0,T];B^{d/2})$, we actually have $$\rho\in {\tilde{L}^\infty}([0,T];B^{d/2}).$$ The same arguments are valid for the sequence $(g^\ell)_{\ell\in{\Bbb{N}}}$.
From the definition of $({\mathbf{u}}_{\mathrm{ls}}^\ell)_{\ell\in{\Bbb{N}}}$, it is clear that $({\mathbf{u}}_{\mathrm{ls}}^\ell)_{\ell\in{\Bbb{N}}}$ tends to the solution ${\mathbf{u}}_{\mathrm{ls}}$ of in ${\tilde{L}^\infty}([0,t];B^{d/2-1,d/2})\cap L^1([0,T]; B^{d/2+1,d/2+2})$.
Since $(\tilde{{\mathbf{u}}}^\ell)_{\ell\in{\Bbb{N}}}$ is uniformly bounded in ${\tilde{L}^\infty}([0,T]; B^{d/2-1,d/2})$ and equicontinuous on $[0,T]$ with values in $B^{d/2-1,d/2}+B^{d/2-3/2,d/2-1/2}$, it enable us to conclude that $(\tilde{{\mathbf{u}}}^\ell)_{\ell\in{\Bbb{N}}}$ converges, up to an extraction, to some function $\tilde{{\mathbf{u}}}\in {\tilde{L}^\infty}([0,T];B^{d/2-1})\cap L^1([0,T];B^{d/2+1})$.
Thus, we can pass to the limit in the system and setting ${\mathbf{u}}:=\tilde{{\mathbf{u}}}+{\mathbf{u}}_{\mathrm{ls}}$. Then, $(\rho,g,{\mathbf{u}})$ satisfies the system .
*Step 5: Continuities in time.*
From the first equation of , we get $\partial_t\rho\in L^2([0,T];B^{d/2-1,d/2})$ which implies $\rho\in \mathcal{C}([0,T];B^{d/2-1,d/2})$. So does $g$ in the same space. For ${\mathbf{u}}$, we can derive, from the third equation of , that $\partial_t{\mathbf{u}}\in (L^1+L^2)([0,T]; B^{d/2-1,d/2})$ which yields ${\mathbf{u}}\in \mathcal{C}([0,T];B^{d/2-1,d/2})$.
Uniqueness
----------
Let $(\rho_1,g_1,{\mathbf{u}}_1)$ and $(\rho_2,g_2,{\mathbf{u}}_2)$ be two solutions in $F_T^1$ of with the same initial data. Without loss of generality, we can assume that $(\rho_2,g_2,{\mathbf{u}}_2)$ is the solution constructed in the previous subsection such that $$1+\inf_{(t,x)\in[0,T]\times{\Bbb{R}}^d} \rho_2(t,x)>0.$$ We want to prove that $(\rho_2,g_2,{\mathbf{u}}_2)\equiv (\rho_1,g_1,{\mathbf{u}}_1)$ on $[0,T]\times{\Bbb{R}}^d$. To this goal, we shall estimate the discrepancy $(\delta\rho,\delta g,\delta{\mathbf{u}}):=(\rho_2-\rho_1,g_2-g_1,{\mathbf{u}}_2-{\mathbf{u}}_1)$ with respect to a suitable norm, satisfying $$\begin{aligned}
\label{eq.sys.uni}
\left\{\begin{aligned}
&\partial_t\delta\rho+{\mathbf{u}}_2\cdot\nabla\delta\rho+\delta{\mathbf{u}}\cdot\nabla\rho_1 =\delta\rho{\mathrm{div}}{\mathbf{u}}_2+(\rho_1+1){\mathrm{div}}\delta{\mathbf{u}},\\
&\partial_t\delta g+{\mathbf{u}}_2\cdot\nabla\delta g+\delta{\mathbf{u}}\cdot\nabla g_1 =-\delta g{\mathrm{div}}{\mathbf{u}}_2-(g_1+{\bar{n}}){\mathrm{div}}\delta{\mathbf{u}},\\
&\partial_t\delta{\mathbf{u}}+{\mathbf{u}}_2\cdot\nabla\delta{\mathbf{u}}+\delta{\mathbf{u}}\cdot\nabla{\mathbf{u}}_1 -(1+\rho_2)(\mu\Delta \delta{\mathbf{u}}_(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}})\\
&\qquad-\delta\rho(\mu\Delta{\mathbf{u}}_1+(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}}_1)
+Q(\rho_2,g_2)-Q(\rho_1,g_1)=0,\\
&(\delta\rho,\delta g,\delta{\mathbf{u}})|_{t=0}=(0,0,\mathbf{0}).
\end{aligned}\right.\end{aligned}$$ We shall prove the uniqueness in a larger function space $$\begin{aligned}
F_T:=(\mathcal{C}([0,T];B^{d/2}))^{1+1}\times(\mathcal{C}([0,T];B^{d/2})\cap L^1([0,T]; B^{d/2+1}))^d.\end{aligned}$$
By Proposition \[prop.3\], we get for all $T'\in[0,T]$ $$\begin{aligned}
&{\|\delta\rho\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2})}\\
{\leqslant}& e^{C{\|{\mathbf{u}}_2\|}_{L^1([0,T'];B^{d/2+1})}}\int_0^{T'}\Big({\|\delta{\mathbf{u}}\cdot\nabla \rho_1\|}_{B^{d/2}}+{\|\delta\rho{\mathrm{div}}{\mathbf{u}}_2\|}_{B^{d/2}} \\ &\qquad\qquad\qquad\qquad\qquad\qquad\qquad +{\|(\rho_1+1){\mathrm{div}}\delta{\mathbf{u}})\|}_{B^{d/2}}\Big)d\tau\\
\lesssim& e^{C{\|{\mathbf{u}}_2\|}_{L^1([0,T'];B^{d/2+1})}}\int_0^{T'} \Big[{\|\delta{\mathbf{u}}\|}_{B^{d/2}} {\|\rho_1\|}_{B^{d/2+1}}+{\|\delta\rho\|}_{B^{d/2}}{\|{\mathbf{u}}_2\|}_{B^{d/2+1}} \\ &\qquad\qquad\qquad\qquad\qquad +(1+{\|\rho_1\|}_{B^{d/2}}){\|\delta{\mathbf{u}}\|}_{B^{d/2+1}}\Big]d\tau.\end{aligned}$$ Using the Gronwall inequality, it yields $$\begin{aligned}
\label{eq.uni.4}
\begin{aligned}
&{\|\delta\rho\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2})}\\
\lesssim &e^{C{\|{\mathbf{u}}_2\|}_{L^1([0,T'];B^{d/2+1})}}\int_0^{T'}\big[{\|\delta{\mathbf{u}}\|}_{B^{d/2}} {\|\rho_1\|}_{B^{d/2+1}}+(1+{\|\rho_1\|}_{B^{d/2}}) {\|\delta{\mathbf{u}}\|}_{B^{d/2+1}}\big]d\tau\\
{\leqslant}&C_T\left({\|\delta{\mathbf{u}}\|}_{L^2([0,T']; B^{d/2})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})}\right),
\end{aligned}\end{aligned}$$ where $C_T$ is independent of $T'$.
Similarly, we have $$\begin{aligned}
\label{eq.uni.5}
{\|\delta g\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2})}{\leqslant}C_T\left({\|\delta{\mathbf{u}}\|}_{L^2([0,T']; B^{d/2})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})}\right).\end{aligned}$$
Applying Lemmas \[lem.moment\] and \[lem.fgbesov\] to the third equation of , it yields $$\begin{aligned}
&{\|\delta{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2-1})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})}\\
{\leqslant}& C e^{C\int_0^{T'}[{\|{\mathbf{u}}_1\|}_{B^{d/2+1}}+{\|{\mathbf{u}}_2\|}_{B^{d/2+1}}]d\tau}\int_0^{T'} \Big({\|\delta\rho\|}_{B^{d/2}}{\|{\mathbf{u}}_1\|}_{B^{d/2+1}} \\ &\qquad\qquad\qquad\quad\qquad\qquad\qquad\qquad\qquad+{\|Q(\rho_2,g_2)-Q(\rho_1,g_1)\|}_{B^{d/2-1}}\Big)d\tau.\end{aligned}$$ By Lemma \[lem.comp\], we get $$\begin{aligned}
{\|Q(\rho_2,g_2)-Q(\rho_1,g_1)\|}_{B^{d/2-1}}\lesssim (1+{\|(\rho_1,\rho_2,g_1,g_2)\|}_{B^{d/2}})^3({\|\delta\rho\|}_{B^{d/2}}+{\|\delta g\|}_{B^{d/2}}).\end{aligned}$$ Thus, it follows that $$\begin{aligned}
\label{eq.uni.6}
\begin{aligned}
&{\|\delta{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2-1})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})}\\
{\leqslant}& C_T(T'+T'^{1/2})({\|\delta\rho\|}_{L^\infty([0,T'];B^{d/2})}+{\|\delta g\|}_{L^\infty([0,T'];B^{d/2})}),
\end{aligned}\end{aligned}$$ since ${\mathbf{u}}_1\in L^2([0,T];B^{d/2+1})$ by Lemma \[lem.inter\].
From -, it yields, with the help of Lemma \[lem.inter\], that $$\begin{aligned}
&{\|\delta{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2-1})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})}\\
{\leqslant}& C_T(T'+T'^{1/2})({\|\delta{\mathbf{u}}\|}_{L^2([0,T']; B^{d/2})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})})\\
{\leqslant}& C_T(T'+T'^{1/2})({\|\delta{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T']; B^{d/2-1})}+{\|\delta{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1})}).\end{aligned}$$ Therefore, if we choose $T'$ so small that $C_T(T'+T'^{1/2})<1$, then we obtain that $(\delta\rho,\delta g,\delta{\mathbf{u}})=(0,0,\mathbf{0})$ on the time interval $[0,T']$. As in the proof of uniqueness for global well-posedness, we can extend $T'$ to $T$ by the translation with respect to the time variable, i.e. $(\delta\rho,\delta g,\delta{\mathbf{u}})=(0,0,\mathbf{0})$ on the time interval $[0,T]$.
A continuation criterion
------------------------
\[prop.cc\] Under the hypotheses of Theorem \[thm.4\], assume that the system has a solution $(\rho,g,{\mathbf{u}})$ on $[0,T)\times{\Bbb{R}}^d$ which belongs to $F_{T'}^1$ for all $T'<T$ and satisfies $$\begin{aligned}
\rho, g\in L^\infty([0,T);B^{d/2,d/2+1}), \quad \inf_{(t,x)\in [0,T)\times{\Bbb{R}}^d} \rho(t,x)>-1, \quad \int_0^T {\|\nabla{\mathbf{u}}\|}_{\infty}dt<\infty.
\end{aligned}$$ Then, there exists some $T^*>T$ such that $(\rho,g,{\mathbf{u}})$ may be continued on $[0,T^*]\times{\Bbb{R}}^d$ to a solution of which belongs to $F_{T^*}^1$.
Recall that ${\mathbf{u}}$ satisfies $$\begin{aligned}
{\mathbf{u}}_t+{\mathbf{u}}\cdot\nabla{\mathbf{u}}-(1+\rho)(\mu\Delta{\mathbf{u}}+(\mu+\lambda)\nabla{\mathrm{div}}{\mathbf{u}})+Q(\rho,g)=0, \quad {\mathbf{u}}|_{t=0}={\mathbf{u}}_0.
\end{aligned}$$ By Lemma \[lem.moment\], we get, for $T'<T$, that $$\begin{aligned}
&{\|{\mathbf{u}}\|}_{{\tilde{L}^\infty}([0,T'];B^{d/2-1,d/2})} +\underline{\nu}{\|{\mathbf{u}}\|}_{L^1([0,T'];B^{d/2+1,d/2+2})}\\
{\leqslant}&C e^{C\int_0^{T'}\left({\|\nabla{\mathbf{u}}\|}_\infty +{\|\rho\|}_{B^{d/2+1}}^2\right)dt}\left({\|{\mathbf{u}}_0\|}_{B^{d/2-1,d/2}}+\int_0^{T'} {\|\rho\|}_{B^{d/2,d/2+1}}dt\right)
\end{aligned}$$ for some constant $C$ depending only on $d$ and viscosity coefficients. Thus, there exists a constant ${\varepsilon}>0$ such that with initial data $(\rho(T-{\varepsilon}),g(T-{\varepsilon}),{\mathbf{u}}(T-{\varepsilon}))$ yields a solution on $[0,2{\varepsilon}]$. Since the solution $(\rho,g,{\mathbf{u}})$ is unique on $[0,T)$, this provides a continuation of $(\rho,g,{\mathbf{u}})$ beyond $T$.
Littlewood-Paley theory and Besov spaces
========================================
This section is devoted to recall some properties of Littlewood-Paley theory and Besov spaces which will be used in this paper. For more details, one can see [@Danchin01; @HHL] and references therein.
Let $\psi : {\Bbb{R}}^d \to [0,1]$ be a radial smooth cut-off function valued in $[0,1]$ such that $$\begin{aligned}
\psi(\xi)=\left\{
\begin{array}{ll}
1, &{|\xi|}{\leqslant}3/4,\\
\text{smooth}, &3/4<{|\xi|}<4/3,\\
0, &{|\xi|}{\geqslant}4/3.
\end{array}
\right.\end{aligned}$$ Let $\varphi(\xi)$ be the function $$\begin{aligned}
\varphi(\xi):=\psi(\xi/2)-\psi(\xi).\end{aligned}$$ Thus, $\psi$ is supported in the ball ${\{\xi\in{\Bbb{R}}^d:
{|\xi|}{\leqslant}4/3\}}$, and $\varphi$ is also a smooth cut-off function valued in $[0,1]$ and supported in the annulus $\{\xi\in{\Bbb{R}}^d:
3/4{\leqslant}{|\xi|}{\leqslant}8/3\}$. By construction, we have $$\begin{aligned}
\sum_{k\in{\mathbb Z}}\varphi(2^{-k}\xi)=1, \quad \forall
\xi\neq 0.\end{aligned}$$ One can define the dyadic blocks as follows. For $k\in{\mathbb Z}$, let $$\begin{aligned}
{\triangle_k }f:={\mathscr{F}}^{-1}\varphi(2^{-k}\xi){\mathscr{F}}f.\end{aligned}$$ The formal decomposition $$\begin{aligned}
\label{lpd}
f=\sum_{k\in{\mathbb Z}}{\triangle_k }f\end{aligned}$$ is called homogeneous Littlewood-Paley decomposition. Nevertheless, is true modulo polynomials, in other words (cf.[@Pee76]), if $f\in{\mathscr{S}}'({\Bbb{R}}^d)$, then $\sum_{k\in{\mathbb Z}}{\triangle_k }f$ converges modulo $\mathscr{P}[{\Bbb{R}}^d]$ and holds in ${\mathscr{S}}'({\Bbb{R}}^d)/\mathscr{P}[{\Bbb{R}}^d]$.
Let $s\in{\Bbb{R}}$. For $f\in{\mathscr{S}}'({\Bbb{R}}^d)$, we write $$\begin{aligned}
{\|f\|}_{{B^{s}}}=\sum_{k\in{\mathbb Z}}
2^{ks}{\|{\triangle_k }f\|}_{2}.\end{aligned}$$
A difficulty comes from the choice of homogeneous spaces at this point. Indeed, ${\|\cdot\|}_{{B^{s}}}$ cannot be a norm on $\{f\in{\mathscr{S}}'({\Bbb{R}}^d): {\|f\|}_{{B^{s}}}<\infty\}$ because ${\|f\|}_{{B^{s}}}=0$ means that $f$ is a polynomial. This enforces us to adopt the following definition for homogeneous Besov spaces (cf. [@Danchin01]).
Let $s\in{\Bbb{R}}$ and $m=-[{{d/2}}+1-s]$. If $m<0$, then we define ${B^{s}}({\Bbb{R}}^d)$ as $$\begin{aligned}
{B^{s}}=\Big\{f\in{\mathscr{S}}'({\Bbb{R}}^d):
{\|f\|}_{{B^{s}}}<\infty \text{ and } f=\sum_{k\in{\mathbb Z}}{\triangle_k }f
\text{ in } {\mathscr{S}}'({\Bbb{R}}^d)\Big\}.\end{aligned}$$ If $m{\geqslant}0$, we denote by $\mathscr{P}_m$ the set of $d$ variables polynomials of degree less than or equal to $m$ and define $$\begin{aligned}
{B^{s}}=\Big\{f\in{\mathscr{S}}'({\Bbb{R}}^d)/\mathscr{P}_m:
{\|f\|}_{{B^{s}}}<\infty \text{ and } f=\sum_{k\in{\mathbb Z}}{\triangle_k }f
\text{ in } {\mathscr{S}}'({\Bbb{R}}^d)/\mathscr{P}_m\Big\}.\end{aligned}$$
For the composition of functions, we have the following estimates.
\[lem.comp\] Let $s>0$ and $u\in {B^{s}}\cap L^\infty$. Then, it holds
Let $F\in W_{loc}^{[s]+2,\infty}({\Bbb{R}}^d)$ with $F(0)=0$. Then $F(u)\in{B^{s}}$. Moreover, there exists a function of one variable $C_0$ depending only on $s$ and $F$, and such that $$\begin{aligned}
{\|F(u)\|}_{{B^{s}}}{\leqslant}C_0({\|u\|}_{L^\infty}){\|u\|}_{{B^{s}}}.\end{aligned}$$
If $u,\, v\in{B^{{{d/2}}}}$, $(v-u)\in {B^{s}}$ for $s\in(-{{d/2}},{{d/2}}]$ and $G\in W_{loc}^{[{{d/2}}]+3,\infty}({\Bbb{R}}^d)$ satisfies $G'(0)=0$, then $G(v)-G(u)\in {B^{s}}$ and there exists a function of two variables $C$ depending only on $s$, $N$ and $G$, and such that $$\begin{aligned}
{\|G(v)-G(u)\|}_{{B^{s}}}{\leqslant}C({\|u\|}_{L^\infty},
{\|v\|}_{L^\infty})\left({\|u\|}_{{B^{{{d/2}}}}}+{\|v\|}_{{B^{{{d/2}}}}}\right)
{\|v-u\|}_{{B^{s}}}.\end{aligned}$$
We also need hybrid Besov spaces for which regularity assumptions are different in low frequencies and high frequencies [@Danchin01]. We are going to recall the definition of these new spaces and some of their main properties.
Let $s,\,t\in{\Bbb{R}}$. We define $$\begin{aligned}
{\|f\|}_{{B^{s,t}}}=\sum_{k{\leqslant}0}2^{ks}{\|{\triangle_k }f\|}_{2}
+\sum_{k>0}2^{kt}{\|{\triangle_k }f\|}_{2}.\end{aligned}$$ Let $m=-[{{d/2}}+1-s]$, we then define $$\begin{aligned}
{B^{s,t}}({\Bbb{R}}^d)
=&{\{f\in{\mathscr{S}}'({\Bbb{R}}^d): {\|f\|}_{{B^{s,t}}}<\infty\}},
\quad \text{if } m<0,\\
{B^{s,t}}({\Bbb{R}}^d)
=&{\{f\in{\mathscr{S}}'({\Bbb{R}}^d)/\mathscr{P}_m: {\|f\|}_{{B^{s,t}}}<\infty\}},
\quad \text{if } m {\geqslant}0.\end{aligned}$$
We have the following inclusions for hybrid Besov spaces.
We have ${B^{s,s}}={B^{s}}$.
If $s{\leqslant}t$ then ${B^{s,t}}={B^{s}}\cap{B^{t}}$. Otherwise, ${B^{s,t}}={B^{s}}+{B^{t}}$.
The space ${B^{0,s}}$ coincides with the usual inhomogeneous Besov space $B_{2,1}^s$.
If $s_1{\leqslant}s_2$ and $t_1{\geqslant}t_2$, then ${B^{s_1,t_1}}\hookrightarrow {B^{s_2,t_2}}$.
Let us now recall some useful estimates for the product in hybrid Besov spaces.
\[lem.fgbesov\] Let $s_1,\, s_2>0$ and $f,\,g\in L^\infty\cap {B^{s_1,s_2}}$. Then $fg\in {B^{s_1,s_2}}$ and $$\begin{aligned}
{\|fg\|}_{{B^{s_1,s_2}}}\lesssim
{\|f\|}_{L^\infty}{\|g\|}_{{B^{s_1,s_2}}}
+{\|f\|}_{{B^{s_1,s_2}}}{\|g\|}_{L^\infty}.\end{aligned}$$ Let $s\in (-d/2,d/2]$, $f\in B^{d/2}$ and $g\in B^s$, then $fg\in B^s$ and $${\|fg\|}_{B^s}\lesssim {\|f\|}_{B^{d/2}}{\|g\|}_{B^s}.$$
Let $s_1, s_2, t_1, t_2{\leqslant}{{d/2}}$ such that $\min(s_1+s_2, t_1+t_2)>0$, $f\in {B^{s_1,t_1}}$ and $g\in
{B^{s_2,t_2}}$. Then $fg\in {B^{s_1+s_2-1,t_1+t_2-1}}$ and $$\begin{aligned}
{\|fg\|}_{{B^{s_1+s_2-{{d/2}},t_1+t_2-{{d/2}}}}}\lesssim
{\|f\|}_{{B^{s_1,t_1}}}{\|g\|}_{{B^{s_2,t_2}}}.\end{aligned}$$
For $\alpha,\beta\in{\Bbb{R}}$, let us define the following characteristic function on ${\mathbb Z}$: $$\begin{aligned}
\tilde{\varphi}^{\alpha,\beta}(k)=\left\{
\begin{array}{ll}
\alpha,\quad &\text{if } k{\leqslant}0,\\
\beta, & \text{if } k{\geqslant}1.
\end{array}
\right.\end{aligned}$$ Then, we can recall the following lemma.
\[lem.innner\] Let $F$ be an homogeneous smooth function of degree $m$. Suppose that $-d/2<s_1,t_1,s_2,t_2{\leqslant}1+d/2$. The following two estimates hold: $$\begin{aligned}
&{|(F(D){\triangle_k }({\mathbf{v}}\cdot \nabla a),F(D){\triangle_k }a)|}\\
&\qquad\qquad\lesssim \gamma_k2^{-k(\tilde{\varphi}^{s_1,s_2}(k)-m)}{\|{\mathbf{v}}\|}_{{B^{{{d/2}}+1}}} {\|a\|}_{{B^{s_1,s_2}}}{\|F(D){\triangle_k }a\|}_{2},\\
&{|(F(D){\triangle_k }({\mathbf{v}}\cdot\nabla a),{\triangle_k }b)+({\triangle_k }({\mathbf{v}}\cdot\nabla b),F(D){\triangle_k }a)|}\\
&\qquad\qquad\lesssim \gamma_k{\|{\mathbf{v}}\|}_{{B^{{{d/2}}+1}}}\times\big( 2^{-k\tilde{\varphi}^{t_1,t_2}(k)}{\|F(D){\triangle_k }a\|}_{2}{\|b\|}_{{B^{t_1,t_2}}}\\
&\qquad\qquad\qquad+2^{-k(\tilde{\varphi}^{s_1,s_2}(k)-m)}{\|a\|}_{{B^{s_1,s_2}}}
{\|{\triangle_k }b\|}_{2}\big),
\end{aligned}$$ where $(\cdot,\cdot)$ denotes the $2$-inner product, $\sum_{k\in{\mathbb Z}} \gamma_k{\leqslant}1$ and the operator $F(D)$ is defined by $F(D)f:={\mathscr{F}}^{-1} F(\xi){\mathscr{F}}f$.
In the context of this paper, we also need to use the interpolation spaces of hybrid Besov spaces together with a time space such as $L^p([0,T);{B^{s,t}})$. Thus, we have to introduce the Chemin-Lerner type space (cf. [@CheLer]) which is a refinement of the space $L^p([0,T);{B^{s,t}})$.
Let $p\in [1,\infty]$, $T\in(0,\infty]$ and $s_1,\, s_2\in{\Bbb{R}}$. Then we define $$\begin{aligned}
{\|f\|}_{{\tilde{L}^p}([0,T);{B^{s,t}})}=\sum_{k{\leqslant}0}2^{ks}{\|{\triangle_k }f\|}_{L^p([0,T); L^2)}+\sum_{k> 0}2^{kt}{\|{\triangle_k }f\|}_{L^p([0,T); L^2)}.\end{aligned}$$
Noting that Minkowski’s inequality yields ${\|f\|}_{L^p([0,T);{B^{s,t}})}{\leqslant}{\|f\|}_{{\tilde{L}^p}([0,T);{B^{s,t}})}$, we define spaces ${\tilde{L}^p}([0,T);{B^{s,t}})$ as follows $$\begin{aligned}
{\tilde{L}^p}([0,T);{B^{s,t}})={\{f\in L^p([0,T);{B^{s,t}}): {\|f\|}_{{\tilde{L}^p}([0,T);{B^{s,t}})}<\infty\}}.\end{aligned}$$ If $T=\infty$, then we omit the subscript $T$ from the notation ${\tilde{L}^p}([0,T);{B^{s,t}})$, that is, ${\tilde{L}^p}({B^{s,t}})$ for simplicity. We will denote by $\tilde{\mathcal{C}}([0,T);{B^{s,t}})$ the subset of functions of ${\tilde{L}^\infty}([0,T);{B^{s,t}})$ which are continuous on $[0,T)$ with values in ${B^{s,t}}$.
Let us observe that $L^1([0,T);{B^{s,t}})={\tilde{L}^1}([0,T);{B^{s,t}})$, but the embedding $${\tilde{L}^p}([0,T);{B^{s,t}})\subset L^p([0,T);{B^{s,t}})$$ is strict if $p>1$.
We will use the following interpolation property which can be verified easily (cf. [@BCDbook; @Ber76]).
\[lem.inter\] Let $s,t,s_1,t_1, s_2,t_2\in{\Bbb{R}}$ and $p,p_1,p_2\in[1,\infty]$. We have $$\begin{aligned}
{\|f\|}_{{\tilde{L}^p}([0,T);{B^{s,t}})}{\leqslant}{\|f\|}_{{\tilde{L}^{{p_1}}}([0,T);{B^{s_1,t_1}})}^\theta {\|f\|}_{{\tilde{L}^{{p_2}}}([0,T);{B^{s_2,t_2}})}^{1-\theta},\end{aligned}$$ where $\frac{1}{p}=\frac{\theta}{p_1}+\frac{1-\theta}{p_2}$, $s=\theta s_1+(1-\theta)s_2$ and $t=\theta t_1+(1-\theta)t_2$.
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors would like to thank the referees for their valuable comments and Dr. L. Yao for helpful comments on the original version of the manuscript. Hao’s work was partially supported by the National Natural Science Foundation of China (grant 11171327), and the Youth Innovation Promotion Association, Chinese Academy of Sciences. Li’s work was partially supported by NSFC grant 11171228 and 11011130029, and the AHRDIHL Project of Beijing Municipality (No. PHR 201006107).
[99.]{}
, [*Fourier Analysis and Nonlinear Partial Differential Equations*]{}, GMW 343, Springer-Verlag, Berlin Heidelberg, 2011. , [*Interpolation Spaces, An Introduction*]{}, GMW 223, Springer-Verlag, Berlin Heidelberg, 1976.
, [*Fundamental of Multipule Flow*]{}, Cambridge University Press, New York, 2005.
, [*Flot de champs de vecteurs non lipschitziens et équations de Navier-Stokes*]{}, J. Differential Equations, 121 (1992), pp. 314–328.
, [*Global existence in critical spaces for compressible Navier-Stokes equations*]{}, Invent. Math., 141 (2000), pp. 579–614.
[——]{}, [*Global existence in critical spaces for flows of compressible viscous and heat-conductive gases*]{}, Arch. Ration. Mech. Anal., 160 (2001), pp. 1–39. [——]{}, [*Local theory in critical spaces for compressible viscous and heat-conductive gases*]{}, Commun. Partial Differential Equations, 26 (2001), pp. 1183–1233.
[——]{}, [*Well-Posedness in critical spaces for barotropic viscous fluids with truly not constant density*]{}, Commun. Partial Differential Equations, 32 (2007), pp. 1373–1397.
, [*Global weak solutions for a viscous liquid-gas model with transition to single-phase gas flow and vacuum*]{}, Nonlinear Anal., 70 (2009), pp. 3864–3886. , [*Global existence of weak solutions for a viscous two-phase model*]{}, J. Differential Equations, 245 (2008), pp. 2660–2703.
[——]{}, [*Global weak solutions for a viscous liquid-gas model with singular pressure law*]{}, Commun. Pure Appl. Anal., 8 (2009), pp. 1867–1894.
, [*Cauchy problem for viscous rotating shallow water equations*]{}, J. Differential Equations, 247 (2009), pp. 3234–3257.
, [*One-dimensional drift-flux model and constitutive equations for relative motion between phases in various two-phase flow regimes*]{}, Argonne National Lab Report, ANL 77–47, October 1977.
, [*Thermo-fluid Dynamics of Two-Phase Flow*]{}, Springer, New York, 2006.
, [*Multiphase flow dynamics, Vol. 1. Fundamentals*]{}, Springer-Verlag, Berlin, 2005; [*Vol. 2. Thermal and mechanical interactions*]{}, Springer-Verlag, Berlin, 2005.
, [*New thoughts on Besov spaces*]{}, Duke University Mathematical Series [1]{}, Durham N. C., 1976.
, [*Computational Methods for Multiphase Flow*]{}, Cambridge University Press, 2007.
, [*One-Dimensional Two-Phase Flow*]{}, McGraw Hill, New York, 1979.
, [*A blow-up criterion for a 2D viscous liquid-gas two-phase flow model*]{}, J. Differential Equations, 250 (2011), pp. 3362–3378. [——]{}, [*Existence and asymptotic behavior of global weak solutions to a 2D viscous liquid-gas two-phase flow model*]{}, SIAM J. Math. Anal., 42 (2010), pp. 1874–1897.
, [*Free boundary value problem for a viscous two-phase model with mass-dependent viscosity*]{}, J. Differential Equations, 247 (2009), pp. 2705–2739.
[——]{}, [*Existence and uniqueness of global weak solution to a two-phase flow model with vacuum*]{}, Math. Ann., 349 (2010), pp. 903–928. , [*On the dispersed two-phase flow in the laminar flow regime*]{}, Chemical Engineering Science, 19 (1964), pp. 897–917.
, [*Average volumetric concentration in two-phase systems*]{}, J. Heat Transfer, 87 (1965), pp. 453–468.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In this paper, we study the dynamic assortment optimization problem under a finite selling season of length $T$. At each time period, the seller offers an arriving customer an assortment of substitutable products under a cardinality constraint, and the customer makes the purchase among offered products according to a discrete choice model. Most existing work associates each product with a real-valued fixed mean utility and assumes a multinomial logit choice (MNL) model. In many practical applications, feature/contextual information of products is readily available. In this paper, we incorporate the feature information by assuming a linear relationship between the mean utility and the feature. In addition, we allow the feature information of products to change over time so that the underlying choice model can also be non-stationary. To solve the dynamic assortment optimization under this changing contextual MNL model, we need to simultaneously learn the underlying unknown coefficient and make the decision on the assortment. To this end, we develop an upper confidence bound (UCB) based policy and establish the regret bound on the order of ${\widetilde}{O}(d\sqrt{T})$, where $d$ is the dimension of the feature and ${\widetilde}{O}$ suppresses logarithmic dependence. We further establish a lower bound $\Omega(d\sqrt{T}/{K})$, where $K$ is the cardinality constraint of an offered assortment, which is usually small. When $K$ is a constant, our policy is optimal up to logarithmic factors. In the exploitation phase of the UCB algorithm, we need to solve a combinatorial optimization for assortment optimization based on the learned information. We further develop an approximation algorithm and an efficient greedy heuristic. The effectiveness of the proposed policy is further demonstrated by our numerical studies.
**keywords**: Dynamic assortment optimization, regret analysis, contextual information, bandit learning, upper confidence bounds.
author:
- 'Xi Chen [^1]'
- Yining Wang
- Yuan Zhou
bibliography:
- 'refs.bib'
title:
- Dynamic Assortment Optimization with Changing Contextual Information
- 'Supplementary Material for: Dynamic Assortment Optimization with Changing Contextual Information'
---
Introduction
============
In operations, an important research problem facing a retailer is the selection of products/advertisements for display. For example, due to the limited shelf space, stocking restrictions, or available slots on a website, the retailer needs to carefully choose an assortment from the set of substitutable products. In an assortment optimization problem, choice model plays an important role since it characterizes a customer’s choice behavior. However, in many scenarios, customers’ choice behavior (e.g., mean utilities of products) is not given as *a priori* and cannot be easily estimated due to the insufficiency of historical data. This motivates the research of dynamic assortment optimization, which has attracted a lot of attentions from the revenue management community in recent years. A typical dynamic assortment optimization problem assumes a finite selling horizon of length $T$ with a large $T$. At each time period, the seller offers an assortment of products (with the size upper bounded by $K$) to an arriving customer. The seller observes the customer’s purchase decision, which further provides useful information for learning utility parameters of the underlying choice model. The multinomial logit model (MNL) has been widely used in dynamic assortment optimization literature, see, e.g., [@Caro2007; @Rusmevichientong2010; @Saure2013; @Agrawal16MNLBandit; @Agrawal17Thompson; @Chen:18tight; @Chen:18near].
In the age of e-commerce, side information of products is widely available (e.g., brand, color, size, texture, popularity, historical selling information), which is important in characterizing customers’ preferences for products. Moreover, some features are not static and could change over time (e.g., popularity score or ratings). The feature/contextual information of products will facilitate accurate assortment decisions that are tailored to customers’ preferences. In particular, we assume at each time $t=1,\ldots, T$, each product $j$ is associated with a $d$-dimensional feature vector $v_{tj} \in \mathbb{R}^d$. To incorporate the feature information, following the classical conditional logit model [@McFa73], we assume that the mean utility of product $j$ at time $t$ (denoted by $u_{tj}$) obeys a linear model $$\label{eq:linear}
u_{tj}= v_{tj}^\top\theta_0.$$ Here, $\theta_0 \in \mathbb{R}^d$ is the unknown coefficient to be learned. Based on this linear structure of the mean utility, we adopt the MNL model as the underlying choice model (see Section \[sec:model\] and Eq. for more details). As compared to the standard MNL, this *changing contextual MNL* model not only incorporates rich contextual information but also allows the utility to evolve over time. The changing utility is an attractive property as it captures the reality in many applications but also brings new technical challenges in learning and decision-making. For example, in existing works of [@Agrawal16MNLBandit] for plain MNL choice models, upper confidence bands are constructed by providing the same assortment repetitively to incoming customers until a no-purchase activity is observed. Such an approach, however, can no longer be applied to MNL with changing contextual information as the utility parameters of products constantly evolve with time. To overcome such challenges, we propose a policy that performs optimization at every single time period, without repetitions of assortments in general.
Our model also allows the revenue for each product $j$ to change over time. In particular, we associate the revenue parameter $r_{tj}$ for the product $j$ at time $t$.
This model generalizes the widely adopted (generalized) linear contextual bandit from machine learning literature (see, e.g., [@filippi2010parametric; @chu11; @Abbasi-Yadkori:2011; @Agrawal:13; @li2017provably] and references therein) in a non-trivial way since the MNL cannot be written in a generalized linear model form (when an assortment contains more than one product, see Section \[sec:related\] for more details). It is also worthwhile noting that this model incorporates a personalized MNL model proposed by [@Wang:17:person] as a special case, where each product $j$ is associated with a fixed but unknown coefficient $\theta(j)$ and each arriving customer at time $t$ with an observable feature vector $x_t$ (see Section \[sec:related\] for a more detailed discussion). On the other hand, we choose to motivate our model from product contextual information since in practice, obtaining products’ features is usually easier (and less sensitive) than extracting customers’ preferences.
Given this contextual MNL choice model, the key challenge is how to design a policy that simultaneously learns the unknown coefficient $\theta_0$ and sequentially makes the decision on offered assortment. The performance of a dynamic policy is usually measured by the *regret*, which is defined as the gap between the expected revenue generated by the policy and the oracle expected revenue when $\theta_0$ (and thus the mean utilizes) is known as a priori.
The first contribution of the paper is the construction of an upper confidence bound (UCB) policy. Our UCB policy is based on the maximum likelihood estimator (MLE) and thus is named MLE-UCB. Although UCB has been a well-known technique for bandit problems, how to adopt this high-level idea to solve a problem with specific structures certainly requires technical innovations (e.g., how to build a confidence interval varies from one problem to another). In particular, our MLE-UCB contains two stages. The first stage is a *pure exploration stage* in which assortments are randomly offered and a *“pilot MLE”* is computed based on the observed purchase actions. As we will show in Lemma \[lem:pilot\], this pilot estimator serves as a good initial estimator of $\theta_0$. After the exploration phase, the MLE-UCB enters the *simultaneous learning and decision-making phase*. We carefully construct an upper confidence bound of the expected revenue when offering an assortment. The added interval is based on the Fisher information matrix of the computed MLE from the previous step. Then we solve a combinatorial optimization problem to search the assortment that maximizes the upper confidence bound. By observing the customer’s purchase action based on the offered assortment, the policy updates the estimated MLE. In this update, we propose to compute a *“local MLE”*, which requires the solution to be close enough to our pilot estimator. The local MLE plays an important role in MLE-UCB policy since it guarantees that the obtained estimator at each time period is also close to the unknown true coefficient $\theta_0$.
Under some mild assumptions on features and coefficients, we are able to establish a regret bound ${\widetilde}{O}(d \sqrt{T})$, where the ${\widetilde}{O}$ notation suppresses logarithmic dependence on $T$, $K$ (cardinality constraint), and some other problem dependent parameters[^2]. One remarkable aspect of our regret bound is that our regret has no dependence on the total number of products $N$ (not even in a logarithmic factor). This makes the result attractive to online applications where $N$ is large (e.g., online advertisement). Moreover, it is also worthwhile noting the dependence of $K$ is only through a logarithmic term.
Our second contribution is to establish the lower bound result $\Omega(d\sqrt{T}/K)$. When the maximum size of an assortment $K$ is small (which usually holds in practice), this result shows that our policy is almost optimal.
Moreover, at each time period in the exploitation phase, our UCB policy needs to solve a combinatorial optimization problem, which searches for the best assortment (under the cardinality constraint) that minimizes the upper confidence bound of the expected revenue. Given the complicated structure of the upper confidence bound, there is no simple solution for this combinatorial problem. When $K$ is small and $N$ is not too large, one can directly search over all the possible sets with the size less than or equal to $K$. In addition to the solution of solving the combinatorial optimization exactly, [the third contribution of the work is to provide an approximation algorithm based on dynamic programming that runs in polynomial time with respect to $N$, $K$, $T$.]{} Although the proposed approximation algorithm has a theoretical guarantee, it is still not efficient for dealing with large-scale applications. To this end, we further describe a computationally efficient greedy heuristic for solving this combinatorial optimization problem. The heuristic algorithm is based on the idea of local search by greedy swapping, with more details described in Sec. \[subsec:heuristic\].
Related work {#sec:related}
------------
Due to the popularity of data-driven revenue management, dynamic assortment optimization, which adaptively learns unknown customers’ choice behavior, has received increasing attention in the past few years. Motivated by fast-fashion retailing, the work by [@Caro2007] first studied dynamic assortment optimization problem, but it makes a strong assumption that the demands for different product are independent. Recent works by [@Rusmevichientong2010; @Saure2013; @Agrawal16MNLBandit; @Agrawal17Thompson; @Chen:18tight; @Chen:18near] incorporated MNL models into dynamic assortment optimization and formulated the problem into a online regret minimization problem. In particular, for capacitated MNL, [@Agrawal16MNLBandit] and [@Agrawal17Thompson] proposed UCB and Thompson sampling techniques and established the regret bound ${\widetilde}{O}(\sqrt{NT})$ (when $T \gg N^2$). [@Chen:18tight] further established a matching lower bound of $\Omega(\sqrt{NT})$. It is interesting to compare our regret to the bound for the standard MNL case. When the total number of products $N$ is much larger than $d$ (i.e., $N > d^2$), by incorporating the contextual information, the regret reduces from ${\widetilde}{O}(\sqrt{NT})$ to ${\widetilde}{O}(d \sqrt{T})$. The latter one only depends on $d$ and is completely independent of the total number of products $N$, which also demonstrates the usefulness of the contextual information. [@Chen:18dynamic] further studied the dynamic assortment optimization under nested logit models. We also note that to highlight our key idea and focus on the balance between learning of $\theta_0$ and revenue maximization, we study the stylized dynamic assortment optimization problems following the existing literature [@Rusmevichientong2010; @Saure2013; @Agrawal16MNLBandit; @Agrawal17Thompson], which ignore operations considerations such as price decisions and inventory replenishment.
There is another line of recent research on investigating personalized assortment optimization.
By incorporating the feature information of each arriving customer, both the static and dynamic assortment optimization problems are studied in [@Chen2015] and [@Wang:17:person], respectively. It is worthwhile noting that although we do not approach our work from a personalized perspective[^3], the *personalized MNL* considered in [@Wang:17:person] can be viewed as a special of our model.
In particular, the personalized MNL assumes that each product $j$ is associated with an unknown coefficient $\theta(j) \in \mathbb{R}^D$. When a customer arrives at time $t$ with the *observed feature $x_t$*, the utility of product $j$ at time $t$ is $u_{tj}=x_t^\top \theta(j)$. Now we explain how to specialize our model to obtain the personalized MNL. Let us define $\theta_0:=\{\theta(1), \ldots, \theta(N)\} \in \mathbb{R}^{DN}$ and the feature vector $v_{tj}:=(0, \ldots, x_{t}, \ldots, 0) \in \mathbb{R}^{DN}$, which is a concatenation of $N$ $D$-dimensional vectors with the $j$-th vector being $x_t$ and all other vectors being 0. Then according to our linear model in Eq. , we have $u_{tj}= v_{tj}^\top\theta_0= x_t^\top \theta(j)$, which recovers the personalized MNL model. Using our regret bound ${\widetilde}{O}(d\sqrt{T})$ with $d=DN$ as the dimensionality of $\theta_0$, we directly obtain the regret ${\widetilde}{O}(DN \sqrt{T})$ for the dynamic assortment optimization under the personalized MNL. As compared to the Bayesian regret bound ${\widetilde}{O}(DN\sqrt{KT})$ in [@Wang:17:person] (see Theorem 3.3. therein), our approach still saves a factor of $\sqrt{K}$.
We also remark that our results require a slightly stronger assumption on the contextual information vectors $\{v_{tj}\}$ compared to [@Wang:17:person], which allows customer feature vectors $\{x_t\}$ to be adversarially chosen. More specifically, [a stochastic assumption]{} is imposed on $\{v_{tj}\}$ only during the *pure exploration* phase of our proposed policy. [After this pure exploration phase, the feature vectors $\{v_{tj}\}$ can also be adversarially chosen.]{} We refer the readers to Sec. \[subsec:regret\] for further details.
In addition, the developed techniques in our work and [@Wang:17:person] are different, Our policy is based on UCB, while the policy in [@Wang:17:person] is based on Thompson sampling. Furthermore, other research studies personalized assortment optimization in an adversarial setting rather than stochastic setting. For example, [@Golrezaei2014; @Chen:16recom] assumed that each customer’s choice behavior is known, but that the customers’ arriving sequence (or customers’ types) can be adversarially chosen and took the inventory level into consideration. Since the arriving sequence can be arbitrary, there is no learning component in the problem and both [@Golrezaei2014] and [@Chen:16recom] adopted the competitive ratio as the performance evaluation metric.
Another field of related research is the *contextual bandit* literature, in which the linear contextual bandit has been widely studied as a special case (see, e.g., [@Dani08stochastic; @Rusmevichientong:10:bandit; @chu11; @Abbasi-Yadkori:2011; @Agrawal:13] and references therein). Some recent work extends the linear contextual bandit to *generalized linear bandit* [@filippi2010parametric; @li2017provably], which assumes a generalized linear reward structure. In particular, the reward $r$ of pulling an arm given the observed feature vector of this arm $x$ is modeled by $$\mathbb E[r|x] = \sigma(x^\top\theta_0),
\label{eq:glm}$$ for an unknown linear model $\theta_0$ and a known link function $\sigma:\mathbb R\to\mathbb R$. For example, for a linear contextual bandit, $\sigma$ is the identity mapping, i.e., $\mathbb E[r|x]=\sigma(x^\top\theta_0)$. For the logistic contextual bandit, we have $r \in \{0,1\}$ and $\Pr(r=1|x)=\frac{\exp(x^\top\theta_0)}{1+\exp(x^\top\theta_0)}$. [ In a standard generalized linear bandit problem (see, e.g., [@li2017provably]) with $N$ arms, it is assumed that a context vector $v_{tj}$ is revealed at time $t$ for each arm $j \in [N]$. Given a selected arm $i_t \in [N]$ at time $t$, the expected reward follows Eq. , i.e., $\mathbb{E}[r_t | v_{t,i_t} ]=\sigma(v_{t,i_t}^\top\theta_0)$. At first glance, our contextual MNL model is a natural extension of the generalized linear bandit to the MNL choice model. However, when the size of an assortment $K\geq 2$, the contextual MNL *cannot* be written in the form of Eq. (\[eq:glm\]) and the denominator in the choice probability (see Eq. in the next section) has a more complicated structure. Therefore, our problem is technically *not* a generalized linear model and is therefore more challenging. Moreover, in contextual bandit problems, only one arm is selected by the decision-maker at each time period. In contrast, each action in an assortment optimization problem involves a set of items, which makes the action space more complicated.]{}
Notations and paper organization
--------------------------------
Throughout the paper, we adopt the standard asymptotic notations. In particular, we use $f(\cdot) \lesssim g(\cdot)$ to denote that $f(\cdot) = O(g(\cdot))$. Similarly, by $f(\cdot) \gtrsim g(\cdot)$, we denote $f(\cdot) = \Omega(g(\cdot))$. We also use $f(\cdot) \asymp g(\cdot)$ for $f(\cdot) = \Theta(g(\cdot))$. Throughout this paper, we will use $C_0, C_1, C_2，\ldots$ to denote universal constants. For a vector $v$ and a matrix $M$, we will use $\|v\|_2$ and $\|M\|_{\op}$ to denote the vector $\ell_2$-norm and the matrix spectral norm (i.e., the maximum singular value), respectively. Moreover, for a real-valued symmetric matrix $M$, we denote the maximum eigenvalue and the minimum eigenvalue of $M$ by $\lambda_{\max}(M)$ and $\lambda_{\min}(M)$, respectively, and define $\|v\|_M^2:=v^T M v$ for any given vector $v$. For a given integer $N$, we denote the set $\{1,\ldots, N\}$ by $[N]$.
The rest of the paper is organized as follows. In Section \[sec:model\], we introduce the mathematical formulation of our models and define the regret. In Section \[sec:UCB\], we describe the proposed MLE-UCB policy and provide the regret analysis. The lower bound result is provided in Section \[sec:lower\]. In Section \[sec:comb\], we investigate the combinatorial optimization problem in MLE-UCB and propose the approximation algorithm and greedy heuristic. The multivariate case of the approximation algorithm is relegated to the appendix. In Section \[sec:numerical\], we provide the numerical studies. The conclusion and future directions are discussed in Section \[sec:conclusion\]. Some technical proofs are provided in the online supplementary material.
The problem setup {#sec:model}
=================
There are $N$ items, conveniently labeled as $1,2,\cdots,N$. At each time $t$, a set of time-sensitive “feature vectors” $v_{t1}, v_{t2}, \cdots, v_{tN}\in\mathbb R^d$ and revenues $r_{t1},\cdots,r_{tN}\in[0,1]$ are observed, reflecting time-varying changes of items’ revenues and customers’ preferences. A retailer, based on the features $\{v_{ti}\}_{i=1}^N$ and previous purchasing actions, picks an assortment $S_t\subseteq[N]$ under the cardinality constraint $|S_t|\leq K$ to present to an incoming customer; the retailer then observes a purchasing action $i_t\in S_t\cup\{0\}$ and collects the associated revenue $r_{i_t}$ of the purchased item (if $i_t=0$ then no item is purchased and zero revenue is collected).
We use an MNL model with features to characterize how a customer makes choices. Let $\theta_0\in\mathbb R^d$ be an *unknown* time-invariant coefficient. For any $S\subseteq[N]$, the choice model $p_{\theta_0,t}(\cdot|S)$ is specified as (let $r_0=0$ and $v_{t0}=0$) $$p_{\theta_0,t}(j|S) = \frac{\exp\{v_{tj}^\top\theta_0\}}{1+\sum_{k\in S} \exp\{v_{tk}^\top\theta_0\}} \;\;\;\;\;\;\forall j\in S\cup\{0\}.
\label{eq:context_MNL}$$ For simplicity, in the rest of the paper we use $p_{\theta,t}(\cdot|S)$ to denote the law of the purchased item $i_t$ conditioned on given assortment $S$ at time $t$, parameterized by the coefficient $\theta\in\mathbb R^d$. The expected revenue $R_t(S)$ of assortment $S\subseteq[N]$ at time $t$ is then given by $$\label{eq:rev}
R_t(S) := \mathbb E_{\theta_0,t}[r_{tj}|S] = \frac{\sum_{j\in S}r_{tj}\exp\{v_{tj}^\top\theta_0\}}{1+\sum_{j\in S}\exp\{v_{tj}^\top \theta_0\}}.$$ [Note that throughout the paper, we use $\mathbb E_{\theta_0,t}[\cdot |S]$ to denote the expectation with respect to the choice probabilities $p_{\theta_0,t}(j|S)$ defined in Eq. .]{}
Our objective is to design policy $\pi$ such that the regret $$\label{eq:regret}
\mathrm{Regret}(\{S_t\}_{t=1}^T) =\mathbb E^\pi\sum_{t=1}^T R_t(S_t^*) - R_t(S_t) \;\;\;\;\;\text{where}\;\;S_t^*=\arg\max_{S\subseteq[N], |S|\leq K} R_t(S)$$ is minimized. [Here, $S_t^*$ is an optimal assortment chosen when the full knowledge of choice probabilities is available (i.e., $\theta_0$ is known).]{}
An MLE-UCB policy and its regret {#sec:UCB}
================================
[[Pure exploration]{}]{}: for $t=1,\cdots,T_0$, pick $S_t=\{\ell_t\}$ for a single product $\ell_t$ sampled uniformly at random from $\{1,\cdots,N\}$ and record purchasing actions $(i_1,\cdots,i_{T_0})$
Compute a pilot estimator using global MLE: $\theta^*\in\arg\max_{\theta\in\mathbb R^d} \sum_{t'=1}^{T_0} \log p_{\theta,t}(i_{t'}|S_{t'})$
Remark: the expectations admit the following closed-form expressions: $\mathbb E_{\theta,t}[r_{tj}|S] = \sum_{j\in S} p_{\theta,t}(j|S)r_{tj} =\frac{\sum_{j\in S}r_{tj}\exp\{v_{tj}^\top\theta\}}{1+\sum_{j\in S}\exp\{v_{tj}^\top\theta\}}$; $\mathbb E_{\theta,t}[v_{tj}|S] = \sum_{j\in S}p_{\theta,t}(j|S)v_{tj} = \frac{\sum_{j\in S}v_{tj}\exp\{v_{tj}^\top\theta\}}{1+\sum_{j\in S}\exp\{v_{tj}^\top\theta\}}$; $\mathbb E_{\theta,t}[v_{tj}v_{tj}^\top|S] = \sum_{j\in S}p_{\theta,t}(j|S)v_{tj}v_{tj}^\top = \frac{\sum_{j\in S}v_{tj}v_{tj}^\top\exp\{v_{tj}^\top\theta\}}{1+\sum_{j\in S}\exp\{v_{tj}^\top\theta\}}$.
We propose an MLE-UCB policy, described in Algorithm \[alg:mle-ucb\].
The policy can be roughly divided into two phases. In the first *pure exploration* phase, the policy selects assortments uniformly at random, consisting of only *one item*. The objective of the pure exploration is to establish a “pilot” estimator of the unknown coefficient $\theta_0$, i.e., a good initial estimator for $\theta_0$. For the simplicity of the analysis, we choose one item for each assortment in this phase, which facilitates us to adapt existing analysis in [@filippi2010parametric; @li2017provably] as the MNL-logit choice model reduces to a *generalized linear model* when only one item is present in the assortment. In the second phase, we use a UCB-type approach that selects $S_t$ as the assortment maximizing an *upper bound* $\overline R_t(S_t)$ of the expected revenue $R_t(S_t)$. Such upper bounds are built using a *local Maximum Likelihood Estimation (MLE)* of $\theta_0$. In particular, in Step 5, instead of computing an MLE, we compute a local MLE, where the estimator ${\widehat}{\theta}_{t-1}$ lies in a ball centered at the pilot estimator $\theta^*$ with a radius $\tau$. This localization also simplifies the technical analysis based on Taylor expansion, which benefits from the constraint that ${\widehat}\theta_{t-1}$ is not too far away from $\theta^*$.
To construct the confidence bound, we introduce the matrices ${\widehat}M_t({\widehat}\theta_{t-1}|S)$ and ${\widehat}I_{t-1}({\widehat}\theta_{t-1})$ in Step 6 of Algorithm \[alg:mle-ucb\], which are empirical estimates of the *Fisher’s information* matrices $-\mathbb E[\nabla_\theta^2\log p(\cdot|\theta)]$ corresponding to the MNL choice model $p(\cdot|S_t)$. The population version of the Fisher’s information matrices are presented in Eq. in Sec. \[sec:ana\_local\]. These quantities play an essential role in classical statistical analysis of maximum likelihood estimators (see, e.g., [@van1998asymptotic]).
The proposed MLE-UCB policy has three hyper-parameters: the coefficient $\omega>0$ that controls the lengths of confidence intervals of $R_t(S)$, the number of pure exploration iterations $T_0$, and the radius $\tau_0$ in the local MLE formulation. While theoretical values of $\omega,T_0$ and $\tau$ are given in Theorem \[thm:mle-ucb\], which potentially depend on several unknown problem parameters, in practice we recommend the usage of $T_0=\max\{d\log T, T^{1/4}\}$, $\omega=\sqrt{d\log T}$ and $\tau=1/K$.
In the rest of this section, we give a regret analysis that shows an ${\widetilde}O(d\sqrt{T})$ upper bound on the regret of the MLE-UCB policy. Additionally, we prove a lower bound of ${\widetilde}O(d\sqrt{T}/K)$ in Sec. \[sec:negative\] and show how the combinatorial optimization in Step 7 can be approximately computed efficiently in Sec. \[sec:comb\].
Regret analysis {#subsec:regret}
---------------
To establish rigorous regret upper bounds on Algorithm \[alg:mle-ucb\], we impose the following assumptions:
1. There exists a constant $\nu$ such that $\|v_{tj}\|_2\leq\nu$ for all $t$ and $j$. Moreover, for all $t\leq T_0$ and $j \in [N]$, $v_{tj}$ are i.i.d. generated from an unknown distribution with the density $\mu$ satisfying that $\lambda_{\min}(\mathbb E_\mu vv^\top) \geq \lambda_0$ for some constant $\lambda_0>0$;
2. There exists a constant $\rho<\infty$ such that for all $t\in[T]$ and $S\subseteq[N]$ [with $|S|\leq K$]{}, $\frac{p_{\theta_0,t}(j|S)}{p_{\theta_0,t}(j'|S)} \leq \rho$ for all $j,j'\in S\cup\{0\}$.
The item (A1) assumes that the contextual information vectors $\{v_{tj}\}$ in the pure-exploration phase with $t \leq T_0$ are *randomly generated* from a non-degenerate density. It also places a standard boundedness condition on $\{v_{tj}\}$ for all time periods $t$. Note that after the pure-exploration phase, we allow the contextual vectors $\{v_{tj}\}$ to be adversarially chosen, only subject to boundedness conditions. (A2) additionally assumes a bounded ratio between the probability of choosing any two different items in an arbitrary assortment set. [We remark that if $\|\theta_0\|_2\leq C$, then the boundedness assumption in (A1) implies (A2) with $\rho\leq e^{2 \max\{1,C\nu\}}$.]{}
We are now ready to state our main result that upper bounds the worst-case accumulated regret of the proposed MLE-UCB policy in Algorithm \[alg:mle-ucb\].
Suppose that $T_0\asymp \max\{\nu^2d\log T/\lambda_0^2, \rho^2(d+\log T)/(\tau^2\lambda_0)\}$ and $\tau\asymp 1/\sqrt{\rho^2\nu^2K^2}$ in Algorithm \[alg:mle-ucb\], then the regret of the MLE-UCB policy is upper bounded by $$\label{eq:thm-main}
C_1\left[ d\sqrt{T}\cdot \log(\lambda_0^{-1}\rho\nu TK)+ {d^2\lambda_0^{-2}\rho^4\nu^2 K^2\log T}\right] + C_2,$$ where $C_1, C_2>0$ are universal constants. \[thm:mle-ucb\]
In addition to universal constants, the regret upper bound established in Theorem \[thm:mle-ucb\] has two terms. The first term, $d\sqrt{T}\cdot \log(\lambda_0^{-1}\rho\nu TK)$, is the main regret term that scales as ${\widetilde}O(d\sqrt{T})$ dropping logarithmic dependency. The second $d^2\lambda_0^{-2}\rho^4\nu^2K^2\log T$ term is a minor term, because it only scales logarithmically with the time horizon $T$. One remarkable aspect of Theorem \[thm:mle-ucb\] is the fact that the regret upper bound has *no* dependency on the total number of items $N$ (even in a logarithmic term). This is an attractive property of the proposed policy, which allows $N$ to be very large, even exponentially large in $d$ and $K$.
Proof sketch of Theorem \[thm:mle-ucb\]
---------------------------------------
We provide a proof sketch of Theorem \[thm:mle-ucb\] in this section. The proofs of technical lemmas are relegated to the online supplement.
The proof is divided into four steps. In the first step, we analyze the pilot estimator $\theta^*$ obtained from the pure exploration phase of Algorithm \[alg:mle-ucb\], and show as a corollary that the true model $\theta_0$ is feasible to all subsequent local MLE formulations with high probability (see Corollary \[cor:tau\]). In the second step, we use an $\varepsilon$-net argument to analyze the estimation error of the local MLE. Afterwards, we show in the third step that an upper bound on the estimation error ${\widehat}\theta_{t-1}-\theta_0$ implies an upper bound on the estimation error of the expected revenue $R_t(S)$, hence showing that ${\overline}R_t(S)$ are valid upper confidence bounds. Finally, we apply the *elliptical potential lemma*, which also plays a key role in linear stochastic bandit and its variants, to complete our proof.
### Analysis of pure exploration and the pilot estimator
Our first step is to establish an upper bound on the estimation error $\|\theta^*-\theta_0\|_2$ of the pilot estimator $\theta^*$, built using pure exploration data. It should be noted that in the pure exploration phase ($t\in\{1,\cdots,T_0\}$), the assortments $\{S_t\}_{t=1}^{T_0}$ only consist of one item. Therefore the observation model reduces to a standard *generalized linear model* with the sigmoid function $\sigma(x)=1/(1+e^{-x})= e^x/(1+e^x)$ as the link function, which is essentially a logistic regression model of observing 1 if the customer makes a purchase. Because the choice model in the pure exploration phase reduces to a generalized linear model, we can cite existing works to upper bound the error $\|\theta^*-\theta_0\|_2$. In particular, the following lemma is cited from [@li2017provably Eq. (18)], adapted to our model and parameter settings. The details on how to adapt the result from [@li2017provably] provided in the supplementary material.
With probability $1-\delta$ it holds that $$\|\theta^*-\theta_0\|_2 \leq \frac{2}{\kappa}\sqrt{\frac{d+\log(1/\delta)}{\lambda_{\min}(V)}}
\;\;\;\;\text{where}\;\;\kappa = \frac{1}{2e(1+\rho)}\;\;\text{and}\;\; V = \sum_{t=1}^{T_0}v_{t,i_t}v_{t,i_t}^\top.$$ \[lem:pilot\]
The following corollary immediately follows Lemma \[lem:pilot\], by lower bounding $\lambda_{\min}(V)$ using standard matrix concentration inequalities. Its proof is again deferred to the supplementary material.
There exists a universal constant $C_0>0$ such that for arbitrary $\tau\in(0,1/2]$, if $T_0 \geq C_0\max\{\nu^2d\log T/\lambda_0^2, \rho^2 (d+\log T)/(\tau^2\lambda_0)\}$ then with probability $1-O(T^{-1})$, $\|\theta^*-\theta_0\|_2\leq \tau$. \[cor:tau\]
The purpose of Corollary \[cor:tau\] is to establish a connection between the number of pure exploration iterations $T_0$ and the critical radius $\tau$ used in the local MLE formulation. It shows a lower bound on $T_0$ in order for the estimation error $\|\theta^*-\theta_0\|_2$ to be upper bounded by $\tau$ with high probability, which certifies that the true model $\theta_0$ is also a *feasible* local estimator in our MLE-UCB policy. This is an important property for later analysis of local MLE solutions ${\widehat}\theta_{t-1}$.
### Analysis of the local MLE {#sec:ana_local}
The following lemma upper bounds a Mahalanobis distance between ${\widehat}\theta_t$ and $\theta_0$. For convenience, we adopt the notation that $r_{t0}=0$ and $v_{t0}=0$ for all $t$ throughout this section. We also define $$\begin{aligned}
I_t(\theta) & := &\sum_{t'=1}^t M_{t'}(\theta), \label{eq：pop_fisher} \\
M_{t'}(\theta)& := &-\mathbb E_{\theta_0,t'}[\nabla^2_\theta \log p_{\theta,t'}(j|S_{t'})] \nonumber \\
& = &\mathbb E_{\theta_0,t'}[v_{t'j}v_{t'j}^\top] - \{\mathbb E_{\theta_0,t'}v_{t'j}\}\{\mathbb E_{\theta,t'}v_{t'j}\}^\top
- \{\mathbb E_{\theta,t'}v_{t'j}\} \{\mathbb E_{\theta_0,t'}v_{t'j}\}^\top + \{\mathbb E_{\theta,t'}v_{t'j}\}\{\mathbb E_{\theta,t'}v_{t'j}\}^\top \nonumber\end{aligned}$$ where $\mathbb E_{\theta,t'}$ denotes the expectation evaluated under the law $j\sim p_{\theta,t'}(\cdot|S_{t'})$; that is, $p_{\theta,t'}(j|S_{t'}) =\exp\{v_{t'j}^\top\theta\}/(1+\sum_{k\in S_{t'}}\exp\{v_{t'j}^\top\theta\})$ for $j\in S_{t'}$ and $p_{\theta,t'}(j|S_{t'})=0$ for $j\notin S_{t'}$.
Suppose $\tau\leq 1/\sqrt{8\rho \nu^2K^2}$. Then there exists a universal constant $C>0$ such that with probability $1-O(T^{-1})$ the following holds uniformly over all $t=T_0,\cdots,T-1$: $$({\widehat}\theta_t-\theta_0)^\top I_t(\theta_0)({\widehat}\theta_t-\theta_0) \leq C\cdot d\log(\rho\nu TK).$$ \[lem:mle\]
For $\theta=\theta_0$, the expression of $M_{t'}(\theta)$ can be simplified as $M_{t'}(\theta_0) = \mathbb E_{\theta_0,t'}[v_{t'j}v_{t'j}^\top] - \{\mathbb E_{\theta_0,t'}v_{t'j}\}\{\mathbb E_{\theta_0,t'}v_{t'j}\}^\top$.
The complete proof of Lemma \[lem:mle\] is given in the supplementary material, and here we provide some high-level ideas behind our proof.
Our proof is inspired by the classical convergence rate analysis of M-estimators [@van1998asymptotic Sec. 5.8]. The main technical challenge is to provide *finite-sample* analysis of several components in the proof of [@van1998asymptotic Sec. 5.8].
In particular, for any $\theta\in\mathbb R^d$, consider $$F_{t}(\theta) := \sum_{t'\leq t}f_{t'}(\theta)\;\;\;\;\text{where}\;\;
f_{t'}(\theta) := \mathbb E_{\theta_0,t'}\left[\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}\right] = \sum_{j\in S_{t'}\cup\{0\}} p_{\theta_0,t'}(j|S_{t'})\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}$$ and its “sample” version $${\widehat}F_t(\theta) := \sum_{t'\leq t} {\widehat}f_{t'}(\theta)\;\;\;\;\text{where}\;\;
{\widehat}f_{t'}(\theta) := \log\frac{p_{\theta,t'}(i_{t'}|S_{t'})}{p_{\theta_0,t'}(i_{t'}|S_{t'})}.$$
It is easy to verify by definition that $F_t({\widehat}\theta_{t})\geq F_t(\theta_0)=0$ and ${\widehat}F_t({\widehat}\theta_{t})\leq {\widehat}{F}_t(\theta_0)=0$, because $F_t(\cdot)$ is a Kullback-Leibler divergence, $\theta_0$ is feasible to the local MLE formulation and ${\widehat}\theta_{t-1}$ is the optimal solution. On the other hand, it can be proved that $|F_t(\theta)-{\widehat}F_t(\theta)|$ is small for all $\theta$ with high probability, by using concentration inequalities for self-normalized empirical process (note that $\mathbb E{\widehat}f_{t'}(\theta) = f_{t'}(\theta)$ for any $\theta$). Moreover, by constructing a local quadratic approximation of $F_t(\cdot)$ around $\theta_0$, we can show that $F_t(\theta)-F_t(\theta_0)$ is large when $\theta$ is far away from $\theta_0$.
[Following the above observations, we can use proof by contradiction to prove Lemma \[lem:mle\], which essentially claims that ${\widehat}\theta_{t}$ and $\theta_0$ are close under the quadratic distance $\|\cdot\|_{I_t(\theta_0)}$. Suppose by contradiction that ${\widehat}\theta_{t}$ and $\theta_0$ are far apart, which implies that $|F_t({\widehat}\theta_{t})-F_t(\theta_0)|$ is large. On the other hand, by the fact that ${\widehat}F_t({\widehat}\theta_{t})\leq 0 = {F}_t(\theta_0) \leq F_t({\widehat}\theta_{t})$, we have $$|F_t({\widehat}\theta_{t})-F_t(\theta_0)| = |F_t({\widehat}\theta_{t})| \leq |F_t({\widehat}\theta_{t})-{\widehat}F_t({\widehat}\theta_{t}) |.$$ By the established concentration result, we have $|F_t(\theta)-{\widehat}F_t(\theta)|$ is small for all $\theta$ with high probability (including $\theta={\widehat}\theta_{t}$). This leads to the desired contradiction.]{}
### Analysis of upper confidence bounds
The following technical lemma shows that the upper confidence bounds constructed in Algorithm \[alg:mle-ucb\] are valid with high probability. Additionally, we establish an upper bound on the discrepancy between ${\overline}R_t(S)$ and the true value $R_t(S)$ defined in Eq. .
Suppose $\tau$ satisfies the condition in Lemma \[lem:mle\]. With probability $1-O(T^{-1})$ the following holds uniformly for all $t>T_0$ and $S\subseteq[N]$, $|S|\leq K$ such that
1. ${\overline}R_t(S)\geq R_t(S)$;
2. $|{\overline}R_t(S)-R_t(S)| \lesssim \min\{1,\omega\sqrt{\|I_{t-1}^{-1/2}(\theta_0) M_t(\theta_0|S) I_{t-1}^{-1/2}(\theta_0)\|_\op}\}$.
\[lem:ucb\]
At a higher level, the proof of Lemma \[lem:ucb\] can be regarded as a “finite-sample” version of the classical *Delta’s method*, which upper bounds estimation error of some functional $\varphi$ of parameters, i.e., $|\varphi({\widehat}\theta_{t-1})-\varphi(\theta_0)|$ using the estimation error of the parameters themselves ${\widehat}\theta_{t-1}-\theta_0$. The complete proof is relegated to the supplementary material.
### The elliptical potential lemma {#sec:ellipitical}
Let $S_t^*$ be the assortment that maximizes the expected revenue $R_t(\cdot)$ (defined in Eq. ) at time period $t$, and $S_t$ be the assortment selected by Algorithm \[alg:mle-ucb\]. Because $R_t(S)\leq {\overline}R_t(S)$ for all $S$ (see Lemma \[lem:ucb\]), we have the following upper bound for each term in the regret (see Eq. ): $$R_t(S_t^*)-R_t(S_t)
\leq ({\overline}R_t(S_t^*)-{\overline}R_t(S_t)) + ({\overline}R_t(S_t)-R_t(S_t)) \leq {\overline}R_t(S_t)-R_t(S_t),
\label{eq:ucb-triangle}$$ where the last inequality holds because ${\overline}R_t(S_t^*)-{\overline}R_t(S_t) \leq 0$ (note that $S_t$ maximizes ${\overline}{R}_t(\cdot)$).
Subsequently, invoking Lemma \[lem:ucb\] and the Cauchy-Schwarz inequality, we have $$\begin{aligned}
\sum_{t=T_0+1}^T& R_t(S_t^*)-R_t(S_t)
\lesssim \sqrt{d\log(\rho\nu TK)}\cdot \sum_{t=T_0+1}^T \sqrt{\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op\}}\nonumber\\
&\lesssim \sqrt{dT\log(\rho\nu TK)\cdot\sum_{t=T_0+1}^T\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op^2 \}}.\label{eq:ucb-regret-decomp}\end{aligned}$$
The following lemma is a key result that upper bounds $\sum_{t=T_0+1}^T\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op^2\}$. It is usually referred to as the *elliptical potential lemma* and has found many applications in contextual bandit-type problems (see, e.g., [@Dani08stochastic; @Rusmevichientong2010; @filippi2010parametric; @li2017provably]).
It holds that $$\sum_{t=T_0+1}^T\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op^2\} \leq
4\log\frac{\det I_{T}(\theta_0)}{\det I_{T_0}(\theta_0)} \lesssim d\log(\lambda_0^{-1}\rho\nu).$$ \[lem:elliptical\]
The proof of Lemma \[lem:elliptical\] is placed in the supplementary material. It is a routine proof following existing proofs of elliptical potential lemmas using matrix-determinant rank-1 updates.
We are now ready to give the final upper bound on $\mathrm{Regret}(\{S_t\}_{t=1}^T)$ defined in Eq. . Note that the total regret incurred by the pure exploration phase is upper bounded by $T_0$, because the revenue parameters $r_{tj}$ are normalized so that they are upper bounded by 1. In addition, as the failure event of ${\overline}R_t(S)\leq R_t(S)$ for some $S$ occurs with probability $1-O(T^{-1})$, the total regret accumulated under the failure event is $O(T^{-1})\cdot T = O(1)$. Further invoking Eq. (\[eq:ucb-regret-decomp\]) and Lemma \[lem:elliptical\], we have $$\begin{aligned}
\mathrm{Regret}(\{S_t\}_{t=1}^T)
&\leq T_0 + O(1) + \mathbb E\sum_{t=T_0+1}^T R_t(S_t^*)-R_t(S_t)\nonumber\\
&\lesssim O(1) + \frac{\nu^2d\log T}{\lambda_0^2} + \frac{\rho^2(d+\log T)}{\tau^2\lambda_0} + d\sqrt{T}\cdot \log(\lambda_0^{-1}\rho\nu TK)\nonumber\\
&\lesssim O(1) + {d^2\lambda_0^{-2}\rho^4\nu^2 K^2\log T}+ d\sqrt{T}\cdot \log(\lambda_0^{-1}\rho\nu TK).\end{aligned}$$
Lower bound {#sec:negative}
===========
\[sec:lower\] To complement our regret analysis in Sec. \[subsec:regret\], in this section we prove a *lower* bound for worst-case regret. Our lower bound is information theoretical, and therefore applies to *any* policy for dynamic assortment optimization with changing contextual features.
Suppose $d$ is divisible by 4. There exists a universal constant $C_0>0$ such that for any sufficiently large $T$ and policy $\pi$, there is a worst-case problem instance with $N\asymp K\cdot 2^d$ items and uniformly bounded feature and coefficient vector (i.e., $\|v_{ti}\|_2\leq 1$ and $\|\theta_0\|_2\leq 1$ for all $i\in[N]$, $t\in[T]$) such that the regret of $\pi$ is lower bounded by $C_2\cdot d\sqrt{T}/K$. \[thm:lower\]
Theorem \[thm:lower\] essentially implies that the ${\widetilde}O(d\sqrt{T})$ regret upper bound established in Theorem \[thm:mle-ucb\] is tight (up to logarithmic factors) in $T$ and $d$. Although there is an $O(K)$ gap between the upper and lower regret bounds, in practical applications $K$ is usually small and can be generally regarded as a constant. It is an interesting technical open problem to close this gap of $O(K)$.
We also remark that an $\Omega(d\sqrt{T})$ lower bound was established in [@Dani08stochastic] for contextual linear bandit problems. However, in assortment selection, the reward function is *not* coordinate-wise decomposable, making techniques in [@Dani08stochastic] not directly applicable. In the following subsection, we provide a high-level proof sketch of Theorem \[thm:lower\], with complete proofs of technical lemmas relegated to the supplementary material.
Proof sketch of Theorem \[thm:lower\]
-------------------------------------
At a higher level, the proof of Theorem \[thm:lower\] can be divided into three steps (separated into three different sub-sections below). In the first step, we construct an *adversarial parameter* set and reduce the task of lower bounding the *worst-case* regret of any policy to lower bounding the *Bayes risk* of the constructed parameter set. In the second step, we use a “counting argument” similar to the one developed in [@Chen:18tight] to provide an explicit lower bound on the Bayes risk of the constructed adversarial parameter set, and finally we apply *Pinsker’s inequality* (see, e.g., [@tsybakov2009introduction]) to derive a complete lower bound.
### Adversarial construction and the Bayes risk
Let $\epsilon\in(0, 1/d\sqrt{d})$ be a small positive parameter to be specified later. For every subset $W\subseteq[d]$, define the corresponding parameter $\theta_W\in\mathbb R^d$ as $[\theta_W]_i=\epsilon$ for all $i\in W$, and $[\theta_W]_i=0$ for all $i\notin W$. The parameter set we consider is $$\theta \in \Theta := \{\theta_W: W\in\mathcal W_{d/4}\} := \{\theta_W: W\subseteq[d], |W|=d/4\}.$$ Note that $d/4$ is a positive integer because $d$ is divisible by 4, as assumed in Theorem \[thm:lower\]. Also, to simplify notation, we use $\mathcal W_k$ to denote the class of all subsets of $[d]$ whose size is $k$.
The feature vectors $\{v_{ti}\}$ are constructed to be invariant across time iterations $t$. For each $t$ and $U\in\mathcal W_{d/4}$, $K$ identical feature vectors $v_{U}$ are constructed as (recall that $K$ is the maximum allowed assortment capacity) $$[v_U]_i = 1/\sqrt{d} \;\;\;\;\text{for}\;\; i\in U; \;\;\;\;\;\;
[v_U]_i = 0 \;\;\;\;\text{for}\;\; i\notin U.$$
It is easy to check that with the condition $\epsilon\in(0,1/\sqrt{d})$, $\|\theta_W\|_2\leq 1$ and $\|v_U\|_2\leq 1$ for all $W,U\in\mathcal W_{d/4}$. Hence the worst-case regret of any policy $\pi$ can be lower bounded by the worst-case regret of parameters belonging to $\Theta$, which can be further lower bounded by the “average” regret over a uniform prior over $\Theta$: $$\begin{aligned}
\sup_{v,\theta}\mathbb E_{v,\theta}^\pi\sum_{t=1}^TR(S_\theta^*)-R(S_t)
&\geq \max_{\theta_W\in\Theta} \mathbb E_{v,\theta_W}^\pi\sum_{t=1}^TR(S_{\theta_W}^*)-R(S_t)\nonumber\\
&\geq \frac{1}{|\mathcal W_{d/4}|}\sum_{W\in\mathcal W_{d/4}}\mathbb E_{v,\theta_W}^\pi\sum_{t=1}^TR(S_{\theta_W}^*)-R(S_t).
\label{eq:bayes-risk}\end{aligned}$$
Here $S_\theta^*$ is the optimal assortment of size at most $K$ that maximizes (expected) revenue under parameterization $\theta$. By construction, it is easy to verify that $S_{\theta_W}^*$ consists of all $K$ items corresponding to feature $v_W$. We also employ constant revenue parameters $r_{ti}\equiv 1$ for all $t\in[T]$, $i\in[N]$.
### The counting argument
In this section we drive an explicit lower bound on the Bayes risk in Eq. (\[eq:bayes-risk\]). For any sequences $\{S_t\}_{t=1}^T$ produced by the policy $\pi$, we first describe an alternative sequence $\{{\widetilde}S_t\}_{t=1}^T$ that provably enjoys less regret under parameterization $\theta_W$, while simplifying our analysis.
Let $v_{U_1},\cdots,v_{U_L}$ be the distinct feature vectors contained in assortment $S_t$ (if $S_t=\emptyset$ then one may choose an arbitrary feature $v_U$) with $U_1,\cdots,U_L\in\mathcal W_{d/4}$. Let $U^*$ be the subset among $U_1,\cdots,U_L$ that maximizes $\langle v_{U^*},\theta_W\rangle$, where $\theta_W$ is the underlying parameter. Let ${\widetilde}S_t$ be the assortment consisting of all $K$ items corresponding to feature $v_U^*$. We then have the following observation:
$R(S_t) \leq R({\widetilde}S_t)$ under $\theta_W$. \[prop:reduction-surrogate-st\]
Because $r_{tj}\equiv 1$ in our construction, we have $R(S_t)=(\sum_{j\in S_t}u_{j})/(1+\sum_{j\in S_t}u_j)$ where $u_j=\exp\{v_j^\top\theta_W\}$ under $\theta_W$. Clearly $R(S)$ is a monotonically non-decreasing function in $u_j$. By replacing all $v_j\in S_t$ with $v_{U^*}\in{\widetilde}S_t$, the $u_j$ values do not decrease and therefore the Proposition holds true.
To simplify notation we also use ${\widetilde}U_t$ to denote the unique $U^*\in\mathcal W_{d/4}$ in ${\widetilde}S_t$. We also use $\mathbb E_W$ and $\mathbb P_W$ to denote the law parameterized by $\theta_W$ and policy $\pi$. The following lemma gives a lower bound on $R({\widetilde}S_t)-R(S_{\theta_W}^*)$ by comparing it with $W$, which is also proved in the supplementary material.
Suppose $\epsilon\in(0,1/d\sqrt{d})$ and define $\delta := d/4 - |{\widetilde}U_t\cap W|$. Then $$R(S_{\theta_W}^*)-R({\widetilde}S_t) \geq\frac{\delta\epsilon}{4K\sqrt{d}} .$$ \[lem:R-lb\]
Define random variables ${\widetilde}N_i := \sum_{t=1}^T{\boldsymbol }1\{i\in{\widetilde}U_t\}$. Lemma \[lem:R-lb\] immediately implies $$\mathbb E_W\sum_{t=1}^TR(S_{\theta_W}^*)-R({\widetilde}S_t)
\geq \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{4} - \sum_{i\in W}\mathbb E_W[{\widetilde}N_i] \right), \;\;\;\;\;\forall W\in\mathcal W_{d/4}.
\label{eq:count1}$$
Denote $\mathcal W_{d/4}^{(i)} := \{W\in\mathcal W_{d/4}: i\in W\}$ and $\mathcal W_{d/4-1} := \{W\subseteq[d]: |W|=d/4-1\}$. Averaging both sides of Eq. (\[eq:count1\]) with respect to all $W\in\mathcal W_{d/4}$ and swapping the summation order, we have $$\begin{aligned}
\frac{1}{|\mathcal W_{d/4}|}\sum_{W\in\mathcal W_{d/4}}&\mathbb E_{W}\sum_{t=1}^TR(S_{\theta_W}^*)-R(S_t)
\geq \frac{\epsilon}{4K\sqrt{d}} \frac{1}{|\mathcal W_{d/4}|}\sum_{W\in\mathcal W_{d/4}}\left(\frac{dT}{4} - \sum_{i\in W}\mathbb E_W[{\widetilde}N_i]\right)\nonumber\\
&= \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{4} - \frac{1}{|\mathcal W_{d/4}|} \sum_{i=1}^d\sum_{W\in\mathcal W_{d/4}^{(i)}}\mathbb E_W[{\widetilde}N_i]\right)\nonumber\\
&= \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{4} - \frac{1}{|\mathcal W_{d/4}|} \sum_{W\in\mathcal W_{d/4-1}}\sum_{i\notin W}\mathbb E_{W\cup\{i\}}[{\widetilde}N_i]\right)\nonumber\\
&\geq \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{4} - \frac{|\mathcal W_{d/4-1}|}{|\mathcal W_{d/4}|} \max_{W\in\mathcal W_{d/4-1}}\sum_{i\notin W}\mathbb E_{W\cup\{i\}}[{\widetilde}N_i]\right)\\
&= \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{4} - \frac{|\mathcal W_{d/4-1}|}{|\mathcal W_{d/4}|} \max_{W\in\mathcal W_{d/4-1}}\sum_{i\notin W}\mathbb E_{W}[{\widetilde}N_i] + \mathbb E_{W\cup\{i\}}[{\widetilde}N_i]-\mathbb E_W[{\widetilde}N_i]\right).\end{aligned}$$
Note that for any fixed $W$, $\sum_{i\notin W}\mathbb E_W[{\widetilde}N_i] \leq \sum_{i=1}^d\mathbb E_W[{\widetilde}N_i]\leq dT/4$. Also, $|\mathcal W_{d/4-1}|/|\mathcal W_{d/4}| = \binom{d}{d/4-1}/\binom{d}{d/4}=\frac{d/4}{3d/4+1} \leq 1/3$. Subsequently, $$\frac{1}{|\mathcal W_{d/4}|}\sum_{W\in\mathcal W_{d/4}}\mathbb E_{W}\sum_{t=1}^TR(S_{\theta_W}^*)-R(S_t)
\geq \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{6} - \max_{W\in\mathcal W_{d/4-1}}\sum_{i\notin W}|\mathbb E_{W\cup\{i\}}[{\widetilde}N_i]-\mathbb E_W[{\widetilde}N_i]|\right).
\label{eq:count2}$$
### Pinsker’s inequality
In this section we concentrate on upper bounding $|\mathbb E_{W\cup\{i\}}[{\widetilde}N_i]-\mathbb E_W[{\widetilde}N_i]|$ for any $W\in\mathcal W_{d/4-1}$. Let $P=\mathbb P_W$ and $Q=\mathbb P_{W\cup\{i\}}$ denote the laws under $\theta_W$ and $\theta_{W\cup\{i\}}$, respectively. Then $$\begin{aligned}
\big|\mathbb E_P[{\widetilde}N_i]-\mathbb E_Q[{\widetilde}N_i]\big|
& \leq \sum_{j=0}^Tj\cdot \big|P[{\widetilde}N_i=j] - Q[{\widetilde}N_i=j]\big|\\
&\leq T\cdot \sum_{j=0}^T\big|P[{\widetilde}N_i=j] - Q[{\widetilde}N_i=j]\big|\\
&\leq T\cdot \|P-Q\|_{\mathrm{TV}} \leq T\cdot\sqrt{\frac{1}{2}{\mathrm{KL}}(P\|Q)},\end{aligned}$$ where $\|P-Q\|_{\mathrm{TV}}=\sup_A|P(A)-Q(A)|$ is the total variation distance between $P$, $Q$, ${\mathrm{KL}}(P\|Q)=\int(\log{\mathrm d}P/{\mathrm d}Q){\mathrm d}P$ is the Kullback-Leibler (KL) divergence between $P$, $Q$, and the inequality $\|P-Q\|_{\mathrm{TV}} \leq \sqrt{\frac{1}{2}{\mathrm{KL}}(P\|Q)}$ is the celebrated Pinsker’s inequality.
For every $i\in[d]$ define random variables $N_i := \sum_{t=1}^T \frac{1}{K}\sum_{v_{U}\in S_t}{\boldsymbol }1\{i\in U\}$. The next lemma upper bound the KL divergence, which is proved in the supplementary material.
For any $W\in\mathcal W_{d/4-1}$ and $i\in[d]$, ${\mathrm{KL}}(P_W\|P_{W\cup\{i\}}) \leq C_{{\mathrm{KL}}}\cdot \mathbb E_W[N_i] \cdot\epsilon^2/{d}$ for some universal constant $C_{{\mathrm{KL}}}>0$. \[lem:kl\]
Combining Lemma \[lem:kl\] and Eq. (\[eq:count2\]), we have $$\frac{1}{|\mathcal W_{d/4}|}\sum_{W\in\mathcal W_{d/4}}\mathbb E_{W}\sum_{t=1}^TR(S_{\theta_W}^*)-R(S_t)
\geq \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{6} - T\sum_{i=1}^d\sqrt{C_{{\mathrm{KL}}}\mathbb E_W[N_i]\epsilon^2/d}\right).$$
Further using Cauchy-Schwartz inequality, we have $$\sum_{i=1}^d\sqrt{C_{{\mathrm{KL}}}\mathbb E_W[N_i]\epsilon^2/d}
\leq \sqrt{d}\cdot \sqrt{\sum_{i=1}^d C_{{\mathrm{KL}}}\mathbb E_W[N_i]\epsilon^2/d},$$ which is further upper bounded by $\sqrt{d}\cdot \sqrt{C_{{\mathrm{KL}}}T\epsilon^2/4}$ because $\sum_{i=1}^d\mathbb E_W[N_i]\leq dT/4$. Subsequently, $$\frac{1}{|\mathcal W_{d/4}|}\sum_{W\in\mathcal W_{d/4}}\mathbb E_{W}\sum_{t=1}^TR(S_{\theta_W}^*)-R(S_t) \geq \frac{\epsilon}{4K\sqrt{d}}\left(\frac{dT}{6} -T \sqrt{C_{{\mathrm{KL}}}'dT\epsilon^2}\right),$$ where $C_{{\mathrm{KL}}}'=C_{{\mathrm{KL}}}/4$. Setting $\epsilon = \sqrt{d/144C_{{\mathrm{KL}}}'T}$ we complete the proof of Theorem \[thm:lower\].
The combinatorial optimization subproblem {#sec:comb}
=========================================
The major computational bottleneck of our algorithm is its Step \[alg:step6\], which involves solving a *combinatorial* optimization problem. For notational simplicity, we equivalently reformulate this problem as follows: $$\begin{aligned}
\max_{S\subseteq[N], |S|\leq K} &\mathrm{ESTR}(S) + \min\left\{1, \omega\cdot \mathrm{CI}(S)\right\} \;\;\;\;\;\text{where}\;\;\mathrm{ESTR}(S):=\frac{\sum_{j\in S}r_{tj}{\widehat}u_{tj}}{1+\sum_{j\in S}{\widehat}u_{tj}}\;\;\text{and}\;\;
\label{eq:comb-opt}\\
& \mathrm{CI}(S) := \sqrt{\left\|\frac{\sum_{j\in S}{\widehat}u_{tj} x_{tj} x_{tj}^\top}{1+\sum_{j\in S}{\widehat}u_{tj}} -\left(\frac{\sum_{j\in S}{\widehat}u_{tj}x_{tj}}{1+\sum_{j\in S}{\widehat}u_{tj}}\right)\left(\frac{\sum_{j\in S}{\widehat}u_{tj}x_{tj}}{1+\sum_{j\in S}{\widehat}u_{tj}}\right)^\top \right\|_\op}.\nonumber\end{aligned}$$
Here ${\widehat}u_{tj} := \exp\{v_{tj}^\top{\widehat}\theta_{t-1}\}$ and $x_{tj} := {\widehat}I_{t-1}^{-1/2}({\widehat}\theta_{t-1})v_{tj}$, both of which can be pre-computed before solving Eq. (\[eq:comb-opt\]). A brute-force way to compute Eq. (\[eq:comb-opt\]) is to enumerate all subsets $S\subseteq[N]$, $|S|\leq K$ and select the one with the largest objective value. Such an approach is *not* an efficient (polynomial-time) algorithm and is therefore not scalable.
In this section we provide two alternative methods for (approximately) solving the combinatorial optimization problem in Eq. (\[eq:comb-opt\]). Our first algorithm is based on discretized dynamic programming and enjoys rigorous approximation guarantees. The second algorithm is a computationally efficient greedy heuristic. Although the greedy heuristic does not have rigorous guarantees, our numerical result suggests it works reasonably well (see Sec. \[sec:numerical\]).
Approximation algorithms for assortment optimization {#sec:approx}
----------------------------------------------------
In this section we introduce algorithms with polynomial running times and rigorous approximation guarantees for the optimization task described in Eq. (\[eq:comb-opt\]). We first formally introduce the concept of *$(\alpha,\varepsilon,\delta)$-approximation* to characterize the approximation performance, and show that such approximation guarantees imply certain upper bounds on the final regret.
Fix $\alpha\geq 1$, $\varepsilon\geq 0$ and $\delta\in[0,1)$. An algorithm is an *$(\alpha,\varepsilon,\delta)$-approximation algorithm* if it produces ${\widehat}S\subseteq[N]$, $|{\widehat}S|\leq K$ such that with probability at least $1-\delta$, $$\mathrm{ESTR}({\widehat}S) + \min\{1, \alpha\omega\cdot \mathrm{CI}({\widehat}S)\} + \varepsilon \geq \mathrm{ESTR}(S^*) + \min\{1,\omega\cdot\mathrm{CI}(S^*)\},
\label{eq:approx}$$ where $S^*$ is the assortment set maximizing the actual objective in Eq. (\[eq:comb-opt\]) [^4]. \[defn:approx\]
The following lemma shows how $(\alpha,\varepsilon,\delta)$-approximation algorithms imply an upper bound on the accumulated. It is proved using standard analysis of UCB type algorithms, with the complete proof given in the supplementary material.
Suppose an $(\alpha,\varepsilon,\delta)$-approximation algorithm is used instead of exact optimization in the MLE-UCB policy at each time period $t$. Then its regret can be upper bounded by $$\alpha\cdot \mathrm{Regret}^* + \varepsilon T + \delta T^2 + O(1),$$ where $\mathrm{Regret}^*$ is the regret upper bound shown by Theorem \[thm:mle-ucb\] for Algorithm \[alg:mle-ucb\] with exact optimization in Step \[alg:step6\]. \[lem:approx\]
In the rest of this section we introduce our proposed approximation algorithm and the approximation guarantee. To highlight the main idea of the approximation algorithm, we only describe how the algorithm operates in the *univariate* ($d=1$) case, while leaving the general multivariate ($d>1$) case to the appendix.
Our approximation algorithm can be roughly divided into three steps. In the first step, we use a “discretization” trick to approximate the objective function using “rounded” parameter values. Such rounding motivates the second step, in which we define “reachable states” and present a simple yet computationally expensive brute-force method to enumerate all reachable states, and establish approximation guarantees for such methods. [This brute-force method is only presented for illustration purposes and will be replaced by a dynamic programing algorithm proposed in the third step.]{} In particular, a *dynamic programming* algorithm is developed to compute which states are “reachable” in polynomial time.
### The discretization trick
In the univariate case, $\{x_{tj}\}$ are scalars and therefore $x_{tj}x_{tj}^\top$ is simply $x_{tj}^2$. Let $\Delta>0$ be a small positive discretization parameter to be specified later. For all $i\in[N]$, define $$\mu_i := \left[\frac{{\widehat}u_{ti}}{\Delta}\right]\Delta, \;\;
\alpha_i := \left[\frac{{\widehat}u_{ti}x_{ti}}{\Delta}\right]\Delta,\;\;
\beta_i := \left[\frac{{\widehat}u_{ti}x_{ti}^2}{\Delta}\right]\Delta,\;\;
\gamma_i := \left[\frac{{\widehat}u_{ti}r_{ti}}{\Delta}\right]\Delta,
\label{eq:muabg}$$ where $[a]$ denotes the nearest integer a real number $a$ is rounded into. Intuitively, $\mu_i$ is the real number closest to ${\widehat}u_{ti}$ that is an *integer* multiple of the discretization parameter $\Delta$, and similarly for $\alpha_i,\beta_i,\gamma_i$.
The motivation for the definitions of $\{\mu_i,\alpha_i,\beta_i,\gamma_i\}$ is their *sufficiency* in computing the objective function $\mathrm{ESTR}(S)+\min\{1,\omega\cdot \mathrm{CI}(S)\}$. Indeed, for any $S\subseteq[n]$, $|S|\leq K$, define $\mu=\sum_{j\in S}\mu_j$, $\alpha = \sum_{j\in S}\alpha_j$, $\beta=\sum_{j\in S}\beta_j$, $\gamma=\sum_{j\in S}\gamma_j$ and $${\widehat}{\mathrm{ESTR}}(S) := \frac{\gamma}{1+\mu}, \;\;\;\;\;\;
{\widehat}{\mathrm{CI}}(S) := \max\left\{0,\sqrt{\frac{\beta}{1+\mu}-\left(\frac{\alpha}{1+\mu}\right)^2}\right\}.$$
Following the definition of $\mathrm{ESTR}(S)$ and $\mathrm{CI}(S)$, it is easy to see that ${\widehat}{\mathrm{ESTR}}(S)\to \mathrm{ESTR}(S)$ and ${\widehat}{\mathrm{CI}}(S)\to\mathrm{CI}(S)$ as $\Delta\to 0^+$. The following lemma gives a more precise control of the error between ${\widehat}{\mathrm{ESTR}}(S)$, ${\widehat}{\mathrm{CI}}(S)$ and $\mathrm{ESTR}(S)$, $\mathrm{CI}(S)$ using the values of $\Delta$ and the maximum utility parameter in $S$.
For any $S\subseteq[N]$, $|S|\leq K$, suppose $U = \max_{j\in S}\{1, {\widehat}{u}_{tj}\}$ and $\Delta = \epsilon_0 U / K$ for some $\epsilon_0>0$. Suppose also $|x_{tj}|\leq\nu$ for all $t,j$. Then $$\big|\mathrm{ESTR}(S)-{\widehat}{\mathrm{ESTR}}(S)\big| \leq 6\epsilon_0\;\;\;\;\;\text{and}\;\;\;\;\;
\big|\mathrm{CI}(S)-{\widehat}{\mathrm{CI}}(S)\big| \leq \sqrt{24 \epsilon_0} (1 +\nu),$$ \[lem:approx-1d-estr-ci\]
The complete proof of Lemma \[lem:approx-1d-estr-ci\] is relegated to the supplementary material.
### Reachable states and a brute-force algorithm
To apply the estimation error bounds in Lemma \[lem:approx-1d-estr-ci\] one needs to first enumerate $q\in[N]$ giving rise to the item in $S$ with the largest utility parameter ${\widehat}u_{tq}$. After such an element $q$ is enumerated, the discretization parameter $\Delta=\epsilon_0 U/K=\epsilon_0\max\{1,{\widehat}u_{tq}\}/K$ can be determined and discretized values $\mu_i,\alpha_i,\beta_i,\gamma_i$ can be computed for all $i\in[N]/\backslash\{q\}$. It is also easy to verify that there are at most $O(K/\epsilon)$ possible values of $\mu_i,\gamma_i$, $O(K\nu/\epsilon)$ possible values of $\alpha_i$ and $O(K\nu^2/\epsilon)$ possible values of $\beta_i$ (recall that $\nu$ is the upper bound of $|x_{tj}$ for all $t$ and $j$).
For any $i\in[N]\cup\{0\}$, $k\in[K]\cup\{0\}$ and $\mu,\alpha,\beta,\gamma\geq 0$ being *integer* multiples of $\Delta$, we use a tuple $\varsigma_i^k(\mu,\alpha,\beta,\gamma)$ to denote a *state*. [Here the indices $i$ and $k$ mean that the assortment $S\subseteq\{1,2,\cdots,i\}$ and $|S|=k$.]{} Clearly there are at most $O(NK^5\nu^3/\epsilon^4)$ different types of states. A state $\varsigma_i^k(\mu,\alpha,\beta,\gamma)$ can be either *reachable* or *non-reachable*, as defined below:
Let $q\in[N]$ be the enumerated item with maximal utility parameter and $U=\max\{1,{\widehat}u_{tq}\}$, $\Delta=\epsilon_0 U/K$. A state $\varsigma_i^k(\mu,\alpha,\beta,\gamma)$ is *reachable* if there exists $S\subseteq[N]$ satisfying the following:
1. $S\subseteq\{1,2,\cdots,i\}$ and $|S|=k$;
2. ${\widehat}u_{tj}\leq {\widehat}u_{tq}$ for all $j\in S$;
3. if $i\geq q$ then $q\in S$;
4. $\mu=\sum_{j\in S}\mu_j$, $\alpha=\sum_{j\in S}\alpha_j$, $\beta=\sum_{j\in S}\beta_j$ and $\gamma=\sum_{j\in S}\gamma_j$.
On the other hand, a state $\varsigma_i^k(\mu,\alpha,\beta,\gamma)$ is *non-reachable* if at least one condition above is violated. \[defn:reachable\]
A simple way to find all reachable states is to enumerate all $S\subseteq[N]$, $|S|\leq K$ and verify the three conditions in Definition \[defn:reachable\]. While such a procedure is clearly computationally intractable, in the next section we will present a dynamic programming approach to compute all reachable states in polynomial time. After all reachable states are computed, enumerate over every $q\in[N]$ and reachable $\zeta_N^k(\cdot,\cdot,\cdot,\cdot)$ for $k\in[K]$ and find ${\widehat}S$ that maximizes ${\widehat}{\mathrm{ESTR}}({\widehat}S)+\min\{1,\omega\cdot{\widehat}{\mathrm{CI}}({\widehat}S)\}$. The following corollary establishes the approximation guarantee for ${\widehat}S$, following Lemma \[lem:approx-1d-estr-ci\].
Let ${\widehat}S\subseteq[N]$, $|{\widehat}S|\leq K$ be a subset corresponding to a reachable state $\varsigma_N^k(\cdot,\cdot,\cdot,\cdot)$ for some $k\in[K]$, $q\in[N]$, that maximizes ${\widehat}{\mathrm{ESTR}}({\widehat}S)+\min\{1,\omega\cdot{\widehat}{\mathrm{CI}}({\widehat}S)\}$. Then $${\mathrm{ESTR}}({\widehat}S)+\min\{1,\omega\cdot{\mathrm{CI}}({\widehat}S)\}
\geq \max_{S\subseteq[N],|S|\leq K}\mathrm{ESTR}(S)+\min\{1,\omega\cdot\mathrm{CI}(S)\} - (6\epsilon_0 + \omega(1+\nu)\sqrt{24\epsilon_0}).$$ \[cor:approx-1d-estr-ci\]
Corollary \[cor:approx-1d-estr-ci\] follows easily by plugging in the upper bounds of estimation error in Lemma \[lem:approx-1d-estr-ci\]. By setting $\epsilon_0=\min\{\varepsilon/12, \varepsilon^2/96\omega^2(1+\nu)^2\}$, the algorithm that produces ${\widehat}S$ satisfies $(1,\varepsilon,0)$-approximation as defined in Definition \[defn:approx\].
### A dynamic programming method for computation of reachable states
In this section we describe a *dynamic programming* algorithm to compute reachable states in polynomial time. The dynamic programming algorithm is exact and deterministic, therefore approximation guarantees in Corollary \[cor:approx-1d-estr-ci\] remain valid.
The first step is again to enumerate $q\in[N]$ corresponding to the item in $S$ with the largest utility parameter ${\widehat}u_{tq}$, and calculating the discretization parameter $\Delta=\epsilon\max\{1,{\widehat}u_{tq}\}/K$. Afterwards, reachable states are computed in an iterative manner, from $i=0,1,\cdots$ until $i=N$. The initialization is that $\varsigma_0^0(0,0,0,0)$ is reachable. Once a state $\varsigma_i^k(\mu,\alpha,\beta,\gamma)$ is determined to be reachable, the following two states are *potentially* reachable: $$\varsigma_{i+1}^k(\mu,\alpha,\beta,\gamma) \;\;\;\;\;\text{and}\;\;\;\;\; \varsigma_{i+1}^{k+1}(\mu+\mu_{i+1},\alpha+\alpha_{i+1}, \beta+\beta_{i+1}, \gamma+\gamma_{i+1}).$$
The first future state $\varsigma_{i+1}^k(\mu,\alpha,\beta,\gamma)$ corresponds to the case of $i+1\notin S$. To determine when such a state is reachable, we review the conditions in Definition \[defn:reachable\] and observe that whenever $i+1\neq q$, the decision $i+1\notin S$ is legal because $q$ must belong to $S$ whenever $i\geq q$ [(note that $q$ is the item in $S$ with the largest estimated utility)]{}. The second future state $\varsigma_{i+1}^{k+1}(\mu+\mu_{i+1},\alpha+\alpha_{i+1}, \beta+\beta_{i+1}, \gamma+\gamma_{i+1})$ corresponds to the case of $i+1\in S$. Reviewing again the conditions listed in Definition \[defn:reachable\], such a state is reachable if $k+1\leq K$ (meaning that there is still room to include a new item in $S$) and ${\widehat}u_{t,i+1}\leq {\widehat}u_{tq}$ (meaning that the new item ($i+1$) to be included has an estimated utility smaller than ${\widehat}u_{tq}$). Combining both cases, we arrive at the following updated rule of reachability:
1. If $i+1\neq q$, then $\varsigma_{i+1}^k(\mu,\alpha,\beta,\gamma)$ is reachable;
2. If $k<K$ and ${\widehat}u_{t,i+1}\leq {\widehat}u_{tq}$, then $\varsigma_{i+1}^{k+1}(\mu+\mu_{i+1},\alpha+\alpha_{i+1},\beta+\beta_{i+1},\gamma+\gamma_{i+1})$ is reachable.
Algorithms \[alg:approx-1d\] and \[alg:approx-1d-restrict\] give pseudo-codes for the proposed dynamic programming approach of computing reachable states and an approximate optimizer of ${\widehat}{\mathrm{ESTR}}(S)+\min\{1,\omega\cdot{\widehat}{\mathrm{CI}}(S)\}$.
Initialization: compute $\mu_i,\alpha_i,\beta_i,\gamma_i$ for all $i\in[N]$ as in Eq. (\[eq:muabg\]) Declare $\varsigma_1^0(0,0,0,0)$ as reachable
For all reachable states $\varsigma_N^k(\mu,\alpha,\beta,\gamma)$, trace back the actual assortment $S\subseteq[N]$, $|S|\leq K$ and select the one with the largest ${\widehat}{\mathrm{ESTR}}(S)+\min\{1,\omega\cdot{\widehat}{\mathrm{CI}}(S)\}$ as the output ${\widehat}S$.
Among ${\widehat}{S}_1, \dots, {\widehat}{S}_N$, select the one with the largest ${\widehat}{\mathrm{ESTR}}(S)+\min\{1,\omega\cdot{\widehat}{\mathrm{CI}}(S)\}$ as the output ${\widehat}S$.
Finally, we remark on the time complexity of the proposed algorithm. Because the items $j$ we consider in the assortment satisfy $|{\widehat}u_{ti}| \leq U$, $|r_{ti}| \leq 1$, and $|x_{ti}| \leq \nu$, and all $\mu_i,\alpha_i,\beta_i,\gamma_i$ are integral multiples of $\Delta$, we have (1) $\mu_i$ and $\gamma_i$ take at most $O(K\epsilon_0^{-1})$ possible values; (2) $\alpha_i$ takes at most $(K \nu \epsilon_0^{-1})$ possible values; and (3) $\beta_i$ takes at most $(K \nu^2 \epsilon_0^{-1})$ values. Therefore, the total number of states $\varsigma_i^k(\cdot,\cdot,\cdot,\cdot)$ for fixed $i\in[N]\cup\{0\}$, $k\in[K]$ can be upper bounded by $O(K^8 \nu^3\epsilon_0^{-4})$. The time complexity of Algorithm \[alg:approx-1d\] is thus upper bounded by $O(K^9N \nu^3\epsilon_0^{-4})$. Alternatively, to achieve $(1,\varepsilon,0)$-approximation, one may set $\epsilon_0=\min\{\varepsilon/12,\varepsilon^2/(96(1+\nu)^2\omega^2)\}$ as suggested by Corollary \[cor:approx-1d-estr-ci\], resulting in a time complexity of $O(K^9 N \nu^3 \max\{\varepsilon^{-4}, (1 + \nu)^8 \omega^8\varepsilon^{-8}\})$.
[This dynamic programming based approximation algorithm can be extended to multivariate feature vector with $d>1$. The details are presented in Appendix \[appsec:multivariate\]. ]{}
Greedy swapping heuristics {#subsec:heuristic}
--------------------------
[While the proposed approximation has rigorous approximation guarantees and runs in polynomial time, the large time complexity still prohibits its application to moderately large scale problem instances.]{} In this subsection, we consider a practically efficient greedy swapping heuristic to approximately solve the combinatorial optimization problem in Eq. (\[eq:comb-opt\]).
At a higher level, the heuristic algorithm is a “local search” method similar to the Lloyd’s algorithm for K-means clustering [@lloyd1982least], which continuously tries to improve an assortment solution by considering local swapping/addition/deletions until no further improvements are possible. A pseudo-code description of our heuristic method is given in Algorithm \[alg:greedy\].
\[alg:greedy\]
Initialization: select $S\subseteq[N]$, $|S|= K$ uniformly at random
While the greedy heuristic does not have rigorous guarantees in general, we would like to mention a special case of $\omega=0$, in which Algorithm \[alg:greedy\] does converge to the optimal assortment $S$ maximizing $\mathrm{ESTR}(S)+\min\{1,\omega\cdot\mathrm{CI}(S)\}$ in polynomial time. More specifically, we have the following proposition which is proved in the supplementary material.
If $\omega=0$, then Algorithm \[alg:greedy\] terminates in $O(N^4)$ iterations and produces an output $S$ that maximizes $\mathrm{ESTR}(S)$. \[prop:greedy-omega-zero\]
Numerical studies {#sec:numerical}
=================
In this section, we present numerical results of our proposed MLE-UCB algorithm. We use the greedy swapping heuristics (Algorithm \[alg:greedy\]) as the subroutine to solve the combinatorial optimization problem in Eq. (\[eq:comb-opt\]). We will also study the quality of the solution of the greedy swapping heuristics.
#### Experiment setup.
The unknown model parameter $\theta_0 \in \R^d$ is generated as a uniformly random unit $d$-dimensional vector. The revenue parameters $\{r_{tj}\}$ for $j \in [N]$ are independently and identically generated from the uniform distribution $[0.5, 0.8]$. For the feature vectors $\{v_{tj}\}$, each of them is independently generated as a uniform random vector $v$ such that $\|v\| = 2$ and $v^\top \theta_0 < -0.6$. Here we set an upper bound of $-0.6$ for the inner product so that the utility parameters $u_{tj} = \exp\{v_{tj}^\top \theta_0\}$ are upper bounded by $\exp(-0.6)\approx 0.55$. We set such an upper bound because if the utility parameters are uniformly large, the optimal assortment is likely to pick very few items, leading to degenerated problem instances. In the implementation of our MLE-UCB algorithm, we simply set $T_0 = \lfloor \sqrt{T} \rfloor$ and $\omega = \sqrt{d \ln (T K)}$.
#### The greedy swapping heuristics.
We first numerically evaluate the solution quality of the greedy swapping heuristic algorithm by focusing on the optimization problem in Eq. . We compare the obtained objective values in Eq. to the proposed greedy heuristic and the optimal solution (obtained by brute-force search). Instead of generating purely random instances, we consider more realistic instances generated from a dynamic assortment planning process. In particular, for a given $T$, we generate a dynamic assortment optimization problem with parameters $N = 10$, $K = 4$ and $d = 5$, and run the MLE-UCB algorithm till the $T$-th time period. Now the combinatorial optimization problem in Eq. (\[eq:comb-opt\]) to be solved at the $T$-th time is kept as one testing instance for the greedy swapping algorithm.
For each $T \in \{50, 200, 800\}$, we generate $1000$ such test instances, and compare the solution of the greedy swapping heuristics with the optimal solution obtained by brute-force search in terms of the objective value in Eq. . Table \[table:greedy-swap\] shows the relative differences between the two solutions at several percentiles, and the mean relative differences. We can see that the approximation quality of the greedy swapping algorithm has already been desirable when $T = 50$, and becomes even better as $T$ grows.
----- -------- ----------------------------- -------- -------- ---------- ----------------- --
mean relative difference in
$94$th $96$th $98$th $99$th $99.5$th objective value
50 0 0.0159 0.0293 0.0393 0.0687 0.00207
200 0 0.0001 0.0040 0.0080 0.0123 0.00024
800 0 0 0 0.0014 0.0037 0.00004
----- -------- ----------------------------- -------- -------- ---------- ----------------- --
: relative differences in terms of objective value in Eq. between the greedy swapping algorithm and the optimal solution.[]{data-label="table:greedy-swap"}
#### Performance of the MLE-UCB algorithm.
In Figure \[fig:1-a\] we plot the average regret (i.e. $\mathrm{regret}/T$) of MLE-UCB algorithm with $N = 1000, K = 10, d = 5$ for the first $T = 10000$ time periods. For each experiment (in both Figure \[fig:1-a\] and other figures), we repeat the experiment for 100 times and report the average value. In Figure \[fig:1-b\] we compare our algorithm with the UCB algorithm for multinomial logit bandit (MNL-UCB) from [@Agrawal16MNLBandit] without utilizing the feature information. Since the MNL-UCB algorithm assumes fixed item utilities that do not change over time, in this experiment we randomly generate one feature vector for each of the $N=1000$ items and this feature vector will be fixed for the entire time span. We can observe that our MLE-UCB algorithm performs much better than MNL-UCB, which suggests the importance of taking the advantage of the contextual information.
[0.5]{} {height="6cm"}
[0.5]{} {height="6cm"}
#### Impact of the dimension size $d$.
We study how the dimension of the feature vector impacts the performance of our MLE-UCB algorithm. We fix $N = 1000$ and $K = 10$ and test our algorithm for dimension sizes in $5, 7, 9, 11, \dots, 25$. In Figure \[fig:2\], we report the average regret at times $T \in \{4000, 6000, 8000, 10000\}$. We can see that the average regret increases approximately linearly with $d$. This phenomenon matches the linear dependency on $d$ of the main term of the regret Eq. of the MLE-UCB.
![Average regret of MLE-UCB for various $N$’s. []{data-label="fig:3"}](comp-d.png){width="98.00000%"}
![Average regret of MLE-UCB for various $N$’s. []{data-label="fig:3"}](comp-N.png){width="98.00000%"}
#### Impact of the number of items $N$.
We compare the performance of our MLE-UCB algorithm for the varying number of items $N$. We fix $K = 10$ and $d = 5$, and test MLE-UCB for $N \in \{1000, 2000, 3000, 4000\}$. In Figure \[fig:3\], we report the average regret for the first $T = 10000$ time periods. We observe that the regret of the algorithm is almost not affected by a bigger $N$. This confirms the fact that the regret Eq. of MLE-UCB is totally independent of $N$.
Conclusion and future directions {#sec:conclusion}
================================
In this paper, we study the dynamic assortment planning problem under a contextual MNL model, which incorporates rich feature information into choice modeling. We propose an upper confidence bound (UCB) algorithm based on the local MLE that simultaneously learns the underlying coefficient and makes the decision on the assortment selection. We establish both the upper and lower bounds of the regret. Moreover, we develop an approximation algorithm and a greedy heuristic for solving the key optimization problem in our UCB algorithm.
There are a few possibilities for future work. Technically, there is still a gap of $1/K$ between our upper and lower bounds on regret. Although the cardinality constraint of an assortment $K$ is usually small in practice, it is still a technically interesting question to close this gap. Second, introducing contextual information into choice model is a natural idea for many online applications. This paper explores the standard MNL model, and it would be interesting to extend this work to contextual nested logit and other popular choice models. Finally, it is interesting to incorporate other operational considerations into the model, such as prices or inventory constraints.
Multivariate approximation algorithm {#appsec:multivariate}
====================================
\[sec:multi\] In this appendix we describe an approximation algorithm for the combinatorial optimization problem studied in Sec. \[sec:approx\] for the general multivariate ($d>1$) case. The multivariate case is dealt with by *randomized* reductions to several univariate problem instances.
More specifically, for any $y\in\mathbb R^d$, $\|y\|_2=1$, a univariate problem instance can be constructed by replacing every occurrences of $x_{ti}$ with $x_{ti}^\top y$. The univariate approximation Algorithm \[alg:approx-1d\] is then invoked on $L$ independent univariate problem instances, each corresponding to a $y$ vector sampled uniformly at random from the $d$-dimensional unit sphere. The $L$ output maximizers ${\widehat}S$ of Algorithm \[alg:approx-1d\] are then compared against each other and the one leading to the largest value of $\mathrm{ESTR}({\widehat}S)+\min\{1,\alpha\omega\cdot\mathrm{CI}(R)\}$ is selected, where $\alpha$ is the preset multiplicative approximation parameter. A pseudo-code description is given in Algorithm \[alg:approx-md\].
Generalize $L$ vectors $y^{(1)},\cdots,y^{(L)}\in\mathbb R^d$ independently and uniformly from the unit sphere
Output ${\widehat}S^{(\ell)}$ that maximizes $\mathrm{ESTR}({\widehat}S^{(\ell)})+\min\{1,\alpha \omega\cdot\mathrm{CI}({\widehat}S^{(\ell)})\}$.
Approximation guarantees
------------------------
The approximation performance of Algorithm \[alg:approx-md\] can be analyzed based on the following observation: if $y$ is close to $y^*$, the leading eigenvector of $$\frac{\sum_{j\in S^*}{\widehat}u_{tj}x_{tj}x_{tj}^\top}{1+\sum_{j\in S^*}{\widehat}u_{tj}} - \left(\frac{\sum_{j\in S^*}{\widehat}u_{tj}x_{tj}}{1+\sum_{j\in S^*}{\widehat}u_{tj}}\right)\left(\frac{\sum_{j\in S^*}{\widehat}u_{tj}x_{tj}}{1+\sum_{j\in S^*}{\widehat}u_{tj}}\right)^\top,$$ where $S^*$ is the exact maximizer of Eq. (\[eq:comb-opt\]), then the reduction to a univariate problem instance $x_{tj}\mapsto x_{tj}^\top y$ does not lose much accuracy. More specifically, we have the following lemma:
Suppose there exists $\ell\in[L]$ such that $\langle y^{(\ell)},y^*\rangle \geq 1/\alpha$ for some $\alpha\geq 1$ in Algorithm \[alg:approx-md\], then $\mathrm{ESTR}({\widehat}S^{(\ell)}) + \min\{1,\alpha\omega\cdot\mathrm{CI}({\widehat}S^{(\ell)})\}+\varepsilon \geq \mathrm{ESTR}(S^*)+\min\{1,\omega\cdot\mathrm{CI}(S^*)\}$, where $\varepsilon>0$ is the approximation parameter of the univariate problem instances. \[lem:approx-md\]
Lemma \[lem:approx-md\] is proved in the supplementary material using elementary linear algebra. At a higher level, Lemma \[lem:approx-md\] shows that when the sampled vector $y^{(\ell)}$ is close to the underlying leading eigenvector $y^*$ (in the sense that the inner product between $y^{(\ell)}$ and $y^*$ is large), the produced subset ${\widehat}S^{(\ell)}$ will have good performance in maximizing the objective function $\mathrm{ESTR}(S)+\min\{1,\omega\cdot\mathrm{CI}(S)\}$.
The following proposition additionally gives the proximity between a random $y$ and $y^*$.
Assume that $d \geq 2$. Let $y^*\in\mathbb R^d$, $\|y^*\|_2=1$ be fixed and $y$ be sampled uniformly at random from the unit $d$-dimensional sphere. Then $$\Pr[\langle y,y^*\rangle \geq 1/\sqrt{d}] = \Omega(1) \;\;\;\;\;\text{and}\;\;\;\;\; \Pr[\langle y,y^*\rangle \geq 1/2] = \exp\{-O(d)\}.$$ \[prop:approx-init\]
Proposition \[prop:approx-init\] is again proved in the supplementary material, using isotropy of $y$ and classical concentration inequalities.
Combining Lemma \[lem:approx-md\] and Proposition \[prop:approx-init\] we can give some recommendations on the choice of $L$ in Algorithm \[alg:approx-md\], which is the number of random $y^{(\ell)}$ vectors sampled. First, if $L\asymp \log(1/\delta)$ initializations are taken, then with probability $1-\delta$ Lemma \[lem:approx-md\] is satisfied with $\alpha=\sqrt{d}$, yielding a $(\sqrt{d}, \epsilon, \delta)$-approximation. Additionally, if $L\asymp e^{O(d)}\log(1/\delta)$ initializations are taken, then with probability $1-\delta$ Lemma \[lem:approx-md\] is satisfied with $\alpha=2$, yielding a $(2,\epsilon,\delta)$-approximation.
Time complexity analysis
------------------------
To achieve a $(\sqrt{d},\varepsilon,\delta)$-approximation $L$ is set to $L\asymp \log(1/\delta)$ and the overall running time of Algorithm \[alg:approx-md\] $O(K^9 N \nu^3 \max\{\varepsilon^{-4}, (1 + \nu)^8 \omega^8\varepsilon^{-8}\} \log \delta^{-1})$. To achieve a $(2,\varepsilon,\delta)$-approximation $L$ is set to $L\asymp e^{O(d)}\log(1/\delta)$ and the overall running time of Algorithm \[alg:approx-md\] is $e^{O(d)} K^9 N \nu^3 \max\{\varepsilon^{-4}, (1 + \nu)^8 \omega^8\varepsilon^{-8}\}$.
Now we use Algorithm \[alg:approx-md\] to solve the combinatorial optimization problem in Step \[alg:step6\] of Algorithm \[alg:mle-ucb\] and examine the cumulative regret. If we let Algorithm \[alg:approx-md\] achieve to $(\sqrt{d},\varepsilon,\delta)$-approximation guarantee with $\varepsilon = T^{-1/2}$ and $\delta = T^{-2}$, the computational time complexity at each time slot will be ${\widetilde}{O}(K^9 N \nu^3 (1+\nu)^8 d^4 T^4)$,[^5] and the cumulative regret will be upper bounded by $O(\sqrt{d}) \cdot \mathrm{Regret}^*$. If we let Algorithm \[alg:approx-md\] to achieve $(1/2,\varepsilon,\delta)$-approximation guarantee with $\varepsilon = T^{-1/2}$ and $\delta = T^{-2}$, the computational time complexity at each time slot will be $e^{O(d)} \cdot {\widetilde}{O}(K^9 N \nu^3 (1+\nu)^8 d^4 T^4)$, and the cumulative regret will be upper bounded by $O(1) \cdot \mathrm{Regret}^*$.
#### Acknowledgement
: [The authors would like to thank Vineet Goyal for helpful discussions, and Zikai Xiong for helping with the numerical studies. ]{}
This supplementary material provides detailed proofs for technical lemmas whose proofs are omitted in the main text.
Proofs of technical lemmas for Theorem \[thm:mle-ucb\] (upper bound)
====================================================================
Proof of Lemma \[lem:pilot\]
----------------------------
With probability $1-\delta$ it holds that $$\|\theta^*-\theta_0\|_2 \leq \frac{2}{\kappa}\sqrt{\frac{d+\log(1/\delta)}{\lambda_{\min}(V)}}
\;\;\;\;\text{where}\;\;\kappa = \frac{1}{2e(1+\rho)}\;\;\text{and}\;\; V = \sum_{t=1}^{T_0}v_{t,i_t}v_{t,i_t}^\top.$$
Because the noise in a logistic regression model is clearly centered and sub-Gaussian with parameter at most $1/4$, it only remains to check [@li2017provably Assumption 1], that $\inf_{\|x\|_2\leq 1, \|\theta-\theta_0\|_2\leq 1}\sigma'(x^\top\theta) \geq \kappa=2e(1+\rho)$ where $\sigma(x)=1/(1+e^{-x})$ is the sigmoid link function. Because $\sigma'(x)=\sigma(x)(1-\sigma(x))$, we have $\sigma'(x^\top\theta) = \wp_\theta(1-\wp_\theta) \geq 0.5\wp_\theta$ where $\wp_\theta = \min\{p_{\theta}(1), 1-p_{\theta}(1)\}$ and $p_\theta(1)=\sigma(x^\top\theta)=1/(1+\exp\{-x^\top\theta\})$. By (A2), we know that $\wp_{\theta_0}\geq 1/(1+\rho)$. Subsequently, for any $\|x\|_2\leq 1$ and $\|\theta-\theta_0\|_2\leq 1$, we have $$\wp_\theta = \frac{1}{1+\exp\{-x^\top\theta\}} = \frac{1}{1+\exp\{-x^\top(\theta-\theta_0)\} \exp\{-x^\top\theta_0\}} \geq \frac{1}{e}\frac{1}{1+\exp\{x^\top\theta_0\}} \geq \frac{1}{e(1+\rho)}.$$ Lemma \[lem:pilot\] is then an immediate consequence of [@li2017provably Eq. (18)].
Proof of Corollary \[cor:tau\]
------------------------------
There exists a universal constant $C_0>0$ such that for arbitrary $\tau\in(0,1/2]$, if $T_0 \geq C_0\max\{\nu^2d\log T/\lambda_0^2, \rho^2 (d+\log T)/(\tau^2\lambda_0)\}$ then with probability $1-O(T^{-1})$, $\|\theta^*-\theta_0\|_2\leq \tau$.
Denote $\Lambda := \mathbb E_\mu xx^\top$ and ${\widehat}\Lambda := V/T_0 = \frac{1}{T_0}\sum_{t=1}^{T_0} x_{t,i_t}x_{t,i_t}^\top$. Clearly $\mathbb E{\widehat}\Lambda = \Lambda$. In addition, because $\|v_{tj}\|_2\leq\nu$ almost surely, $v_{tj}$ are sub-Gaussian random variables with parameter $\nu^2$. By standard concentration inequalities (see, e.g., [@vershynin2012close Proposition 2.1]), we have with probability $1-O(T^{-2})$ that $\|{\widehat}\Lambda-\Lambda\|_\op \lesssim \nu\sqrt\frac{d\log T}{T_0}$. Hence, if $T_0\geq C_0\nu^2 d\log T/\lambda_0^2$ for some sufficiently large universal constant $C_0$, we have $\|{\widehat}\Lambda-\Lambda\|_\op \leq 0.5\lambda_0 = \lambda_{\min}(\Lambda)$ and therefore $\lambda_{\min}(V) = T_0\lambda_{\min}({\widehat}\Lambda) \geq 0.5T_0\lambda_0$. The corollary then immediately follows Lemma \[lem:pilot\].
Proof of Lemma \[lem:mle\]
--------------------------
Suppose $\tau\leq 1/\sqrt{8\rho \nu^2K^2}$. Then there exists a universal constant $C>0$ such that with probability $1-O(T^{-1})$ the following holds uniformly over all $t=T_0,\cdots,T-1$: $$({\widehat}\theta_t-\theta_0)^\top I_t(\theta_0)({\widehat}\theta_t-\theta_0) \leq C\cdot d\log(\rho\nu TK).$$
For any $\theta\in\mathbb R^d$ define $$f_{t'}(\theta) := \mathbb E_{\theta_0,t'}\left[\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}\right] = \sum_{j\in S_{t'}\cup\{0\}} p_{\theta_0,t'}(j|S_{t'})\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}.$$
By simple algebra calculations, the first and second order derivatives of $f_{t'}$ with respect to $\theta$ can be computed as $$\begin{aligned}
\nabla_\theta f_{t'}(\theta) &= \mathbb E_{\theta_0,t'}[v_{t'j}] - \mathbb E_{\theta,t'}[v_{t'j}];\label{eq:grad}\\
\nabla_\theta^2 f_{t'}(\theta) &= -\mathbb E_{\theta_0,t'}[v_{t'j}v_{t'j}^\top] + \{\mathbb E_{\theta_0,t'}v_{t'j}\}\{\mathbb E_{\theta,t'}v_{t'j}\}^\top\nonumber\\
&\;\;\;\; + \{\mathbb E_{\theta,t'}v_{t'j}\}\{\mathbb E_{\theta_0,t'}v_{t'j}\}^\top - \{\mathbb E_{\theta,t'}v_{t'j}\}\{\mathbb E_{\theta,t'}v_{t'j}\}^\top.\label{eq:hessian}
\end{aligned}$$ In the rest of the section we drop the subscript in $\nabla_\theta$, $\nabla^2_\theta$, and the $\nabla$, $\nabla^2$ notations should always be understood as with respect to $\theta$.
Define $F_t(\theta) := \sum_{t'=1}^t f_{t'}(\theta)$. It is easy to verify that $-F_t(\theta)$ is the Kullback-Leibler divergence between the conditional distribution of $(i_1,\cdots,i_t)$ parameterized by $\theta$ and $\theta_0$, respectively. Therefore, $F_t(\theta)$ is always non-positive. Note also that $F_t(\theta_0) = 0$, $\nabla F_t(\theta_0)=0$, $\nabla^2 f_{t'}(\theta) = -M_{t'}(\theta)$ and $\nabla^2 F_t(\theta) \equiv -I_t(\theta)$. By Taylor expansion with Lagrangian remainder, there exists ${\overline}\theta_t=\alpha\theta_0+(1-\alpha){\widehat}\theta_t$ for some $\alpha\in(0,1)$ such that $$F_t({\widehat}\theta_t) = -\frac{1}{2}({\widehat}\theta_t-\theta_0)^\top I_t({\overline}\theta_t) ({\widehat}\theta_t-\theta_0).
\label{eq:taylor_F}$$
Our next lemma shows that, if ${\overline}\theta_t$ is close to $\theta_0$ (guaranteed by the constraint that $\|{\widehat}\theta_t-\theta^*\|_2\leq\tau$), then $I_t({\overline}\theta_t)$ can be *spectrally* lower bounded by $I_t(\theta_0)$. It is proved in the supplementary material.
Suppose $\tau\leq 1/\sqrt{8\rho \nu^2K^2}$. Then $I_t({\overline}\theta_t)\succeq \frac{1}{2}I_t(\theta_0)$ for all $t$. \[lem:Itheta-Itheta0\]
As a corollary of Lemma \[lem:Itheta-Itheta0\], we have $$F_t({\widehat}\theta_t) \leq -\frac{1}{4}({\widehat}\theta_t-\theta_0)^\top I_t(\theta_0)({\widehat}\theta_t-\theta_0).
\label{eq:taylor-F-population}$$
On the other hand, consider the “empirical” version ${\widehat}F_t(\theta) := \sum_{t'=1}^t{\widehat}f_{t'}(\theta)$, where $${\widehat}f_{t'}(\theta) := \log\frac{p_{\theta,t'}(i_{t'}|S_{t'})}{p_{\theta_0,t'}(i_{t'}|S_{t'})}.$$
It is easy to verify that ${\widehat}F_t(\theta_0)=0$ remains true; in addition, for any *fixed* $\theta\in\mathbb R^d$, $\{{\widehat}F_t(\theta)\}_t$ forms a *martingale* [^6] and satisfies $\mathbb E{\widehat}F_t(\theta)=F_t(\theta)$ for all $t$. This leads to our following lemma, which upper bounds the *uniform* convergence of ${\widehat}F_t(\theta)$ towards $F_t(\theta)$ for all $\|\theta-\theta_0\|\leq 2\tau$.
Suppose $\tau \leq 1/\sqrt{8\rho^2 \nu^2K^2}$. Then there exists a universal constant $C>0$ such that with probability $1-O(T^{-1})$ the following holds uniformly for all $t\in\{T_0+1,\cdots,T\}$ and $\|\theta-\theta_0\|_2\leq 2\tau$: $$\big|{\widehat}F_t(\theta)-F_t(\theta)\big| \leq C\left[d\log(\rho\nu TK) + \sqrt{|F_t(\theta)| d\log(\rho\nu TK)}\right].
\label{eq:hatF-uniform}$$ \[lem:hatF-uniform\]
Lemma \[lem:hatF-uniform\] can be proved by using a standard $\varepsilon$-net argument. Since the complete proof is quite involved, we defer it to the supplementary material.
We are now ready to prove Lemma \[lem:mle\]. By Eq. (\[eq:hatF-uniform\]) and the fact that ${\widehat}F_t({\widehat}\theta_t)\leq 0\leq F_t({\widehat}\theta_t)$, we have $$|F_t({\widehat}\theta_t)| \leq |{\widehat}F_t({\widehat}\theta_t)- F_t({\widehat}\theta_t)| \lesssim d\log(\rho\nu TK) + \sqrt{|F_t({\widehat}\theta_t)| d\log(\rho\nu TK)}.$$ Subsequently, $$|F_t({\widehat}\theta_t)| \lesssim d\log(\rho\nu NT).$$ In addition, because $F_t({\widehat}\theta_t)\leq 0$, by Eq. (\[eq:taylor-F-population\]) we have $$-\frac{1}{2}({\widehat}\theta_t-\theta_0)^\top I_t(\theta_0) ({\widehat}\theta_t-\theta_0) \geq F_t({\widehat}\theta_t) \geq d\log(\rho\nu TK).$$ Lemma \[lem:mle\] is thus proved.
### Proof of Lemma \[lem:Itheta-Itheta0\] {#proof-of-lemma-lemitheta-itheta0 .unnumbered}
Suppose $\tau\leq 1/\sqrt{8\rho \nu^2K^2}$. Then $I_t({\overline}\theta_t)\succeq \frac{1}{2}I_t(\theta_0)$ for all $t$.
Because ${\widehat}\theta_t$ is a feasible solution of the local MLE, we know $\|{\widehat}\theta_t-\theta^*\|_2\leq\tau$. Also by Corollary \[cor:tau\] we know that $\|\theta^*-\theta_0\|_2\leq\tau$ with high probability. By triangle inequality and the definition of ${\overline}\theta_t$ we have that $\|{\overline}\theta_t-\theta_0\|_2 \leq 2\tau$.
To prove $I_t({\overline}\theta_t)\succeq \frac{1}{2}I_t(\theta_0)$ we only need to show that $M_{t'}({\overline}\theta_t)-M_{t'}(\theta_0)\preceq \frac{1}{2}M_{t'}(\theta_0)$ for all $1\leq t'\leq t$. This reduces to proving $$\{\mathbb E_{{\overline}\theta_t,t'}v_{t'j}-\mathbb E_{\theta_0,t'}v_{t'j}\}\{\mathbb E_{{\overline}\theta_t,t'}v_{t'j}-\mathbb E_{\theta_0,t'}v_{t'j}\}^\top \preceq
\frac{1}{2} \mathbb E_{\theta_0,t'}\left[(v_{t'j}-\mathbb E_{\theta_0,t'}v_{t'j})(v_{t'j}-\mathbb E_{\theta_0,t'}v_{t'j})^\top\right].
\label{eq:preceq-1}$$
Fix arbitrary $S_{t'}\subseteq[N]$, $|S_{t'}|=J\leq K$ and for convenience denote $x_1,\cdots,x_J\in\mathbb R^d$ as the feature vectors of items in $S_{t'}$ (i.e., $\{v_{t'j}\}_{j\in S_{t'}}$). Let also $p_{\theta_0}(j)$ and $p_{{\overline}\theta_t}(j)$ be the probability of choosing action $j\in[J]$ corresponding to $x_j$ parameterized by $\theta_0$ or ${\overline}\theta_t$. Define ${\overline}x := \sum_{j=1}^Jp_{\theta_0}(j)x_j$, $w_j := x_j - {\overline}x$ and $\delta_j := p_{{\overline}\theta_t}(j)-p_{\theta_0}(j)$. Recall also that $x_0=0$ and $w_0=-{\overline}x$. Eq. (\[eq:preceq-1\]) is then equivalent to $$\left\{\sum_{j=0}^J\delta_j w_j\right\}\left\{\sum_{j=0}^J\delta_j w_j\right\}^\top \preceq \frac{1}{2}\sum_{j=0}^J p_{\theta_0}(j)w_jw_j^\top.
\label{eq:preceq-2}$$
Let $L={\mathrm{span}}\{w_j\}_{j=0}^J$ and $H\in\mathbb R^{L\times d}$ be a whitening matrix such that $H(\sum_j p_{\theta_0}(j)w_jw_j^\top)H^\top = I_{L\times L}$, where $I_{L\times L}$ is the identity matrix of size $L$. Denote ${\widetilde}w_j := Hw_j$. We then have $\sum_{j=0}^J p_{\theta_0}(j) {\widetilde}w_j{\widetilde}w_j^\top = I_{L\times L}$. Eq. (\[eq:preceq-2\]) is then equivalent to $$\left\|\sum_{j=0}^J\delta_j {\widetilde}w_j\right\|_2^2 \leq \frac{1}{2}.$$
On the other hand, by (A2) we know that $p_{\theta_0}(j)\geq 1/\rho K$ for all $j$ and therefore $\|{\widetilde}w_j\|_2 \leq \sqrt{\rho K}$ for all $j$. Subsequently, we have $$\left\|\sum_{j=0}^J\delta_j {\widetilde}w_j\right\|_2^2 \leq\left(\max_j |\delta_j|\cdot \sum_{j=0}^J\|{\widetilde}w_j\|_2\right)^2 \leq \max_j|\delta_j|^2\cdot \rho K^2.$$
Recall that $\delta_i = p_{{\overline}\theta_t}(i)-p_{\theta_0}(i)$ where $p_\theta(i) = \exp\{x_i^\top\theta\}/(1+\sum_{j\in S_{t'}}\exp\{x_i^\top\theta\})$. Simple algebra yields that $\nabla_\theta p_\theta(i) = p_{\theta}(i)[x_i - \mathbb E_\theta x_j]$, where $\mathbb E_\theta x_j = \sum_{j\in S_{t'}}p_\theta(j)x_j$. Using the mean-value theorem, there exists ${\widetilde}\theta_t={\widetilde}\alpha{\overline}\theta_t+(1-{\widetilde}\alpha)\theta_0$ for some ${\widetilde}\alpha\in(0,1)$ such that $$\delta_i = \langle\nabla_\theta p_{{\widetilde}\theta_t}(i), {\widehat}\theta_t-\theta_0\rangle = p_{{\widetilde}\theta_t}(i)\langle x_i-\mathbb E_{{\widetilde}\theta_t}x_j, {\overline}\theta_t-\theta_0\rangle.
\label{eq:deltai}$$ Because $\|x_{ti}\|_2\leq \nu$ almost surely for all $t\in[T]$ and $i\in[N]$, we have $$\max_j |\delta_j|^2\cdot \rho K^2 \leq 4\cdot\max_i\|x_i\|_2^2\cdot \|{\overline}\theta_t-\theta_0\|_2^2\cdot \rho K^2 \leq 4\rho \nu^2 K^2\cdot \tau^2.
\label{eq:xi-all}$$ The lemma is then proved by plugging in the condition on $\tau$.
### Proof of Lemma \[lem:hatF-uniform\] {#proof-of-lemma-lemhatf-uniform .unnumbered}
Suppose $\tau \leq 1/\sqrt{8d\rho^2 \nu^2K^2}$. Then there exists a universal constant $C>0$ such that with probability $1-O(T^{-1})$ the following holds uniformly for all $t\in\{T_0+1,\cdots,T\}$ and $\|\theta-\theta_0\|_2\leq 2\tau$: $$\big|{\widehat}F_t(\theta)-F_t(\theta)\big| \leq C\left[d\log(\rho\nu TK) + \sqrt{|F_t(\theta)| d\log(\rho\nu TK)}\right].
$$
We first consider a *fixed* $\theta\in\mathbb R^d$, $\|\theta-\theta_0\|_2\leq 2\tau$. Define $$\mathcal M := \max_{t'\leq t}|{\widehat}f_{t'}(\theta)| \;\;\;\;\;\text{and}\;\;\;\;\; \mathcal V^2 := \sum_{t'=1}^t \mathbb E_{j\sim\theta_0,t'}\left|\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}\right|^2.$$ Using an Azuma-Bernstein type inequality (see, for example, [@fan2015exponential Theorem A], [@freedman1975tail Theorem (1.6)]), we have $$\big|{\widehat}F_t(\theta)-F_t(\theta)\big| \lesssim \mathcal M\log(1/\delta) + \sqrt{\mathcal V^2\log(1/\delta)} \;\;\;\;\;\;\text{with probability $1-\delta$}.$$
The following lemma upper bounds $\mathcal M$ and $\mathcal V^2$ using $F_t(\theta)$ and the fact that $\theta$ is close to $\theta_0$. It will be proved right after this proof.
If $\tau\leq 1/\sqrt{8\rho^2 \nu^2K^2}$ then $\mathcal M\leq 1$ and $\mathcal V^2\leq 8 |F_t(\theta)|$. \[lem:MV\]
Suppose $\tau$ satisfies the condition in Lemma \[lem:MV\]. Then for any $\|\theta-\theta_0\|_2\leq 2\tau$, $$\big|{\widehat}F_t(\theta)-F_t(\theta)\big| \lesssim \log(1/\delta) + \sqrt{|F_t(\theta)|\log(1/\delta)} \;\;\;\;\;\;\text{with probability $1-\delta$}.$$ \[cor:concentration-Ft\]
Our next step is to construct an $\epsilon$-net over $\{\theta\in\mathbb R^d: \|\theta-\theta_0\|_2\leq 2\tau\}$ and apply union bound on the constructed $\epsilon$-net. This together with a deterministic perturbation argument delivers *uniform* concentration of ${\widehat}F_t(\theta)$ towards $F_t(\theta)$.
For any $\epsilon>0$, let $\mathcal H(\epsilon)$ be a finite covering of $\{\theta\in\mathbb R^d:\|\theta-\theta_0\|_2\leq 2\tau\}$ in $\|\cdot\|_2$ up to precision $\epsilon$. That is, $\sup_{\|\theta-\theta_0\|_2\leq 2\tau} \min_{\theta'\in\mathcal H(\epsilon)} \|\theta-\theta'\|_2\leq\epsilon$. By standard covering number arguments (e.g., [@van2000empirical]), such a finite covering set $\mathcal H(\epsilon)$ exists whose size can be upper bounded by $\log|\mathcal H(\epsilon)|\lesssim d\log(\tau/\epsilon)$. Subsequently, by Corollary \[cor:concentration-Ft\] and the union bound, we have with probability $1-O(T^{-1})$ that $$\big|{\widehat}F_t(\theta)-F_t(\theta)\big|\lesssim d\log(T/\epsilon) + \sqrt{{|F_t(\theta)|d\log(T/\epsilon)}} \;\;\;\;\;\;\forall T_0<t\leq T, \theta\in\mathcal H(\epsilon).
\label{eq:net}$$
On the other hand, with probability $1-O(T^{-1})$ such that Eq. (\[eq:deltai\]) holds, we have for arbitrary $\|\theta-\theta'\|_2\leq\epsilon$ that $$\begin{aligned}
\big|{\widehat}F_t(\theta)-{\widehat}F_t(\theta')\big|
& \leq t\cdot\sup_{t'\leq t,j\in S_{t'}\cup\{0\}}\bigg|\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta',t'}(j|S_{t'})}\bigg|\nonumber\\
&\leq t\cdot\sup_{t'\leq t,j\in S_{t'}\cup\{0\}}\frac{|p_{\theta,t'}(j|S_{t'})-p_{\theta',t'}(j|S_{t'})|}{p_{\theta',t'}(j|S_{t'})}\label{eq:cover-intermediate-1}\\
&\leq 2\rho TK\cdot \sup_{t'\leq t, j\in S_{t'}\cup\{0\}} \big|p_{\theta,t'}(j|S_{t'})-p_{\theta',t'}(j|S_{t'})\big|\label{eq:cover-intermediate-2}\\
&\leq 2\rho TK\cdot \sup_{t'\leq t,j\in[N]}4\|v_{t'j}\|_2^2\cdot \|\theta-\theta'\|_2\nonumber\\
&\lesssim \rho TK\cdot \nu^2 \cdot \epsilon.\label{eq:cover-intermediate-3}
\end{aligned}$$ Here Eq. (\[eq:cover-intermediate-1\]) holds because $\log(1+x)\leq x$; Eq. (\[eq:cover-intermediate-2\]) holds because $p_{\theta',t'}(j|S_{t'}) \geq p_{\theta_0,t'}(j|s_{t'}) - |p_{\theta',t'}(j|S_{t'})-p_{\theta_0,t'}(j|S_{t'})|
\geq 1/2\rho K$ thanks to (A2) and Eq. (\[eq:xi-all\]).
Combining Eqs. (\[eq:net\],\[eq:cover-intermediate-3\]) and setting $\epsilon\asymp 1/(\rho\nu^2 TK)$ we have with probability $1-O(T^{-1})$ that $$\big|{\widehat}F_t(\theta)-F_t(\theta)\big|\lesssim d\log(\rho\nu TK) + \sqrt{|F_t(\theta)|d\log(\rho\nu TK)}\;\;\;\;\forall T_0<t\leq T, \|\theta-\theta_0\|_2\leq2\tau,$$ which is to be demonstrated in Lemma \[lem:hatF-uniform\].
### Proof of Lemma \[lem:MV\] {#proof-of-lemma-lemmv .unnumbered}
If $\tau\leq 1/\sqrt{8\rho^2 \nu^2K^2}$ then $\mathcal M\leq 1$ and $\mathcal V^2\leq 8 |F_t(\theta)|$.
We first derive an upper bound for $M$. By (A2), we know that $p_{\theta_0,t'}(j|S_{t'})\geq 1/\rho K$ for all $j$. Also, Eqs. (\[eq:deltai\],\[eq:xi-all\]) shows that $|p_{\theta,t'}(j|S_{t'})-p_{\theta_0,t'}(j|S_{t'})| \leq 4\nu^2\cdot \tau^2$. If $\tau^2\leq 1/\sqrt{8\rho \nu^2K}$ we have $|p_{\theta,t'}(j|S_{t'})-p_{\theta_0,t'}(j|S_{t'})| \leq 0.5p_{\theta_0,t'}(j|S_{t'})$ and therefore $|{\widehat}f_{t'}(\theta)|\leq \log^22\leq 1$.
We next give upper bounds on $\mathcal V^2$. Fix arbitrary $t'$, and for notational simplicity let $p_j=p_{\theta_0,t'}(j|S_{t'})$ and $q_j=p_{\theta,t'}(j|S_{t'})$. Because $\log(1+x)\leq x$ for all $x\in(-1,\infty)$, we have $$\mathbb E_{j\sim\theta_0,t'}\left|\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}\right|^2 = \sum_{j\in S_{t'}\cup\{0\}} p_j\log^2\left(1+\frac{q_j-p_j}{p_j}\right) \leq
\sum_{j\in S_{t'}\cup\{0\}} \frac{(q_j-p_j)^2}{p_j}.$$ On the other hand, by Taylor expansion we know that for any $x\in(-1,\infty)$, there exists ${\overline}x\in(0,x)$ such that $\log(1+x)=x-x^2/2(1+{\overline}x)^2$. Subsequently, $$\begin{aligned}
-f_{t'}(\theta)
&= -\mathbb E_{j\sim\theta_0,t'}\left[\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}\right] = -\sum_{j\in S_{t'}\cup \{0\}}p_j\log\left(1+\frac{q_j-p_j}{p_j}\right)\\
&= -\sum_{j\in S_{t'}\cup\{0\}} p_j\left(\frac{q_j-p_j}{p_j} -\frac{1}{2(1+{\overline}\delta_j)^2} \frac{|q_j-p_j|^2}{p_j^2} \right)\\
&\geq \frac{1}{2(1+\max_j{|p_j-q_j|/p_j})^2}\cdot \sum_{j\in S_{t'}\cup\{0\}}\frac{(q_j-p_j)^2}{p_j}.
\end{aligned}$$ Here ${\overline}\delta_j \in(0, (q_j-p_j)/p_j)$ and the last inequality holds because $\sum_j p_j=\sum_jq_j=1$.
By Eqs. (\[eq:deltai\]) and (\[eq:xi-all\]), we have that $|q_j-p_j|^2\leq 4 \nu^2\cdot\tau^2$. In addition, (A2) implies that $p_j\geq 1/\rho K$ for all $j$. Therefore, if $\tau\leq1/\sqrt{4\rho^2\nu^2K^2}$ we have $|p_j-q_j|/p_j\leq 1$ for all $j$ and hence $$\mathbb E_{j\sim\theta_0,t'}\left|\log\frac{p_{\theta,t'}(j|S_{t'})}{p_{\theta_0,t'}(j|S_{t'})}\right|^2\leq
\sum_{j\in S_{t'}\cup\{0\}} \frac{(q_j-p_j)^2}{p_j} \leq 8|f_{t'}(\theta)|.$$ Summing over all $t'=1,\cdots,t$ and noting that $f_{t'}(\theta)$ is always non-positive, we complete the proof of Lemma \[lem:MV\].
Proof of Lemma \[lem:ucb\]
--------------------------
Suppose $\tau$ satisfies the condition in Lemma \[lem:mle\]. With probability $1-O(T^{-1})$ the following holds uniformly for all $t>T_0$ and $S\subseteq[N]$, $|S|\leq K$ such that
1. ${\overline}R_t(S)\geq R_t(S)$;
2. $|{\overline}R_t(S)-R_t(S)| \lesssim \min\{1,\omega\sqrt{\|I_{t-1}^{-1/2}(\theta_0) M_t(\theta_0|S) I_{t-1}^{-1/2}(\theta_0)\|_\op}\}$.
Without explicit clarification, all statements are conditioned on the success event in Lemma \[lem:mle\], which occurs with probability $1-O(T^{-1})$ if $\tau$ is sufficiently large and satisfies the condition in Lemma \[lem:mle\].
We present below a key technical lemma in the proof of Lemma \[lem:ucb\], which is an upper bound on the absolute value difference between $R_t(S) := \mathbb E_{\theta_0,t}[r_{tj}|S]$ and ${\widehat}R_t(S) := \mathbb E_{{\widehat}\theta_{t-1},t}[r_{tj}|S]$ using $I_{t-1}(\theta_0)$ and $M_t(\theta_0|S)$, where $I_{t-1}(\theta)=\sum_{t'=1}^{t-1}M_{t'}(\theta)$ and $M_{t'}(\theta) =
\mathbb E_{\theta_0,t'}[v_{t'j}v_{t'j}^\top] - \{\mathbb E_{\theta_0,t'}v_{t'j}\}\{\mathbb E_{\theta,t'}v_{t'j}\}^\top - \{\mathbb E_{\theta,t'}v_{t'j}\} \{\mathbb E_{\theta_0,t'}v_{t'j}\}^\top + \{\mathbb E_{\theta,t'}v_{t'j}\}\{\mathbb E_{\theta,t'}v_{t'j}\}^\top$. This key lemma can be regarded as a finite sample version of the celebrated *Delta’s method* (e.g., [@van1998asymptotic]) used widely in classical statistics to estimate and/or infer a functional of unknown quantities.
For all $t>T_0$ and $S\subseteq[N]$, $|S|\leq K$, it holds that $|{\widehat}R_t(S)-R_t(S)| \lesssim\sqrt{d\log(\rho\nu TK)}\cdot \sqrt{\|I_{t-1}^{-1/2}(\theta_0) M_t(\theta_0|S) I_{t-1}^{-1/2}(\theta_0)\|_\op}$, where in $\lesssim$ notation we only hide numerical constants. \[lem:delta-method\]
Below we state our proof of Lemma \[lem:delta-method\], while deferring the proof of some detailed technical lemmas to the supplementary material. Fix $S\subseteq[N]$. We use $\mathfrak R_t(\theta) = \mathbb E_{\theta,t}[r_{tj}] = [\sum_{j\in S}r_{tj}\exp\{v_{tj}^\top\theta\}]/[1+\sum_{j\in S}\exp\{v_{tj}^\top\theta\}]$ to denote the expected revenue of assortment $S$ at time $t$, evaluated using a specific model $\theta\in\mathbb R$. Then $$\begin{aligned}
\nabla_\theta\mathfrak R_t(\theta)
&= \frac{\sum_{j\in S}r_{tj}\exp\{v_{tj}^\top\theta\}(1+\sum_{j\in S}\exp\{v_{tj}^\top\theta\})^2 - (\sum_{j\in S}r_{tj}\exp\{v_{tj}^\top\theta\})(\sum_{j\in S}\exp\{v_{tj}^\top\theta\})}{(1+\sum_{j\in S}\exp\{v_{tj}^\top\theta\})^2}\nonumber\\
&= \mathbb E_{\theta,t}[r_{tj}v_{tj}] - \{\mathbb E_{\theta,t}r_{tj}\}\{\mathbb E_{\theta,t}v_{tj}\}.
\end{aligned}$$
By the mean value theorem, there exists ${\widetilde}\theta_{t-1}=\theta_0+\xi({\widehat}\theta_{t-1}-\theta_0)$ for some $\xi\in(0,1)$ such that $$\begin{aligned}
\big|{\widehat}R_t(S)-R_t(S)\big|
&= \big|\mathfrak R_t({\widehat}\theta_{t-1})-\mathfrak R_t(\theta_0)\big|
= \big|\langle \nabla\mathfrak R_t({\widetilde}\theta_{t-1}), {\widehat}\theta_{t-1}-\theta_0\rangle\big|\nonumber\\
&= \sqrt{({\widehat}\theta_{t-1}-\theta_0)^\top[\nabla\mathfrak R_t({\widetilde}\theta_{t-1})\nabla\mathfrak R_t({\widetilde}\theta_{t-1})^\top)]({\widehat}\theta_{t-1}-\theta_0)}.
\label{eq:delta-intermediate-1}
\end{aligned}$$
Recall that $\nabla\mathfrak R_t({\widetilde}\theta_{t-1}) = \mathbb E_{{\widetilde}\theta_{t-1},t}[r_{tj}v_{tj}] - \{\mathbb E_{{\widetilde}\theta_{t-1},t}r_{tj}\}\{\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj}\}= \mathbb E_{{\widetilde}\theta_{t-1},t}[(r_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}r_{tj})(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})]$. Subsequently, by Jenson’s inequality and the fact that $r_{tj}\in[0,1]$ almost surely, $$\begin{aligned}
\nabla\mathfrak R_t({\widetilde}\theta_{t-1})\nabla\mathfrak R_t({\widetilde}\theta_{t-1})^\top
&\preceq \mathbb E_{{\widetilde}\theta_{t-1,t}}\left[(r_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}r_{tj})^2(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})^\top\right]\nonumber\\
&\preceq \mathbb E_{{\widetilde}\theta_{t-1,t}}\left[(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})^\top\right]
= {\widehat}M_t({\widetilde}\theta_{t-1}|S).
\label{eq:delta-intermediate-2}
\end{aligned}$$
Define ${\widehat}M_t(\theta|S) := \mathbb E_{\theta,t}[(v_{tj}-\mathbb E_{\theta,t}v_{tj})(v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top]$, where $S\subseteq[N]$ is the assortment supplied at iteration $t$. Combining Eqs. (\[eq:delta-intermediate-1\],\[eq:delta-intermediate-2\]) with Lemma \[lem:mle\], we have $$\big|{\widehat}R_t(S)-R_t(S)\big|
\lesssim \sqrt{d\log(\rho\nu NT)}\cdot \sqrt{\|I_{t-1}(\theta_0)^{-1/2}{\widehat}M_t({\widetilde}\theta_{t-1}|S)I_{t-1}(\theta_0)^{-1/2}\|_\op}.
\label{eq:delta-intermediate-3}$$
It remains to show that ${\widehat}M_t({\widetilde}\theta_{t-1}|S)$ and $M_t(\theta_0|S)$ are close, for which we first recall the definitions of both quantities: $$\begin{aligned}
{\widehat}M_t({\widetilde}\theta_{t-1}|S) &= \mathbb E_{{\widetilde}\theta_{t-1,t}}\left[(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})(v_{tj}-\mathbb E_{{\widetilde}\theta_{t-1},t}v_{tj})^\top\right];\\
M_t(\theta_0|S) &= \mathbb E_{\theta_0,t}[v_{tj}v_{tj}^\top] - \{\mathbb E_{\theta_0,t}v_{tj}\}\{\mathbb E_{\theta_0,t}v_{tj}\}^\top = {\widehat}M_t(\theta_0|S).
\end{aligned}$$
The next lemma shows that under suitable conditions ${\widehat}M_t({\widetilde}\theta_{t-1}|S)$ is close to ${\widehat}M_t(\theta_0|S)=M_t(\theta_0|S)$, implying that $\frac{1}{4}M_t(\theta_0|S)\preceq {\widehat}M_t({\widetilde}\theta_{t-1}|S) \preceq 4M_t(\theta_0|S)$. It is proved in the supplementary material.
Suppose $\tau \leq 1/\sqrt{8\rho^2\nu^2 K^2}$. Then $\frac{1}{4}M_t(\theta_0|S)\preceq {\widehat}M_t({\widetilde}\theta_{t-1}|S) \preceq 4M_t(\theta_0|S)$ for all $t$, $S$ and $\theta$. \[lem:close-M\]
As a consequence of Lemma \[lem:close-M\], the right-hand side of Eq. (\[eq:delta-intermediate-3\]) can be upper bounded by $$\sqrt{d\log(\rho\nu TK)}\cdot \sqrt{4\|I_{t-1}(\theta_0)^{-1/2} M_t(\theta_0|S)I_{t-1}(\theta_0)^{-1/2}\|_\op}.$$
Lemma \[lem:delta-method\] is thus proved. We are now ready to prove Lemma \[lem:ucb\]. By Lemma \[lem:delta-method\], we know that with high probability $$\big|{\widehat}R_t(S)-R_t(S)\big| \lesssim \sqrt{d\log(\rho\nu TK)}\cdot \sqrt{\|I_{t-1}(\theta_0)^{-1/2} M_t(\theta_0|S)I_{t-1}(\theta_0)^{-1/2}\|_\op}$$
In addition, by Lemma \[lem:close-M\] and the fact that $\|{\widehat}\theta_{t-1}-\theta_0\|_2\leq\tau$ thanks to the local MLE formulation, we have $\frac{1}{4}M_t(\theta_0|S)\preceq {\widehat}M_t({\widehat}\theta_{t-1}|S)\preceq 4M_t(\theta_0|S)$ and subsequently $\frac{1}{4}I_{t-1}(\theta_0)\preceq{\widehat}I_{t-1}({\widehat}\theta_{t-1})\preceq 4I_{t-1}(\theta_0)$ because $I_{t-1}(\cdot)$ and ${\widehat}I_{t-1}(\cdot)$ are summations of $M_{t'}(\cdot)$ and ${\widehat}M_{t'}(\cdot)$ terms. Setting $\omega \gtrsim \sqrt{d\log(\rho\nu TK)}$ we proved that ${\overline}R_t(S)\geq R_t(S)$. The second property of Lemma \[lem:ucb\] can be proved similarly, by invoking the spectral similarities between $I_{t-1}(\cdot)$, $M_{t'}(\cdot)$ and ${\widehat}I_{t-1}(\cdot)$, ${\widehat}M_{t'}(\cdot)$.
### Proof of Lemma \[lem:close-M\] {#proof-of-lemma-lemclose-m .unnumbered}
Suppose $\tau \leq 1/\sqrt{8\rho^2\nu^2 K^2}$. Then $\frac{1}{4}M_t(\theta_0|S)\preceq {\widehat}M_t({\widetilde}\theta_{t-1}|S) \preceq 4M_t(\theta_0|S)$ for all $t$, $S$ and $\theta$.
Define ${\overline}M_t(\theta|S) := \mathbb E_{\theta_0,t}[(v_{tj}-\mathbb E_{\theta,t}v_{tj})(v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top]$, where only the outermost expectation is replaced by taking with respect to the probability law under $\theta_0$. Denote also ${\widetilde}w_j := v_{tj}-\mathbb E_{\theta,t}v_{tj}$. Then ${\overline}M_t(\theta|S) = \sum_j p_{\theta_0,t}(j){\widetilde}w_j{\widetilde}w_j^\top$ and ${\overline}M_t(\theta|S)-{\widehat}M_t(\theta|S) = \sum_j \delta_j{\widetilde}w_j{\widetilde}w_j^\top$, where $\delta_j = p_{\theta_0,t}(j)-p_{\theta,t}(j)$. By Eq. (\[eq:deltai\]) and the fact that $\|v_{ti}\|_2\leq\nu$, $\|\theta-\theta_0\|_2\leq\tau$, we have $$\max_j|\delta_j|\leq \sqrt{4 \nu^2\cdot \tau}.
\label{eq:propclosem-intermediate1}$$
On the other hand, by (A2) we know that $\min_jp_{\theta_0,t}(j)\geq 1/\rho K$ and therefore $${\overline}M_t(\theta|S) = \sum_j p_{\theta_0,t}{\widetilde}w_j{\widetilde}w_j^\top \succeq \frac{1}{\rho K}\sum_j {\widetilde}w_j{\widetilde}w_j^\top.
\label{eq:propclosem-intermediate2}$$
Combining Eqs. (\[eq:propclosem-intermediate1\],\[eq:propclosem-intermediate2\]) and the fact that ${\overline}M_t(\theta|S)-{\widehat}M_t(\theta|S) = \sum_j \delta_j{\widetilde}w_j{\widetilde}w_j^\top$, we have ${\overline}M_t(\theta|S)-{\widehat}M_t(\theta|S)\preceq {\overline}M_t(\theta|S)/2$ and ${\widehat}M_t(\theta|S)-{\overline}M_t(\theta|S)\preceq {\overline}M_t(\theta|S)/2$, provided that $\tau \leq 1/\sqrt{8\rho^2\nu^2 K^2}$. This also implies $\frac{1}{2}{\overline}M_t(\theta|S)\preceq{\widehat}M_t(\theta|S)\preceq 2{\overline}M_t(\theta|S)$.
We next prove that $\frac{1}{2}M_t(\theta_0|S)\preceq {\overline}M_t(\theta|S)\preceq 2M_t(\theta_0|S)$ which, together with $\frac{1}{2}{\overline}M_t(\theta|S)\preceq{\widehat}M_t(\theta|S)\preceq 2{\overline}M_t(\theta|S)$ established in the previous section, implies Lemma \[lem:close-M\]. Recall the definitions that $$\begin{aligned}
M_t(\theta_0|S) &= \mathbb E_{\theta_0,t}\left[(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})^\top\right];\\
{\overline}M_t(\theta|S) &= \mathbb E_{\theta_0,t}\left[(v_{tj}-\mathbb E_{\theta,t}v_{tj})(v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top\right].
\end{aligned}$$
Adding and subtracting $\mathbb E_{\theta,t}v_{tj}, \mathbb E_{\theta_0,t}v_{tj}$ terms, we have $$\begin{aligned}
&{\overline}M_t(\theta|S)-M_t(\theta_0|S)\\
&= \mathbb E_{\theta_0,t}\left[(v_{tj}-\mathbb E_{\theta_0,t}v_{tj}+\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})(v_{tj}-\mathbb E_{\theta_0,t}v_{tj}+\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top\right]\\
&\;\;\;\; - \mathbb E_{\theta_0,t}\left[(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})^\top\right]\\
&= \mathbb E_{\theta_0,t}\left[(\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})^\top\right]
+ \mathbb E_{\theta_0,t}\left[(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})(\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top\right]\\
&\;\;\;\; + (\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})(\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top\\
&= (\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})(\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top.
\end{aligned}$$
By Eq. (\[eq:preceq-1\]) in the proof of Lemma \[lem:Itheta-Itheta0\], we have that $$(\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})(\mathbb E_{\theta_0,t}v_{tj}-\mathbb E_{\theta,t}v_{tj})^\top
\lesssim \frac{1}{2}\mathbb E_{\theta_0,t}[(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})(v_{tj}-\mathbb E_{\theta_0,t}v_{tj})^\top] = \frac{1}{2}M_t(\theta_0|S)$$ provided that $\tau \leq 1/\sqrt{8\rho^2\nu^2 K^2}$, thus implying $\frac{1}{2}M_t(\theta_0|S)\preceq {\overline}M_t(\theta|S)\preceq 2M_t(\theta_0|S)$.
Proof of Lemma \[lem:elliptical\]
---------------------------------
It holds that $$\sum_{t=T_0+1}^T\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op^2\} \leq
4\log\frac{\det I_{T}(\theta_0)}{\det I_{T_0}(\theta_0)} \lesssim d\log(\lambda_0^{-1}\rho\nu).$$
Denote $A_t := I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)$ as $d$-dimensional positive semi-definite matrices with eigenvalues sorted as $\sigma_1(A_t)\geq\cdots\geq\sigma_d(A_t)\geq 0$. By simple algebra, $$\begin{aligned}
\sum_{t=T_0+1}^T&\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op^2\}
= \sum_{t=T_0+1}^T\min\{1,\sigma_1(A_t)^2\}\nonumber\\
&\leq \sum_{t=T_0+1}^T2\log(1+\sigma_1(A_t)^2) \leq \sum_{t=T_0+1}^T4\log(1+\sigma_1(A_t)).
\label{eq:epl-intermediate-1}
\end{aligned}$$
On the other hand, note that $I_t(\theta_0) = I_{t-1}(\theta_0) + M_t(\theta_0|S_t)
= I_{t-1}(\theta_0)^{1/2}[I_{d\times d} + A_t] I_{t-1}(\theta_0)^{1/2}$. Hence, $$\log \det I_t(\theta_0) =\log \det I_{t-1}(\theta_0) + \sum_{j=1}^d\log(1+\sigma_j(A_t)).
\label{eq:epl-intermediate-2}$$
Comparing Eqs. (\[eq:epl-intermediate-1\]) and (\[eq:epl-intermediate-2\]), we have $$\sum_{t=T_0+1}^T\min\{1,\|I_{t-1}^{-1/2}(\theta_0)M_t(\theta_0|S_t)I_{t-1}^{-1/2}(\theta_0)\|_\op^2\} \leq 4\log\frac{\det I_T(\theta_0)}{\det I_{T_0}(\theta_0)},$$ which proves the first inequality in Lemma \[lem:elliptical\].
We next prove the second inequality in Lemma \[lem:elliptical\]. Because assortments have size 1 throughout the pure exploration phase ($t\leq T_0$), we have $$\begin{aligned}
I_{T_0}(\theta_0) = \sum_{t=1}^{T_0} p_{\theta_0,t}(j_t)(1-p_{\theta_0,t}(j_t))^2 v_{t,j_t}v_{t,j_t}^\top
\geq\frac{1}{(1+\rho)^3}\cdot \sum_{t=1}^{T_0} v_{t,j_t}v_{t,j_t}^\top,
\end{aligned}$$ where the last inequality holds thanks to assumption (A2), which implies $p_{\theta_0,t}(j_t) \in [1/(1+\rho), \rho/(1+\rho)]$. In addition, by the proof of Corollary \[cor:tau\], with high probability $\lambda_{\min}(\sum_{t=1}^{T_0}v_{t,j_t}v_{t,j_t}^\top)\geq 0.5 T_0\lambda_0$, where $\lambda_0>0$ is a parameter specified in assumption (A1). Therefore, $$\det I_{T_0}(\theta_0) \gtrsim [T_0\lambda_0/\rho^3]^d.
\label{eq:epl-intermediate-3}$$
On the other hand, because $\max_{t,j}\|v_{tj}\|_2 \leq \nu$ we have $I_T(\theta_0) \lesssim T\cdot \nu^2$ and subsequently $$\det I_T(\theta_0) \lesssim [\nu^2 T]^d.
\label{eq:epl-intermediate-4}$$
Combining Eqs. (\[eq:epl-intermediate-3\]) and (\[eq:epl-intermediate-4\]) we proved the second inequality in Lemma \[lem:elliptical\].
Proofs of technical lemmas for Theorem \[thm:lower\] (lower bound)
==================================================================
Proof of Lemma \[lem:R-lb\]
---------------------------
Suppose $\epsilon\in(0,1/d\sqrt{d})$ and define $\delta := d/4- |{\widetilde}U_t\cap W|$. Then $$R(S_{\theta_W}^*)-R({\widetilde}S_t) \geq\frac{\delta\epsilon}{4K\sqrt{d}} .$$
Let $v=v_W$ and ${\widehat}v=v_{{\widetilde}U_t}$ be the corresponding feature vectors. Then $$\begin{aligned}
R(S_{\theta_W}^*) - R({\widetilde}S_t)
&= \frac{K\exp\{v^\top\theta_W\}}{1+K\exp\{v^\top\theta_W\}} - \frac{K\exp\{{\widehat}v^\top\theta_W\}}{1+K\exp\{{\widehat}v^\top\theta_W\}} \\
&= \frac{K[\exp\{v^\top\theta_W\} - \exp\{{\widehat}v^\top\theta_W\}]}{(1+K\exp\{v^\top\theta_W\})(1+K\exp\{{\widehat}v^\top\theta_W\})}\\
&\geq \frac{\exp\{v^\top\theta_W\} - \exp\{{\widehat}v^\top\theta_W\}}{2Ke}.
\end{aligned}$$
Here the last inequality holds because $\max(\exp\{v^\top\theta_W\},\exp\{{\widehat}v^\top\theta_W\})\leq e$. In addition, by Taylor expansion we know that $1+x\leq e^x\leq 1+x+x^2/2$ for all $x\in[0,1]$. Subsequently, $$\begin{aligned}
R(S_{\theta_W}^*) - R({\widetilde}S_t)
&\geq \frac{(v-{\widehat}v)^\top \theta_W - ({\widehat}v^\top\theta_W)^2/2}{2Ke}
\geq \frac{\delta\epsilon/\sqrt{d} - (\sqrt{d}\epsilon)^2/2}{2Ke}.
\end{aligned}$$
Finally, noting that $d\epsilon^2/2\leq \delta\epsilon/2\sqrt{d}$ provided that $\epsilon\in(0,1/d\sqrt{d})$, we finish the proof of Lemma \[lem:R-lb\].
Proof of Lemma \[lem:kl\]
-------------------------
For any $W\in\mathcal W_{d/4-1}$ and $i\in[d]$, ${\mathrm{KL}}(P_W\|P_{W\cup\{i\}}) \leq C_{{\mathrm{KL}}}\cdot \mathbb E_W[N_i] \cdot\epsilon^2/{d}$ for some universal constant $C_{{\mathrm{KL}}}>0$.
Fix a time $t$ with policy’s assortment choice $S_t$, and define $n_i(S_t) := \sum_{v_U\in S_t}{\boldsymbol }1\{i\in U\}/K$. Let $\{p_j\}_{j\in S_t\cup\{0\}}$ and $\{q_j\}_{j\in S_t\cup \{0\}}$ be the probabilities of purchasing item $j$ under parameterization $\theta_W$ and $\theta_{W\cup\{i\}}$, respectively. Then $${\mathrm{KL}}(P_W(\cdot|S_t)\|P_{W\cup\{i\}}(\cdot|S_t))
= \sum_{j\in S_t\cup\{0\}} p_j\log\frac{q_j}{p_j}
\leq \sum_{j}p_j\frac{p_j-q_j}{q_j}
\leq \sum_j \frac{|p_j-q_j|^2}{q_j},
\label{eq:kl-intermediate-1}$$ where the only inequality holds because $\log(1+x)\leq x$ for all $x>-1$. Because $q_j\geq e^{-1}/(1+Ke) \geq 1/(2Ke^2)$ for all $j\in S_t\cup\{0\}$, Eq. (\[eq:kl-intermediate-1\]) is reduced to $${\mathrm{KL}}(P_W(\cdot|S_t)\|P_{W\cup\{i\}}(\cdot|S_t)) \leq 2e^2 K\cdot \sum_{j\in S_t\cup\{0\}} |p_j-q_j|^2.
\label{eq:kl-intermediate-2}$$
We next upper bound $|p_j-q_j|$ separately. First consider $j=0$. We have $$\begin{aligned}
|p_j-q_j|
&= \left|\frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_W\}} - \frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_{W\cup\{i\}}\}}\right|\\
&\leq \frac{1}{(1+K/e)^2}\cdot 2\sum_{j\in S_t}\big|v_j^\top(\theta_W-\theta_{W\cup\{i\}})\big|\\
&\leq \frac{2Kn_i(S_t)\epsilon/\sqrt{d}}{(1+K/e)^2}\leq \frac{8e^2n_i(S_t)\epsilon}{K\sqrt{d}}.
\end{aligned}$$
Here the first inequality holds because $e^x\leq 1+2x$ for all $x\in[0,1]$.
For $j>0$ corresponding to $v_j=v_U$ where $i\notin U$, we have $$\begin{aligned}
|p_j-q_j| &= \left|\frac{\exp\{v_U^\top\theta_W\}}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_W\}} - \frac{\exp\{v_U^\top\theta_{W\cup\{i\}}\}}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_{W\cup\{i\}}\}}\right|\\
&\leq \left|\frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_W\}} - \frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_{W\cup\{i\}}\}}\right|\\
&\leq \frac{8e^2n_i(S_t)\epsilon}{K\sqrt{d}}.
\end{aligned}$$
Here the first inequality holds because $\exp\{v_U^\top\theta_W\} = \exp\{v_U^\top\theta_{W\cup\{i\}}\} \leq 1$, since $i\notin U$.
For $j>0$ corresponding to $v_j=v_U$ and $i\in U$, we have $$\begin{aligned}
|p_j-q_j| &= \left|\frac{\exp\{v_U^\top\theta_W\}}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_W\}} - \frac{\exp\{v_U^\top\theta_{W\cup\{i\}}\}}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_{W\cup\{i\}}\}}\right|\\
&\leq \exp\{v_u^\top\theta_{W\cup\{i\}}\}\cdot \left|\frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_W\}} - \frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_{W\cup\{i\}}\}}\right|\\
&\;\; + \big|\exp\{v_u^\top\theta_W\} - \exp\{v_u^\top\theta_{W\cup\{i\}}\}\big|\cdot \left|\frac{1}{1+\sum_{j\in S_t}\exp\{v_j^\top\theta_W\}}\right|\\
&\leq \frac{8e^2n_i(S_t)\epsilon}{K\sqrt{d}} + \frac{\epsilon}{\sqrt{d}}\cdot \frac{1}{1+K/e}.
\leq \frac{8e^2n_i(S_t)\epsilon}{K\sqrt{d}} + \frac{2e\epsilon}{K\sqrt{d}}.
\end{aligned}$$
Combining all upper bounds on $|p_j-q_j|$ and Eq. (\[eq:kl-intermediate-2\]), we have $$\begin{aligned}
{\mathrm{KL}}(P_W(\cdot|S_t)\|P_{W\cup\{i\}}(\cdot|S_t))
&\leq 2e^2K\cdot \left[\frac{128e^4n_i(S_t)^2\epsilon^2}{K^2d}(1+K)+ Kn_i(S_t)\cdot \frac{8e^4\epsilon^2}{K^2d} \right]\nonumber\\
&\lesssim n_i(S_t)\epsilon^2/{d}.
\end{aligned}$$
Here the last inequality holds because $n_i(S_t) \leq 1$. Note also that $N_i = \sum_{t=1}^Tn_i(S_t)$ by definition, and subsequently summing over all $t=1$ to $T$ we have $${\mathrm{KL}}(P_W\|P_{W\cup\{i\}}) \lesssim \mathbb E_W[N_i]\cdot \epsilon^2/{d},$$ which is to be demonstrated.
Proofs of approximation algorithms
==================================
Proof of Lemma \[lem:approx\]
-----------------------------
Suppose an $(\alpha,\varepsilon,\delta)$-approximation algorithm is used instead of exact optimization in the MLE-UCB policy at each time period $t$. Then its regret can be upper bounded by $$\alpha\cdot \mathrm{Regret}^* + \varepsilon T + \delta T^2 + O(1),$$ where $\mathrm{Regret}^*$ is the regret upper bound shown by Theorem \[thm:mle-ucb\] for Algorithm \[alg:mle-ucb\] with exact optimization in Step \[alg:step6\].
By union bound, we know the approximation guarantee in Eq. (\[eq:approx\]) *for all $t$* with probability at least $1-\delta T$. In the event of failure, the accumulated regret is upper bounded by $T$ almost surely, because the regret incurred by each time period $t$ is at most $1$. This gives rises to the $\delta T^2$ term in Lemma \[lem:approx\], and in the rest of the proof we shall assume Eq. (\[eq:approx\]) holds for all $t$.
Let $S_t^*$ be the solution to the exact optimization problem in Step \[alg:step6\] of Algorithm \[alg:mle-ucb\], $S_t^\#$ be the assortment with the optimal revenue the same step, and ${\widehat}S_t$ be the solution by an $(\alpha,\varepsilon,\delta)$-approximation algorithm.
For each $t > T_0$, we bound the expected regret incurred at time $t$ by $$\begin{aligned}
&\mathrm{ESTR}(S^\#_t) - \mathrm{ESTR}(S_t)\\
=& \left( \mathrm{ESTR}(S^\#_t) + \min\{1 + \omega \cdot \mathrm{CI}(S^\#_t)\} \right) - \left( \mathrm{ESTR}(S^*_t) + \min\{1 + \omega \cdot \mathrm{CI}(S^*_t)\} \right) \\
& \qquad + \left( \mathrm{ESTR}(S^*_t) + \min\{1 + \omega \cdot \mathrm{CI}(S^*_t)\}\right) - \left(\mathrm{ESTR}(S_t) + \min\{1 + \alpha \omega \cdot \mathrm{CI}(S_t)\} - \varepsilon\right) \\
& \qquad + \left(\varepsilon + \min\{1 + \alpha \omega \cdot \mathrm{CI}(S_t)\} - \min\{1 + \omega \cdot \mathrm{CI}(S^\#_t)\} \right)\\
\leq & \varepsilon + \min\{1 + \alpha \omega \cdot \mathrm{CI}(S_t)\} - \min\{1 + \omega \cdot \mathrm{CI}(S^\#_t)\} \leq \varepsilon + \min\{1 + \alpha \omega \cdot \mathrm{CI}(S_t)\} .
\end{aligned}$$
Therefore, the total expected regret is bounded by $$\begin{aligned}
T_0 + \sum_{t = T_0 + 1}^{T} \left(\varepsilon + \min\{1 + \alpha \omega \cdot \mathrm{CI}(S_t)\} \right)
\leq \varepsilon T + T_0 + \alpha \sum_{t = T_0 + 1}^{T} \min\{1 + \omega \cdot \mathrm{CI}(S_t)\},
\end{aligned}$$ which, by the same analysis in Section \[sec:ellipitical\], can be bounded by $\alpha \cdot \mathrm{Regret}^* + \varepsilon T$.
Proof of Lemma \[lem:approx-1d-estr-ci\]
----------------------------------------
For any $S\subseteq[N]$, $|S|\leq K$, suppose $U = \max_{j\in S}\{1, {\widehat}{u}_{tj}\}$ and $\Delta = \epsilon_0 U / K$ for some $\epsilon_0>0$. Suppose also $|x_{tj}|\leq\nu$ for all $t,j$. Then $$\big|\mathrm{ESTR}(S)-{\widehat}{\mathrm{ESTR}}(S)\big| \leq 6\epsilon_0\;\;\;\;\;\text{and}\;\;\;\;\;
\big|\mathrm{CI}(S)-{\widehat}{\mathrm{CI}}(S)\big| \leq \sqrt{24 \epsilon_0} (1 +\nu),$$
We first prove the upper bound on $|\mathrm{ESTR}(S)-{\widehat}{\mathrm{ESTR}(S)}|$, which is $$\left|\frac{\sum_{i\in S}{\widehat}u_{ti}r_{ti}}{1+\sum_{i\in S}{\widehat}u_{ti}} - \frac{\sum_{i\in S}\gamma_i}{1+\sum_{i\in S}\mu_i}\right| \leq 6\epsilon_0,
\label{eq:approx-estr}$$ where $\mu_i = [{\widehat}u_{ti}/\Delta]\cdot \Delta$, $\gamma_i = [{\widehat}u_{ti}r_{ti}/\Delta]\cdot \Delta$.
Denote $A := \sum_{i\in S}{\widehat}u_{ti}r_{ti}$ and $B := 1 + \sum_{i\in S}{\widehat}u_{ti}$. Because $r_{ti}\leq 1$, we have $A\leq B$. Let also $\tau_1 := \sum_{i\in S}\gamma_i - A$ and $\tau_2 := 1 + \sum_{i\in S}\mu_i - B$. Because $\max\{|\gamma_i - {\widehat}u_{ti}r_{ti}|, |\mu_i-{\widehat}u_{ti}|\} \leq \Delta$, we have $\max\{|\tau_1|,|\tau_2|\}\leq \Delta\cdot K$. Subsequently, $$\begin{aligned}
\left|\frac{\sum_{i\in S}{\widehat}u_{ti}r_{ti}}{1+\sum_{i\in S}{\widehat}u_{ti}} - \frac{\sum_{i\in S}\gamma_i}{1+\sum_{i\in S}\mu_i}\right|
&= \left|\frac{A}{B} - \frac{A+\tau_1}{B+\tau_2}\right|
= \left|\frac{A\tau_2-B\tau_1}{B(B+\tau_2)}\right|
= \left|\frac{A\tau_2-B\tau_2+B\tau_2-B\tau_1}{B(B+\tau_2)}\right|\\
&\leq \left|\frac{(A-B)\tau_2}{B(B+\tau_2)}\right| + \frac{|\tau_1|+|\tau_2|}{B-|\tau_2|}
\leq \frac{|\tau_1|+2|\tau_2|}{B-|\tau_2|},
\end{aligned}$$ where the last inequality holds because $A\leq B$. Using $B=1+\sum_{i\in S}{\widehat}u_{ti} \geq 1 + {\widehat}{u}_{tq} \geq U$ (since $q \in S$ and $U = \max\{1, u_{tq}\}$, and $\max\{|\tau_1|,|\tau_2|\}\leq \Delta\cdot K = \epsilon_0 U$, we have $$\frac{|\tau_1|+2|\tau_2|}{B-|\tau_2|} \leq \frac{3 \epsilon_0 U}{U - \epsilon_0 U} \leq 6\epsilon_0,$$ provided that $\epsilon_0\in(0,1/2]$. Eq. (\[eq:approx-estr\]) is thus proved.
We next prove the upper bound on $|\mathrm{CI}(S)-{\widehat}{\mathrm{CI}}(S)|$, which is $$\left|\sqrt{\frac{\sum_{i\in S}{\widehat}u_{ti}x_{it}^2}{1+\sum_{i\in S}{\widehat}u_{ti}}-\left(\frac{\sum_{i\in S}{\widehat}u_{ti}x_{ti}}{1+\sum_{i\in S}{\widehat}u_{ti}}\right)^2} - \sqrt[*]{\frac{\sum_{i\in S}\beta_i}{1+\sum_{i\in S}\mu_i} - \left(\frac{\sum_{i\in S}\alpha_i}{1+\sum_{i\in S}u_i}\right)^2}\right| \leq \sqrt{24 \epsilon_0} (1 +\nu),
\label{eq:approx-ci}$$ where $\sqrt[*]{\cdot} = \sqrt{\max\{0,\cdot\}}$, $\mu_i = [{\widehat}u_{ti}/\Delta]\cdot \Delta$, $\alpha_i = [{\widehat}u_{ti}x_{ti}/\Delta]\cdot \Delta$, $\beta_i = [{\widehat}u_{ti}x_{ti}^2/\Delta]\cdot \Delta$.
Denote $C := \frac{\sum_{i\in S}{\widehat}u_{ti}x_{ti}}{1+\sum_{i\in S}{\widehat}u_{ti}}$ and $D := \frac{\sum_{i\in S}{\widehat}u_{ti}x_{ti}^2}{1+\sum_{i\in S}{\widehat}u_{ti}}$. Because $|x_{ti}|\leq\nu$ for all $t$ and $i$, we have $C\in[-\nu,\nu]$ and $D\in[0,\nu^2]$. Denote also $\tau_3 := \frac{\sum_{i\in S}\alpha_i}{1+\sum_{i\in S}\mu_i}-C$ and $\tau_4 := \frac{\sum_{i\in S}\beta_i}{1+\sum_{i\in S}\mu_i} - D$. Using the same analysis as in the proof of Eq. (\[eq:approx-estr\]), we have $|\tau_3| \leq 6 \epsilon_0 (1+\nu)$ and $|\tau_4| \leq 6 \epsilon_0 (1+\nu^2)$.
With the definitions of $C$, $D$, $\tau_3$ and $\tau_4$, the left-hand side of Eq. (\[eq:approx-ci\]) can be re-written as $$\big|\sqrt{D-C^2} - \sqrt[*]{(D+\tau_4) - (C+\tau_3)^2}\big|.
\label{eq:approx-ci-ref}$$
#### Case 1: $D-C^2>-(\tau_4-2\tau_3C-\tau_3^2)$.
In this case, we have $$\begin{aligned}
\text{Eq.~(\ref{eq:approx-ci-ref})}
&= \frac{|\tau_4-2\tau_3C-\tau_3^2|}{\sqrt{D-C^2} + \sqrt{D-C^2+(\tau_4-2\tau_3 C-\tau_3^2)}}\leq \sqrt{|\tau_4-2\tau_3 C-\tau_3^2|}\\
&\leq \sqrt{6\epsilon_0 (1 + \nu^2) + 2 \cdot 6\epsilon_0 (1 + \nu)^2 + 6\epsilon_0 (1+\nu)} \leq \sqrt{24 \epsilon_0} (1 +\nu).
\end{aligned}$$
#### Case 2: $D-C^2\leq -(\tau_4-2\tau_3C-\tau_3^2)$.
In this case, we have $(D+\tau_4)-(C+\tau_3)^2\leq 0$ and subsequently $$\text{Eq.~(\ref{eq:approx-ci-ref})}
= \sqrt{D-C^2} \leq \sqrt{|\tau_4-2\tau_3 C-\tau_3^2|} \leq \sqrt{24 \epsilon_0} (1 +\nu).$$
Combining both cases we prove Eq. (\[eq:approx-ci\]).
Proof of Lemma \[lem:approx-md\]
--------------------------------
Suppose there exists $\ell\in[L]$ such that $\langle y^{(\ell)},y^*\rangle \geq 1/\alpha$ for some $\alpha\geq 1$ in Algorithm \[alg:approx-md\], then $\mathrm{ESTR}({\widehat}S^{(\ell)}) + \min\{1,\alpha\omega\cdot\mathrm{CI}({\widehat}S^{(\ell)})\}+\varepsilon \geq \mathrm{ESTR}(S^*)+\min\{1,\omega\cdot\mathrm{CI}(S^*)\}$, where $\varepsilon>0$ is the approximation parameter of the univariate problem instances.
For each assortment $S$, define $\mathrm{CI}^{(\ell)} (S)$ by $$\mathrm{CI}^{(\ell)} (S) :=
{y^{(\ell)}}^\top \left(\frac{\sum_{j\in S}{\widehat}u_{tj}x_{tj}x_{tj}^\top}{1+\sum_{j\in S}{\widehat}u_{tj}} - \left(\frac{\sum_{j\in S}{\widehat}u_{tj}x_{tj}}{1+\sum_{j\in S}{\widehat}u_{tj}}\right)\left(\frac{\sum_{j\in S}{\widehat}u_{tj}x_{tj}}{1+\sum_{j\in S}{\widehat}u_{tj}}\right)^\top\right) y^{(\ell)},$$ Since $\mathrm{CI}^{(\ell)} (S) \leq \mathrm{CI} (S)$, we have $$\label{eq:approx-md-eq1}
\mathrm{ESTR}({\widehat}S^{(\ell)}) + \min\{1,\alpha\omega\cdot\mathrm{CI}({\widehat}S^{(\ell)})\}+\varepsilon \geq \mathrm{ESTR}({\widehat}S^{(\ell)}) + \min\{1,\alpha \omega\cdot\mathrm{CI}^{(\ell)}({\widehat}S^{(\ell)})\}+\varepsilon .$$ By the approximation guarantee of Algorithm \[alg:approx-1d\], we have $$\label{eq:approx-md-eq2}
\mathrm{ESTR}({\widehat}S^{(\ell)}) + \min\{1,\alpha\omega\cdot\mathrm{CI}^{(\ell)}({\widehat}S^{(\ell)})\}+\varepsilon \geq \mathrm{ESTR}( S^*) + \min\{1,\alpha\omega\cdot\mathrm{CI}^{(\ell)}( S^*)\} .$$ Since $\langle y^{(\ell)},y^*\rangle \geq 1/\alpha$, we have $\mathrm{CI}^{(\ell)} (S^*) \geq (1/\alpha) \cdot\mathrm{CI}(S^*)$. Therefore, $$\label{eq:approx-md-eq3}
\mathrm{ESTR}( S^*) + \min\{1,\alpha\omega\cdot\mathrm{CI}^{(\ell)}( S^*)\} \geq \mathrm{ESTR}( S^*) + \min\{1,\omega\cdot\mathrm{CI}( S^*)\} .$$
The lemma is proved by combining Eq. , Eq. , and Eq. .
Proof of Proposition \[prop:greedy-omega-zero\]
-----------------------------------------------
If $\omega=0$, then Algorithm \[alg:greedy\] terminates in $O(N^4)$ iterations and produces an output $S$ that maximizes $\mathrm{ESTR}(S)$.
We first show that when the algorithm terminates with $\mathrm{ESTR}(S) = r$, $S$ is one of the optimal assortments. Suppose $S$ is not an optimal assortment, i.e. there exists $S^\#$ such that $\mathrm{ESTR}(S^\#) > r$, we show that the algorithm will not terminate. By the definition of $\mathrm{ESTR}(\cdot)$ we have $\sum_{i \in S^\#} {\widehat}{u}_{ti} (r_{ti} - r) > r$ and $\sum_{i \in S} {\widehat}{u}_{ti} (r_{ti} - r) = r$. By comparing $S^\#$ and $S$, one can find a new candidate assortment $S'$ via swapping, adding, or deleting an item from/to $S$ such that $\sum_{i \in S'} {\widehat}{u}_{ti} (r_{ti} - r) > r$. Therefore, $\mathrm{ESTR}(S') > r$ and the algorithm will not terminate.
It remains to show that the algorithm terminates in $O(N^4)$ iterations.
For each $r \in [0, 1]$, we define a total order $\geq_r$ on $[N] \cup \{\bot\}$, where $[N]$ corresponds to the $N$ items and $\bot$ is a special element with the definition ${\widehat}{u}_{t\bot} = r_{t\bot} = 0$ for convenience, as follows: $i \geq_r j$ if and only if ${\widehat}{u}_{ti} (r_{ti} - r) \geq {\widehat}{u}_{tj} (r_{tj} - r)$ (and consequently $i >_r j$ if and only if ${\widehat}{u}_{ti} (r_{ti} - r) > {\widehat}{u}_{tj} (r_{tj} - r)$). It is straightforward to verify that there exists $O(N^2)$ section points $\theta_0 < \theta_1 < \theta_2 < \dots < \theta_{L - 1} < \theta_L = 1$ so that for any two $r_1, r_2$ that sandwiched by the same pair of neighboring section points (i.e. $\exists \ell \in [L]: r_1, r_2 \in (\theta_{\ell - 1}, \theta_\ell)$), we have $\geq_{r_1} \equiv \geq_{r_2}$. Indeed, one can set the section points to be the solutions to the equalities ${\widehat}{u}_{ti} (r_{ti} - r) = {\widehat}{u}_{tj} (r_{tj} - r)$ for every pair of $i, j \in [N] \cup \{\bot\}$.
We will show that if $\mathrm{ESTR}(S) \in (\theta_{\ell-1}, \theta_\ell)$ for some $\ell \in [L]$, after at most $O(N^2)$ iterations, either the algorithm terminates or $\mathrm{ESTR}(S) \geq \theta_{\ell}$. This directly leads to an $O(N^4)$ upper bound on the total number of iterations that the algorithm performs. We pick an arbitrary $r \in (\theta_{\ell-1}, \theta_\ell)$ and define the following two potential functions: $I(S) = |\{(i, j): i \in S, j \in [N] \backslash S, i <_r j \}|$, and $J(S) = |\{i \in S: i <_r \bot\}|$. We have the following observations:
- When a swapping operation is performed on $S$, $I(S)$ strictly decreases and $J(S)$ does not increase.
- When a deletion operation is performed on $S$, $I(S)$ increases by at most $N$ and $J(S)$ strictly decreases.
- When an addition operation is performed on $S$, $I(S)$ increases by at most $N$ and $J(S)$ does not increase.
We let $F(S) = I(S) + (2N+1) \cdot J(S) \in [0, O(N^2)]$. Suppose there are $a$ swapping operations, $b$ deletion operations, and $c$ addition operations done in total, $S$ is the assortment that the algorithm begins with and $T$ is the last assortment satisfying $\mathrm{ESTR}(T) < \theta_\ell$. Observe that there are at most $c \leq b + N$ addition operations. Together with the three observations above, we have $$\begin{gathered}
0 \leq F(T) \leq F(S) - a + N b - (2N + 1) b + N c \leq F(S) - a + N b - (2N + 1) b + N (b + N) \\
\leq F(S) + N^2 - a - b \leq O(N^2) - a - b .
\end{gathered}$$ In total, we have $a + b \leq O(N^2)$. Therefore, the total number of iterations where $\mathrm{ESTR}(S) \in (\theta_{\ell -1}, \theta_\ell)$ is $a + b + c \leq a + 2b + N \leq O(N^2)$.
Proof of Proposition \[prop:approx-init\]
-----------------------------------------
Assume that $d \geq 2$. Let $y^*\in\mathbb R^d$, $\|y^*\|_2=1$ be fixed and $y$ be sampled uniformly at random from the unit $d$-dimensional sphere. Then $$\Pr[\langle y,y^*\rangle \geq 1/\sqrt{d}] = \Omega(1) \;\;\;\;\;\text{and}\;\;\;\;\; \Pr[\langle y,y^*\rangle \geq 1/2] = \exp\{-O(d)\}.$$
Assume without loss of generality that $y^* = (1, 0, 0, \dots, 0)$, and let $y$ be sampled as follows. Sample $z_i \sim N(0, 1)$ independently for each $i \in [d]$, and let $y = z / \|z\|_2$. Now, $\langle y, y^* \rangle = z_1 / \|z\|_2$.
We first prove $\Pr[\langle y,y^*\rangle \geq 1/\sqrt{d}] = \Pr[z_1 / \|z\|_2 \geq 1 / \sqrt{d}] = \Omega(1)$. Note that when $z_1 \geq 10$ and $ \sqrt{z_2^2 + \dots + z_d^2} \leq 5\sqrt{d}$, we have $z_1 / \|z\|_2 = 1 / \sqrt{1 + (z_2^2 + \dots + z_d^2)/z_1^2} \geq 1 / \sqrt{1 + (5 \sqrt{d})^2 / 10^2} \geq 1 / \sqrt{d}$, where the last inequality holds for $d \geq 2$. Therefore, $$\begin{aligned}
\Pr[z_1 / \|z\|_2 \geq 1/\sqrt{d}] &\geq \Pr\left[z_1 \geq 10 \wedge \sqrt{z_2^2 + \dots + z_d^2} \leq 5\sqrt{d}\right] \\
&= \Pr[z_1 \geq 10] \cdot \Pr\left[ \sqrt{z_2^2 + \dots + z_d^2} \leq 5\sqrt{d}\right] = \Omega(1) .
\end{aligned}$$
Now we prove $\Pr[\langle y,y^*\rangle \geq 1/2] = \Pr[z_1 / \|z\|_2 \geq 1 / 2] = \exp\{-O(d)\}$. Similarly, when $z_1 \geq 5\sqrt{d}$ and $ \sqrt{z_2^2 + \dots + z_d^2} \leq 5\sqrt{d}$, we have $z_1 / \|z\|_2 = 1 / \sqrt{1 + (z_2^2 + \dots + z_d^2)/z_1^2} \geq 1 / \sqrt{1 + 1} > 1 / 2$. Therefore, $$\begin{aligned}
\Pr[z_1 / \|z\|_2 \geq 1/2] &\geq \Pr\left[z_1 \geq 5\sqrt{d} \wedge \sqrt{z_2^2 + \dots + z_d^2} \leq 5\sqrt{d}\right] \\
&= \Pr[z_1 \geq 5\sqrt{d}] \cdot \Pr\left[ \sqrt{z_2^2 + \dots + z_d^2} \leq 5\sqrt{d}\right] \\
&= \exp\{-O(d)\} \cdot \Omega(1) = \exp\{-O(d)\}.
\end{aligned}$$
[^1]: Author names listed in alphabetical order.
[^2]: For the ease of presentation in the introduction, we only present the dominating term under the common scenario that the selling horizon $T$ is larger than the dimensionality $d$ and the cardinality constraint $K$. Please refer to Theorem \[thm:mle-ucb\] for a more explicit expression of the obtained regret.
[^3]: This is because in some applications, the product features are easier to obtain by the seller as compared to customer features.
[^4]: [We slightly abuse the notation $S^*$ here following the optimization convention that $S^*$ denotes the optimal solution. Note that $S^*$ is different from $S_t^*$ in , where the latter means the assortment that maximizes the expected revenue at time $t$.]{}
[^5]: A polylogarithmic factor dependent on $T, K, \delta^{-1}, \nu, \rho$ is hidden in the ${\widetilde}{O}(\cdot)$ notation.
[^6]: $\{X_k\}_k$ forms a martingale if $\mathbb E[X_{k+1}|X_1,\cdots,X_k]=X_k$ for all $k$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Based on Density Functional Theory (DFT) with non-collinear-magnetism formulations, we have calculated the magnetism of single atom of 3d transition metals and the magnetic anisotropy of supported Ni chains on Au(110)-(1$\times2$) surface. Our results show that surface relaxations enhance the orbital moments of left-end elements (Ti,V) and quench the orbital moments of right-end elements (Fe,Co,Ni) on the Au(110)-(1$\times$2) surface. This is because Ti and V atoms raise their positions above the trough of the reconstructed Au(110) surface and Fe, Co, Ni atoms trap more deeply in the trough after the surface relaxations. From the study of magnetism of Ni one-dimensional atomic chains, their magnetic anisotropy are closely relate to their orbital quenching. The easy magnetized axis changes from the direction parallel to the chains to the direction perpendicular to the Ni chains when they absorb on the surface. These one-dimensional Ni chains in the trough of the reconstructed Au(110) surface have ferromagnetic order and are favored with chain-length less than 6.'
address:
- ' Key Laboratory of Materials Physics, Institute of Solid State Physics, Chinese Academy of Sciences, 230031-Hefei, P. R. China'
- 'Department of Physics, Fudan University, 200433-Shanghai, P. R. China'
author:
- Wei Fan
- 'Xin-Gao Gong'
title: 'Magnetism of 3d Transition Metals and Magnetic Anisotropy of one-dimension Ni Chains on Au(110)-(1$\times2$): A Noncollinear $ab-initio$ Density-Functional-Theory Study'
---
,
Non-Collinear Density Functional Theory ,Surface Magnetism , one-dimensional Magnetism ,Magnetic Crystalline Anisotropy 75.70.Ak ,75.30.Gw ,75.75.+a ,71.15.Mb
\[Intro\] Introduction
======================
The high-density magnetic recording and the memory devices require the strong magnetic anisotropy. The low-dimensional materials due to the reduced dimensionality generally have a favorable direction and the physical properties along this direction are generally different from the other directions. Thus we have the most possibilities to find the strong magnetic anisotropy materials from these low-dimensional materials. On the other hand, the reduced dimensionality reduces the atomic coordination number and enhances the spin and orbital magnetic moments of the materials. Since the pioneering experiments for the magnetism of Fe strips on the W(110) and Cu(111) surfaces [@Elmers1; @Shen1; @Hauschild1; @Pratzer1], the researches along this direction have been extended to others quasi-one-dimensional systems such as the monatomic magnetic chains on vicinal surface [@Gambardella1; @Gambardella2; @Gambardella3] and Co magnetic dot-chain on Ru(0001) surface [@Li1]. The transition from in-plane to out-of-plane magnetic anisotropy had been found in Fe nano-structure when it is approach to one-dimensional limit [@Boeglin1]. Besides the celebrating properties such as the strong magnetic anisotropy, the enhanced spin and orbital moments, some new phenomena have been found, such as the temperature and time dependent magnetization [@Elmers1; @Shen1] which make the realistic applications of these novel experimental devices face new serious problem such as the instability of magnetic structure.
It is valuable to investigate theoretically the magnetic stability of surface-supported novel structures [@Dreysse1]. Many theoretical and computational methods have been used to studied the magnetism of one-dimensional structure on surface such as the KKR [@Bellini1; @Wildberger1; @Lazarovits1; @Eisenbach1], and TB-LMTO [@Komelj1] and PAW methods [@Spisak1; @Spisak2] based on Density Functional Theory, and the numerically tight-binding self-consistent calculation [@Dovila1] and analytically tight-binding calculation [@Druzinic1]. Some common arguments have been found such as the enhancement of surface magnetism, the magnetic anisotropy are close related to the orbital moments and the spin-orbit coupling interaction (SOC).
The orbital polarization [@Brooks1; @Eriksson1; @Solovyev1] can remedy the small orbital moments in the calculations of density functional theory to approach the experimental values, and especially reproduce the correct phase transition in Ce companying with the volume collapses [@Eriksson1]. The orbital polarization plays the opposite role to crystal field which quenches the orbital moment to small value in crystal due to the splitting of ground state energy levels. However there are still many contradictions among the already known theoretical results. The density functional calculations with OP (orbital polarization) + SOC (spin-orbit coupling) have given the correct orbital moments but too large MAE (Magnetic Anisotropy Energy) for tetragonal and cubic Ni. If the OP term switches off the theoretical value of the magnetic anisotropy energy is close its experimental values [@Hjortstarn1]. The magnetic anisotropy energy of CoPt calculated using the so-called c-RPA [@Solovyev2] are weakly dependent on the new introduced OP interaction. The magnetic anisotropy energies are surprisingly consistent with experiments even only under LSDA approximation without the OP effects. The orbital moments of transition metals are well consistent with experiments. This method has been combined with PAW method to calculate quasi-particle spectrum of materials [@Fuchs1].
Based on Brook’s theory, the magnetic anisotropy is induced by the spin-orbital interaction [@Brooks2]. Although the orbital moments are underestimated, it’s still valuable to study the magnetic anisotropy properties in the absence of (OP) orbital polarization term. Theoretical works based on Density Functional Theory have also discovered others interesting properties of surface supported magnetic chains. Such as in Ref. [@Spisak2], the authors have studied the magnetism of the ultrathin wires of fifth and sixth row elements supported on Cu(117) and Ag(117) vicinal surfaces. They found that only metals with a half-filled $d$ band are found to have magnetic order on Cu(117) surface, additionally the ferromagnetic order is energetically stable. On the contrary, on the Ag(117) surface, the anti-ferromagnetic order is stable. The magnetism of mono-atomic chain is strongly dependent on their local environment absorbed on the surfaces.
Besides the high-index vicinal surfaces such as Ag(117), the reconstructed Au(110)-(1$\times$2) surface decorated with one-dimensional troughs along the closed-packed A-\[1$\bar{1}$0\] direction can be used as a template to grow one-dimensional nano-structure. Experimentally, the distribution of lengths of Ni chains shows that most Ni chains are short than 6 [@Hitzke1]. Our molecular simulation additionally shows that the lattice misfit between Ni chains and the Au substrate can induce the instability of Ni chains and makes them break into short segments [@Fan1].
In this work we study the magnetism of mono-atomic Ni chains supported on Au(110)-(1$\times$2) surface using the same theoretical method as in Ref. [@Spisak2]. Our studies focus on the magnetic anisotropy of surface-supported magnetic atomic chains. We perform the non-collinear magnetic calculations including the spin-orbit coupling interaction but without orbital polarization interaction. We have calculated the magnetic anisotropy energies of the supported Ni chains and found an off-plane easy magnetization axis perpendicular to the chains and the surface. On the contrary, the easy magnetization axes of the free-standing Ni chains are parallel to these chains. The change of the easy magnetization axes of the surface-supported chains are also found in the others systems [@Dovila1]. Our results also indicate the closed relationship between the magnetic anisotropy and the orbital magnetism.
The paper is arranged as follow. In the next section, we introduce the theoretical methods used in this paper. The third section will present the results for single 3d transition metal atoms absorbed on Au(110)-(1$\times$2) and Au(111) surface. The magnetism of a single magnetic atom on a surface is important to understand the magnetism of materials with more complex structures such as the one-dimensional nano-structure. The reduced dimensionality and coordination number for surface still enhance the magnetism of single absorbed atoms [@Gross1; @Gross2; @Hanf1; @Riegel1; @Ortega1]. The results of short one-dimensional mono-atomic Ni chains are included in the forth section. Our studies focus on the magnetic anisotropy of Ni mono-atomic chains. Finally, we conclude our results.
\[methods\] Theoretical Methods: Non-collinear Density Functional Theory and PAW methods
========================================================================================
We have calculated magnetism of Ni$_{n}$(n=1-5) chains supported on the Au(110)-(1$\times$2) surface based on Density Functional Theory [@Hohenberg1; @Kohn1] and the Methods of Projection of Augmentation Wave (PAW) [@Blochl1] with the plane-wave base set and GGA Perdew-Burke-Ernzerhof’s exchange-correlation potential [@Perdew1]. The PAW methods used in VASP program [@Kresse1; @Kresse3; @Hobbs1] is as accurate as frozen-core all-electron methods and hopefully improved to include all-electron relaxation. Our calculations include the non-collinear effects and the spin-orbit coupling which is proved important to heavy metals such as gold. In non-collinear formula, a magnetic moment as a vector can point to any direction in space. The orbital polarization (OP) interaction isn’t included in our calculations.
The basic theory used in this paper is Hobbs’s fully unconstrained non-collinear DFT formulas based on plane wave basis sets and all-electron PAW methods [@Hobbs1; @Hobbs2]. Instead of distributing local quantization axes for every atoms, which is used in programs based on atom sphere approximation (ASA) or others analogous methods [@Sandratskii1], Hobbs’s theory adopts only one global quantization axis and the vectors of magnetic electronic density are varying smoothly and continuously, which is more suited for the calculations of itinerant magnetism in transition metal materials.
In non-collinear density functional theory [@Hobbs1; @Hobbs2; @Sandratskii1], the total energy is the functional of a density matrix $n_{\alpha\beta}$ where $\alpha$ and $\beta$ are the spin index along a defined quantization axis. $\alpha=\uparrow$ or $\downarrow$ represent the spin angular momentum point to the positive direction or negative direction of the quantization axis. The sum of the diagonal elements is the total charge density, that is $n_{_{Tr}}=\sum_{\alpha}n_{\alpha\alpha}$. In a non-collinear magnetic system, the off-diagonal elements are non-zero. The 2$\times$2 matrix can be expanded using the complex matrix basis ($I$, $\sigma_{x}$, $\sigma_{y}$, $\sigma_{z}$) and expressed as $$\begin{aligned}
n_{\alpha\beta}=[n_{_{Tr}}\delta_{\alpha\beta}+\vec m
\cdot\vec\sigma_{\alpha\beta}]/2 ,
\end{aligned}$$ where $\vec m=\sum_{\alpha\beta}n_{\beta\alpha}\cdot\vec\sigma_{\alpha\beta}$ is the magnetic electronic density and $\vec\sigma$ is the Pauli matrix. In PAW methods, the pseudo-density matrix can be expressed using the pseudo-wave functions as $$\begin{aligned}
\tilde n_{\alpha\beta}(\vec{r})=\sum_{n}f_{n}\langle\tilde\Psi_{n}^{\beta}|\vec{r}\rangle
\langle\vec{r}|\tilde\Psi_{n}^{\alpha}\rangle ,
\end{aligned}$$ where $f_{n}$ is the occupation number. The total electronic density matrix $n_{\alpha\beta}=\tilde n_{\alpha\beta}+n^{1}_{\alpha\beta}-\tilde n^{1}_{\alpha\beta}$, where $n^{1}_{\alpha\beta}$ and $\tilde n^{1}_{\alpha\beta}$ are the on-site electronic density matrixes. The total energy as the sum of three parts $E=\tilde{E}+E^{1}-\tilde{E}^{1}$, where $\tilde E$ the smooth functional of the pseudo-density matrix $\tilde n_{\alpha\beta}(\vec{r})$ is evaluated on a regular grid, $E^{1}$ and $\tilde{E}^{1}$ the functionals of on-site electronic density matrixes $n^{1}_{\alpha\beta}$ and $\tilde n^{1}_{\alpha\beta}$ are calculated on radical support grid around every atom.
The Kohn-Sham Hamilton and Kohn-Sham equation can be obtained from the variation of total energy to the soft pseudo-density matrix and written as $$\begin{aligned}
H^{\alpha\beta}\left[n\right]&=&-\frac{1}{2}\Delta \delta^{\alpha\beta}+\tilde{v}^{\alpha\beta}_{eff}
\\ \nonumber
&+& \sum_{(ij)}|\tilde{p}_{i}\rangle(\hat{D}^{\alpha\beta}_{ij}+^{1}D^{\alpha\beta}_{ij}
-^{1}\tilde{D}^{\alpha\beta}_{ij})\langle\tilde{p}_{j}| ,
\\ \nonumber
\sum_{\beta}H^{\alpha\beta}|\tilde{\Psi}^{\beta}_{n}\rangle&=&
\varepsilon_{n}S^{\alpha\alpha}|\tilde{\Psi}_{n}^{\alpha}\rangle .
\end{aligned}$$ where $S^{\alpha\alpha}$ is the overlapping operator, $\tilde{D}^{\alpha\beta}_{ij}$, $^{1}\hat{D}^{\alpha\beta}_{ij}$ and $^{1}D^{\alpha\beta}_{ij}$ are the nonlocal interaction obtained by the PAW transformation, which are equivalent to the nonlocal pseudo-potential in the ultrasoft-potential methods. In non-collinear DFT formulas, the one-electron effective potential $\tilde{v}^{\alpha\beta}_{eff}=v_{H}(r)\delta_{\alpha\beta}+v_{xc}(r)\delta_{\alpha\beta}
+\vec{b}(r)\cdot\vec{\sigma}^{\alpha\beta}$, which is the summation of electrostatic potential $v_{H}(r)$, the nonmagnetic part of GGA exchange-correlation correction $v_{xc}(r)$ and the magnetic exchange-correlation potential $\vec{b}(r)\cdot\vec{\sigma}^{\alpha\beta}$, where $\vec{b}=\delta E_{xc}/\delta \vec{m}(r)$. The magnetic exchange-correlation potential will potentially enhance magnetic moments in non-collinear calculations.
The spin-orbit coupling enters into PAW formulas by all-electron part of PAW Hamiltonian and is expressed as $$\begin{aligned}
H_{SOI}^{\alpha\beta}=\frac{\hbar^{2}}{(2m_{e}c)^{2}}\sum_{ij}
\langle\phi_{i}|\frac{1}{r}\frac{dV_{sphere}}{dr}|\phi_{j}\rangle
|\tilde{p}_{i}\rangle\vec{\sigma}_{\alpha\beta}\cdot\vec{L}_{ij}
\langle\tilde{p}_{j}|.
\end{aligned}$$ where $\tilde{p}_{i}$ is the projector function and $\phi_{i}$ all-electron wave function. The one-electron effective potential is modified $\tilde{v}^{\alpha\beta}_{eff}\rightarrow\tilde{v}^{\alpha\beta}_{eff}+\tilde{H}_{SOI}^{\alpha\beta}$, and the same time, one part of nonlocal PAW potentials is modified according to $\hat{D}^{\alpha\beta}_{ij}\rightarrow\hat{D}^{\alpha\beta}_{ij}=\sum_{L}
\int(\tilde{v}^{\alpha\beta}_{eff}(r)+\tilde{H}_{SOI}^{\alpha\beta}(r))\hat{Q}_{ij}^{L}(r)dr$. The roles of spin-orbit interaction in $^{1}\hat{D}^{\alpha\beta}_{ij}$ and $^{1}D^{\alpha\beta}_{ij}$ are similar to $\hat{D}^{\alpha\beta}_{ij}$, however they are calculated only within atom sphere. In the calculations of real material, they adopt the values in the calculations of isolate atom and don’t update in Kohn-Sham self-consistent loop. More information on non-collinear density functional theory within PAW methods can be found in the reference [@Hobbs1].
The atoms can have their initial magnetic moments by constructing the magnetic electronic density and the magnetic moments around each of atoms are equal to their desired values. If the initial total magnetic moments point $\vec{n}$ direction, generally the direction of total moment changes in Kohn-Sham self-consistent loop. We can use the constrained DFT calculations to fix the direction of magnetic moments. Our self-consistent calculations show that the directions of magnetic moments change very small in ferromagnetic materials although using the unconstrained DFT calculations (see the table.(\[table2\]) in section \[Ni\_chain\_15\] ). In real ferromagnetic materials, the directions of magnetic moments are hard to change, once the magnetic domains have formed. This is because the change of magnetization direction must include the changes of magnetic moments of all magnetic atoms in the materials. An external magnetic field can force to change the direction of magnetic moments. Of course there exists an easy magnetization axis which is stable direction of magnetic moments. We need the ’strongest’ magnetic field to change direction of magnetic moments away from the direction of the easy magnetization axis. However for the materials with more complex magnetic structure such as the canted and spiral magnetic structures, the relaxations of directions of magnetic moments are intrinsic, the use of unconstrained DFT is essentially important [@Hobbs1; @Hobbs2].
\[mag\_3d\] The magnetism of 3d transition metal atoms absorbed on Au(110)-(1$\times$2) and Au(111) surface
===========================================================================================================
The missing row Au(110) surface reconstructs from Au(110) surface by having missed closed-packed Au chains every others, and forms one-dimensional trough along the close-packed A-\[1$\bar{1}$0\] direction. We construct a crystal slab with 4 atomic layers including 28 gold atoms. At the first step we optimized the lattice constant of gold crystal which is about 4.2$\AA$. In the procession of optimization, the RMM-DIIS algorithm [@Kresse1] is applied to the total-energy minimum with plane-wave energy cutoff 229.9eV and the 6x6x6 Monkhorst-Pack K points mesh. Now the size of super-cell is 11.8$\AA$$\times$8.4$\AA$$\times$5.94$\AA$.
We add a vacuum layer by extending the super-cell along C-\[110\] direction. The size along C-\[110\] direction of super-cell including vacuum layer is equal to 16$\AA$. The lattice constant of surface is generally smaller than that of the crystal. Thus we optimized the slab with vacuum layer together. Based on the optimized lattice constant, the size of the super-cell is 11.523$\AA$$\times$8.148$\AA$$\times$15.52$\AA$. The vacuum layer is about twice thicker than thickness of the slab. For the optimized Au(111) surface, the size of super-cell is about 11.5$\AA$$\times$9.97$\AA$$\times$20$\AA$ including 64 gold atoms. In above optimizations, we only change the lattice constant. The initial structures are constructed by placing the 3d transition metal atoms on the hollow sites of the Au(111) and Au(110)-(1$\times$2) surface in the trough (Fig. \[fig1\]). The initial structures are optimized for all atoms except for the bottom atoms by having used the conjugate-gradient Methods. The K-points meshes of 2$\times$2$\times$1 or $\Gamma$ point are used to sampling the first Brillouin-zone in the corresponding calculations of the electronic structure for Au(110)-(1$\times$2) and Au(111) surfaces respectively. In this work we use smaller k-points mesh to more efficiently optimize surface structures.
We use both collinear and non-collinear density functional calculations but mainly present the non-collinear results. In non-collinear DFT calculation, we use the optimized structures from collinear DFT calculation. Based on the relaxed structures, we calculate more exactly the electronic structures using the RMM-DIIS algorithm [@Kresse1] with the (6$\times$6$\times$1) Monkhorst-Pack grids sampling the first Brillouin zone. If the changes of the total energies are smaller than 0.0001eV between two electronic self-consistent (SC) steps the SC-loops break. We set the Methfessel-Paxton smearing width equal to 0.20eV to accelerate the speed of convergence. The energy-cutoffs of plane-wave for different atoms are summarized in table (\[table1\]). The magnetic moments for a single atom are calculated by considering the electrons in a Wigner-Seitz sphere centered at the position of the atom. The Wigner Seitz Radii for 3d transition metal atoms are also summarized in table (\[table1\]). We choose the C-\[110\] direction as the quantization axis. We emphasize the influence of surface deformation on the orbital moments of the absorbed atoms. The orbital polarization interaction omits in this work but includes the spin-orbit coupling interaction. We obtain smaller orbital moments compared with pervious density functional calculation including the orbital polarization interaction [@Nonas1].
\[mag\_3d\_relax\] The influence of structural relaxations on the magnetism of a single absorbed 3d transition atom
-------------------------------------------------------------------------------------------------------------------
It’s seem that collinear DFT calculation is enough to the study of magnetism of a single transition metal absorbed on the gold surface. However, if we want to know the direction of magnetic moments relative to the surface, we must use the non-collinear DFT formulas. These results for single absorbed atom will be compared with the results of one-dimensional atomic chains. At first step, we calculate the magnetic moments of the absorbed 3d atoms on the Au(110)-(1$\times$2) and Au(111) surfaces including the spin-orbit coupling interactions and non-collinear calculations. From Fig. \[fig2\], we can see that the spin moments reach the maximum in the middle of the group. The large spin moment of the absorbed Mn atom is about 4.163$\mu_{_{B}}$ on the Au(110)-(1$\times$2) surface. We can see from the left panel of the figure that the spin moment of the absorbed Cr atom is the second largest when including the spin-orbit coupling interactions and non-collinear calculations. The change of magnetic moments per absorbed magnetic atom across the 3d row has been found in many other [*ab initio*]{} calculations such as 3d atoms on Au(001), Ag(001) [@Cabria1] and Cu(001) [@Stepanyuk1] surfaces. Our results are consistent very well with these results although the energy cutoffs of plane-waves are not too large in our calculations.
Both the electronic correlations and the crystal field have significant influence on the orbital moments of the absorbed 3d atoms. If the electronic correlation is stronger than the crystal fields, the orbital moment is large, otherwise small. The absorbed Ti,V,Co atoms have visible orbital moments 0.086$\mu_{_{B}}$, 0.082$\mu_{_{B}}$ and 0.092$\mu_{_{B}}$ on the Au(110)-(1$\times$2) surface, and 0.006$\mu_{_{B}}$, 0.071$\mu_{_{B}}$ and 0.247$\mu_{_{B}}$ on the Au(111) surface respectively. Cabria, et.al obtained larger orbital moment (about 0.5$\mu_{_{B}}$) of Fe and Co atoms absorbed on Au(001) surface using the spin polarization relativistic KKR methods including the orbital polarization term [@Cabria1; @Nonas1]. The usual exchange-correlation potentials (LDA or GGA) underestimate the electronic correlations such as the Coulomb correlation and the orbital polarization, thus the obtained orbital moments in this work are generally small compared with experimental values [@Brooks1; @Solovyev1; @Eriksson1; @Nonas1] although our work includes the spin-orbit coupling interaction. The small values of orbital moments of the absorbed Cr atoms in our calculations are not related to the crystal field but to its electronic structure. In the individual Cr atom, five 3d-electrons half-fill the 3d states, the total orbital moment is very small.
The coordination number of surface atoms is generally smaller than those in bulk. So the magnetism of surface is generally stronger than that in bulk. The supported atoms on the surface are in the environment similar to the surface atoms. Based on the same logic, the supported atoms on the surface have possibly strong magnetism. The lack of the orbital polarization isn’t the obstacle to study the structural influence on the orbital moments. In order to clearly illustrate the effects of the crystal deformations (or the cubic distortions), we compare the results on the deformed surface with that on the prefect Au(110)-(1$\times$2) surface. The absorbed 3d atoms on the prefect surface still modify their positions to reach their stable positions, while the surface atoms are fixed. We find from Fig. \[fig3\] that the surface relaxations generally quench the orbital moments of the absorbed 3d atoms except for the absorbed V and Ti atoms. The changes of the orbital moments in response to the relaxations are closely related to the changes of the depth of the 3d atoms embedded in the trough of the reconstructed Au(110) surface. The absorbed atoms are deeper in the through, their orbital moments are quenched to smaller values due to the stronger crystal field. The absorbed V atom rises above the top row of the trough after the relaxations and its orbital moment enhances, opposes to the orbital quenching for the other 3d atoms except for the absorbed Ti atom. This is due to the weaker crystal field above the surface than that in the trough. The absorbed Ti atom has almost the same height as the top row after the relaxations and its orbital moment slightly increases. Thus our results indicate that the surface relaxation decreases the orbital moments of the absorbed 3d atoms with the excess half-filled 3d states and increases the orbital moments of the absorbed atoms with less half-filled 3d states on Au(110)-(1$\times$2) surface.
The 3d density of states of the absorbed 3d atoms on the two surfaces are shown in Fig. \[fig4\]. The absorbed Ti, V, Co atoms on the Au(110)-(1$\times$2) surface and V, Co atoms on the Au(111) surface have large 3d densities of states at the Fermi energies and large orbital moments too. We get the same arguments on the magnetism of Ni$_{n}$ (n=1-5) chains. Large density of state $N(E_{f})$ at Fermi energy is advantaged to the formation of the orbital moments. The single Ni atom has small magnetic moment, however we can see in section \[Ni\_Chain\] the Ni clusters such as one-dimensional Ni chains have significant magnetic moments on the Au surface. The exchange interactions between Ni atoms prevent from the decay of magnetic moments of Ni atoms.
\[Ni\_Chain\] The magnetism of Ni atomic chains
===============================================
The initial Ni chains are located in the trough along the A-\[1$\bar{1}$0\] direction with nearest-nearby distance about 2.88$\AA$ the same as the nearest-nearby distance of the missed Au chain \[Fig. \[fig1\]\]. The surface slab includes 4 atomic layers with 56 Au atoms and the size of the super-cell is 23.046$\AA$$\times$8.148$\AA$$\times$15.52$\AA$. We have relaxed the initial structures using the conjugate-gradient Methods. The 2$\times$2$\times$1 K-points mesh is used to sampling Brillouin-zone in the corresponding electronic-structure calculations. Based on the relaxed structures, we calculated their electronic structures more exactly using RMM-DIIS algorithm with the (4$\times$4$\times$1) Monkhorst-Pack grids sampling the first Brillouin zone. The other information is the same as in the calculations for single 3d atoms.
We have also calculated the magnetism of free-standing Ni chains by removing all surface atoms but still preserving in the same box cell. There are mirror atomic chains because of periodic boundary condition along three directions. The interaction of free-standing chains with its mirror atomic chains are small and negligible if the box cell is large enough. We have only chosen short chains Ni$_{n}$ where $n$ less than 6 and large box size $23 \AA$ along the direction parallel the chain so that the ends of chains are far away from the boundaries of the box cell. The energetically favorable configuration of one-dimensional free-standing chains has zigzag shapes and not straight line. The Ni$_{n}$ atomic chains in the trough of Au(110)-(1$\times$2) surface are straight. As we optimize the structures of free-standing chains, all atoms are only allowed modified their position along the direction of chains and the optimized free-standing Ni chains still keep a straight line shape. We try to compare the results of surface-support straight chains with those of free-standing straight chains. The quantization axis is still along the C-\[110\] direction in this section.
\[Ni\_chain\_15\] Magnetism of Ni$_{n}$ (n=1-5)
-----------------------------------------------
All Ni atoms in the cell have their initial magnetic moments in our calculations. We calculate the total energies, the spin and orbital magnetic moments of the supported Ni chains when they are magnetized along A-\[1$\bar{1}$0\] direction parallel to the chains or C-\[110\] direction perpendicular to the chains respectively. The magnetization of an atomic chain in this paper means that the magnetic moments for every Ni atom align along the same direction. The values of initial moments of Ni atoms are all the same, however the values after the self-consistent calculations are probably not the same for all Ni atoms. The orientations of magnetic moment may manually change to simulate the different magnetizing directions. In a single self-consistent calculation, the directions of magnetic moments change very small for the ferromagnetic Ni chains although we have used the unconstrained Density Functional calculations (see the table.(\[table2\])).
The chains become short after the relaxations and the average nearest-nearby distance of Ni$_{5}$ chain is about 2.75 $\AA$ shorter than 2.88 $\AA$ that of initial structure \[the insert figure of Fig.\[fig5\](b) and Fig.\[fig1\]\]. We choose two magnetizing directions: A-\[1$\bar{1}$0\] direction parallel to the chain and C-\[110\] direction perpendicular to the chain. The size-dependence magnetic moments are shown in Fig. \[fig5\] (a). From this figure we can see that the spin moments of a single absorbed Ni atom are small, 0.147$\mu_{_{B}}$ for the magnetization along C-\[110\] direction and 0.127$\mu_{_{B}}$ for A-\[1$\bar{1}$0\] direction. The spin moments pre atom of the Ni$_{2}$ chain increase to 0.455 $\mu_{_{B}}$ for A-\[1$\bar{1}$0\] direction and 0.464 $\mu_{_{B}}$ for C-\[110\] directions. The spin moments of Ni$_{3}$ have slightly decrease for C-\[110\] direction but large decrease for A-\[1$\bar{1}$0\] direction. The spin moments decrease slightly and try to keep a constant as the chain length increases above 4. Our results show that the spin moments of Ni atoms on the reconstructed Au(110) surface are smaller than the value 0.675$\mu_{B}$ in crystal Ni having been calculated using the same methods in our work.
The magnetic order is an important aspect of material magnetism. Experimentally the long ferromagnetic order is found for Co chains stabilized above a finite temperature on Pt(997) surface [@Gambardella1]. Above the temperature, the long ferromagnetic order is destroyed and changed into the short ferromagnetic fragments. In our calculations, the short Ni chains are ferromagnetic. The Fig.\[fig6\] shows the ferromagnetic order of the surface supported Ni$_{5}$ chains for the magnetization for C-\[110\] and A-\[1$\bar{1}$0\] directions. The spin-polarizations of surface Gold atoms near the Ni chain are induced by the magnetic Ni atoms, which almost disappear for Gold atoms far from the Ni chain. We have also calculated the anti-ferromagnetic configurations, non-magnetic configurations and other non-collinear magnetic configurations and found their energies higher than that of the ferromagnetic configuration. Thus our calculations show that the Ni chains with ferromagnetic order are most energetically favorable. It is interesting to find from this figure that, when magnetizing the Ni chains parallel to the chain, the orbital moments are larger than that perpendicular to the chains. The spin moments for every Ni atom of the ferromagnetic Ni chain almost don’t change their directions after the Kohn-Sham self-consistent calculations for all three magnetization directions \[table. (\[table2\])\]. The spin moment of the middle atom is smaller than that of other atoms and thus the short Ni chain isn’t in prefect ferromagnetic order. The larger spin-moments for atoms at two ends are because of the smaller coordination number compared with the middle atoms.
The interaction between supported Ni$_{n}$ chain and surface can be measured with the interaction energy $E_{I}=E^{tot}-E^{s}-E^{f}$ where $E^{tot}$ is the total energy, $E^{s}$ the energy of slab with relaxed surface and $E^{f}$ the energy of free-standing chain. We generally calculate the interaction energy per atom, which removes the effects of simple summation. The interaction energies are generally negative. Fig. \[fig5\](b) shows that the interaction energies of the Ni chains with the surface increase with the chain lengths. In our classical molecular dynamics simulation, we have found the oscillation of interaction energy for longer chains induced by lattice misfit [@Fan1]. These results indicate that short Ni chains are more energetically favorable on Au(110)-(1$\times$2) surface. This is consistent with the experiments [@Hitzke1] in which most of Ni chains are shorter than 6. Additionally, we calculate the density of states of the Ni$_{n}$ (n=1-5) chains. From the Fig. \[fig7\], we can see that the densities of states of 3d states of Ni$_{5}$ chain changes very small when the magnetization changes from C-\[110\] direction to A-\[1$\bar{1}$0\] direction and the spin magnetic moment does also change very small \[Fig. \[fig7\]\]. From the figure \[fig5\], the changes of orbital moments are more significant if the magnetizing directions change. Although the orbital moments are underestimated due to the absence of (OP) orbital polarization term, our results still show that magnetic anisotropy are closely related to the orbital degree of freedom. For compact magnetic islands on metal surface such as Co/Pt(111) [@Gambardella4], the orbital moments decrease with increasing island sizes. This is because only atoms at the edge of island have significant contributions to the total orbital moments. However, for one-dimensional island all atoms have significant contribution to the total orbital moments. The orbital moments per atom keep constant as the lengths of chains beyond 4.
\[Ni\_chain\_aniso\] The magnetic anisotropy of Ni$_{5}$ chain
--------------------------------------------------------------
In the above subsection we have chosen two magnetization directions A-\[1$\bar{1}$0\] and C-\[110\]. In this subsection, we study the magnetic anisotropy of the Ni$_{5}$ chain and show that C-\[110\] direction is along the easy magnetization axis. The magnetic anisotropy energies in S$_{1}$ and S$_{2}$ planes are defined as $\delta E_{1}=E^{C}-E^{A}$ and $\delta E_{2}=E^{C}-E^{B}$ respectively. $E^{A}$, $E^{B}$ and $E^{C}$ are the total energies when magnetizing the chain along A-\[1$\bar{1}$0\] ($\theta=0^{\circ}$), B-\[001\] ($\phi=0^{\circ}$)and C-\[110\] ($\theta=90^{\circ}$) directions [@Force1]. The magnetic anisotropy of one-dimensional structure is more prominent than compact structures such as surface and bulk. Fig.\[fig8\] shows the total energies, the spin and orbital moments change with two magnetization angles in above two planes. We find from Fig. \[fig8\](a,b) that there are weak magnetic anisotropy in S$_{1}$ plane and relative strong the magnetic anisotropy in S$_{2}$ plane. The magnetic anisotropy energies are 0.158(meV/atom) and 0.524(meV/atom) respectively in the two planes. The shape anisotropy determines the large difference of magnetic anisotropy energy in two plane because of one-dimension characteristic of mono-atomic chain. In the same plane such as the plane S$_{1}$, the magnetic anisotropy is determined by spin-orbit coupling. The easy magnetization directions in both S$_{1}$ and S$_{2}$ planes are all along C-\[110\] direction ($\theta = 90^{\circ}$ and $\phi = 0^{\circ}$), which is perpendicular to the chain and the surface. The magnetic anisotropy energies can be represented as $K_{0}+K_{1}sin^2(\theta)$ in S$_{1}$ plane and $K'_{0}+K'_{1}sin^2(\phi)$ in S$_{2}$ plane [@Autes1; @Brooks2]. The parameters $K_{0}$=0.158 (meV), $K_{1}$=-0.158 (meV) in S$_{1}$ and $K'_{0}$=0.0 (meV), $K'_{1}$=0.62 (meV) in S$_{2}$ are obtained by fitting the magnetic anisotropy curves using above two functions. The magnetic anisotropy isn’t good fit with $K_{0}+K_{1}sin^2(\theta)$ in S$_{1}$. The underlaying reasons are (1) there isn’t simple cubic symmetry in the region where the one-dimensional atomic chain absorbed; (2) the heterogeneously distributed the nearest nearby distances along the chain induced by the lattice misfit between the chain and the substrate. On the other hand, the value of $K_{1}$ is close to the lowest limit of allowed error in total energy about $\pm$0.1meV, thus more accurate calculations are required. Our results also show that $K_{1}$ (or $K'_{1}$) isn’t equal to the simple $\delta E_{1}=E^{C}-E^{A}$ (or $\delta E_{2}=E^{C}-E^{B}$).
Experimentally, the values of magnetic anisotropy energy are from 1 meV to 10 meV for magnetic clusters on metal surface and from 0.1 mV to 1.0 meV for magnetic film. Based on our calculations, the magnetic anisotropy energies for Ni chains on the reconstructed Au(110) surface are close to the values of magnetic film. This is because the atoms on Ni chain replace the atoms of first layer of original surface and have almost the same coordinate number in their monolayer film. The experiment in Co/Pt(997) has shown that the anisotropy energy about 2 meV for the monatomic Co chains and 0.14 meV for the monolayer Co film on Pt(997) surface [@Gambardella1]. From Fig. \[fig8\](e) we can find that the orbital moments reach the minimum when magnetizing the Ni chain along the easy magnetization direction. Fig. \[fig8\](f) also shows that the orbital moments almost keep a constant when the magnetization angle $\phi$ changes in S$_{2}$ plane. The directions of orbital moments for every Ni atoms on the chain are almost parallel to their spin moments when the magnetizations are along A-\[1$\bar{1}$0\] and C-\[110\] axes, or anti-parallel to their spin moments for the magnetization along B-\[001\] axis. From the curve of magnetic anisotropy energy in $S_{2}$ plane in Fig. \[fig8\](b) it’s most energetically unfavorable when the spin moment of the Ni chain points to B-\[001\] direction ($\phi=90^{\circ}$).
\[Ni\_chain\_free\] The magnetic anisotropy of free-standing Ni$_{5}$ chain and Discussion
------------------------------------------------------------------------------------------
In order to illustrate the surface influence on the magnetic anisotropy, as a comparison, we also calculate the magnetic anisotropy of free-standing Ni$_{5}$ chain in the same box. All calculating details are the same as those of the surface-supported Ni$_{5}$ chain except for removing the surface atoms. The average nearest-neighbor distance is about 2.196 $\AA$ shorter than 2.77 $\AA$ for the surface-support Ni$_{5}$ chain and 2.5 $\AA$ in Ni crystal. The changes of densities of states of 3d states for free-standing Ni chain are that the majority-spin-up densities of states are far from to the Fermi Energy. The larger densities of states at Fermi energy contributed by the minority-spin-down DOS indicate that there are larger spin and orbital moment magnetic moments for free-standing Ni chain.
We only consider the changes of the total energies with the angle $\theta$. From the Fig. \[fig8\] (g) we can see that the easy magnetization direction is now parallel to the chain which is different from perpendicular to the chain for the surface-supported Ni$_{5}$ chain. For free-standing chain we find $K_{0}$=0.0 (meV) and $K_{1}$=8.4 (meV) in S$_{1}$. The anisotropy energy is large compared with the surface-supported Ni$_{5}$ chain. The change of the sign of $K_{1}$ compared with surface-supported chain means that the easy magnetization axis change in S$_{1}$ plane. The change of the easy magnetization axis had been found in other theoretical works such as Co chains on Pd(110) surface [@Dovila1]. Experimentally, the changes of easy magnetization axis are found in the growth of magnetic film on metal surface when the thickness of film is beyond a critical value $d_{c}$. The transitions from in-pane to perpendicular plane for Ni/Cu(100) [@Schulz5] and inversely from perpendicular to in-plane for Fe/Ag(100) [@Qiu5] and Co/Au(111) [@Allenspach5] are found at critical coverage about 5-7 ML. The transitions are close related with the transitions from films with heterogeneously distributed strains to films with homogenously distributed strains once beyond critical thickness. The atomic chains in this work are found in initial stage of film growth and have the easy magnetization axis perpendicular to the surface. Whether the easy magnetization axis changes after a monolayer had completely grown on the surface is interesting topic in our following work.
The spin moments keep almost constant when the magnetization angle changes from $\theta=0^{\circ}$ to $\theta=90^{\circ}$. The spin moment about 1.14$\mu_{B}$ per atom is significantly large than its values in Ni crystal 0.675$\mu_{B}$. The behavior of the orbital moment is similar to the surface-supported Ni$_{5}$ chain, that is, reaches the minimum when perpendicularly magnetizing the chain. The underlaying reason is geometric, that is, the orbital currents (or Molecular Currents) of neighbor atoms have large overlapping and their cancelations are also large which is disadvantageous to form larger orbital moments. The values of orbital moments from 0.2$\mu_{B}$ to 0.6$\mu_{B}$ are larger one power of the values of surface supported Ni atoms. The easy magnetization axis of free-standing Ni chain A-\[1$\bar{1}$0\] is different from the easy magnetization axis of surface-supported chain C-\[110\]. The orbital moment reaches its maximum along the easy magnetization axis A-\[1$\bar{1}$0\], $\theta=0^{\circ}$ for free standing chain, which is consistent with the results for magnetic monolayer, [@Solovyev1; @Bruno1] but reaches the minimum along the easy magnetization C-\[110\] axis $\theta=90^{\circ}$ for the surface-supported chain. Thus our results show that the surface effects change the easy magnetization axis and decrease both the magnetic anisotropy energy and the magnetic moment of free standing chains because of the increase of coordination number for every atom on a surface.
\[conclusion\] Conclusion
==========================
Based on the Density Functional Theory, we have calculated the magnetic anisotropy of the supported Ni chains and free-standing chain. The easy magnetization axis of the supported Ni chain is perpendicular to the chain and surface but parallel to the chain for free standing chain. The different easy magnetization axis for the supported and free standing chain is also found in density functional calculations of other system. Our results indicate the closed relationship between the magnetic anisotropy and orbital moments of the Ni chains. In the environments of crystal, the orbital moment is generally quenched to small value due to the splitting of crystal field. Compared with spin moment, orbital moment is more sensitive to the change of magnetization direction. The orbital quenching changes with the change of magnetized directions. The magnetic anisotropy of materials are determined both by the orbital quenching and spin-orbit coupling. For a single absorbed atom, the surface relaxations generally deform the surface and modify the positions of the absorbed atoms. If the relaxations lift the absorbed atom out of the trough on the constructed Au(110) surface, the orbital moments enhance due to weaker crystal fields and if the relaxations make the absorbed atoms deeper embedded in the trough the orbital moments quench to smaller values due to the stronger crystal fields.
[**Acknowledgement**]{}
The one of authors (W. Fan) is greatly indebted to Prof. Q. Q. Zheng, Dr. J. L.Wang, Prof. L. J. Zou and Prof. Z. Zeng for their helpful discussions. This work run on computers in Center for computational science, Hefei Institutes of Physical Science and on LSSC in Institute of Computational Mathematics and Scientific/Engineering Computing, Chinese Academy of Sciences. X. G. Gong is additionally supported by the National Science Foundation of China, the special funds for major state basic research and CAS projects and Knowledge Innovation Program of Chinese Academy of Sciences under KJCX2-SW-W11.
[99]{} H. J. Elmers, J. Hauschild, H. Höche, U. Gradmann, H. Bethge, D. Heuer, and U. Köhler, Phys. Rev. Lett. [**73**]{} (1994) 898. J. Shen, R. Skomski, M. Klaua, H. Jenniches, S. S. Manoharan, and J. Kirschner, Phys. Rev. B [**56**]{} (1997) 2340. J. Hauschild, H. J. Elmers, U. Gradmann, Phys. Rev. B [**57**]{} (1998) R677. M. Pratzer, H. J. Elmers, M. Bode, O. Pietzsch, A. Kubetzka, and R. Wiesendanger, Phys. Rev. Lett. [**87**]{} (2001) 127201. P. Gambardella, A. Dallmeyer, K. Maiti, M. C. Malagoli, W. Eberdardt, K. Kern, and C. Carbone, Nature (London) [**416**]{} (2002) 301. P. Gambardella, M.B lanc, L. Bürgi, K. Kuhnke, and K. Kern, Surf. Sci. [**449**]{} (2000) 93. P. Gambardella, M. Blanc, K. Kuhnke, and K. Kern, Phys. Rev. B [**61**]{} (2000) 2254. Dongqi Li, Chengtao Yu, J. Pearson, and S. D. Bader, Phys. Rev. B [**66**]{} (2002) 020404. C. Boeglin, S. Stanescu, J. P. Deville, P. Ohresser, and N. B. Brookes, Phys. Rev. B [**66**]{} (2002) 014439. H. Dreyssé, C. Demangeat, Surf. Sci. Rep. [**28**]{} (1997) 65. V. Bellini, N. Papanikolaou, R. Zeller, and P. H. Dederichs, Phys. Rev. B [**64**]{} (2001) 094403. K. Wildberger, V. S. Stepanyuk, P. Lang, R. Zeller, and P. H. Dederichs, Phys. Rev. Lett. [**75**]{} (1995) 509. B. Lazarovits, L. Szunyogh, P. Weinberger, B. Ujfalussy, Phys. Rev. B [**68**]{} (2003) 024433. M. Eisenbach, B. L. Györffy, G. M. Stocks, and B. Újfalussy, Phys. Rev. B [**65**]{} (2002) 144424. M. Komelj, C. Ederer, J. W. Davenport, and M. Fähnle, Phys. Rev. B [**66**]{} (2002) 140407. D. Spišák and J. Hafner, Phys. Rev. B[**65**]{} (2002) 235405. D. Spišák and J. Hafner, Phys. Rev. B[**67**]{} (2003) 214416. J. Dorantes-Dávila and G. M. Pastor, Phys. Rev. Lett, [**81**]{} (1998) 208. R. Druzinic and W. Hübner, Phys. Rev. B [**55**]{} (1997) 347. M. S. S. Brooks, P. J. Kelly, Phys. Rev. Lett. [**51**]{} (1983) 1708 . O. Eriksson, M. S. S. Brooks, B. Johansson, Phys. Rev. B [**41**]{} (1990) 7311. I. V. Solovyev, A. I. Liechtenstein, K. Terakura, Phys. Rev. Lett. [**80**]{} (1998) 5758. O. Hjortstarm, K. Baberschke, J. M. Wills, B. Johansson, and O. Eriksson, Phys. Rev. B [**55**]{} (1997) 15026. I. V. Solovyev, Phys. Rev. Lett. [**95**]{} (2005) 267205. (or arXiv:cond-mat/0510100) F. Fuchs, J. Furthmüller, F. Bechstedt, M. Shishkin, and G. Kresse, Preprint-arXiv:cond-mat/0604447, 2006 Harvey Brooks, Phys. Rev. [**58**]{} (1940) 909. A. Hitzke, M. B. Hugenschmidt, R. J. Behm, Surface Science, [**389**]{} (1997) 8. W. Fan, X. G. Gong, and W. M. Lau, Phys. Rev. B [**66**]{} (2002) 115418. K. D. Gross, D. Riegel, and R. Zeller, Phys. Rev. Lett. [**63**]{} (1989) 1176. K. D. Gross, D. Riegel, and R. Zeller, Phys. Rev. Lett. [**65**]{} (1990) 3044. M. C. Hanf, C. Pirri, J. C. Peruchetti, D. Bolmont, and G. Gewinner, Phys. Rev. B[**36**]{} (1987) 4487. D. Riegel, L. Büermann, K. D. Gross, M. Luszik-Bhadra, and S. N. Mishra, Phys. Rev. Lett. [**62**]{} (1989) 316. J. E. Ortega and F. J. Himpsel Phys. Rev. B[**47**]{} (1993) 16441. P. Hohenberg and W. Kohn, Phys. Rev. [**136**]{} (1964) B864. W. Kohn and L. J. Sham, Phys. Rev. [**140**]{} (1965) A1133. P. E. Blöchl, Phys. Rev. B[**50**]{} (1994) 17953. J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. [**77**]{} (1996) 3865. G. Kresse and J. Furthmuller, Comput. Mater. Sci. [**6**]{} (1996) 15. G. Kresse and D. Joubert, Phys. Rev. B [**59**]{} (1999) 1758. D. Hobbs, G. Kesse, J. Hafner, Phys. Rev. B [**62**]{} (2000) 11556; M. Marsman, [*Magnetism*]{}, the talks and hands-on sessions of VASP Workshop, Vienna (2003). D. Hobbs, J. Hafner, D. Spišák Phys. Rev. B [**68**]{} (2003) 014407. L. M. Sandratskii, Adances in Physics, [**47**]{} (1998) 91. The energy cutoffs for plane wave adopt the values distributed by VASP code (CMS group at University of Vienna). B. Nonas, I. Cabria, R. Zeller, and P. H. Dederichs, T. Huhne and H. Ebert, Phys. Rev. Lett. [**86**]{} (2001) 2146. I. Cabria, B. Noas, R. Zeller, and P. H. Dederichs, Phys. Rev. B[**65**]{} (2002) 054414. V. S. Stepanyuk, A. N. Baranov, W. Hergert, and P. Bruno, Phys. Rev. B[**68**]{} (2003) 205422. P. Gambardella, S. Ruspoi, T. Cren, N. Weiss, H. Brune, C. R. Physique, [**6**]{} (2005) 75. B. Schulz and K. Baberschke, Phys. Rev. B [**50**]{} (1994) 13467. Z. Q. Niu, J. Pearson, and S. D. Bader, Phys. Rev. Lett [**70**]{} (1993) 1006. R. Allenspach, M. Stampanoni, and A. Bischof, Phys. Rev. Lett [**65**]{} (1990) 3344. Patrick Bruno, Phys. Rev. B [**39**]{} (1989) 865. Gabriel Autès, Cyrille Barreteau, Daniel Spanjaard, and Marie-Catherine Desjonquères, Preprint:arXiv:cond-mat/0603121, (2006). The magnetic anisotropy energy (MAE) also be measured using the difference of the sums of Kohn-Sham eigenvalues which have been used in LMTO and LAPW program with (ASA) Atom sphere Approximation for monatomic systems. T. Burkert, O. Eriksson, P. James, S.I. Simak, B. Johansson, and L. Nordström, Phys. Rev. B [**69**]{} (1985) 104426. M. Weinert, R.E. Watson, and J.W. Davenport, Phys. Rev B [**32**]{} (1985) 2115. G. H. O. Daalderop, P. J. Kelly, and M. F. H. Schuurmans, Phys. Rev. B [**41**]{} (1990) 11919.
Ti V Cr Mn Fe Co Ni Cu Au
------ ------- ------- ------- ------- ------- ------- ------- ------- -------
WSR 1.323 1.323 1.323 1.323 1.302 1.302 1.286 1.312 1.503
Ecut 178.4 192.6 227.1 269.9 267.9 268.0 269.6 273.2 229.9
: \[table1\]The Wigner Seitz Radius-WSR ($\AA$) and Energy-Cutoff of plane wave-Ecut (eV) for PAW potentials of 3d transition metal T$_{3d}$=V, Ti, Cr, Mn, Fe, Co, Ni, Cu. [@VaspPaw] The real energy cutoff for T$_{3d}$/Au is Ecut(T$_{3d}$/Au)= Max \[Ecut(T$_{3d}$),Ecut(Au)\].
3 cm
A-\[1$\bar{1}$0\] B-\[001\] C-\[110\]
---------- -- ------------------- --------- --------- -- ----------- --------- --------- -- ----------- --------- --------- --
M$_{A}$ M$_{B}$ M$_{C}$ M$_{A}$ M$_{B}$ M$_{C}$ M$_{A}$ M$_{B}$ M$_{C}$
Ni$_{1}$ 0.434 0.0 -0.006 0.0 0.411 0.0 0.017 0.0 0.430
Ni$_{2}$ 0.449 0.0 0.006 0.0 0.426 0.0 -0.012 0.0 0.449
Ni$_{3}$ 0.352 0.0 0.000 0.0 0.323 0.0 0.001 0.0 0.366
Ni$_{4}$ 0.449 0.0 -0.007 0.0 0.452 0.0 0.015 0.0 0.452
Ni$_{5}$ 0.437 0.0 0.005 0.0 0.446 0.0 -0.015 0.0 0.432
L$_{A}$ L$_{B}$ L$_{C}$ L$_{A}$ L$_{B}$ L$_{C}$ L$_{A}$ L$_{B}$ L$_{C}$
Ni$_{1}$ 0.105 0.0 0.000 0.0 -0.085 0.0 0.009 0.0 0.080
Ni$_{2}$ 0.097 0.0 -0.010 0.0 -0.082 0.0 -0.012 0.0 0.093
Ni$_{3}$ 0.092 0.0 0.000 0.0 -0.065 0.0 0.001 0.0 0.060
Ni$_{4}$ 0.097 0.0 0.010 0.0 -0.088 0.0 -0.012 0.0 0.096
Ni$_{5}$ 0.105 0.0 -0.001 0.0 -0.092 0.0 -0.009 0.0 0.079
: \[table2\]The spin and orbital magnetic moments (Bohr) for every atoms of Ni$_{5}$ chain when the magnetization directions are along A-\[1$\bar{1}$0\] ,B-\[001\] and C-\[110\] respectively. This table shows that the directions of spin magnetic moments are almost unchanged in Kohn-Sham self-consistent loop although we have used the unconstrained DFT calculations.
![\[fig1\] The schematic diagram of a Ni chain supported on the Au(110)-(1$\times$2) surface. The trough of the reconstructed surface is along the A-\[1$\bar{1}$0\] direction. The red spheres are the absorbed Ni atoms in the trough. The yellow spheres are the Au atoms of the surface. The plane including both C-\[110\] axis and A-\[1$\bar{1}$0\] axis is named as S$_{1}$, the plane including both C-\[110\] axis and B-\[001\] axis as S$_{2}$. The rotating angle from any vector in S$_{1}$ to A-\[1$\bar{1}$0\] axis labels as $\theta$ and the rotating angle from any vector in S$_{2}$ to C-\[110\] axis labels as $\phi$. $\theta$ and $\phi$ are the magnetization angles in the two planes.](fig1.eps)
![\[fig3\] The effects of structure-relaxation on orbital moments of the absorbed 3d transition metal atoms. (a). The heights of top-row and absorbed 3d atoms before the full-structure relaxations. The stable positions are different for different 3d atoms, which are obtained by the single-atom optimizations on the unrelaxed surface. (b). The heights of top row and absorbed 3d atoms after having done full-structure relaxations. The V atom is higher above the top row after the full-relaxation. (c). The schematic figure defines the heights of top row and absorbed atoms using two arrows pointing to the bottom of the trough. (d). The orbital moments change before and after the full-relaxations.](fig3.eps)
![\[fig4\] The 3d densities of state (DOS) for the 3d atoms absorbed on Au(110)-(1$\times$2) (solid lines) and Au(111) (dot lines ) surface.](fig4.eps)
\[fig4\]
![ \[fig5\] (a) The spin $^{A}$M, $^{C}$M and orbital $^{A}$L, $^{C}$L moments change with the length of Ni chains. $^{A}$M and $^{C}$M aren’t the A and B components of the spin moment $\vec{M}$. They are the spin magnetic moments when the magnetization along the A-\[1$\bar{1}$0\] and C-\[110\] direction and are obtained from different self-consistent calculations. $^{A}$L and $^{A}$L have the same meaning for orbital moments. (b) Interaction energies between the Ni chains and the surface. The larger absolute values indicate the stronger interaction with the surface. ](fig5.eps)
![\[fig6\]The vector graph of the spin magnetic moments of the Ni chain which illustrates the ferromagnetic order of the Ni$_{5}$ chain based on our calculations. Especially, (a) and (b) are corresponding to the magnetization along C-\[110\] and A-\[1$\bar{1}$0\] direction respectively. The smaller spin magnetic moments of Au atoms and the orbital magnetic moments of Ni atoms are multiplied by 40 and 2 respectively to show them with the larger spin moments of Ni atoms together in the same figure clearly.](fig6.eps)
![ \[fig7\] The projected DOS of 3d states of the supported Ni$_{5}$ chain along the directions of magnetization when the chain is magnetized along different directions respectively (the solid lines for the C-\[110\] direction and the circles for the A-\[1$\bar{1}$0\] direction). The spin-up (or spin-down) means along the positive (or negative) direction of two magnetization directions. This shows that spin degree of freedom is less relevant to magneto-crystalline anisotropy. The dots line is the DOS for free-standing Ni chain when magnetizing it along C-\[110\] direction.](fig7.eps)
![ \[fig8\] The anisotropy energies (a,b), the spin (c,d) and orbital moment (e,f) of the supported Ni$_{5}$ chain change with the magnetization angles $\theta$ and $\phi$ in the planes S$_{1}$ and S$_{2}$ defined in above. The easy magnetization direction is perpendicular to the surface $\theta=90^{\circ}$ and $\phi=0^{\circ}$. The anisotropy energies (g), spin moment (h) and orbital moment (i) of the free standing Ni$_{5}$ chain change with the magnetization angles $\theta$ in the planes S$_{1}$. The easy magnetization direction is parallel to the chain $\theta=0^{\circ}$. Error bars in (a,b) represent the lowest limit of allowed error in total energy about $\pm$0.1meV. The magnetic anisotropy energy curves are fitting using function $K_{0}+K_{1}sin^2(\theta)$ for (a,g) and $K_{0}+K_{1}sin^2(\phi)$ for (b)(red dash-lines). ](fig8.eps)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We introduce a natural correspondence between artinian monomial almost complete intersections in three variables and punctured hexagonal regions. We use this correspondence to investigate the algebras for the presence of the weak Lefschetz property. In particular, we relate the field characteristics in which such an algebra fails to have the weak Lefschetz property to the prime divisors of the enumeration of signed lozenge tilings of the associated punctured hexagonal region. On the one side this allows us to establish the weak Lefschetz property in many new cases. On the other side we can determine some of the prime divisors of the enumerations by means of an algebraic argument.
For numerous classes of punctured hexagonal regions we find closed formulae for the enumerations of signed lozenge tilings, and thus the field characteristics in which the associated algebras fail to the have the weak Lefschetz property. Further, we offer a conjecture for a closed formula for the enumerations of signed lozenge tilings of symmetric punctured hexagonal regions. These formulae are exploited to lend further evidence to a conjecture by Migliore, Miró-Roig, and the second author that classifies the [*level*]{} artinian monomial almost complete intersections in three variables that have the weak Lefschetz property in characteristic zero. Moreover, the formulae are used to generate families of algebras which never, or always, have the weak Lefschetz property, regardless of field characteristic. Finally, we determine (in one case, depending on the presence of the weak Lefschetz property) the splitting type of the syzygy bundle of an artinian monomial almost complete intersection in three variables, when the characteristic of the base field is zero.
Our results convey an intriguing interplay between problems in algebra, combinatorics, and algebraic geometry, which raises new questions and deserves further investigation.
address: 'Department of Mathematics, University of Kentucky, 715 Patterson Office Tower, Lexington, KY 40506-0027, USA'
author:
- 'David Cook II${}^{\star}$, Uwe Nagel'
title: Enumerations deciding the weak Lefschetz property
---
[[ [=<10 0]{}:[=<10 0]{}, [=<10 0]{}. JanuaryFebruaryMarchAprilMayJuneJulyAugustSeptemberOctoberNovemberDecember ]{}]{}
[^1]
Introduction {#sec:intro}
============
The starting point of this paper has been an intriguing conjecture in [@MMN] on the weak Lefschetz property of certain algebras. Though the presence of this property implies considerable restrictions on invariants of the algebra, many algebras are expected to have the weak Lefschetz property. However, establishing this property is often rather difficult. In this paper we make progress on the above conjecture and illustrate the depth of the problem by considering a larger class of algebras and relating the problem to [*a priori*]{} seemingly unrelated questions in combinatorics and algebraic geometry. This builds on the work of many authors (e.g., [@BK], [@CGJL], [@CN], [@LZ], and [@MMN]).
Throughout this work we consider in particular the question of how the weak Lefschetz property of a certain $K$-algebra $A$ depends on the characteristic of the field $K$. We begin by relating the algebra $A$ to two square integer matrices, $N$ and $Z$, where the entries of $N$ are binomial coefficients and $Z$ is a zero-one matrix. We show that $A$ has the weak Lefschetz property if and only if the determinant of either of these matrices does not vanish modulo the characteristic of $K$. Next, we establish that the determinant of $N$ enumerates signed lozenge tilings of a punctured hexagonal region and that the determinant of $Z$ enumerates signed perfect matchings of a bipartite graph associated to the same punctured hexagonal region. The relation to the weak Lefschetz property implies that both determinants have the same prime divisors; in fact, we show that their absolute values are the same by using combinatorial arguments. Finally, we show that in certain cases deciding the presence of the weak Lefschetz property is equivalent to determining the splitting type of some semistable rank three vector bundles on the projective plane.
We now describe the contents of this paper in more detail. Let $R = K[x_1, \ldots, x_n]$ be the standard graded $n$-variate polynomial ring over the infinite field $K$, and let $A$ be a standard graded $K$-algebra over $R$. We say $A$ is [*artinian*]{} if $A$ is finite dimensional as a vector space over $K$. Further, an artinian $K$-algebra $A$ is said to have the [*weak Lefschetz property*]{} if there exists a linear form $\ell \in [A]_1$ such that, for all integers $d$, the multiplication map $\times \ell: [A]_d \rightarrow [A]_{d+1}$ has maximal rank, that is, the map is injective or surjective. Such a linear form is called a [*Lefschetz element*]{} of $A$.
The weak Lefschetz property has been studied extensively for many reasons, especially for the relation to the Hilbert function (see, e.g., [@BMMNZ], [@HMNW], [@MZ], and [@ZZ]). A convenient way to encode the Hilbert function of an artinian $K$-algebra $A$ is the [*$h$-vector*]{}, a finite sequence $h(A) = (h_0, \ldots, h_e)$ of positive integers $h_i = \dim_K{[A]_i}$. Using this notation, one immediate consequence ([@HMNW Remark 3.3]) of $A$ having the weak Lefschetz property is that the $h$-vector of $A$ is [*strictly unimodal*]{}. Further, the positive part of the first difference of $h(A)$ is $h(A/\ell A)$, where $\ell$ is any Lefschetz element of $A$.
The weak Lefschetz property is known to be subtle to both deformations (see, e.g., [@CN-2010], [@Mi], and [@MMN]) but also to field characteristic. The latter, considering the weak Lefschetz property in positive characteristic, is an exciting and active direction of research. Migliore, Miró-Roig, and the second author [@MMN], as well as Zanello and Zylinski [@ZZ], began explorations into the connection between the weak Lefschetz property and positive characteristic, and also posed several interesting questions.
In [@CN], the authors found a connection between certain families of level artinian monomial almost complete intersections and lozenge tilings of hexagons; independently, Li and Zanello [@LZ] found a similar connection for artinian monomial complete intersections (see also Corollary \[cor:ci\]). However, both were without combinatorial bijection until one was found by Chen, Guo, Jin, and Liu [@CGJL]; Boyle, Migliore, and Zanello [@BMZ] have pushed this connection further. Brenner and Kaid [@BK-2010] also consider artinian monomial complete intersections in three variables with generators all of the same degree. We also note that in their study of pure $O$-sequences Boij, Migliore, Miró-Roig, the second author, and Zanello [@BMMNZ] have explored the relation between the weak Lefschetz property and pure $O$-sequences.
In this paper we extend the connection found by Chen, Guo, Jin, and Liu to a connection between artinian monomial almost complete intersections in three variables and lozenge tilings of more general regions that we call punctured hexagons. In Section \[sec:toolchain\] we gather a few useful tools for dealing with the weak Lefschetz property. In Section \[sec:aci\] we introduce the algebras we are interested in: artinian monomial almost complete intersections in three variables. If the syzygy bundle is not semistable, then the algebra has the weak Lefschetz property in characteristic zero ([@BK]). Thus we focus on the algebras that have semistable syzygy bundles, which we classify numerically (Proposition \[pro:semistable\]). Then we prove that such an algebra has the weak Lefschetz property if and only if a particular map between the peak homogeneous components of the algebra is a bijection (Corollary \[cor:one-map\]). Using this, we show that to each of the studied algebras $A$ we can associate a zero-one matrix $Z_A$ such that $A$ has the weak Lefschetz property in positive characteristic $p$ if and only if $p$ is not a prime divisor of the determinant of $Z_A$ (Proposition \[pro:wlp-zero-one\]). We also describe a matrix $N_A$ with binomial entries that also has the analogous property (Proposition \[pro:wlp-binom\]). Moreover, we demonstrate that a rather simple algebraic argument can be used to determine some of the prime divisors of the determinants for both $Z_A$ and $N_A$ (Proposition \[pro:wlp-p\]).
In Section \[sec:ph\] we organise the monomials generating the peak homogeneous components of such an algebra in a plane. It turns out that the monomials fill a punctured hexagon (Theorem \[thm:interlace-amaci\]). Using the well-known bijection between lozenge tilings and non-intersecting lattice paths, and the Lindström-Gessel-Viennot theorem ([@GV], [@GV-1989], [@Li]) on non-intersecting lattice paths, we show that the determinant of the binomial matrix $N_A$ is the enumeration of the signed lozenge tilings of the punctured hexagon, up to sign (Theorem \[thm:nilp-matrix\]). Furthermore, using another well-known bijection between lozenge tilings and perfect matchings (see, e.g., [@Ku]), we argue that the determinant of the zero-one matrix $Z_A$ is an enumeration of the signed perfect matchings of the associated bipartite graph, up to sign (Theorem \[thm:bip-matrix\]).
In Section \[sec:signs\] we use the aforementioned connections to show that the determinant of the zero-one matrix $Z_A$ and the binomial matrix $N_A$ are the same, up to sign (Theorem \[thm:det-Z-N\]). Moreover, in a special case of Kasteleyn’s theorem [@Ka] about enumerating perfect matchings, when the puncture has an even side-length, then the determinant and the permanent of $Z_A$ are also the same, up to sign (Corollary \[cor:det-Z-per-Z\]).
In Section \[sec:det\] we prove first that the determinant of $N_A$ is non-zero when the puncture is of even side-length (Theorem \[thm:M-even\]), thus establishing the weak Lefschetz property in many new cases. We then find closed formulae for the determinants when the puncture is trivial (Proposition \[pro:M-zero\]), when any one side of the hexagonal region has length zero (Proposition \[pro:C-zero\]), when a vertex of the puncture touches one of the sides of the region (Proposition \[pro:C-maximal\]), and when a side of the puncture touches one of the sides of the region (Proposition \[pro:gamma-zero\]). We close with a complete description of when the region is symmetric. In particular, we show that when certain parity conditions hold the determinant is zero (Proposition \[pro:symmetry-zero\]) and we provide a conjecture for a closed formula of the determinant when the same parity conditions fail (Conjecture \[con:symmetry\]).
In Section \[sec:central\] we explore two different ways to centralise the puncture. We call the puncture [*axis-central*]{} when it is central along each of the three axes, independently. Using very involved computations, Ciucu, Eisenkölbl, Krattenthaler, and Zare [@CEKZ] found closed formulae for the enumerations and signed enumerations of regions with an axis-central puncture; therein axis-central is called simply “central”. We use these closed formulae to describe the permanents of the zero-one matrices $Z_A$ (Corollary \[cor:axis-central-perm-Z\]) and the determinants of both matrices (Corollary \[cor:axis-central-det-Z\]), $Z_A$ and $N_A$, when the puncture is axis-central. We call the puncture [*gravity-central*]{} when its vertices are equidistant from the sides of the containing hexagon; this condition is equivalent to the associated algebra being level, that is, its socle is concentrated in one degree. Using this observation we provide further evidence for a conjecture by Migliore, Miró-Roig, and the second author [@MMN] about the presence of the weak Lefschetz property for level artinian monomial almost complete intersections in characteristic zero (Proposition \[pro:level-wlp\]).
In Section \[sec:interesting\] we describe a method, for any positive integer $n$, to generate a subfamily of algebras whose associated matrices have determinant $n$ (Proposition \[pro:det-n\]). From this we generate a subfamily of algebras which [*always*]{} have the weak Lefschetz property, regardless of the field characteristic (Corollary \[cor:C-zero-unique\]); we also describe a different subfamily of algebras which always have the weak Lefschetz property (Proposition \[pro:gamma-zero-unique\]). Moreover, we describe the unique algebras which retain certain properties yet have minimal multiplicity (Example \[exa:minimal-multiplicity\]).
In Section \[sec:splitting-type\] we explicitly determine (in one case, depending on the presence of the weak Lefschetz property) the splitting type of all artinian monomial almost complete intersections. In particular, we consider separately the cases when the syzygy bundle is non-semistable (Proposition \[pro:st-nss\]) and semistable (Propositions \[pro:st-nmod3\] and \[pro:st-mod3\]). Moreover, in the case of ideals associated to punctured hexagons, we relate the weak Lefschetz property to a number of other problems in algebra, combinatorics, and algebraic geometry (Theorem \[thm:equiv\]).
Finally, in Appendix \[sec:hyper-calculus\] we provide a technique, a “picture-calculus”, for working with hyperfactorials, a basic unit for the aforementioned closed formulae. We demonstrate that several nice polynomials can be written as ratios of products of hyperfactorials (Proposition \[pro:hyper-f\] and Corollary \[cor:hyper-eo-f\]). Further, this shows that MacMahon’s formula for the number of lozenge tilings of a (non-punctured) hexagon is a polynomial in one of the side-lengths when the other two are fixed (Corollary \[cor:mac-poly\]).
Compiling the tool-chain {#sec:toolchain}
========================
Let $R = K[x_1, \ldots, x_n]$ be the standard graded $n$-variate polynomial ring over the infinite field $K$, and let $A$ be an artinian standard graded $K$-algebra over $R$. Then the minimal free resolution of $A$ ends with the free module $\bigoplus_{i=1}^{m} R(-t_i)^{r_i},$ where $0 < t_1 < \cdots < t_m$ and $0 < r_i$ for all $i$. In this case, $A$ is called [*level*]{} if $m = 1$, the [*socle degrees*]{} of $A$ are $t_i - n$, for all $i$, and the [*socle type*]{} of $A$ is the sum $\sum_{i=1}^{m} t_i$.
We recall that once multiplication by a general linear form is surjective, then it remains surjective.
[[@MMN Proposition 2.1(a)]]{} \[pro:surj\] Let $A = R/I$ be an artinian standard graded $K$-algebra, and let $\ell$ be a general linear form. If the map $\times \ell: [A]_{d} \rightarrow [A]_{d+1}$ is surjective, then $\times \ell: [A]_{d+1} \rightarrow [A]_{d+2}$ is surjective.
This generalises to modules generated in degrees that are sufficiently small.
\[lem:mod-surj\] Let $M$ be an $R$-module generated in degrees bounded by $e$, and let $\ell$ be a general linear form. If the map $\times\ell: [M]_d \rightarrow [M]_{d+1}$ is surjective, and $d \geq e$, then the map $\times\ell: [M]_{d+1} \rightarrow [M]_{d+2}$ is surjective.
Consider the sequence $$[M]_d \stackrel{\times\ell}{\longrightarrow} [M]_{d+1} \rightarrow [M/\ell M]_{d+1} \rightarrow 0.$$ Notice the first map is surjective if and only if $[M/\ell M]_{d+1} = 0.$ By assumption the map is surjective, so $[M/\ell M]_{d+1} = 0$. Hence $[M/\ell M]_{d+2}$ is zero unless there is a generator of $M$ with degree beyond $d$. However, the assumption is that no generators exist with degree beyond $d.$
From this we get a result analogous to [@MMN Proposition 2.1(b)] for non-level algebras.
\[pro:inj\] Let $A = R/I$ be an artinian standard graded $K$-algebra, and let $\ell$ be a general linear form. If the map $\times \ell: [A]_{d-1} \rightarrow [A]_{d}$ is injective, and $d$ is no greater than the smallest socle degree of $A$, then $\times \ell: [A]_{d-2} \rightarrow [A]_{d-1}$ is injective.
The $K$-dual of $A$, $M$, is a shift of the canonical module of $A$ and is generated in degrees that are a linear shift of the socle degrees of $A$. Consider now the map $\times \ell: [M]_i \rightarrow [M]_{i+1}$. Using Lemma \[lem:mod-surj\] we see that once $i$ is at least as large as the largest degree in which $M$ is generated, and the map is surjective, then the map is surjective thereafter. The result then follows by duality.
Further recall that a monomial algebra has the weak Lefschetz property exactly when the sum of the variables is a Lefschetz element.
[[@MMN Proposition 2.2]]{} \[pro:mono\] Let $A = R/I$ be an artinian standard graded $K$-algebra with $I$ generated by monomials. Then $A$ has the weak Lefschetz property if and only if $x_1 + \cdots + x_n$ is a Lefschetz element of $A$.
Hence, the weak Lefschetz property can be decided for monomial ideals, in a small number of cases, by simple invariants. The following lemma is a generalisation of [@LZ Proposition 3.7].
\[lem:wlp-p\] Let $A = R/I$ be an artinian standard graded $K$-algebra with $I$ generated by monomials. Suppose that $a$ is the least positive integer such that $x_i^a \in I$, for $1 \leq i \leq n$, and suppose that the Hilbert function of $R/I$ weakly increases to degree $s+1$. Then, for any positive prime $p$ such that $a \leq p^m \leq s+1$ for some positive integer $m$, $A$ fails to have the weak Lefschetz property in characteristic $p$.
By Proposition \[pro:mono\], we need only consider $\ell = x_1 + \cdots + x_n$. Suppose the characteristic of $K$ is $p$, then by the Frobenius endomorphism $\ell \cdot \ell^{p^m-1} = \ell^{p^m} = x_1^{p^m} + \cdots + x_n^{p^m}$. Moreover, as $a \leq p^m$, then $\ell^{p^m} = 0$ in $A$ while $\ell \neq 0$ in $A$. Hence $\times\ell^{p^m-1}: [A]_{1} \rightarrow [A]_{p^m}$ is not injective and thus $A$ does not have the weak Lefschetz property.
Further, for monomial ideals, if the weak Lefschetz property holds in characteristic zero, then it holds for almost every characteristic.
\[lem:wlp-ae\] Let $I$ be an artinian monomial ideal in $R$. If $R/I$ has the weak Lefschetz property when $\operatorname{char}{K} = 0$, then $R/I$ has the weak Lefschetz property for $\operatorname{char}{K}$ sufficiently large.
By Proposition \[pro:mono\], we need only consider $\ell = x_1 + \cdots + x_n$. As $R/I$ is artinian, then there are finitely many maps that need to be checked for the maximal rank property, and this in turn implies finitely many determinants that need to be computed. Further, because of the form of $\ell$, the matrices in question are all zero-one matrices. Thus, the determinants to be checked are integers. Simply let $p$ be the smallest prime larger than all prime divisors of the determinants, then the determinants are all non-zero modulo $p$ and so $R/I$ has the weak Lefschetz property if $\operatorname{char}{K} \geq p$.
And (pseudo-)conversely, again for monomial ideals, if the weak Lefschetz property holds in some positive characteristic, then it holds for characteristic zero.
\[lem:wlp-p-zero\] Let $I$ be an artinian monomial ideal in $R$. If $R/I$ has the weak Lefschetz property when $\operatorname{char}{K} = p > 0$, then $R/I$ has the weak Lefschetz property for $\operatorname{char}{K} = 0$.
The proof is the same as that of Lemma \[lem:wlp-ae\] except we notice that if an integer $d$ is non-zero modulo a prime $p$, then $d$ is not zero.
Last, we note that any artinian ideal in two variables has the weak Lefschetz property. This was proven for characteristic zero in [@HMNW Proposition 4.4] and then for arbitrary characteristic in [@MZ Corollary 7], though it was not specifically stated therein, as noted in [@LZ Remark 2.6]. We provide a brief, direct proof of this fact to illustrate the weak Lefschetz property. Unfortunately, the simplicity of this proof fails in three variables, even for monomial ideals.
\[pro:2-wlp\] Let $R = K[x,y]$, where $K$ is an infinite field with [*arbitrary*]{} characteristic. Every artinian algebra in $R$ has the weak Lefschetz property.
Assume $I = (g_1, \ldots, g_t)$ and the given generators are minimal. Let $s = \min\{\deg{g_i} {\; | \;}1 \leq i \leq t\}$. Then $h(R/I)$, the $h$-vector of $R/I$, strictly increases by one from $h_0$ to $h_{s-1}$ and $h_{s-1} \geq h_s$, thus the positive part of the first difference of $h(R/I)$, ${\Delta^{\hspace{-0.333em} +}\hspace{-0.167em}}h(R/I)$, is $s$ ones. Moreover, for a general linear form $\ell \in R$, $R/(I,\ell) \cong K[x]/J$ where $J = (x^s)$ so $h(R/(I, \ell))$ is $s$ ones, that is, ${\Delta^{\hspace{-0.333em} +}\hspace{-0.167em}}h(R/I) = h(R/(I,\ell))$. Hence $R/I$ has the weak Lefschetz property with Lefschetz element $\ell$.
Almost complete intersections {#sec:aci}
=============================
Here we restrict to artinian monomial almost complete intersections in three variables. These are the ideals discussed in [@BK Corollary 7.3] and [@MMN Section 6].
Let $K$ be an infinite field, and consider the ideal $$I_{a,b,c,\alpha,\beta,\gamma} = (x^a, y^b, z^c, x^\alpha y^\beta z^\gamma)$$ in $R = K[x,y,z]$, where $0 \leq \alpha < a, 0 \leq \beta < b,$ and $0 \leq \gamma < c$. If $\alpha = \beta = \gamma = 0$, then we define $I_{a,b,c,0,0,0}$ to be $(x^a, y^b, z^c)$ which is a complete intersection and is studied extensively in [@LZ] and [@CGJL]. Assume at most one of $\alpha, \beta,$ and $\gamma$ is zero.
[[@MMN Proposition 6.1]]{} \[pro:amaci-props\] Let $I = I_{a,b,c,\alpha,\beta,\gamma}$ be defined as above. Assume, without loss of generality, that $0 \leq \alpha \leq \beta \leq \gamma$.
1. If $\alpha = 0$, then $R/I$ has socle type $2$ with socle degrees $a+\beta+c-3$ and $a+b+\gamma -3$; thus $R/I$ is level if and only if $b - \beta = c - \gamma$.
2. If $\alpha > 0$, then $R/I$ has socle type $3$ with socle degrees $\alpha + b + c -3$, $a+\beta+c-3$, and $a+b+\gamma -3$; thus $R/I$ is level if and only if $a - \alpha = b - \beta = c - \gamma$.
3. Moreover, the minimal free resolution of $R/I$ has the form $$\label{eqn:resolution}
0
\rightarrow
\begin{array}{c} R(-a-b-\gamma) \\ \oplus \\ R(-a-\beta-c) \\ \oplus \\ R^n(-\alpha-b-c) \end{array}
\rightarrow
\begin{array}{c} R(-a-\beta-\gamma) \\ \oplus \\ R(-\alpha-b-\gamma) \\ \oplus \\ R(-\alpha-\beta-c) \\ \oplus \\
R(-a-b) \\ \oplus \\ R(-a-c) \\ \oplus \\ R^n(-b-c) \end{array}
\rightarrow
\begin{array}{c} R(-\alpha-\beta-\gamma) \\ \oplus \\ R(-a) \\ \oplus \\ R(-b) \\ \oplus \\ R(-c) \end{array}
\rightarrow
R
\rightarrow
R/I
\rightarrow
0$$ where $n = 1$ if $\alpha > 0$ and $n = 0$ if $\alpha = 0$.
Moreover, we see that in characteristic zero the weak Lefschetz property follows for certain choices of the parameters.
[[@MMN Theorem 6.2]]{} \[pro:amaci-not-3\] Let $K$ be an algebraically closed field of characteristic zero. Then $R/I_{a,b,c,\alpha,\beta,\gamma}$ has the weak Lefschetz property if $a + b + c + \alpha + \beta + \gamma \not\equiv 0 \pmod{3}$.
Semi-stability
--------------
The syzygy module $\operatorname{syz}{I}$ of $I = I_{a,b,c,\alpha,\beta,\gamma}$ fits into the exact sequence $$0
\longrightarrow
\operatorname{syz}{I}
\longrightarrow
R(-\alpha-\beta-\gamma) \oplus R(-a) \oplus R(-b) \oplus R(-c)
\longrightarrow
I_{a,b,c,\alpha,\beta,\gamma}
\longrightarrow
0.$$ The sheafification $\widetilde{\operatorname{syz}{I}}$ is a rank 3 bundle on ${\mathbb{P}}^2$, and it is called the [*syzygy bundle*]{} of $I$. Recall that a vector bundle $E$ on projective space is said to be [*semistable*]{} if, for every coherent subsheaf $F \subset E$, the following inequality holds: $$\frac{c_1(F)}{rk(F)} \leq \frac{c_1(E)}{rk(E)}.$$
We analyse when $I_{a,b,c,\alpha,\beta,\gamma}$ has a semistable syzygy bundle. (Note, the slightly awkward definition of $s$ in the following is kept for consistency with [@MMN Section 7], the starting point of this work.)
\[pro:semistable\] Let $K$ be an algebraically closed field of characteristic zero. Further, let $I = I_{a,b,c,\alpha,\beta,\gamma}$, and define the following rational numbers $$\begin{split}
s :=& \frac{1}{3}(a + b + c + \alpha + \beta + \gamma)-2, \\
A :=& s+2-a, \\
B :=& s+2-b, \\
C :=& s+2-c, \mbox{ and}\\
M :=& s + 2 - (\alpha + \beta + \gamma).
\end{split}$$ Then $I$ has a semistable syzygy bundle if and only if the following conditions all hold:
1. $0 \leq M$,
2. $0 \leq A \leq \beta + \gamma$,
3. $0 \leq B \leq \alpha + \gamma$, and
4. $0 \leq C \leq \alpha + \beta$.
Using [@Br Corollary 7.3] we have that $I$ has a semistable syzygy bundle if and only if
1. $\max\{a, b, c, \alpha + \beta + \gamma\} \leq s+2$,
2. $\min\{\alpha + \beta + c, \alpha + b + \gamma, a + \beta + \gamma\} \geq s+2$, and
3. $\min\{a+b, a+c, b+c\} \geq s+2$.
Notice that condition (a) is equivalent to $A, B, C,$ and $M$ being non-negative. Moreover, condition (b) is equivalent to the upper bounds on $A, B,$ and $C$. We claim that condition (c) follows directly from condition (a).
Indeed, by condition (a) we have that $C+M \geq 0$ and so $A+B+C+M = s+2 \geq A+B = 2(s+2) - a - b$, thus $a + b \geq s+2$. Similarly, we have $a + c \geq s+2$ and $b+c \geq s+2$. Thus condition (c) holds if condition (a) holds.
This gives further conditions on the parameters that force the weak Lefschetz property in characteristic zero (see [@BK Theorem 3.3]). This extends [@MMN Lemma 6.7].
\[cor:wlp-ss\] Let $K$ be an algebraically closed field of characteristic zero, and let $I = I_{a,b,c,\alpha,\beta,\gamma}$. If any of the conditions (i)-(iv) in Proposition \[pro:semistable\] fail, then $R/I$ has the weak Lefschetz property.
The above definitions of $s, A, B, C,$ and $M$ are not without purpose. Before going further, we make a few comments about the given parameters.
\[rem:sabcm\] Suppose $s, A, B, C,$ and $M$ are defined as in Proposition \[pro:semistable\]. Then clearly $s$ is an integer if and only if $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$; if $s$ is an integer, then so are $A, B, C,$ and $M$. Further, $A+B+C+M = s+2$ and $A+B+C=\alpha + \beta + \gamma$.
Associated matrices
-------------------
Given the minimal free resolution of $R/I$ (see (\[eqn:resolution\])), we can easily compute the $h$-vector of $R/I$ as a weighted sum of binomial coefficients dependent only on the parameters $a,b,c,\alpha, \beta,$ and $\gamma$.
We say $h(A)$ has [*twin peaks*]{} if there exists an integer $s$ such that $h_s = h_{s+1}$. When $I_{a,b,c,\alpha,\beta,\gamma}$ has parameters as in Proposition \[pro:semistable\] and $s$ is an integer, then the algebras $R/I_{a,b,c,\alpha,\beta,\gamma}$ always have twin peaks and the peaks are bounded by the socle degrees. This extends the results in [@MMN Lemma 7.1] wherein the level algebras $R/I_{a,b,c,\alpha,\beta,\gamma}$ with twin peaks are identified.
\[lem:twin-peaks\] Assume the parameters of $I = I_{a,b,c,\alpha,\beta,\gamma}$ satisfy the conditions in Proposition \[pro:semistable\] and suppose $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$. Then $R/I$ has twin peaks in degrees $s$ and $s+1$. Moreover, $s+1$ is bounded above by the socle degrees of $R/I$.
The upper bounds on $A, B,$ and $C$ are exactly those required to force the ultimate and penultimate terms in the minimal free resolution of $R/I$, given in Proposition \[pro:amaci-props\](iii), to not contribute to the computation of the $h$-vector for degrees up to $s+1$. Moreover, as $A,B,C,$ and $M$ are non-negative, and using $\binom{n+1}{2} - \binom{n}{2} = n$ for $n \geq 0$, then $$\begin{split}
h_{s+1} - h_s = & \left(\binom{s+3}{2} - \binom{A+1}{2} - \binom{B+1}{2} - \binom{C+1}{2} - \binom{M+1}{2}\right) \\
& - \left(\binom{s+2}{2} - \binom{A}{2} - \binom{B}{2} - \binom{C}{2} - \binom{M}{2}\right) \\
= & s+2 - (A + B + C + M) \\
= & 0.
\end{split}$$
Suppose, without loss of generality, that $\alpha \leq \beta \leq \gamma$. The socle degrees of $R/I$ are $\alpha + b + c -3$, $a+\beta+c-3$, and $a+b+\gamma -3$, with the first removed if $\alpha = 0$. The following argument shows that $\alpha + b + c - 3$ is at least $s+1$, however, with a simple changing of names it can be used to show that each of the socle degrees is at least $s+1$.
As we are considering the socle degree $\alpha + b + c - 3$, we may assume $\alpha \geq 1$. Notice that $\alpha + b + c - 3 = 2A + B + C + 2M + \alpha - 3$, which is at least $s+1 = A+B+C+M-1$ exactly when $A+M+\alpha \geq 2$. If $A+M\geq 1$, then we are done. Suppose $A+M = 0$, then $A = M = 0$ and $b+c = \alpha+\beta+\gamma$. Moreover, since $b > \beta$ and $c > \gamma$, then $\alpha+\beta+\gamma = b+c \geq \beta+\gamma + 2$. Thus $\alpha \geq 2$.
An immediate consequence of the previous lemma is that exactly one map need be considered for each algebra in order to determine the presence of the weak Lefschetz property.
\[cor:one-map\] Assume the parameters of $I = I_{a,b,c,\alpha,\beta,\gamma}$ satisfy the conditions in Proposition \[pro:semistable\] and suppose $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$. Then $R/I$ has the weak Lefschetz property if and only if the map $\times(x+y+z): [R/I]_s \rightarrow [R/I]_{s+1}$ is injective (or surjective).
This follows immediately from Lemma \[lem:twin-peaks\] by using Propositions \[pro:surj\]–\[pro:mono\].
This leads to the definition of two matrices with determinants that determine the weak Lefschetz property. The first is a zero-one matrix and the second is a matrix of binomial coefficients.
\[pro:wlp-zero-one\] Assume the parameters of $I = I_{a,b,c,\alpha,\beta,\gamma}$ satisfy the conditions in Proposition \[pro:semistable\] and suppose $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$.
Then there exists a matrix $Z = Z_{a,b,c,\alpha,\beta,\gamma}$ such that
1. $Z$ is a square integer matrix of size $h_s$,
2. $R/I$ has the weak Lefschetz property if and only if $\det{Z} \not\equiv 0 \pmod{\operatorname{char}{K}}$, and
3. the entries of $Z$ are given by $$(Z)_{i,j} = \left\{
\begin{array}{ll}
1 & n_j \mbox{ is a multiple of } m_i, \\[0.3em]
0 & \mbox{otherwise},
\end{array}
\right.$$ where $\{m_1, \ldots, m_{h_s}\}$ and $\{n_1, \ldots, n_{h_s}\}$ are the monomial bases of $[R/I]_s$ and $[R/I]_{s+1}$, respectively, and are given in lexicographic order.
We notice that the map $\times(x+y+z): [R/I]_s \rightarrow [R/I]_{s+1}$ can be represented as a matrix $Z$ with rows and columns indexed by fixed monomial bases of $[R/I]_s$ and $[R/I]_{s+1}$, respectively. This follows immediately from viewing $[R/I]_d$ as a vector space over $K$.
Claim (i) follows from Lemma \[lem:twin-peaks\] wherein it is shown that $h_s = h_{s+1}$. Since $Z$ is square, then the injectivity of $\times(x+y+z): [R/I]_s \rightarrow [R/I]_{s+1}$ is equivalent to $Z$ being invertible, that is, equivalent to $\det{Z}$ being non-zero in $K$. Thus, claim (ii) follows from Corollary \[cor:one-map\] wherein it is shown that the injectivity of the map $\times(x+y+z): [R/I]_s \rightarrow [R/I]_{s+1}$ exactly determines the presence of the weak Lefschetz property for $R/I$. Claim (iii) follows immediately from the construction of the map.
The following generalises the results in [@MMN Theorem 7.2 and Corollary 7.3].
\[pro:wlp-binom\] Assume the parameters of $I = I_{a,b,c,\alpha,\beta,\gamma}$ satisfy the conditions in Proposition \[pro:semistable\], and suppose $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$.
Then there exists a matrix $N = N_{a,b,c,\alpha,\beta,\gamma}$ such that
1. $N$ is a square integer matrix of size $C+M$,
2. $R/I$ has the weak Lefschetz property if and only if $\det{N} \not\equiv 0 \pmod{\operatorname{char}{K}}$, and
3. the entries of $N$ are given by $$(N)_{i,j} = \left\{
\begin{array}{ll}
\displaystyle \binom{c}{A + j - i} & \mbox{if } 1 \leq i \leq C, \\[0.8em]
\displaystyle \binom{\gamma}{A+C-\beta + j - i} & \mbox{if } C + 1 \leq i \leq C+M.
\end{array}
\right.$$
Notice that $R/(I,x+y+z) \cong S/J$, where $S = K[x,y]$ and $$J = (x^a, y^b, (x+y)^c, x^\alpha y^\beta (x+y)^\gamma).$$ Thus the sequence $$[R/I]_d \xrightarrow{\times(x+y+z)} [R/I]_{d+1} \rightarrow [R/(I, x+y+z)]_{d+1} \rightarrow 0$$ implies that $\times(x+y+z): [R/I]_s \rightarrow [R/I]_{s+1}$ is injective exactly when $[S/J]_{s+1} = 0$. Hence it suffices to show that all $s+2$ monomials of the form $x^iy^j$ where $i+j=s+1$ are in $J$.
Clearly if $i \geq a$ or $j \geq b$, then $x^iy^j$ is in $J$. This leaves $s+2 - (s+2-a) - (s+2-b) = s+2 - A - B = C+M$ monomials that are not trivially in $J$. Thus there are $C+M$ equations and unknowns, all of which only involve the non-monomial terms (after reduction by the monomial terms). Associated to this system of equations is a square integer matrix of size $C+M$, call it $N$. Then $N$ is invertible if and only if $\det{N}$ is non-zero in $K$. Thus, claims (i) and (ii) hold.
There are $s+2-c = C$ ways to scale $(x+y)^c$ and $s+2-(\alpha+\beta+\gamma) = M$ ways to scale $x^\alpha y^\beta (x+y)^\gamma$ to be degree $s+1$. In both cases consider the binomial coefficient indexed by the degree of $y$. Then $(N)_{i,j}$ is the coefficient on $x^{a-j}y^{A+j-1}$ in the scaling $x^{C-i}y^{i-1}(x+y)^c$ for $1 \leq i \leq C$, i.e., $\binom{c}{A+j-i}$, and in the scaling $x^{C+M-i}y^{i-C-1}x^\alpha y^\beta (x+y)^\gamma$ for $C+1 \leq i \leq C+M$, i.e., $\binom{\gamma}{A+C-\beta+j-i}$. Thus claim (iii) holds.
Clearly $\det{Z_{a,b,c,\alpha,\beta,\gamma}}$ and $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ must both be either zero or have the same set of prime divisors. We can determine a few of the prime divisors from the known failure of the weak Lefschetz property.
\[pro:wlp-p\] Assume the parameters of $I = I_{a,b,c,\alpha,\beta,\gamma}$ satisfy the conditions in Proposition \[pro:semistable\], and suppose $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$. If $K$ has positive characteristic $p$ and their exists a positive integer $m$ such that $$\max\{a,b,c\} \leq p^m \leq s+1 = \frac{1}{3}(a + b + c + \alpha + \beta + \gamma)-1,$$ then
1. $R/I$ fails to have the weak Lefschetz property,
2. $p$ is a prime divisor of the determinant of $Z_{a,b,c,\alpha,\beta,\gamma}$, and
3. $p$ is a prime divisor of the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$.
By Lemma \[lem:twin-peaks\], the Hilbert function of $R/I$ weakly increases to degree $s+1$, hence part (i) follows by Lemma \[lem:wlp-p\]. Parts (ii) and (iii) then follow from Propositions \[pro:wlp-zero-one\] and \[pro:wlp-binom\], respectively.
In the next section we will see a nice combinatorial interpretation for both matrices as well as the defined values $s, A, B, C,$ and $M$.
Punctured hexagons and friends {#sec:ph}
==============================
Recall the definition of $s, A, B, C,$ and $M$, and the conditions thereon, from Proposition \[pro:semistable\]. In this section we assume, without exception, that $I = I_{a,b,c,\alpha,\beta,\gamma}$ has parameters matching these conditions and further that $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$.
Punctured hexagons
------------------
Notice that every monomial in $[R]_d$ is of the form $x^i y^j z^k$ where $i,j,$ and $k$ are non-negative integers such that $i+j+k=d$. Hence we can organise the monomials in $[R]_d$ into a triangle of side-length $d+1$ with $x^d$ at the lower-center, $y^d$ at the upper-right, and $z^d$ at the upper-left. (See Figure \[fig:mono-tri\].)
![The monomial triangle for $[R]_3$[]{data-label="fig:mono-tri"}](tri-mono)
Notice that we can interlace the monomials of $[R]_{d-1}$ within the monomials of $[R]_{d}$. If we stay consistent with our orientation (i.e., largest power of $x$ at the lower-center, largest power of $y$ at the upper-right, and largest power of $z$ at the upper-left), then two monomials are adjacent if and only if one divides the other. (See Figure \[fig:interlace-tri\].) We call such a figure the [*interlaced basis region of $[R]_{d-1}$ and $[R]_{d}$*]{}.
![The interlaced basis region of $[R]_2$ and $[R]_3$[]{data-label="fig:interlace-tri"}](tri-lace)
If we compute the interlaced basis region of $[R/I_{a,b,c,\alpha,\beta,\gamma}]_{s}$ and $[R/I_{a,b,c,\alpha,\beta,\gamma}]_{s+1}$, then we get a punctured hexagonal region.
\[thm:interlace-amaci\] Let $I = I_{a,b,c,\alpha,\beta,\gamma}$ satisfy the conditions in Proposition \[pro:semistable\], and suppose $a + b + c + \alpha + \beta + \gamma \equiv 0 \pmod{3}$. Then the interlaced basis region $H_{a,b,c,\alpha,\beta,\gamma}$ of $[R/I]_{s}$ and $[R/I]_{s+1}$ is in the shape of a hexagon with side-lengths (in clockwise cyclic order, starting at the bottom) $$(A, B+M, C, A+M, B, C+M)$$ and with a puncture in the shape of an equilateral triangle of side-length $M$. The puncture has sides parallel to the sides of the hexagon of lengths $A+M, B+M,$ and $C+M$. Moreover, the puncture is located $\alpha, \beta,$ and $\gamma$ units from the sides of length $A+M, B+M,$ and $C+M$, respectively. (See Figure \[fig:punctured-hexagon\].)
![$H_{a,b,c,\alpha,\beta,\gamma}$, the interlaced basis region of $[R/I]_s$ and $[R/I]_{s+1}$[]{data-label="fig:punctured-hexagon"}](hex)
The interlaced basis region of $[R/I]_s$ and $[R/I]_{s+1}$ corresponds to a spatial placement of the monomials of the associated components of $R/I$. As $I$ is a monomial ideal, we can easily get restrictions on the monomials $x^iy^jz^k$ in the region:
1. The generator $x^a$ forces $0 \leq i < a$; this corresponds to the lower-center missing triangle which has side-length $s+2 - a = A$.
2. The generator $y^b$ forces $0 \leq j < b$; this corresponds to the upper-right missing triangle which has side-length $s+2 - b = B$.
3. The generator $z^c$ forces $0 \leq k < c$; this corresponds to the upper-left missing triangle which has side-length $s+2 - c = C$.
4. The generator $x^\alpha y^\beta z^\gamma$ forces one of $i < \alpha, j < \beta$, or $k < \gamma$ to also hold; this corresponds to the center missing triangle, which has side-length $s+2 - \alpha - \beta - \gamma = M$. This further forces the particular placement of the puncture.
Moreover, the conditions in Proposition \[pro:semistable\] force the regions to have non-negative side-lengths and to not overlap.
\[rem:bijection\] The ideals $I = I_{a,b,c,\alpha,\beta,\gamma}$ in Theorem \[thm:interlace-amaci\] are in bijection with their hexagonal regions (assuming a fixed orientation and assuming a puncture of side-length zero is still considered to be in a particular position). Suppose we have a punctured hexagonal region, as in Figure \[fig:punctured-hexagon\], with parameters $A,B,C,M,\alpha,$ and $\beta$. Then $a = B+C+M$, $b = A+C+M$, $c = A+B+M$, and $\gamma = A+B+C-(\alpha + \beta)$.
Moreover, we notice that, in characteristic zero, these ideals are exactly the artinian monomial almost complete intersections which do not immediately have the weak Lefschetz property from Proposition \[pro:amaci-not-3\] or Proposition \[pro:semistable\].
Notice that by Lemma \[lem:twin-peaks\] we have $h_s = h_{s+1}$, so the region $H_{a,b,c,\alpha,\beta,\gamma}$ has the same number of upward pointing triangles as it has downward pointing triangles. In particular, it may then be possible to tile the region by lozenges (i.e., rhombi with unit side-lengths and angles of $60^\circ$ and $120^\circ$; we also note a pair of alternate names used in the literature: calissons and diamonds).
Non-intersecting lattice paths
------------------------------
We follow [@CEKZ Section 5] (similarly, [@Fi Section 2]) to translate lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$ to families of non-intersecting lattice paths. An example of a lozenge tiling and its associated family of non-intersecting lattice paths is given in Figure \[fig:hex-nilp\].
![Example of a lozenge tiling and its associated family of non-intersecting lattice paths[]{data-label="fig:hex-nilp"}](tile "fig:")\
[*Hexagon tiling by lozenges*]{}
![Example of a lozenge tiling and its associated family of non-intersecting lattice paths[]{data-label="fig:hex-nilp"}](nilp-fin "fig:")\
[*Family of non-intersecting lattice paths*]{}
In order to transform a lozenge tiling of a punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$ into a family of non-intersecting lattice paths, we follow three simple steps (see Figure \[fig:hex-to-nilp\]):
1. Mark the midpoints of the triangle edges parallel to the sides of length $C$ and $C+M$ with vertices. Further, label the midpoints, always moving lower-left to upper-right,
1. along the hexagon side of length $C$ as $A_1, \ldots, A_C$,
2. along the puncture as $A_{C+1}, \ldots, A_{C+M}$, and
3. along the hexagon side of length $C+M$ as $E_1, \ldots, E_{C+M}$.
2. Using the lozenges as a guide, we connect any pair of vertices that occur on a single lozenge.
3. Thinking of motion parallel to the side of length $A$ as horizontal and motion parallel to the side of length $B$ as vertical, we orthogonalise the lattice (and paths) and consider the lower-left vertex as the origin.
![Example of converting lozenge tilings to families of non-intersecting lattice paths[]{data-label="fig:hex-to-nilp"}](nilp-1 "fig:")\
[*(i) Mark midpoints with vertices and label particular vertices*]{}
![Example of converting lozenge tilings to families of non-intersecting lattice paths[]{data-label="fig:hex-to-nilp"}](nilp-2 "fig:")\
[*(ii) Connect vertices using the tiling*]{}
\
![Example of converting lozenge tilings to families of non-intersecting lattice paths[]{data-label="fig:hex-to-nilp"}](nilp-3 "fig:")\
[*(iii) Orthogonalise the path family*]{}
![Example of converting lozenge tilings to families of non-intersecting lattice paths[]{data-label="fig:hex-to-nilp"}](nilp-fin "fig:")\
[*The family by itself*]{}
Given the above transformation of $H_{a,b,c,\alpha,\beta,\gamma}$ to the integer lattice, we see that $A_i$ and $E_j$ have easy to compute coordinates: $$A_i = \left\{
\begin{array}{ll}
(i-1, B+M+i-1) & \mbox{if } 1 \leq i \leq C, \\[0.3em]
(\beta + i - C - 1, B - \alpha + i - 1) & \mbox{if } C+1 \leq i \leq C+M,
\end{array}
\right.$$ and $$E_j = (A + j - 1, j - 1) \mbox{ for } 1 \leq j \leq C+M.$$
Now we associate to each family of non-intersecting lattices paths a permutation and use it to assign a sign to the family of paths.
\[def:nilp-sign\] Let $L$ be a family of non-intersecting lattice paths as above, and let $\lambda \in {\mathfrak{S}}_{C+M}$ be the permutation so that $A_i$ is connected to $E_{\lambda(i)}$. We define the [*sign*]{} of $L$ to be the signature (or sign) of the permutation $\lambda$. That is, $\operatorname{sgn}{L} := \operatorname{sgn}{\lambda}$.
Now we are ready to use a beautiful theorem relating (signed) enumerations of families of non-intersecting lattice paths with determinants. In particular, we use a theorem first given by Lindström in [@Li Lemma 1] and stated independently in [@GV-1989 Theorem 1] by Gessel and Viennot. Stanley gives a very nice exposition of the topic in [@St-EC Section 2.7].
Here we give a specialisation of the theorem to the case when all edges have the same weight—one. In particular, this result is given in [@CEKZ Lemma 14].
\[thm:lgv\] Let $A_1, \ldots, A_n, E_1, \ldots, E_n$ be distinct lattice points on ${\mathbb{N}}_0^2$ where each $A_i$ is above and to the left of every $E_j$. Then $$\det_{1\leq i, j \leq n} (P(A_i \rightarrow E_j)) = \sum_{\lambda \in {\mathfrak{S}}_n} \operatorname{sgn}(\lambda) P^+_\lambda(A\rightarrow E),$$ where $P(A_i \rightarrow E_j)$ is the number of lattice paths from $A_i$ to $E_j$ and, for each permutation $\lambda \in {\mathfrak{S}}_n$, $P^+_\lambda(A \rightarrow E)$ is the number of families of non-intersecting lattice paths with paths going from $A_i$ to $E_{\lambda(i)}$.
Thus, we have an enumeration of the signed lozenge tilings of a punctured hexagon with signs given by the non-intersecting lattice paths.
\[thm:nilp-matrix\] The enumeration of signed lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$, with signs given by the signs of the associated families of non-intersecting lattice paths (Definition \[def:nilp-sign\]), is given by $\det N_{a,b,c,\alpha,\beta,\gamma}$, where the matrix $N_{a,b,c,\alpha,\beta,\gamma}$ is defined in Proposition \[pro:wlp-binom\].
Notice that the number of lattice paths from $(u,v)$ to $(x,y)$, where $u \leq x$ and $v \geq y$, is given by $\binom{x-u+v-y}{x-u}$ as there are $x-u + v-y$ steps and $x-u$ must be horizontal steps (equivalently, $v-y$ must be vertical steps). Thus the claim follows immediately from the steps above.
However, we need not consider all $(C+M)!$ permutations $\lambda \in {\mathfrak{S}}_{C+M}$ as the vast majority will always have $P^+_\lambda(A \rightarrow E) = 0$. Given our choice of $A_i$ and $E_j$ the only possible choices of $\lambda$ are given by $$\lambda_k = \left(
\begin{array}{cccc|cccc|cccc}
1 & 2 & \cdots & k & k+1 & k+2 & \cdots & C & C+1 & C+2 & \cdots & C+M \\
1 & 2 & \cdots & k & M+k+1 & M+k+2 & \cdots & C+M & k+1 & k+2 & \cdots & M+k
\end{array}
\right),$$ where $0 \leq k \leq C$ and $k$ corresponds to the number of lattice paths that go below the puncture. In particular, the three parts of $\lambda_k$ correspond to the paths going below, above, and starting from the puncture. We call these permutations the [*admissible permutations*]{} of $H_{a,b,c,\alpha,\beta,\gamma}$.
We will use this connection to compute determinants in Section \[sec:det\], but first we look at an alternate combinatorial connection.
Perfect matchings
-----------------
Lozenge tilings of a punctured hexagon can be associated to perfect matchings on a bipartite graph. This connection was first used by Kuperberg in [@Ku] to study symmetries on plane partitions. An example of a lozenge tiling and its associated perfect matching of edges is given in Figure \[fig:hex-bip\].
![Example of a lozenge tiling and its associated perfect matching of edges[]{data-label="fig:hex-bip"}](tile "fig:")\
[*(i) Hexagon tiling by lozenges*]{}
![Example of a lozenge tiling and its associated perfect matching of edges[]{data-label="fig:hex-bip"}](bip-fin "fig:")\
[*(ii) Perfect matching of edges*]{}
In order to transform a lozenge tiling of a punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$ into a perfect matching of edges, we follow three simple steps (see Figure \[fig:hex-to-bip\]):
1. Put a vertex at the center of each triangle.
2. Connect the vertices whose triangles are adjacent.
3. Select the edges which the lozenges cover–this set is the perfect matching.
![Example of converting lozenge tilings to perfect matchings of edges[]{data-label="fig:hex-to-bip"}](bip-1 "fig:")\
[*(i) Put vertices in triangle centers*]{}
![Example of converting lozenge tilings to perfect matchings of edges[]{data-label="fig:hex-to-bip"}](bip-2 "fig:")\
[*(ii) Connect vertices of adjacent triangles*]{}
\
![Example of converting lozenge tilings to perfect matchings of edges[]{data-label="fig:hex-to-bip"}](bip-3 "fig:")\
[*(iii) Select edges covered by lozenges*]{}
![Example of converting lozenge tilings to perfect matchings of edges[]{data-label="fig:hex-to-bip"}](bip-fin "fig:")\
[*The perfect matching by itself*]{}
Notice that the graph associated to the punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$ is a bipartite graph with colour classes given by monomials in $[R/I]_{s}$ and $[R/I]_{s+1}$. Thus we can represent this bipartite graph by a bi-adjacency matrix with rows enumerated by the monomials in $[R/I]_{s}$ and columns enumerated by the monomials in $[R/I]_{s+1}$. We fix the order on the monomials to be the lexicographic order. Clearly then the matrix $Z_{a,b,c,\alpha,\beta,\gamma}$ from Proposition \[pro:wlp-zero-one\] is the bi-adjacency matrix described here.
Consider the permanent of $Z = Z_{a,b,c,\alpha,\beta,\gamma}$, that is, $$\operatorname{perm}{Z} = \sum_{\pi \in {\mathfrak{S}}_{h_s}} \prod_{i=1}^{h_s} (Z)_{i,\pi(i)}.$$ As $Z$ has entries which are either zero or one, we see that all summands in $\operatorname{perm}{Z}$ are either zero or one. Moreover, each non-zero summand corresponds to a perfect matching, as it corresponds to an isomorphism between the two colours classes of the bipartite graph, namely, the monomials in $[R/I]_{s}$ and $[R/I]_{s+1}$. Thus, $\operatorname{perm}{Z}$ enumerates the perfect matchings of the bipartite graph associated to $H_{a,b,c,\alpha,\beta,\gamma}$, and hence $\operatorname{perm}{Z}$ also enumerates the lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$.
\[pro:bip-matrix\] The number of lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$ is $\operatorname{perm}{Z_{a,b,c,\alpha,\beta,\gamma}}.$
Since each perfect matching is an isomorphism between the two colour classes, it can be seen as a permutation $\pi \in {\mathfrak{S}}_{h_s}$. As with Definition \[def:nilp-sign\], it is thus natural to assign a sign to a given perfect matching.
\[def:pm-sign\] Let $P$ be a perfect matching of the bipartite graph associated to $H_{a,b,c,\alpha,\beta,\gamma}$, and let $\pi \in {\mathfrak{S}}_{h_s}$ be the associated permutation (as described above). We define th [*sign*]{} of $P$ to be the signature of the permutation $\pi$. That is, $\operatorname{sgn}{P} := \operatorname{sgn}{\pi}$.
Since the sign is the sign that is used in computing the determinant of the matrix $Z_{a,b,c,\alpha,\beta,\gamma}$, we get an enumeration of the signed lozenge tilings of a punctured hexagon with signs given by the perfect matchings.
\[thm:bip-matrix\] The enumeration of signed perfect matchings of the bipartite graph associated to $H_{a,b,c,\alpha,\beta,\gamma}$, with signs given by the signs of the related perfect matchings, is given by $\det Z_{a,b,c,\alpha,\beta,\gamma}$, where the matrix $Z_{a,b,c,\alpha,\beta,\gamma}$ is defined in Proposition \[pro:wlp-zero-one\].
\[rem:dimers\] Kasteleyn [@Ka] provided, in 1967, a general method for computing the number of perfect matchings of a planar graph as a determinant. Moreover, he provided a classical review of methods and applications of enumerating perfect matchings. Planar graphs, such as the “honeycomb graphs” described here, are studied for their connections to physics; in particular, honeycomb graphs model the bonds in dimers (polymers with only two structural units) and perfect matchings correspond to so-called [*dimer coverings*]{}. Kenyon [@Ke] gives a modern recount of explorations on dimer models, including random dimer coverings and their limiting shapes.
\[rem:wlp-p\] Recall that Proposition \[pro:wlp-p\] provides a numerical constraint that determines some of the prime divisors of the determinants of the matrices $Z_{a,b,c,\alpha,\beta,\gamma}$ and $N_{a,b,c,\alpha,\beta,\gamma}$ by means of some algebra deciding the weak Lefschetz property for the algebra $R/I_{a,b,c,\alpha,\beta,\gamma}$. Hence, by Theorems \[thm:nilp-matrix\] and \[thm:bip-matrix\], we see that information from algebra can indeed be used to determine some of the prime divisors of the enumerations of signed lozenge tilings and of signed perfect matchings.
Finally, we note that in [@Pr], Propp gives a history of the connections between lozenge tilings (of non-punctured hexagons), perfect matchings, plane partitions, non-intersecting lattice paths.
Interlude of signs {#sec:signs}
==================
In the preceding section we discussed three related combinatorial structures from which we can extract the primes $p$ for which the algebras $R/I_{a,b,c,\alpha,\beta,\gamma}$ fail to have the weak Lefschetz property. Therein we discussed two different ways to assign a sign to a lozenge tiling: by the associated family of non-intersecting lattice paths (Definition \[def:nilp-sign\]) and by the associated perfect matching (Definition \[def:pm-sign\]). We now show that the two signs indeed agree.
Fix a hexagonal region $H = H_{a,b,c,\alpha,\beta,\gamma}$, and fix a lozenge tiling $T$ of $H$. As discussed in Section \[sec:ph\], we associate to the tiling $T$ a family of non-intersecting lattice paths $L_T$ and a perfect matching $P_T$. Moreover, we introduced a permutation $\lambda_T \in {\mathfrak{S}}_{C+M}$ associated to $L_T$ (see Definition \[def:nilp-sign\]) and a permutation $\pi_T \in {\mathfrak{S}}_{h_s}$ associated to $P_T$ via $Z_{a,b,c,\alpha,\beta,\gamma}$ (see Definition \[def:pm-sign\]).
We first notice that “rotating” particular lozenge groups of $T$ do not change the permutation associated to the non-intersecting lattice paths.
\[lem:perm-rotate\] Let $T$ be a lozenge tiling of $H_{a,b,c,\alpha,\beta,\gamma}$. Pick any triplet of lozenges in $T$ which is either an [*up*]{} or a [*down*]{} grouping, as in Figure \[fig:rotatable\],
![[*up*]{} and [*down*]{} lozenge groups with lattice path pieces superimposed[]{data-label="fig:rotatable"}](rotatable)
and let $U$ be $T$ with the triplet exchanged for the other possibility (i.e., rotated $180^\circ$). Then $U$ is a lozenge tiling of $H_{a,b,c,\alpha,\beta,\gamma}$ and $\lambda_T = \lambda_U$. Moreover, $\pi_U = \tau \pi_T$, for some three-cycle $\pi \in {\mathfrak{S}}_{h_s}$.
First, we note that if $T$ is a lozenge tiling of $H_{a,b,c,\alpha,\beta,\gamma}$ then clearly so is $U$ as the change does not modify any tiles besides the three in the triplet.
Next, notice that exchanging the triplet in $T$ for its rotation only modifies the associated family of non-intersecting lattice paths in one path. Moreover, it does not change the starting or ending points of the path, merely the order in which it gets there, that is, either right then down or down then right. Thus, $\lambda_T = \lambda_U$.
Last, suppose, without loss of generality, that our chosen triplet is an [*up*]{} lozenge group. Label the three upward pointing triangles in the triplet $i, j, k$ as in Figure \[fig:rotatable-labeled\].
![An [*up*]{} lozenge group with labeling[]{data-label="fig:rotatable-labeled"}](rotatable-labeled)
Thus we see that $\pi_U(i) = \pi_T(k)$, $\pi_U(j) = \pi_T(i)$, $\pi_U(k) = \pi_T(j)$, and $\pi_U(m) = \pi_T(m)$ for $m$ not $i, j,$ or $k$. Hence $\pi_U = \tau \pi_T$ where $\tau$ is the three-cycle $(\pi_T(k), \pi_T(j),\pi_T(i))$.
It follows that two lozenge tilings that have the same $\lambda$ permutation have $\pi$ permutations with the same sign.
\[pro:sign-agree\] For each $H_{a,b,c,\alpha,\beta,\gamma}$ there exists a constant $i \in \{1, -1\}$ such that for all lozenge tilings $T$ of $H_{a,b,c,\alpha,\beta,\gamma}$ the expression $\operatorname{sgn}{L_T} = i \cdot \operatorname{sgn}{P_T}$ holds, where $L_T$ is the family of non-intersecting lattice paths associated to $T$ and $P_T$ is the family of perfect matchings associated to $T$.
[*Step 1*]{}:
Let $T$ and $U$ be two lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$ with $\lambda_T = \lambda_U$. As $\lambda_T = \lambda_U$, then the families of non-intersecting lattice paths associated to $T$ and $U$ start and end at the same places. Hence $T$ can be modified by a series of, say $n$, rotations, as in Lemma \[lem:perm-rotate\], to $U$. Thus $$\pi_U = \tau_n \tau_{n-1} \cdots \tau_1 \pi_T,$$ where $\tau_1, \ldots, \tau_n \in {\mathfrak{S}}_{h_s}$ are three cycles by Lemma \[lem:perm-rotate\]. As $\operatorname{sgn}\tau_i = 1$ for $1 \leq i \leq n$, and $\operatorname{sgn}$ is a group homomorphism, we see that $\operatorname{sgn}\pi_T = \operatorname{sgn}\pi_U$.
[*Step 2*]{}:
By the comments following Theorem \[thm:nilp-matrix\] we only need to consider the admissible permutations $\lambda_0, \ldots, \lambda_C$. Moreover, $\operatorname{sgn}\lambda_k = (-1)^{M(C-k)}$ so $\operatorname{sgn}\lambda_k = (-1)^M\operatorname{sgn}\lambda_{k+1}$.
Let $T$ and $U$ be two lozenge tilings of $H = H_{a,b,c,\alpha,\beta,\gamma}$ with $\lambda_T = \lambda_k$ and $\lambda_U = \lambda_{k+1}$. That is, $\operatorname{sgn}\lambda_T = (-1)^M \operatorname{sgn}\lambda_U$. First, $\alpha \geq C-k$ by the existence of $T$ as $C-k$ paths go above the puncture and so must go through a gap of size $\alpha$, and similarly $\beta \geq k+1$ by the existence of $U$.
By [*Step 1*]{}, we may pick $T$ and $U$ however we wish, as long as $\lambda_T = \lambda_k$ and $\lambda_U = \lambda_{k+1}$. In particular, let $T$, and similarly $U$, be defined as follows (see Figure \[fig:maxmin-tiling\]):
![An example of tilings $T$ and $U$ of $H_{9,8,9,4,3,3}$, for $k = 1$, which are “minimal” below the puncture and “maximal” everywhere else; both tilings have the regions of similarity highlighted.[]{data-label="fig:maxmin-tiling"}](maxmin-tiling-1 "fig:")\
[*(i) The tiling $T$*]{}
![An example of tilings $T$ and $U$ of $H_{9,8,9,4,3,3}$, for $k = 1$, which are “minimal” below the puncture and “maximal” everywhere else; both tilings have the regions of similarity highlighted.[]{data-label="fig:maxmin-tiling"}](maxmin-tiling-2 "fig:")\
[*(ii) The tiling $U$*]{}
1. The $C-k$ paths above the puncture ($C-k-1$ for $U$) always move right before moving down.
2. The $M$ paths from the puncture always move right before moving down.
3. The $k$ paths below the puncture ($k+1$ for $U$) always move down before moving right.
With the idea of [*up*]{} and [*down*]{} triplets from Lemma \[lem:perm-rotate\], we can say a path is “minimal” if it contains no [*up*]{} triplets and a path is “maximal” if it contains no [*down*]{} triplets. Thus, $T$ and $U$ are “minimal” below the puncture and “maximal” everywhere else.
Given this choice, $T$ and $U$ have exactly the same paths for the top $C-k-1$ paths above the puncture and the bottom $k$ paths below the puncture. Hence we can trim off these paths to make two new tilings, $T'$ and $U'$, of $H' = H_{B+M+1, A+M+1, c, \alpha - (C-k-1), \beta - k, \gamma}$. Notice that $H$ and $H'$ have the same $A, B, M,$ and $\gamma$, only $C, \alpha,$ and $\beta$ have changed; in particular, $C' = 1$. See Figure \[fig:T-U-difference\] parts (i) and (ii) for an example of the tilings $T'$ and $U'$ with their region-of-difference highlighted in bold.
![The punctured hexagon $H_{7,6,9,3,2,3}$; both tilings have the region-of-difference highlighted.[]{data-label="fig:T-U-difference"}](T-U-diff-1 "fig:")\
[*(i) The tiling $T'$*]{}
![The punctured hexagon $H_{7,6,9,3,2,3}$; both tilings have the region-of-difference highlighted.[]{data-label="fig:T-U-difference"}](T-U-diff-2 "fig:")\
[*(ii) The tiling $U'$*]{}
Clearly then $T'$ and $U'$ differ in four ways: (i) the upper path in $T'$ except the small overlap near the end, (ii) the lower path in $U'$, (iii) the position of the bend in the puncture-paths, and (iv) the part past the bend of the bottom puncture-path in $T'$. The difference between $T'$ and $U'$ is exactly $2(A+B+M)+M-1$ tiles; moreover the region-of-difference forms a cycle so that there exists a $(2(A+B+M)+M-1)$-cycle, $\sigma$, such that $\pi_{T'} = \sigma \pi_{U'}$. We then have $$\operatorname{sgn}\pi_{T'} = (-1)^{2(A+B+M)+M-1-1}\operatorname{sgn}\pi_{U'} = (-1)^M\operatorname{sgn}\pi_{U'}.$$ That is, $\operatorname{sgn}\pi_{T} = (-1)^M\operatorname{sgn}\pi_{U}$. Since $\operatorname{sgn}\lambda_T = (-1)^M \operatorname{sgn}\lambda_U$, the claim follows.
We conclude that $Z_{a,b,c,\alpha,\beta,\gamma}$ and $N_{a,b,c,\alpha,\beta,\gamma}$ have the same determinant, up to sign.
\[thm:det-Z-N\] Consider the punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$. Then $$|\det{Z_{a,b,c,\alpha,\beta,\gamma}}| = |\det{N_{a,b,c,\alpha,\beta,\gamma}}|.$$
Combine Theorems \[thm:nilp-matrix\] and \[thm:bip-matrix\] via Proposition \[pro:sign-agree\].
Moreover, when the puncture is of even length, the determinant and permanent of $Z_{a,b,c,\alpha,\beta,\gamma}$ are the same.
\[cor:det-Z-per-Z\] Consider the punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$. If $M$ is even, then $$\operatorname{perm}{Z_{a,b,c,\alpha,\beta,\gamma}} = |\det{Z_{a,b,c,\alpha,\beta,\gamma}}|.$$
A simple analysis of the proof of Proposition \[pro:sign-agree\] implies that when $M$ is even then $\operatorname{sgn}{\pi_T} = \operatorname{sgn}{\pi_U}$ for all tilings $T$ and $U$ of $H_{a,b,c,\alpha,\beta,\gamma}$. Thus, the enumeration of signed lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$ is, up to sign, the enumeration of (unsigned) lozenge tilings of $H_{a,b,c,\alpha,\beta,\gamma}$. Thus, the claim follows from Proposition \[pro:bip-matrix\] and Theorem \[thm:bip-matrix\].
\[rem:det-Z-per-Z\] We make a pair of remarks regarding the preceding corollary.
1. The corollary can be viewed as a special case of Kasteleyn’s theorem on enumerating perfect matchings [@Ka]. To see this, notice that when $M$ is even, then all “faces” of the bipartite graph have size congruent to $2 \pmod{4}$.
2. The corollary extends [@CGJL Theorem 1.2], where punctured hexagons with trivial puncture (i.e., $M = 0$) are considered. We further note that [@Ke Section 3.4] provides, independently, essentially the same proof as [@CGJL], and the proof of Lemma \[lem:perm-rotate\] builds on this technique.
We conclude this section with some observations on the signs introduced here.
Let $T$ be a lozenge tiling of $H_{a,b,c,\alpha,\beta,\gamma}$, and let $L_T$ and $P_T$ be the associated family of non-intersecting lattice paths and perfect matching, respectively. By Proposition \[pro:sign-agree\], we may assume that $\operatorname{sgn}L_T = \operatorname{sgn}P_T$. Thus we may assign to $T$ the sign $\operatorname{sgn}T = \operatorname{sgn}L_T$.
Recall that there are $C$ admissible permutations $\lambda_0, \ldots, \lambda_C$ (see the discussion after Theorem \[thm:nilp-matrix\]) associated to $H_{a,b,c,\alpha,\beta,\gamma}$. Further, $\operatorname{sgn}{\lambda_k} = (-1)^{M(C-k)}$ and so if $M$ is even then $\operatorname{sgn}{\lambda_k} = 1$ for all $k$. Hence, we need only consider $M$ odd. In this case, $\operatorname{sgn}{\lambda_k} = 1$ if and only if $C-k$ is even. Thus, the sign of $T$ is $(-1)^{C-k}$.
![Example of interpreting the sign[]{data-label="fig:interp-signs"}](nilp-sign "fig:")\
[*(i) The sign of a family of non-intersecting lattice paths*]{}
![Example of interpreting the sign[]{data-label="fig:interp-signs"}](tile-sign "fig:")\
[*(ii) The sign of a lozenge tiling*]{}
![Example of interpreting the sign[]{data-label="fig:interp-signs"}](bip-sign "fig:")\
[*(iii) The sign of a perfect matching*]{}
By definition of $\lambda_k$, $C-k$ is the number of lattice paths in the family that go above the puncture; see Figure \[fig:interp-signs\](i). For the lozenge tiling $T$, $C-k$ is the number of edges of lozenges of $T$ that touch the line formed by extending the edge of the puncture parallel to the side of length $C$ to the side of length $A+M$; see Figure \[fig:interp-signs\](ii). Note that this interpretation is in line with the definition of the statistic $n(\cdot)$ in [@CEKZ Section 2]. Last, for the perfect matching, $C-k$ is the number of [*non*]{}-selected edges that correspond to those on the edge described for lozenge tilings; see Figure \[fig:interp-signs\](iii).
Determinants {#sec:det}
============
We continue to use the notation introduced in Proposition \[pro:semistable\] and Theorem \[thm:interlace-amaci\]. Throughout this section we assume that $A, B, C,$ and $M$ meet conditions (i)-(iv) of Proposition \[pro:semistable\] and $a+b+c+\alpha+\beta+\gamma \equiv 0 \pmod{3}$.
We will discuss properties of the determinant of the matrix $N_{a,b,c,\alpha,\beta,\gamma}$ given in Proposition \[pro:wlp-binom\] using Theorem \[thm:nilp-matrix\]. In particular, we are chiefly interested in whether the determinant is zero and if we can compute an upper bound on the prime divisors. In some cases we can explicitly compute the determinant.
A few properties
----------------
First, a brief remark about the polynomial nature of the determinants.
\[rem:polynomial\] The argument in [@CEKZ Section 6] demonstrates that for fixed $A, B,$ and $C$ and $\alpha, \beta,$ and $\gamma$ satisfying certain restraints, then the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$ is polynomial in $M$, the side-length of the puncture of $H_{a,b,c,\alpha,\beta,\gamma}$, for $M$ of a fixed parity. This argument centers around an alternate bijection between the lozenge tilings and non-intersecting lattice paths.
We note that the argument is completely independent of the restrictions on $\alpha, \beta,$ and $\gamma$. Thus, their argument can be easily seen to generalise to show that, for fixed $A,B,C,\alpha,\beta,$ and $\gamma$, the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$ is polynomial in $M$, for $M$ of a fixed parity.
We demonstrate that every punctured hexagonal region $H_{a,b,c,\alpha,\beta,\gamma}$ has at least one tiling.
\[lem:tilings-exist\] Every region $H_{a,b,c,\alpha,\beta,\gamma}$ has at least one lozenge tiling.
In this case, it is easier to show there exists a family $L$ of non-intersecting lattice paths. In particular, it is sufficient to show that the sum of the maximum number of paths that can go above and below the puncture is at least $C$. By analysis of $H_{a,b,c,\alpha,\beta,\gamma}$, we see that at most $\min\{C, \beta, B+C-\alpha\}$ paths can go below the puncture and at most $\min\{C, \alpha, A+C-\beta\}$ paths can go above the puncture. However, as $0 \leq A,B,C$ and $C \leq \alpha + \beta$, then $\min\{C, \beta, B+C-\alpha\} + \min\{C, \alpha, A+C-\beta\} \geq C$.
Thus when $M$ is even, the determinant is always positive.
\[thm:M-even\] If $a+b+c$ is even, then $M$ is even and $\det{N_{a,b,c,\alpha,\beta,\gamma}} > 0$. Thus $$I_{a,b,c,\alpha,\beta,\gamma} = (x^a, y^b, z^c, x^\alpha y^\beta z^\gamma)$$ has the weak Lefschetz property in characteristic zero and when the characteristic is sufficiently large.
Recall the definition of the admissible partitions $\lambda_k$, for $0 \leq k \leq C$ (see the discussion following Theorem \[thm:nilp-matrix\]). Since $M$ is even, then $\operatorname{sgn}{\lambda_k} = 1$ for $0 \leq k \leq C$ and hence $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is the number of tilings of $H_{a,b,c,\alpha,\beta,\gamma}$. Thus, by Lemma \[lem:tilings-exist\], $\det{N_{a,b,c,\alpha,\beta,\gamma}} > 0$.
Mahonian determinants {#ss:macmahon}
---------------------
MacMahon computed the number of plane partitions (finite two-dimensional arrays that weakly decrease in all columns and rows) in an $A \times B \times C$ box as (see, e.g., [@Pr Page 261]) $$\operatorname{Mac}(A,B,C) := \frac{{\mathcal{H}}(A) {\mathcal{H}}(B) {\mathcal{H}}(C) {\mathcal{H}}(A+B+C)}{{\mathcal{H}}(A+B) {\mathcal{H}}(A+C) {\mathcal{H}}(B+C)},$$ where $A,B,$ and $C$ are non-negative integers and ${\mathcal{H}}(n) := \prod_{i=0}^{n-1}i!$ is the [*hyperfactorial*]{} of $n$. David and Tomei proved in [@DT] that plane partitions in an $A \times B \times C$ box are in bijection with lozenge tilings in an non-punctured hexagon of side-lengths $(A,B,C,A,B,C)$. We note that Propp states on [@Pr Page 258] that Klarner was likely the first to have observed this. See Figure \[fig:pp-tile\] for an illustration of the connection.
![An example of a $3 \times 6 \times 5$ plane partition and its associated lozenge tiling (with light grey as the top faces of the boxes)[]{data-label="fig:pp-tile"}](pp-tile-1 "fig:")\
![An example of a $3 \times 6 \times 5$ plane partition and its associated lozenge tiling (with light grey as the top faces of the boxes)[]{data-label="fig:pp-tile"}](pp-tile-2 "fig:")\
We can use MacMahon’s formula to compute the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$ in many cases. Also, note that the prime divisors of $\operatorname{Mac}(A,B,C)$ are sharply bounded above by $A+B+C-1$. A first case is when the puncture is trivial. This extends [@CN Theorem 4.5] where the level algebras of this family are considered.
\[pro:M-zero\] If $a+b+c=2(\alpha+\beta+\gamma)$, then $M = 0$ and $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is $$\operatorname{Mac}(A,B,C).$$ Thus, $I_{a,b,c,\alpha,\beta,\gamma}$ has the weak Lefschetz property if the characteristic of $K$ is zero or at least $A+B+C=\alpha+\beta+\gamma = \frac{1}{2}(a+b+c)$.
![When the puncture has side-length zero, the region is a simple hexagon.[]{data-label="fig:hex-no-puncture"}](hex-no-puncture)
When $M = 0$ then there is no puncture in the region $H_{a,b,c,\alpha,\beta,\gamma}$. Hence the region is a simple hexagon with side-lengths $(A,B,C,A,B,C)$, exactly the region to which MacMahon’s formula applies.
This result allows us to recover some earlier results about complete intersections.
\[cor:ci\] If $a+b+c$ is even, then the complete intersection $J = (x^a, y^b, z^c)$ has the weak Lefschetz property if and only if the characteristic of $K$ is not a prime divisor of $\operatorname{Mac}(A,B,C)$. That is, the algebra $R/J$ has the weak Lefschetz property if the characteristic of $K$ is zero or at least $A+B+C=\alpha+\beta+\gamma = \frac{1}{2}(a+b+c)$.
Set $\alpha = \frac{1}{2}(-a+b+c),$ $\beta = \frac{1}{2}(a-b+c),$ $\gamma = \frac{1}{2}(a+b-c),$ and consider $I = (x^a, y^b, z^c, x^\alpha y^\beta z^\gamma)$. Then Proposition \[pro:M-zero\] applies to $I$ and the mixed term, $x^\alpha y^\beta z^\gamma$, has total degree $s+2$. Thus we have that $[R/I]_i \cong [R/J]_i$ for $i \leq s+1$. That is, the twin peaks of $R/I$ are isomorphic to the twin peaks of the complete intersection $R/J$. Hence $R/J$ has the weak Lefschetz property if and only if $R/I$ has the weak Lefschetz property, and Proposition \[pro:M-zero\] gives the claim.
In particular, the corollary recovers [@LZ Theorem 3.2(1)] when combined with Proposition \[pro:wlp-binom\] and [@CGJL Theorem 1.2] when combined with Corollary \[cor:det-Z-per-Z\]. Further, the special case in [@LZ Theorem 4.2] can be recovered if we set $a = \beta + \gamma, b = \alpha + \gamma,$ and $c = \alpha + \beta$.
MacMahon’s formula can be used again in another special case, when $C = 0$. (Notice if $A$ or $B$ is zero, then we can simply relabel the sides to ensure $C$ is zero.) We notice this extends [@CN Theorem 4.3] where the level algebras of this family are considered.
\[pro:C-zero\] If $c = \frac{1}{2}(a+b+\alpha+\beta+\gamma)$, then $C = 0$ and $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is $$\operatorname{Mac}(M, A-\beta, B-\alpha).$$ Thus, $I_{a,b,c,\alpha,\beta,\gamma}$ has the weak Lefschetz property if the characteristic of $K$ is zero or at least $A+B+M-\alpha-\beta=c-\alpha-\beta$.
![When $C$ is zero, the lightly shaded region has tiles that are fixed, leaving the only variation in the darkly shaded region.[]{data-label="fig:hex-C-zero"}](hex-C-zero)
In this case, it is easier to consider families of non-intersecting lattice paths. In particular, since $C = 0$, then the only starting points, the $A_i$, are those on the puncture. Further, since lattice paths must move only right and down, then we can focus on the isolated region between the puncture and the bottom-right edge. If we convert this region back into a punctured hexagon, then it is just a hexagon without a puncture and with side-lengths $(M, A+C-\beta, B+C-\alpha, M, A+C-\beta, B+C-\alpha)$.
\[rem:C-zero\] Notice that in the preceding proof, we show that the only possible lattice paths come from the puncture to the opposite edge. Converting this back to the language of lozenge tilings, we see this means that a large region of the figure has fixed tiles leaving only a small region in which variation can occur. See Figure \[fig:hex-C-zero\] for an illustration of this.
Further, given the condition in Proposition \[pro:C-zero\], we see that the pure power of $z$, $z^c$, has total degree $c = s+2$. Thus, if we let $I = I_{a,b,c,\alpha,\beta,\gamma}$, then we have that $[R/I]_i \cong [R/J]_i$ for $i \leq s+1$, where $J = (x^a, y^b, x^\alpha y^\beta z^\gamma)$. Thus, the twin peaks of $R/I$ are isomorphic to the twin peaks of the non-artinian algebra $R/J$.
\[cor:injective-pure-z-missing\] Let $J = (x^a, y^b, x^\alpha y^\beta z^\gamma)$ and $c = \frac{1}{2}(a+b+\alpha+\beta+\gamma)$, with parameters still suitably restricted. Then the map $$[R/J]_i \stackrel{\times (x+y+z)}{\longrightarrow} [R/J]_{i+1}$$ is injective for $i \leq c$.
Further, MacMahon’s formula can be used when $C$ is maximal, that is, $C = \alpha + \beta$.
\[pro:C-maximal\] If $c = \frac{1}{2}(a+b+\gamma) - \alpha-\beta$, then $C = \alpha + \beta$ and $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is $$\operatorname{Mac}(A,B,C+M).$$ Thus, $I_{a,b,c,\alpha,\beta,\gamma}$ has the weak Lefschetz property if the characteristic of $K$ is zero or at least $A+B+C+M=s+2=c+\alpha+\beta$.
In this case, it is easier to consider families of non-intersecting lattice paths. In particular, since $C = \alpha + \beta$, then $\gamma = A+B$ and so the puncture has a point touching the side labeled $C$; see Figure \[fig:hex-C-maximal\].
![When $C$ is maximal, the lightly shaded region has tiles which are fixed, leaving the only variation in the darkly shaded region.[]{data-label="fig:hex-C-maximal"}](hex-C-maximal)
Thus the lattice paths starting from $A_1, \ldots, A_\beta$ have the first $M$ moves being down and the lattice paths starting from $A_{\beta+1}, \ldots, A_C$ have the first $M$ moves being right. However, we then see that each $A_i$ “starts” on the same line, the line running through the lower-right side of the puncture. If we convert the region-of-interest back into a punctured hexagon, then it is a simple hexagon with side-lengths $(A, B, C+M, A, B, C+M)$.
The next case considered, when the mixed term is in two variables, needs a special determinant calculation which may be of independent interest.
\[lem:split-binom-det\] Let $T$ be an $n$-by-$n$ matrix defined as follows $$(T)_{i,j} = \left\{
\begin{array}{ll}
\displaystyle \binom{p}{q + j - i} & \mbox{if } 1 \leq j \leq m, \\[0.8em]
\displaystyle \binom{p}{q + r + j - i} & \mbox{if } m + 1 \leq j \leq n, \\
\end{array}
\right.$$ where $p,q,r,$ and $m$ are non-negative integers and $1 \leq m \leq n$. Then $$\det{T} = \operatorname{Mac}(m,q,r) \operatorname{Mac}(n-m, p-q-r, r) \frac{{\mathcal{H}}(q+r){\mathcal{H}}(p-q){\mathcal{H}}(n+r){\mathcal{H}}(n+p)}{{\mathcal{H}}(n+p-q){\mathcal{H}}(n+q+r){\mathcal{H}}(p){\mathcal{H}}(r)}.$$
In this case, we can use [@CEKZ Equation (12.5)] to evaluate $\det{T}$ to be $$\prod_{1\leq i < j \leq n} (L_j - L_i) \prod_{i=1}^n \frac{(p+i-1)!}{(n+p-L_i)!(L_i-1)!},$$ where $L_j = q+j$ if $1 \leq j \leq m$ and $L_j = q+r+j$ if $m + 1 \leq j \leq n$. If we split the products in the previously displayed equation relative to the split in $L_j$, then we get the following equations: $$\begin{split}
\prod_{1\leq i < j \leq n} (L_j - L_i)
= & \left(\prod_{1\leq i < j \leq m} (j - i)\right) \left(\prod_{m< i < j \leq n} (j - i)\right) \left(\prod_{1\leq i \leq m < j \leq n} (r+j-i)\right) \\[0.3em]
= & \left({\mathcal{H}}(m)\right) \left({\mathcal{H}}(n-m)\right) \left(\frac{{\mathcal{H}}(n+r) {\mathcal{H}}(r)}{{\mathcal{H}}(n+r-m) {\mathcal{H}}(m+r)}\right)
\end{split}$$ and $$\begin{split}
\prod_{i=1}^n \frac{(p+i-1)!}{(n+p-L_i)!(L_i-1)!}
= & \left( \prod_{i=1}^{n}(p+i-1)! \right) \left( \prod_{i=1}^m \frac{1}{(n+p-q-i)!(q+i-1)!}\right) \\[0.3em]
& \left( \prod_{i=m+1}^{n} \frac{1}{(n+p-q-r-i)!(q+r+i-1)!}\right) \\[0.3em]
= & \left( \frac{{\mathcal{H}}(n+p)}{{\mathcal{H}}(p)} \right) \left( \frac{{\mathcal{H}}(n+p-m-q){\mathcal{H}}(q)}{{\mathcal{H}}(n+p-q){\mathcal{H}}(m+q)} \right) \\[0.3em]
& \left( \frac{{\mathcal{H}}(p-q-r){\mathcal{H}}(m+q+r)}{{\mathcal{H}}(n+p-m-q-r){\mathcal{H}}(n+q+r)} \right).
\end{split}$$ Bringing these equations together we have that $\det{T}$ is [$$\frac{{\mathcal{H}}(m){\mathcal{H}}(q){\mathcal{H}}(r){\mathcal{H}}(m+q+r)}{{\mathcal{H}}(m+r){\mathcal{H}}(m+q)} \frac{{\mathcal{H}}(n-m){\mathcal{H}}(p-q-r){\mathcal{H}}(n+p-m-q)}{{\mathcal{H}}(n+r-m){\mathcal{H}}(n+p-m-q-r)} \frac{{\mathcal{H}}(n+r){\mathcal{H}}(n+p)}{{\mathcal{H}}(p){\mathcal{H}}(n+p-q){\mathcal{H}}(n+q+r)},$$ ]{} which, after minor manipulation, yields the claimed result.
\[rem:split-binom-det\] Lemma \[lem:split-binom-det\] generalises the result of [@LZ Lemma 2.2] where the case $r = 1$ is discussed. Further, when $r = 0$, then $\det{T} = \operatorname{Mac}(n, p-q, q)$, as expected (see the running example, $\det \binom{a+b}{a-i+j}$, in [@Kr]).
The case when the mixed term has only two variables follows immediately.
\[pro:gamma-zero\] If $\gamma = 0$, then $|\det{N_{a,b,c,\alpha,\beta,\gamma}}|$ is $$\operatorname{Mac}(\beta-A,A,M) \operatorname{Mac}(\alpha - B, B, M) \frac{{\mathcal{H}}(A+M){\mathcal{H}}(B+M){\mathcal{H}}(C+M){\mathcal{H}}(A+B+C+M)}{{\mathcal{H}}(a){\mathcal{H}}(b){\mathcal{H}}(c){\mathcal{H}}(M)}.$$ Thus, the type $2$ ideal $$I_{a,b,c,\alpha,\beta,0} = (x^a, y^b, z^c, x^\alpha y^\beta)$$ has the weak Lefschetz property if the characteristic of $K$ is zero or at least $A+B+C+M$.
![When $\gamma$ is zero, the starting points $A_{C+1}, \ldots, A_{C+M}$ coincide with the $M$ consecutive ending points $E_{A-\beta+1}, \ldots, E_{A-\beta+M}$.[]{data-label="fig:hex-gamma-zero"}](hex-gamma-zero)
As $\gamma = 0$, $N = N_{a,b,c,\alpha,\beta,\gamma}$ has entries given by $$(N)_{i,j} = \left\{
\begin{array}{ll}
\binom{c}{A + j - i} & \mbox{if } 1 \leq i \leq C, \\[0.3em]
\left\{
\begin{array}{ll}
1 & \mbox{if } j = i + \beta - A - C \\
0 & \mbox{if } j \neq i + \beta -A - C\\
\end{array}
\right\} & \mbox{if } C + 1 \leq i \leq C+M.
\end{array}
\right..$$ Further, if we define the matrix $T$ by $$(T)_{i,j} = \left\{
\begin{array}{ll}
\binom{c}{A + j - i} & \mbox{if } 1 \leq j \leq \beta - A, \\[0.3em]
\binom{c}{A + M + j - i} & \mbox{if } \beta - A + 1 \leq j \leq C \\
\end{array}
\right.,$$ then $|\det{N}| = |\det{T}|$ due to the structure of the lower-part of $N$. Thus, if we let $p = c, q = A, r = M, m = \beta - A,$ and $n = C$, then by Lemma \[lem:split-binom-det\] we have the desired determinant evaluation.
Moreover, $\alpha + M$ and $\beta + M$ are smaller than $A+B+C+M$, so the prime divisors of $\det{N}$ are strictly bounded above by $A+B+C+M$.
\[rem:gamma-zero\] Proposition \[pro:gamma-zero\] deserves a pair of comments:
1. The evaluation of the determinant includes two Mahonian terms and a third non-Mahonian term. It should be noted that both hexagons associated to the Mahonian terms actually show up in the punctured hexagon.
![The darkly shaded hexagons correspond to the two Mahonian terms in the determinantal evaluation.[]{data-label="fig:hex-gamma-zero-highlight"}](hex-gamma-zero-highlight)
See Figure \[fig:hex-gamma-zero-highlight\] where the darkly shaded hexagons correspond to the Mahonian terms. It is not clear (to us) where the third term comes from, though it may be of interest to note that if one subtracts $M$ from each hyperfactorial, before the evaluation, then what remains is $\operatorname{Mac}(A,B,C)$.
2. We notice the proposition also extends [@MMN Lemma 6.6] where it was shown that the associated almost complete intersection always has the weak Lefschetz property in characteristic zero (i.e., the determinant is non-zero). That is, all level type $2$ artinian monomial almost complete intersections in $R$ have the weak Lefschetz property in characteristic zero.
Exploring symmetry
------------------
When $a = b$ (equivalently, $A = B$) and $\alpha = \beta$, then $H_{a,a,c,\alpha,\alpha,\gamma}$ is symmetric; see Figure \[fig:hex-symmetric\]. In this case, $c$ is even exactly when $M = \frac{1}{3}(2a+c-4\alpha-2\gamma)$ is even; similarly, $\gamma$ is even exactly when $C = \frac{1}{3}(2a - 2c + 2\alpha + \gamma)$ is even. Moreover, $\alpha = A + \frac{1}{2}(C-\gamma)$.
![When $a=b$ and $\alpha = \beta$, then $H_{a,a,c,\alpha,\alpha,\gamma}$ is symmetric.[]{data-label="fig:hex-symmetric"}](hex-symmetric)
When $C$ and $M$ are odd, we can exploit symmetry to show $\det{N_{a,a,c,\alpha,\alpha,\gamma}}$ is $0$. This result extends the evaluation in [@MMN Corollary 7.4] and offers a (more) direct combinatorial proof, rather than one based on linear algebra.
\[pro:symmetry-zero\] If $c$ and $\gamma$ are odd, $a=b$, and $\alpha = \beta$, then $H_{a,a,c,\alpha,\alpha,\gamma}$ is symmetric with an odd puncture (i.e., $M$ odd; see Figure \[fig:hex-symmetric\]) and $\det{N_{a,a,c,\alpha,\alpha,\gamma}}$ is 0. Thus, $$I_{a,a,c,\alpha,\alpha,\gamma} = (x^a, y^a, z^c, x^\alpha y^\alpha z^\gamma)$$ never has the weak Lefschetz property, regardless of characteristic.
Recall the admissible partitions of $H_{a,a,c,\alpha,\alpha,\gamma}$ are $\lambda_0, \ldots, \lambda_C$. For $0 \leq i \leq \frac{C-1}{2}$ we see that $P^+_{\lambda_i}(A \rightarrow E) = P^+_{\lambda_{C-i}}(A \rightarrow E)$ by symmetry, and further that $\operatorname{sgn}{\lambda_{i}} = -\operatorname{sgn}{\lambda_{C-i}}$, as $\operatorname{sgn}{\lambda_k} = (-1)^{M(C-k)}$ and $C$ is odd. Hence, $\det{N_{a,b,c,\alpha,\beta,\gamma}} = \sum_{i = 0}^C{\operatorname{sgn}{\lambda_i} P^+_{\lambda_i}(A \rightarrow E)} = 0.$
From the preceding proof we see that if we consider $c$ even instead of $c$ odd (i.e., $M$ even instead of $M$ odd), then $\det{N_{a,a,c,\alpha,\alpha,\gamma}}$ is even, when $\gamma$ is odd (i.e., $C$ is odd).
Recall the definitions of $A, B, C,$ and $M$ from Proposition \[pro:semistable\], $H_{a,a,c,\alpha,\alpha,\gamma}$ from Theorem \[thm:interlace-amaci\], and $N_{a,b,c,\alpha,\beta,\gamma}$ from Proposition \[pro:wlp-binom\]. If $C$ or $M$ is even, then the region $H_{a,a,c,\alpha,\alpha,\gamma}$ is symmetric and we offer the following conjecture for a closed form for $\det{N_{a,a,c,\alpha,\alpha,\gamma}}$. Notice that in this case $\alpha = A + \frac{1}{2}(C-\gamma)$.
\[con:symmetry\] Suppose $a=b$ and $\alpha = \beta$ so $H_{a,a,c,\alpha,\alpha,\gamma}$ is symmetric. If $c$ or $\gamma$ is even, then $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is $$(-1)^{M{\left\lceil\frac{C}{2}\right\rceil}}
\times \frac{ {\mathcal{H}}(M+C) {\mathcal{H}}(M+\gamma) {\mathcal{H}}(M+A+{\left\lfloor\frac{C}{2}\right\rfloor}) {\mathcal{H}}(M + A + {\left\lceil\frac{C}{2}\right\rceil}) {\mathcal{H}}(M+2A+C) }
{ {\mathcal{H}}(M+2A) {\mathcal{H}}^2(M+A+C) {\mathcal{H}}^2(M + \frac{C+\gamma}{2}) }$$ $$\times \frac{ {\mathcal{H}}({\left\lfloor\frac{M}{2}\right\rfloor}) {\mathcal{H}}({\left\lfloor\frac{M}{2}\right\rfloor} + A) {\mathcal{H}}({\left\lfloor\frac{M}{2}\right\rfloor} + \frac{C+\gamma}{2}) {\mathcal{H}}({\left\lfloor\frac{M}{2}\right\rfloor} + A + \frac{C-\gamma}{2}) }
{ {\mathcal{H}}({\left\lfloor\frac{M+C}{2}\right\rfloor}) {\mathcal{H}}({\left\lfloor\frac{M+\gamma}{2}\right\rfloor}) {\mathcal{H}}({\left\lfloor\frac{M+C}{2}\right\rfloor} + A) {\mathcal{H}}({\left\lfloor\frac{M-\gamma}{2}\right\rfloor} + A) }$$ $$\times \frac{ {\mathcal{H}}({\left\lceil\frac{M}{2}\right\rceil}) {\mathcal{H}}({\left\lceil\frac{M}{2}\right\rceil} + A) {\mathcal{H}}({\left\lceil\frac{M}{2}\right\rceil} + \frac{C+\gamma}{2}) {\mathcal{H}}({\left\lceil\frac{M}{2}\right\rceil} + A + \frac{C-\gamma}{2}) }
{ {\mathcal{H}}({\left\lceil\frac{M+C}{2}\right\rceil}) {\mathcal{H}}({\left\lceil\frac{M+\gamma}{2}\right\rceil}) {\mathcal{H}}({\left\lceil\frac{M+C}{2}\right\rceil} + A) {\mathcal{H}}({\left\lceil\frac{M-\gamma}{2}\right\rceil} + A) }$$ $$\times \frac{ {\mathcal{H}}(A - {\left\lfloor\frac{\gamma}{2}\right\rfloor}) {\mathcal{H}}({\left\lfloor\frac{C}{2}\right\rfloor}) {\mathcal{H}}({\left\lfloor\frac{\gamma}{2}\right\rfloor}) {\mathcal{H}}(A - {\left\lceil\frac{\gamma}{2}\right\rceil}) {\mathcal{H}}({\left\lceil\frac{C}{2}\right\rceil}) {\mathcal{H}}({\left\lceil\frac{\gamma}{2}\right\rceil}) }
{ {\mathcal{H}}(\gamma) {\mathcal{H}}^2(A + \frac{C-\gamma}{2}) }.$$
Further, the ideal $$I_{a,a,c,\alpha,\alpha,\gamma} = (x^a, y^a, z^c, x^\alpha y^\alpha z^\gamma)$$ has the weak Lefschetz property when the characteristic of $K$ is zero or at least $2A+C+M$.
\[rem:symmetry\] The above symmetry conjecture deserves a few remarks.
1. Note that by Remark \[rem:polynomial\], $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is polynomial in $M$. Further, the conjectured form of the determinant would imply that the polynomial factors completely into linear terms and has degree $AC + \left\lfloor \frac{\gamma}{2}(C-\frac{\gamma}{2}) \right\rfloor$.
2. If Conjecture \[con:symmetry\] were shown to hold, then it would complete the $(-1)$-enumeration of symmetric punctured hexagons when combined with Proposition \[pro:symmetry-zero\],
3. As expected, the conjecture corresponds to Proposition \[pro:C-zero\] when $C = 0$, to Proposition \[pro:C-maximal\] when $A = \frac{1}{2}\gamma$ (this implies $\alpha = \frac{1}{2}C$ and so $C = 2\alpha$, which is maximal), and to Proposition \[pro:gamma-zero\] when $\gamma = 0$. Moreover, when $A=C=\gamma$, then $H_{a,a,c,\alpha,\alpha,\gamma}$ has an [*axis-central*]{} puncture (see Section \[sub:axis\]) and the conjecture corresponds to Corollary \[cor:axis-central-det-Z\].
4. When $C$ is even and $M$ is odd, then using the $f_{a,b}(c)$ from Proposition \[pro:hyper-f\] and $f^e_{a,b}(c)$ and $f^o_{a,b}(c)$ from Corollary \[cor:hyper-eo-f\], we can rewrite the monic (as a polynomial in $M$) part of the conjecture as
$
\frac{f^o_{\frac{C+\gamma}{2}, \frac{C+\gamma}{2}}(M)
\cdot f^e_{\frac{|C-\gamma|}{2},\frac{|C-\gamma|}{2}}(M+\min(C,\gamma))
\cdot f^e_{\min(C, \gamma), \max(C, \gamma)}(M+2A-\gamma)
\cdot f_{|A-\gamma|, |A-\gamma|}(M+C-\gamma + 2\min(A, \gamma))}
{f_{\left|\frac{1}{2}C+\gamma-A\right|, \left|\frac{1}{2}C+\gamma-A\right|}(M+\min(2A-\gamma, C+\gamma))}.
$
(Carefully note that the input parameter in each of the polynomials $f$ above is odd as $M$ is odd.)
We give an example of using the symmetry conjecture.
\[exa:symmetry\] Consider $A = B = 8$, $C = 6$, $\gamma = 2$, and $M$ even. Then $\alpha = \beta = 10$, $a = b = 14 + M$, and $c = 16 + M$. Moreover, the region $H_{14+M, 14+M, 16+M, 10, 10, 2}$ is symmetric and does not fall into the case of Remark \[rem:symmetry\](iii).
Supposing Conjecture \[con:symmetry\] holds, then $H_{14+M, 14+M, 16+M, 10, 10, 2}$ has a $(-1)$-enumeration of $$\frac{1}{-2^{34} 3^{16} 5^6 7^6} \times (M+1)(M+3)^3(M+4)^2(M+5)^3(M+7)$$ $$\times (M+12)^2(M+13)^4(M+14)^6(M+15)^5(M+16)^6(M+17)^3(M+18)^4(M+19)(M+20)^2.$$ Thus, $I_{14+M, 14+M, 16+M, 10, 10, 2} = (x^{14+M}, y^{14+M}, z^{16+M}, x^{10} y^{10} z^2)$ has the weak Lefschetz property when the characteristic of the ground field is $0$ or at least $M+21$.
So far, in every case where we can bound the prime divisors of $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ from above, we can do so linearly in the parameters (actually, always by at most $s+2$). This may, however, not always be the case. We provide the following example to demonstrate that this is true, but also as a contrast to the symmetry conjecture, where some restrictions lead to a (conjectured) closed form.
\[exa:non-linear-bound\] Consider the level and type $3$ algebra given by $R/I$, where $$I_{1+t, 4+t, 7+t, 1,4,7} = (x^{1+t}, y^{4+t}, z^{7+t}, x y^4 z^7)$$ and $t \geq 4$. By Remark \[rem:polynomial\], we have that $\det{N} = \det{N_{1+t, 4+t, 7+t, 1,4,7}}$ is a polynomial in $t$. Hence we can use interpolation to determine the polynomial in terms of $t$; in particular, $\det{N_{1+t, 4+t, 7+t, 1,4,7}}$ is $$\frac{4}{{\mathcal{H}}(7)} \cdot \left\{ \begin{array}{ll}
(t-3) (t-2) (t-1)^3 t^3 (t+1)^2 (t+2) (t+4) (t+6) (t^2 + 6t - 1) & \mbox{ if $t$ is odd;} \\[0.3em]
(t-2)^2 (t-1)^2 t^4 (t+1)^2 (t+2) (t+5) (t+7) (t^2 + 2t - 9) & \mbox{ if $t$ is even.}
\end{array} \right.$$
In 1857, Bouniakowsky conjectured that for every irreducible polynomial $f \in {\mathbb{Z}}[t]$ of degree at least 2 with common divisor $d = \gcd\{f(i) {\; | \;}i \in {\mathbb{Z}}\}$, there exists infinitely many integers $t$ such that $\frac{1}{d}f(t)$ is prime. We note that the weaker Fifth Hardy-Littlewood conjecture, which states that $t^2 + 1$ is prime for infinitely many positive integers $t$, is a special case of the Bouniakowsky conjecture.
When $t$ is odd, the determinant has the quadratic factor $t^2 + 6t - 1$. If we let $t = 2k+1$, then this factor becomes $2(2k^2 + 8k +3)$, which is an irreducible polynomial over ${\mathbb{Z}}[k]$ with common divisor $2$ (when $k = 4$ then the polynomial evaluates to $134 = 2\cdot 67$). Hence the quadratic factor of the determinant is prime for infinitely many odd integers $t$, assuming the Bouniakowsky conjecture. Similarly the quadratic factor of the determinant for $t$ even is prime for infinitely many even integers $t$, again assuming the Bouniakowsky conjecture.
Hence, assuming the Bouniakowsky conjecture, for large enough $t$, the upper bound on the prime divisors of the determinant grows quadratically in $t$.
The above example falls in to the case of Proposition \[pro:level-wlp\](ii)(a) or the second open case immediately following the proposition, depending on the parity of $t$.
Centralising the puncture {#sec:central}
=========================
In this section we consider two subtlety different ways to centralise the puncture of a punctured hexagon. The first, [*axis-central*]{}, forces the puncture to be centered along each axis, individually. The second, [*gravity-central*]{}, forces the puncture to be the same distance, simultaneously, from the three sides of the hexagon that are parallel to the puncture-sides.
Throughout this section we assume, in addition to the conditions in Proposition \[pro:semistable\] and $a+b+c+\alpha+\beta+\gamma \equiv 0 \pmod{3}$, that $I_{a,b,c,\alpha,\beta,\gamma}$ has type $3$, that is, $\alpha$, $\beta$, and $\gamma$ are non-zero.
Axis-central {#sub:axis}
------------
We define a punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$ to have an [*axis-central*]{} puncture if the puncture is “central” as defined in [@CEKZ Section 1]. Specifically, for each side of the puncture, the puncture-side should be the same distance from the parallel hexagon-side as the puncture-vertex opposite the puncture-side is from the other parallel hexagon-side; see Figure \[fig:axis-central\](i). However, when $c$ has a different parity than both $a$ and $b$, then an adjustment has to be made; in particular, translate the puncture parallel to the hexagon-side of length $C$ one-half unit toward the side of length $A$; see Figure \[fig:axis-central\](b).
![A punctured hexagon with an axis-central puncture.[]{data-label="fig:axis-central"}](axis-central-parity-same "fig:")\
[*(i) The parity of $c$ agrees with $a$ and $b$.*]{}
![A punctured hexagon with an axis-central puncture.[]{data-label="fig:axis-central"}](axis-central-parity-diff "fig:")\
[*(ii) The parity of $c$ differs from $a$ and $b$.*]{}
When $H_{a,b,c,\alpha,\beta,\gamma}$ has an axis-central puncture, then the ideal has a nice form. Suppose first that $a$, $b$, and $c$ have the same parity. Then $\alpha = a - M - \alpha$ so $a = 2\alpha + M$; similarly, $b = 2\beta + M$ and $c = 2\gamma + M$. Thus, if we set $t = M$, then $$I_{2\alpha+t,2\beta+t, 2\gamma+t, \alpha,\beta,\gamma} = (x^{2\alpha+t}, y^{2\beta+t}, z^{2\gamma+t}, x^\alpha y^\beta z^\gamma).$$ The conditions in Proposition \[pro:semistable\] simplify to $\alpha \leq \beta + \gamma$, $\beta \leq \alpha + \gamma$, $\gamma \leq \alpha + \beta$, and $t \geq 0$.
Now, suppose the parity of $c$ differs from that of both $a$ and $b$. Then $\alpha = a - M - \alpha + 1$, $\beta = b - M - \beta - 1$, and $\gamma = c- M - \gamma$, so $a = 2\alpha + M-1$, $b = 2\beta +M+1$, and $c = 2\gamma +M$. Thus, if we set $t = M$, then $$I_{2\alpha+t-1,2\beta+t+1, 2\gamma+t, \alpha,\beta,\gamma} = (x^{2\alpha+t-1}, y^{2\beta+t+1}, z^{2\gamma+t}, x^\alpha y^\beta z^\gamma).$$ The conditions in Proposition \[pro:semistable\] simplify to $\alpha \leq \beta + \gamma+1$, $\beta \leq \alpha + \gamma-1$, $\gamma \leq \alpha + \beta$, and $t \geq 0$.
Much to our fortune, the determinants of $N_{a,b,c,\alpha,\beta,\gamma}$ have been calculated for punctured hexagons with axis-central punctures. We recall the four theorems here, although we forgo the exact statements of the determinant evaluations; the explicit evaluations can be found in [@CEKZ].
[[@CEKZ Theorems 1, 2, 4, & 5]]{} \[thm:CEKZ-1245\] Let $A, B, C,$ and $M$ be non-negative integers and let $H$ be the associated hexagon with an axis-central puncture. Then
1. The number of lozenge tilings of $H$ is $\operatorname{CEKZ}_1(A,B,C,M)$ if $A, B,$ and $C$ share a common parity.
2. The number of lozenge tilings of $H$ is $\operatorname{CEKZ}_2(A,B,C,M)$ if $A, B,$ and $C$ do not share a common parity.
3. The number of signed lozenge tilings of $H$ is
1. $\operatorname{CEKZ}_4(A,B,C,M)$ if $A, B,$ and $C$ are all even, and
2. $0$ if $A, B,$ and $C$ are all odd.
4. The number of signed lozenge tilings of $H$ is $\operatorname{CEKZ}_5(A,B,C,M)$ if $A, B,$ and $C$ do not share a common parity.
Moreover, the four functions $\operatorname{CEKZ}_i$ are polynomials in $M$ which factor completely into linear terms. Further, each can be expressed as a quotient of products of hyperfactorials and, in each case, the largest hyperfactorial term is ${\mathcal{H}}(A+B+C+M)$.
Thus, we calculate the permanent of $Z_{a,b,c,\alpha,\beta,\gamma}$.
\[cor:axis-central-perm-Z\] Let $H_{a,b,c,\alpha,\beta,\gamma}$ be a hexagon with an axis-central puncture. Then $$\operatorname{perm}{Z_{a,b,c,\alpha,\beta,\gamma}} = \left\{ \begin{array}{ll}
\operatorname{CEKZ}_1(A,B,C,M) & \mbox{if $a,b,$ and $c$ share a common parity;} \\[0.3em]
\operatorname{CEKZ}_2(A,B,C,M) & \mbox{otherwise.}
\end{array} \right.$$
This follows from Proposition \[pro:bip-matrix\] and Theorem \[thm:CEKZ-1245\].
Moreover, we calculate the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$, and thus can completely classify when the algebra $R/I_{a,b,c,\alpha,\beta,\gamma}$ has the weak Lefschetz property.
\[cor:axis-central-det-Z\] Let $H_{a,b,c,\alpha,\beta,\gamma}$ be a hexagon with an axis-central puncture. If $M$ is even, then $$\det{N_{a,b,c,\alpha,\beta,\gamma}} = \left\{ \begin{array}{ll}
\operatorname{CEKZ}_1(A,B,C,M) & \mbox{if $a,b,$ and $c$ share a common parity;} \\[0.3em]
\operatorname{CEKZ}_2(A,B,C,M) & \mbox{otherwise.}
\end{array} \right.$$ If $M$ is odd, then $$\det{N_{a,b,c,\alpha,\beta,\gamma}} = \left\{ \begin{array}{ll}
\operatorname{CEKZ}_4(A,B,C,M) & \mbox{if $a,b,c,$ and $s+2$ share a common parity;} \\[0.3em]
\raisebox{-0.6em}{0} & \mbox{if $a,b,$ and $c$ share a common parity} \\[-0.25em] & \mbox{different from the parity of $s+2$;} \\[0.3em]
\operatorname{CEKZ}_5(A,B,C,M) & \mbox{if $a,b,$ and $c$ do not share a common parity.}
\end{array} \right.$$
Thus, $R/I_{a,b,c,\alpha,\beta,\gamma}$ always fails to have the weak Lefschetz property if $a,b,c,$ and $M$ are odd, regardless of the field characteristic. Otherwise, $R/I_{a,b,c,\alpha,\beta,\gamma}$ has the weak Lefschetz property if the field characteristic is zero or at least $A+B+C+M$.
This follows from Theorem \[thm:nilp-matrix\] and Theorem \[thm:CEKZ-1245\].
As we will see in the following subsection, having a gravity-central puncture is equivalent to the associated algebra being level.
\[que:axis-central-algebraically\] Consider the punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$. Is there an algebraic property $P$ of algebras such that $H_{a,b,c,\alpha,\beta,\gamma}$ has an axis-central puncture if and only if $R/I_{a,b,c,\alpha,\beta,\gamma}$ has property $P$?
Gravity-central
---------------
We define a punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$ to have a [*gravity-central*]{} puncture if the vertices of the puncture are each the same distance from the perpendicular side of the hexagon; see Figure \[fig:gravity-central\].
![A punctured hexagon with a gravity-central puncture.[]{data-label="fig:gravity-central"}](gravity-central)
That is, we have that $B+C-\alpha = A+C -\beta = A+B - \gamma$, which simplifies to the relation $a-\alpha = b-\beta = c-\gamma$, and this is exactly the condition in Proposition \[pro:amaci-props\](ii) for $R/I_{a,b,c,\alpha,\beta,\gamma}$ to be level and type $3$. Thus, if we let $t$ be this common difference, then we can rewrite $I_{a,b,c,\alpha,\beta,\gamma}$ as $$I_{\alpha+t,\beta+t, \gamma+t, \alpha,\beta,\gamma} = (x^{\alpha+t}, y^{\beta+t}, z^{\gamma+t}, x^\alpha y^\beta z^\gamma).$$ Without loss of generality, assume $0 < \alpha \leq \beta \leq \gamma$. Then the conditions in Proposition \[pro:semistable\] simplify to $t \geq \frac{1}{3}(\alpha+\beta+\gamma)$ and $\gamma \leq 2(\alpha + \beta)$.
The ideals $I_{\alpha+t,\beta+t,\gamma+t,\alpha,\beta,\gamma}$ are studied extensively in [@MMN Sections 6 & 7]. In particular, [@MMN Conjecture 6.8] makes a guess as to when $R/I_{\alpha+t,\beta+t,\gamma+t,\alpha,\beta,\gamma}$ has the weak Lefschetz property in characteristic zero. We recall the conjecture here, though we present it in a different but equivalent form.
\[conj:level-wlp\] Consider the ideal $I_{\alpha+t,\beta+t, \gamma+t, \alpha,\beta,\gamma}$ in $R$ where $K$ has characteristic zero, $0 < \alpha \leq \beta \leq \gamma \leq 2(\alpha+\beta)$, $t \geq \frac{1}{3}(\alpha+\beta+\gamma)$, and $\alpha + \beta + \gamma$ is divisible by three.
If $(\alpha,\beta,\gamma,t)$ is not $(2,9,13,9)$ or $(3,7,14,9)$, then $R/I_{\alpha+t,\beta+t, \gamma+t, \alpha,\beta,\gamma}$ fails to have the weak Lefschetz property if and only if $t$ is even, $\alpha + \beta + \gamma$ is odd, and $\alpha = \beta$ or $\beta = \gamma$.
[@MMN Conjecture 6.8] is presented in a format that does not elucidate the reasoning behind it. We present the conjecture differently so it says that the weak Lefschetz property fails in two exceptional cases and also when a pair of parity conditions and a symmetry condition hold.
We add further support to the conjecture.
\[pro:level-wlp\] Let $I = I_{\alpha+t,\beta+t, \gamma+t, \alpha,\beta,\gamma}$ be as in Conjecture \[conj:level-wlp\]. Then
1. $R/I$ fails to have the weak Lefschetz property when $t$ is even, $\alpha + \beta + \gamma$ is odd, and $\alpha = \beta$ or $\beta = \gamma$;
2. $R/I$ has the weak Lefschetz property when
1. $t$ and $\alpha + \beta + \gamma$ have the same parity, or
2. $t$ is odd and $\alpha = \beta = \gamma$ is even.
Part (i) follows from Proposition \[pro:symmetry-zero\] (also by [@MMN Corollary 7.4]). Part (ii)(a) implies $M$ is even and so follows by Theorem \[thm:M-even\]. Part (ii)(b) follows from [@CEKZ Theorem 4], which is recalled here in Theorem \[thm:CEKZ-1245\](4)(i).
We note that Conjecture \[conj:level-wlp\] remains open in two cases, both of which are conjectured to have the weak Lefschetz property:
1. $t$ even, $\alpha + \beta + \gamma$ odd, and $\alpha < \beta < \gamma$;
2. $t$ odd, $\alpha + \beta + \gamma$ even, and $\alpha < \beta$ or $\beta < \gamma$.
\[rem:level-symmetry\] Notice that the second open case in the above statement is solved for the cases when $\alpha = \beta$ or $\beta = \gamma$ if Conjecture \[con:symmetry\] is true.
Axis- and gravity-central
-------------------------
Suppose $a,b,$ and $c$ have the same parity. Then the punctured hexagons that are both axis- and gravity-central are precisely those with $a = b = c = \alpha + t$ and $\alpha = \beta = \gamma$. In this case, we strengthen [@MMN Corollary 7.6].
\[cor:a-t\] Consider the level, type $3$ algebra $A$ given by $$R/I_{\alpha + t, \alpha + t, \alpha + t, \alpha, \alpha, \alpha} = R/(x^{\alpha + t}, y^{\alpha + t}, z^{\alpha + t}, x^\alpha y^\alpha z^\alpha),$$ where $t \geq \alpha$. Then $A$ fails to have the weak Lefschetz property in characteristic zero if and only if $\alpha$ is odd and $t$ is even.
In [@Kr-DPP], Krattenthaler described a bijection between cyclically symmetric lozenge tilings of the punctured hexagon considered in the previous corollary and descending plane partitions with specified conditions.
If $c$ has a different parity than $a$ and $b$, then $\alpha - 1 = \beta + 1 = \gamma$. Thus for $\alpha \geq 3$ and $M$ non-negative we have that the ideals of the form $$I_{2\alpha + M, 2\alpha + M-2, 2\alpha + M-1, \alpha, \alpha, \alpha} = (x^{2\alpha + M}, y^{2\alpha + M-2}, z^{2\alpha + M-1}, x^\alpha y^{\alpha-2} z^{\alpha-1}),$$ are precisely those that are both axis- and gravity-central.
Interesting families and examples {#sec:interesting}
=================================
In this section, we give several interesting families and examples.
Large prime divisors
--------------------
Throughout the two preceding sections, when we could bound the prime divisors of $\det{N}$ above, we bounded them above by (at most) $s+2$. However, this need not always be the case, as demonstrated in Example \[exa:non-linear-bound\]. We provide here a few exceptional-looking though surprisingly common cases.
\[exa:large-primes\] Recall that $s+2 = \frac{1}{3}(a+b+c+\alpha+\beta+\gamma)$. In each case, we specify the parameter set by a sextuple $(a,b,c,\alpha,\beta,\gamma)$.
1. Consider the parameter set $(4,6,6,1,1,3)$. Then $s+2 = 7$ and $\det{N} = 11$. This is the smallest $s+2$ so that $\det{N}$ has a prime divisor greater than $s+2$.
2. For the parameter set $(20, 20, 20, 3, 8, 13)$, we get $s+2 = 28$ and $$\det{N} = 2\cdot 3^{2}\cdot 5^{3}\cdot 7\cdot 11\cdot 17^{2}\cdot 19^{6}\cdot 23^{5}\cdot 20554657.$$ Hence $\det{N}$ is divisible by a prime that is over $700000$ times large than $s+2$. Moreover, $20554657$ is greater than the multiplicity of the associated algebra.
3. Consider the parameter set $(7,12,13,1,7,2)$. Then $s+2 = 14$ and $\det{N} = 13\cdot17\cdot23$. This is the smallest $s+2$ so that $\det{N}$ has more than one prime divisor greater than $s+2$.
4. Last, for the parameter set $(8,12,15,2,8,5)$, we get $s+2 =17$ and $\det{N} = 2\cdot11\cdot13^2\cdot179\cdot197$. In this case, notice that $\det{N}$ has two prime divisors both greater than $a+b+c+\alpha+\beta+\gamma$, the sum of the generating degrees of $R/I_{a,b,c,\alpha,\beta,\gamma}$.
Given the previous example and Example \[exa:non-linear-bound\], it seems unlikely that there is a reasonably simple closed formula for the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$ in general, as opposed to the case of a symmetric region (see Conjecture \[con:symmetry\]).
Fixed determinants
------------------
For any positive integer $n$, there is an infinite family of punctured hexagons with exactly $n$ tilings. Note the algebras are type $2$ if $\beta$ is zero or $ c= n+\beta+1$ and type $3$ otherwise.
\[pro:det-n\] Let $n$ be a positive integer. If $\beta \geq 0$ and $c \geq n+\beta+1$, then $$\det{N_{c-\beta-1, \beta+2, c, c-n-\beta-1, \beta, n}} = n.$$ Hence the ideal $$I_{c-\beta-1, \beta+2, c, c-n-\beta-1, \beta, n} = (x^{c-\beta-1}, y^{\beta + 2}, z^c, x^{c-n-\beta-1} y^\beta z^n)$$ has the weak Lefschetz property when the characteristic of $K$ is either zero or not a prime divisor of $n$.
In this case, $s = c-2$, $A = \beta + 1$, $B = c - \beta - 2$, $C = 0$, and $M = 1$.
Using Proposition \[pro:C-zero\] we have that $$\det{N_{c-\beta-1, \beta+2, c, c-n-\beta-1, \beta, n}} = \operatorname{Mac}(M, A-\beta, B-\alpha) = \operatorname{Mac}(1, 1, n-1) = n.$$ Alternatively, from Proposition \[pro:wlp-binom\] we have that $$N_{c-\beta-1, \beta+2, c, c-n-\beta-1, \beta, n} = \left( \binom{\gamma}{A+C-\beta} \right) = \left( \binom{n}{1} \right) = \left( n \right).$$ Clearly then the determinant is $n$.
Thus for any prime $p$, Proposition \[pro:det-n\] provides infinitely many monomial almost complete intersections that fail to have the weak Lefschetz property exactly when the field characteristic is $p$.
A result of Proposition \[pro:det-n\] is an infinite (in fact, two dimensional) family whose members have a unique tiling. Note that the algebras are type $2$ if $\beta$ is zero or $c = \beta + 2$ and type $3$ otherwise.
\[cor:C-zero-unique\] If $\beta \geq 0$ and $c\geq \beta+2$, then $\det{N_{c-\beta-1, \beta+2, c, c-\beta-2, \beta, 1}}$ is 1. That is, $$I_{a,b,c,\alpha,\beta,\gamma} = (x^{c-\beta-1}, y^{\beta+2}, z^c, x^{c-\beta-2} y^\beta z)$$ has the weak Lefschetz property independent of the field characteristic.
Another family whose members have a unique tiling comes from Proposition \[pro:gamma-zero\]. Note that it is a three dimensional family but also that all of the associated algebras are type $2$.
\[pro:gamma-zero-unique\] If $a = b = \alpha + \beta + c$ and $\gamma = 0$, then $A = B = 0$ (see Figure \[fig:hex-A-B-gamma-zero\]) and $\det{N_{a,b,c,\alpha,\beta,\gamma}}$ is 1. That is, $$I_{a,b,c,\alpha,\beta,\gamma} = (x^{\alpha + \beta + c}, y^{\alpha + \beta + c}, z^c, x^\alpha y^\beta)$$ has the weak Lefschetz property independent of the field characteristic.
![When $A = B = \gamma = 0$, then $H_{a,b,c,\alpha,\beta,\gamma}$ has a unique tiling.[]{data-label="fig:hex-A-B-gamma-zero"}](hex-A-B-gamma-zero)
This follows from Proposition \[pro:gamma-zero\].
Several questions were asked in [@MMN], two of which we can answer in the affirmative.
\[rem:q-and-a\] Question 8.2(2c) asked if there exist non-level almost complete intersections which never have the weak Lefschetz property. The almost complete intersection $$R/I_{5,5,3,2,2,1} = R/(x^5, y^5, z^3, x^2y^2z)$$ is non-level and never has the weak Lefschetz property, regardless of field characteristic, as $\det{N_{5,5,3,2,2,1}} = 0$ by Proposition \[pro:symmetry-zero\].
Further, we notice here that Question 8.2(2b) in [@MMN] is answered in the affirmative by the comments following Question 7.12 in [@MMN]. In particular, $I_{11, 18, 22, 2, 9, 13}$ is a level almost complete intersection which has odd socle degree (39) and never has the weak Lefschetz property, as $\det{N_{11,18,22,2,9,13}} = 0$.
Minimal multiplicity
--------------------
The Huneke-Srinivasan Multiplicity Conjecture, which was proven by Eisenbud and Schreyer [@ES Corollary 0.3], shows that the multiplicity of a Cohen-Macaulay module gives nice bounds on the possible shifts of the Betti numbers. Moreover, as the algebras $A$ can be viewed as finite dimensional vector spaces, then the multiplicity [*is*]{} the dimension of $A$ as a vector space. Thus, algebras that have minimal multiplicity while retaining a particular property are the smallest, in the above sense, examples one can generate.
\[exa:minimal-multiplicity\] Possibly of interest are a few cases of minimal multiplicity with regard to the weak Lefschetz property.
The following examples never have the weak Lefschetz property, that is, the determinant of their associated matrix $N_{a,b,c,\alpha,\beta,\gamma}$ is 0. Note that both examples are type $3$.
1. The unique level ideal with minimal multiplicity is $$I_{3,3,3,1,1,1} = (x^3, y^3, z^3, xyz).$$ Its Hilbert function is $(1,3,6,6,3)$ and so it has multiplicity $19$. It is worth noting that this ideal is extensively studied in [@BK Example 3.1] and is the basis for an exploration of the subtlety of the Lefschetz properties in [@CN-2010].
2. The unique non-level ideal with minimal multiplicity is $$I_{5,5,3,2,2,1} = (x^5, y^5, z^3, x^2y^2z).$$ Its Hilbert function is $(1,3,6,9,12,12,9,4,1)$ and so it has multiplicity $57$. Further, this ideal is the example given in Remark \[rem:q-and-a\].
Moreover, the following examples always have the weak Lefschetz property, regardless of the base field characteristic. That is to say, the determinant of their associated matrix $N_{a,b,c,\alpha,\beta,\gamma}$ is 1.
1. The two level ideals with minimal multiplicity are $$I_{1,2,3,0,1,2} = (x, y^2, z^3, yz^2) \mbox{ and } I_{1,3,3,0,1,1} = (x, y^3, z^3, yz).$$ Both ideals have Hilbert function $(1,2,2)$ and thus multiplicity $5$. However, both ideals are isomorphic to ideals in $K[y,z]$.
2. The unique level, type $2$ ideal without $x$ as a generator and with minimal multiplicity is $$I_{2,2,3,1,1,0} = (x^2, y^2, z^3, xy).$$ Its Hilbert function is $(1,3,3,2)$ and so it has multiplicity $9$.
3. The unique level, type $3$ ideal with minimal multiplicity is $$I_{3,3,6,1,1,4} = (x^3, y^3, z^6, xyz^4).$$ Its Hilbert function is $(1,3,6,8,9,9,7,3)$ and so it has multiplicity $46$.
4. The unique non-level, type $2$ ideal with minimal multiplicity is $$I_{2,2,3,0,1,1} = (x^2, y^2, z^3, yz).$$ Its Hilbert function is $(1,3,3,1)$ and so it has multiplicity $8$.
5. The unique non-level, type $3$ ideal with minimal multiplicity is $$I_{2,2,4,1,1,2} = (x^2, y^2, z^4, xyz^2).$$ Its Hilbert function is $(1,3,4,4,2)$ and so it has multiplicity $14$.
Notice that example (ii) and (iv) in the above enumeration differ only slightly in the mixed term yet one is level and the other is not. It should also be noted that all of the above examples were found via an exhaustive search in the finite space of possible ideals using Macaulay2 [@M2].
Splitting type and regularity {#sec:splitting-type}
=============================
Throughout this section we assume $K$ is an algebraically closed field of characteristic zero.
Recall the definition of the ideals given in Section \[sec:aci\]; consider $$I = I_{a,b,c,\alpha,\beta,\gamma} = (x^a, y^b, z^c, x^\alpha y^\beta z^\gamma),$$ where $0 \leq \alpha < a, 0 \leq \beta < b, 0 \leq \gamma < c,$ and at most one of $\alpha, \beta,$ and $\gamma$ is zero. In this section we consider the splitting type of the syzygy bundles of the artinian algebras $R/I$, regardless of any extra conditions on the parameters.
Recall, also from Section \[sec:aci\], that the syzygy module $\operatorname{syz}{I}$ of $I$ is defined by the exact sequence $$0
\longrightarrow
\operatorname{syz}{I}
\longrightarrow
R(-\alpha-\beta-\gamma) \oplus R(-a) \oplus R(-b) \oplus R(-c)
\longrightarrow
I_{a,b,c,\alpha,\beta,\gamma}
\longrightarrow
0$$ and the syzygy bundle $\widetilde{\operatorname{syz}{I}}$ on ${\mathbb{P}}^2$ of $I$ is the sheafification of $\operatorname{syz}{I}$. Its restriction to the line $H \cong {\mathbb{P}}^1$ defined by $\ell = x+y+z$ splits as ${\mathcal{O}}_H(-p) \oplus {\mathcal{O}}_H(-q) \oplus {\mathcal{O}}_H(-r)$. The arguments in [@MMN Proposition 2.2] (recalled here in Proposition \[pro:mono\]) imply that $(p, q, r)$ is the splitting type of the restriction of $\widetilde{\operatorname{syz}{I}}$ to a general line. Thus, we call $(p,q,r)$ the [*generic splitting type*]{} of $\operatorname{syz}{I}$.
In order to compute the generic splitting type of $\operatorname{syz}{I}$, we use the observation that $R/(I, \ell) \cong S/J$, where $S = K[x,y]$, and $J = (x^a, y^b, (x+y)^c, x^\alpha y^\beta (x+y)^\gamma)$. Define $\operatorname{syz}{J}$ by the exact sequence $$\label{eqn:syz-J}
0
\longrightarrow
\operatorname{syz}{J}
\longrightarrow
S(-\alpha-\beta-\gamma) \oplus S(-a) \oplus S(-b) \oplus S(-c)
\longrightarrow
J
\longrightarrow
0$$ using the possibly non-minimal set of generators $\{x^a, y^b, (x+y)^c, x^\alpha y^\beta (x+y)^\gamma\}$ of $J$. Then $\operatorname{syz}{J} \cong S(-p) \oplus S(-q) \oplus S(-r)$. The Castelnuovo-Mumford regularity of a homogeneous ideal $I$ is denoted by $\operatorname{reg}{I}$.
\[rem:splitting-type\] For later use, we record the following facts on the generic splitting type $(p,q,r)$ of $\operatorname{syz}{I_{a,b,c,\alpha,\beta,\gamma}}$.
1. As the sequence in (\[eqn:syz-J\]) is exact, we see that $p + q + r = a+b+c+\alpha+\beta+\gamma$.
2. Further, if any of the generators of $J$ are extraneous, then the degree of that generator is one of $p, q,$ and $r$.
3. As regularity can be read from the Betti numbers of $R/J$, we get that $\operatorname{reg}{J} + 1 = \max\{p,q,r\}$.
Before moving on, we prove a useful lemma.
\[lem:reg-2AMACI\] Let $S = K[x,y]$, where $K$ is a field of characteristic zero, and let $a,b,\alpha,\beta,$ and $\gamma$ be non-negative integers with $\alpha + \beta + \gamma < a+b$. Without loss of generality, assume that $0 < a-\alpha \leq b - \beta$. Then $\operatorname{reg}{(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma)}$ is $$\left\{ \begin{array}{ll}
a + \beta + \gamma - 1 & \mbox{ if } \alpha = 0 \mbox{ and } 0 < \gamma \leq b - \beta - a; \\
\alpha + b - 1 & \mbox{ if } 0 < \alpha, \gamma \leq b - \beta + \alpha - a, \mbox{and } 0 < \beta \mbox{ or } 0 < \gamma; and \\
\left\lceil \frac{1}{2}(a+b+\alpha+\beta+\gamma)\right\rceil-1 & \mbox{ if } \gamma > b - \beta + \alpha - a.\\
\end{array} \right.$$
Further still, we always have $\operatorname{reg}{(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma)} \leq \left\lceil \frac{1}{2}(a+b+\alpha+\beta+\gamma)\right\rceil-1$.
We proceed in three steps.
First, consider $\gamma = 0$, $0 < \alpha$, and $0 < \beta$. Then by the form of the minimal free resolution of the quotient algebra $S/(x^a, y^b, x^\alpha y^\beta)$ we have that $\operatorname{reg}{(x^a, y^b, x^\alpha y^\beta)} = \alpha + b - 1$.
Second, consider $\gamma > 0$ and $\alpha = \beta = 0$. By [@HMNW Proposition 4.4], the algebra $S/(x^a, y^b)$ has the strong Lefschetz property in characteristic zero. Thus the Hilbert function of $S/(x^a, y^b, (x+y)^\gamma)$ is $$\dim_K{[S/(x^a, y^b, (x+y)^\gamma)]_j} = \max\{0, \dim_K{[S/(x^a, y^b)]_j} - \dim_K{[S/(x^a,y^b)]_{j-\gamma}}\}.$$ By analysing when the difference becomes non-positive, we get that the regularity is $a + \gamma - 1$ if $\gamma \leq b-a$ and $\left\lceil \frac{1}{2}(a+b+\gamma)\right\rceil-1$ if $\gamma > b-a$.
Third, consider $\gamma > 0$ and $0 < \alpha$ or $0 < \beta$. Notice that $$(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma):x^\alpha y^\beta = (x^{a-\alpha}, y^{b-\beta}, (x+y)^\gamma).$$ Considering the short exact sequence $$0 \rightarrow [S/(x^{a-\alpha}, y^{b-\beta}, (x+y)^\gamma)](-\alpha-\beta) \stackrel{\times x^\alpha y^\beta}{\longrightarrow}
S/(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma) \rightarrow S/(x^a, y^b, x^\alpha y^\beta) \rightarrow 0,$$ where the first map is multiplication by $x^\alpha y^\beta$, we obtain $$\operatorname{reg}{(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma)} = \max\{\alpha + \beta + \operatorname{reg}{(x^{a-\alpha}, y^{b-\beta}, (x+y)^\gamma)},
\operatorname{reg}{(x^a, y^b, x^\alpha y^\beta)}\}.$$ The claims then follows by simple case analysis.
Recall that the semistability of $\operatorname{syz}{I_{a,b,c,\alpha,\beta,\gamma}}$ is completely determined by the parameters $a,b,c,\alpha,\beta,\gamma$ in Proposition \[pro:semistable\].
Non-semistable syzygy bundle
----------------------------
We first consider the case when the syzygy bundle is not semistable. We distinguish three cases. It turns out that in two cases, at least one of the generators of $J$ is extraneous.
\[pro:st-nss\] Let $K$ be a field of characteristic zero and suppose $I = I_{a,b,c,\alpha,\beta,\gamma}$ is an ideal of $R$. Let $J = (x^a, y^b, (x+y)^c, x^\alpha y^\beta (x+y)^\gamma)$ be an ideal of $S$. We assume, without loss of generality, that $a \leq b \leq c$ so that $C \leq B \leq A$.
1. If $M < 0$, then the generator $x^\alpha y^\beta (x+y)^\gamma$ of $J$ is extraneous. The generic splitting type of $\operatorname{syz}{I}$ is $(a+c, b, \alpha+\beta+\gamma)$ if $c \leq b-a$ and $(\left\lfloor \frac{1}{2}(a+b+c) \right\rfloor, \left\lceil \frac{1}{2}(a+b+c) \right\rceil, \alpha+\beta+\gamma)$ if $c > b-a$.
2. If $M \geq 0$ and $C < 0$, then the generator $(x+y)^c$ of $J$ is extraneous. The generic splitting type of $\operatorname{syz}{I}$ is $(a+b+\alpha+\beta+\gamma - r - 1, r+1, c)$, where $r = \operatorname{reg}{(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma)}$ (which is given in Lemma \[lem:reg-2AMACI\]).
3. If $M \geq 0, C \geq 0$, and $A > \beta + \gamma$, then the only destabilising sub-bundle of $\operatorname{syz}{I}$ is $\operatorname{syz}{(x^a, x^\alpha y^\beta z^\gamma)}$ and so the generic splitting type of $\operatorname{syz}{I}$ is $(\left\lfloor \frac{1}{2}(\alpha+b+c) \right\rfloor,$ $\left\lceil \frac{1}{2}(\alpha+b+c) \right\rceil, a + \beta + \gamma)$.
Assume $M < 0$, then $\frac{1}{2}(a+b+c) < \alpha+\beta+\gamma$ and when $c \geq a+b$ then $$a+b-1 \leq \frac{1}{2}(a+b+c) - 1 < \alpha + \beta + \gamma.$$ By Lemma \[lem:reg-2AMACI\] the regularity of $(x^a, y^b, (x+y)^c)$ is $a+b-1$ when $c \geq a+b$ and $\lceil \frac{1}{2}(a+b+c)\rceil - 1$ otherwise; hence we have that $x^\alpha y^\beta (x+y)^\gamma$ is contained in $(x^a, y^b, (x+y)^c)$ and the first claim follows.
Assume $M \geq 0$ and $C < 0$, then $2(\alpha+\beta+\gamma) \leq a+b+c$, $c \geq \frac{1}{2}(a+b+\alpha+\beta+\gamma)$, and when $\alpha+\beta+\gamma \geq a+b$ then $2(\alpha+\beta+\gamma) \leq a+b+c$ implies $c \geq a+b$. By Lemma \[lem:reg-2AMACI\], the regularity of $(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma)$ is $a+b-1$ if $\alpha+\beta+\gamma \geq a+b$ and at most $\lceil \frac{1}{2}(a+b+\alpha+\beta+\gamma) \rceil - 1$ otherwise; hence we have that $(x+y)^c$ is contained in $(x^a, y^b, x^\alpha y^\beta (x+y)^\gamma)$ and the second claim follows.
Last, assume $M \geq 0, C \geq 0$, and $A > \beta + \gamma$. Note that since $A+B+C = \alpha+\beta+\gamma$ we then have that $B+C < \alpha$ and, in particular, $B < \alpha + \gamma$ and $C < \alpha + \beta$. Using Brenner’s combinatorial criterion for the semi-stability of syzygy bundles of monomial ideals (see [@Br Corollary 6.4]), we see that that $\mathcal{S} = \operatorname{syz}{(x^a, x^\alpha y^\beta z^\gamma)} \cong R(-r)$, where $r = a +\beta+\gamma$, is the only destabilising sub-bundle of $\operatorname{syz}{I}$. Further, $(\operatorname{syz}{I})/\mathcal{S}$ is a semistable rank two vector bundle, so by Grauert-Mülich theorem, the quotient has generic splitting type $(p,q)$ where $0 \leq q-p \leq 1$. Thus, if we consider the short exact sequence $$0 \longrightarrow \mathcal{S} \longrightarrow \operatorname{syz}{I} \longrightarrow (\operatorname{syz}{I})/\mathcal{S} \longrightarrow 0,$$ then the third claim follows after restricting to $\ell$.
In the third case, when $A > \beta + \gamma$, the associated ideal $J \subset S$ may be minimally generated by four polynomials, unlike in the other two cases.
\[exa:st-nss-4mingen\] Consider the ideals $$I_{4,5,5,3,1,1} = (x^4, y^5, z^5, x^3yz) \mbox{ and } J = (x^4, y^5, (x+y)^5, x^3y(x+y))$$ in $R$ and $S$, respectively. Notice that in this case, $0 \leq C \leq B \leq A$, $0 \leq M$, and $A > \beta + \gamma$ so the syzygy bundle of $R/I_{4,5,5,3,1,1}$ is non-semistable and its generic splitting type is determined in Proposition \[pro:st-nss\](iii). Further, $J$ is minimally generated by the four polynomials $x^4,$ $y^5,$ $xy^3(2x+y)$, and $x^3y^2.$
Semistable syzygy bundle
------------------------
Order the entries of the generic splitting type $(p,q,r)$ of the semistable syzygy bundle $\widetilde{\operatorname{syz}{I}}$ such that $p \leq q \leq r$. Then by Grauert-Mülich theorem we have that $r - q$ and $q - p$ are both non-negative and at most 1. Moreover, [@BK Theorem 2.2] specialises in our case.
\[thm:wlp-semistable-splitting-type\] Let $I = I_{a,b,c,\alpha,\beta,\gamma}$. If $R/I$ has the weak Lefschetz property, then $p = q$ or $q = r$ and $r - p \leq 1$; otherwise $q = p + 1$ and $r = p + 2$.
When $a+b+c+\alpha+\beta+\gamma \not\equiv 0 \pmod{3}$, then the generic splitting type of $\operatorname{syz}{I}$ and regularity of $J$ can be computed easily.
\[pro:st-nmod3\] Let $R = K[x,y,z]$ where $K$ is a field of characteristic zero. Suppose $I = I_{a,b,c,\alpha,\beta,\gamma}$ is an ideal of $R$ with a semistable syzygy bundle and let $J = (x^a, y^b, (x+y)^c, x^\alpha y^\beta (x+y)^\gamma)$ be an ideal of $S$. Let $k = \left\lfloor \frac{1}{3}(a+b+c+\alpha+\beta+\gamma) \right\rfloor$. Then $\operatorname{reg}{J} = k$ and the generic splitting type of $\operatorname{syz}{I}$ is $$\left\{ \begin{array}{ll}
(k,k,k+1) & \mbox{ if } a+b+c+\alpha+\beta+\gamma = 3k+1, \mbox{ and} \\[0.3em]
(k,k+1,k+1) & \mbox{ if } a+b+c+\alpha+\beta+\gamma = 3k+2.
\end{array} \right.$$
Let $(p,q,r)$ be the generic splitting type of $\operatorname{syz}{I}$, so $a+b+c+\alpha+\beta+\gamma = 3(s+2) = p+q+r$. By Proposition \[pro:amaci-not-3\], $R/I$ has the weak Lefschetz property so $p = q$, $q = r$, and $r - p \leq 1$. Clearly if $p = q = r$ then $p+q+r=3p$ is 0 modulo 3 so cannot be $a+b+c+\alpha+\beta+\gamma$.
If $p = q < r$, then $r = p+1$ and $p+q+r = 3p+1$. This matches the case when $a+b+c+\alpha+\beta+\gamma = 3k+1$, so $p = k$ and the splitting type of $\operatorname{syz}{I}$ is $(k,k,k+1)$. Similarly, if $p < q = r$, then $q = r = p + 1$ and $p+q+r = 3p+2$. This matches the case when $a+b+c+\alpha+\beta+\gamma = 3k+2$, so $p = k$ and the splitting type of $\operatorname{syz}{I}$ is $(k,k+1,k+1)$.
In both cases, we have that $k-1 \leq \operatorname{reg}{J} \leq k$ by Remark \[rem:splitting-type\](iii). However, we see that $\dim_K{[R/I]_{k-2}} < \dim_K{[R/I]_{k-1}}$ so $\dim_K{[R/(I,x+y+z)]_{k-1}} = \dim_K{[S/J]_{k-1}} > 0$ and thus $\operatorname{reg}{J} > k-1$. Hence $\operatorname{reg}{J} = k$.
The generic splitting type of $I_{a,b,c,\alpha,\beta,\gamma}$, when the ideal is associated to a punctured hexagon, depends on thew ideal having the weak Lefschetz property.
\[pro:st-mod3\] Let $R = K[x,y,z]$ where $K$ is a field of characteristic zero. Suppose $I = I_{a,b,c,\alpha,\beta,\gamma}$ is an ideal of $R$ with a semistable syzygy bundle (see Proposition \[pro:semistable\]) and $a+b+c+\alpha+\beta+\gamma \equiv 0 \pmod{3}$. Let $J = (x^a, y^b, (x+y)^c, x^\alpha y^\beta (x+y)^\gamma)$ be an ideal of $S$ and let $s+2 = \frac{1}{3}(a+b+c+\alpha+\beta+\gamma)$. Then
1. If $R/I$ has the weak Lefschetz property, then the generic splitting type of $\operatorname{syz}{I}$ is $(s+2,s+2,s+2)$ and $\operatorname{reg}{J} = s+1$.
2. If $R/I$ does not have the weak Lefschetz property, then the generic splitting type of $\operatorname{syz}{I}$ is $(s+1, s+2, s+3)$ and $\operatorname{reg}{J} = s+2$.
Let $(p,q,r)$ be the generic splitting type of $\operatorname{syz}{I}$, so $a+b+c+\alpha+\beta+\gamma = 3(s+2) = p+q+r$.
Assume that $R/I$ has the weak Lefschetz property. Suppose $p \neq q$, then $q = r = p + 1$ and $p+q+r = 3p + 2$, similarly, if $q \neq r$, then $p = q$ and $r = p+1$ so $p +q +r = 3p+1$; neither case is 0 modulo 3, hence cannot be $3(s+2)$. Thus $p = q = r = s+2$. Further we then see that $\operatorname{reg}{J} = s+1$ by Remark \[rem:splitting-type\](iii).
Now assume $R/I$ fails to have the weak Lefschetz property. Then $p + q + r = 3p + 3 = 3(s+2)$ so $p+1 = s+2$ and $p = s+1$. Thus, the generic splitting type of $\operatorname{syz}{I}$ must be $(s+1, s+2, s+3)$. As $R/I$ has twin-peaks at $s+1$ and $s+2$ by Corollary \[cor:one-map\], we see that $\operatorname{reg}{J} \leq s+1$ if and only if $R/I$ has the weak Lefschetz property; so $\operatorname{reg}{J} \geq s+2$. However, by Remark \[rem:splitting-type\](iii) we have that $\operatorname{reg}{J} + 1 \leq s+3$ so $\operatorname{reg}{J} \leq s+2$, hence $\operatorname{reg}{J} = s+2$.
This proposition can be combined with the results in the previous sections to compute the generic splitting type of many of syzygy bundles of the artinian algebras $R/I_{a,b,c,\alpha,\beta,\gamma}$.
\[exa:syzygy\] Consider the ideal $I_{7,7,7,3,3,3} = (x^7, y^7, z^7, x^3 y^3 z^3)$ which never has the weak Lefschetz property, by Proposition \[pro:symmetry-zero\]. The generic splitting type of $\operatorname{syz}{I_{7,7,7,3,3,3}}$ is $(9, 10, 11)$. Notice that the similar ideal $I_{6,7,8,3,3,3} = (x^6, y^7, z^8, x^3 y^3 z^3)$ has the weak Lefschetz property in characteristic zero as $\det{N_{6,7,8,3,3,3}} = -1764$ and the generic splitting type of $\operatorname{syz}{I_{6,7,8,3,3,3}}$ is $(10,10,10)$.
If $I = I_{a,b,c,\alpha,\beta,\gamma}$ is not associated to a punctured hexagon, then we have seen in Proposition \[pro:amaci-not-3\] and Corollary \[cor:wlp-ss\] that $R/I$ has the weak Lefschetz property in characteristic zero. We summarise part of our results by pointing out that in the case when $I$ is associated to a punctured hexagon then deciding the presence of the weak Lefschetz property is equivalent to determining other invariants of the algebra.
\[thm:equiv\] Let $R = K[x,y,z]$ where $K$ is a field of arbitrary characteristic. Let $I = I_{a,b,c,\alpha,\beta,\gamma}$ be associated to a punctured hexagon; in particular, $a+b+c+\alpha+\beta+\gamma \equiv 0 \pmod{3}$ and $\operatorname{syz}{I}$ is semistable (see Proposition \[pro:semistable\]). Set $s = \frac{1}{3}(a+b+c+\alpha+\beta+\gamma) - 2$.
Then the following conditions are equivalent:
1. The algebra $R/I$ has the weak Lefschetz property;
2. the regularity of $S/J$ is $s$;
3. the determinant of $N_{a,b,c,\alpha,\beta,\gamma}$ (i.e., the enumeration of signed lozenge tilings of the punctured hexagon $H_{a,b,c,\alpha,\beta,\gamma}$) modulo the characteristic of $K$ is non-zero; and
4. the determinant of $Z_{a,b,c,\alpha,\beta,\gamma}$ (i.e., the enumeration of signed perfect matchings of the bipartite graph associated to $H_{a,b,c,\alpha,\beta,\gamma}$) modulo the characteristic of $K$ is non-zero.
Moreover, if the characteristic of $K$ is zero, then there is one further equivalent condition:
1. The generic splitting type of $\operatorname{syz}{I}$ is $(s+2,s+2,s+2)$.
Combine Corollary \[cor:one-map\], Propositions \[pro:wlp-zero-one\] and \[pro:wlp-binom\], Theorems \[thm:nilp-matrix\] and \[thm:bip-matrix\], and Proposition \[pro:st-mod3\].
This relates the weak Lefschetz property to a number of other problems in algebra, combinatorics, and algebraic geometry.
Jumping lines
-------------
Recall that a [*jumping line*]{} is a line, $L = 0$, over which the syzygy bundle splits differently than in the generic case, $x+y+z = 0$. Since $I = I_{a,b,c,\alpha,\beta,\gamma}$ is a monomial ideal it is sufficient to consider the two cases $z = 0$ and $y+z=0$.
\[pro:jumping-lines\] Let $R = K[x,y,z]$ where $K$ is a field of characteristic zero and let $I = I_{a,b,c,\alpha,\beta,\gamma}$ be an ideal of $R$. The splitting type of $\operatorname{syz}{I}$ on the line $z = 0$ is $(c, \alpha + b, a + \beta)$ if $\gamma = 0$ and $(c, \alpha + \beta + \gamma, a+b)$ if $\gamma > 0$. And the splitting type of $\operatorname{syz}{I}$ on the line $y+z = 0$ is $(c, a + \beta + \gamma, \alpha + b)$ if $\beta + \gamma < b \leq c$ and $(c, \alpha + \beta + \gamma, a + b)$ if $b \leq \min\{c, \beta + \gamma\}$.
All four cases follow immediately by analysing the monomial algebra $S/J$ isomorphic to $R/(I, L)$, where $L = 0$ is the splitting line, and using Lemma \[lem:reg-2AMACI\] to compute the regularities.
Hyperfactorial calculus {#sec:hyper-calculus}
=======================
Throughout this manuscript, the [*hyperfactorial*]{} function ${\mathcal{H}}$ on non-negative integers $n$, defined by $${\mathcal{H}}(n) := \prod_{i=0}^{n-1}i!,$$ has been a key ingredient in many of the formulae. In this appendix we will highlight the uses and structure of the hyperfactorial and describe a useful “picture-calculus” approach to working with hyperfactorials.
Notice that, for $n \geq 0$, ${\mathcal{H}}(n)$ can also be seen as $\prod_{k=1}^{n-1} k^{n-k}.$ Thus if we place the numbers $1$ to $n-1$ in a right-triangular grid with legs of length $n-1$ (see Figure \[fig:hyper-tri-6\]), then the hyperfactorial of $n$ is the product of all the numbers in the grid.
![${\mathcal{H}}(6) = 34560$ represented as a triangular grid, both specifically and as a more generic shape[]{data-label="fig:hyper-tri-6"}](hyper-tri-6)
Using this pictorial representation of the hyperfactorial, various identities involving hyperfactorials become more transparent. The first identity is simple, but very useful.
\[pro:hyper-f\] Let $a$ and $b$ be non-negative integers with $a \leq b$. Then the polynomial $f_{a,b}(c) \in {\mathbb{Z}}[c]$, given by $$f_{a,b}(c) := \prod_{i=1}^{a}(c+i)^i \prod_{i=1}^{b-a}(c+a+i)^a \prod_{i=1}^{a}(c+b+i)^{a-i},$$ is equal to $$\frac{{\mathcal{H}}(a+b+c) {\mathcal{H}}(c)}{{\mathcal{H}}(a+c) {\mathcal{H}}(b+c)},$$ for positive integers $c$.
We proceed with a proof by picture-calculus; note that in each case we choose the $-1$ inherent to the hyperfactorial to go with the term which contains $c$ and that we represent the numbers that are present by a grey shaded region.
In Figure \[fig:hyper-f\](i), we consider ${\mathcal{H}}(a+b+c)$ divided by ${\mathcal{H}}(b+c)$; note that we align the triangles at their bottom points. We then multiply by ${\mathcal{H}}(c)$, seen in Figure \[fig:hyper-f\](ii); note that we align the top edge of the new triangle with the top edge of the triangle associated to ${\mathcal{H}}(b+c)$. Last, we divide by ${\mathcal{H}}(a+c)$ creating a parallelogram, seen in Figure \[fig:hyper-f\](iii).
![A picture-calculus proof that $f_{a,b}(c) = \frac{{\mathcal{H}}(a+b+c) {\mathcal{H}}(c)}{{\mathcal{H}}(a+c) {\mathcal{H}}(b+c)}$[]{data-label="fig:hyper-f"}](hyper-f-1 "fig:")\
[*(i) ${\mathcal{H}}(a+b+c) / {\mathcal{H}}(b+c)$*]{}
![A picture-calculus proof that $f_{a,b}(c) = \frac{{\mathcal{H}}(a+b+c) {\mathcal{H}}(c)}{{\mathcal{H}}(a+c) {\mathcal{H}}(b+c)}$[]{data-label="fig:hyper-f"}](hyper-f-2 "fig:")\
[*(ii) Multiply by ${\mathcal{H}}(c)$*]{}
![A picture-calculus proof that $f_{a,b}(c) = \frac{{\mathcal{H}}(a+b+c) {\mathcal{H}}(c)}{{\mathcal{H}}(a+c) {\mathcal{H}}(b+c)}$[]{data-label="fig:hyper-f"}](hyper-f-3 "fig:")\
[*(iii) Divide by ${\mathcal{H}}(a+c)$*]{}
Notice that the parallelogram is $a$ units tall, $b$ units long, and is shifted to be $c$ units from the left edge. Further,
1. The left grey triangular region corresponds to $\prod_{i=1}^{a}(c+i)^i$;
2. The central grey rectangular region corresponds to $\prod_{i=1}^{b-a}(c+a+i)^a$; and
3. The right grey triangular region corresponds to $\prod_{i=1}^{a}(c+b+i)^{a-i}$.
Thus, this region is exactly the polynomial $f_{a,b}(c)$ evaluated at the integer $c$.
\[exa:hyper-f\] For example, notice that $$f_{3,3}(c) = (c+1)(c+2)^2(c+3)^3(c+4)^2(c+5)$$ and $$f_{3,5}(c) = (c+1)(c+2)^2(c+3)^3(c+4)^3(c+5)^3(c+6)^2(c+7).$$
A key example of using Proposition \[pro:hyper-f\] is with MacMahon’s formula (see, e.g., [@CEKZ Equation (1.1)]). MacMahon’s formula for the number of lozenge tilings of a hexagon with side-lengths $(a,b,c,a,b,c)$, where $a, b,$ and $c$ are positive integers, is given by $$\operatorname{Mac}(a,b,c) = \frac{{\mathcal{H}}(a) {\mathcal{H}}(b) {\mathcal{H}}(c) {\mathcal{H}}(a+b+c)}{{\mathcal{H}}(a+b) {\mathcal{H}}(a+c) {\mathcal{H}}(b+c)},$$ Thus, for fixed $a$ and $b$, MacMahon’s formula is a polynomial in $c$.
\[cor:mac-poly\] Let $a$ and $b$ be non-negative integers with $a \leq b$. Then $\operatorname{Mac}(a,b,c)$ is equal to a polynomial in $c$, when evaluated at positive integers; in particular, $$\operatorname{Mac}(a,b,c) = \frac{{\mathcal{H}}(a) {\mathcal{H}}(b)}{{\mathcal{H}}(a+b)} f_{a,b}(c)$$ for positive integers $c$.
This follows immediately from Proposition \[pro:hyper-f\] after noticing that $\frac{{\mathcal{H}}(a) {\mathcal{H}}(b)}{{\mathcal{H}}(a+b)}$ is independent of $c$.
When considering polynomials such as $f_{a,b}(c)$, we may want only the terms where the factors are all of the form $(c+i)$, where $i$ has a fixed parity. To do this, we define the [*even part of the hyperfactorial*]{} of $n$, a positive integer, to be the even terms in the product ${\mathcal{H}}(n)$, that is $${\mathcal{H}}_e(n) := \prod_{i=0}^{n-1} \prod_{j=1}^{\left \lfloor \frac{1}{2}i \right \rfloor} 2j,$$ and we define the [*odd part of the hyperfactorial*]{} of $n$ to be $${\mathcal{H}}_o(n) := \frac{{\mathcal{H}}(n)}{{\mathcal{H}}_e(n)}.$$
We notice though, that ${\mathcal{H}}_e(n)$ can be written in terms of hyperfactorials, after an appropriate scaling.
\[lem:hyper-e\] For positive integers $n$, the even part of the hyperfactorial of $n$, ${\mathcal{H}}_e(n)$, is $$2^{\binom{\left \lfloor \frac{1}{2}n \right \rfloor}{2} + \binom{\left \lceil \frac{1}{2}n \right \rceil}{2}}
{\mathcal{H}}\left(\left \lfloor \frac{1}{2}n \right \rfloor\right) {\mathcal{H}}\left(\left \lceil \frac{1}{2}n \right \rceil \right).$$
By definition, ${\mathcal{H}}_e(n)$ is the product of the even columns of the pictorial representation of ${\mathcal{H}}(n)$. In Figure \[fig:hyper-e\](i), we see the product of ${\mathcal{H}}\left(\left \lfloor \frac{1}{2}n \right \rfloor\right)$ and ${\mathcal{H}}\left(\left \lceil \frac{1}{2}n \right \rceil \right)$ and in part (ii) we see this same product after simplification, that is, condensing the columns to be contiguous. In Figure \[fig:hyper-e\](iii), we multiply each element of the triangular representation by $2$; since there are $\binom{\left \lfloor \frac{1}{2}n \right \rfloor}{2} + \binom{\left \lceil \frac{1}{2}n \right \rceil}{2}$ terms, then we are simply scaling ${\mathcal{H}}\left(\left \lfloor \frac{1}{2}n \right \rfloor\right) {\mathcal{H}}\left(\left \lceil \frac{1}{2}n \right \rceil \right)$ by $2^{\binom{\left \lfloor \frac{1}{2}n \right \rfloor}{2} + \binom{\left \lceil \frac{1}{2}n \right \rceil}{2}}$. This is exactly the even columns of ${\mathcal{H}}(n)$.
![A picture-calculus proof of an identity of ${\mathcal{H}}_e(n)$[]{data-label="fig:hyper-e"}](hyper-e-1 "fig:")\
[*(i) ${\mathcal{H}}\left(\left \lfloor \frac{1}{2}n \right \rfloor\right) {\mathcal{H}}\left(\left \lceil \frac{1}{2}n \right \rceil \right)$*]{}
![A picture-calculus proof of an identity of ${\mathcal{H}}_e(n)$[]{data-label="fig:hyper-e"}](hyper-e-2 "fig:")\
[*(ii) After simplification*]{}
![A picture-calculus proof of an identity of ${\mathcal{H}}_e(n)$[]{data-label="fig:hyper-e"}](hyper-e-3 "fig:")\
[*(iii) After scaling*]{}
Now we can find the polynomials which represent $f_{a,b}(c)$ with only factors of the form $(c+i)$, where $i$ has a fixed parity, present.
\[cor:hyper-eo-f\] Let $a$ and $b$ be non-negative integers with $a \leq b$. Set $\underline{a} = \left\lfloor \frac{1}{2}a \right\rfloor$ and $\underline{b} = \left\lfloor \frac{1}{2}b \right\rfloor$. Then the polynomial $f^e_{a,b}(c) \in {\mathbb{Z}}[c]$, given by $$f^e_{a,b}(c) := \prod_{i=1}^{\underline{a}}(c+2i)^{2i}
\prod_{i=1}^{\underline{b}-\underline{a}} (c+2\underline{a}+2i)^a
\prod_{i=1}^{\underline{a}}(c+2\underline{b}+2i)^{b-2\underline{b}+a-2i},$$ is equal to $$\left\{ \begin{array}{ll}
\displaystyle \frac{{\mathcal{H}}_e(a+b+c) {\mathcal{H}}_e(c)}{{\mathcal{H}}_e(a+c) {\mathcal{H}}_e(b+c)} & \mbox{ if $c$ is even and} \\[0.8em]
\displaystyle \frac{{\mathcal{H}}_o(a+b+c) {\mathcal{H}}_o(c)}{{\mathcal{H}}_o(a+c) {\mathcal{H}}_o(b+c)} & \mbox{ if $c$ is odd},
\end{array} \right.$$ for positive integers $c$.
Further, the polynomial $f^o_{a,b}(c) \in {\mathbb{Z}}[c]$, given by $$f^o_{a,b}(c) := \frac{f_{a,b}(c)}{f^e_{a,b}(c)}$$ is equal to $$\left\{ \begin{array}{ll}
\displaystyle \frac{{\mathcal{H}}_o(a+b+c) {\mathcal{H}}_o(c)}{{\mathcal{H}}_o(a+c) {\mathcal{H}}_o(b+c)} & \mbox{ if $c$ is even and} \\[0.8em]
\displaystyle \frac{{\mathcal{H}}_e(a+b+c) {\mathcal{H}}_e(c)}{{\mathcal{H}}_e(a+c) {\mathcal{H}}_e(b+c)} & \mbox{ if $c$ is odd},
\end{array} \right.$$ for positive integers $c$.
Notice first that $f^e_{a,b}(c)$ is defined to be the factors of $f_{a,b}(c)$ of the form $(c+i)$, where $i$ is even. The hyperfactorial representation then follows when considering $(c+i)$ would then be even exactly when $c$ is even.
The second claim follows similarly as the first.
\[exa:hyper-feo\] For example, notice that $$f^e_{3,5}(c) = (c+2)^2(c+4)^3(c+6)^2$$ and $$f^o_{3,5}(c) = (c+1)(c+3)^3(c+5)^3(c+7).$$
The authors would like to acknowledge the invaluable nature of the computer algebra system Macaulay2 [@M2]. Macaulay2 was used extensively throughout both the original research and the writing of this manuscript.
[99]{}
M. Boij, J. Migliore, R. Miró-Roig, U. Nagel, F. Zanello, [*On the shape of a pure $O$-sequence*]{}, Mem. Amer. Math. Soc. (to appear).
B. Boyle, J. Migliore, F. Zanello, [*More plane partitions enumerated by the Weak Lefschetz Property*]{}, in preparation.
H. Brenner, [*Looking out for stable syzygy bundles*]{}, Adv. Math. [**219**]{} (2008), 401–427.
H. Brenner, A. Kaid, [*Syzygy bundles on ${\mathbb{P}}^2$ and the Weak Lefschetz Property*]{}, Illinois J. Math. [**51**]{} (2007), 1299–1308.
H. Brenner, A. Kaid, [*A note on the weak Lefschetz property of monomial complete intersections in positive characteristic*]{}, Collect. Math. [**62**]{} (2011), 85–93.
C. Chen, A. Guo, X. Jin, G. Liu, [*Trivariate monomial complete intersections and plane partitions*]{}, J. Commut. Algebra (to appear), available at [arXiv:1008.1426](http://arxiv.org/abs/1008.1426).
M. Ciucu, T. Eisenkölbl, C. Krattenthaler, D. Zare, [*Enumerations of lozenge tilings of hexagons with a central triangular hole*]{}, J. Combin. Theory Ser. A [**95**]{} (2001), 251–334.
D. Cook II, U. Nagel, [*The weak Lefschetz property, monomial ideals, and lozenges*]{}, Illinois J. Math (to appear).
D. Cook II, U. Nagel, [*Hyperplane sections and the subtlety of the Lefschetz properties*]{}, J. Pure Appl. Algebra [**216**]{} (2012), no. 1, 108–114; [doi:10.1016/j.jpaa.2011.05.007](http://dx.doi.org/10.1016/j.jpaa.2011.05.007).
G. David, C. Tomei, [*The problem of the calissons*]{}, Amer. Math. Monthly [**96**]{} (1989), 429–431.
D. Eisenbud, F.O. Schreyer, [*Betti numbers of graded modules and cohomology of vector bundles*]{}, J. Amer. Math. Soc. [**22**]{} (2009), 859–888.
I. Fischer, [*Enumeration of Rhombus Tilings of a Hexagon which Contain a Fixed Rhombus in the Centre*]{}, J. Combin. Theory Ser. A [**96**]{} (2001), 31–88.
I. Gessel, G. Viennot, [*Binomial determinants, paths, and hook length formulae*]{}, Adv. in Math. [**58**]{} (1985), 300–321.
I. Gessel, X. G. Viennot, [*Determinants, paths and plane partitions*]{}, preprint (1989); available at [http://people.brandeis.edu/gessel/homepage/papers/](http://people.brandeis.edu/~gessel/homepage/papers/).
D. R. Grayson, M. E. Stillman, [*Macaulay 2, a software system for research in algebraic geometry*]{}, Available at <http://www.math.uiuc.edu/Macaulay2/>.
T. Harima, J. Migliore, U. Nagel, J. Watanabe, [*The weak and strong Lefschetz properties for Artinian $K$-algebras*]{}, J. Algebra [**262**]{} (2003), 99–126.
P. W. Kasteleyn, [*Graph theory and crystal physics*]{}. In [*Graph Theory and Theoretical Physics*]{}, 43–110. [*Academic Press*]{}, London, 1967.
R. Kenyon, [*Lectures on dimers*]{}. In [*Statistical mechanics*]{}, 191–230. IAS/Park City Math. Ser., [**16**]{}, [*Amer. Math. Soc.*]{}, Providence, RI, 2009.
C. Krattenthaler, [*Advanced determinant calculus*]{}, Sém. Lothar. Combin. [**42**]{} (“The Andrews Festschrift”) (1999), Art. B42q.
C. Krattenthaler, [*Descending plane partitions and rhombus tilings of a hexagon with a triangular hole*]{}, European J. Combin. [**27**]{} (2006), no. 7, 1138–1146.
G. Kuperberg, [*Symmetries of plane partitions and the permanent-determinant method*]{}, J. Combin. Theory Ser. A [**68**]{} (1994), no. 1, 115–151.
J. Li, F. Zanello, [*Monomial complete intersections, the weak Lefschetz property and plane partitions*]{}, Discrete Math. [**310**]{} (2010), 3558–3570.
B. Lindström, [*On the vector representations of induced matroids*]{}, Bull. London Math. Soc. [**5**]{} (1973), 85–90.
J. Migliore, [*The Geometry of the Weak Lefschetz Property and Level Sets of Points*]{}, Canad. J. Math. Vol. [**60**]{} (2008), 391–411.
J. Migliore, R. Miró-Roig, U. Nagel, [*Monomial ideals, almost complete intersections and the weak Lefschetz property*]{}, Trans. Amer. Math. Soc. [**363**]{} (2011), 229–257.
J. Migliore, F. Zanello, [*The Hilbert functions which force the weak Lefschetz property*]{}, J. Pure Appl. Algebra [**210**]{} (2007), 465–471.
J. Propp, [*Enumeration of matchings: problems and progress*]{}, Mathematical Sciences Research Institute Publications [**38**]{}, [*Cambridge University Press*]{} (1999), 255–291.
R. Stanley, [*Enumerative combinatorics. Vol. 1.*]{}, Cambridge Studies in Advanced Mathematics [**49**]{}, [*Cambridge University Press*]{}, Cambridge, 1997.
F. Zanello, J. Zylinski, [*Forcing the Strong Lefschetz and the Maximal Rank Properties*]{}, J. Pure Appl. Algebra [**213**]{} (2009), 1026–1030.
[^1]: Part of the work for this paper was done while the authors were partially supported by the National Security Agency under Grant Number H98230-09-1-0032.\
${}^{\star}$ Corresponding author.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present x-ray powder diffraction (XRPD) and neutron diffraction measurements on the slightly underdoped iron pnictide superconductor Ba$_{1-x}$K$_x$Fe$_2$As$_2$, $T_{\rm c}=32$K. Below the magnetic transition temperature $T_{\rm m}=70$K, both techniques show an additional broadening of the nuclear Bragg peaks, suggesting a weak structural phase transition. However, macroscopically the system does not break its tetragonal symmetry down to 15K. Instead, XRPD patterns at low temperature reveal an increase of the anisotropic microstrain proportionally in all directions. We associate this effect with the electronic phase separation, previously observed in the same material, and with the effect of lattice softening below the magnetic phase transition. We employ density functional theory to evaluate the distribution of atomic positions in the presence of dopant atoms both in the normal and magnetic states, and to quantify the lattice softening, showing that it can account for a major part of the observed increase of the microstrain.'
author:
- 'D.S.Inosov'
- 'A.Leineweber'
- Xiaoping Yang
- 'J.T. Park'
- 'N.B.Christensen'
- 'R.Dinnebier'
- 'G.L.Sun'
- 'Ch.Niedermayer'
- 'D.Haug'
- 'P. W. Stephens'
- 'J.Stahn'
- 'C.T. Lin'
- 'O.K.Andersen'
- 'B.Keimer'
- 'V. Hinkov'
bibliography:
- 'Orthorhombicity.bib'
title: |
Suppression of the structural phase transition and lattice softening\
in slightly underdoped Ba$_{1-x}$K$_x$Fe$_2$As$_2$ with electronic phase separation
---
Introduction
============
The recent discovery of superconductivity (SC) in layered iron arsenides [@KamiharaWatanabe08; @WenMu08; @ChenLi08; @TakahashiIgawa08; @ChenWu08; @KitoEisaki08; @ZhiAnWei08] served as a powerful impetus in the search for novel superconductors with high critical temperatures. Within this new family of compounds, the record holders for the highest known $T_{\rm c}$ are the electron-doped 1111-compounds Gd$_{1-x}$Th$_x$FeAsO [@WangLi08] and Sr$_{1-x}$Sm$_x$FeAsF [@WuXie09] (both with optimal $T_{\rm c}=56$ K), whereas among the so-called 122-compounds the highest $T_{\rm c}$ of 38K was reached in the hole-doped Ba$_{1-x}$K$_x$Fe$_2$As$_2$ near the optimal doping of $x\approx0.5$ [@RotterTegel08; @WuLiu08]. The parent compounds ($x=0$) of both types of arsenides order antiferromagnetically (AFM) below a spin-density-wave (SDW) transition at temperatures $T_{\rm m}$ in the range between 140 and 200K, as seen by neutron scattering [@CruzHuang08; @ZhaoHuang08; @ZhaoHuang08PRB; @HuangZhao08; @HuangQiu08; @ChenRen09; @GoldmanArgyriou08; @SuLink09] and local-probe methods, such as $\mu$SR [@AczelBaggio08; @GokoAczel08; @KlaussLuetkens08; @JescheCaroca08] and $^{57}$Fe Mössbauer spectroscopy [@RotterTegel08PRB; @KlaussLuetkens08]. At low doping levels, this SDW transition is always accompanied by a structural phase transition at $T_{\rm s}$$\gtrsim$$T_{\rm m}$ from a high-temperature tetragonal (T) to a low-temperature orthorhombic or monoclinic structure, which manifests itself as a longitudinal splitting of the in-plane nuclear Bragg peaks $(hh0)\,_{\rm T}$ both in neutron scattering and x-ray diffraction experiments [@CruzHuang08; @ZhaoHuang08; @ZhaoHuang08PRB; @HuangZhao08; @HuangQiu08; @ChenRen09; @RotterTegel08PRB; @JescheCaroca08; @GoldmanArgyriou08; @YanKreyssig08; @NiBudko08; @NiNandi08]. Up to now, to the best of our knowledge, there have been no reports of any iron pnictides, neither among parent nor doped compounds, where magnetic order would be observed without the development of a structural distortion.
![Neutron diffraction data showing the temperature-dependent longitudinal broadening of the (110)$_{\rm T}$ nuclear Bragg reflection () overlayed with the intensity of the $\bigl(\frac{1}{\text{\protect\raisebox{1pt}{2}}}\,\frac{1}{\text{\protect\raisebox{1pt}{2}}}\,\bar{3}\bigr)_{\rm T}$ magnetic Bragg peak ().[]{data-label="fig:broadening"}](Broadening.pdf){width="\columnwidth"}
Though it is commonly acknowledged that the magnetic and structural order parameters in iron pnictides are intimately coupled, the details of the relationship between the two phase transitions still remain a puzzle. On the one hand, in 1111-compounds (but not in 122-compounds [@HuangQiu08]) the structural phase transition precedes the magnetic one [@CruzHuang08; @ZhaoHuang08], suggesting itself as the driving force for the magnetic anisotropy of the SDW phase. On the other hand, the experimentally observed structural distortion cannot be reproduced in non-magnetic calculations [@MazinJohannes09; @MazinJohannesBoeri08]. Therefore most theories consider the SDW instability an intrinsic property of the electronic system, driven either by the nesting of the electron- and hole-like Fermi surface sheets [@ChubukovEfremov08; @KorshunovEremin08; @MazinSingh08; @DongZhang08; @KurokiOnari08; @YareskoLiu08] or by the local superexchange interactions in the framework of the Heisenberg model [@Yildirim08; @SiAbrahams08; @MaLu08; @FangYao08; @XuMuller08]. Both scenarios imply that the structural phase transition occurs as a consequence of the AFM ordering, and its somewhat higher transition temperature is explained as a response to anisotropic AFM fluctuations that persist even above $T_{\rm m}$ [@FangYao08; @XuMuller08; @MazinJohannes09; @ZabolotnyyInosov09].
In this paper, we combine neutron scattering and x-ray powder diffraction (XRPD) experiments, along with theoretical calculations, to study the interplay between the magnetic and structural phase transitions in a slightly underdoped 122-compound Ba$_{1-x}$K$_x$Fe$_2$As$_2$ (BKFA), $T_{\rm c}=32$K, in which the onset of a phase-separated magnetic order occurs at $T_{\rm m}=70$K according to our recent study performed on the same samples [@ParkInosov09]. Our experimental evidence indicates that macroscopically the sample preserves its tetragonal symmetry down to 15K, well below $T_{\rm m}$. Instead of the structural transition to an orthorhombic phase at low temperatures, seen in more underdoped BKFA samples [@ChenRen09; @RotterTegel08PRB], here the lattice reacts to the magnetic order only microscopically, by an increase of the microstrain as observed in our XRPD measurements, without a macroscopic breakdown of the lattice symmetry. We argue that such an effect is most probably related to a softening of the lattice below the magnetic phase transition in comparison to the high-temperature non-magnetic state, whereas the phase-separated coexistence of twinned magnetic domains and the non-magnetic phase [@ParkInosov09] suppresses the structural phase transition beyond the experimentally detectable limit, in spite of a relatively high SDW transition temperature.
Sample preparation
==================
The single crystals of BKFA used for the present study were grown using Sn as flux in a zirconia crucible sealed in a quartz ampoule filled with Ar. A mixture of Ba, K, Fe, As, and Sn in a wt. ratio of BKFA:Sn = 1:85 was heated in a box furnace up to 850$^\circ$C and kept constant for 2–4 hours to soak the sample in a homogeneous melt. The cooling rate of 3$^\circ$C/h was then applied to decrease the temperature to 550$^\circ$C, and the grown crystals were then decanted from the flux [@SunSunLin08]. Sample characterization by resistivity and dc susceptibility measurements [@ParkInosov09] revealed a sharp SC transition at $T_{\rm c,\,onset}=(32\pm1)$ K, reproducible among different samples from the same batch. The same samples have been extensively studied by angle-resolved photoelectron spectroscopy (ARPES) [@ZabolotnyyInosov09; @EvtushinskyInosov09; @EvtushinskyInosovNJP] and muon-spin rotation ($\mu$SR) [@ParkInosov09; @KhasanovAmato09].
The neutron diffraction measurements were done on a $\sim$30mg single crystal with in-plane ($hh0$) and out-of-plane ($00l$) mosaicities better than 1.5° and 2.5°, respectively, as determined from the full width at half maximum (FWHM) of the rocking curves. A few smaller single crystals from the same batch were ground into powder for XRPD analysis. The sample was then prepared by sprinkling a small amount of the powder onto a flat brass sample holder.
![Panels (a) and (b) present XRPD data measured at 300K and 15K, respectively. (i) Scattered x-ray intensity as a function of the diffraction angle $2\Theta$ ($\lambda=0.7$Å) fitted to the tetragonal *I*4/*mmm* space group. For $2\Theta>17^\circ$ the plots are enlarged by a factor of three. The fit includes a few wt.% of tetragonal $\beta$-tin from the flux as an impurity phase and some reflections of the brass sample holder as indicated by the reflection markers in (ii). (iii) The difference $\Delta$ between the experimental points and the fitting curve. The insets show tensor surfaces representing the normalized anisotropic microstrain distribution along different crystallographic directions. The distance of the surface from the origin corresponds the squared full width at half maximum (FWHM) of the microstrain $B_\varepsilon^2$ along the corresponding directions in real space. The $x$-$z$ and $x$--0.5pt$y$ cross-sections of both surfaces are shown in panel (c) for comparison.[]{data-label="fig:xray"}](XRPD.pdf){width="\columnwidth"}
Neutron diffraction
===================
We have measured the longitudinal width of the (110)$_{\rm T}$ nuclear Bragg reflection as a function of temperature, which is plotted in Fig.\[fig:broadening\] together with the intensity of the $\bigl(\frac{1}{\text{\protect\raisebox{1pt}{2}}}\,\frac{1}{\text{\protect\raisebox{1pt}{2}}}\,\bar{3}\bigr)_{\rm T}$ magnetic Bragg peak. One can clearly see the broadening of the nuclear Bragg peak at low temperatures, with an onset at $T_{\rm m}$, which perfectly follows the magnetic intensity, and amounts to $\sim$20% as $T\rightarrow0$. The most straightforward explanation for such broadening would be a weak orthorhombic distortion that leads to a splitting of the peak that is masked by the experimental resolution, as was also previously observed whenever the AFM order was suppressed either by doping, as in CeFeAsO$_{0.94}$F$_{0.06}$ at low temperature \[Ref., Fig.2(d)\], or by temperature, as in the parent compound LaFeAsO at $T=138$K \[Ref., Fig.4(inset)\].
To check this interpretation, we have performed XRPD measurements of the same samples, with subsequent analysis of the microstrain anisotropy, which is known to be helpful in detecting minute structural distortions related to possible phase transitions [@Leineweber06; @Leineweber07].
X-ray powder diffraction
========================
The XRPD data for the structure refinement were collected at room temperature and at 15K, as shown in Fig.\[fig:xray\] (a) and (b). The sample was placed in a closed cycle cryostat. X-rays of 0.7Å wavelength were selected by a double Si(111) monochromator. The wavelength and zero-point error were calibrated using 8 precisely measured peaks of the NBS1976 flat plate alumina standard. The diffracted beam was analyzed by reflection from a Ge(111) crystal before a NaI scintillation detector. Data were taken at each $2\Theta$ step of 0.005° from 3° to 38.6° at room temperature and 2° to 52° at 15K. The sample was rocked during the measurement for better particle statistics. All data were normalized for storage ring current decay by an ionization chamber monitor.
XRPD data were analyzed using the program TOPAS (*Bruker-AXS*). Both high- and low-temperature data could be interpreted in terms of a tetragonal *I*4/*mmm* space group symmetry both at room temperature and at $T=15$ K (see Fig.\[fig:xray\]). As impurity phases, a few wt.% of tetragonal $\beta$-tin from the flux and some reflections of the brass sample holder were included in the refinement. The analysis of the anisotropic peak broadening in the powder pattern due to a microstrain distribution was performed using the Cartesian parametrization by Leineweber [@Leineweber06; @Leineweber07].
The lattice parameters of the sample, as determined from XRPD by Rietveld refinement using the fundamental parameters approach of TOPAS [@ChearyCoelho05], are $a$=$b$=3.9111(1)Å and $c$=13.3392(6)Å at room temperature and $a$=$b$=3.90075(7)Å and $c$=13.2476(3)Å at 15K, which corresponds to a 1.2% decrease in the unit cell volume at low temperature. From the dependence of the lattice parameters on doping [@RotterPangerl08], the average potassium content of $x=0.4$ could be determined, in agreement with the results of our energy-dispersive x-ray analysis. No evidence was found for an orthorhombic distortion of the tetragonal lattice at low temperature. This conclusion is based on the absence of any orthorhombic splitting of the Bragg reflections and the refinement of the lattice parameters. The isotropic microstrain distribution in the $(hk0)$ plane also does not hint at an orthorhombic distortion.
![The $2\sqrt{2}\,a\times2\sqrt{2}\,b\times c$ supercell with 50% of the Ba atoms randomly substituted by the K dopants that we used in our density functional calculations. The calculated statistical distributions of the five interatomic distances, which are marked in the figure, are presented in Fig.\[fig:histograms\].[]{data-label="fig:unitcell"}](UnitCell.pdf){width="0.8\columnwidth"}
The microstrain distribution represents the statistics of the deviations ${\scriptstyle\Delta}d$ of the interplanar spacings from their average values, normalized by the average spacings $d$, i.e. of the strain $\varepsilon={\scriptstyle\Delta}d/d$, over the investigated specimen as a function of the crystallographic direction. Tensor surfaces representing the squared FWHM of the anisotropic microstrain distribution $B_\varepsilon^2$ along different crystallographic directions are shown as insets in Fig.\[fig:xray\] (a) and (b), whereas panel (c) shows the $x$-$z$ \[tetragonal (*ac*) plane\] and $x$--0.5pt$y$ \[tetragonal (*ab*) plane\] cross-sections of both surfaces. The largest microstrains of the crystalline lattice both at 300K and at 15K are found in the $c$-direction ($|B_\varepsilon|_\perp=0.9$% and 1.1%, respectively$)$ as compared to the average in-plane values of $|B_\varepsilon|_\parallel=0.65$% and 0.82%. The flowerlike shape of the $x$-$z$ cross-section indicates a negative correlation between the in-plane ($hk0$) and the out-of-plane ($00l$) directions, which agrees with the opposite changes of the $a$ and $c$ lattice constants upon the variation of doping [@RotterPangerl08].
![Histograms of the calculated interatomic distances for the supercell as defined in Fig. \[fig:unitcell\]. The plots at the top of each panel are a result of the normal-state (non-magnetic) calculation, whereas the up-side-down plots below represent the magnetic ground state. The solid lines are fits to a normal distribution.[]{data-label="fig:histograms"}](Histograms.pdf){width="\columnwidth"}
The low-temperature increase of the microstrain amounts to $\sim$20% relative to the corresponding values at room temperature both in the $c$-direction and in-plane. In other words, to a good approximation the two tensor surfaces are geometrically similar to each other, which would not be expected in the case of a weak orthorhombic distortion, as it should instead broaden only the in-plane peaks. Moreover, at both temperatures no considerable in-plane anisotropy is observed \[i.e.anisotropy in the $x$-$y$ plane, see Fig.2(c)\], which would be a sign for the onset of an orthorhombic phase transition, e.g. for an incomplete orthorhombic reflection splitting. Such an anisotropy (cf. also Ref., Fig.1) characteristically precedes tetragonal-to-orthorhombic structural phase transitions in Pb$_3$O$_4$ [@DinnebierCarlson03] and La$_2$NiO$_4$ [@RodriguezCarvajal91].
This lets us conclude that the origin of the microstrain at both temperatures is not related to a macroscopic structural transition to orthorhombic symmetry, but rather should be attributed to an increase of the microscopic distortions of the lattice. The microstrain distribution quantitatively represents the response of the lattice to structural defects, such as chemical inhomogeneities or dislocations, which are unavoidable in any real material. Therefore an increase of the microstrain below the magnetic transition can either indicate that the lattice becomes softer, i.e. increases its response to the local stresses upon entering the AFM state, or that the local stresses themselves increase, causing a proportional increase of the microstrain.
In the studied compound, both mechanisms could be important. On the one hand, in the case of lattice softening, one would expect its direct influence on the phonon mode frequencies. Indeed, such an effect has been reported in the phonon spectra of two similar 122-compounds: polycrystalline Sr$_{0.6}$K$_{0.4}$Fe$_2$As$_2$ and Ca$_{0.6}$Na$_{0.4}$Fe$_2$As$_2$ [@MittalSu08]. There, softening of phonon modes below 10meV has been observed by inelastic neutron scattering upon cooling from 300K to 140K, despite the decrease of the unit cell volume at low temperature. More recently, softening and narrowing of several phonon modes below the spin density wave transition was also observed by Raman scattering in underdoped Sr$_{1-x}$K$_x$Fe$_2$As$_2$ and in the parent BaFe$_2$As$_2$ single crystals [@RahlenbeckSun09]. On the other hand, the phase-separated coexistence of AFM and paramagnetic phases in this material [@ParkInosov09] and the presence of twin AFM domain boundaries [@MazinJohannes09] should lead to an increase of local stresses below $T_{\rm m}$ due to the magnetic anisotropy of individual AFM domains.
To quantify the relative role of these two possible causes of the increased microstrain, we present here an estimation of the lattice softening across the magnetic transition based on our density functional calculations and show that it is comparable in magnitude and therefore could possibly provide a considerable contribution to the additional microstrain observed in the XRPD measurements.
Density functional calculations
===============================
Density functional calculations were performed using the projector augmented-wave [@Blochl94; @KresseJoubert99] method in the framework of the generalized gradient approximation [@PerdewBurke97; @PerdewBurke96]. We have chosen a large $2\sqrt{2}\,a\times2\sqrt{2}\,b\times c$ supercell, where 50% of the Ba atoms were substituted by K to model a random distribution of the dopants, as shown in Fig.\[fig:unitcell\]. Using the Vienna *ab initio* simulation package (VASP) plane wave code [@KresseHafner93; @KresseHafner94; @KresseFurthmueller96prb; @KresseFurthmueller96cms; @KresseFurthmueller99], we have carried out the crystal structure optimization of the cell parameters and all ionic positions within the supercell to determine their displacements from the high-symmetry positions due to the introduced chemical disorder. The unit cell volume was fixed during the course of structure optimization to the experimental room-temperature value for the parent compound, as derived from $a$=$b$=3.9625Å, $c$=13.0168Å [@RotterTegel08PRB], which also agrees well with our room-temperature value measured for BKFA. The cutoff energy of the plane-wave expansion was 367 eV, and the Brillouin zone sampling mesh was 4$\times$4$\times$4 with its origin at the $\Gamma$ point. In the final optimized geometry, no forces on the atoms exceeded 0.01eV/Å.
First, a non-magnetic calculation was performed, which represents the high-temperature (normal) state. As expected, it revealed no deviations from the tetragonal symmetry, and resulted in the lattice parameters of $a$=$b$=3.927Å and $c$=13.258Å, which are reasonably consistent with the results of the XRPD structure refinement discussed above.
The results of the crystal structure optimization are presented in Fig.\[fig:histograms\] (top plot in each panel), which shows histograms of the five interatomic distances, as defined in Fig.\[fig:unitcell\], fitted to a Gaussian distribution (solid lines). Naturally, the largest atomic displacements due to potassium substitution are observed in the Ba/K plane itself, where the distances $x_{\rm Ba/K}$ between the neighboring Ba/K atoms vary by 0.176Å or 4.48% (0.0448 r.l.[0.5pt]{}u., where r.l.[0.5pt]{}u. stands for relative lattice units), as estimated by the FWHM of the distribution. Within the FeAs block of layers, the out-of-plane atomic displacements (the buckling of the As and Fe planes) are the largest, and amount to 0.068Å (1.65% or 0.0051 r.l.[0.5pt]{}u.) for the As layer and 0.088Å (1.32% or 0.0066 r.l.[0.5pt]{}u.) for the Fe layer. In-plane distortions are notably smaller: 0.023Å (0.60%) and 0.013Å (0.45%), respectively.
Finally, we performed a spin-polarized calculation for the low-temperature striped AFM state. Collinear magnetic moments were self-consistently determined within the calculation. The corresponding histograms are shown in the same figure at the bottom of each panel. The most noticeable effect is the splitting of the Fe-Fe nearest neighbor distance, $x_\text{Fe}$, which indicates the tendency of the system towards an orthorhombic distortion despite the presence of the dopants. As already mentioned above, such a transition is however suppressed macroscopically in the sample due to the presence of twin AFM domains and phase separation. In addition, one sees that the nearest-neighbor Fe-Fe inter-layer distance $z_\text{Fe}$ increases, while that of As, $z_\text{As}$, decreases, which corresponds to the stretching of the Fe-As tetrahedra.
Of more relevance for the present paper is the small but not negligible increase in the width of the distribution for every interatomic distance, as compared to the normal state, which we associate with the sought lattice softening effect. The variation of interatomic distances in the Ba/K plane increases by 3% to 0.182Å (4.67% or 0.0467 r.l.[0.5pt]{}u.). The buckling of the As and Fe planes increases to 0.097Å (2.38% or 0.0073 r.l.[0.5pt]{}u.) and 0.099Å (1.46% or 0.0073 r.l.[0.5pt]{}u.), respectively, which represents an increase by 43% and 12% relative to the corresponding normal-state values. In-plane distortions increase to 0.032Å (0.82%) for the As layer and 0.019Å (0.67%) for the Fe layer ($\sim$40% of relative increase in the width).
The observed changes in FWHM of the distributions between the AFM and the normal states are statistically significant and are observed consistently for all five considered interatomic distances. For the FeAs block of layers, they amount to $\sim$0.002 r.l.[0.5pt]{}u. on average both in- and out-of-plane. This is comparable with the increase of the microstrain $|B_\varepsilon|_\perp(15\,K)-|B_\varepsilon|_\perp(300\,K)\!=0.002$ and $|B_\varepsilon|_\parallel(15\,K)-|B_\varepsilon|_\parallel(300\,K)\!=0.0017$ observed in our XRPD experiment. We can therefore conclude that the softening of the lattice associated with the SDW transition provides a major contribution to the observed effect. An additional contribution from the increase of the local stresses at the AFM domain boundaries due to weak local distortions within each domain can not be excluded, however.
Summary and discussion
======================
We have presented an example of an iron pnictide superconductor, which does not break its tetragonal crystal symmetry macroscopically upon entering the magnetically ordered state. This conclusion is based on x-ray powder diffraction measurements with subsequent analysis of the microstrain anisotropy. Instead, we have observed a low-temperature increase of the microstrain proportionally in all crystallographic directions, which has a magnetic origin and mostly originates from the softening of the crystal lattice below the SDW phase transition. A detailed analysis of the lattice structure in the presence of randomly distributed dopant atoms has been presented both in the normal and AFM states, confirming this conclusion. This does not exclude that a weak orthorhombic distortion possibly happens on a microscopic scale within each AFM domain, leading to an increase of the local stresses at the domain boundaries, whereas the mesoscopic electronic phase separation [@ParkInosov09] suppresses the breakdown of the tetragonal symmetry on lateral scales larger than the typical size of the AFM domains. It can be argued that the observed magnetic state of the lattice represents a crossover between the well-developed orthorhombic phase at low doping levels and the normal tetragonal phase typical for the overdoped region of the phase diagram, where no magnetic transition occurs even at lowest temperatures.
Acknowledgments {#acknowledgments .unnumbered}
===============
The experimental work was performed at the *Morpheus* diffractometer and RITA-II spectrometer, both at the Swiss spallation source SINQ, Paul Scherrer Institut (PSI), Villigen, Switzerland, and the X16C beamline at the National Synchrotron Light Source, Brookhaven National Laboratory, USA. We acknowledge financial support from DFG in the consortium FOR538, as well as from the Bundesministerium für Bildung und Forschung (BMBF), and the Fonds der Chemischen Industrie (FCI). Special thanks to A. Yaresko and O. Khvostikova for helpful discussions.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A comparative study of optical spectra of Type Ia supernovae (SNe Ia) obtained near 1 week, 3 weeks, and 3 months after maximum light is presented. Most members of the four groups that were defined on the basis of maximum light spectra in Paper II (core normal, broad line, cool, and shallow silicon) develop highly homogeneous postmaximum spectra, although there are interesting exceptions. Comparisons with SYNOW synthetic spectra show that most of the spectral features can be accounted for in a plausible way. The fits show that 3 months after maximum light, when SN Ia spectra are often said to be in the nebular phase and to consist of forbidden emission lines, the spectra actually remain dominated by resonance scattering features of permitted lines, primarily those of Fe II. Even in SN 1991bg, which is said to have made a very early transition to the nebular phase, there is no need to appeal to forbidden lines at 3 weeks postmaximum, and at 3 months postmaximum the only clear identification of a forbidden line is \[Ca II\] 7291, 7324. Recent studies of SN Ia rates indicate that most of the SNe Ia that have ever occurred have been “prompt” SNe Ia, produced by young ($\sim 10^8$ yr) stellar populations, while most of the SNe Ia that occur at low redshift today are “tardy”, produced by an older (several Gyrs) population. We suggest that the shallow silicon SNe Ia tend to be the prompt ones.'
author:
- 'David Branch, David J. Jeffery, Jerod Parrent, E. Baron, M. A. Troxel, V. Stanishev, Melissa Keithley, Joshua Harrison, and Christopher Bruner'
title: 'Comparative Direct Analysis of Type Ia Supernova Spectra. IV. Postmaximum'
---
15[[$\Delta$]{}$m_{15}$]{} 10[$V_{10}$(Si II)]{} 575[$W(5750)$]{} 610[$W(6100)$]{} 100[the 6100 Å absorption]{}
INTRODUCTION
============
This is the fourth in a series of papers on a comparative direct analysis of the optical spectra of Type Ia supernovae (SNe Ia). Paper I (Branch et al. 2005) was concerned with a time series of spectra of the spectroscopically normal SN 1994D, Paper II (Branch et al. 2006) was devoted to spectra obtained near maximum light, and Paper III (Branch et al. 2007) concentrated on premaximum spectra. This paper is about postmaximum spectra, which form in the deeper, lower velocity ejecta.
In Paper II we divided the maximum-light ($B$ maximum) spectra into four groups: core normal, broad line, cool, and shallow silicon (denoted CN, BL, CL, and SS, respectively, in this paper). The group assignments were made on the basis of measurements of the (pseudo) equivalent widths of absorption features near 5750 and 6100, as well as on the appearance (depth, width, and shape) of the 6100absorption, which is produced by the Si II 6355 transition. Although we framed the presentation and discussion in terms of the four groups, in the end we concluded that for the most part the spectra appeared to have a continuous distribution of properties, rather than consisting of discrete subtypes (the extreme SS SN 2002cx–likes being an apparent exception, and the extreme CL SN 1991bg–likes being a possible exception). In Paper III we found that to a large extent the premaximum spectra exhibited the defining characteristics of the four groups.
In this paper, as in the previous ones, we confine our attention to optical spectra, from the Ca II H and K feature in the blue ($\sim3700$) to the Ca II infrared triplet (Ca II IR3) in the red ($\sim9000$). All spectra have been corrected for the redshifts of the host galaxies, and mild smoothing has been applied to some of the spectra. All spectra have been flattened according to the local normalization procedure of Jeffery et al. (2007), which eliminates all significant broad band continuum variations, including those caused by interstellar reddening and observational error. Thus the locally normalized spectra allow valid comparisons of intrinsic line features. But note that a locally normalized spectrum is not a unique representation of the line spectrum. It depends on the exact prescription of the local normalization procedure. We use exactly the same procedure for all spectra of this paper.
We examined all of the SN Ia postmaximum spectra available to us and selected three samples: a “1 week postmax” sample, consisting of one spectrum each of 21 SNe Ia observed between day +6 and day +8 with respect to the time of maximum brightness in the $B$ band; a “3 week postmax” sample of 19 SNe Ia observed between day +19 and day +23; and a “3 months postmax” sample of 15 SNe Ia observed between day +81 and day +98 (an interval of day +80 to day +100 yielded no additional spectra). The SNe Ia and the epochs of the selected spectra are listed in Table 1. Many comparisons of subsets of these spectra have appeared in the literature. Our goal is to provide a more systematic and comprehensive comparison.
Spectra of four of the SNe Ia of Table 1 were not available for previous papers of this series. Sadakane et al. (1996) published spectra of SN 1995D obtained shortly after maximum light, but it is difficult to decide from their Figure 3 whether SN 1995D should be assigned to the CNs or the SSs. On the basis of the report by Benetti & Mendes de Oliveira (1995) that in a 1 week premax spectrum the 5750 Å absorption was unusually weak if present at all, we tentatively assign SN 1995D to the SS group. Pastorello et al. (2007) referred to SN 2004eo as a transitional SN Ia because its properties placed it between the three groups (faint, low velocity gradient, and high velocity gradient) of Benetti et al. (2005). In our classification, the maximum light spectrum of SN 2004eo is not quite CN; it resembles that of SN 1989B (Paper II), and like SN 1989B it is placed in the CL group although it does not contain the blue “Ti II trough” of the more extreme members of the CL group. The maximum light spectrum of SN 2006X (Wang et al. 2007) is that of an extreme BL SN Ia, such as SN 1984A. Hicken et al. (2007) showed that an early spectrum of SN 2006gz had the strongest C II 6580 absorption yet seen in a SN Ia, and argued that super–Chandrasekhar mass ejection is required. Although no spectrum was obtained within three days of maximum light, at earlier epochs SN 2006gz showed characteristics of a SS event, and we tentatively classify it as such.
CALCULATIONS
============
Continuing our attempt to provide an internally consistent quantification of SN Ia spectra, we have used the parameterized resonance–scattering synthetic–spectrum code, SYNOW, to fit the spectra of Table 1. The excitation temperature is fixed at a nominal value of 7000 K, as was done for the postmaximum SN 1994D spectra of Paper I. A change of procedure from our previous work with SYNOW on postmaximum spectra, including that of Paper I, is that instead of using exponential (Paper I) or power–law radial optical depth distributions, we use a flat distribution with an imposed maximum velocity, $v_{\rm max}$, for every ion. In the outer layers of the ejected matter, a decreasing density is dictated by hydrodynamical models (and by mass conservation: the density must start decreasing somewhere), but in the deeper layers the density gradient is less steep. Since the optical depth distribution depends also on composition, excitation, and ionization gradients, a flat optical depth distribution is not unreasonable. For further economy of parameters, for a given spectrum we use the same value of $v_{\rm
max}$ for all singly ionized members of the iron group, from Sc II to Ni II (with the exception of the 1 week postmax spectrum of SN 2000cx; § 3.) The price to be paid is that the fits are not optimized, but the fitting procedure becomes more efficient (there is no need to vary the $v_{\rm e}$ parameter) and the parameters become easier to compare: for each ion we have just a reference–line optical depth and the velocity interval in which the lines of the ion are forming in the synthetic spectrum, the minimum velocity being either the velocity at the photosphere $v_{\rm phot}$ or a higher detachment velocity (i.e., detached from the photosphere). Since we use only one optical depth component for each ion, in this paper it is not necessary to use the HV (high velocity) and PV (photospheric velocity) terminology of previous papers of this series.
Consider the effects, when switching from an exponential to a flat optical depth distribution, on the relative importance of lines of the same ion but of different strengths. In the flat case, the optical depths of the strongest lines need to be smaller than in the exponential case, since they maintain the same optical depths all the way to $v_{\rm max}$. Lines that are somewhat weaker, but whose optical depths still remain above unity in the flat case, become relatively more important, because they now form over just as large a velocity interval as the strongest lines. Lines whose optical depths are reduced from above unity in the exponential case to well below unity in the flat case no longer conspicuously affect the spectrum.
As an example, Figure 1 compares synthetic spectra for Fe II, with flat and exponential optical depth distributions. The flat case has the parameters that we use in § 4 for the day +22 spectrum of the CN SN 1996X: $\tau$(Fe II)=12, $v_{\rm phot} = 6000$ , and $v_{\rm
max}=13,000$ . The exponential case has the parameters that we used in Paper I for the day +24 spectrum of the CN SN 1994D: $\tau$(Fe II)=200, $v_{\rm phot} = 9000$ , and exponential $e$–folding velocity $v_e = 1000$ . Note that the flat case requires a lower value of $v_{\rm phot}$ than the exponential case. While the two synthetic spectra are similar, there are some significant differences, such as the near disappearance in the flat case of the absorption that appears near 5380 Å in the exponential case, and the higher flux peak near 4660 Å in the flat case.
ONE WEEK POSTMAX
================
One week postmax is in the middle of what we referred to in Paper I as the postmaximum “Si II phase” (from two to 12 days past maximum, for SN 1994D), because the 6100 Å absorption is deep and apparently unblended, at least in its core. At 1 week postmax the spectra are not radically different from at maximum light.
Figure 2 shows the 1 week postmax spectra of the CNs. The spectra are very similar. A log plot such as Figure 2 is convenient for comparing multiple spectra at once, but since the spectra are displaced from each other, it can be difficult to appreciate the degree of the homogeneity. Figure 3 directly compares the spectra of SN 1998bu and SN 1996X, to show how strikingly similar these two spectra are. Even the difference at the red end of the SN 1998bu spectrum is not necessarily real, because observed spectra sometimes have problems at their ends, and also because the local normalization technique can mildly warp the ends of the spectra (Jeffery et al. 2007).
An example SYNOW fit is shown in Figure 4. The fit is to the spectrum of the CN SN 1996X (selected because it is the best spectrum of Figure 2, not because it is the best fit). The fitting parameters for SN 1996X (and other selected spectra of Figure 2) are in Table 2. Apart from the flat optical depth distribution, the fit of Figure 4 is a conventional one. The ions are the same as were used for the 1 week postmax spectrum of SN 1994D in Paper I, except that here Cr II also is included. Lines of O I, Si II, S II, Ca II, and Fe II produce most of the features in the synthetic spectrum, and we are confident that these ions are present in the observed spectrum. The presence in the observed spectrum of Na I, Mg II, Cr II, Fe III, and Co II is not definite, but we use them in the synthetic spectrum because they are plausible and they improve the fit.
The three main discrepancies in Figure 4 are familiar problems with SYNOW fits. First, for Ca II we choose parameters to fit the observed IR3 feature, because it is less blended and more sensitive to optical depth than the H and K feature; in this case, this causes the synthetic H and K absorption to be too strong. Second, the flux minimum from about 6900 to 7100 Å has no counterpart in the synthetic spectrum. The only identification that we can suggest is \[O II\] 7320,7330, which was discussed in the context of SN 1991T by Fisher et al. (1999) and of the Type Ic SN 1994I by Millard et al. (1999). However, at later epochs this discrepancy becomes stronger, and invoking \[O II\] to solve it would imply excessive mass and kinetic energy of oxygen (at least for an exploding white dwarf). More likely, what we are seeing in Figure 4 is an early, mild manifestation of a discrepancy that is due to our simplifying assumption of resonant scattering. Third, the synthetic spectrum lacks absorption from about 7600 to 7800 Å. The observed absorption in this region presumably is due to Mg II 7890, blended with O I 7772, but in SYNOW spectra the synthetic absorption usually is too blue, even when Mg II is undetached, as it is here.
Most of the spectra of Figure 2 are so similar that it is not worth reporting fitting parameters for every spectrum. The exception is SN 2004S, which has much stronger high velocity Ca II absorption than the others. Krisciunas et al. (2007) termed SN 2004S a clone of SN 2001el, another CN with exceptionally strong high velocity Ca II features (Wang et al. 2003; Kasen et al. 2003; Mattila et al. 2005; Paper II). Krisciunas et al. noted that at maximum light SN 2004S had an unusually low Si II absorption blueshift for a spectrum containing the usual SN Ia features, and that the blueshift decreased more rapidly than in SN 2001el. We find that at 1 week postmax, all absorptions other than those of Ca II are less blueshifted than in the other spectra of Figure 2; consequently our fit for SN 2004S has a low value of $v_{\rm phot}= 7000$ (Table 2), instead of 11,000 as used for the others of Figure 2.
Figure 5 shows the 1 week postmax spectra of the BLs (plus the spectrum of the CN SN 1996X for comparison). At this epoch the spectrum of SN 2002er is like that of a CN. SN 1992A retains some of its BL characteristics and remains mildly different from CN. The spectra of SN 2002bf, SN 2006X, and SN 1984A are obviously different from CN, but similar to each other. Their 6100 Å absorptions are deeper, broader, and more blueshifted, and they have deep absorptions from about 4700 to 5100 Å. A fit to SN 2002bf is shown in Figure 6. The 6100 Å absorption and the deep absorption from 4700 to 5100 Å are matched by using a high $v_{\rm max}$ of 21,000 for Si II and Fe II, compared to a typical value of 15,000 for the CNs at 1 week postmax. The Fe II features of SN 2005bf are so prominent that it is difficult to determine whether lines of Mg II, Fe III, and Co II are present.
The 1 week postmax spectra of the CLs are shown in Figure 7. At this epoch SN 1989B is like that of a CN and SN 2004eo is only mildly different. The spectrum of the extreme CL SN 1999by, a SN 1991bg–like (Garnavich et al. 2004), is more highly evolved than the others of the 1 week postmax sample, in fact it has more in common with the 3 week postmax spectra of the CNs than with their 1 week postmax spectra. A fit for SN 1999by is shown in Figure 8. The presence of Si II, Ca II, Ti II, and Fe II is definite, while Mg I, Sc II, and Cr II are plausible and improve the fit. There are three main discrepancies. First, the synthetic absorption near 5340 Å, produced by Cr II, is not deep enough, even though Cr II is too strong in several other places. (A corresponding feature in the CNs of the 3 week postmax sample is discussed in § 4.) Second, the observed absorption near 5820 Å has practically no counterpart in the synthetic spectrum. The only identification we can offer is Si II 5972, but it hardly appears in the synthetic spectrum even though the absorption feature of the 6355 Si II reference line is too strong. Third, the synthetic flux peaks near 4000 and 4600 Å are too high; this may be a consequence of the flat optical depth distribution (see the 4600 Å peak in Figure 1.) Our present SYNOW fit differs from the one presented in Garnavich et al. in that we do not use Ca I, Ti II makes no significant contribution to the synthetic spectrum at wavelengths longer than 5300 Å, and we do invoke Mg I, Sc II, and Cr II.
Figure 9 shows the 1 week postmax spectra of the SSs. SN 2006gz is intermediate between the CN SN 1996X and the SS SN 1999aa, which has weaker Si II, S II, and Ca II absorptions. Hicken et al. (2007) identified C II in spectra of SN 2006gz obtained 10 or more days before maximum light, after which the feature was not identifiable, consistent with the lack of evidence for C II in the spectrum shown in Figure 9. The direct comparison of SN 1999ac and SN 1996X in Figure 10 is interesting. SN 1999ac has Si II and S II absorptions that are less blueshifted than in SN 1996X, but Fe II features from 4700 to 5100 Å that are more blueshifted. A fit for SN 1999ac is shown in Figure 11. The presence of O I, Si II, S II, Ca II, and Fe II in SN 1999ac is definite; Na I, Mg II and Fe III are plausible. To account for the above mentioned peculiarity, we use a low photospheric velocity of 6000 and a higher $v_{\rm max}$ for Fe II (15,000 ) than for most of the other ions (see Table 2). The spectra of SN 1999ac have been discussed extensively by Garavini et al. (2005). They emphasized the unusually low Si II blueshift, consistent with our use of a low photospheric velocity. They also concluded that Fe II lines formed at unusually low velocities, but our $v_{\rm max}$ of 15,000 is the same as we use for the CN SN 1996X.
The Ca II IR3 feature of SN 2000cx obviously differs from the others of Figure 9. Otherwise, the closest match to SN 2000cx is SN 1999aa, but there are conspicuous differences. The fit for SN 2000cx shown in Figure 12 is unusual because, as in Paper II, we resort to detached high–velocity Ti II, Cr II, and Fe II. In fact, in the synthetic spectrum most of the photospheric features are at least mildly detached (Table 2). Still, the fit is not very good. Although the synthetic Ca II H and K absorption is much too strong, the IR3 does not appear. The observed multicomponent absorption of the IR3 has been discussed by Li et al. (2001), Wang et al. (2003), Kasen et al. (2003), and Thomas et al. (2004), and is known to have involved asymmetrical structures at high velocity.
THREE WEEKS POSTMAX
===================
Three weeks postmax is in the middle of what we referred to in Paper I as the “Si II–to–Fe II transition phase” (from 14 to 28 days past maximum, for SN 1994D) because the core of the 6100 Å absorption is present but flanked by strengthening Fe II features. The 3 week postmax spectra are very different from the 1 week postmax spectra.
Figure 13 shows the 3 week postmax spectra of the CNs, and a fit for SN 1996X is shown in Figure 14. The ions are the same as were used for SN 1994D at comparable epochs in Paper I. Apart from the Ca II and Na I features, the dominant ion is Fe II, but Si II, Cr II, and to a lesser extent Co II are also needed. Fitting parameters for SN 1996X and other selected spectra of the 3 week postmax sample are given in Table 3.
The degree of homogeneity in Figure 13 is very high, except in two respects. First, SN 2001el and SN 2004S still have strong high velocity Ca II H and K absorptions (and most of the absorption features of SN 2004S continue to be somewhat less blueshifted than in the others). Second, the relative heights of the flux peaks that flank the 5350 Å absorption are not all the same. The relative heights of these two peaks were discussed in Paper I: in SN 1994D the ratio of the peak on the right of the 5350 Å absorption to the peak on the left increased steadily from day +15, when the peaks were roughly equally high, to day +28, when the peak on the right was much higher. Our fits accounted nicely for this evolution, in terms of strengthening Fe II and Cr II lines. In this respect, some of the spectra of Figure 13 appear to be spectroscopically “earlier” than some of the others. For example, the day +21 spectrum of SN 1998aq (Fig. 13) has equally high peaks, like the day +15 spectrum of SN 1994D (Paper I), and closely resembles that spectrum in other respects also. Even CN spectra can get mildly out of phase, with respect to the time of B–band maximum.
The 3 week postmax spectra of the BLs are shown in Figure 15. At this epoch SN 1981B is like CN and so is SN 2002er, apart from its deep high velocity Ca II H and K absorption (which is not matched by the appearance of the IR3 absorption). Even the extreme BL SN 1984A appears to differ only mildly from CN, although the limited wavelength coverage does not allow us to see how the Ca II lines are behaving.
Figure 16 shows the 3 week postmax spectra of the CLs. At this epoch SN 1989B and SN 2004eo are like CN. SN 1986G remains mildly different, e.g., its 5750 Å absorption is shallow, and the ratio of the flux peaks flanking the 5350 Å absorption is very large. SN 1991bg continues to show obvious differences. A fit for SN 1991bg is shown in Figure 17. All 7 ions used in the synthetic spectrum are considered to be definite. In the blue the synthetic spectrum is a complex blend of Ti II, Cr II, and Fe II. Not only does SN 1991bg have Ti II while CNs do not, but also Cr II plays a more important role than it does in the CNs.
The 3 week postmax spectra of the SSs are shown in Figure 18. SN 1999ee is similar to CN, although with a strong high velocity Ca II H and K absorption. SN 1999aa also is similar to CN except for the spectroscopically “early” ratio of the flux peaks flanking the 5350 Å absorption. In this respect, as in others, the day +19 spectrum of SN 1999aa in Figure 18 is a good match to the day +14 spectrum of SN 1994D (Paper I).
As shown in previous papers (Chornock et al. 2006; Phillips et al. 2007; Stanishev et al. 2007), at multiple epochs the spectra of SN 2002cx and SN 2005hk are very similar. In Figure 18, the spectra of both have been artifically blueshifted by 5000 , to approximately align their absorption features with those of the other SSs. The direct comparison in Figure 19 of SN 2005hk and the CN SN 1996X shows that although the features of SN 2005hk are narrower, when the spectrum of SN 2005hk is artificially blueshifted, the features of the two spectra show a strong correspondence. Not only the absorptions but also the flux peaks match up well. If the flux peaks of SN 2005hk were true emission peaks, at the rest wavelengths of the lines that produce them, then after the blueshifting, the peaks would be bluer than those of SN 1996X. This is in accord with the maxim that in heavily blended resonance–scattering spectra, “absorptions trump emissions”, i.e., flux peaks do not necessarily occur at the rest wavelengths of the strongest lines, because they are strongly influenced by absorption components of mainly redward lines. Figure 20 shows a fit to the (not artificially blueshifted) spectrum of SN 2005hk. The ions used are the same ones as used for the 3 week postmax spectrum of the CN SN 1996X, except that Si II is not needed for SN 2005hk. Note that it is not $v_{\rm phot}$, but $v_{\rm max}$ for Cr II, Fe II, and Co II that differs by 5000 from that of SN 1996X (Table 2). Apart from the two Ca II features and the one Na I feature, the synthetic spectrum is a complex blend of Fe II, Cr II, and Co II, just as it is for SN 1996X.
As can be seen in Figure 18, at this epoch SN 2000cx remains different from CN in several ways, e.g., the peaks that appear near 4800 and 4920 Å in SN 1996X are smeared out in SN 2000cx. In this respect, SN 2000cx resembles the BL SN 2002bf of the 1 week premax sample (§ 2). To fit the spectrum of SN 2000cx we use a high maximum velocity of 18,000 for Fe II and Cr II (Table 2). A discussion of SYNOW fits to day +15 and +32 spectra, which have better wavelength coverage, can be found in Branch et al. (2004).
THREE MONTHS POSTMAX
====================
Three months postmax is within what we referred to in Paper I as the “Fe II phase” (from 50 to 115 days after maximum for SN 1994D, with these particular limiting epochs being determined by the availability of spectra) because the 6100 Å absorption is nearly gone at 50 days and the spectrum is mainly shaped by Fe II lines. The 3 months postmax spectra are quite different from the 3 weeks postmax spectra. The spectra of the five CNs (Fig. 21) are very much alike. It is often assumed that at this epoch the spectrum is composed of optically thin, collisionally excited forbidden lines, but the SYNOW fit for SN 2003du in Figure 22 is good enough to firmly establish that whatever the underlying source of the emission, permitted lines of Na I, Ca II, and Fe II are mainly responsible for shaping the spectrum (see also the discussion in Paper I). However, our fit is very poor from 6600 to 7800 Å. It may be that the broad flux minimum near 6750 Å is not an absorption feature, and that the flux peaks from 7270 to 7700 are produced by strong net emission. Bowers et al. (1997) modeled the day +95 optical spectrum (along with the day 92 infrared spectrum) of SN 1995D assuming that the spectrum consists only of forbidden–line emissions. Their identifications of the flux peaks are shown in Figure 22 (SN 1995D and SN 2003du have the same peaks, and for some of the peaks Bowers et al. did not suggest identifications). In the blue, the Bowers et al. fit was not successful. However, our poor fit from 6600 to 7800 Å may signal that in this wavelength range forbidden emission lines are emerging without strong modification by permitted–line scattering. On the other hand, if we make the Fe II lines stronger, thereby making our fit worse in the blue, we do begin to obtain a better fit in the 6600 to 7800 Å interval. This could be a consequence of a wavelength dependent variation in the depth of the photosphere. At present, line identifications in this wavelength range are uncertain.
At this epoch the four BL events of the 3 months postmax sample (Fig. 23), including the extreme BL SN 2006X, are similar to CN.
Among the CLs (Fig. 24), SN 1989B is like CN, SN 1986G differs mildly, and the extreme CL SN 1991bg remains distinctly different. The fit for SN 1991bg in Figure 25 shows that permitted–line scattering, mainly by Fe II, shapes the spectrum in the blue, but the fit fails from 6000 to 8000 Å. The only clear evidence for forbidden lines is the strong emission near 7250 Å, due to \[Ca II\] (Filippenko et al. 1992a).
At this epoch the probable SS SN 1995D and the extreme SS SN 1991T (Fig. 26) are like CN, and even the maverick SN 2000cx is only mildly different.
DISCUSSION
==========
The spectra that appear in this series of papers are simply the good spectra that are available to us, so the samples are affected not only by observational bias in favor of supernovae that are bright and have slowly declining light curves, but also by the observer bias in favor of obtaining and publishing spectra of unusual events. Nevertheless, these samples clearly indicate a trend toward increasing spectroscopic conformity at later epochs. Among the 24 SNe Ia of the maximum light sample (Paper II), only 7 were admitted to the (by construction) highly homogeneous CN group. By contrast, of the 12 SNe Ia of the 3 months postmax sample, only SN 1991bg is distinctly different from the CNs. (SN 2002cx–likes would be different too, if they were in the sample.) This spectroscopic convergence suggests that although the outer layers of SNe Ia are diverse in various ways, the deep layers of most SNe Ia are fundamentally the same.
In this paper, the line identifications for the CNs are much like they were in Paper I for SN 1994D, although the fitting parameters are different because of our present use of flat optical depth distributions. The conclusion that permitted lines strongly affect the spectrum as late as 3 months postmax leads to a caveat to the statement above that the deep layers of most SNe Ia are fundamentally the same: at three months postmax the strongest permitted lines are strong, which entails line profile saturation and a limited sensitivity of the spectra to physical differences in the ejecta. If the spectra of the 3 months postmax sample consisted simply of optically thin forbidden emission lines, saturation would not be an issue and the observed spectroscopic homogeneity would imply a startling (unbelievable?) degree of physical homogeneity. The degree of diversity should increase at still later times as the permitted lines weaken and the forbidden lines emerge unscathed. The transition from resonance–scattering to forbidden–line dominated spectra will be the subject of a future paper.
As in previous papers of this series, we find that the BL SNe Ia have essentially the same line identifications as the CNs, but with more line optical depth at high velocities. The distinct characteristics of the extreme BL events SN 1984A, SN 2002bf, and SN 2006X at 1 week postmax raises the issue of whether they are a discrete subgroup, but only in terms of the properties of their outer layers. There is no evidence that their deep layers are unusual: by 3 months postmax, SN 2006X was like a CN.
The spectrum of the extreme CL SN 1991bg has been thought to have undergone an early transition to the nebular phase (Filippenko et al. 1992a; Leibundgut et al. 1993; Turatto et al. 1996; Mazzali et al. 1997). According to our interpretation (Fig. 17) there is no need to appeal to forbidden lines in the 3 week postmax spectrum (Fig. 17), and even in the 3 month postmax spectrum the only obvious one is \[Ca II\] 7291, 7324. In spite of its peculiarities, the spectrum of SN 1991bg does increasingly resemble the spectra of other SNe Ia after maximum light, and in all respects SN 1986G appears to be a link between SN 1991bg–likes and normal SNe Ia. As discussed in Paper I, the spectroscopic peculiarities of SN 1991bg–likes may largely (but not entirely) reflect a temperature threshold below which key ionization ratios change abruptly. The issue of whether SN 1991bg–likes are a distinct subgroup remains open.
The 1 week and 3 week postmax spectra of the SS SN 2000cx are unique among SNe Ia observed so far. Our identifications of high velocity Ti II and especially Cr II are not definite, but for nonstandard spectra, nonstandard identifications are to be expected, and no alternative identifications have been suggested. The resemblance of the 3 months postmax spectrum of SN 2000cx to CNs suggests, though, that its deeper layers are like those of other SNe Ia. The SN 2002cx-likes, on the other hand, appear to be radically different from other SNe Ia, even in their deeper layers. Pending the discovery of SNe Ia having less mild SN 2002cx–like properties, the SN 2002cx–likes appear to be a discrete subgroup. Yet the resemblance of their spectra, when artificially blueshifted, to other SNe Ia (Fig. 19) suggests kinship.[^1]
Although this series of papers is confined to spectroscopy, it is worth emphasizing that the spectroscopic diversity is bound to contribute to the photometric diversity. Branch et al. (2004) discussed the effects of the putative high velocity Ti II of SN 2000cx on the evolution its $B-V$ color. Another good example is the 1 week postmax spectra of the extreme BLs SN 2002bf, SN 1984A, and SN 2006X (Figures 5 and 6). The deep Fe II absorption from 4700 to 5100 Å, near the interesection of the B and V bands, throws flux right into the peak of the V band (thus obscuring the S II absorptions) and is bound to affect the B-V color evolution, with implications for attempts to estimate the interstellar extinction on the basis of broad band photometry (e.g., Wang et al. 2007 for SN 2006X).
In recent years it has been shown (Mannucci et al. 2005, 2006; Scannapieco & Bildsten 2005; Sullivan et al. 2006) that some SNe Ia, called prompt, are produced by a young ($\sim 10^8$ yr) stellar population while others, called tardy, are produced by an older population (several Gyr). Most the SNe Ia that have ever occurred were prompt, but at the present epoch most of the SNe Ia at low redshift are tardy. This raises the question of which of our spectroscopic groups are prompt and which are tardy. Since the extreme CL events such as SN 1991bg tend to occur in old populations and the extreme SS events such as SN 1991T tend to occur in young populations (Sullivan et al. 2006 and references therein), if we are to break our groups into a minority of prompts and a majority of tardies, then the members of the SS group should tend to be prompts and the members of the other three groups tend to be tardies. Yet there is no obvious spectroscopic separation between the SS group and the CN group. The CN group was defined to emphasize the strong spectroscopic homogeneity of its members, but several of the SSs (e.g., SN 1999ee, SN 1999ac, SN 1999aa) are not very different from CN. As discussed in Paper II, it may be that there are two evolutionary paths to SNe Ia that require different amounts of time but produce two families of SNe Ia that have overlapping distributions of their properties.
That being said, Quimby et al. (2007) suggested that even among normal SNe Ia (in the broad sense of the term) there may be two distinct classes of events: those that have a smoothly declining density distribution in their outer layers and therefore show a blueshift of the 6100 Å absorption that decreases smoothly with time, and those in which silicon is largely confined to a shell–like density structure and therefore have a more constant blueshift of the 6100 Å absorption. The smooth density structure is characteristic of deflagration and delayed detonation models, while a shell is characteristic of pulsating delayed detonation and tamped detonation models. The suggestion of Quimby et al. was motivated by a prolonged period of nearly constant blueshift in SN 2005hj, but they point out that some other SNe Ia such as SN 1999aa and SN 2000cx are similar in this respect. In our classification, these three events are SS. More well observed SS and CN SNe Ia are needed to determine whether SS and CN SNe Ia are distinctly different in this respect.
SYNOW is useful for establishing line identifications and information on the velocity intervals in which the lines are forming, but it is no substitute for detailed calculations of spectra of hydrodynamical models. At present no 1D explosion model has been to shown to account well for the observed spectral evolution, not even for the CNs (Baron et al. 2006). Spectropolarimetry and explosion modeling tells us that asymmetries are present, but the spectra of 3D explosion models cannot yet even be calculated in full detail (although a start has been made by calculating the spectra of 3D models for angle averaged compositions; Baron et al. 2007). The homogeneity of the CNs shows that there is a standard, repeatable SN Ia explosion model that does not have large compositional inhomogeneities near or below the maximum photospheric velocity of about 12,000 — but what is this model? And there are many additional questions, such as: what variation on the CN model is responsible for the extra high velocity matter in the BLs, and the putative bizarre high velocity features of SN 2000cx, and the distinctive properties of the SN 2002cx–likes? We have hardly begun on the path toward understanding the various manifestions of spectroscopic diversity of SNe Ia.
We are grateful to all observers who have provided spectra. This work has been supported by NSF grants AST 0506028 and AST 0707704, NASA LTSA grant NNG04GD36G, and DOE grant DEFG02-07ER41517.
Barbon, R., Iijima, T., & Rosino, L. 1989, A&A, 220, 83
Baron, E., Bongard, S., Branch, D., & Hauschildt, P. H. 2006, ApJ, 645, 480
Baron, E., Jeffery, D. J, & Branch, D., Bravo, E., Garcia–Senz, D., & Hauschildt, P. H., 2007, ApJ, in press (arXiv:0709.4177)
Benetti, S., & Mendes de Oliveira, C. 1995, IAU Circ. No. 6135
Benetti, S., et al. 2004, MNRAS, 348, 261
Benetti, S., et al. 2005, ApJ, 623, 1011
Bowers, E., et al. 1997, MNRAS, 290, 663
Branch, D., et al. 1983, ApJ, 270, 123
Branch, D., Baron, E., Hall, N., Melakayil, M., & Parrent, J. 2005, PASP, 117, 545 (Paper I)
Branch, D., et al. 2003, AJ, 126, 1489
—–. 2004, PASP, 116, 903
—–. 2006, PASP, 118, 560 (Paper II)
—–. 2007, PASP, 119, 709 (Paper III)
Chornock, D., Filippenko, A. V., Branch, D., Foley, R. J., Jha, S., & Li, W. 2006, PASP, 118, 722
Cristiani, S., et al. 1992, A&A, 259, 63
Elias–Rosa, N., et al. 2006, MNRAS, 369, 1880
Filippenko, A. V., et al. 1992a, AJ, 104, 1543
Filippenko, A. V. 1997, in Proc. NATO Advanced Study Institute, Thermonuclear Supernovae, ed. P. Ruiz–Lapuente, R. Canal, & J. Isern (Dordrecht: Kluwer), 1
Fisher, A., Branch, D., Hatano, K., & Baron, E. 1999, MNRAS, 304, 67
Garavini, G., et al. 2004, AJ, 128, 387
—–. 2005, AJ, 130, 2278
Garnavich, P. M., et al. 2004, ApJ, 613, 1120
Gomez, G., & Lopez, R. 1998, AJ, 115, 1096
Hamuy, M., et al. 2002, AJ, 124, 417
Hicken, M., Garnavich, P. M., Prieto, J. L., Blondin, S., DePoy, D. L., Kirshner, R. P., & Parrent, J., ApJ, 669, 17
Howell, D. A., & Nugent, P. 2004, in Cosmic Explosions in Three Dimensions, eds. P. Höflich, P. Kumar, & J. C. Wheeler (Cambridge, CUP), p. 151
Howell, D. A., et al. 2006, Nature, 443, 308
Jeffery, D. J., Ketchum, W., Branch, D., Baron, E., Elmhamdi, A., & Danziger, I. J. 2007, ApJS, 171, 493
Jha, S., et al. 1999, ApJS, 125, 73
Kasen, D., et al. 2003, ApJ, 593, 788
Kirshner, R. P., et al. 1993, 415, 589
Kotak, R., et al. 2006, A&A, 436, 1021
Krisciunas, K., et al. 2007, AJ, 133, 58
Leibundgut, B., et al. 1993, AJ, 105, 301
Leonard, D. C., Li, W., Filippenko, A. V., Foley, R. J., & Chornock, R. 2005, ApJ, 632, 450
Li, W., et al. 1999, AJ, 117, 2709
—–. 2001, PASP, 113, 1178
—–. 2003, PASP, 115, 453
Mannucci, F., Della Valle, M., & Panagia, N. 2006, MNRAS, 370, 773
Mannucci, F., et al. 2005, A&A, 433, 807
Mattila, S., et al. 2005, A&A, 443, 649
Mazzali, P. A., et al. 1997, MNRAS, 284, 151
Millard, J., et al. 1999, ApJ, 527, 746
Pastorello et al. 2007, MNRAS, 377, 1531
Patat, F., et al. 1996, MNRAS, 278, 111
Phillips, M. M., et al. 2007, PASP, 119, 360
Quimby, R., Höflich, P., & Wheeler, J. C., ApJ, 666, 1083
Sadakane, K., et al. 1996, PASJ, 48, 51
Salvo, M. E., Cappellaro, E., Mazzali, P. A., Benetti, S., Danziger, I. J., Pata, F., & Turatto, M. 2001, MNRAS, 321, 254
Scannapieco, E., & Bildsten, L. 2005, ApJ, 629, L85
Stanishev, V., et al. 2007, in preparation
Sullivan, M., et al. 2006, ApJ, 648, 868
Thomas, R. C., Branch, D., Baron, E., Nomoto, K., Li, W., & Filippenko, A. V. 2004, ApJ, 601, 1019
Turatto, M., et al. 1996, MNRAS, 283, 1
Wang, L., et al. 2003, ApJ, 591, 1110
Wang, X., et al. 2007, ApJ, in press (arXiv:0708.0140)
Wells, L. A., et al. 1994, AJ, 108, 2233
![Synthetic spectra for Fe II with flat ([*solid line*]{}) and exponential ([*dashed line*]{}) optical depth distributions are compared. The parameters for the flat case are $\tau$(Fe II)=12, $v_{\rm phot} = 6000$ , and $v_{\rm max}=13,000$ . For the exponential case they are $\tau$(Fe II)=200, $v_{\rm phot} = 9000$ , and $v_e = 1000$ .](f1.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![ The 1 week–postmax spectra of the CN SN 1998bu ([*solid line*]{}) from Jha et al. (1999) and the CN SN 1996X ([*dashed line*]{}) from Salvo et al. (2001) are compared.](f3.eps){width=".8\textwidth"}
![The 1 week postmax spectrum of the CN SN 1996X ([ *solid line*]{}) from Salvo et al. (2001) compared with a synthetic spectrum ([*dashed line*]{}).](f4.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![The 1 week postmax spectrum of the BL SN 2002bf ([*solid line*]{}) from Leonard et al. (2005) compared with a synthetic spectrum ([*dashed line*]{}). Unlabelled absorption features in the synthetic spectrum are produced by Fe II.](f6.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![The 1 week–postmax spectrum of the CL SN 1999by ([ *solid line*]{}) from Garnavich et al. (2004) compared with a synthetic spectrum ([*dashed line*]{}).](f8.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![ The 1 week postmax spectra of the SS SN 1999ac ([*solid line*]{}) from Garavini et al. (2005) and the CN SN 1996X ([*dashed line*]{}) from Salvo et al. (2001) are compared. The arrows point to the lower blueshifts of Si II and S II and the higher blueshifts of Fe II, in SN 1999ac.](f10.eps){width=".8\textwidth"}
![The 1 week postmax spectrum of the SS SN 1999ac ([*solid line*]{}) from Garavini et al. (2005) compared with a synthetic spectrum ([*dashed line*]{}).](f11.eps){width=".8\textwidth"}
![The 1 week postmax spectrum of the SS SN 2000cx ([*solid line*]{}) from Li et al. (2001) compared with a synthetic spectrum ([*dashed line*]{}). The synthetic absorption near 8000 Å is produced by O I 8446 but the neighboring observed absorptions are produced by the Ca II IR3.](f12.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![The 3 week–postmax spectrum of the CN SN 1996X ([*solid line*]{}) from Salvo et al. (2001) compared with a synthetic spectrum ([*dashed line*]{}).](f14.eps){width=".8\textwidth"}
{width=".8\textwidth"}
{width=".8\textwidth"}
![The 3 week–postmax spectrum of the CL SN 1991bg([*solid line*]{}) from Filippenko et al. (1992a) compared with a synthetic spectrum ([*dashed line*]{}).](f17.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![ The 3 week–postmax spectra of the SS SN 2005hk ([*solid line*]{}) from Stanishev et al. (2007) but artificially blueshifted by 5000 , and the CN SN 1996X ([*dashed line*]{}) from Salvo et al. (2001) are compared.](f19.eps){width=".8\textwidth"}
![The 3 week–postmax spectrum of the SS SN 2005hk ([*solid line*]{}) from Stanishev et al. (2007) compared with a synthetic spectrum ([*dashed line*]{}). Unlabelled absorption features in the synthetic spectrum are produced by Fe II.](f20.eps){width=".8\textwidth"}
{width=".8\textwidth"}
![The 3 month–postmax spectrum of the CN SN 2003du ([*solid line*]{}) from Stanishev et al. (2007) compared with a synthetic spectrum ([*dashed line*]{}). The forbidden–line identifications that appear above some of the flux peaks are from Bowers et al. (1997) but these are not used in the SYNOW synthetic spectrum. Unlabelled absorption features in the synthetic spectrum are produced by Fe II.](f22.eps){width=".8\textwidth"}
{width=".8\textwidth"}
{width=".8\textwidth"}
![The 3 month–postmax spectrum of the CL SN 1991bg ([ *solid line*]{}) from Filippenko et al. (1992a) compared with a synthetic spectrum ([*dashed line*]{}).](f25.eps){width=".8\textwidth"}
{width=".8\textwidth"}
[lcll]{}
1981B BL & 22, 93 & NGC 4526 & Branch et al. 1983\
1984A BL & 8, 19 & NGC 4419 & Barbon et al. 1989\
1986G CL & 21, 90 & NGC 5128 & Cristiani et al. 1992\
1989B CL & 8, 19, 92 & NGC 3627 & Wells et al. 1994\
1990N CN & 7, 21 & NGC 4639 & 7: Leibundgut et al. 1991; 21: Filippenko et al. 1992b\
1991M BL & 81 & IC 1151 & Gomez & Lopez 1998\
1991T SS & 83 & NGC 4527 & A. V. Filippenko, unpublished\
1991bg CL & 19, 91 & NGC 4374 & Filippenko et al. 1992a\
1992A BL & 6 & NGC 1380 & Kirshner et al. 1993\
1994D CN & 7, 19, 87 & NGC 4526 & 7, 19: Patat et al. 1996; 87: Filippenko 1997\
1994ae CN & 89 & NGC 3370 & Bowers et al. 1997\
1995D SS & 96 & NGC 2962 & Bowers et al. 1997\
1996X CN & 7, 22, 87 & NGC 5061 & Salvo et al. 2001\
1997br SS & 8 & ESO 576-G40 & Li et al. 1999\
1998aq CN & 7, 21, 91 & NGC 3982 & Branch et al. 2003\
1998bu CN & 8 & NGC 3368 & Jha et al. 1999\
1999aa SS & 6, 19 & NGC 4469 & Garavini et al. 2004\
1999ac SS & 8 & NGC 2848 & Garavini et al. 2005\
1999by CL & 7 & NGC 2841 & Garnavich et al. 2004\
1999ee SS & 22 & IC 5179 & Hamuy et al. 2002\
2000cx SS & 7, 20, 89 & NGC 524 & Li et al. 2001\
2001el CN & 20 & NGC 1448 & Wang et al. 2003\
2002bf BL & 7 & & Leonard et al. 2005\
2002bo BL & 82 & NGC 3190 & Benetti et al. 2004\
2002cx SS & 21 & & Li et al. 2003\
2002er BL & 6, 20 & UGC 10743 & Kotak et al. 2006\
2003cg CN & 7, 23 & NGC 3169 & Elias–Rosa et al. 2006\
2003du CN & 7, 84 & UGC 9391 & Stanishev et al. 2007\
2004S CN & 8, 19 & & Krisciunas et al. 2007\
2004eo CL & 7, 21 & NGC 6928 & Pastorello et al. 2007\
2005hk SS & 21 & UGC 272 & Stanishev et al. 2007\
2006X BL & 6, 98 & NGC 4321 & Wang et al. 2007\
2006gz SS &7 & IC 1277 & Hicken et al. 2007\
[lccccccc]{}
() & 11,000 & 7000 & 11,000 & 6000 & 6000 & 11,000\
$\tau$(O I) & 0.25/\[17\] & 0.2/\[16\] & 0.2/\[15\] & 0.3/\[14\] & 0.1/\[16\] & 0.2/\[17\]\
$\tau$(Na I) & 0.3/\[15\] & 0.3/\[12\] & 0.5/\[14\] & & 0.1/\[14\] & 0.2/\[13\]\
$\tau$(Mg I) & & & & 1/\[13\] & &\
$\tau$(Mg II) & 0.6/\[15\] & 0.2/\[13\] & & & 0.5/\[14\] &\
$\tau$(Si II) & 2/\[15\] & 0.5/\[13\] & 3/\[21\] & 1.2/\[12\] & 0.5/\[14\] & 1/13\[15\]\
$\tau$(S II) & 0.8/\[13\] & 0.3/\[11\] & 0.2/\[15\] & & 0.3/\[10\] & 0.3/13\[15\]\
$\tau$(Ca II) & 30/\[17\] & 15/\[24\] & 30/\[23\] & 100/\[14\] & 25/\[20\] & 2/\[23\]\
$\tau$(Sc II) & & & & 2/\[11\] & &\
$\tau$(Ti II) & & & & 0.8/\[11\] & & 0.2/23\[25\]\
$\tau$(Cr II) & & & & 2/\[11\] & & 0.3/21\[23\]\
$\tau$(Fe II) & 1/\[15\] & 0.8/\[13\] & 1.5/\[21\] & 7/\[11\] & 0.8/\[15\] & 0.3/19\[22\]\
$\tau$(Fe III) & 0.3/\[15\] & 0.3/\[13\] & & & & 0.3/13\[15\]\
$\tau$(Co II) & 0.6/\[15\] & 0.6/\[13\] & & & &\
$\tau$(Ni II) & & & & & & 0.1/13\[15\]\
[lcccc]{}
() & 6000 & 6000 & 4000 &7000\
$\tau$(O I) & & 0.6/8\[13\] & &\
$\tau$(Na I) & 0.4/7\[17\] & 0.6/10\[13\] & 0.2/6\[9\] & 0.6/\[18\]\
$\tau$(Si II) & 0.7/8\[13\] & 1.2/7\[11\] & & 0.7/9\[16\]\
$\tau$(Ca II) & 50/\[16\] & 500/9\[14\] & 50/\[9\] &\
$\tau$(Ti II) & & 3/\[11\] & &\
$\tau$(Cr II) & 6/\[13\] & 10/\[11\] & 10/\[8\] & 0.3/\[18\]\
$\tau$(Fe II) & 12/\[13\] & 10/\[11\] & 30/\[8\] & 0.8/\[18\]\
$\tau$(Fe III) & & & & 0.5/\[15\]\
$\tau$(Co II) & 6/\[13\] & & 10/\[8\] &\
[lcc]{}
() & 7000 & 3000\
$\tau$(Na I) & 4/\[15\] & 4/\[15\]\
$\tau$(Ca II) & 10,000/\[14\] & 1000/\[12\]\
$\tau$(Fe II) & 8/\[12\] & 2/\[10\]\
$\tau$(Cr II) & & 0.5/\[10\]\
[^1]: An anonymous referee points out that another possible discrete subgroup could be SNe Ia that have very slow light curve decline rates, such as SN 2001ay (Howell & Nugent 2004), SN 2003fg (Howell et al. 2006), and SN 2006gz (Hicken et al. 2007). Subclassifying SNe Ia by means of photometric characteristics is beyond the scope of this paper, but we note that in our spectroscopic classification scheme, SN 2001ay is broad line (Paper II) while SN 2003fg (Paper III) and SN 2006gz (this paper) are shallow silicon.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The 5G New Radio (NR) standard for wireless communications supports the millimetre-wave (mmWave) spectrum to yield unprecedented improvement of the access network capacity. However, intermittent blockages in the mmWave signal may degrade the system performance and lead to the under-utilisation of the allocated resources. To circumvent this problem, the transmission slot-time shall be adjusted according to the blockage condition, avoiding the resource under-utilisation. In this paper, we propose that the 5G NR flexible numerology should be applied to adapt the slot-time in order to mitigate the blockage effects. We validate this claim by analysing the expected data rate of a mmWave system, under a range of blockage scenarios. We show that different blockage scenarios may require different numerologies to produce best performance, and that the correct choice of numerology may improve this performance by as much as hundreds of Mbps. Our results carry insights important for the design of blockage-aware scheduling mechanisms for 5G.'
author:
-
bibliography:
- 'myreferences.bib'
title: 'Application of Flexible Numerology to Blockage Mitigation in 5G-mmWave Networks'
---
millimetre-wave networks, self-body blockage, flexible numerology, performance analysis, blockage mitigation.
Introduction
============
The fifth-generation (5G) of mobile networks is being developed to boost the available mobile speeds to multi-Gbps, and, consequently, provide support for the increasing user traffic demands [@itu2018setting]. To achieve this goal, 5G networks will use the wide bandwidths available in frequencies. The challenge is that ordinary objects (e.g., human bodies, furniture) that are to signals transmitted over microwave frequencies become blockages when the same signals are transmitted over .
Blockages in signal propagation are related to severe attenuation of the signal power (in certain cases, the blockage may add as much as of attenuation [@maccartney2018rapid]), which can lead to radio link failures and consequent disconnection in the communication. This issue has mostly been addressed in the literature to date by deployment strategies that allow the network to exploit spatial macro-diversity, i.e., increasing the communication robustness by enabling the user to receive a signal from distinct points in space. These deployment strategies include: reflective surfaces [@narayanan2017coverage], relay nodes [@kim2017relay], dense networks [@bai2015coverage], ceiling-mounted [@firyaguna2017coverage], and movable [@gapeyenko2018effects] that can position themselves in a way that increases the likelihood of having an operating in . In order to achieve spatial macro-diversity, the layer mechanisms should properly coordinate the network nodes and allocate the transmission resources (e.g., time, frequency, space) according to the blockage condition. Yet, as we show in our numerical results, the intermittency of blockage events may cause system performance degradation and lead to resource under-utilisation if a fixed is considered. In this paper, we consider adaptable transmission times for blockage mitigation, using the flexible proposed for the 5G . The ability of adjustment is enabled by the 5G access technology. It works with a flexible transmission frame system, in which the configuration of , i.e., and , is flexible, as illustrated in Figure \[fig:example\_scheduling\_5G\]. According to the 5G terminology [@5gnr_38211], such configuration is referred to as the *flexible numerology* and the supported numerologies are listed in Table \[tab:resource\_type\]. Originally flexible numerology was introduced to enable service-level differentiation, i.e., network slicing for different 5G use cases [@shafi2017tutorial].
![Exemplary application of different numerologies in a 5G frame. The resource blocks are allocated for and within a bandwidth $b$ .[]{data-label="fig:example_scheduling_5G"}](figures/example_scheduling_5G.pdf){width=".95\linewidth"}
Herein, we propose an alternative application for flexible numerology. Our claim is that different numerologies will fare better under blockage conditions, and hence may be used to improve the mmWave user performance. We verify our claim by analysing the mmWave link performance using the numerologies available for 5G mmWave systems, under a range of blockage scenarios, defined and empirically-validated in [@yoo2017channel]. Our results show that there is a trade-off between the high transmission efficiency, achieved with longer , and the high probability of LOS transmission, achieved using shorter . In consequence, the same numerology used for two different blockage scenarios (office and car-park) leads to opposing conclusions about the system performance, and that the correct choice of numerology may improve this performance by as much as hundreds of Mbps. Effectively, we identify conditions under which it may be favourable to use a given numerology, which shall provide insights important for the design of blockage-aware scheduling mechanisms for 5G.
The rest of the paper is organised as follows. In Section \[sec:related\_work\], we present the state-of-the-art and how our work goes beyond it. In Section \[sec:system\_model\], we describe our system model. In Section \[sec:performance\_metric\], we describe our performance metric. In Section \[sec:numerical\_results\] we analyse the link performance comparing the numerologies and the blockage conditions. Finally, we draw our conclusions in Section \[sec:conclusion\].
[lrrr]{} & & &\
& & &\
0 & 1 & 4.690 & 180\
1 & 0.5 & 2.345 & 360\
2\* & 0.25 & 1.172 & 720\
3\* & 0.125 & 0.586 & 1440\
4\* & 0.0625 & 0.293 & 2880\
\
\[tab:resource\_type\]
Related Work {#sec:related_work}
============
The literature on blockage mitigation in mmWave communication is mostly focused on techniques that rely on spatial macro-diversity. Such techniques allow the transmitter to find an alternative physical path for the mmWave signal when the primary path fails due to a blockage event. The main techniques considered are: (i) *reflectors*: usage of surfaces made of materials that reflect the mmWave signal to cover an obstructed spot through a path [@kwon2018multibeam; @feng2017dealing]; (ii) *relays*: forwarding the transmission to a relay node that has a path with the [@wu2017coverage; @yang2018sense]; (iii) *movable*: moving the location during the transmission to a position where there is a path [@bao2017blockage]; (iv) *multi-connectivity*: associating the with multiple , so the can have a path served by a backup [@tatino2018maximum; @petrov2018achieving].
It is the responsibility of the layer to coordinate the extra communication nodes (e.g., relay nodes, neighbour ), and provide a smooth handover between the , relays, or reflectors when the mmWave signal power fades due to blockage [@tesema2017multiconnectivity; @giordani2016multiconnectivity; @petrov2017dynamic; @polese2017improved]. However, the intermittent blockages together with fixed may lead to poor utilisation of the transmission resources. Therefore, to avoid this under-utilisation, we propose the application of flexible numerology to mitigate blockage effects through layer transmission time adaptation.
In state-of-the-art flexible numerology has been applied to improve the network latency where the is optimised according to a latency deadline restriction [@ibrahim2016numerology] and according to the traffic pattern [@patriciello20185g]. Also, it has been applied to improve the frame spectral efficiency when multiplexing different types of services, e.g., and [@lagen2018subband; @you2018resource].
System Model {#sec:system_model}
============
We consider a single cell, with an installed on the ceiling or a lamppost, transmitting a 5G frame to a at a distance $d_\mathrm{A}$ in the horizontal plane. The is installed at a height $h_\mathrm{A}$ above the UE level, as illustrated in Figure \[fig:body\_blockage\_side\_view\]. This setup shall generalise over the two deployment scenarios considered in [@yoo2017channel], but is also in-line with the 3GPP-defined scenarios for 5G mmWave system evaluation [@5gnr_38913]: indoor office with ceiling-mounted access points and outdoor car-park with lamppost mounted access points. We assume the resource allocation decision in the is made every , which we refer to as the **. For ease of exposure, we consider the link performance as experienced by a single user, attached to a single cell. The cell’s bandwidth is $b$ and can be filled using flexible numerology $\mu$ with resource blocks of bandwidth $b_\mu$ and $t_\mu$, as illustrated in Figure \[fig:example\_scheduling\_5G\].
[.55]{} ![Body blockage model. An inside the blockage free zone is never blocked by the body regardless of its orientation. (a) For the given body and heights, an is inside the blockage free zone when $d_\mathrm{A} < z_\mathrm{B}$. (b) Outside this zone, the is blocked when its orientation lies in the shadowed cone of width $\varphi_\mathrm{B}$.[]{data-label="fig:body_blockage_model"}](figures/body_blockage_side_view.pdf "fig:"){width="\linewidth"}
[.45]{} ![Body blockage model. An inside the blockage free zone is never blocked by the body regardless of its orientation. (a) For the given body and heights, an is inside the blockage free zone when $d_\mathrm{A} < z_\mathrm{B}$. (b) Outside this zone, the is blocked when its orientation lies in the shadowed cone of width $\varphi_\mathrm{B}$.[]{data-label="fig:body_blockage_model"}](figures/body_blockage_top_view.pdf "fig:"){width="\linewidth"}
Blockage Probability Model
--------------------------
Whether or not the path to the is blocked during a slot within the depends on the blockage probability. We define the blockage probability as the probability of a slot being blocked during a given interval. This probability is given by the probability of self-body blockage as in [@firyaguna2017coverage]: $$p = \left\{\;
\begin{split}
& \frac{1}{\pi} \arctan\left(\frac{w_\mathrm{B}}{2 r_\mathrm{B}}\right), \; d_\mathrm{A} \geq z_\mathrm{B}; \\
& 0, \; \text{otherwise};
\end{split}
\; \right.
\label{eq:prob_blockage}$$ where $w_\mathrm{B}$ is the body width, $r_\mathrm{B}$ is the distance between the body and the UE, and $h_\mathrm{B}$ is the distance between the UE level and the top of the body, and $z_\mathrm{B}=r_\mathrm{B} \frac{h_\mathrm{A}}{h_\mathrm{B}}$ is the self-body blockage free zone radius, as illustrated in Figure \[fig:body\_blockage\_model\].
Signal-to-Noise Ratio
---------------------
For modelling of the mmWave signal propagation, we consider the experimentally-validated channel model proposed in [@yoo2017channel]. The model consists of the path loss and the composite Gamma-Nakagami-m fading, whose parameters take one of two values depending on whether the user body blocks the path, meaning that the model parameters change with the random blockage state. The fading and path-loss coefficients for that model were estimated from the experimental data collected for a mmWave operating at in [@yoo2017channel].
We define the set of the two possible blockage states as $ \chi = \mbox{\footnotesize $\{\mathrm{LOS},\mathrm{NLOS}\}$}$. Hence, given the blockage state $X=x \: \in \chi$, we can define the path loss as $l_x = \ell_x \cdot \left(\sqrt{d_\mathrm{A}^2+h_\mathrm{A}^2}\right)^{-\nu_x}$, where $\sqrt{d_\mathrm{A}^2+h_\mathrm{A}^2}$ is the Euclidean distance from the to the , $\ell_x$ is the path loss at one metre distance under free space propagation, and $\nu_x$ is the attenuation exponent.
Instead of treating each fading component individually, we consider the fading gain as a single random variable $H_x$ with a composite fading distribution, as obtained in [@laourine2009performance]. This approach allows us to define the complementary cumulative distribution function (ccdf) of the SNR in (\[eq:snr\]) based on the formula in [@laourine2009performance (15)] as follows: $$\begin{gathered}
F^c_{Y|X}(y;\:\bar{y}_x,m_x,\alpha_x,\beta_x) \\
= \mathsf{A} \Gamma(m_x) \sum\limits_{i=0}^{m_x} \frac{2^i y^{m_x-i}}{(BD)^i(m_x-i)!} \frac{\mathcal{K}_{m_x-i+\frac{1}{2}}(\mathsf{B}\sqrt{\mathsf{C}+\mathsf{D}y})}{(\mathsf{B}\sqrt{\mathsf{C}+\mathsf{D}y})^{m_x-i+\frac{1}{2}}}
\label{eq:ccdf_snr}
\end{gathered}$$ where $\bar{y}_x = \frac{\rho}{\sigma} \: l_x$ is the SNR without the fading component, the variable $m_x$ is the Nakagami-m fading parameter, ($\alpha_x,\beta_x$) are the Gamma shadowing parameters, $\mathcal{K}_{o}(\cdot)$ is the modified Bessel function of the second kind of order $o$, and the used constants are: $$\begin{split}
\mathsf{A} &= \frac{(\alpha_x \Bar{y}_x)^\frac{1+2m_x}{4}}{\Gamma(m_x)} \sqrt{\frac{2\alpha_x\beta_x}{\pi}} \exp(\alpha_x\beta_x) \left(\frac{m_x}{\Bar{y}_x}\right)^{m_x}, \\
\mathsf{B} &= \beta_x \sqrt{\frac{\alpha_x}{\Bar{y}_x}}, \\
\mathsf{C} &= \alpha_x \Bar{y}_x, \\
\mathsf{D} &= 2m_x/\beta_x,
\end{split}$$ where $\Gamma(\cdot)$ is the Gamma function.
Thus, we define the signal-to-noise ratio (SNR) $Y$, conditioned on $X=x$, as: $$Y_x = \frac{\rho}{\sigma} \: l_x \: H_x,
\label{eq:snr}$$ where $\rho$ is the transmit power, $\sigma$ is the noise power, and $H_x$ is the fading gain.
Transmission Efficiency and Slot Aggregation Efficiency
-------------------------------------------------------
We define the transmission efficiency $\eta_\mu \in [0,1]$ of a resource block of type $\mu$ as the decrease in the spectral efficiency for shorter due to inter-symbol interference caused by shorter [@lahetkangas2014achieving]. The longer the , the greater the transmission efficiency, i.e., $\eta_i > \eta_j$ for all $i > j$.
We also define the slot aggregation efficiency $\zeta_\mu \in [0,1]$ as the ratio between the number of symbols that are used for data transmission and the total number of symbols.
Performance Analysis {#sec:performance_metric}
====================
We evaluate the performance of our system in terms of the expected data rate. To calculate it, we consider the probability of blockage, the spectral efficiency of the channel, the transmission efficiency of each resource block, and the slot aggregation efficiency.
We consider that the $t_\mu$ is a multiple of $\Delta t$ (see Figure \[fig:time\_structure\]), which is the blockage interval. Within this interval, the blockage event has probability $p$ as described in (\[eq:prob\_blockage\]) and is independent of the previous interval, but once the first blockage event happens, we assume that all the following slots within the $\tau$ are also blocked[^1]. Consequently, a slot, that contains $k$ coherence intervals ($t_\mu = k \cdot \Delta t$), is transmitted in if no blockage has occurred in previous slots and in each of its own coherence intervals. Thus, the probability of the $i$-th slot being in is: $$\begin{split}
\Pb[X_i = \mbox{\scriptsize $\mathrm{LOS}$}] = (1-p)^{k \cdot i} = (1-p)^{i \frac{t_\mu}{\Delta t}}.
\label{eq:prob_los}
\end{split}$$
![The $\tau$ contains a sequence of slots with duration $t_\mu$. The is a multiple of the blockage interval $\Delta t$. When a blockage occurs during one $\Delta t$ interval within a slot, the entire slot and the following ones are considered , as the blockage duration is expected to be significantly larger than the .[]{data-label="fig:time_structure"}](figures/time_structure.pdf){width=".9\linewidth"}
The spectral efficiency $S$ can be expressed as $S=\log(1+Y)$, and has only non-negative values. Hence, the expected value conditioned to the blockage state of the $i$-th slot can be represented by this integral which can be efficiently computed numerically: $$\Eb[S_{x_i}|X_i=x_i] = \int\limits_0^\infty F_{Y|X}^c(2^s-1; \: \bar{y}_x,m_x,\alpha_x,\beta_x) \: \dd{s},
\label{eq:expected_se}$$ where $F_{Y|X}^c(\cdot)$ is ccdf of the SNR defined in (\[eq:ccdf\_snr\]). Thus, the expected spectral efficiency with respect to the blockage state of the $i$-th slot is: $$\begin{split}
\Eb_{X_i,S}\left[S_{X_i}\right] &= \sum\limits_{x_i \in \chi} \Pb[X_i=x_i] \Eb[S_{x_i}|X_i=x_i] \\
&\overset{(a)}= (1-p)^{i \frac{t_\mu}{\Delta t}} \cdot \Eb[S_\mathrm{LOS}|X=\mbox{\scriptsize $\mathrm{LOS}$}] \\
&+ \big(1-(1-p)^{i \frac{t_\mu}{\Delta t}}\big) \cdot \Eb[S_\mathrm{NLOS}|X=\mbox{\scriptsize $\mathrm{NLOS}$}],
\label{eq:expected_se_x}
\end{split}$$ where (a) comes from using (\[eq:prob\_blockage\]) and from the fact that the channel is constant with the same blockage state.
Finally, the expected data rate using the numerology $\mu$ ($\bar{R_\mu}$) of a slot aggregation of $\xi_\mu$ slots is given by the expectation of the sum of the spectral efficiency of each slot multiplied by the frame bandwidth and the transmission and slot aggregation efficiencies: $$\bar{R_\mu} = b \cdot \zeta_\mu \cdot \eta_\mu \cdot \Eb_{X,S}\left[\sum\limits_{i=1}^{\xi_\mu} S_{X_i}\right].$$ Then, using (\[eq:expected\_se\_x\]), $\bar{R_\mu}$ becomes as expressed in (\[eq:expected\_rate\]):
$$\begin{split}
\bar{R_\mu} &= b \, \zeta_\mu \, \eta_\mu \, \sum\limits_{i=1}^{\xi_\mu} \Eb_{X,S}\left[S_{X_i}\right]
= b \, \zeta_\mu \, \eta_\mu \, \sum\limits_{i=1}^{\xi_\mu} \left( (1-p)^{i \frac{t_\mu}{\Delta t}} \, \Eb[S_\mathrm{LOS}|X=\mbox{\scriptsize $\mathrm{LOS}$}] + \big(1-(1-p)^{i \frac{t_\mu}{\Delta t}}\big) \, \Eb[S_\mathrm{NLOS}|X=\mbox{\scriptsize $\mathrm{NLOS}$}] \right) \\
&= b \, \zeta_\mu \, \eta_\mu \, \Big( \xi_\mu \, \Eb[S_\mathrm{NLOS}|X=\mbox{\scriptsize $\mathrm{NLOS}$}] +
\frac{(1-p)^{ \frac{t_\mu}{\Delta t}}((1-p)^{\xi_\mu \frac{t_\mu}{\Delta t}}-1)}{(1-p)^{ \frac{t_\mu}{\Delta t}}-1} \left( \Eb[S_\mathrm{LOS}|X=\mbox{\scriptsize $\mathrm{LOS}$}] - \Eb[S_\mathrm{NLOS}|X=\mbox{\scriptsize $\mathrm{NLOS}$}] \right) \Big)
\label{eq:expected_rate}
\end{split}$$
Numerical Results {#sec:numerical_results}
=================
In this section, we show the benefits of the application of the flexible numerology to blockage mitigation in 5G NR mmWave systems. We compare the results for three types of resource blocks supported by mmWave NR (i.e., $\mu=2,3,4$ according to 5G NR Rel-15 [@5gnr_38211]), under different blockage scenarios.
We consider two blockage scenarios: a UE held in a pocket (, *UE in pocket*), and a UE operated with the hand (, *UE in hand*). The *UE in pocket* scenario is a severe blockage condition where the body is obstructing half of the , thus, the probability of blockage is $p=0.5$. The *UE in hand* scenario is a common blockage condition where the user is, for example, operating an app in the mobile phone, and the body obstructs a smaller angle than in the *UE in pocket* scenario. We consider two environments: an indoor open *office* and an outdoor *car park*. This setup reflects the scenarios and environments characterised in [@yoo2017channel], for which coefficients of the fading and path loss models we use were estimated. These coefficients are listed on the left side of Table \[tab:model\_parameters\] (based on [@yoo2017channel Table I]).
We assume that the transmission efficiency $\eta_\mu$ decreases by 5% with each increment in the numerology $\mu$ (i.e., $\eta_2=1.00$, $\eta_3 = 0.95$ and $\eta_4 = 0.90$). Accurate values for the numerology-dependent transmission efficiency can be obtained following the calculations presented in [@lahetkangas2014achieving]. We assume that the slot structure is as described in [@mogensen2013small], where each slot consists of 14 symbols. In a slot aggregation of $\xi_\mu = \tau/t_\mu$ slots, the two first symbols are used for downlink and uplink control, the third is used for demodulation reference signal, and the rest of the symbols is for data. Thus, the slot aggregation efficiency is $\zeta_\mu = 1 - \frac{3 t_\mu}{14 \tau}$. We set the frame bandwidth as [@5gnr_38211] and we evaluate the performance with the (unless specified otherwise), as in the legacy LTE scheduling. For our analysis, we set as the shortest among the numerologies considered. All other fixed system parameters are shown in Table \[tab:model\_parameters\] (right side).
[lcrcr|crcr|cc|lcr]{} &\
& & & & Transmit Power & $\rho$ & 20 dBm\
& & & & & & & Noise Density & $\sigma/b$ & -174 dBm/Hz\
& & & $\nu$ & & $\alpha$ & & $\alpha$ & & & & Body Width & $w_\mathrm{B}$ & 40 cm\
Office & & 45.1 & & 57.4 & & 0.15 & & 0.20 & & & Body Height\* & $h_\mathrm{B}$ & 40 cm\
Car Park & & 48.7 & & 88.8 & & 0.11 & & 0.23 & & & AP Height\* & $h_\mathrm{A}$ & 5 m\
Environment and Blockage Impact
-------------------------------
In this subsection, we evaluate the impact of two environments (an *office* and a *car park*), with distinct channel characteristics, and two blockage scenarios (*UE in hand* and *UE in pocket*) on the expected data rate of mmWave communication. We also compare the scenarios where the UE is close to the AP () and where the UE is far from the AP (). The results are shown in Figure \[fig:avgR\_bar\_environment\].
[.9]{}
[.9]{}
From the left side of Figure \[fig:avgR\_bar\_hand\_ta1ms\], we see that the expected data rate is higher when using a of (blue bar), compared to other s, for a user close to the AP and operating the UE with the hand. In this case, the blockage probability is very low, allowing for high transmission efficiency of the resource block with long slot duration to have a more significant impact on the data rate than the blockage. On the right-hand side of that same figure, we observe a decrease in the expected data rate because of the dual-effect of increased path loss and blockage probability for a user further away from the AP. The *car park* environment suffers more from blockages compared to the *office* environment as there is less power in the signal, likely due to lack of reflecting/scattering environment. Thus, short (, green bar) mitigates the blockage effects by increasing the expected number of slots in LOS and, then, yields better performance in the *car park* environment.
From Figure \[fig:avgR\_bar\_pocket\_ta1ms\], we note that the expected data rate has similar trends when we vary the distance between the UE and the AP, as the considered range of distances have little effect on the blockage probability for the UE in the pocket. We see that the user in the *office* environment achieves highest expected data rate using , and in the *car park* environment, the user achieves highest expected data rate using in both cases (UE close to AP and far from AP) of the *UE in pocket* scenario.
Scheduling Interval Impact
--------------------------
In LTE networks, the scheduler makes the allocation decision every , which has a fixed duration of , and the decision is valid until the next [@capozzi2013downlink]. The 5G NR allows slot aggregation, in which the aggregation duration can span two or more slots to reduce control overhead. Hence, the scheduling decision interval is no longer fixed and can vary with the slot aggregation size. Here, we evaluate the impact of the aggregation overhead reduction by comparing the s (short )[^2] and (long )[^3]. The results considering the *car park* environment are shown in Figure \[fig:avgR\_bar\_scheduling\].
[.9]{}
[.9]{}
From Figure \[fig:avgR\_bar\_scheduling\], we observe that, in most cases, a short achieves better performance with short (green bar) compared to long (blue bar). The only exception, as we see from the left side of Figure \[fig:avgR\_bar\_carpark\_hand\], is the scenario of a user close to the AP operating the UE with the hand. In this scenario, the best performance is achieved by long with long . This is because the effects of increased transmission efficiency (long ) and reduced overhead (long ) have a higher impact on the expected data rate than the low blockage probability. In other cases, i.e., UE far from AP in *UE in hand* scenario and both cases in *UE in pocket* scenario (see Figure \[fig:avgR\_bar\_carpark\_pocket\]), the blockage probability is high and has more impact on the expected data rate than the overhead reduction. Thus, there is no benefit in increasing the in those cases.
To sum up, the appropriate combination of numerology and scheduling interval is essential to achieve the best performance. For example, in the cases where the blockage has more significant impact on the expected data rate (e.g., UE far from AP or UE in pocket), the use of short with short is recommended to avoid prolonged exposure to blockage interruptions. On the other hand, in the cases where the blockage has less impact (e.g., UE close to AP), the use of long with long is recommended to take advantage of the reduced overhead and the high transmission efficiency.
Conclusion {#sec:conclusion}
==========
In this paper, we proposed that 5G NR flexible numerology be used to mitigate the negative effects of body blockage in 5G systems. We presented an analytical framework that allowed us to show and understand the benefits of our proposed application of flexible numerology. We showed that different blockage conditions require different combinations of numerology and slot aggregation to achieve the best performance, as presented in Table \[tab:recommendation\]. The effectiveness of flexible numerology in mmWave blockage scenarios is a consequence of the trade-off between the high transmission efficiency, achieved with long , and the high probability of LOS transmission, achieved using short .
-- ---------- ---------- ------------- -------------
UE close UE far UE close UE far
to AP from AP to AP from AP
$\mu=2$ $\mu=2$ $\mu=2$ $\mu=4$
$\tau=5$ $\tau=5$ $\tau=5$ $\tau=0.25$
$\mu=2$ $\mu=2$ $\mu=4$ $\mu=4$
$\tau=5$ $\tau=5$ $\tau=0.25$ $\tau=0.25$
-- ---------- ---------- ------------- -------------
: Recommended Numerology $\mu$ and Scheduling Interval $\tau$ (in )[]{data-label="tab:recommendation"}
This work is a stepping stone to further studies on the application of flexible numerology to blockage mitigation in 5G-mmWave networks. Further work is needed to investigate the implications of multiple users or services sharing the frame, as well as how the effectiveness of flexible numerology affects the blockage mitigation via macro-diversity. Nonetheless, the results we have shown thus far should motivate the development of new scheduling algorithms/policies for the 5G NR mmWave frame.
Acknowledgement {#acknowledgement .unnumbered}
===============
This publication has emanated from the research conducted within the scope of *NEMO (Enabling Cellular Networks to Exploit Millimetre-wave Opportunities)* project financially supported by the Science Foundation Ireland (SFI) under Grant No. 14/US/I3110 and with partial support of the European Regional Development Fund under Grant No. 13/RC/2077. We are thankful to Danny Finn who kindly assisted us with the Wolfram Mathematica^^ software.
{#ap:scripts}
All scripts used to generate the presented results were written in Wolfram Mathematica^^ and are available in *https://github.com/firyaguna/wolfram-flexible-numerology*.
[^1]: For 5G NR it has been shown in [@jain2018driven] that the body blockage duration can be in the order of (due to low mobility of pedestrians) versus the duration frames.
[^2]: A of allows to allocate at least one of the longest () considered.
[^3]: Any longer between leads to similar results.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
In this paper we exhibit two infinite families of trees $\{T^1_n\}_{n \geq 17}$ and $\{T^2_n\}_{n \geq 17}$ on $n$ vertices, such that $T^1_n$ and $T^2_n$ are non-isomorphic, co-spectral, with co-spectral complements, and the right-angled Coxeter groups (RACGs) based on $T^1_n$ and $T^2_n$ have the same geodesic growth with respect to the standard generating set. We then show that the spectrum of a tree is not sufficient to determine the geodesic growth of the RACG based on that tree, by providing two infinite families of trees $\{S^1_n\}_{n \geq 11}$ and $\{S^2_n\}_{n \geq 11}$, on $n$ vertices, such that $S^1_n$ and $S^2_n$ are non-isomorphic, co-spectral, with co-spectral complements, and the right-angled Coxeter groups (RACGs) based on $S^1_n$ and $S^2_n$ have distinct geodesic growth.
Asymptotically, as $n\rightarrow \infty$, each set $T^i_n$, or $S^i_n$, $i=1,2$, has the cardinality of the set of all trees on $n$ vertices. Our proofs are constructive and use two families of trees previously studied by B. McKay and C. Godsil.
2000 Mathematics Subject Classification: 20E08, 20F65.
Key words: geodesic growth, Coxeter group, RACG, regular languages.
author:
- 'Laura Ciobanu & Alexander Kolpakov'
title: 'Geodesic growth of right-angled Coxeter groups based on trees'
---
Introduction
============
The *geodesic growth function* of a group $G$ with respect to a finite generating set $S$ counts, for each positive integer $n$, the number of geodesics of length $n$ starting at the identity $1_G$ in the Cayley graph of $G$ with respect to $S$. The *geodesic growth series* is the formal power series that takes the values of the geodesic growth function as its coefficients (see Definition \[defgrowth\]).
The groups that we consider in this paper, right-angled Coxeter groups, or RACGs, are known to have a regular language of geodesics with respect to the standard generating sets, and therefore rational geodesic growth series (see [@JJJ] or [@BjornerBrenti Theorem 4.8.3] for proofs of these facts). Our goal here is to obtain more specific data concerning the geodesic growth of RACGs. Namely, we are interested in extracting information about the geodesic growth series from the defining graph of the group. It is known that non-isomorphic graphs define non-isomorphic RACGs [@rad], and that non-isomorphic RACGs can have equal geodesic growth [@AntolinCiobanu]. However, we are interested in knowing how much the similarities or differences between two defining graphs influence the geodesic growth of the corresponding RACGs.
Two RACGs $G_i = G(\Gamma_i)$, $i=1,2$, with non-isomorphic defining graphs $\Gamma_i$, may have equal standard growth (see Definition \[defgrowth\] (2)); this can be determined by computing the $f$-polynomials of the graphs $\Gamma_i$ [@Davis Proposition 17.4.2]. However, their geodesic growth exhibits more subtle properties [@AntolinCiobanu], and in general it is not known which graph theoretic conditions completely determine the geodesic growth of a RACG. The examples of non-isomorphic graphs $\Gamma_i$ defining RACGs with equal geodesic growth in [@AntolinCiobanu] are degree-regular and have cycles. In this paper we consider the case when the $\Gamma_i$’s are trees. Although these are some of the simplest classes of graphs, we already encounter a phenomenon that shows a great difference between the behaviour of the standard growth and that of the geodesic growth. The standard growth of each RACG $G_i$ is determined solely by the number of vertices (or edges) in the respective tree $\Gamma_i$, while the geodesic growth can be distinct even for two co-spectral trees (which might have co-spectral complements as well). Recall that two graphs are [*co-spectral*]{} if the characteristic polynomials of their adjacency matrices are the same.
In this paper we count and compute with the help of an automaton generating the geodesic language in a RACG based on a tree $T$. This automaton reflects some of the path information from $T$. In general, a good deal of combinatorial information about $T$ can be extracted from its spectrum. However, we encounter two rather different behaviours: on one hand we produce (infinitely many) pairs of trees $T_1$ and $T_2$ which are non-isomorphic, co-spectral, with co-spectral complements, and the RACGs $G(T_i)$, $i=1,2$, based on them have the same geodesic growth; on the other hand we obtain (infinitely many) pairs of trees $S_1$ and $S_2$ which are non-isomorphic and co-spectral, with co-spectral complements, whose respective RACGs $G(S_i)$ have distinct geodesic growth. This shows that the spectrum of a tree alone does not determine the geodesic growth of the RACG based on that particular tree.
The following two theorems, proved in Sections \[s:racg\] and \[s:racg-diff-growth\], respectively, are the main results of the paper:
\[thm1\] There exist two families of trees $\mathcal{T}^1_n = \{ T^1_1, T^1_2, \dots \}$ and $\mathcal{T}^2_n = \{ T^2_1, T^2_2, ... \}$ on $n\geq 17$ vertices such that for all $i\geq 1$:
1. $T^1_i$ and $T^2_i$ are not isomorphic, but co-spectral, with co-spectral complements, and
2. the RACGs $G(T^1_i)$ and $G(T^2_i)$ have equal geodesic growth series.
Theorem \[thm1\] thus answers positively Question 1 in [@AntolinCiobanu Section 8].
\[thm2\] There exist two families of trees $\mathcal{S}^1_n = \{ S^1_1, S^1_2, \dots \}$ and $\mathcal{S}^2_n = \{ S^2_1, S^2_2, ... \}$ on $n\geq 11$ vertices such that for all $i\geq 1$:
1. $S^1_i$ and $S^2_i$ are not isomorphic, but co-spectral, with co-spectral complements, and
2. the RACGs $G(S^1_i)$ and $G(S^2_i)$ have distinct geodesic growth series.
We note that McKay showed in [@McKay] that the trees $T^1_i$ and $T^2_i$ in Theorem \[thm1\] are simultaneously co-spectral, have co-spectral complements and co-spectral line graphs. On the contrary, the trees $S^1_i$ and $S^2_i$ from Theorem \[thm2\] might have line graphs with distinct spectra. Moreover, by a result of McKay [@McKay], in both Theorem \[thm1\] and Theorem \[thm2\] the cardinality of each family $\mathcal{T}^i_n$ and $\mathcal{S}^i_n$ tends asymptotically to the cardinality of the set $\Upsilon_n$ of all trees on $n$ vertices: $$\frac{\mathrm{card}\, \mathcal{T}^i_n}{\mathrm{card}\, \Upsilon_n} \rightarrow 1 \mbox{\,\,\, and \,\,\,} \frac{\mathrm{card}\, \mathcal{S}^i_n}{\mathrm{card}\, \Upsilon_n} \rightarrow 1, \mbox{\,\,\, as \,\,\,} n\rightarrow \infty, \ i=1, 2.$$ The two theorems above, together with McKay’s results, lead us to the following conjecture:
If two trees are simultaneously co-spectral, have co-spectral complements and co-spectral line graphs, then they have the same geodesic growth.
Section \[s:equal\_special\] contains details about the computation of several kinds of geodesics, and their numbers, for the trees introduced in Section \[s:racg\].
The authors have created a Python code $\mathrm{Monty}$[^1] [@Pyth], which performs the computations needed for the results in this paper by using either SAGE standard routines for symbolic computation of determinants and rational expressions of growth series, or the Berlekamp-Massey algorithm for restoring the rational expression for a growth series from a sufficient number of its coefficients, as a more efficient approach. Our Python code constructs finite-state automata that accept the geodesic languages in RACGs based on triangle-free graphs (in this case, trees), and then proceeds to determine the respective growth series.
Definitions and notation {#s:def}
========================
Let $S$ be a finite set and $S^*$ the free monoid on $S$. We identify $S^*$ with the set of words over $S$, that is, finite sequences of elements of $S$. We use $| . |$ to denote word length.
Let $G={\left\langleS\right\rangle}$ be a group generated by $S$. For an element $g$ of $G$, denote by $|g|_S$ the word length of $g$ with respect to $S$. Given $w\in S^*$, we denote by $\overline{w}$ the image of $w$ in $G$ under the natural projection $S^* \rightarrow G$.
A word $w$ over an alphabet $S$ is [*geodesic*]{} in $G = \langle S \rangle$ if $|w|=|\overline{w}|_S$. The set of geodesics in $G$ with respect to $S$ will be denoted by $Geo(G)$ or $Geo(S)$.
Let ${\Gamma}={\Gamma}(G,S)$ be a simple (no loops, no multiple edges) graph with vertex set $V(\Gamma)=S$ (or simply $V$) and edge set $E(\Gamma)$ (or simply $E$), where $E \subseteq V \times V$. The RACG based on $\Gamma$ is given by the presentation $$\langle s \in S \mid s^2=1 \ \forall s\in S, \ \textrm{and}\ (ss')^2=1, \, \forall \{s, s'\} \in E \rangle.$$ It is easy to see that for any two involutions $s$ and $s'$ the relation $(ss')^2=1$ implies $ss'=s's$. This leads to another possible presentation for RACGs: $\langle s \in S \mid s^2=1, ss'=s's, \, \forall \{s, s'\} \in E \rangle.$
In the present paper we use the same letters for the vertices of ${\Gamma}$ and the corresponding generators of the group $G({\Gamma})$.
The [*star*]{} of a vertex $v\in V$ in $\Gamma$, denoted by $\operatorname{St}_{\Gamma}(v)$, or $\operatorname{St}(v)$ if the ambient graph is clear in the given context, is the set of vertices in $\Gamma$ that are adjacent to $v$. That is, $$\operatorname{St}_{\Gamma}(v)=\{w \in V \mid \{v, w\} \in E \}.$$
We now describe a finite deterministic automaton that recognises geodesics in RACGs (see [@hu] for definitions of languages and automata). We define such automata in the standard way, as quintuples $(Q,\Sigma, \delta, q_0, F)$, where $Q$ is the finite set of *states*, $\Sigma$ the input alphabet, $\delta$ the *transition function*, $q_0$ the *initial* or *start* state, and $F$ the set of *final* or *accepting* states. The following definition is a simplified version of Proposition 4.1 in [@AntolinCiobanu].
\[def:DFA\] Let $\Gamma=(V,E)$ be a tree. The deterministic finite state automaton recognising the geodesics in $G(\Gamma)$ is $A=(Q, S, \delta, \{\emptyset\}, F)$, where the set of states is $Q=\{\emptyset\} \cup V \cup E \cup \{ \rho\}$, $\rho$ is the unique “fail” state, and $\{\emptyset\}$ is the start state. The input alphabet is $S=V$, the set of accept states $F$ is $\emptyset \cup V \cup E$, and the transition function ${\delta}:Q \times S \rightarrow Q$ is given by
1. ${\delta}(\sigma, s)=(\operatorname{St}(s) \cap \sigma) \cup \{s\}$, for $s \notin \sigma$;
2. ${\delta}(\sigma,s)=\rho$, otherwise.
Any set $L$ of words over an alphabet $\Sigma$ gives rise to a *strict growth function* $f_L:\mathbb{N} \rightarrow \mathbb{N}$, defined by $$f_L(n) := |\{W \in L \mid |W| = n\}|.$$
\[defgrowth\] Let $G$ be a group generated by $S$. Then we define the following:
1. The *geodesic growth function* $f_{Geo(G)}:\mathbb{N}\rightarrow \mathbb{N}$ is given by $$f_{Geo(G)}(r):= f_{Geo(G,S)}(r)=|\{w\in S^* \mid |w|=|\overline{w}|_S=r\}|,$$ and the *geodesic growth series* of $G$ equals $$\mathcal{G}_{(G,S)}(t)=\sum_{r=0}^\infty f_{Geo(G)}(r)\, t^r.$$
2. The *spherical (standard) growth function* $\sigma_{(G,S)}:\mathbb{N}\rightarrow \mathbb{N}$ is given by $$\sigma_{(G,S)}(r)=|\{g\in G \mid |g|_S=r\}|,$$ and the *spherical (standard) growth series* of $G$ equals $$\Sigma_{G}(t):=\Sigma_{(G,S)}(t)=\sum_{r=0}^\infty \sigma_{(G,S)}(r)\, t^r.$$
We recall that the [*$f$-polynomial*]{} of a graph $\Gamma$ is the generating function for the number of cliques (that is, complete subgraphs) of size $i$ in $\Gamma$: $f(t)=f_0+f_1t+f_2t^2+ \dots $, where $f_i$ is the number of $i$-cliques in $\Gamma$. We consider the empty set to be a clique on zero vertices, and therefore $f_0=1$. We remark that the spherical (standard) growth function of a RACG is determined by the $f$-polynomial of its defining graph [@Davis Proposition 17.4.2]. We want to contrast this to the fact that the geodesic growth function is not uniquely determined by the $f$-polynomial, or even by the spectrum of the defining graph, as the following sections show.
RACGs with equal geodesic growth {#s:racg}
================================
In this section we prove Theorem \[thm1\] by giving an explicit construction of the families of trees $\mathcal{T}^i_n$, $i=1,2$.
We define the *coalescence* $\tau \cdot \sigma$ of two rooted trees $\tau$ and $\sigma$ to be the tree which results from merging $\tau$ and $\sigma$ at their roots. The tree $\tau \cdot \sigma$ has as vertex set the union of the vertex sets of $\tau$ and $\sigma$, and as root the identification of the roots of $\tau$ and $\sigma$, as shown in Fig. \[fig:trees0\].
(0)[1]{} (1)[2]{} (2)[3]{} (2)[4]{}
(0)(1) (1)(2) (2)(3) (2)(4)
(0)[A]{} (A)[B]{} (B)[C]{} (B)[D]{}
(A)(B) (B)(C) (B)(D)
(0)[1]{} (1)[2]{} (2)[3]{} (2)[4]{}
(0)(1) (1)(2) (2)(3) (2)(4)
(1)[A]{} (1)[B]{} (1)[C]{}
(1)(A) (1)(B) (1)(C)
Consider the trees $T_1$ and $T_2$, both rooted at $0$, first described by McKay in [@McKay], as shown in Fig. \[fig:trees1\]. In [@McKay] the following fact is proved:
(0)[1]{} (0)[2]{} (0)(1) (0)(2) (1)[3]{} (1)[4]{} (2)[5]{} (1)(3) (1)(4) (2)(5) (3)[6]{} (4)[7]{} (5)[8]{} (5)[9]{} (3)(6) (4)(7) (5)(8) (5)(9) (8)[10]{} (9)[11]{} (10)[12]{} (11)[13]{} (8)(10) (9)(11) (10)(12) (11)(13) (13)[14]{} (13)[15]{} (13)(14) (13)(15)
(0)[1]{} (0)[2]{} (0)(1) (0)(2) (1)[3]{} (2)[4]{} (2)[5]{} (1)(3) (2)(4) (2)(5) (3)[6]{} (3)[7]{} (3)(6) (3)(7) (4)[8]{} (5)[9]{} (8)[10]{} (9)[11]{} (4)(8) (5)(9) (8)(10) (9)(11) (11)[12]{} (11)[13]{} (11)(12) (11)(13) (12)[14]{} (13)[15]{} (12)(14) (13)(15)
\[thm:McKay1\] Let $T$ be a rooted tree with at least two vertices and with root labelled $0$. Then the trees $\Gamma_i = T\cdot T_i$, $i=1,2$, are not isomorphic, but are co-spectral. Also, their complements $\overline{\Gamma_i}$ and line graphs $L(\Gamma_i)$, $L(\overline{\Gamma_i})$, $\overline{L(\Gamma_i)}$, $i=1,2$, are respectively co-spectral.
Now let $G_1:=G(T_1)$ and $G_2:=G(T_2)$ be the RACGs associated to the trees $T_1$ and $T_2$ defined above. First note that since the trees $T_1$ and $T_2$ are isomorphic as graphs, the groups $G_1$ and $G_2$ are isomorphic. However, $T_1$ and $T_2$ are not isomorphic as rooted trees.
Let $\tau$ be a tree with $n$ vertices. Fix a labelling $\{0, \dots, n-1\}$ of the vertices of $\tau$, and suppose that $0$ represents $\tau$’s root. Define $\Gamma_1=\tau \cdot T_1$ and $\Gamma_2=\tau \cdot T_2$ to be the trees obtained as the coalescence of $\tau$ with $T_i$ at vertex $0$, and let $G(\Gamma_1)$ and $G(\Gamma_2)$ be the RACGs based on $\Gamma_1$ and $\Gamma_2$, respectively. Since $\Gamma_1$ and $\Gamma_2$ are non-isomorphic, $G(\Gamma_1)$ and $G(\Gamma_2)$ are non-isomorphic, as well.
The following lemma is the key ingredient of Theorem \[thm1\].
\[equal\_growth\] The groups $G(\Gamma_1)$ and $G(\Gamma_2)$ have the same geodesic growth series.
In order to simplify the exposition in the proof of Lemma \[equal\_growth\] we use the notation below.
\[not:series\]
Let $G$ be a group with generating set $T$ containing the letter $0$. Denote the set of words in $Geo(T)$ starting with $0$ by $Geo_0(T)$, the set of words in $Geo(T)$ ending in $0$ by $Geo^0(T)$, and the set of words in $Geo(T)$ ending and starting with $0$ by $Geo_0^0(T)$.
[*(of Lemma \[equal\_growth\])*]{} Notice that $f_{Geo(\Gamma_1)}(r)$ (respectively $f_{Geo(\Gamma_2)}(r)$) is equal to the number of all words of length $r$ in $\Gamma_1^*$ (respectively, $\Gamma_2^*$) minus the number of those words of length $r$ in $\Gamma_1^*$ that are not geodesics. We denote the number of non-geodesics by $\overline{f_{Geo(\Gamma_1)}}(r)$ (and $\overline{f_{Geo(\Gamma_2)}}(r)$, respectively). Since $| \Gamma_1| = |\Gamma_2|$, clearly $f_{Geo(\Gamma_1)}(r)=f_{Geo(\Gamma_2)}(r)$ if and only if $\overline{f_{Geo(\Gamma_1)}}(r)=\overline{f_{Geo(\Gamma_2)}}(r)$.
We now show that $\overline{f_{Geo(\Gamma_1)}}(r)=\overline{f_{Geo(\Gamma_2)}}(r)$ for all $r \geq 1$. Any word in $\Gamma_1^*$ can be written as $w_1u_1w_2 \dots u_n$, where $u_i \in (\tau\setminus\{0\})^*$ and $w_i \in T_1^*$. A non-geodesic word $w$ in $\Gamma_1^*$ belongs to one of the following sets (or is of the type), depending on its form:
- $w$ contains non-geodesics $u_i \in (\tau\setminus\{0\})^*$ or $w_j \in T_1^* $, or both, for some $1\leq i,j \leq n$, or
- all $w_i$ and $u_i$ are geodesic on their respective alphabets, and there exists $1\leq j <n$ such that $0 u_j 0$ is a subword of $w$ with $u_j \in (\operatorname{St}_{\Gamma_1}(0) \cap \tau)^*$, or
- all $w_i$ and $u_i$ are geodesic on their respective alphabets, and $w$ contains a subword of the form $s 0 s$, where $s \in (\operatorname{St}_{\Gamma_1}(0) \cap \tau)^*$.
Notice that the set of non-geodesics is then $A \cup B \cup C$, where $A \cap(B \cup C) =\emptyset$ and $B\cap C \neq \emptyset$.
The number of words in $A$ depends only on the geodesic growth series of $\tau$ and $T_1$, and thus it will be equal to the number of words of type $A$ in $\Gamma_2^*$.
The computations in Section \[s:equal\_special\] show that the following identities hold: $f_{Geo(T_1)}(r)=f_{Geo(T_2)}(r)$, $f_{Geo_0(T_1)}(r)=f_{Geo_0(T_1)}(r)$, $f_{Geo^0(T_1)}(r)=f_{Geo^0(T_2)}(r)$ and $f_{Geo_0^0(T_1)}(r)=f_{Geo_0^0(T_2)}(r)$ for all $r \geq 1$, as a result of (\[equalities\]). This means that there is a length-presenting bijection $\phi_0$ between $Geo_0(T_1)$ and $Geo_0(T_2)$, i.e. for each geodesic $w=0v \in Geo_0(T_1)$ there is a geodesic $w'=0v'=\phi_0(w) \in Geo_0(T_2)$, and $|w|=|w'|$. Analogously, there is a length-preserving bijection $\phi^0:Geo^0(T_1) \rightarrow Geo^0(T_2)$, and a bijection $\phi_0^0:Geo_0^0(T_1) \rightarrow Geo_0^0(T_2)$. This means there is a length-preserving bijection $\phi:Geo_0(T_1)\cup Geo^0(T_1)\rightarrow Geo_0(T_2)\cup Geo^0(T_2)$ between those geodesics starting or ending with $0$, for which we have that $\phi |_{Geo_0} = \phi_0$, $\phi |_{Geo^0} = \phi^0$ and $\phi |_{Geo_0^0} = \phi_0^0$. By the computations in (\[equalities\]) we have that $f_{(Geo_0 \cup Geo^0)(T_1)}(r)=f_{(Geo_0 \cup Geo^0)(T_2)}(r)$ by the standard formula for the cardinality of the union of two sets, and since $f_{Geo(T_1)}(r)=f_{Geo(T_2)}(r)$, we also have that $f_{Geo \setminus (Geo_0 \cup Geo^0)(T_1)}(r)=f_{Geo \setminus (Geo_0 \cup Geo^0)(T_2)}(r)$. Thus, there is a length-preserving bijection $\psi$ between $Geo \setminus (Geo_0 \cup Geo^0)(T_1)$ and $Geo \setminus (Geo_0 \cup Geo^0)(T_2)$.
The bijection $\phi$ can be extended to the set $Geo(T_1)$ of all geodesics on $T_1$ by letting $\phi(w)=\psi(w)$ for all $w \in Geo \setminus (Geo_0 \cup Geo^0)(T_1)$, and then furthermore extended to $Geo(T_1) \cup Geo(\tau \setminus \{0\})$ by letting $\phi(w)=w$ for all $w \in Geo(\tau \setminus \{0\}) $.
Then $\phi$ provides a bijection between the words of type $B$ in $\Gamma_1^*$ and the words of type $B$ in $\Gamma_2^*$. To see this, associate to each $w_1u_1w_2 \dots u_n$ the word $\phi(w_1)\phi(u_1) \dots \phi(w_n)\phi(u_n)=\phi(w_1)u_1 \dots \phi(w_n)u_n$. Then $w_i$ ends in $0$, $w_{i+1}$ starts with $0$, and $u_i\in \operatorname{St}_{\tau}(0)^*$ if and only if $\phi(w_i)$ ends in $0$, $\phi(w_{i+1})$ starts with $0$, and $\phi(u_i)=u_i\in \operatorname{St}_{\tau}(0)^*$, by definition.
It is immediate to see that $\phi$ also provides a bijection between the words of type $C$, and between the words of type $B \cap C$ in $\Gamma_1$ and $\Gamma_2$, respectively. Thus, there is a length-preserving bijection between the words of type $A \cup B \cup C$ (*i.e.* all non-geodesic words) in $\Gamma_1$ and $\Gamma_2$. This concludes the proof of the lemma.
[*(of Theorem \[thm1\])*]{} Let $\Upsilon_k = \{ \tau_1, \tau_2, \dots \}$ be the set of non-isomorphic trees on $k \geq 2$ vertices. For $i=1,2$ the two families of trees $\mathcal{T}^i_n = \{\tau \cdot T_i | \tau \in \Upsilon_k \}$, $k=n-15$, satisfy Theorem \[thm1\].
From Theorem \[thm:McKay1\] we already know that $\Gamma_1=\tau \cdot T_1$ and $\Gamma_2=\tau \cdot T_2$ are co-spectral, for any $\tau \in \Upsilon_k$. By Lemma \[equal\_growth\] the groups $G(\Gamma_1)$ and $G(\Gamma_2)$ have the same geodesic growth series.
\[thm:McKay2\] Let $p_i(n)$, $i=1,2$, be the proportion of trees on $n$ vertices that have $T_i$ as a limb. Then $p_1(n) = p_2(n)$ for all $n$ and $\lim_{n\rightarrow \infty} p_i(n) = 1$.
From Theorem \[thm:McKay2\], we obtain that $\frac{\mathrm{card}\, \mathcal{T}^i_n}{\mathrm{card}\, \Upsilon_n} \rightarrow 1$, as $n \rightarrow \infty$.
Computing the numbers of special geodesics {#s:equal_special}
==========================================
In this section we prove, by concrete computations, the equalities between the numbers of special geodesics required in the proof of Lemma \[equal\_growth\].
Let $T_1$ and $T_2$ be as in Figure \[fig:trees1\], and recall Notation \[not:series\]. For each group $G_i = G(T_i)$, $i=1,2$, we construct a finite automaton $A_i$ accepting the geodesic language of $G_i$, as described in Definition \[def:DFA\], and with the help of $A_i$ we compute the growth series $$\gamma_{G_1}(t) = \sum_{r\geq 0}\, f_{Geo(T_1)}(r)\,t^r,\,\, _0\gamma_{G_1}(t) = \sum_{r\geq 0}\, f_{Geo_0(T_1)}(r)\,t^r,$$ $$^0\gamma_{G_1}(t) = \sum_{r\geq 0}\, f_{Geo^0(T_1)}(r)\,t^r, \mbox{ and } ^0_0\gamma_{G_1}(t) = \sum_{r\geq 0}\, f_{Geo_0^0(T_1)}(r)\,t^r.$$ We also compute the analogous growth series $\gamma_{G_2}(t)$, $_0\gamma_{G_2}(t)$, $^0\gamma_{G_2}(t)$ and $^0_0\gamma_{G_2}(t)$ for $G_2$, where the series coefficients are given by the sequences $Geo(T_2)(r)$, $Geo(T_2)_0(r)$, $Geo(T_2)^0(r)$ and $Geo(T_2)^0_0(r)$, respectively.
Our computations, which we elaborate upon below, show that $$\label{equalities}
\gamma_{G_1}(t) = \gamma_{G_2}(t), \ _0\gamma_{G_1}(t) = {_0\gamma_{G_2}(t)}, \ ^0\gamma_{G_1}(t)= {^0\gamma_{G_2}(t)} \ \textrm{and} \ ^0_0\gamma_{G_1}(t) = {^0_0\gamma_{G_2}(t)}.$$
These identities prove the equality of the corresponding numbers of geodesics.
Let $A$ be a deterministic finite-state automaton with accepting states $q_i$, $i=0,1,\dots, N$, where $q_0$ is the start state and the “fail” state is denoted by $q$. Let $M = M(A)$ be the transition matrix of the automaton $A$. Computing the generating function $\gamma_A (t)$ of $A$ is a standard technique, and the formula for $\gamma_A(t)$ is $$\label{DFAseries}
\gamma_A(t) = \frac{e^T\, M w}{\det(I - t M)},$$ where $I$ is the $N \times N$ identity matrix, and $e=(1, 0, \dots, 0)^T$ and $w=(1,1, \dots, 1)^T$ are two vectors in $\mathbb{Z}^N$.
Now let $A_i$ be the deterministic finite-state automata accepting the language of geodesics in $G_i$, $i=1,2$, and let $M_i = M(A_i)$ be the transition matrix of $A_i$. Then the geodesic growth series $\gamma_{G_i}(t)$ is a rational function (determined by the equality (\[DFAseries\])), and the coefficients of its numerator and denominator can be easily computed, see [@Epstein]. The Python code $\mathrm{Monty}$ [@Pyth] may perform the above computation either by finding the symbolic determinant $\det(I - t M_i)$ (usually slow), or by applying the Berlekamp-Massey algorithm (a faster one). This Python code starts by creating the finite-state automaton $A_i$, given a triangle-free graph $T_i$ (in this case, a tree), and then proceeds to determining $\gamma_{G_i}(t)$.
Since $G_1$ and $G_2$ are isomorphic, the equality $\gamma_{G_1}(t) = \gamma_{G_2}(t)$ is immediate. Here we provide an explicit formula for these identical growth series. The output of $\mathrm{Monty}$ for both trees $T_1$ and $T_2$ consists of two finite-state automata $A_i$ which are isomorphic, as one only renumbers the vertices of $T_2$ in order to obtain $T_1$. Each $A_i$ has $32$ states and $466$ transition arrows, and the corresponding growth series are
&\_[G\_1]{}(t) = \_[G\_2]{}(t) = (1+t)(1 + 2t - 2t\^3 - 4t\^4 - t\^5)(1 + 5t + 10t\^2 + 9t\^3 - 5t\^4 - 26t\^5 - 34t\^6&\
& - 22t\^7 - t\^8 + 7t\^9 + 4t\^[10]{}) (1 - 8t - 85t\^2 - 243t\^3 - 222t\^4 + 332t\^5 + 1194t\^6 + 1349t\^7 + 132t\^8&\
& - 1510t\^9 - 2008t\^[10]{} - 1088t\^[11]{} + 28t\^[12]{} + 359t\^[13]{} + 170t\^[14]{} + 15t\^[15]{})\^[-1]{}.&
Now we compute the functions $_0\gamma_{G_i}(t)$, $i=1,2$, which will turn out to be equal, as well. However, in this case the equality is not known to hold beforehand: even though the groups $G_1$ and $G_2$ are isomorphic, the image of vertex $0$ in $T_1$ under the canonical isomorphism does not correspond to vertex $0$ in $T_2$. We shall explicitly compute the generating function $_0\gamma_{G_i}(t)$ for the number of geodesic words starting with $0$ accepted by each $A_i$, $i=1,2$, and then compare these functions. Suppose that the start state of each $A_i$ is $q_0 = \emptyset$. Let the corresponding transition function $\delta_i$ be so that $\delta_i(q_0, 0) = q_{k_i} = \{0\}$. If a word $w$ labels a path from $q_{k_i}$ to an accept state, then the word $0w$ is a geodesic word in $G_i$ starting with $0$. Thus, we have to compute the generating function $_0\alpha_{G_i}(t)$ for the number of words starting at $q_{k_i}$ and ending at an accept state: $$_0\alpha_{G_i}(t) = \frac{e^T\, M_i\, w}{\det(I - t M_i)},$$ where $e = (0, \dots, 0, \underbrace{1}_{k_i}, 0, \dots, 0)^T$, and $w = (1, 1, \dots, 1)^T$. Then we use the fact that $_0\gamma_{G_i}(t) = t \cdot {_0\alpha_{G_i}(t)}$. By symmetry, we get $^0\gamma_{G_i}(t) = {_0\gamma_{G_i}(t)}$, $i=1,2$.
By using $\mathrm{Monty}$ we obtain
&\_0\_[G\_1]{}(t) = [\_0\_[G\_2]{}(t)]{} = t (1+t) (1 + 2t - 2t\^3 - 4t\^4 - t\^5) (1 + 4t + 4t\^2 - 3t\^3 - 9t\^4 - 5t\^5 + 3t\^6&\
& + t\^7 - 3t\^8 - 3t\^9) (1 - 8t - 85t\^2 - 243t\^3 - 222t\^4 + 332t\^5 + 1194t\^6 + 1349t\^7 + 132t\^8&\
& - 1510t\^9 - 2008t\^[10]{} - 1088t\^[11]{} + 28t\^[12]{} + 359t\^[13]{} + 170t\^[14]{} + 15t\^[15]{})\^[-1]{}.&
Finally, it remains to compute the growth series $^0_0\gamma_{G_i}(t)$.
There are two kinds of words forming disjoint subsets of the geodesic language of $G_i$ that we are interested in:
- the words $w$ starting at the accept state $q_{k_i} = \delta(q_0, 0) = \{0\}$ of $A_i$ and coming back to it: then the geodesic word $0w$ starts and ends with a “0” (since there are only $0$-transitions leading to $q_{k_i} = \{0\}$ and no $0$-transition coming out of $q_{k_i}$), by Definition \[def:DFA\] (i).
- the words $w$ starting at the state $q_{k_i} = \{0\}$ and ending at a state $q_{l_i}$ such that $\delta(q_{l_i}, 0) = q_{m_i} \neq q_{k_i}$ and $q_{m_i}$ is an accept state: then the word $0w0$ will be a geodesic word starting and ending with a “0”.
Let $^0_0\alpha_{G_i}(t)$ be the generating function for the words of type (I), and $^0_0\beta_{G_i}(t)$ be that for the words of type (II). Then, $^0_0\gamma_{G_i}(t) = t\cdot {^0_0\alpha_{G_i}(t)} + t^2 \cdot {^0_0\beta_{G_i}(t)}$. We have that $$^0_0\alpha_{G_i}(t) = \frac{e^T\, M_i\, w}{\det(I - t M_i)},$$ with $e = w = (0, \dots, 0, \underbrace{1}_{k_i}, 0, \dots, 0)^T$.
Analogously, $$^0_0\beta_{G_i}(t) = \frac{e^T\, M_i\, w}{\det(I - t M_i)},$$ with $e = (0, \dots, 0, \underbrace{1}_{k_i}, 0, \dots, 0)^T$, and $w$ having a “1” at position $l_i$ for all states $q_{l_i}$ as described above, and zeroes at all other places.
By using $\mathrm{Monty}$ we obtain
&\^0\_0\_[G\_1]{}(t) = [\^0\_0\_[G\_2]{}(t)]{} = t(1 - 5t - 94t\^2 - 374t\^3 - 456t\^4 + 955t\^5 + 4275t\^6 + 5652t\^7 - 1617t\^8&\
& - 16773t\^9 - 24255t\^[10]{} - 7337t\^[11]{} + 26583t\^[12]{} + 45100t\^[13]{} + 26181t\^[14]{} - 12789t\^[15]{} - 34553t\^[16]{}&\
& - 24957t\^[17]{} - 3147t\^[18]{} + 8130t\^[19]{} + 6288t\^[20]{} + 1398t\^[21]{} - 458t\^[22]{} - 284t\^[23]{} - 24t\^[24]{})(1 + 3t&\
& + 2t\^2 - 3t\^3 - 9t\^4 - 8t\^5 + 4t\^7 + 3t\^8 - t\^9)\^[-1]{} (1 - 8t - 85t\^2 - 243t\^3 - 222t\^4 + 332t\^5 + 1194t\^6&\
& + 1349t\^7 + 132t\^8 - 1510t\^9 - 2008t\^[10]{} - 1088t\^[11]{} + 28t\^[12]{} + 359t\^[13]{} + 170t\^[14]{} + 15t\^[15]{})\^[-1]{}.&
Thus, the numbers of special geodesics in $G_1$ and $G_2$ coincide.
RACGs with different geodesic growth {#s:racg-diff-growth}
====================================
In this section we give a proof of Theorem \[thm2\]. Our construction will be analogous to that in Section \[s:racg\], although we shall use different trees, $S_1$ and $S_2$, in order to construct the families $\mathcal{S}^i_n = \{ \tau \cdot S_i | \tau \in \Upsilon_k \}$, $k = n-9$. Namely, we will use the rooted trees in Fig. \[fig:trees2\].
(0)[1]{} (0)[2]{} (0)(1) (0)(2) (1)[3]{} (2)[4]{} (2)[5]{} (1)(3) (2)(4) (2)(5) (3)[6]{} (5)[7]{} (7)[8]{} (7)[9]{} (3)(6) (5)(7) (7)(8) (7)(9)
(0)[1]{} (0)[2]{} (0)(1) (0)(2) (1)[3]{} (1)[4]{} (2)[5]{} (2)[6]{} (1)(3) (1)(4) (2)(5) (2)(6) (3)[7]{} (7)[8]{} (8)[9]{} (3)(7) (7)(8) (8)(9)
We set $\Gamma_1 = \tau \cdot S_1$ and $\Gamma_2 = \tau \cdot S_2$, where $\tau$ is an arbitrary tree with $n$ vertices labelled $\{0,\dots,n-1\}$. Suppose that $0$ represents $\tau$’s root. Then the following holds:
\[thm:Godsil\] The trees $\Gamma_1$ and $\Gamma_2$ are not isomorphic, but are co-spectral and have co-spectral complements. Their line graphs $L(\Gamma_i)$, $L(\overline{\Gamma_i})$ and $\overline{L(\Gamma_i)}$ are not necessarily co-spectral.
The trees $\Gamma_1$ and $\Gamma_2$ are not isomorphic, since $\Gamma_1$ has the rooted tree $\sigma$ depicted in Fig. \[fig:trees3\] as a limb more times than $\Gamma_2$ does.
(0)[2]{} (0)(2) (2)[4]{} (2)[5]{} (2)(4) (2)(5) (5)[7]{} (7)[8]{} (7)[9]{} (5)(7) (7)(8) (7)(9)
The characteristic polynomials of $\tau$, $S_i$ and $\Gamma_i$ ([*i.e.*]{} the characteristic polynomials of the adjacency matrices of these graphs) are related by $$\phi_{\Gamma_i}(t) = \phi_{\tau}(t) \cdot \phi_{S_i\setminus \{0\}}(t) + \phi_{\tau\setminus \{0\}}(t) \cdot \phi_{S_i}(t) - t\cdot \phi_{\tau\setminus \{0\}}(t) \cdot \phi_{S_i\setminus \{0\}}(t),$$ according to [@McKay Lemma 2.2 (i)]. Given that the trees $S_1$ and $S_2$ are isomorphic as graphs (though not as rooted trees), we obtain that $\Gamma_1$ and $\Gamma_2$ share the same characteristic polynomial ([*i.e.*]{} are co-spectral).
The fact that $\overline{\Gamma_1}$ and $\overline{\Gamma_2}$ are co-spectral follows from the above and [@McKay Theorem 3.1 (i)].
Now, by letting $\tau$ be the line graph on two vertices we obtain that $L(\Gamma_1)$, $L(\overline{\Gamma_1})$ and $\overline{L(\Gamma_1)}$ have different characteristic polynomials from $L(\Gamma_2)$, $L(\overline{\Gamma_2})$ and $\overline{L(\Gamma_2)}$, respectively.
\[lem3\] Let $\tau$ be a rooted tree with at least two vertices, and let $\Gamma_1 = \tau \cdot S_1$ and $\Gamma_2 = \tau \cdot S_2$. The RACGs $G(\Gamma_1)$ and $G(\Gamma_2)$ have distinct geodesic growth series.
We will prove that $f_{Geo(\Gamma_1)}(10) \neq f_{Geo(\Gamma_2)}(10)$. We use the same argument as in the proof of Lemma \[equal\_growth\], that is, $f_{Geo(\Gamma_1)}(r)=f_{Geo(\Gamma_2)}(r)$ if and only if $\overline{f_{Geo(\Gamma_1)}}(r)=\overline{f_{Geo(\Gamma_2)}}(r)$, where $\overline{f_{Geo(\Gamma_i)}}(r)$ denotes the numbers of non-geodesics of length $r$, $r \geq 1$. In this proof we use the result of our explicit computations with Monty, which shows that $f_{Geo_0(\Gamma_1)}(r)= f_{Geo_0(\Gamma_2)}(r)$ for $r<8$, but $f_{Geo_0(\Gamma_1)}(8) = f_{Geo^0(\Gamma_1)}(8) = 8919523$ for $G_1$ and $f_{Geo_0(\Gamma_2)}(8) = f_{Geo^0(\Gamma_2)}(8) = 8919522$ for $G_2$.
Any word in $\Gamma_1^*$ has the form $w_1u_1w_2 \dots u_n$, where $u_i \in (\tau\setminus \{0\})^*$, $w_i \in S_1^*$, and $w_i, u_i$ non-empty except for perhaps $w_1$ and $u_n$. A non-geodesic word $w$ in $\Gamma_1^*$ belongs to one of the following sets (or is of the type), depending on its form:
- it either contains non-geodesics $w_i \in (\tau \setminus \{0\})^*$ or $u_j \in S_1^* $, or both, for some $1\leq i,j \leq n$, or
- all $w_i$ and $u_i$ are geodesic on their respective alphabets, and there exists $1\leq j <n$ such that $0 u_j 0$ is a subword of $w$ and $u_j \in (\operatorname{St}_{\Gamma_1}(0) \cap \tau)^*$, or
- all $w_i$ and $u_i$ are geodesic on their respective alphabets, and $w$ contains a subword of the form $s 0 s$, where $s \in (\operatorname{St}_{\Gamma_1}(0) \cap \tau)^*$.
For the remaining discussion we only consider words $w$ of length $10$, and call the number of non-empty subwords $u_i$ or $w_i$ the *syllable length* of $w$. The number of words of type (A) depends only on the geodesic growth series of $\tau$ and $S_1$, and thus it will be equal to the number of words of type (A) in $\Gamma_2$. Notice that if the syllable length of $w$ is $\leq 2$ there are no words of type (B) or (C), so we obtain the same numbers of words in $\Gamma_1$ and $\Gamma_2$. If the syllable length of $w$ is $>4$ then all $w_i, u_i$ are shorter than or equal to $6$, and our computations show that the numbers of special geodesics of length less than or equal to $6$ in $\Gamma_1$ and $\Gamma_2$ coincide. Thus, a discrepancy may appear only when the syllable length is $3$ or $4$. The words of type (B) or (C) with syllable length $4$ have the form $w= w_1 u_1 w_2 u_2$ (or $w= u_1 w_2 u_2 w_3$), where $w_1$ ends in $0$ and $w_2$ starts with $0$, or $w_2=0$. But in this case $|w_1|, |w_2| \leq 7$, and our computations show that $f_{Geo_0(\Gamma_1)}(r)=f_{Geo_0(\Gamma_2)}(r)$, up to $r=7$.
Next, we consider the non-geodesics $w$ of syllable length $3$. If they have the form $w=u_1 w_1 u_2$, then $w_1=0$, and the numbers of geodesics $u_i$ is the same since they are all written over the same alphabet determined by the tree $\tau$, so no discrepancy in the numbers of non-geodesics occurs. Thus, it remains to count the words of the form $w=w_1 u_1 w_2$, where $|w_1| \leq 8$, $w_1$ ends in $0$, $|u_1| \geq 1$, $u_1 \in (\operatorname{St}_{\Gamma_1}(0) \cap \tau)^*$, and $w_2$ starts with $0$, $|w_2| \leq 8$. Again, if $|w_i|\leq 7$ we obtain the same numbers of non-geodesics. However, a discrepancy occurs when $|w_1|=8$ or $|w_2|=8$.
Indeed, the number of such words is $2(f_{Geo_0(\Gamma_1)}(8)\deg_{\tau}(0))$ in $\Gamma_1^*$ and $2(f_{Geo_0(\Gamma_2)}(8)\deg_{\tau}(0))$ in $\Gamma_2^*$. By using Monty, we obtained $f_{Geo_0(\Gamma_1)}(8) = f_{Geo^0(\Gamma_1)}(8) = 8919523$ for $G_1$, but $f_{Geo_0(\Gamma_2)}(8) = f_{Geo^0(\Gamma_2)}(8) = 8919522$ for $G_2$, so the numbers of non-geodesics in these two groups are distinct.
Now we can finish the proof of our second main result.
[*(of Theorem \[thm2\])*]{} This is a straightforward consequence of Theorem \[thm:Godsil\] and Lemma \[lem3\].
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors gratefully acknowledge the support received from the Swiss National Science Foundation: L.C. was supported by PP00P2-144681 (SNSF Professorship); A.K. was supported by P300P2-151316 (Advanced Post-Doc Mobility) and P300P2-151316/2 (CH-Link). The authors are also thankful to Prof. Ruth Kellerhals (University of Fribourg, Switzerland), Prof. Michelle Bucher (University of Geneva, Switzerland), Prof. Robert Young (Courant Institute of Mathematical Sciences, New York, USA) and Dr. Alexey Talambutsa (Steklov Mathematical Institute, Moscow, Russia) for fruitful discussions.
Appendix {#appendix .unnumbered}
========
In this section we give the Python code “Monty” that we used in our computations, with comments and remarks. Its on-line copy [@Pyth] can be downloaded as a SAGE worksheet from the second author’s [web-page.](www.sashakolpakov.wordpress.com/list-of-papers)
We begin by defining the automaton $A$ that recognizes the language of geodesics of a given RACG $G$ whose defining graph $\Gamma$ is given as input.
def Automaton(t):
# takes a triangle-free graph, returns the automaton recognising
# the resp. RACG as a digraph
CliqueComplex = t.clique_complex(); C = list();
for s in CliqueComplex.faces().values():
for f in s: C.append(set(f));
n = len(C); a = DiGraph(); a.add_vertices(range(n));
for i in range(n):
for v in t.vertices():
if not(C[i].issuperset([v])):
st = set(t.vertex_boundary([v])).union([v]);
d = set([v]).union(st.intersection(C[i]));
k = C.index(d);
a.add_edge((i,k));
a.set_vertices({i : C[i] for i in range(n)})
return a;
Given the automaton $A$, we then compute the growth function of its accepted language (in this case, the geodesic language for $G$ with respect to $S$).
def GrowthFunc(a):
# takes a geodesic automaton, returns the resp. (geodesic) growth function
am = a.adjacency_matrix();
n = am.nrows();
R = FractionField(PolynomialRing(QQ, 't'));
R.inject_variables();
m = diagonal_matrix([1]*n) - t*am;
M = (1/m.det())*m.adjoint();
ee = [0]*(n-1);
ee.append(1);
e = vector(ee);
w = vector([1]*n);
func = e*M*w;
return func.numerator().factor()/func.denominator().factor();
A clique $c$ in the defining graph $\Gamma$ of the RACG $G = G(\Gamma)$ corresponds to a state $q_c$ in the automaton $A$. We need the following auxiliary function in order to determine the index of $q_c$ represented as a vertex of the digraph $A$ (the automaton) created by the procedure `Automaton`.
def Index(a, c):
# takes a geodesic automaton `a', a clique `c' in the resp. defining graph,
# returns the vertex of the automaton `a' corresponding to `c'
l = None;
for v in a.vertices():
if a.get_vertex(v) == c:
l = v;
return l;
Below we compute the growth function ${_0}\alpha(t)_G$ as described in the proof of Theorem \[thm1\].
def GrowthFuncStart0(a):
# takes a geodesic automaton, returns the growth function for geodesic words
# starting at the state q, where $\delta(Start, 0) = q$
am = a.adjacency_matrix();
n = am.nrows();
R = FractionField(PolynomialRing(QQ, 't'));
R.inject_variables();
m = diagonal_matrix([1]*n) - t*am;
M = (1/m.det())*m.adjoint();
ind = Index(a, set([0]));
ee = [0]*n;
ee[ind] = 1;
e = vector(ee);
w = vector([1]*n);
func = e*M*w;
return func.numerator().factor()/func.denominator().factor();
Now we define a function that takes as input the geodesic automaton $A$ for a RACG $G = G(\Gamma)$, a list of cliques $l = [c_0, c_1, \dots, c_k]$ in the respective defining graph $\Gamma$ and returns the growth function for geodesic words starting at the state $q = \delta(q_0, 0)$ that bring $A$ to any of the states described by the cliques in $l$.
def GrowthFuncStart0End(a, l):
# takes an automaton and a list of cliques `l' in the resp. defining graph
# as input, returns the growth function for geodesic words starting at
# state q, where $\delta(Start, 0) = q$, and ending at any of the states
# corresponding to cliques in `l'
am = a.adjacency_matrix();
n = am.nrows();
R = FractionField(PolynomialRing(QQ, 't'));
R.inject_variables();
m = diagonal_matrix([1]*n) - t*am;
M = (1/m.det())*m.adjoint();
ee = [0]*n;
ind = Index(a, set([0]));
ee[ind] = 1;
e = vector(ee);
ww = [0]*n;
for c in l:
ind = Index(a, set(c));
ww[ind] = 1;
w = vector(ww);
func = e*M*w;
return func.numerator().factor()/func.denominator().factor();
By using a suitable list of cliques $l$ we can compute the functions ${_0^0}\alpha(t)_G$ and ${_0^0\beta(t)}_G$. Namely, in the proof of Theorem \[thm1\], we find
# the function ${_0^0}\alpha(t)_{G_1}$
a001 = GrowthFuncStart0End(a1, [[0]]);
# the function ${_0^0}\alpha(t)_{G_2}$
a002 = GrowthFuncStart0End(a2, [[0]]);
# the function ${_0^0}\beta(t)_{G_1}$
b001 = GrowthFuncStart0End(a1, [[1], [1,3], [1,4], [2], [2,5]]);
# the function ${_0^0}\beta(t)_{G_2}$
b002 = GrowthFuncStart0End(a2, [[1], [1,3], [2], [2,4], [2,5]]);
The list `l = [[1], [1,3], [1,4], [2], [2,5]]` above contains the cliques of $\Gamma_1$ corresponding to the accept states $q$ of the geodesic automaton $A_1$ for $G_1 = G(\Gamma_1)$ such that $\delta(p, 0) = q$, for a state $p$. The list `l = [[1], [1,3], [2], [2,4], [2,5]]` contains the cliques of $\Gamma_2$ with analogous properties, corresponding to the states of the geodesic automaton $A_2$ for the RACG $G_2 = G(\Gamma_2)$.
The above described Python procedures are also used to perform the computations in the proof of Theorem \[thm2\].
The on-line version of Monty [@Pyth] contains a variation of the `GrowthFunc` procedure, called `GrowthFuncBM`, that uses the Berlekamp-Massey algorithm for faster computing.
[99]{}
Y. Antolín and L. Ciobanu, , European J. Combin. [**34**]{} (2013), no. 5, 859–874.
P. Bahls, , Imperial College Press, London, 2005. xiv+176 pp.
A. Bjorner and F. Brenti, , Graduate Texts in Mathematics, 231. Springer, New York, 2005. xiv+363 pp.
R. Charney, , Math. Ann. [**301**]{} (1995), no. 2, 307–324.
M. Davis, , Princeton Univ. Press, (2008), 1–594.
D.B.A. Epstein, J.W. Cannon, D.F. Holt, S. Levy, M.S. Paterson, and W.P. Thurston, , Jones and Bartlett, 1992.
D.B.A. Epstein, A. R. Iano-Fletcher, and U. Zwick, , Experimental Mathematics [**5**]{}, 4 (1996), 297–315.
C. Godsil, , available on-line at http://www.math.uwaterloo.ca/ $\sim$cgodsil/pdfs/cospectral.pdf
D.F. Holt and S. Rees, , , [**104**]{} (2012) 486–512.
J. Hopcroft, and J.D. Ullman, Introduction to automata theory, languages, and computation, Addison-Wesley Series in Computer Science, Addison-Wesley Publishing Co., Reading, Mass., 1979.
A. Kolpakov, and L. Ciobanu, , available on-line at [www.sashakolpakov.wordpress.com/list-of-papers/](www.sashakolpakov.wordpress.com/list-of-papers/)
J. Loeffler, J. Meier, and J. Worthington, , Internat. J. Algebra Comput. [**12**]{} (2002), 6, 747–754.
B. McKay, , , [**3**]{} (1977) 219–232.
D. G. Radcliffe, , , [**3**]{} (2003) 1079–1088.
*Laura Ciobanu\
Department of Mathematics\
University of Neuchâtel\
Rue Emile - Argand 11\
CH-2000 Neuchâtel, Switzerland\
*
*e-mail*[:]{}`[email protected]`\
*Alexander Kolpakov\
Department of Mathematics\
University of Toronto\
40 St. George Street\
M5S 2E4 Toronto ON, Canada\
*
*e-mail*[:]{}`[email protected]`\
=6.2in =8.3in =9000 =1.2mm
[^1]: this is not the given name of the code, which would be obviously too posh for such a petty thing, but a reference name, which is seemingly good for any Python code.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Compressive sensing is considered a huge breakthrough in signal acquisition. It allows recording an image consisting of $N^2$ pixels using much fewer than $N^2$ measurements if it can be transformed to a basis where most pixels take on negligibly small values. Standard compressive sensing techniques suffer from the computational overhead needed to reconstruct an image with typical computation times between hours and days and are thus not optimal for applications in physics and spectroscopy. We demonstrate an adaptive compressive sampling technique that performs measurements directly in a sparse basis. It needs much fewer than $N^2$ measurements without any computational overhead, so the result is available instantly.'
author:
- 'Marc Aßmann$^{\ast}$'
- Manfred Bayer
title: Compressive adaptive computational ghost imaging
---
Computational ghost imaging (CGI) is a novel imaging technique that has received significant attention during the last few years[@Erkmen2010]. It is a consequent further development of conventional ghost imaging [@Pittman1995; @Ferri2010] which allows to record spatially resolved images using a detector without spatial resolution. In conventional ghost imaging the image is recorded using two spatially correlated light beams, one object and one reference beam. The object beam illuminates the object to be imaged and is then collected using a bucket detector. The reference beam never interacts with the object and is recorded using a pixelated device offering spatial resolution. As both beams are spatially correlated the coincidence count signal allows one to retrieve a ghost image of the object. Ghost imaging using both entangled photons [@Pittman1995] or classical light [@Bennink2002; @Valencia2005; @Cheng2004] as the spatially correlated twin beam source has been demonstrated. A seminal paper by Shapiro [@Shapiro2008] clarified that the sole purpose of the reference beam lies in determining the illumination pattern at the object position at each instant, while the object beam gives data about the transmission of this pattern through the object. Therefore, if one can create a deterministic illumination pattern at the object position, the reference beam becomes obsolete and CGI using just a single beam and a single pixel detector [@Duarte2008] becomes possible. This approach has been realized using deterministic speckle patterns created using a spatial light modulator (SLM) [@Bromberg2009]. It has also been demonstrated that this technique also offers the possibility to perform compressive sensing [@Katz2009; @Zerom2011], that is it is possible to reconstruct an image consisting of $N^2$ pixels using much less than $N^2$ measurements by utilizing the fact that natural images are typically sparse[@Candes2008]: When transformed to an appropriate basis, they contain many coefficients that are zero or close to it. In practice, the transmission measured for each speckle pattern constitutes a projection of the object image and compressive sensing is performed by utilizing an algorithm which checks all the possible images which are consistent with the projections performed and finds the image which is the sparsest one. Usually the $L1$-norm serves as a measure of sparsity and the image which minimizes it, is the optimal reconstruction of the object. However, this method still has some drawbacks. The time taken by the reconstruction algorithm can become very long for large images and one needs to compute the speckle pattern at the object position by using the Fresnel-Huygens propagator on the phase pattern imprinted on the SLM. While the latter is not a big problem - one can calculate the speckle pattern once and reuse the phase pattern masks - the computational overhead, given by the computational effort once all measurements have been made, is a huge problem. The overhead becomes especially problematic considering typical problems in spectroscopy (e.g. pump-probe spectroscopy), where many similar images need to be taken, while one experimental parameter is changed. Here, it is desirable to have the reconstructed image directly, so one can use this information when taking the next image. For example one could adaptively scan the previous image for regions of large values or strongly varying values and scan these areas with higher resolution in the next image.
results {#results .unnumbered}
=======
The adaptive compressive CGI algorithm {#the-adaptive-compressive-cgi-algorithm .unnumbered}
--------------------------------------
We demonstrate an alternative way to perform compressive CGI (CCGI) without any computational overhead once all measurements have been performed by using an adaptive measurement scheme. We follow a promising strategy for adaptive compressive sensing that suggests replacing the random speckle patterns by directly using the patterns that form the sparse basis [@Averbuch2012]. We start the discussion of our strategy by recalling the properties of the 2D Haar wavelet transform of square images consisting of $N\times N$ pixels. The wavelet decomposition procedure is schematically depicted in figure \[Monkey\]. The decomposition of the image $I(x,y)$ is performed seperately for rows and columns. At first each row is divided into $\frac{N}{2}$ pairs of adjacent pixels. The partial wavelet transform $T'(x,y)$ now consists of the sum and the difference of these adjacent pixels according to the following rules for $x<\frac{N}{2}$: $$\begin{aligned}
T'(x,y) & = & I(2x,y)+I(2x+1,y) \\
T'(x+\frac{N}{2},y) & = & I(2x,y)-I(2x+1,y).\end{aligned}$$ Repeating that procedure for each column in $T'$ according to similar rules for $y<\frac{N}{2}$ gives the full transform $T(x,y)$: $$\begin{aligned}
T(x,y) & = & T'(x,2y)+T'(x,2y+1) \\
T(x,y+\frac{N}{2}) & = & T'(x,2y)-T'(x,2y+1).\end{aligned}$$ The resulting transform now consists of four quadrants. The upper left quadrant represents a coarse version of the original image, while the other three quadrants contain information about horizontal, vertical and diagonal edges. One may now continue and perform another wavelet transform on the upper left quadrant and iteratively repeat this procedure until the pixel in the upper left corner contains the mean intensity of the picture and all other pixels contain information about edges. Now each additional transform performed corresponds to a coarser scale $j$ with wavelet coefficients spanning over larger regions, but carrying information over a smaller range of frequencies. Such wavelet representations are efficient in terms of image compression. Only a small portion of natural images consists of edges and only wavelet coefficients corresponding to regions with sharp edges are large, therefore only few large coefficients are sufficient to approximate the full image. As can be seen in figure \[Monkey\], the number of large wavelet coefficients (shown in white) is rather small.
![512$\times$512 pixel baboon test image (left panel) and its one-step (middle panel) and complete (right panel) wavelet transform. For the transform absolute values of the wavelet coefficients are shown. White regions correspond to large wavelet values and mark regions with strong edges. Every wavelet coefficient at scale $j$ contains information about four pixels of the coarse image of size $\frac{N}{j}\times\frac{N}{j}$. Also, every wavelet coefficient has four children wavelet coefficients at scale $j-1$ which contain information about the same range of the image.[]{data-label="Monkey"}](Monkeyfig1.eps){width="0.95\linewidth"}
This strategy becomes interesting as the wavelet transformation is hierarchic. Every parent coefficient at some coarse scale has four children coefficients at the next finer scale covering the same spatial region. As it is very likely that the children wavelet coefficients belonging to parent coefficients which are small will also be small, this offers a great opportunity for image compression in terms of wavelet trees [@Shapiro1993] by cutting of these trees at an adequate scale. We follow a similar strategy and first take a coarse image of size $\frac{N}{j}\times \frac{N}{j}$. Experimentally, this is realized by inserting a phase-only SLM (Holoeye-Pluto) in the path of a laser beam polarized such that the SLM only introduces a phase shift to it. The phase pattern imprinted on the SLM is the Fourier transform of a square superposed with the phase map of a lens. As a consequence, in the focal plane behind the SLM the square is recovered in the spatial intensity pattern of the light beam. We precomputed 87040 of such phase patterns using an iterative numerical technique based on the adaptive-additive algorithm [@Dufresne2001]. 65536 of these form the pixels of a 256$\times$256 ($j$=1) pixel square. The other patterns form the pixels of squares of the same size, but consisting of fewer (128$\times$128 ($j$=2), 64$\times$64 ($j$=3) and 32$\times$32 ($j$=4)), but larger pixels of size $2^{(2(j-1))}$, respectively. The object to be imaged is placed at the focal plane of the SLM ($f$=36cm) and the transmission through that object is measured. Under the conditions used throughout the manuscript, the whole square has a side length of 32mm. When the coarse image is taken, we perform a one-step wavelet transform on it. Now we check the absolute values of the wavelet coefficients corresponding to edges against a predefined threshold $I_j$. If the values are larger than $I_j$, the four children wavelet values at the next finer scale $j$-1 are measured too. As each wavelet coefficient spans over exactly four pixels at its scale, it is never necessary to perform more than four measurements in real space to determine any wavelet value. Once all the measurements at the finer scale have been performed, a new finer image can be constructed. It consists of the newly measured transmission values for regions containing sharp edges and of the transmission values already measured at a coarser scale for regions without edges. Now another one-step wavelet transform is performed on this finer image and again all wavelet values are checked against a new threshold $I_{j-1}$. This process is repeated until the finest picture at scale $j$=1 is constructed. A summary of the imaging process is presented in fig. \[example\].
![Summary of the CCGI scheme: First, a low resolution real space image is taken (upper right panel). The wavelet transform of that image is created (middle left panel). Large wavelet coefficients are shown in white, small ones in black. For each wavelet coefficient larger than the chosen threshold, its four children coefficients are determined. See the upper left panel for exemplaric parent (capital letters) and corresponding children wavelets (corresponding lower case letters) across different scales. The measurement of a children wavelet coefficient requires four real space measurements at a finer scale. After all target wavelet coefficients at the finer scale are measured (middle right panel), the procedure continues with the next finer scale until the target scale $j$=1 is reached or no wavelet coefficient is larger than the threshold value (lower left panel). The result is then converted back to a real space image using the inverse wavelet transform (lower right image). For this example the number of measurements needed is roughly 40$\%$ of the number of pixels present in the image. Note that the upper right, lower left and lower right sector of the wavelet transform correspond to horizontal, vertical and diagonal edges, respectively. Wavelet coefficients have been multiplied by 8 to enhance contrast.[]{data-label="example"}](example2.eps){width="0.6\linewidth"}
Experimental results {#experimental-results .unnumbered}
--------------------
We tested the CCGI algorithm using a metal plate containing twelve holes as a test target. We chose to use a threshold which becomes sharper at finer scales ($I_{j-1}$=2$I_j$) and varied the initial threshold $I_4$, resulting in images of differing quality. The results are shown in figure \[Lochblende\]. Here the transmission maps quantized to 256 greyscales are shown in terms of the decreased acquisition rate $\alpha$, which is the total number of measurements performed on all scales divided by the total number of pixels present on the finest scale ($N^2$=65536). The transmission is normalized to the empty space transmission to account for possible inhomogeneities introduced by the SLM.
![Normalized transmission maps through a metal plate containing twelve holes. The large holes have a diameter of 2mm, while the smaller ones have a diameter of 1mm. $\alpha$ gives the decreased acquisition rate. A faithful image of the plate is already possible with approximately 5-7$\%$ of the measurements required to record every single pixel in full resolution.[]{data-label="Lochblende"}](Lochblendefig2.eps){width="0.95\linewidth"}
As can be seen, the image is reproduced quite well at relatively small $\alpha$. At around 2$\%$ the quality is already sufficient for distinguishing the holes and counting their number. For $\alpha$ around 4$\%$ the image already looks reasonable. For $\alpha$ around 7$\%$ the recorded image shows good quality. For larger $\alpha$ only small improvements are seen. However, to get a more quantitative measure of the recorded image quality, we calculated the mean squared error $$\sigma^2=\frac{1}{N^2}\sum_{i,j} [T(x,y)-R(x,y)]^2$$ of the measured image as compared to the reference construction drawing $R(x,y)$ of the metal plate containing the holes.
![Mean squared error versus the number of measurements taken. The residual error saturates for $\alpha>$7$\%$.[]{data-label="MSE"}](MSEfig3.eps){width="0.95\linewidth"}
The impression that the image quality does not improve significantly for $\alpha >$7$\%$ is verified. The mean squared error roughly follows an exponential decay and saturates approximately at a value of 0.055 for large $\alpha$. A closer examination of this residual error shows that it is mainly caused by the edges of the holes. In contrast to the construction drawing, the edges between full transmission and zero transmission are not positioned at pixel borders. Therefore the pixels at the edges show some intermediate transmission and introduce some deviations from the reference image.
![Normalized transmission map through a 1951 USAF resolution test chart at $\alpha$=0.369.[]{data-label="USAF"}](USAFPlatefig4.eps){width="0.95\linewidth"}
The number of necessary measurements needed for near optimal reconstruction of an image obviously depends on the number of large wavelet coefficients that image carries. In order to demonstrate the adaptivity of our technique, we kept the threshold setting used for measuring the metal plate at $\alpha$=0.131 in figure \[Lochblende\] and imaged a more complex object - a 1951 USAF resolution test chart. The recorded image is shown in figure \[USAF\]. The image quality is still good, but the algorithm automatically took almost three times more measurements than were needed for the metal plate, resulting in $\alpha$=36.9$\%$. The image resolution is reasonable. One pixel has a length of about 125$\mu$m in real space which is roughly the size of the lines on the test plate which can still be resolved. Nevertheless the image shows some artifacts. These are a consequence of a weakness of the algorithm used. Strictly periodic structures like several parallel lines placed next to each other may look like having no edge at all at a coarser scale. However, such problems may be overcome by more advanced algorithms relying on taking more than just the parent wavelet value into account when deciding on whether a certain wavelet value should be measured or not [@Averbuch2012].\
Our technique offers a wide range of advantages. As it is adaptive, one has control over the quality of the image in advance by choosing the thresholds. The algorithm does not require any additional computationally intensive recovery algorithms needed for standard compressive sensing techniques using pseudorandom illumination patterns. Our technique works reasonable fast. The SLM can be operated at up to 60Hz. Photodiodes and readout circuitry working on the same timescale are common today, opening up the possibility to record images of reasonable resolution and quality within few minutes. In particular experimental techniques requiring single pixel detectors like lock-in detection for sensitive pump-probe measurements may benefit from our results. Spatially resolved measurements are to the best of our knowledge not carried out with high resolution using such techniques due to the long measurement durations that would be needed. Reducing the number of necessary measurements by a factor of at least fifteen opens up the way to perform such measurements with high spatial resolution. If the duration of the measurement is more crucial than the image quality, our approach also allows one to perform a fixed number of finer measurements for a preset number of largest wavelet coefficients at each scale instead of using thresholds. In that way it is possible to take an image in a fixed amount of time.
Discussion {#discussion .unnumbered}
==========
In conclusion we have developed an adaptive CCGI technique that allows us to record images using a single pixel camera at an acquisition rate fifteen to twenty times below the Shannon limit by recording the image directly in a sparse basis. A number of further research directions arise from our work. Compressive imaging techniques are not limited to recording image information, but have also found usage far beyond simple imaging applications in fields like quantum process tomography [@Shabani2011; @Liu2012], optical encryption [@Clemente2010], fluorescence microscopy and hyperspectral imaging [@Studer2012]. Also our approach is quite generic. Optimized approaches which also take the magnitude of neighboring wavelet coefficients into account [@Averbuch2012] instead of just the parent wavelet coefficient may lead to increased image quality or smaller values of $\alpha$. Also, it is well known that especially designed measurement matrices can drastically reduce the number of needed measurements for exact image reconstruction using seeded belief propagation techniques[@Krzakala2012]. Finally, it is not strictly necessary to use precomputed phase patterns, but one could compute them on the fly, thereby allowing one to even choose an adaptive wavelet basis. Yet, the greatest strength of our approach lies in drastically reducing the needed measurement time for high-resolution images using single-pixel detectors without having any need for computational image reconstruction.\
We would like to conclude this paper by a comparison between CCGI and standard random Gaussian matrix based compressive sensing techniques (RMCS) to identify the strengths and weaknesses of our approach in more detail. Obviously, having the image available once all the measurements are taken, is an advantage, but it is also introduces a drawback: CCGI needs to use a predefined sparse basis, while RMCS will automatically find an adequate sparse basis during reconstruction. Accordingly, it is typically possible to achieve a near-ideal reconstruction with fewer measurements using RMCS. However, it should be noted that the exact number of measurements needed for near-ideal reconstruction is usually not known a priori as it depends on the sparsity of the image. The needed number of measurements is also unknown in CCGI, but specifying the desired image quality in terms of the threshold $I_j$ ensures that not too many measurements are made. The exact number of measurements needed for CCGI is hard to predict as it depends on how sparse an image is in the wavelength basis. For very sparse images, the penalty can be as much as 50%. For less sparse images, the differences are less drastic. However, for a comparable number of measurements, CCGI-based methods tend to achieve a better signal to noise ration than RMCS methods do. See [@Averbuch2012] for a detailed comparison of a slightly modified version of CCGI with state of the art RMCS techniques. Another important benchmark is the performance of compressive sensing techniques in the presence of noise. In CCGI noise can become a severe problem if the noise magnitude becomes comparable to the threshold chosen. CCGI is therefore not the method of choice for measuring images containing strong noise or weak signals. Another issue is scalabilty. Going to larger images, increases the necessary number of measurements a computations during data acquisition in CCGI and the complexity of the minimization problem in RMCS. However, for all the image sizes we examined, the time needed for performing the measurements was always so much longer than the time needed for performing the wavelet transforms and building the sampling queue that no delay was noticeable. In summary, although other compressive sensing methods based on Gaussian random matrices approaches typically need fewer measurements than most (but not all [@Monajemi2012]) techniques using deterministic matrices, having the result immediately renders adaptive spatially resolved pump-probe spectroscopy and other delicate spectroscopic techniques with high resolution possible. Therefore, we suggest that our technique is well suited for specialized complex problems in physics and spectroscopy which are a priori known to be reasonably sparse in the wavelet basis, while RMCS methods are a much better choice for taking single images, for images where noise is an issue and for images where the sparsest basis is unknown.
Methods {#methods .unnumbered}
=======
The objects to be imaged were placed at the focal plane of the SLM ($f$=36cm) and the transmission through the objects was measured by a standard commercial photo diode onto which the transmission through the object was focused and a Keithley 2000 multimeter was used for measuring the photo diode output. The hole test plate used consisted of twelve holes. Six of these holes had a diameter of 2mm, while the other six holes had a diameter of 1mm. Their average separation was around 1.5mm. The laser used for the transmission measurements was a pulsed Ti:Sapph laser emitting pulses wih a duration of approximately 2ps at a wavelength of 800nm.
[10]{} url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
& . ** ****, ().
, , & . ** ****, ().
, , & . ** ****, ().
, & . ** ****, ().
, , & . ** ****, ().
& . ** ****, ().
. ** ****, ().
*et al.* . ** ****, ().
, & . ** ****, ().
, & . ** ****, ().
, , & . ** ****, ().
& . ** ****, ().
, & . ** ****, ().
. ** ****, ().
, , , & . ** ****, ().
*et al.* . ** ****, ().
, , , & . ** ****, ().
, , , & . ** ****, ().
*et al.* . ** ****, ().
, , , & . ** ****, ().
, , , & . ** ****, ().
Author Contributions {#author-contributions .unnumbered}
====================
M.A. designed the experiment and analyzed the data. M.A. and M.B. wrote the manuscript.
Competing financial interests {#competing-financial-interests .unnumbered}
=============================
The authors declare no competing financial interests.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A comprehensive universal description of the rotational-vibrational spectrum for two identical particles of mass $m$ and the third particle of mass $m_1$ in the zero-range limit of the interaction between different particles is given for arbitrary values of the mass ratio $m/m_1$ and the total angular momentum $L$. It is found that the number of vibrational states is determined by the functions $L_c(m/m_1)$ and $L_b(m/m_1)$. Explicitly, if the two-body scattering length is positive, the number of states is finite for $L_c(m/m_1) \le L \le L_b(m/m_1)$, zero for $L > L_b(m/m_1)$, and infinite for $L < L_c(m/m_1)$. If the two-body scattering length is negative, the number of states is zero for $L \ge L_c(m/m_1)$ and infinite for $L < L_c(m/m_1)$. For the finite number of vibrational states, all the binding energies are described by the universal function $\varepsilon_{L N}(m/m_1) = {\cal E}(\xi, \eta)$, where $\xi = \displaystyle\frac{N - 1/2}{\sqrt{L(L + 1)}}$, $\eta = \displaystyle\sqrt{\frac{m}{m_1 L (L + 1)}}$, and $N$ is the vibrational quantum number. This scaling dependence is in agreement with the numerical calculations for $L > 2$ and only slightly deviates from those for $L = 1, 2$. The universal description implies that the critical values $L_c(m/m_1)$ and $L_b(m/m_1)$ increase as $0.401 \sqrt{m/m_1}$ and $0.563 \sqrt{m/m_1}$, respectively, while the number of vibrational states for $L \ge L_c(m/m_1)$ is within the range $N \le N_{max} \approx 1.1 \sqrt{L(L + 1)} + 1/2$.'
author:
- 'O. I. Kartavtsev'
- 'A. V. Malykh'
bibliography:
- 'fermions.bib'
title: 'Universal description of the rotational-vibrational spectrum of three particles with zero-range interactions '
---
Introduction {#Introduction}
============
The universal low-energy few-body dynamics of two-species compounds is of much interest both for atomic and many-body physics. In this respect, the study of the three-body energy spectrum gives insight into the role of triatomic molecules and few-body scattering. The area of applications includes the investigation of multi-component ultra-cold quantum gases, e. g., binary Fermi-Bose [@Ospelkaus06; @Karpiuk05] and Fermi [@Shin06; @Chevy06; @Iskin06] mixtures and of impurities embedded in a quantum gas [@Cucchietti06; @Kalas06], which are presently under thorough experimental and theoretical study. In addition, one should mention the reactions with negative atomic and molecular ions [@Penkov99; @Jensen03].
The universal isotopic dependence of the three-body energy spectrum was multiply discussed [@Efimov73; @Ovchinnikov79; @Li06; @DIncao06; @Shermatov03], nevertheless, the main objective was the description of Efimov’s spectrum. Recently, the infinite number of the $1^+$ bound states was predicted [@Macek06] for three identical fermions with the resonant $p$-wave interaction. Concerning the low-energy scattering, one should mention a two-hump structure in the isotopic dependence of the three-body recombination rate of two-component fermions [@Petrov03; @Petrov05a; @Kartavtsev07] and the two-component model for the three-body recombination near the Feshbach resonance [@Kartavtsev02].
The main aim of the paper is a comprehensive description of the finite three-body rotational-vibrational spectrum in the zero-range limit of the interaction between different particles. Both qualitative and numerical results are obtained by using the solution of hyper-radial equations (HREs) [@Macek68; @Kartavtsev99; @Kartavtsev06]. The detailed study of the bound states and scattering problems for the total angular momentum $L = 1$ was presented in [@Kartavtsev07].
Outline of the approach {#approach}
=======================
Particle 1 of mass $m_1$ and two identical particles 2 and 3 of mass $m$ are described by using the scaled Jacobi variables ${\mathbf x} =
\sqrt{2\mu}\left({\mathbf r}_2 - {\mathbf r}_1\right),\
{\mathbf y} = \sqrt{2\tilde\mu}[{\mathbf r}_3 -
(m_1 {\mathbf r}_1 + m {\mathbf r}_2)/(m_1 + m)]$ and the corresponding hyper-spherical variables $x = \rho\cos\alpha$, $y = \rho\sin\alpha$, $\hat{\mathbf x} = {\mathbf x}/x$, and $\hat{\mathbf y} = {\mathbf y}/y$, where ${\mathbf r}_i$ is the position vector of the $i$th particle and $\mu = m m_1/(m + m_1)$ and $\tilde{\mu} = m (m + m_1)/(m_1 + 2m)$ are the reduced masses. In the universal low-energy limit, only the s-wave interaction between different particles will be taken into account provided the s-wave interaction is forbidden between two identical fermions and is strongly suppressed between two heavy bosons in the states of $L > 0$. The two-body interaction is defined by imposing the boundary condition at the zero inter-particle distance, which depends on a single parameter, e. g., the two-body scattering length $a$ [@Kartavtsev07]. This type of interaction is known in the literature as the zero-range potential [@Demkov88], the Fermi [@Wodkiewicz91] or Fermi-Huang [@Idziaszek06] pseudo-potential, and an equivalent approach is used in the momentum-space representation [@Braaten03]. The units $\hbar = 2\mu = |a| = 1$ are used throughout; thus, the binding energy becomes the universal function depending on the mass ratio $m/m_1$ and the rotational-vibrational quantum numbers $L$ and $N$. In view of the wave-function symmetry under permutation of identical particles, a sum of two interactions between different particles is expressed by a single boundary condition at the zero distance between particles $1$ and $2$, $$\begin{aligned}
\label{bch}
\lim_{\alpha\rightarrow \pi/2}\left[ \frac{\partial }{\partial\alpha} -
\tan\alpha - \rho \frac{a}{|a|} \right]\Psi = 0 \ . \end{aligned}$$
The problem under study is conveniently treated by using the expansion of the properly symmetrized wave function, $$\label{Psi} \Psi = (1 + S\widehat{P})
\frac{ Y_{LM}(\hat{\mathbf y})}{\rho^{5/2}\sin 2\alpha} \sum_{n = 1}^{\infty}
f_n(\rho)\varphi_n^L(\alpha, \rho) \ ,$$ which leads to the hyper-radial equations for the functions $f_n(\rho)$ [@Kartavtsev07]. Here $\widehat{P}$ denotes permutation of the identical particles 2 and 3, $S = 1$ and $S = -1$ if these particles are bosons and fermions, respectively, $Y_{LM}(\hat{\mathbf y})$ is the spherical function. The action of $\widehat{P}$ on the angular variables in the limit $\alpha \to \pi/2$ is given by $\widehat{P}Y_{LM}(\hat{\mathbf y}) \to
(-1)^L Y_{LM}(\hat{\mathbf y})$ and $\widehat{P}\alpha \to \omega$, where $\omega = \arcsin (1+ m_1/m)^{-1}$. The functions $\varphi_n^L(\alpha, \rho)$ in the expansion (\[Psi\]) are the solutions of the equation on a hypersphere (at fixed $\rho$), $$\label{eqonhyp1}
\left[\frac{\partial^2}{\partial \alpha^2} - \frac{L(L + 1)}{\sin^2\alpha}
+ \gamma^2_n(\rho)\right]\varphi_n^L(\alpha,\rho) = 0 \ ,$$ complemented by the boundary conditions $\varphi_n^L(0, \rho) = 0$ and $$\label{bconhyp}
\lim_{\alpha\rightarrow \pi/2}
\left(\frac{\partial}{\partial\alpha} - \rho \frac{a}{|a|} \right)
\varphi_n^L(\alpha, \rho) = S(-)^L\frac{2}{\sin 2\omega}
\varphi_n^L(\omega, \rho) \ ,$$ where a set of discrete eigenvalues $\gamma_n^2(\rho)$ plays the role of the effective channel potentials in a system of the hyper-radial equations [@Kartavtsev07]. The functions satisfying Eq. (\[eqonhyp1\]) and the zero boundary condition are straightforwardly expressed [@Bateman53] via the Legendre function $$\begin{aligned}
\label{varphi}
\varphi_n^L(\alpha, \rho) = \sqrt{\sin\alpha}
Q_{\gamma_n(\rho) - 1/2}^{L + 1/2} (\cos\alpha ) \equiv
\phi_{L, \gamma_n(\rho)}(\alpha)\ .\end{aligned}$$ The functions $ \phi_{L, \gamma}(\alpha)$ are odd functions on both variables $\gamma$ and $\alpha$ satisfying the recurrent relations $\sin\alpha\ \phi_{L + 1, \gamma}(\alpha) = (\gamma - L - 1)\cos\alpha \
\phi_{L, \gamma}(\alpha) - (\gamma + L)\phi_{L, \gamma - 1}(\alpha )$, which follow from those for the Legendre functions. It is convenient to write $\phi_{L, \gamma}(\alpha) =
A_{L, \gamma}(\cot\alpha)\sin\gamma\alpha +
B_{L, \gamma}(\cot\alpha)\cos\gamma\alpha$, where $A_{L, \gamma}(x)$ and $B_{L, \gamma}(x)$ are simple polynomials on $\gamma$ and $x$, which are explicitly given for few lowest $L$ by $A_{0, \gamma}(x) = 1$, $B_{0,\gamma}(x) = 0$, $A_{1, \gamma}(x) = -x$, $B_{1,\gamma}(x) = \gamma$, $A_{2, \gamma}(x) = 1 - \gamma^2 + 3x^2$, $B_{2,\gamma}(x) = -3\gamma x$, $A_{3, \gamma}(x) = 3x(2\gamma^2 - 3 - 5x^2)$, and $B_{3,\gamma}(x) =
\gamma (15x^2 + 4 - \gamma^2)$. Substituting (\[varphi\]) into the boundary condition (\[bconhyp\]) and using the identity $\phi_{L+1, \gamma}(\pi/2) = \frac{\partial
\phi_{L, \gamma}(\alpha)}{\partial\alpha} \Big|_{\alpha=\pi/2}$ one comes to the transcendental equation for $\gamma_n^2(\rho)$, $$\begin{aligned}
\label{transeq}
\rho \frac{a}{|a|}\ \phi_{L, \gamma} (\pi/2) = \phi_{L + 1, \gamma} (\pi/2)
- \frac{2 S (-)^L}{\sin 2\omega}\phi_{L, \gamma}(\omega) \ .\end{aligned}$$
The attractive lowest effective potential determined by $\gamma_1^2(\rho)$ plays the dominant role for the binding-energy and low-energy-scattering calculations, while the effective potentials in the upper channels for $n \ge 2$ contain the repulsive term $\gamma_n^2(\rho)/\rho^2$ and are of minor importance. Thus, a fairly good description will be obtained by using the one-channel approximation for the total wave function (\[Psi\]) where the first-channel radial function satisfies the equation [@Kartavtsev07] $$\label{system1}
\left[\frac{d^2}{d \rho^2} - \frac{\gamma_1^2(\rho) - 1/4}{\rho^2} + E \right]
f_1(\rho) = 0 \ .$$ Note that the diagonal coupling term is omitted in Eq. (\[system1\]), which does not affect the final conclusions and leads to the calculation of a lower bound for the exact three-body energy. Our calculations [@Kartavtsev07] shows that the one-channel approximation provides better than few percent overall accuracy of the binding energy.
The most discussed feature [@Efimov73; @Ovchinnikov79; @Li06; @DIncao06] of the three-body system under consideration is the infinite number of the bound states for small $L$ and large $m/m_1$ (more precisely, for the finite interaction radius $r_0$ the number of states unrestrictedly increases with increasing $|a|/r_0$). As the effective potential in (\[system1\]) is approximately given by $(\gamma_1^2(0) - 1/4)/\rho^2$ at small $\rho$, the number of vibrational states is finite (infinite) if $\gamma_1^2(0) > 0$ ($\gamma_1^2(0) < 0$). According to Eq. (\[transeq\]), $\gamma_1^2(0)$ decreases with increasing $m/m_1$ and becomes zero at the critical value $(m/m_1)_{cL}$. Thus, one can define the step-like function $L_c(m/m_1)$, which increases by unity at the points $(m/m_1)_{cL}$, so that the number of vibrational states is infinite for $L < L_c(m/m_1)$ and finite for $L \ge L_c(m/m_1)$. Solving Eq. (\[transeq\]) at $\gamma_1 \to 0$ and $\rho \to 0$, one obtains the exact values $(m/m_1)_{cL}$, which approximately equal $13.6069657$, $38.6301583$, $75.9944943$, $125.764635$, and $187.958355$ for $L = 1 - 5$. Originally, the dependence $L_c(m/m_1)$ was discussed in [@Efimov73].
Analyzing the eigenvalue equation (\[transeq\]) one concludes that for $a > 0$ and $S(-)^L = -1$ the effective potential exceeds the threshold energy $E = -1$, $\gamma_1^2(\rho)/\rho^2 > -1$, therefore, the bound states only exist if either two identical particles are bosons and $L$ is even or two identical particles are fermions and $L$ is odd. Furthermore, one obtains the trivial answer if $a < 0$ and $L \ge L_c(m/m_1)$, for which $\gamma_1^2(\rho) > 0$ and there are no three-body bound states.
Numerical results {#bound}
=================
The mass-ratio dependence of the binding energies $\varepsilon_{L N}(m/m_1)$ for $L \ge L_c(m/m_1)$ and $a > 0$ is determined numerically by seeking the square-integrable solutions to Eq. (\[system1\]). Mostly, the properties of the energy spectrum are similar to those for $L = 1$, which were carefully discussed in [@Kartavtsev07]. For given $L$, there is the critical value of $m/m_1$ at which the first bound state arise, in other words, there are no three-body bound states for $L \ge L_b(m/m_1)$, where the step-like function $L_b(m/m_1)$ undergoes unity jumps at those critical values. Furthermore, all the bound states arise at some values of $m/m_1$ being the narrow resonances just below them. For the mass ratio near these values, the binding energies and resonance positions depend linearly and the resonance widths depend quadratically on the mass-ratio excess. Exactly at these values one obtains the threshold bound states, whose wave functions are square-integrable with a power fall-off at large distances. A set of these values of $m/m_1$ (more precisely, the lower bounds for them) is obtained numerically and presented in Table \[tab1\]. With increasing $m/m_1$, the binding energies monotonically increase reaching the finite values (shown in Table \[tab1\]) at $(m/m_1)_{cL}$; just below $(m/m_1)_{cL}$ they follow the square-root dependence on the difference $m/m_1 - (m/m_1)_{cL}$. Correspondingly, the number of the vibrational states increases with increasing $m/m_1$ taking the finite number $N_{max}$ at $(m/m_1)_{cL}$ and jumping to infinity beyond $(m/m_1)_{cL}$; in the present calculations $N_{max} = L + 1$ for $L < 9$ and $N_{max} = L + 2$ for $10 \le L \le 12$.
$N$ $L = 1$ $L = 2$ $L = 3$ $L = 4$ $L = 5$
----- --------- --------- --------- --------- ---------
1 7.9300 22.342 42.981 69.885 103.06
2 12.789 31.285 55.766 86.420 123.31
3 - 37.657 67.012 101.92 142.82
4 - - 74.670 115.08 160.64
5 - - - 123.94 175.48
6 - - - - 185.51
1 5.906 12.68 22.59 35.59 52.16
2 1.147 1.850 2.942 4.392 6.216
3 - 1.076 1.417 1.920 2.566
4 - - 1.057 1.273 1.584
5 - - - 1.049 1.206
6 - - - - 1.045
: Upper part: Mass ratios for which the $N$th bound state of the total angular momentum $L$ arises. Lower part: Binding energies $\varepsilon_{L N}$ for the mass ratio fixed at $(m/m_1)_{cL}$. []{data-label="tab1"}
Universal description of the spectrum {#largeL}
=====================================
A comprehensive description of the spectrum is obtained by using the large-$L$ (correspondingly, large-$m/m_1$) asymptotic expression for the binding energies $\varepsilon_{L N}(m/m_1)$. Taking the quasi-classical solution of Eq. (\[eqonhyp1\]) satisfying the zero boundary condition, $$\begin{aligned}
\label{qcphi}
\phi_{L, i\kappa}(\alpha) = \exp\left(\kappa \arccos\frac{x \cos\alpha}
{\sqrt{1 + x^2}}\right) \left(\frac{\sqrt{1 + x^2 \sin^2\alpha} - \cos\alpha}
{\sqrt{1 + x^2 \sin^2\alpha} + \cos\alpha}\right)^{L/2 + 1/4} \ , \end{aligned}$$ where $\gamma_1 = i\kappa$ and $x = \kappa/(L + 1/2)$, one writes the eigenvalue equation (\[transeq\]) in the form, $$\begin{aligned}
\label{qceigenv}
\frac{\rho }{L + 1/2} = \sqrt{1 + x^2} -
\frac{2 \exp\left(\kappa \arcsin\frac{x \cos\omega}{\sqrt{1 + x^2}}\right)}
{(L + 1/2)\sin 2\omega} \left(\frac{\sqrt{1 + x^2 \sin^2\omega} - \cos\omega}
{\sqrt{1 + x^2 \sin^2\omega} + \cos\omega}\right)^{L/2 + 1/4} \ . \end{aligned}$$ In the limit of large $L$ and $m/m_1$ the eigenvalue equation (\[qceigenv\]) reduces to $$\begin{aligned}
\label{adeigenv}
\rho\cos\omega = u - e^{-u} \ , \end{aligned}$$ where $u = \cos\omega\sqrt{\kappa^2 + (L + 1/2)^2}$. Notice that taking the limit $\kappa \to 0$ and $\rho \to 0$ in (\[adeigenv\]) one immediately obtains the relation $\cos\omega_{cL} = u_0/(L + 1/2)$, where $\sin\omega_{cL} =
(m/m_1)_{cL}/[1 + (m/m_1)_{cL}]$ and $u_0 \approx 0.567143$ is the root of the equation $u = e^{-u}$; as a result, one finds the asymptotic dependence $(m/m_1)_{cL} \approx 6.2179(L + 1/2)^2$ and the inverse relation $L_c(m/m_1) \approx (u_0 \sqrt{2m/m_1} - 1)/2 \approx
0.40103\sqrt{m/m_1} - 1/2$. Now the asymptotic dependence $\varepsilon_{L N}(m/m_1)$ for large $L$ and $m/m_1$ can be obtained by the quasi-classical solution of (\[system1\]) with $\gamma_1^2(\rho) = (L + 1/2)^2 -
[u(\rho)/\cos\omega]^2$ and $u(\rho)$ determined by (\[adeigenv\]), $$\label{qcint}
\displaystyle\int_{u_-}^{u_+} du \frac{1 + e^{-u}}{u -e^{-u}}
\sqrt{u^2 - \varepsilon_{L N} (u - e^{-u})^2 - L(L + 1)\cos^2\omega} =
\pi (N - 1/2)\cos\omega \ ,$$ where $u_-$ and $u_+$ are zeros of the integrand.
Following Eq. (\[qcint\]), one expects to express the binding energies via the universal function $\varepsilon_{L N}(m/m_1) = {\cal E}(\xi, \eta)$ of two scaled variables $\xi = \displaystyle\frac{N - 1/2}{\sqrt{L(L + 1)}}$ and $\eta = \displaystyle\sqrt{\frac{m}{m_1 L(L + 1)}}$. This two-parameter dependence is confirmed by the numerical calculations (up to $m/m_1 \sim 700$), which reveal that the calculated energies for $L > 2$ lie on a smooth surface as shown in Fig. \[figen\_univ\]. Even for the smallest $L = 1, 2$ the calculated energies are in good agreement with the two-parameter dependence showing only a slight deviation from the surface.
The variables $\xi $ and $\eta $ take values within the area limited by the line $\xi = 0$, the line $\eta = \eta_{max} \approx \sqrt{2}/u_0
\approx 2.493574$ stemming from the condition of finiteness of bound states $L \ge L_c(m/m_1)$, and the line ${\cal E}(\xi, \eta ) = 1$ expressing the condition of arising of the bound states at the two-body threshold. As shown in Fig. \[figen\_univ\], the smallest value $\eta = \eta_{min}$ is at $\xi = 0$, which corresponds to the condition of arising of the first bound state in the large-$L$ limit. To find it one requires that $u_+ = u_- \equiv u_b$ at $\varepsilon_{L N} = 1$ in Eq. (\[adeigenv\]), which leads to $\eta_{min} \approx
\sqrt{2/(u_b^2 - 1)} \approx 1.775452$, where $u_b \approx 1.278465$ is the root of the equation $u = 1 + e^{-u}$. This gives the asymptotic dependence for arising of the first bound state, $L_b \approx \eta_{min}^{-1}\sqrt{m/m_1} - 1/2 \approx 0.563237
\sqrt{m/m_1} - 1/2$. At the line $\eta = \eta_{max}$ the variable $\xi$ takes its largest value $\xi_{max}$, which determines the large-$L$ dependence of the number of the vibrational states $N_{max}$ for a given $L$. The calculation of the quasi-classical integral (\[qcint\]) gives $u_- = u_0$, $u_+ \approx 2.872849$, $\xi_{max} = \displaystyle\frac{1}{\pi u_0}\int_{u_-}^{u_+}
\frac{1 + e^{-u}}{u e^u - 1} \sqrt{e^u(2u - u_0^2 e^u) - 1}\ du
\approx 1.099839 $, and the large-$L$ estimate $N_{max} = \xi_{max} \sqrt{L(L + 1)} + 1/2$. Taking the entire part of this expression, one can predict that the dependence $N_{max} = L + 1$ for $L < 10$ changes to $N_{max} = L + 2$ at $L = 10$, which is in agreement with the numerical result.
The universal surface ${\cal E}(\xi, \eta )$ is bound by three lines, which are described by fitting the calculated energies for $L \ge 3$ to simple dependencies plotted in Fig. \[figen\_univ\]. As a result, the line defined by ${\cal E}(\xi, \eta ) = 1$ is fairly well fitted to $\eta = (\eta_{min} + a\xi)[1 - c \xi (\xi - \xi_{max})]$, where $a = (\eta_{max} - \eta_{min})/\xi_{max} \approx 0.652933$ is fixed by the evident condition ${\cal E}(\xi_{max}, \eta_{max}) = 1$ and the only fitted parameter is $c = 0.1865$. Furthermore, the analysis shows that the next boundary line defined by $\eta = \eta_{max}$ is described by ${\cal E}^{-1/2}(\xi, \eta_{max}) =
a_1 \xi(1 - a_2 \xi) [1 - c_1 \xi(\xi - \xi_{max})]$, where $a_1 = 1 + u_0 \approx 1.56714$ is fixed by the asymptotic behaviour of the integral (\[qcint\]) at $L \to \infty$ and $\eta = \eta_{max}$, $a_2 \approx 0.38171$ is fixed by the condition $a_1 \xi_{max}(1 - a_2 \xi_{max}) = 1$, and the only fitted parameter is $c_1 = 0.1881$. In particular, at the critical mass ratio the binding energy of the deep states in the limit of large $L$ is described by ${\cal E}(\xi, \eta_{max}) \to (a_1 \xi)^{-2}$, i. e., $\varepsilon_{N L}[(m/m_1)_{cL}] = \frac{L(L + 1)}{(N - 1/2)^2(1 + u_0)^2}$. The third boundary line at $\xi \to 0$ is described by the dependence ${\cal E}^{-1/2}(0, \eta) = a_3 \sqrt{\eta_{max} - \eta}
[1 + c_2 (\eta - \eta_{min})]$, where $a_3 = 1/\sqrt{\eta_{max} - \eta_{min}}
\approx 1.18$ is fixed by ${\cal E}(0, \eta_{min}) = 1$ and the only fitted parameter is $c_2 = 0.3992$.
Conclusion {#Conclusion}
==========
The presented results complemented by the accurate calculations for $L = 1$ [@Kartavtsev07] provide in the universal low-energy limit a comprehensive description of the rotational-vibrational spectrum of three two-species particles with the short-range interactions. Essentially, all the binding energies are described by means of the universal function ${\cal E}(\xi, \eta )$ for those $L_c(m/m_1) \le L \le L_b(m/m_1)$ which correspond to the finite number of vibrational states. One expects that the universal picture should be observed in the limit $|a| \to \infty$, e. g., if the potential is tuned to produce the loosely bound two-body state as discussed in [@Blume05; @Kartavtsev06].
It is of interest to discuss briefly the effect of the finite, though small enough interaction radius $r_0 \ll a$. For $L < L_c(m/m_1)$ Efimov’s infinite energy spectrum is extremely sensitive to the interaction radius $r_0$ and to the interaction in the vicinity of the triple-collision point, whereas for $L\ge L_c(m/m_1)$ the binding energies depend smoothly on the interaction parameters provided $r_0 \ll a$. For this reason, one expects not an abrupt transition from the finite to infinite number of bound states for $L = L_c(m/m_1)$ but a smeared off dependence for any finite value of $r_0/a$.
It is worthwhile to mention that arising of the three-body bound states with increasing mass ratio is intrinsically connected with the oscillating behaviour of the $2 + 1$ elastic-scattering cross section and the three-body recombination rate. In particular, for $L = 1$ it was shown in [@Kartavtsev07] that two interference maxima of the scattering amplitudes are related to the arising of two three-body bound states. Analogously, the dependence of the scattering amplitudes on the mass ratio for higher $L$ would exhibit the number of interference maxima which are related to arising of up to $N_{max} = 1.099839 \sqrt{L(L + 1)} + 1/2$ bound states.
Concerning possible observations of the molecules containing two heavy and one light particles in the higher rotational states, one should mention the ultra-cold mixtures of $^{87}\mathrm{Sr}$ with lithium isotopes [@Kartavtsev07] and mixtures of cesium with either lithium or helium. In particular, for $^{133}\mathrm{Cs}$ and $^6\mathrm{Li}$ the mass ratio $m/m_1 \approx 22.17$ is just below the value $m/m_1 = 22.34$ at which the $L = 2$ bound state arises and $m/m_1 \approx 33.25$ for $^{133}\mathrm{Cs}$ and $^4\mathrm{He}$ is above the value $m/m_1 = 31.29$, which corresponds to arising of the second $L = 2$ bound state. Also, a complicated rotational-vibrational spectrum and significant interference effects are expected for the negatively charged atomic and molecular ions for which the typical total angular momentum up to $L \sim 100$ becomes important due to the large mass ratio.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'New results from a large survey of H$\alpha$ emission-line galaxies at $z=0.84$ using the Wide Field CAMera on UKIRT and a custom narrow-band filter in the J band are presented as part of the HiZELS survey. The deep narrow-band images reach an effective flux limit of F$_{{\rm H}\alpha}\sim10^{-16}$ergs$^{-1}$cm$^{-2}$ in a co-moving volume of $1.8 \times 10^5$Mpc$^3$, resulting in the largest and deepest survey of its kind ever done at $z\sim1$. There are 1517 potential line emitters detected across $\sim1.4$ square degrees (in two fields: COSMOS and UKIDSS UDS), of which 743 are selected as H$\alpha$ emitters, based on their photometric and spectroscopic redshifts. These are then used to calculate the H$\alpha$ luminosity function, which is well-fitted by a Schechter function with $L^*=10^{42.26 \pm 0.05}$ergs$^{-1}$, $\phi^*$=$10^{-1.92 \pm 0.10}$ Mpc$^{-3}$ and $\alpha=-1.65\pm0.15$, and are used to estimate the volume average star formation rate at $z=0.845$, $\rho_{SFR}$: $0.15\pm0.01$M$_{\odot}$yr$^{-1}$Mpc$^{-3}$ (corrected for 15% AGN contamination and integrated down to 2.5 M$_{\odot}$yr$^{-1}$). These results robustly confirm a strong evolution of $\rho_{SFR}$ from the present day out to $z\sim1$ and then flattening to $z\sim2$ using a single star-formation indicator: H$\alpha$ luminosity. Out to $z\sim 1$, both the characteristic luminosity and space density of the H$\alpha$ emitters increase significantly; at higher redshifts, $L^*$ continues to increase, but $\phi^*$ decreases. The $z=0.84$ H$\alpha$ emitters are mostly disk galaxies ($82\pm3$%), while $28\pm4$% of the sample show signs of merger activity; mergers account for $\sim 20$% of the total integrated $\rho_{SFR}$ at this redshift. Irregulars and mergers dominate the H$\alpha$ luminosity function above $L^*$, while disks are dominant at fainter luminosities. These results demonstrate that it is the evolution of “normal“ disk galaxies that drives the strong increase in the star formation rate density from the current epoch to $z \sim 1$, although the continued strong evolution of $L^*$ beyond $z=1$ suggests an increasing importance of merger activity at higher redshifts.'
author:
- |
D. Sobral$^{1}$[^1], P. N. Best$^{1}$, J. E. Geach$^{2}$, Ian Smail$^{2}$, J. Kurk$^{3}$, M. Cirasuolo$^{1}$, M. Casali$^{4}$, R. J. Ivison$^{1,5}$, K. Coppin$^{2}$ & G. B. Dalton$^{6,7}$\
$^{1}$SUPA, Institute for Astronomy, Royal Observatory of Edinburgh, Blackford Hill, Edinburgh, EH9 3HJ, UK\
$^{2}$Institute of Computational Cosmology, Durham University, South Road, Durham, DH1 3LE, UK\
$^{3}$Max-Planck-Institut f[ü]{}r Astronomie, K[ö]{}nigstuhl, 17 D-69117, Heidelberg, Germany\
$^{4}$European Southern Observatory, Karl-Schwarzschild-Strasse 2, D-85738 Garching, Germany\
$^{5}$Astronomy Technology Centre, Royal Observatory of Edinburgh, Blackford Hill, Edinburgh, EH9 3HJ, UK\
$^{6}$Astrophysics, Department of Physics, Keble Road, Oxford, OX1 3RH, UK\
$^{7}$Space Science and Technology, Rutherford Appleton Laboratory, HSIC, Didcot, OX11 0QX, UK
bibliography:
- 'bibliography.bib'
date: 'Accepted 2009 May 22. Received 2009 May 11; in original form 2009 January 25'
title: ' HiZELS: a high redshift survey of H$\alpha$ emitters. II: the nature of star-forming galaxies at z=0.84 [^2] '
---
\[firstpage\]
galaxies: high-redshift, galaxies: luminosity function, cosmology: observations, galaxies: evolution.
Introduction
============
Understanding the basic features of galaxy formation and evolution requires unveiling the volume-averaged star formation rate as a function of epoch, $\rho_{SFR}$, its distribution function within the galaxy population, and its variation with environment. At the present, surveys of the star-formation rate density suggest a rise of $(1+z)^4$ out to at least $z\sim1$ [e.g. @1996ApJ...460L...1L; @con; @1998AAS...192.5104H; @2000ApJ...544..641H; @2004ApJ...615..209H; @2006ApJ...651..142H] indicating that most of the stars in galaxies today formed at $z>1$.
In the local Universe, studies have demonstrated that star formation is strongly dependent on the environment. While clusters of galaxies seem to be primarily populated by passively-evolving galaxies, star-forming galaxies are mainly found in less dense environments [e.g. @2002MNRAS.334..673L]. Star-forming galaxies have also been found to have lower masses than passive galaxies [@1996AJ....112..839C]. How do these environmental and mass dependencies change with cosmic time? When did they start to be noticeable, and how do they affect the evolution of galaxies, clusters and the star formation rate density of the Universe as a whole? How much of the evolution of the cosmic star formation rate density is associated with the evolution of star-forming galaxies and how much is driven by galaxy merger activity? In order to properly answer such questions it is mandatory to conduct observational surveys at high redshift, which can then be used to test theoretical models of galaxy evolution. These can be performed with a large variety of techniques and instrumentation.
There are many different star formation indicators which have been widely used both to select and study star-forming galaxies, such as the Ultra-Violet (UV) light emitted by young stars, the energy that is absorbed and then re-emitted in the form of emission lines (such as H$\alpha$ or \[O[ii]{}\]3727), or in the sub-mm and infra-red, or even radio emission from supernovae. Ideally, the use of different star formation tracers would provide consistent answers to most of the questions that remain unanswered. Unfortunately, studies with different star formation tracers suffer different biases and selection effects, which contribute to considerable discrepancies. These problems are also amplified by the effects of cosmic variance in most of the current samples. Another issue is the difficulty in correcting for extinction, especially for UV and optical wavelengths, which can lead to large systematic uncertainties in the star formation densities derived from measurements in these wavebands.
No single indicator provides a perfect view of the evolution of the star-formation rate density [@kewe], but using different tracers is not the solution either. Ideally, one needs to choose a star-formation indicator that can be applied from the present to high redshift, is relatively immune to dust extinction, and which has sufficient sensitivity to ensure that the derived star-formation rate distribution does not require large extrapolations for faint sources below the sensitivity limit. The H$\alpha$ luminosity is arguably the best candidate to achieve these goals and large-area surveys with a sensitivity of a few solar masses per year can now be done with current instrumentation out to $z\sim2$. This contrasts with the equivalent star formation rate limits for dust-independent tracers such as radio, far-infrared or sub-millimeter observations which vary from $\sim10$–100M$_{\odot}$yr$^{-1}$ at the same redshifts for large area surveys.
Large H$\alpha$ surveys have been carried out since the 1970’s [e.g. @cohen76; @kenkent; @Roma90; @Gava91; @galo; @yan; @alpha1; @doherty]. With the rise of large-format imaging cameras in the optical and in the near-infrared, narrow-band filters can now be used to undertake deep surveys for emission-line objects in large volumes which cover the great majority of the cosmic history, out to $z\sim2.5$ using H$\alpha$ [e.g. @2008ApJ...677..169V; @2008arXiv0805.2861G]. The surveys identify sources on the basis of the strength of their emission line and thus represent roughly a star-formation rate selected sample. Hence, by using a set of narrow-band filters, it is possible to apply a single technique to target H$\alpha$ emitters across a wide range of redshifts, gathering representative samples at each epoch with a uniform selection. Those samples are powerful tools for tracing the evolution in the star-formation rate density across the expected peak of star-formation in the history of the Universe, which is one of the main aims of the High-$z$ Emission Line Survey (HiZELS).
HiZELS is a panoramic extragalactic survey using the wide area coverage of the Wide Field CAMera (WFCAM, Casali et al. 2007) instrument on the 3.8-m UK InfraRed Telescope (UKIRT). The survey utilizes a set of existing and custom-made narrow-band filters in the $J$, $H$ and $K$ bands to detect emission line galaxies over $\sim 5$ square degrees of extragalactic sky. The H$_2$S(1) narrow-band filter is being used to target H$\alpha$ emitting galaxies at $z=2.23$, which are predominantly star-forming systems (Geach et al. 2008). In addition, narrow-band filters in the $J$ and $H$ bands (hereafter NB$_J$ and NB$_H$) have been custom-designed to target the \[O[ii]{}\]3727 and \[O[iii]{}\]5007 emission lines in galaxies at the same redshift as the H$_2$S(1) H$\alpha$ survey. Together, the three sets of filters help to detail the properties of the line emitters at $z=2.23$, while the NB$_J$ and NB$_H$ filters deliver identically-selected H$\alpha$ samples at $z=0.84$ and 1.47 respectively.
---------- -------- ------------- ----------- ----------- -------- ----------------------- ------------- -- --
Field Filter R.A. Dec. Int. time FHWM Dates $m_{lim}$
[(J2000)]{} (J2000) (ks) ($''$) (2007) (3$\sigma$)
COSMOS 1 NB$_J$ 100000 +021030 19.7 1.0 14–16 Jan 21.7
COSMOS 2 NB$_J$ 100052 +021030 21.6 1.0 13, 14 Jan 21.6
COSMOS 3 NB$_J$ 100000 +022344 19.0 0.9 15–17 Jan 21.7
COSMOS 4 NB$_J$ 100053 +022344 17.2 0.9 15, 17 Jan, 13–16 Feb 21.6
COSMOS 1 $J$ 100000 +021030 5.7 0.9 14–16 Jan 22.8
COSMOS 2 $J$ 100052 +021030 6.9 1.0 13, 14 Jan 22.8
COSMOS 3 $J$ 100000 +022344 5.7 0.9 15–17 Jan 22.8
COSMOS 4 $J$ 100053 +022344 5.1 0.9 15, 17 Jan, 13–16 Feb 22.6
UDS NE NB$_J$ 021829 $-$045220 21.2 1.0 18, 20 Oct 21.6
UDS NW NB$_J$ 021736 $-$045220 22.6 1.0 19, 21 Oct 21.7
UDS SE NB$_J$ 021829 $-$050553 20.0 1.0 19 Oct 21.5
UDS SW NB$_J$ 021738 $-$050534 22.5 1.0 20, 21 Oct 21.6
---------- -------- ------------- ----------- ----------- -------- ----------------------- ------------- -- --
\[obs\]
This paper presents deep narrow-band imaging using the NB$_J$ filter at $\lambda = 1.211\umu$m, as part of HiZELS, over $\sim 0.7\deg^2$ of the in the SXDF Subaru-XMM–UKIDSS Ultra Deep Survey [@2007MNRAS.379.1599L] field (UDS), and $\sim 0.8\deg^2$ in the Cosmological Evolution Survey [@2007ApJS..172....1S; @2007ApJS..172..196K] field (COSMOS). This corresponds to an area coverage which is $\sim 8$ times larger and twice the depth of the survey by [@2008ApJ...677..169V], the best previous emission-line survey at $z\sim1$.
The paper is organised in the following way. §2 outlines the details of the observations, and describes the data reduction, photometric calibration, source extraction and survey limits. §3 presents the narrowband selection criteria, the final sample and the photometric and spectroscopic redshift analysis. Results are presented in §4: the H$\alpha$ luminosity function, the star-formation rate density, the morphologies of the H$\alpha$ emitters and their relation and contribution to the H$\alpha$ luminosity function. Finally, §5 outlines the conclusions. An H$_0=70$kms$^{-1}$Mpc$^{-1}$, $\Omega_M=0.3$ and $\Omega_{\Lambda}=0.7$ cosmology is used and all magnitudes are in the Vega system, except if noted otherwise.
OBSERVATIONS and DATA REDUCTION
===============================
Both fields were observed with WFCAM on UKIRT using a set of custom narrow-band J filters ($\lambda = 1.211\umu$m, $\delta\lambda = 0.014\umu$m), during 2007 January 13–17 and February 13,14 & 16 for the COSMOS field and October 19–24 for UKIDSS UDS. WFCAM’s standard “paw-print” configuration of four $2048\times2048$ $0.4''$pixel$^{-1}$ detectors offset by $\sim20'$ can be macrostepped four times to cover a contiguous region of $\sim55'\times55'$ . For each field (COSMOS and UDS) 4 paw-prints, or $\sim0.8 \deg^2$ were mapped with narrowband exposures of $\sim21$kspixel$^{-1}$. The seeing varied between $0.8$–$1.0''$ during the observing nights, and conditions were photometric. The NDR (Non Destructive Read) mode was used for all narrow-band observations to minimise the effects of cosmic rays in long exposures. The observations were obtained following a 14-point jitter pattern for UDS (to match the UKIDSS observing strategy) and a 9-point jitter sequence for COSMOS. Broad-band $J$-band observations were also obtained in the COSMOS field, but were not necessary for the UDS due to the availability of UKIDSS data. A summary of the observations is given in Table \[obs\].
Data reduction
--------------
In order to accomplish the survey goals, a dedicated pipeline has been developed: PfHiZELS. This was done following the method presented in [@2008arXiv0805.2861G], improving several steps and producing a user-friendly, completely automated pipeline. It consists of a set of [python]{} scripts which implement a complete reduction of WFCAM data, taking advantage of several publicly available packages, as detailed below. Small samples of data were also reduced by hand using [iraf]{} (including [dimsum]{} and [xdimsum]{}), which allowed a comparison with the automatic data reduction and showed that the results are comparable.
Dark frames were median combined for each observing night and the individual science frames were then dark subtracted. Next, frames were median combined in each jitter sequence of 14 (UDS) or 9 (COSMOS) without any offset, to produce a rough first pass flat. The latter was then used to produce a badpixel mask for each chip by flagging pixels which deviate by more than $3\sigma$ from the median value. Each frame was then flattened, and an individual badpixel mask (for each frame) was produced.
In order to produce better flats, [SExtractor]{} was used to identify objects on the flattened images, and source masks were created and used to produce second-pass flat fields. These were then used to create the final flattened frames.
A world coordinate system was fitted to each frame by querying the USNO A2.0 catalog. This fitted, on average, $\sim 100$ objects per frame and rotated images appropriately for each WFCAM chip. Frames were then de-jittered and co-added with [SWarp]{} (Bertin 1998), which performs a background mesh-based sky subtraction optimized for narrow-band data. It should be pointed out that WFCAM frames suffer from significant cross-talk artifacts, which manifest themselves as toroidal features at regular (every 128) pixel intervals from sources, in the read-out direction. Furthermore, as these are linked to a “physical” location, they can only be removed from the source catalog (see §\[extraction\]).
Photometry calibration
----------------------
Narrow- and broad-band images were photometrically calibrated (independently) by matching $\sim70$ stars with $m_J=11$–16 per frame from the 2MASS All-Sky catalog of Point Sources [@2003tmc..book.....C] which are unsaturated in narrow-band frames. Based on the photometric calibration, the absolute calibration is expected to be accurate to $<3\%$ when compared to 2MASS. We have also double-checked the offset between zero-points by comparing them to the expected values, roughly given by $-2.5\log\frac{\delta \lambda_J\times \delta t_J}{\delta \lambda_{NB_J}\times \delta t_{NB_J}}$, where $\delta \lambda$ is the filter width and $\delta t$ the exposure time per frame. For convenience, after this step, narrow and broad-band images were normalised to give them the same zero-point.
Source Extraction and Survey Limits {#extraction}
-----------------------------------
The survey is made up of a mosaic of eight WFCAM pointings (4 each for UDS and COSMOS), i.e., $8\times4\times13.7'\times13.7'$ tiles. The UKIDSS UDS $J$-band image does not overlap entirely with the full narrow-band image, leading to a total overlapping area of $0.70 \deg^2$, while for COSMOS the overlap is $0.78 \deg^2$.
Sources were extracted using [SExtractor]{} . Optimal parameters were found by running a large number of different extractions which converged to a set of parameters that allowed the extraction of all obvious sources down to the 3-$\sigma$ limit in each frame and minimized the extraction of noise/artifact features. The extraction included an optimized sky subtraction, and fixed photometry apertures of $3''$ (diameter) were used. Several tests were done using [SExtractor]{} in dual mode (using frames in one band to detect the sources and measuring on the other band) and single mode. The first option was used for COSMOS (where data were taken in both bands and so the frames were extremely well registered), while for UDS (where UKIDSS data was used for the $J$ band) the extraction was done on both bands independently, followed by a match using a simple $2.0''$ criteria. In 11 cases there was more than one match; for these, a careful local astrometry solution was calculated based on nearby sources clearly identified in both bands and the correct source was then clear in all cases (the correct match was always the original closest). Furthermore, tests were run on COSMOS frames to show that both options produce comparable samples, and so this is not likely to produce any significant difference between the extraction for UDS and COSMOS. Narrowband sources with no clear $J$ band detection were retained but there are few of these as the $J$ coverage is significantly deeper than the NB$_J$. As the observations have slightly different FHWM, different total exposure times, and because each WFCAM chip has slightly different properties, objects were extracted down to each chip’s limit, which was then confirmed using Monte Carlo simulations (see §\[compl\]).
![Two narrow-band excess objects selected from the NB$_J$ imaging with different colour significances (upper panels), and a strong isolated emitter (lower panels). Images are $\sim20''\times20''$. \[excess\]](./figs/excess_without_markers_COSMOS2.jpg){width="8.2cm"}
In order to clean spurious sources from the catalogue (essential to remove cross-talk artifacts), the frames were visually inspected showing that sources brighter than $\sim12$mag are surrounded by large numbers of artifacts detected within “bright halos”, as well as cross-talk, while fainter sources (up to 16mag) show only cross-talk features. Sources fainter than 16mag do not produce any detectable cross-talk at the typical depth of the observations. Cross-talk sources and detections in the halo regions were removed from the catalogue separately for each frame, which greatly simplifies their identification (Geach et al. 2008).
Catalogue sources detected in regions with less than 85% of the total integration time were removed; which still assures a complete overlap between frames. When a source was catalogued in more than one final image, the catalogue entry with a higher exposure time was selected.
The average 3-$\sigma$ depth of the entire set of NB$_J$ frames is 21.6mag, with $J$ depths being $J=23.4$mag (UKIDSS UDS DR3) and $J=22.7$mag (COSMOS). The narrow-band imaging detects a total of 21773 objects in COSMOS across $0.76\deg^2$ and 15449 in UDS across $0.68 \deg^2$. These areas include the removal of regions in which cross-talk and other artifacts caused by bright objects are located.
SELECTION
=========
Narrowband excess selection
---------------------------
Emission line systems are initially selected according to the significance of their $(J-$NB$_J)$ colour, as they will have $(J-$NB$_J)>0$. However, true emitters need to be distinguished from those with positive colours due to scatter in the magnitude measurements and this is done by quantifying the significance of the narrowband excess. The parameter $\Sigma$ quantifies the excess compared to the random scatter expected for a source with zero colour [@1995MNRAS.273..513B]. In other words, a source can be considered to have a genuine narrow-band excess if:
$$c_{NB_J}-c_{J}>\Sigma\delta$$
where $c_{{\rm NB}_J}$ and $c_{J}$ are the counts for the NB$_J$ and $J$ bands, respectively, and $\delta$ is the combined photometric error:
$$\delta=\sqrt{\pi r^2(\sigma_{{\rm NB}_J}^2+\sigma_{J}^2)}.$$
The approach is similar to a standard signal-to-noise selection. Colour and $\Sigma$ significances are related by
$$J-{{\rm NB}_J}=-2.5\log(1-\Sigma\delta10^{-0.4(ZP-{{\rm NB}_J})}),$$
where $ZP$ is the zeropoint of both frames, assuming that the sky variation is the dominant feature contributing to the errors in photometry. This has been tested by measuring counts in randomly placed apertures ($\sim 10000$ for each frame), confirming the hypothesis.
{width="8.2cm"}
{width="8.2cm"}
As the NB$_J$ filter does not fall at the centre of the $J$-band, objects with redder colours will tend to have a negative $(J-$NB$_J)$ colour, while bluer sources will have $(J-$NB$_J)>0$. This will affect the selection of emission line objects, but can be corrected for by considering the broad-band colours of each source. To do this, $(J-$NB$_J)$ was plotted as a function of $(z-J)$ (COSMOS) or $(J-K)$ (UDS) colour and a linear fit was derived, determining trends with a slope of $(J-$NB$_J)\sim 0.06 (J-K)$ for UDS and $(J-$NB$_J)\sim 0.11 (z-J)$ for COSMOS. This was then used to correct the NB$_J$ magnitudes and thus the resultant $(J-$NB$_J)$ colours. Sources with no $z$ or $K$ data – either because they are too bright or too faint – were corrected assuming a median colour (based on sources at similar magnitudes).
The flux of an emission line, F$_{{\rm line}}$, and the equivalent width, EW$_{{\rm line}}$, can then be expressed as: $${\rm F}_{{\rm line}}=\delta_{NB_J}\frac{f_{NB_J}-f_J}{1-(\Delta\lambda_{NB_J}/\Delta\lambda_{J})}$$ and $${\rm EW}_{{\rm line}}=\delta_{NB_J}\frac{f_{NB_J}-f_J}{f_J-f_{NB_J}(\Delta\lambda_{NB_J}/\Delta\lambda_{J})}$$ where $\Delta\lambda_{NB_J}$ and $\Delta\lambda_{J}$ are the widths of the broad and narrow-band filters, and $f_{NB_J}$ and $f_{J}$ are the flux densities measured for each band. EW$_{{\rm line}}$ is simply the ratio of the line flux and continuum flux density.
The selection of emission-line candidates is done imposing two conditions. Firstly, the NB$_J$ sources are considered candidate emitters if they present a colour excess significance of $\Sigma>2.5$. Secondly, only NB$_J$ detections with EW$_{{\rm line}}>50$Å (corresponding to $(J-$NB$_J)>0.3$) are selected. This is done to avoid including bright foreground objects with a large significance and a steep continuum across the $J$ band , and was chosen to reflect the general scatter around the zero colour a bright magnitudes. A comparison with spectroscopic data (see §\[red\]) indicates that these criteria maximise the completeness of the sample without introducing a significant number of false emitters. Figure \[colour-magC\] shows the colour-magnitude diagrams with the selection criteria for COSMOS and UDS, respectively. Figure \[excess\] presents two examples of emitters.
Complete sample
---------------
Narrow-band detections below the 3-$\sigma$ threshold were not considered. When there was no $>3\sigma$ $J$-band detection, a 3-$\sigma$ upper limit for the $(J-$NB$_J)$ colour was computed, as shown in Figure \[colour-magC\]. The average 3-$\sigma$ line flux limit over both fields is $8\times10^{-17}$ergs$^{-1}$cm$^{-2}$. The complete sample has 1517 excess sources out of all 37222 NB$_J$ detections in the entire area, with 824 being detected in COSMOS (corresponding to 1084 emitters per $\deg^2$) and 693 in UDS (1020 emitters per $\deg^2$). The potential emitters were all visually inspected in both $J$-band (when detected) and NB$_J$, and 48 were removed from the sample as they were flagged as spurious. The majority of these (34) correspond to artifacts caused by bright stars that are on the edges of two or more frames simultaneously. The remaining 14 sources removed were low S/N detections in noisy regions of the NB$_J$ image.
Photometric redshift analysis {#foto}
-----------------------------
In order to select H$\alpha$ emitters at $z=0.84$, one needs to separate them from other line emitters at different redshifts. The detection of H$\beta$ and \[O[iii]{}\]5007 emitters at $z\sim1.4$–1.5 is expected, together with \[O[ii]{}\]3727 at $z=2.23$ and other emission lines. Multi-wavelength data, photometric and spectroscopic redshifts available for both COSMOS [@2007ApJS..172..117M; @zCOSMOS] and UDS [@2008arXiv0804.3471C] are therefore used to distinguish between different emission lines and also to evaluate how robust the emitter selection criteria are. In order to do this, photometric redshifts for the emission line candidates were taken from the photometric redshift catalogues for COSMOS and UDS [@2007ApJS..172..117M; @2008arXiv0804.3471C]. For UDS, the overlap with the photometric redshift catalogues reduces the area coverage to $0.54\deg^2$ due to the required overlap with Subaru optical data and further masking around bright stars: 133 sources are “lost” as they are in the excluded area. In addition, for 14 sources photometric redshifts are not available in the catalogues due to i/$K_s$ magnitude limits (3 in COSMOS, 11 in UDS): while these appear to be real sources (although there is a chance they are spurious), they are likely to be at redshifts higher than $z=0.84$ and so are excluded from this analysis of H$\alpha$ emitters. The photometric redshift matches result in a final sample of 1370 excess sources (821 in COSMOS, 549 in UDS) meeting all of the selection criteria across a total of $1.30\deg^2$.
{width="8.2cm"}
{width="8.2cm" height="6.3cm"}
The photometric redshifts in COSMOS are particularly good for selecting H$\alpha$ emitters at $z=0.84$ as the available data are ideal to probe the 4000Å break at that redshift. The deeper near-infrared data used in the UDS photometric redshifts provide better accuracy for higher redshift sources, and are accurate enough to probe the proportion of $z\sim1.4$–1.5 \[O[iii]{}\]5007 and H$\beta$ emitters. Photometric redshifts for COSMOS present $\sigma(\Delta z ) = 0.03$, where $\Delta z$ = $(z_{phot}-z_{spec})/(1+z_{spec})$. The fraction of outliers, defined as sources with $\Delta z>3\sigma(\Delta z)$, is lower than 3%, based on results from $z$-COSMOS (see §\[red\]). For UDS, the photometric redshifts have $\sigma(\Delta z) = 0.04$, with 2% of outliers.
Figure \[spectroz\] shows the photometric redshift distribution for the selected narrow-band emitters in COSMOS and UDS, demonstrating very good agreement between the two fields, despite the completely different photometric analyses – done with different codes, bands and by completely independent teams. Both samples peak at $z\sim0.85$, implying that the majority of the narrow-band excess sources are indeed H$\alpha$ emitters. In addition to this, there is another peak at $z\sim1.4$–1.5 in UDS and the same – but at a slightly higher photo-z – in COSMOS. This is interpreted as a significant population of \[O[iii]{}\]5007 emitters at $z\sim1.4$ and/or H$\beta$ emitters at $z\sim1.5$. However, the photometric redshifts are not accurate enough to clearly distinguish between these two populations, and thus the slight difference between UDS and COSMOS might be just caused by the use of different bands and methods giving different performance at $z>1$.
Emission Line $\lambda$ (Å) $<z>$ Number Selected as H$\alpha$
----------------- --------------- ------- -------- -----------------------
H$\alpha$ 6563 0.845 93 88
H$\beta$ 4861 1.49 2 0
$[$O[iii]{}$]$ 5003 1.42 4 1
He[i]{} 5876 1.04 1 0
$[$O[i]{}$]$ 6363 0.89 4 0
$[$S[ii]{}$]$ 6717 0.79 2 1
$[$Ar[iii]{}$]$ 7135 0.70 3 0
$[$O[ii]{}$]$ 7325 0.63 1 0
He[ii]{} 8237 0.47 1 0
C[i]{} 9830 0.22 1 0
Unidentified 7 0
\[z\_table\]
Selecting H$\alpha$ emitters at ${z=0.84}$ {#selecting}
------------------------------------------
The sample of candidate H$\alpha$ emitters at $z=0.84$ is derived by considering not only the best-fit photometric-redshift for each source, but also more extensive information contained within the photometric redshift probability distribution. In particular, the H$\alpha$ candidates are defined to be those sources with $z_{min}<0.845<z_{max}$ (where $z_{min}$ and $z_{max}$ are the 1-$\sigma$ redshift limits of the principle peak in the photometric redshift probability distribution). For COSMOS, the photometric redshift selection produces a sample of 477 potential H$\alpha$ emitters, while for UDS the same procedure yields 270 sources (see §\[foto\] for details).
{width="8.2cm"}
{width="8.2cm" height="6.3cm"}
While it is true that this selection can potentially introduce some biases, a careful comparison of the selected H$\alpha$ emitters and the remaining emitters shows no evidence for such an effect. The emission line equivalent width distributions of the selected H$\alpha$ and non-H$\alpha$ emitters are statistically indistinguishable. The sample of non-selected emitters does contain objects which have slightly fainter magnitudes (and correspondingly fainter emission lines) than the selected H$\alpha$ emitters, but the colours of these fainter galaxies are clearly consistent with them simply being at higher redshift (e.g. \[OIII\], H$\beta$ or \[OII\] emitters): they are typically detected at a $\sim$20-30 $\sigma$ (or higher) level in the optical images, and thus it should not be any inaccuracy in the photometric redshifts which affects their selection. One can also see that the photometric redshift distribution in Figure \[spectroz2\] (left panel) is clearly different between the selected H$\alpha$ emitters and the remaining emitters, revealing both a population of lower and higher redshift. The right panel of Figure \[spectroz2\] also reveals that the H$\alpha$ selected emitters present colours which clearly distinguish them from the low-z and higher-z samples of emitters – they also occupy a region in the colour-colour diagram where one expects to find star-forming galaxies at z=0.84. Finally, as examined below, spectroscopic redshifts for $\sim$ 25% of the sample in COSMOS show that only a negligible fraction of the H$\alpha$ emitters are being missed by the photometric redshift selection.
Spectroscopic redshifts and selection robustness {#red}
------------------------------------------------
Both COSMOS and UDS have large spectroscopic surveys underway, but only a limited fraction of those spectroscopic redshifts is currently available. Nevertheless, for the $z$-COSMOS survey [@zCOSMOS] Data Release 2 (DR2), the match with the sources detected on the NB$_J$ frames yields 4600 sources (using a $1.0''$ match criteria), of which 138 (each matching only one $z$-COSMOS source) have been selected as excess sources and hence potential emitters. Twelve of these have unreliable or non-existent redshift determinations. Of the remaining 126, the vast majority (119) have redshifts which place an emission line within the NB$_J$ filter (see Table \[z\_table\] for the detailed list of emitters), confirming the narrow-band excess as an emission line. For the other 7 sources, it was not possible to identify any emission line falling into the NB$_J$ filter. Spectra for these sources were analysed in order to look for potential errors in the redshift determination. In fact, while redshifts for 5 of them are very robust, one source, identified as $z\sim0.4$, seems better fitted by being at $z=0.85$ (with \[O[ii]{}\]3727 being detected, while the fit at lower redshift assumes an emission line where fringing starts to become an issue in the spectrum). The remaining source contains one emission line (identified as \[O[ii]{}\]3727) but with a low signal to noise (S/N$<2.0$). These 7 sources all have a colour excess significance $\Sigma=2.5$–3.0, and at least the 5 robust cases represent the galaxies expected to be randomly scattered into the candidate emitter list at these low $\Sigma$ values. All of the candidate emitters with $\Sigma>3.0$ were confirmed to be real. It is also worth noticing that $z$-COSMOS DR2 is highly biased towards $z<1$ sources, which means that its completeness for potential interlopers (which are more likely to be in this redshift range) is very high, while at the same time it misses most of the emitters at higher redshift. Therefore, the contamination rate within the sample of emitters (i.e., the fraction of non-emitters) is likely to be lower than $\sim6$%, even down to $\Sigma=2.5$.
Tests regarding the selection criteria used were also done by selecting samples with different colour significances and different equivalent width cuts, and comparing the matches for each selection criteria with $z$-COSMOS DR2. As expected, the number of potential interlopers declines rapidly with an increasing colour significance threshold, but so does the sample size. The sample with the largest number of real emitters whilst returning a low ratio between potential interlopers and real emitters is obtained using $\Sigma\sim2.5$ and an equivalent width cut of $\sim50$Å, therefore indicating that the selection criteria used are producing reliable results.
Table \[z\_table\] compares the spectroscopic selection of line emitters with the selection from the photometric redshift catalogue. It can be seen that 88 of the 93 H$\alpha$ emitters are correctly identified ($\sim95$% completeness). On the other hand, 2 non-H$\alpha$ emitters (with fluxes which do not deviate significantly from the median) were found in the photometric redshift selected sample (implying a $\sim98$% reliability), with one of those being a \[S[ii]{}\]6731 emitter. Because this line is very close to H$\alpha$ it is very difficult to completely distinguish these emitters with photometric redshifts alone. However, as seen by the sample which has been presented, the total contamination by these will be very small ($<2$%). The spectroscopic data were used to improve the H$\alpha$ catalogue, removing the 2 non-H$\alpha$ emitters and including those 5 which had not been selected by the photometric redshifts. These few sources present mean and median fluxes slightly higher than the rest of the sample, suggesting that some bright H$\alpha$ emitters could be lost by a photometric redshift selection, due to the emission line contribution to the broad-band fluxes. However, as the brighter H$\alpha$ emitters are likely to contain strong emission lines in the visible (\[O[ii]{}\]3727, H$\beta$ and \[O[iii]{}\]5007), these targets should have a very high completeness in $z$-COSMOS, so this can be interpreted as an upper limit. As these emitters are introduced into the sample, no further specific correction was applied.
While 93 H$\alpha$ emitters are confirmed using $z$-COSMOS, there are 72 $z$-COSMOS galaxies with a spectroscopic redshift which should place the H$\alpha$ line in the NB$_J$ filter. The failure to select these as narrow-band excess sources means that they have a weak or absent H$\alpha$ emission line. These present a mean colour $(J-$NB$_J)=0.1$ corresponding to a mean measured H$\alpha$ flux lower than the survey limit ($3\times 10^{-17}$ergs$^{-1}$cm$^{-2}$ and a mean EW of 20Å). The brightest source has $J=19$ mag. These are likely to be very faint emitters for which a completeness correction is applied later (see §\[compl\]). The \[O[ii]{}\]3727 line fluxes from $z$-COSMOS galaxies confirms this: where they are detected they yield similar line fluxes to the ones that would be estimated for H$\alpha$. On the other hand, some of these sources have a negative $(J-$NB$_J)$ colour. By applying similar selection criteria used for emitters, it seems that at least 2 H$\alpha$ absorbers would be selected.
Narrowband-$K$ and $H$ matches
------------------------------
By design, the custom-built narrow-band filters make it possible to look for line detections in multiple bands to refine redshift estimates. Currently, data are available in H$_{2}$S$_{1}$ for COSMOS (Geach et al. 2008) and NB$_H$ and H$_{2}$S$_{1}$ for UDS (Sobral et al., Geach et al., in prep.)
For COSMOS, it was possible to match 3 sources between excess sources catalogs ($J$ and $K$), confirming a $z=2.23$ redshift for those emitters. This shows that the line being detected in NB$_J$ is \[O[ii]{}\]3727.
For UDS, 18 sources were matched between the NB$_H$ and NB$_J$. These have photometric redshifts consistent with either being H$\alpha$ emitters at $z\sim 1.5$ or H$\alpha$ emitters at $z=2.23$, thus meaning that the line detected in the $J$ band is H$\beta$ or \[O[ii]{}\]3727 (respectively) for these emitters. At least 2 of these are also selected as emitters in the NB$_K$ filter, which confirms $z=2.23$ for them. In addition, there are another 6 matches between the $K$ and the $J$ narrow-band filters, indicating that those sources are at $z=2.23$ as well. Four of these 24 matched sources were found in the photometric redshift-derived H$\alpha$ catalog and were thus removed as the evidence points towards these being at higher redshift.
RESULTS
=======
A total of 743 candidate $z = 0.84$ H$\alpha$ emitters are detected over $1.30\deg^2$ down to an average observed line flux limit of $8\times10^{-17}$ergs$^{-1}$cm$^{-2}$. This sample will now be used to evaluate the H$\alpha$ luminosity function and estimate the star formation rate density at $z=0.84$. The morphological mix of these H$\alpha$ emitters in the COSMOS field will then be investigated, together with their evolution as a function of luminosity and redshift, and their contribution to the total luminosity function and star formation rate density.
H$\alpha$ luminosity function at $z=0.84$ {#lfalpha}
-----------------------------------------
In order to calculate the luminosity function of H$\alpha$ emitters, line fluxes are converted to luminosities by applying:
$$L_{{\rm H_{\alpha}}}=4\pi {\rm D_L^2}{\rm F}_{{\rm H_{\alpha}}}$$
where ${\rm D_L}$ is the luminosity distance, 5367Mpc at $z=0.84$.
The estimate of the source density in a luminosity bin of width $\Delta(\log L)$ centered on $\log L_c$ is given by the sum of the inverse volumes of all the sources in that bin. Therefore, the value of the source density in that bin is
$$\phi(\log(L_c))=\frac{1}{\Delta(\log L)} \sum_{|\log \frac{L_i}{L_c}| < \frac{\Delta(\log L)}{2}} \frac{1}{\Delta(V_{\rm filter})}.$$
Here, $i$ refers to sources and $c$ to the center of each bin. The volume probed is calculated taking into account the survey area and the narrow-band filter width (initially assumed to be a top-hat function across 1.2037–$1.2185\umu$m; see §\[profil\_filt\] for a refined approach), resulting in a co-moving volume of $7.4 \times 10^4$Mpc$^{3}$ (UDS) and $1.04 \times 10^5$Mpc$^{3}$ (COSMOS). As detailed before, this takes into account the removed area due to the presence of bright stars and consequent artifacts, along with noisy areas.
The luminosity functions presented here are fitted with a Schechter function defined by the three parameters: $\alpha$, $\phi ^*$ and $L^*$:
$$\phi(L)dL=\phi^*(L/L^*)^\alpha \exp(-L/L^*)d(L/L^*).$$
### \[NII\] flux contamination correction {#nii}
When computing line fluxes and equivalent widths for the H$\alpha$ line, one must note that the adjacent \[N[ii]{}\] lines at 6548 and 6583Å will also contribute to both quantities, increasing them both. In order to account for this, the relation between the flux ratio F$_{{[\rm N\sc II]}}/$F$_{{\rm H}\alpha}$ and the total measured equivalent width EW(H$\alpha$+\[N[ii]{}\]) derived by Villar et al. (2008) is used. This shows that the fractional \[N[ii]{}\] flux contribution decreases with increasing EW. The relation between $y=\log($F$_{{\rm {[N}{\sc II}]}}/$F$_{{\rm H}\alpha})$ and $x=\log($EW(H$\alpha+[$N[ii]{}$]))$ can be approximated by $y=-5.78+7.63x -3.37x^2+0.42x^3$ for EW$>50$Å. This results in a correction lower than the conservative 33% used by some authors (e.g. Geach et al. 2008), with the median being 25% for this sample with EW$>50$Å.
### Completeness correction {#compl}
Fainter sources and those with weak emission lines might be missed and thus not included in the sample; this will result in the underestimation of the number of emitters, especially at lower luminosities. Furthermore, the completeness rate is highly affected by the selection criteria ($\Sigma>2.5$, EW$>50$Å, $\sigma>3$). To address this problem, a series of simulations were conducted. First, the recovery rate has been studied as a function of magnitude for objects in the same frame, using a Monte Carlo method. For this, 10 different galaxies (both real – taken from the narrow-band images in UDS and COSMOS – and simulated) were used. These were introduced (20 for each run, in any given image) with different input magnitudes into the science frames and then detected using the same extraction parameters as the main catalogues. The recovery rate and the recovered magnitude were then studied. The latter follows the input magnitude reasonably well down to the 3-$\sigma$ limit. The recovery rate, while varying slightly with the type of galaxy used (the artificial and point-like objects, for example, showed a much higher recovery rate at fainter levels than real galaxies), falls off sharply fainter than NB$_J\sim21$. Similar simulations were also done for the $J$ band images.
![The total completeness function used for COSMOS and UDS resulting from several simulations to address the effects of selection and detection. This confirms a $\sim 30$% completeness at a flux limit of $\sim 10^{-16}$ergs$^{-1}$cm$^{-2}$.\[completeness\]](./figs/completeness.jpg){width="8.2cm"}
This information alone cannot be used to correct the luminosity function, as faint narrow-band sources don’t necessarily have faint H$\alpha$ lines and vice-versa. To address this, a second series of simulations were performed to investigate the selection completeness of faint emission lines within detected galaxies. For this, narrow-band detections not classified as emitters were considered not to have any emission line. For each of those sources ($\sim21000$ for COSMOS and $\sim15000$ for UDS) a line flux (in steps of $5\times10^{-18}$ergs$^{-1}$cm$^{-2}$ up to $5\times10^{-15}$ergs$^{-1}$cm$^{-2}$) was added to both NB$_J$ and $J$-band magnitudes, and the selection criteria were applied to the revised magnitudes for each line flux to study the recovery rate. This outputs a colour selection completeness for each line flux. Finally, a combined completeness correction was done combining both simulations described. The combined completeness function, which is used to correct the luminosity function on a bin to bin basis, is shown in Figure \[completeness\]. Slightly higher completeness rates at the same flux level have been claimed in [@2008ApJ...677..169V], for shallower data, but it should be pointed out that these authors use a combination of selection criteria which is significantly different from those used in this study. On the other hand, they only consider the colour-selection completeness, neglecting the detection completeness, and on the other hand, they use sources detected down to just $0.8\sigma$ in NB$_J$ (instead of the 3-$\sigma$ limit used in this work), and with a lower EW ($\sim15$Å) cut; therefore, the results presented here are likely to be more reliable than those by [@2008ApJ...677..169V].
-------------------- ------------------ ------------------- --------------------- ----------------------------- ------------------------------- --
log(L$_{H\alpha})$ \# Sources$^{1}$ log($\Phi$)$^{1}$ log($\Phi$)$^{1,2}$ log($\Phi$) final$^{1,2,3}$ $\Delta$log($\Phi$)$^{1,2,3}$
(erg s$^{-1}$) (Mpc$^{-3}$) (Mpc$^{-3}$) (Mpc$^{-3}$) (Mpc$^{-3}$)
41.7 (41.65–41.75) 101 $-$2.25 $-$1.77 $-$1.77 0.14
41.8 (41.75–41.85) 133 $-$2.13 $-$1.78 $-$1.74 0.10
41.9 (41.85–41.95) 132 $-$2.13 $-$1.88 $-$1.83 0.10
42.0 (41.95–42.05) 110 $-$2.20 $-$2.04 $-$1.96 0.10
42.1 (42.05–42.15) 79 $-$2.35 $-$2.23 $-$2.12 0.11
42.2 (42.15–42.25) 69 $-$2.41 $-$2.32 $-$2.24 0.11
42.3 (42.25–42.35) 39 $-$2.66 $-$2.59 $-$2.48 0.14
42.4 (42.35–42.45) 24 $-$2.87 $-$2.81 $-$2.68 0.18
42.5 (42.45–42.55) 14 $-$3.10 $-$3.06 $-$2.92 0.22
42.6 (42.55–42.65) 9 $-$3.30 $-$3.26 $-$3.11 0.27
42.7 (42.65–42.75) 8 $-$3.50 $-$3.45 $-$3.35 0.28
42.9 (42.75–43.05) 6 $-$4.05 $-$4.04 $-$4.02 0.37
-------------------- ------------------ ------------------- --------------------- ----------------------------- ------------------------------- --
\[mass\_table\]
### Extinction Correction
It is well-known that the H$\alpha$ emission line is not immune to dust extinction, although it is considerably less affected than Ly$\alpha$ or the UV continuum. Measuring the extinction for each source can in principle be done by several methods, ranging from spectroscopic analysis of Balmer decrements to a comparison between H$\alpha$ and far-infrared determined SFRs. For now, however, a conservative A$_{H\alpha}=1$ mag is used, which is the same correction adopted in most of the similar studies done before [e.g. @2003ApJ...586L.115F; @2007ApJ...657..738L; @2008arXiv0805.2861G]. This same correction is also applied to all other datasets being compared here, when possible, in order to give consistency between all studies and search for evolution. §\[aign\] shows that this correction should not be too far from the actual mean extinction for the entire sample.
### Filter profile corrections {#profil_filt}
The narrow-band filter transmission function is not a perfect top-hat (as assumed earlier), and thus the real volume probed varies as a function of intrinsic luminosity: luminous H$\alpha$ emitters will be detected over a larger volume than the fainter ones because they can be detected in the wings of the filters (although they will be detected as fainter sources in these cases). Low luminosity sources will only be detected in the central regions of the filter and thus the effective volume will be smaller.
In order to correct for this when deriving the H$\alpha$ luminosity function, a further set of simulations was run. Firstly, the luminosity function was computed with the corrections described above and the best fit was assigned. This was then assumed to be the true luminosity function, allowing the generation of a set of $\sim 10^5$ H$\alpha$ emitters with a flux distribution given by the measured luminosity function, but spread evenly over the redshift range $z=0.81$–0.87 (assuming no cosmic structure variation or evolution of the luminosity function over this narrow redshift range). The top-hat filter model was then confirmed to recover the input luminosity function perfectly. Next, the true filter profiles were used to study the recovered luminosity function. These simulations showed that the number of brighter sources is underestimated relative to the fainter sources. A mean correction factor between the input luminosity function and the one recovered (as a function of luminosity) was then used to correct each bin. The simulation was run again with the new luminosity function, confirming that the recovered luminosity function is very similar to the input luminosity function.
The filter profiles were also checked against the spectroscopic redshift distribution from $z$-COSMOS DR2. By distributing the artificial H$\alpha$ emitters to reproduce the luminosity function, one can predict the redshift distribution of observed sources for each filter. Those matched well the redshift distribution of the $z$-COSMOS emission line sample, although this sample only contains $\sim100$ emitters at the moment.
### Fully corrected luminosity function {#lf}
The final $z=0.84$ H$\alpha$ luminosity function is presented in Figure \[corr\_lum\_f\] and in Table \[mass\_table\]. The raw luminosity function, without correcting for incompleteness and filter profile biases (correcting only for \[N[ii]{}\] contamination and $A_{H\alpha}=1$ mag of extinction), is also shown. The errors are Poissonian in each bin, combined with an uncertainty on the correction factor assumed to be 10% of the applied correction. Luminosity functions are also computed separately for the two observed fields (Figure \[corr\_lum\_f\]) with these being compared with the combined luminosity function. Both luminosity functions are generally consistent within the errors, although there seems to be a slightly higher density at $z=0.84$ in COSMOS. The on-sky distribution of the H$\alpha$ emitters in the COSMOS and UDS fields is far from homogeneous: they are highly clustered and several high-density regions can be found (a detailed clustering analysis is underway and will be published in Sobral et al. in preparation). Assuming a standard angular correlation function for the H$\alpha$ emitters, parameterised as $w(\theta) = A \theta^{-0.8}$ (with $\theta$ in arcsec), then for a value of $A \approx 1$ (typical for galaxies like these), and following [@1975ApJ...196..647P], the cosmic variance in the number of emitters expected in a 1.3 deg$^2$ sky area due to clustering is estimated to be roughly double the pure Poissonian errors. This leads to (1$\sigma$)uncertainties of $\sim12$% on the total number of detected H$\alpha$ emitters; the corresponding uncertainties for the individual fields are $\sim$15% for UDS and $\sim$13% for COSMOS. These are fully consistent with the actual differences in number densities found. Even for an extremely clustered population, with $A \sim 10-15$, the source count variations would not be larger than $\sim
20$% on fields of this size. Given this, and the consistency between the COSMOS and UDS fields (and also the results of Villar et al 2008), it is safe to say that the results derived are largely robust against cosmic variance.
{width="8.2cm"}
{width="8.2cm" height="6.3cm"}
{width="17.4cm"}
Also plotted in Figure \[corr\_lum\_f\] are other published H$\alpha$ luminosity functions at different redshift; these demonstrate a significant evolution with redshift, linked to an increase of the number density of both faint and bright emitters at least up to $z\sim1$. To quantify this evolution a Schechter function is fitted to the combined $z=0.84$ luminosity function, which yields: $\phi^*=10^{-1.92 \pm 0.10}$Mpc$^{-3}$, $\alpha=-1.65\pm0.15$ and $L^*=10^{42.26 \pm 0.05}$ergs$^{-1}$. These parameters, and a comparison with the best fits for COSMOS and UDS separately, can be found in Table 5. The best fit parameters indicate a steeper faint end slope at $z=0.84$ than the canonical $\alpha=-1.35$ usually assumed; similarly steep faint-end slopes for the luminosity function of star-forming galaxies have recently been found from UV studies [e.g. @2008ApJ...686..230B]. These authors have argued for an increase in the faint-end slope of star-forming galaxies with redshift. However, evidence for a steep faint-end slope in the H$\alpha$ luminosity function ($\alpha\sim-1.6$) has been found even at $z=0.08$ by Ly et al. (2007) as well as in other H$\alpha$ studies at higher redshifts, $z\sim0.2$–1.3 [e.g. @2000AJ....120.2843H; @2003ApJ...586L.115F; @2008AJ....135.1412D]. Figure \[param\_evo\] shows the best-fit $\alpha$ for H$\alpha$ luminosity functions derived at $z=0.02$–1.3. This reveals no clear evidence of evolution of the faint-end slope with cosmic time; indeed, rather than having a random distribution, the values of $\alpha$ seem to cluster into two groups with $\alpha \sim-1.6$ or $\alpha \sim-1.3$. The origin of these discrepant results is unclear, but cosmic variance may play an important role.
Figure \[param\_evo\] also shows the variation of $L^*$ and $\phi^*$ of the H$\alpha$ luminosity function out to $z\sim2$. To derive these values, only LFs of H$\alpha$ surveys which were fitted with the “canonical" $\alpha$=$-$1.35 or that had their data points published (those were then refitted by fixing $\alpha$=$-$1.35) were used, in order to reduce degeneracies and allow a direct comparison between results. It should be noted that some degeneracy will still remain between L$^*$ and $\phi^*$ values. Figure \[param\_evo\] shows a strong evolution in $L^*$, increasing by at least an order of magnitude from the local Universe (Gallego et al. 1995) to $z=2.23$ (Geach et al. 2008). The evolution of $\phi^*$ is somewhat different: while it appears to increase from $z=0$ up to $z=0.84$ (this work and Villar et al. 2008) by one order of magnitude, it would then need to fall at $z>1$ to be consistent with the higher redshift data of Yan et al. (1999) and Geach et al. (2008). Note that the increased value of $\phi^*$ at $z=0.84$ does not arise just due to the degeneracy between $\phi^*$ and $L^*$: a value of $\phi^*=10^{-2.7}$Mpc$^{-3}$ for the current data-set can be strongly rejected, having a probability $<10^{-6}$. The evolution of $L^*$ and $\phi^*$ is therefore revealing important details of the evolution of the H$\alpha$ luminosity function: from $z\sim0$ to $z\sim1$ it seems to be driven by an increase in the number density of both bright and faint emitters, with an increasing population of bright emitters at higher redshift then being responsible for a stronger evolution in $L^*$. These results are also consistent with studies done using 24 $\mu$m data [e.g. @2007ApJ...660...97C]. The evolutionary trends do not change if $\alpha$ is fixed at a higher value ($\alpha=-1.65$ for example).
The star formation rate density at $z=0.84$
-------------------------------------------
### AGN contamination {#aign}
The H$\alpha$ luminosity function previously derived used all the H$\alpha$ emitters from the survey, and while most of such sources are likely to be star-forming galaxies, some of these can also be AGN. Spectra from $z$-COSMOS DR2 were used to explored this. A visual inspection of the 93 available spectra was done to confirm additional emission lines (\[O[ii]{}\]3727, \[O[iii]{}\]5007 and H$\beta$) and the assigned redshift. The line fluxes were then measured using an [idl]{} script. Although the comparison of those lines with the H$\alpha$ line fluxes is influenced by many factors (e.g. H$\alpha$/\[N[ii]{}\] ratio, exact location of the H$\alpha$ line within the filter profile, fraction of emission line light falling into $z$-COSMOS slit) it is noteworthy that both the mean ratio of H$\alpha$/\[O[ii]{}\]3727$=2.27$ and the ratio of H$\alpha$/H$\beta=4.16$ are consistent with an H$\alpha$ extinction of $\sim1$mag or slightly higher.
![Line ratios from the $z$-COSMOS spectra of the $z=0.845$ H$\alpha$ sample. These show that the great majority of the sample is composed of star-forming galaxies (82%), as expected, with 11% showing evidence for being AGN contaminants and 7% being unclassified. The red curves represent the maximum line ratios for a star-forming galaxy (from OB stars with effective temperatures of 60000K (solid line) and 50000K (dashed line)). \[agn\]](./figs/SF_vs_AGN_SN3.jpg){width="8.2cm"}
{width="8.2cm"}
{width="8.2cm"}
In order to estimate the AGN contamination, the \[O[ii]{}\]3727/H$\beta$ and \[O[iii]{}\]5007/H$\beta$ line ratios were used; these have been widely used to separate AGN from star-forming galaxies [e.g. @1997MNRAS.289..419R]. Only spectra with all lines being detected at S/N$>3.0$ were used, which results in a sample of 28 galaxies, mainly due to the low S/N at longer wavelengths where \[O[iii]{}\]5007 and H$\beta$ are found. Figure \[agn\] shows data-points for the line ratios, while the curves represent maximum line ratios for a star-forming galaxy (from OB stars with effective temperatures of 60000K and 50000K). From the sample of 28 H$\alpha$ emitters, 23 seem to be clear star-forming galaxies, while 3 are likely to be AGN contaminants. A $\sim15$% AGN contamination is thus estimated, consistent with that found in other H$\alpha$ studies. The AGN are found to have H$\alpha$ fluxes typical of the rest of the sample.
### Star formation rate density
The observed H$\alpha$ luminosity function can be used to estimate the average star formation rate density, $\rho_ {SFR}$, at $z=0.84$. To do this, the standard calibration of Kennicutt (1998) is used to convert the extinction-corrected H$\alpha$ luminosity to a star formation rate:
$${\rm SFR}({\rm M}_{\odot} {\rm year^{-1}})= 7.9\times 10^{-42} ~{\rm L}_{\rm H\alpha} ~ ({\rm erg s}^{-1}).$$
This assumes continuous star formation, Case B recombination at $T_e = 10^4$K and a Salpeter initial mass function ranging from 0.1–100M$_{\odot}$. All measurements of $\rho_{SFR}$ include a correction of 15% for AGN contamination and an extinction correction A$_{H_\alpha}=1$mag, except where the authors only presented their own extinction corrected luminosity function.
In §\[lf\] a significant evolution in the observed H$\alpha$ luminosity function was observed. The left panel of Figure \[sfr\_evo\] shows how this translates into an evolution in $\rho_{SFR}$ as a function of redshift, for luminosity functions which have been integrated down to L$_{\rm H\alpha}>10^{41.5}$ergs$^{-1}$ (the limit of this survey). The measurement at $z=0.84$ presented in this study ($0.15\pm0.02$M$_{\odot}$yr$^{-1}$Mpc$^{-3}$) demonstrates a strong rise in $\rho_{SFR}$, when compared to the local Universe [@galo; @2003ApJ...591..827P; @2007ApJ...657..738L] and low redshift measurements [e.g. @1998ApJ...495..691T; @sullivan01; @2008AJ....135.1412D; @2008arXiv0807.0101M; @2008MNRAS.383..339W; @2008ApJS..175..128S; @2009arXiv0902.2064S], as suggested by other smaller surveys done at similar redshifts [e.g. @alpha1; @2008ApJ...677..169V]. This rise seems to be slightly steeper than $\rho_{SFR}\sim(1+z)^4$. When compared to higher redshift (e.g. Geach et al. 2008), the observations also support a flattening in $\rho_{SFR}$ around $z\sim 1$, up to at least $z=2.23$. A rise and subsequent flattening of the star formation rate density out to $z\sim2$ has therefore been accurately measured using a single star formation tracer. Cosmic evolution of dust reddening corrections may alter the results slightly but would have to be very strong to change the overall conclusions.
Figure \[sfr\_evo\] also presents the same evolution, but now integrating the entire luminosity function. Caution should be used in interpreting this figure as it involves extrapolating all the luminosity functions and it is critically dependent on the assumed faint-end slope. For this study, for example, $\rho_{SFR}=0.37\pm0.18$M$_{\odot}$yr$^{-1}$Mpc$^{-3}$ using the measured value of $\alpha=-1.65$, but if one adopts $\alpha=-1.35$, $\rho_{SFR}$ is reduced to $0.28\pm0.08$M$_{\odot}$yr$^{-1}$Mpc$^{-3}$. Measurements obtained from a set of other star-formation indicators compiled by Hopkins (2004) are also shown for comparison (corrected by a common extinction factor consistent with the H$\alpha$ extinction correction applied here). This confirms the same rise seen using only H$\alpha$ and the possible flattening at $z\sim1$. However, it appears the H$\alpha$ measurements at $z>0.7$ are finding slightly higher values of $\rho_{SFR}$ when compared to the average of all other measurements, while at $z<0.3$ they measure slightly lower values. This may reflect a number of systematic errors or factors such as evolution in the typical reddening in the star forming population.
The morphology of H$\alpha$ emitters
------------------------------------
### Morphology with [*HST*]{} imaging {#morfo}
The COSMOS field has sensitive [*Hubble Space Telescope*]{} ([*HST*]{}) ACS F814W coverage which provides detailed morphological information on the H$\alpha$ sample. These data are used to study the morphologies of the selected star-forming galaxies. This morphological information can be greatly enriched with colour information, and thus the ACS images are combined with deep Subaru data in $Brz$ bands to produce a pseudo-true colour image. The Subaru images are first registered and transformed to match the [*HST* ]{} images using [iraf]{} and [Python ]{} scripts. The [*HST*]{} image is then used to define the luminance of the true colour image. In this way colour and morphological information were derived on $\sim 500$ H$\alpha$ emitters with sufficient detail that a visual morphology analysis could be undertaken. Figure \[thumbs\_int1\] presents some examples of these.
{width="16.6cm"}
-------------- -------- ----------- ------------ -------------- ------------ ------------- ------------------- -----------
Visual ZEST ZEST ZEST ZEST Visual Visual Visual Visual
Class Early Disks Irregulars Unclassified Total Non-mergers Potential Mergers Mergers
Early 7 3 0 4 14 (3%) 12 0 2
Disks 2 301 15 63 381 (80%) 297 33 51
Irregulars 0 10 44 14 68 (14%) 1 7 60
Unclassified 0 2 0 12 14 (3%) 0 0 0
Total 9 (2%) 316 (67%) 59 (12%) 93 (19%) 477 (100%) 310 (67%) 40 (9%) 113 (24%)
-------------- -------- ----------- ------------ -------------- ------------ ------------- ------------------- -----------
\[morphcomp\_table\]
All of the galaxies within the sample were visually classified. In order to compare those results with morphologies obtained in an automated way (by [zest]{}, kindly supplied by Claudia Scarlata, [@2007ApJS..172..406S]), 4 morphological classes were used: 1) Early-types, 2) Disks/spirals, 3) Irregulars, and 4) Unclassified. Table \[morphcomp\_table\] shows the comparison between [zest]{} and the visual classification, while Figure \[thumbs\_int1\] shows examples of each morphological type. In general, a very good agreement between the visual classifications and [zest]{} was obtained. The main difference is that visual classification was able to reduce the number of unclassified galaxies. Also, visually, it was possible to improve the distinction between disks and irregulars by using the colour information – distinguishing between multiple bulges and bright spiral arms – which resulted in a sightly lower fraction of irregulars in the visual classification.
{width="8.2cm"}
{width="8.2cm" height="6.3cm"}
{width="8.2cm"}
{width="8.2cm" height="6.3cm"}
Due to a considerable number of sources showing evidence of merging activity, the sample was also classified independently into merger classes. A source is classified as a merger when: 1) it presents a clearly disturbed morphology or disturbed disk which is inconsistent with being that of a normal disk galaxy; or 2) the source presents more than 1 bright point-like source and the colour information is inconsistent with one of those being a spiral arm; or 3) there are two or more galaxies which are very close ($<15$kpc). By applying these criteria, a visual classification was done, where each source was classified at least twice. In the end, 3 merger classes where used to classify all sources: the consistent non-mergers (non-mergers); those galaxies classified at least once as probable mergers (potential mergers); and galaxies always classified as mergers (mergers). The distribution of mergers within the previous morphological classifications and the total numbers can be found in Table \[morphcomp\_table\] and examples of classified galaxies can also be found in Figure \[thumbs\_int1\].
From a total of 477 H$\alpha$ emitters, 381 are disks (80%), with 68 being classified as irregulars (14%), 14 are early-types (3%) and a total of 13 (3%) are unclassified as these are too faint. Furthermore, 24% of the sample seems to be populated by clear mergers, with the total merger fraction being estimated as $28\pm4$%; almost all irregulars fall into this class.
### Morphology–H$\alpha$ luminosity relations {#morfo}
The left panel of Figure \[morphdep\] presents the fraction of galaxies classified into each morphology as a function of H$\alpha$ luminosity. There is a clear evolution of morphological type with increasing H$\alpha$ luminosity. While at low star-formation rates, disks dominate the sample completely ($>85$%), at higher H$\alpha$ luminosities ($L>L^*$) irregulars become more significant, reaching 100% in the highest luminosity bin. With the large sample presented in this work it is possible to derive independent luminosity functions for each of those morphological classes. Those can be seen in Figure \[morphLF\] with the best fit Schechter function parameters tabulated in Table 5. They illustrate well the different contributions to the total H$\alpha$ luminosity function by disks and irregulars. Irregulars present a remarkably flat luminosity function, only falling at the highest luminosities, while disks demonstrate a steeper faint end and a much lower space density at the bright end. Overall disks are the dominant contributors to the total $\rho_{SFR}$, with a large number of galaxies producing stars at rates $<10$M$_{\odot}$yr$^{-1}$.
As mentioned before, almost all of the irregulars show clear evidence of merging activity. In fact, separating non-mergers from likely mergers (mergers, together with potential mergers weighted by 0.5), a very similar behaviour is found (Figure \[morphdep\]). While non-mergers dominate at faint luminosities, the contributions from the two populations cross over at $L\sim L^*$, with mergers dominating the bright-end of the luminosity function. The luminosity functions for non-mergers and potential mergers are presented in Figure \[morphLF\]. This reveals that non-mergers present the typical disk luminosity function found before, but mergers have a steeper luminosity function than the irregulars. This means that while irregulars on their own are only important in the bright end of the LF, mergers seem to play a dominant role at the bright end (accounting for$\sim60$% of the $\rho_{SFR}$ there) together with a non-negligible contribution even at the faintest luminosities. Overall, mergers account for $\sim20$% of the total $\rho_{SFR}$ at $z=0.84$.
The relations which have been presented were found to be very robust, with the same results being obtained regardless of the use of the visual or [zest]{} classifications, and independently of the \[N[ii]{}\] correction – a constant correction also produces the same dependences.
### Redshift evolution of the morphology relations {#morfo}
One can compare the morphological mix for the H$\alpha$ sample at $z=0.84$ with that found at lower redshifts. In the local Universe, the morphologies of the H$\alpha$ emitters seem to be somewhat different, with visual morphologies from indicating that disks are 88% of the sample, while early-type galaxies increase their importance (9%) and irregulars drop to only 3%. At $z=0.24$ the narrow-band H$\alpha$ emitters of Shioya et al. (2008) in the COSMOS field can be more directly compared to the $z=0.84$ sample, using [zest]{} morphologies for both (although noting that the rest-frame wavelength of the images is different between the two studies – rest-frame $B$ band at $z=0.84$ and rest-frame $R$ band at $z=0.24$ – which could introduce a small bias). In the [zest]{} classifications at $z=0.24$, disks are still dominant (89%), $\sim$10% of the galaxies have an irregular morphology, and elliptical galaxies account for only 1% of the sample. These results seem to show that while disks are always the dominant population ($>80$%) up to $z\sim1$, there is a significant increase in the irregular fraction from 3% at $z=0$ to 10% at $z=0.24$ and 15% at $z=0.84$.
![Morphology class fractions as a function of H$\alpha$ luminosity for $z=0.24$ (Shioya et al. 2008) and this study. These results find the same morphological dependence at both epochs consistent with an evolution in $L^*$ ($\log(L)\sim0.4$). Morphologies were obtained in an automated way by [zest]{} (Scarlata et al. 2007) for both samples. \[morphdepzlow\]](./figs/correct_scale.jpg){width="8.2cm"}
--------------------------- ------------------ ---------------- ------------------ ------------------------- ---------------------------------- ---------------------------------- --
Sample log$\phi^*$ log$L^*$ $\alpha$ log$\rho_{L \ >41.5}$ $\rho_{SFR \ >41.5}$ $\rho_{SFR \ total}$
(Mpc$^{-3}$) (ergs$^{-1}$) (ergs$^{-1}$Mpc$^{-3}$) (M$_{\odot}$yr$^{-1}$Mpc$^{-3}$) (M$_{\odot}$yr$^{-1}$Mpc$^{-3}$)
Total: COSMOS+UDS $-$1.92$\pm$0.10 42.26$\pm$0.05 $-$1.65$\pm$0.15 40.37$\pm$0.04 0.15$\pm$0.01 0.37$\pm$0.15
UDS $-$1.98$\pm$0.16 42.25$\pm$0.10 $-$1.70$\pm$0.22 40.32$\pm$0.06 0.14$\pm$0.03 0.39$\pm$0.18
COSMOS $-$1.90$\pm$0.12 42.28$\pm$0.07 $-$1.68$\pm$0.18 40.44$\pm$0.05 0.18$\pm$0.03 0.45$\pm$0.22
COSMOS: Disks $-$1.81$\pm$0.13 42.12$\pm$0.06 $-$1.65$\pm$0.22 40.27$\pm$0.05 0.13$\pm$0.02 0.41$\pm$0.18
COSMOS: Irregulars $-$2.90$\pm$0.23 42.58$\pm$0.14 $-$1.27$\pm$0.21 39.70$\pm$0.10 0.03$\pm$0.01 0.04$\pm$0.01
COSMOS: Non-mergers $-$1.96$\pm$0.14 42.16$\pm$0.09 $-$1.71$\pm$0.21 40.19$\pm$0.06 0.10$\pm$0.02 0.33$\pm$0.20
COSMOS: Potential Mergers $-$2.63$\pm$0.13 42.50$\pm$0.10 $-$1.47$\pm$0.20 39.94$\pm$0.09 0.06$\pm$0.01 0.08$\pm$0.04
--------------------------- ------------------ ---------------- ------------------ ------------------------- ---------------------------------- ---------------------------------- --
\[lf\_morph\]
The [zest]{} classifications for the Shioya et al. (2008) sample can also be compared with those for the $z=0.84$ sample to see if there is any evolution in the morphology–H$\alpha$ luminosity relations. Thus Figure \[morphdepzlow\] presents the comparison between $z=0.24$ and $z=0.84$ with the luminosities scaled by $L^*$. A very similar variation in morphological mix with H$\alpha$ luminosity is found at $z=0.24$, with the irregular fraction increasing from $\sim10$% to $\sim35$% with increasing luminosity and with the disk fraction having the opposite behaviour, decreasing from $\sim 90$% to $\sim60$%. The relatively good agreement between the data at these two different redshifts and the fact that the switch-over luminosity appears to be at the same $L/L^*$ at both redshifts seems to point towards a rather simple $L^*$ evolution driving the morphology–H$\alpha$ luminosity relation. However, the Shioya et al. (2008) sample is not able to probe the brightest sources and thus there is no low redshift data to directly compare the complete dominance of irregulars at $L>3 L^*$.
The disk fraction at $z=0.24$ appears to be slightly lower than the disk fraction at $z=0.84$ (for the bins which can be directly compared). Curiously, this is due to an apparent rise in the fraction of early-types. However, this might be simply reflecting classification errors between the two epochs. At least some of the very faint galaxies which have been unclassified at $z=0.84$ (even after visual classification) are likely to be early-types which are just too faint to be seen at $z=0.84$, but sufficiently bright at $z=0.24$ to be classified.
Conclusions
===========
Deep near-infrared narrow-band imaging has been obtained, allowing the selection of line emitting galaxies down to an effective flux limit of F$_{{\rm H}\alpha}\sim10^{-16}$ergs$^{-1}$cm$^{-2}$. This has resulted in the largest and deepest survey of emission line selected star forming galaxies at $z\sim1$, detecting 1517 potential line emitters over an area of $\sim 1.4\deg^2$ in the COSMOS and UDS fields, with 1370 having quality multi-wavelength data available in a region of $1.3\deg^2$. For H$\alpha$ emission line objects this survey probes a co-moving volume of $\sim1.8\times10^5$Mpc$^3$ at $z=0.84$ down to a star formation rate of $\sim3$M$_{\odot}$yr$^{-1}$ (with an $A_{H\alpha}=1$mag extinction correction).
Photometric redshifts for COSMOS and UKIDSS UDS (Mobasher et al. 2007; Cirasuolo et al. 2008) clearly show that the majority of the selected emitters are H$\alpha$ emitters at $z\sim0.85$ with a secondary population of \[O[iii]{}\]5007/H$\beta$ emitters at $z\sim1.4$–1.5. Almost 120 emitters were confirmed spectroscopically, from which 93 are H$\alpha$ at $z=0.84$. The contamination within the sample of emitters is estimated to be lower than $\sim 6$%, and the contamination within the H$\alpha$ sample is much lower ($\sim 0$%) based on the current samples. A total of 743 H$\alpha$ selected emitters (based on their photometric and spectroscopic redshifts) was obtained in the two fields. These were used to calculate the luminosity function after correcting for \[N[ii]{}\] flux contamination, extinction, incompleteness and filter profile biases. The morphologies of these emitters were also investigated. The main conclusions of this work are:
- The H$\alpha$ luminosity function at $z=0.84$ found is well fitted by a Schechter function with $\phi^*=10^{-1.92\pm0.10}$Mpc$^{-3}$, $\alpha=-1.65\pm0.15$ and $L^*=10^{42.26\pm0.05}$ergs$^{-1}$. This demonstrates a strong evolution in the H$\alpha$ luminosity function compared to lower redshifts and agrees reasonably well with previous smaller studies at $z\sim 1$.
- The evolution of the H$\alpha$ luminosity function can be described by an increase in $\phi^*$ and $L^*$, at least out to $z\sim1$, with L$^*$ then continuing to rise up to $z\sim2$ but $\phi^*$ peaking around $z\sim1$ and then decreasing at higher redshifts.
- The integrated luminosity function is used to estimate the cosmic star formation rate density, ($\rho_{SFR}$) at $z=0.84$: $0.15\pm0.01$M$_{\odot}$yr$^{-1}$Mpc$^{-3}$ (corrected for 15% AGN contamination and integrated down to 2.5M$_{\odot}$yr$^{-1}$).
- An accurate determination of the cosmic evolution of $\rho_{SFR}$ has been made using a single star formation tracer (H$\alpha$) from $z=0$ to $z=2.23$. This shows a strong rise up to $z\sim1$ followed by a flattening out to $z\sim2.2$.
- H$\alpha$ emitters at $z=0.84$ are mostly morphologically classed as disks in the rest-frame $B$-band ($82\pm3$%). Irregulars account for $15\pm2$% of the sample and early-type galaxies are only $3\pm1$%. Apparent mergers are a significant fraction of the sample ($28\pm4$%).
- A strong morphology–H$\alpha$ luminosity relation is found at $z$=$0.84$, with the fraction of irregulars rising steadily with luminosity and the fraction of disks falling. Mergers/non-mergers present the same behaviour, and $L^*$ (from the total sample) seems to define a critical switch-over luminosity between the two populations.
- Mergers dominate the bright end of the total H$\alpha$ luminosity function at $z=0.84$ and $\sim20$% of the total $\rho_{SFR}$ is due to their activity.
- A similar morphology–H$\alpha$ luminosity relation is found at lower redshift ($z=0.24$), consistent with a simple L$^*$ evolution.
These results suggest that the evolution of both the H$\alpha$ luminosity function and $\rho_{SFR}$ change significantly in nature beyond $z\sim1$, and that it is entirely plausible that this is driven by the different evolutionary behaviour of two different populations of star forming galaxies. Out to $z\sim1$, the integrated $\rho_{SFR}$ at each redshift is produced predominantly by disk galaxies; it is therefore the evolution of these disk galaxies, rather than that of major mergers, which drives the strong decrease in the cosmic star formation rate density from redshift one to the current epoch – this is in line with other recent results [e.g. @2008ApJ...672..177L]. At these redshifts, the evolution in $\rho_{SFR}$ arises predominantly from an evolution in the characteristic space density of the H$\alpha$ luminosity function ($\phi^*$; which evolves by more than an order of magnitude between $z=0$ and $z=0.84$), rather than a strong evolution in $L^*$. With their relatively quiescent star formation activity, the disk galaxies dominate the H$\alpha$ luminosity function at low luminosities, and are thus responsible both for the $\phi^*$ evolution and for setting the faint-end slope of the luminosity function.
In contrast, irregular and merging galaxies appear to dominate the H$\alpha$ luminosity function above $L^*$, at $z=0.84$ (and also at $z=0.24$), being responsible for more than 50% of the non-extrapolated $\rho_{SFR}$ (integrated down to 2.5M$_{\odot}$yr$^{-1}$). The evolution of these systems controls the bright end of luminosity function, and thus the cosmic evolution of $L^*$. The continued strong evolution of $L^*$ between $z=0.84$ and $z =2.23$, and the decrease in $\phi^*$, suggests an increasing importance of merger-driven star formation activity beyond $z\sim1$, especially as the irregular H$\alpha$ luminosity function seems to be very similar to the one found at $z=2.23$ by Geach et al. (2008) when taking into account an $L^*$ evolution. This is consistent with recent results such as [@2009arXiv0903.3035S]. The completed HiZELS survey will provide statistically significant samples of H$\alpha$ emitters at $z=1.47$ and $z=2.23$ and so provide a direct test of this suggestion.
Acknowledgments {#acknowledgments .unnumbered}
===============
The authors thank the referee for relevant and detailed suggestions that have improved this work. DS would like to thank FCT for financial support (grant: SFRH/BD/36628/2007). PNB & IRS acknowledge the Royal Society. JEG thanks the U.K. Science and Technology Facility Council (STFC) and KC acknowledges for a STFC Fellowship. The authors would also like to thank Claudia Scarlata for kindly supplying [zest]{} morphologies for the COSMOS samples, the ESO Large Program 175.A-0839 ($z$-COSMOS), Jesus Gallego and Pablo P[é]{}rez-Gonz[á]{}lez for helpful discussions, and Andy Adamson, Luca Rizzi and Tim, Thor and Jack for the support on the UKIRT telescope.
\[lastpage\]
[^1]: E-mail: [email protected]
[^2]: Based on observations obtained with the Wide Field CAMera (WFCAM) on the United Kingdom Infrared Telescope (UKIRT)
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'J.F. Jardine'
bibliography:
- 'spt.bib'
date: 'July 19, 2016'
nocite:
- '[@pathcat]'
- '[@Joyal-quasi-cat]'
- '[@Mis-path]'
title: Complexity reduction for path categories
---
Introduction {#introduction .unnumbered}
============
A finite cubical complex $K$ is a subobject $K \subseteq \square^{n}$ of a standard $n$-cell in the category of cubical sets.
The object $\square^{n}$ is represented by the poset $\mathcal{P}(\underline{n})$ of subsets of the set $\underline{n} = \{
1,2, \dots ,n \}$. This poset is an object of the box category $\square$ that defines cubical sets (see, for example, [@J40]). The complex $K$ is defined by a list of non-degenerate cells $\sigma:
\square^{k} \subseteq \square^{n}$. These cells can be identified with poset inclusions $[A,B] \subseteq \mathcal{P}(\underline{n})$ of intervals, where $$[A,B] = \{ F\ \vert\ A \subseteq F \subseteq B \}$$ where $A \subseteq B$ are subsets of $\underline{n}$.
As such, $K$ is a list of intervals $[A,B] \subseteq
\mathcal{P}(\underline{n})$ which is closed under taking subintervals.
Finite cubical complexes are the higher dimensional automata of geometric concurrency theory. In that setting, the vertices of a cubical complex $K$ model the states of a concurrent system, and its $k$-cells represent (where possible) the simultaneous action of $k$ processors. The cells of the ambient $n$-cell which are not in $K$ represent constraints on the system.
The main object of study associated to $K$ in this form of concurrency theory is its collections of execution paths. These paths are the morphisms of the [*path category*]{} $P(K)$.
The path category functor is now well known — it is also called the fundamental category and denoted by $\tau_{1}(K)$ in the higher categories literature [@Joyal-quasi-cat].
The emphasis in concurrency theory is different, and is completely concerned with giving exact specifications of path categories $P(K)$ in the geometric setting described above. Techniques leading to explicit, algorithmic calculations of path categories form the subject of this paper.
The [*triangulation*]{} $\vert K \vert$ of the finite cubical complex $K$ is a finite simplicial complex that is defined by “putting in the missing edges”. More explicitly, $$\vert \square^{n} \vert = (\Delta^{1})^{\times n} = B\mathcal{P}(\underline{n}),$$ is the nerve of the poset $\mathcal{P}(\underline{n})$, and $\vert K
\vert$ is constructed by gluing together such objects along the incidence relations for the cells of $K$.
The path category functor $X \mapsto P(X)$ for simplicial sets $X$ is most succinctly defined to be the left adjoint of the nerve functor. The path category construction for cubical sets is a specialization of this functor, and we can write $$P(K) := P(\vert K \vert)$$ for cubical complexes $K$.
In practice, the objects of $P(K)$ are the vertices of $K$, and the morphisms are equivalence classes of paths in $1$-cells, modulo commutativity conditions that are defined by $2$-cells.
Similarly, the path category $P(L)$ of a finite simplicial complex $L$ has the vertices of $L$ as objects, and has morphisms given by equivalence classes of paths in $1$-simplices, modulo commutativity conditions that are defined by $2$-simplices.
There is an algorithm for computing $P(L)$ for finite simplicial complexes $L \subseteq \Delta^{n}$ that arises from a $2$-category $P_{2}(L)$ that is defined by the simplices of $L$, and for which $P(L)$ is the path component category of $P_{2}(L)$ in the sense that there is a bijection $$P(L)(x,y) \cong
\pi_{0}(P_{2}(L)(x,y))$$ for all vertices $x,y$. The $2$-category $P_{2}(L)$ is defined in [@pathcat].
The algorithm can be summarized as follows:
- Restrict to the $2$-skeleton $\sk_{2}(L)$ of $L$.
- Find all paths (strings of non-degenerate $1$-simplices) $$\omega: v_{0} \xrightarrow{\sigma_{1}} v_{1} \xrightarrow{\sigma_{2}} \dots \xrightarrow{\sigma_{k}} v_{k}$$ in $L$.
- Find all morphisms in the category $P_{2}(L)(v,w)$ for all vertices $v < w$ in $L$ (ordering in $\Delta^{n}$).
- Find the sets of path components for all categories $P_{2}(L)(v,w)$.
This algorithm is the [*path category algorithm*]{}. It has been coded in C and Haskell by M. Misamore — Misamore’s code is published on github.com and hackage.haskell.org. The original test of concept was written by G. Denham in Macaulay 2.
Except for the first step, which is due to a basic result for path categories [@pathcat] that also appears in Lemma \[lem 1\] below, the algorithm is brute force. It works well for toy examples, but it is easy to generate simple examples which output very large lists of morphism sets.
The “necklace” $L \subseteq \Delta^{40}$ be the subcomplex $$\label{eq 1}
\xymatrix{
& 1 \ar[dr] && 3 \ar[dr] &&&& 39 \ar[dr] \\
0 \ar[ur] \ar[rr] && 2 \ar[ur] \ar[rr] && 4 & \dots
& 38 \ar[ur] \ar[rr] && 40
}$$ This is 20 copies of the complex $\partial\Delta^{2}$ glued together. It is visually obvious that there are $2^{20}$ morphisms in $P(L)(0,40)$, and the text file list of morphisms of $P(L)$ consumes 2 GB of disk space.
In general, the size of the path category $P(L)$ can grow exponentially with $L$.
Extreme examples aside, various complexity reduction methods have been developed for the path category algorithm, and the purpose of this note is to give an account of these techniques.
The mathematical results of this paper are quite simple. Most of the statements amount to constructions of subcomplexes $K \subseteq L$ such that the induced functor $P(K) \to P(L)$ between path categories is fully faithful.
Explicitly, this means that if $v,w$ are vertices of $K$, then the induced function $P(K)(v,w) \to P(L)(v,w)$ of morphism sets is a bijection. In this case, the morphism set $P(L)(v,w)$ can be computed in the smaller context given by $K$, which can be much simpler computationally.
Most of the time, $K$ is a “full” subcomplex of $L$. Fullness is a general criterion for the induced functor $P(K) \to P(L)$ to be fully faithful. The concept (appearing in Section 1 of this paper) is used repeatedly, for the method of deletions of sources and sinks from a simplicial complex in Section 2, and for deriving Mismore’s method of removing corners from a cubical complex in Section 3.
Section 4, on refinement of cubical complexes, is the opposite in some sense. The idea is that one can use the data that constructs a finite cubical complex $K$ to construct a more complicated object $K_{\alpha}$ in a way that produces a fully faithful functor $P(K) \to
P(K_{\alpha})$. One expects that this idea will be useful for studies of successive approximations of cubical structures.
The last section, Section 5, gives a first, coarse method for parallelizing the path category algorithm for calculating $P(K)$ for a cubical complex $K$. All vertices of $K$ have a size, or cardinality, that they inherit from the ambient cell $\square^{n}$. The resulting size functor can be used to isolate disjoint full subcomplexes, say $A$ and $B$, for which $P(A)$ and $P(B)$ can be computed independently. All paths $u \to v$ of $K$ which start in $A$ and end in $B$ cross a “frontier subcomplex” whose cells define a coequalizer picture (see (\[eq 4\])) that allows one to compute $P(K)(u,v)$ from the path categories $P(A)$ and $P(B)$.
The size functor is also used in Section 4, and it is very likely to have continuing utility. One can think of this functor as a ticking clock, but the relationship between that “clock” and the higher dimensional automaton concept can be a bit fraught.
Basic results
=============
The first step of the path category algorithm involves a direct appeal to the following result:
\[lem 1\] The inclusion $\sk_{2}(X) \subseteq X$ of the $2$-skeleton of a simplicial set $X$ induces an isomorphism of categories $$P(\sk_{2}(X)) \xrightarrow{\cong} P(X).$$
This result follows from the fact that the nerve $BC$ of a small category $C$ is a $2$-coskeleton [@GJ Lem. 3.5], which means that there is a bijection $$\hom(X,BC) \cong \hom(\sk_{2}(X),BC).$$
Lemma \[lem 1\] is a substantial complexity reduction step, in that it means that one can ignore much of the data for a finite simplicial complex $L$ before computing $P(L)$.
We now discuss a concept and result that has appeared in connection with work on homotopy types of categories [@dyn Lem. 4].
Suppose that $L_{0} \subseteq L$ is a subcomplex of a finite simplicial complex $L$. We say that $L_{0}$ is a [ *full subcomplex*]{} of $L$ if the following conditions hold:
- $L_{0}$ is path-closed in $L$, in the sense that, if there is a path $$v=v_{0} \to v_{1} \to \dots \to
v_{n}=v'$$ in $L$ between vertices $v,v'$ of $L_{0}$, then all $v_{i} \in
L_{0}$,
- if all the vertices of a simplex $\sigma \in L$ are in $L_{0}$ then the simplex $\sigma$ is in $L_{0}$.
\[lem 2\] Suppose that $L_{0}$ is a full subcomplex of $L$. Then the functor $P(L_{0})
\to P(L)$ is fully faithful.
Recall that a functor $F: C \to D$ is [*fully faithful*]{} if all induced functions $$f: C(x,y) \to D(f(x),f(y))$$ of morphism sets are bijections.
The proof of Lemma \[lem 2\] follows from the fact that the path category $P(L)$ is constructed by taking the category freely associated to the graph given by the $1$-skeleton $\sk_{1}(L)$, modulo relations defined by $2$-simplices of $L$ [@pathcat]. The conditions imply that every path in $L$ between vertices $v,w$ of $L_{0}$ consists of simplices which are in $L_{0}$, and that all $2$-simplices which define relations of paths in $L$ between $v,w \in
L_{0}$ are also in $L_{0}$.
\[ex 4\] The inclusions $d^{0}: \partial\Delta^{2} \subseteq
\Lambda^{3}_{0}$ and $d^{3}: \partial\Delta^{2} \subseteq
\Lambda^{3}_{3}$ induced by the respective cofaces $\Delta^{2}
\subseteq \Delta^{3}$ both define full subcomplexes.
In the first case, an argument on orientation says that no path in $\Lambda^{0}_{0}$ that starts and ends in the set of vertices $\{
1,2,3 \}$ can pass through the vertex $0$. The second case is similar.
Suppose that $i \leq j$ in $\mathbf{n}$ and suppose that $L \subseteq
\Delta^{n}$. $L[i,j]$ is the subcomplex of $L$ such that $\sigma \in
L[i,j]$ if and only if all vertices of $\sigma$ are in the interval $[i,j]$ of vertices $v$ such that $i \leq v \leq j$. Then $L[i,j]$ is a full subcomplex of $L$.
Suppose that $v \leq w$ are vertices of $L \subseteq
\Delta^{n}$. Let $L(v,w)$ be the subcomplex of $L$ consisting of simplices whose vertices appear on a path from $v$ to $w$. Then $L(v,w)$ is a full subcomplex of $L$, and of $L[v,w]$.
Sources and sinks
=================
A vertex $v$ is a [*source*]{} of $L$ if there are no non-degenerate $1$-simplices $u \to v$ in $L$. The vertex $z$ is a [*sink*]{} of $L$ if there are no non-degenerate $1$-simplices $z \to w$ in $L$.
Every finite simplicial complex $L \subset \Delta^{n}$ has at least one source and one sink. These are the smallest and largest vertices of $L$, respectively, in the totally ordered set of vertices of the ambient simplex $\Delta^{n}$.
Observe that $0$ is a source of $\Lambda^{3}_{0}$ and $3$ is a sink of $\Lambda^{3}_{3}$. The following result formalizes the assertions made in Example \[ex 4\] above:
\[lem 3\] Suppose that $S$ is a subset of the vertices of $L \subseteq \Delta^{n}$ which consists of sources and sinks. Let $L(-S)$ be the subcomplex of $L$ which consists of simplices which do not have a vertex in $S$. Then $L(-S)$ is a full subcomplex of $L$.
Suppose that $v < v'$ are vertices of $L(-S)$ and suppose that the string of $1$-simplices $$v=v_{0} \to v_{1} \to \dots \to v_{n}=v'$$ is a path of $L$ from $v$ to $w$ consisting of non-degenerate $1$-simplices. Then no intermediate object $v_{i}$, $1 \leq i
\leq n-1$ can be a source or a sink. It follows that all $v_{i} \in
L(-S)$.
A simplex $\sigma$ of $L$ is in $L(-S)$ if and only if none of its vertices are in $S$, by definition.
Suppose that $L$ is the complex $$\xymatrix@=12pt{
&&& v_{3} \\
v_{0} && v_{2} \ar[ur] && v_{4} \ar[ul] \\
& v_{1} \ar[ul] \ar[ur]
}$$ The set $S = \{v_{1},v_{3}\}$ consists of sources and sinks, and $L(-S)$ is discrete on the vertices $v_{0},v_{2},v_{4}$. The isolated point $v_{2}$ is a source and a sink for $L(-S)$. Let $S'= \{ v_{2}
\}$. Then $$P(L)(v_{0},v_{4}) = P(L(-S))(v_{0},v_{4}) = P(L(-S)(-S'))(v_{0},v_{4})
= \emptyset.$$
Thus, removing sources and sinks can create new ones. The process of removing sources and sinks relative to a pair of vertices $v,w$ of $L$ must stop, since $L$ is finite.
\[lem 5\] Suppose that $v < w$ in $L$ and that $S$ consists of sources and sinks of $L$ which are distinct from $v$ and $w$. Then $$L(-S)(v,w) = L(v,w).$$
Suppose that $$\sigma: v=v_{0} \to v_{1} \to \dots \to v_{n}=w$$ is a path from $v$ to $w$ in $L$. Then each intermediate vertex $v_{i}$ is neither a source or a sink, and is therefore not in $S$, so that $v_{i} \in L(-S)$. The subcomplex $L(-S)$ is full so that the path $\sigma$ is in $L(-S)$.
Thus, every vertex of $L(v,w)$ is a vertex of $L(-S)(v,w)$, so that the two complexes have the same set of vertices. These are full subcomplexes of $L$ having the same sets of vertices, so that the inclusion $$L(-S)(v,w) \subseteq L(v,w)$$ is an identity.
\[lem 6\] Suppose that $v \leq w$ in $L$, where $v$ is a source and $w$ is a sink. Suppose given complexes $$L_{n} \subseteq L_{n-1} \subseteq \dots \subseteq L_{0}=L$$ where $v,w \in L_{i+1} = L_{i}(-S_{i})$ and $S_{i}$ is some set of sources and sinks in $L_{i}$. Suppose that $L_{n}$ has a unique source $v$ and a unique sink $w$. Then $L_{n}=L(v,w)$.
The connected component of $v$ in $L_{n}$ has a sink, which must be $w$. All other components would have sources and sinks, and must therefore be empty. It follows that $L_{n}$ is connected.
If $L_{n}$ has a vertex $x$ other than $v,w$ then there are non-degenerate $1$-simplices $$a_{1} \to x \to b_{1}.$$ If $a_{1}$ is a source then $a_{1}=v$. Otherwise, there is a $1$-simplex $a_{2} \to a_{1}$. This procedure must stop, to produce a path $$v = a_{r} \to \dots \to a_{2} \to a_{1} \to x.$$ Similarly, there is a path $$x \to b_{1} \to b_{2} \to \dots \to b_{s} = w.$$
If $L_{n}$ has no vertices other than $v,w$, then $L_{n}$ consists of the $1$-simplex $v \to w$.
It follows that every vertex of $L_{n}$ is on a path from $v$ to $w$, so that $L_{n}(v,w) = L_{n}$. Then Lemma \[lem 5\] implies that $L_{n}(v,w) = L(v,w)$, so that $L_{n} = L(v,w)$.
Suppose that $v \leq w$ in $L$, and start with $L_{0}=L[v,w]$. Let $S_{0}$ be the set of all sources and sinks of $L_{0}$, except for the elements $v,w$, and set $L_{1} =
L_{0}(-S_{0})$. Repeat this procedure inductively to produce a descending chain of complexes $$L_{n} \subseteq L_{n-1} \subseteq \dots \subseteq L_{0}=L[v,w],$$ with $S_{n} = \emptyset$. Then $$L_{n} = L[v,w](v,w) = L(v,w),$$ by Lemma \[lem 6\].
In other words, starting with the full subcomplex $L[v,w]$ we can successively delete sources and sinks to produce $L(v,w)$, which is the minimal full subcomplex of $L$ that computes $P(L)(v,w)$.
Corners
=======
Suppose that $i: K \subseteq \square^{n}$ is a finite cubical complex. The inclusion $i$ induces a functor $$i_{\ast}: P(K) \to P(\square^{n}) = \mathcal{P}(\underline{n}).$$ There is a poset map $t: \mathcal{P}(\underline{n}) \to \mathbb{N}$ that is defined by cardinality, in the sense that $$F \mapsto t(F) = \vert F \vert$$ for all subsets $F$ of $\underline{n}$. The composite functor $$P(K) \xrightarrow{i_{\ast}} \mathcal{P}(\underline{n}) \xrightarrow{t} \mathbb{N}$$ will also be denoted by $t$.
One thinks of the functor $t$ as a sort of time parameter for $K$. This functor also behaves like a total degree.
Suppose that $x$ is a vertex of the finite cubical complex $K$. Say that $x$ is a [*corner*]{} if it belongs to only one maximal cell of $K$.
The following result was proved by M. Misamore in [@Mis-path]. The proof that is given here is quite different.
\[lem 7\] Suppose that $x$ is a corner of $K$, and let $K_{x}$ be the subcomplex of cells which do not have $x$ as a vertex. Then the functor $$P(K_{x}) \to P(K)$$ is fully faithful.
Suppose that $\sigma$ is the unique top cell containing $x$.
If $x$ is either maximal or minimal in $\sigma$, then $x$ is either a sink or a source, respectively, by the uniqueness of $\sigma$. In that case, the functor $P(K_{x}) \to P(K)$ is fully faithful, by Lemma \[lem 3\].
Suppose that $x$ is neither maximal nor minimal in $\sigma$, and suppose that $P$ is a non-degenerate path in $K$ which passes through $x$, as in $$P:\ u = u_{0} \to \dots \to u_{n} = v,$$ where $u,v \in K_{x}$, and $u_{i} = x$. Then $i \ne 0,n$, and there is a unique $i$ such that $u_{i} = x$. In effect, since $P$ is non-degenerate, it induces a system of proper inequalities $$\vert u \vert = \vert u_{0} \vert < \vert u_{1} \vert < \dots < \vert x \vert < \dots < \vert u_{n} \vert = \vert v \vert,$$ in which the number $\vert x \vert$ can only appear once.
Then $u_{i-1}$ and $u_{i+1}$ are in $K_{x}$, and both $1$-simplices $u_{i-1} \to x$ and $x \to u_{i+1}$ are in $\sigma$ since $\sigma$ is the unique maximal cell that contains $x$.
Write $\sigma = [A,B]$.
As subsets of $B \subseteq \underline{n}$, $x = u_{i-1} \cup \{ a \}$ and $u_{i+1} = x \cup \{ b \}$, where $a$ and $b$ are distinct. The resulting $2$-cell $$\label{eq 2}
\xymatrix{
u_{i-1} \ar[r] \ar[d] \ar@{.>}[dr] & x \ar[d] \\
u_{i-1} \cup \{ b \} \ar[r] & u_{i+1}
}$$ in $\sigma$ (hence in $K$) defines a morphism $u_{i-1} \to u_{i}$ in $P(K_{x})$. Define $\psi(P)$ to be the composite of the morphisms $$u = u_{0} \to \dots \to u_{i-1} \to u_{i+1} \to \dots \to u_{n} = v$$ in $P(K_{x})(u,v)$.
The $2$-cell of the picture (\[eq 2\]) is uniquely determined by the path $P$, as is its image $\psi(P)$.
If $Q: u \to v$ is a non-degenerate path which does not pass through $x$, let $\psi(Q)$ be the image of $Q$ in $P(K_{x})(u,v)$. We have therefore determined a function $$\psi: \{ \text{paths}\ u \to v \} \to P(K_{x})(u,v).$$ If there is a $2$-cell between paths $u \to v$ in $K$, then the corresponding images under $\psi$ coincide. We therefore have an induced function $$\psi_{\ast}: P(K)(u,v) \to P(K_{x})(u,v).$$ The composite $$P(K_{x})(u,v) \to P(K)(u,v) \xrightarrow{\psi_{\ast}} P(K_{x})(u,v)$$ is the identity by construction. The construction of $\psi(P)$ for paths $P$ passing through $x$ shows that the function $$P(K_{x})(u,v) \to P(K)(u,v)$$ is surjective, and is therefore a bijection.
Suppose that $x \subseteq \underline{n}$. Then $x$ is an object of the poset $\mathcal{P}(\underline{n})$ and is a vertex of the simplicial set $B\mathcal{P}(\underline{n})$.
Let $\square^{n}_{x}$ be the cubical subcomplex of $\square^{n}$ consisting of cells which do not have $x$ as a vertex.
Let $D_{x}$ be the subcomplex of $B\mathcal{P}(\underline{n})$ consisting of those simplices which do not have $x$ as a vertex. $D_{x}$ is the nerve $B\mathcal{P}(\underline{n})_{x}$ of the full subcategory of $\mathcal{P}(\underline{n})$ with objects not equal to $x$. In particular, the functor $P(D_{x}) \to P(\square^{n})$ is fully faithful.
The isomorphism $\vert \square^{n} \vert \cong
B\mathcal{P}(\underline{n})$ restricts to a monomorphism of simplicial complexes $$\gamma: \vert \square^{n}_{x} \vert \to D_{x}.$$
Observe that if $x$ is neither the minimal element $\emptyset$ nor maximal element $\underline{n}$ of $\mathcal{P}(\underline{n})$, then $\emptyset
\subseteq \underline{n}$ is a $1$-simplex of $D_{x}$ which cannot be in the image of the map $\gamma$.
If $x$ is either the maximal or minimal element of $\mathcal{P}(\underline{n})$, then the map $\gamma$ is an isomorphism. In effect, if $x = \underline{n}$, then a simplex $F_{0}
\subseteq \dots \subseteq F_{k}$ is in $D_{x}$ if and only if $F_{k} \ne
\underline{n}$, and in this case it is in the image of the cell $\vert
[\emptyset,F_{k}] \vert$. The case $x = \emptyset$ is argued similarly.
\[cor 8\] The functor $P(\square^{n}_{x}) \to P(\square^{n})$ is fully faithful, and the induced functor $$\gamma_{\ast}: P(\vert \square^{n}_{x} \vert) \to P(D_{x})$$ is an isomorphism of path categories.
The functor $$i_{\ast}: P(\square^{n}_{x}) \to P(\square^{n})$$ is fully faithful by Lemma \[lem 7\].
The functor $\gamma_{\ast}$ is bijective on vertices, and is also fully faithful by the previous paragraph. It is therefore an isomorphism of categories as claimed.
The Swiss flag ($2$-cells indicated by double arrows, centre region is empty) $$\xymatrix@=10pt{
\bullet \ar[r] \ar@{=>}[dr]& \bullet \ar[r] & \bullet \ar[r] \ar@{=>}[dr]
& \bullet \\
\bullet \ar[r] \ar[u] & \ast \ar[u] & \ast \ar[u] \ar[r] & \bullet \ar[u] \\
\bullet \ar[r] \ar[u] \ar@{=>}[dr] & \ast & \ast \ar[r] \ar@{=>}[dr]
& \bullet \ar[u] \\
\bullet \ar[r] \ar[u] & \bullet \ar[r] \ar[u] & \bullet \ar[r] \ar[u]
& \bullet \ar[u]
}$$ has six corners, one sink, and one source, aside from the initial and terminal vertices. Remove the four “inner” corners to show that there are two morphisms from the initial vertex to the terminal vertex in the corresponding path category.
Refinement
==========
Suppose that $\alpha: \mathcal{P}(\underline{m}) \to
\mathcal{P}(\underline{n})$ is a poset monomorphism that preserves meets and joins.
Every interval $[A,B]$ in $\mathcal{P}(\underline{m})$ determines an interval $[\alpha(A),\alpha(B)]$ in $\mathcal{P}(\underline{n})$, and $\alpha$ restricts to a poset monomorphism $\alpha: [A,B] \to
[\alpha(A),\alpha(B)]$. The assignment $$[A,B] \mapsto [\alpha(A),\alpha(B)]$$ preserves inclusion relations between intervals, and preserves meets and joins of intervals.
The cubical subcomplex of $\square^{n}$ that is generated by the intervals $[\alpha(A),\alpha(B)]$ associated to the intervals $[A,B]$ of $K$ is denoted by $K_{\alpha}$, and there is a simplicial set map $\alpha_{\ast}: \vert K \vert \to \vert K_{\alpha} \vert$ that makes the diagram $$\xymatrix{
\vert K \vert \ar[d] \ar[r]^{\alpha_{\ast}} & \vert K_{\alpha} \vert \ar[d] \\
B\mathcal{P}(\underline{m}) \ar[r]_{\alpha} & B\mathcal{P}(\underline{n})
}$$ commute. The simplicial set map $\alpha_{\ast}$ is induced by the restricted poset morphisms $\alpha: [A,B] \to
[\alpha(A),\alpha(B)]$. These poset morphisms are not face inclusions in general.
Suppose that $K \subseteq \square^{m}$ and $L \subseteq \square^{n}$ are higher dimensional automata. We say that $L$ is a [*refinement*]{} of $K$ if there is a poset monomorphism $\alpha: \mathcal{P}(\underline{m}) \to
\mathcal{P}(\underline{n})$ that preserves meets and joins, and an inclusion $i: K_{\alpha} \subseteq L$ of cubical subcomplexes of $\square^{n}$. In this case, there is a commutative diagram of simplicial set maps $$\xymatrix{
\vert K \vert \ar[d] \ar[r]^{\alpha_{\ast}} & \vert K_{\alpha} \vert \ar[d] \ar[r]^{i_{\ast}}
& \vert L \vert \ar[dl] \\
B\mathcal{P}(\underline{m}) \ar[r]_{\alpha} & B\mathcal{P}(\underline{n})
}$$
\[lem 4x\] Suppose that $\alpha: \mathcal{P}(\underline{m}) \to
\mathcal{P}(\underline{n})$ is a poset monomorphism which preserves meets and joins, and suppose that $K \subseteq \square^{r}$ is a cubical subcomplex.
Then the induced functor $\alpha_{\ast}: P(K) \to P(K_{\alpha})$ is fully faithful.
Suppose that $F$ is a vertex of $K_{\alpha}$. Then $F \subseteq
[\alpha(A),\alpha(B)]$ for some interval $[A,B]$ of $K$, so there is a vertex $B$ of $K$ such that $F \subseteq \alpha(B)$. There is a minimal such $B$, call it $B_{F}$, since $\alpha$ preserves meets.
If $F=\alpha(C)$ for some $C$, then $B_{F} = C$, since $\alpha$ is a monomorphism. In effect, $\alpha(C) \leq \alpha(B_{F}) \leq
\alpha(C)$, so $B_{F} = C$ in this case.
Suppose that $$\omega: \alpha(A) \to F_{1} \to \dots \to F_{k} \to \alpha(B)$$ is a path in $K_{\alpha}$. Then each $F_{i} \to F_{i+1}$ is in an interval $[\alpha(C_{i}),\alpha(D_{i})]$, so that the diagram of inclusions $$\xymatrix{
F_{i} \ar[r] \ar[d] \ar[dr] & F_{i+1} \ar[d] \\
\alpha(B_{F_{i}}) \ar[r] & \alpha(B_{F_{i+1}})
}$$ is in that same interval. The inclusion $\alpha(B_{F_{i}}) \to
\alpha(B_{F_{i+1}})$ is the image of an inclusion $B_{F_{i}} \to
B_{F_{i+1}}$ by the minimality of $B_{F_{i}}$. It follows that the diagram $$\label{eq A}
\xymatrix{
\alpha(A) \ar[r] \ar[dr] \ar[d]_{1} & F_{1} \ar[dr] \ar[r] \ar[d] & \dots \ar[r] \ar[dr]
& F_{k} \ar[r] \ar[d] \ar[dr] & \alpha(B) \ar[d]^{1} \\
\alpha(A) \ar[r] & \alpha(B_{F_{1}}) \ar[r] & \dots \ar[r]
& \alpha(B_{F_{k}}) \ar[r] & \alpha(B)
}$$ defines a homotopy in $\vert K_{\alpha} \vert$ from the path $\omega$ to the path along the bottom, which path is in the image of the function $P(K)(A,B) \to P(K_{\alpha})(\alpha(A),\alpha(B))$, because all displayed simplices are in $\vert K_{\alpha} \vert$.
Suppose given a commutative diagram $$\xymatrix{
\alpha(A) \ar[r] & \dots \ar[r] & F_{i} \ar[rr] \ar[dr]
&& F_{i+1} \ar[r] & \dots \ar[r] & \alpha(B)\\
&&& F \ar[ur]
}$$ where $\omega$ is the path along the top, and the displayed triangle of inclusions defines a $2$-simplex $\sigma$ of $\vert K_{\alpha} \vert$. This $2$-simplex is in some interval $[\alpha(C),\alpha(D)]$, and the corresponding diagram $$\xymatrix{
\alpha(B_{F_{i}}) \ar[rr] \ar[dr] && \alpha(B_{F_{i+1}}) \\
& \alpha(B_{F}) \ar[ur]
}$$ is also in the interval $[\alpha(C),\alpha(D)]$, by minimality. This simplex is the image of a $2$-simplex of $\vert K \vert$.
We have therefore defined a function $$s: P(K_{\alpha}(\alpha(A),\alpha(B)) \to P(K)(A,B)$$ such that the composite $s \cdot \alpha_{\ast}$ is the identity on $P(K)(A,B)$. The construction of the function $s$ and the existence of the homotopies (\[eq A\]) together imply that the function $$\alpha_{\ast}: P(K)(A,B) \to P(K_{\alpha})(\alpha(A),\alpha(B))$$ is surjective. It follows that the function $\alpha_{\ast}$ is a bijection, as required.
Frontier subcomplex
===================
Suppose that $i: K \subseteq \square^{n}$ is a finite cubical complex. Recall from Section 3 that the assignment $F \mapsto \vert F
\vert =: t(F)$ that is defined by cardinality determines a poset morphism $\mathcal{P}(\underline{n}) \to \mathbb{N}$, and hence a composite functor $$t: P(K) \xrightarrow{i_{\ast}} \mathcal{P}(\underline{n}) \xrightarrow{t} \mathbb{N}.$$
Observe that if $v \to w$ is a non-degenerate $1$-simplex of $\vert K
\vert$, then there is a strict containment relation $v \subset w$ as subsets of $\underline{n}$, so that $t(v) < t(w)$. A more precise version of this statement applies to all $1$-cells $v \to w$ of $K$: $t(w) =
t(v) +1$ for such a $1$-cell.
The functor $t$ defines full subcomplexes of the complex $K$ and its triangulation $\vert K \vert$. In particular, if $r < s$ in $\mathbb{N}$, let $K(r,s)$ be the subcomplex of cells whose vertices $F$ satisfy $r \leq t(F) \leq s$, and let $\vert K \vert(r,s)$ be the subcomplex of $\vert K \vert$ whose simplices have vertices $F$ with $t(f)$ in the same range.
Then we have the following:
\[lem 10\]
- $\vert K \vert(r,s)$ is a full subcomplex of $\vert K \vert$.
- The canonical map $$\vert K(r,s) \vert \to \vert K \vert(r,s)$$ is an isomorphism of simplicial complexes.
For statement 1), suppose that $v,w$ are vertices of $\vert K
\vert(r,x)$ and that $$v=v_{0} \to v_{1} \to \dots \to v_{n}=w$$ is a non-degenerate path from $v$ to $w$ in $\vert K \vert$. Then $$r \leq t(v)=t(v_{0}) < t(v_{1}) < \dots < t(v_{n}) = t(w) \leq s,$$ so that all vertices $v_{i}$ are in $\vert K \vert(r,s)$. The higher simplex condition for fullness of $\vert K \vert(r,s)$ is automatic from the definition.
The canonical inclusion of statement 2) arises from the observation that $K(r,s)$ is a union of cells of $K$, and the induced inclusion $\vert K(r,s) \vert \subseteq \vert K \vert$ factors through $\vert K
\vert(r,s)$.
To prove statement 2), it is enough to show that the inclusion $$\vert K(r,s) \vert \to \vert K \vert(r,s)$$ is surjective on non-degenerate simplices. If $\sigma$ is a simplex of $\vert K \vert(r,s)$, it is in the image of the map $\vert \square^{k} \vert \to \vert K \vert$ which is induced by a non-degenerate cell of $K$. The simplex $\sigma$ has the form $$F_{0} \leq F_{1} \leq \dots \leq F_{p}$$ with $$r \leq \vert F_{0} \vert \leq \vert F_{1} \vert \leq \dots \leq \vert F_{p} \vert \leq s,$$ and it follows that the interval $[F_{0},F_{p}]$ defines a cell of $K(r,s)$. The simplex $\sigma$ is therefore in $\vert K(r,s)
\vert$.
Suppose that $K \subseteq \square^{n}$ is a finite cubical complex such that $\sk_{2}(K) = K$, and pick $M$ such that $0 < M < n$. Let $A
= K(0,M)$ and $B = K(M+1,n)$. Then $\vert A \vert$ and $\vert B \vert$ are full subcomplexes of $\vert K \vert$ by Lemma \[lem 10\].
Every path $$v_{0} \to v_{1} \to \dots \to v_{k}$$ in $K$ has a number $r$ (which could be $-1$ or $k$) such that $v_{i} \in A$ for $i \leq r$ and $v_{i} \in B$ for $i \geq r+1$. The [*frontier subcomplex*]{} $L$ is generated by $1$-cells and $2$-cells which have vertices in $A$ and $B$.
Suppose that $u \in A$ and $v \in B$. Suppose that the $1$-cell $\sigma: x \to y$ has $x \in A$ and $y \in B$. Then composition with $\sigma$ defines a map $$\sigma_{\ast}: P(A)(u,x) \times P(B)(y,v) \to P(K)(u,v).$$ Suppose that $\omega: \Delta^{1} \times \Delta^{1} \to K$ is defined by $2$-simplices $\omega_{0}$ and $\omega_{1}$ such that $d_{1}\omega_{0} = d_{1}\omega_{1}$ and $d_{2}(\omega_{0}) \in A$ and $d_{0}(\omega_{1}) \in B$. One of the $2$-simplices $\omega_{0}$ or $\omega_{1}$ could be degenerate.
Consider the picture: $$\xymatrix{
& \sigma(0,0) \ar[r]^{\alpha_{0}} \ar[dd]_{A} \ar[ddr]
& \sigma(1,0) \ar[dd]^{B} \ar@{.>}[dr]^{q_{0}} \\
u \ar[ur]^{p_{0}} \ar@{.>}[dr]_{p_{1}} &&& v \\
& \sigma(0,1) \ar[r]_{\alpha_{1}} & \sigma(1,1) \ar[ur]_{q_{1}}
}$$ There are induced maps $$\omega_{0}: P(A)(u,\sigma(0,0)) \times P(B)(\sigma(1,1),v) \to P(A)(u,\sigma(0,0)) \times P(B)(\sigma(1,0),v)$$ and $$\omega_{0}: P(A)(u,\sigma(0,0)) \times P(B)(\sigma(1,1),v) \to P(A)(u,\sigma(0,1)) \times P(B)(\sigma(1,1),v)$$
These maps define the displayed parallel pair of arrows in the diagram $$\label{eq 4}
\begin{aligned}
\bigsqcup_{\omega\ \text{as above}} P(A)(u,\sigma(0,0)) \times P(B)(\sigma(1,1),v) \rightrightarrows \bigsqcup_{x \xrightarrow{\sigma} y\ \in\ L} &P(A)(u,x) \times P(B)(y,v) \\
&\to P(K)(u,v).
\end{aligned}$$
\[lem 11\] The diagram (\[eq 4\]) is a coequalizer.
The proof of Lemma \[lem 11\] is essentially by inspection.
In practical terms, Lemma \[lem 11\] says that one can compute $P(K)$ by first computing $P(A)$ and $P(B)$ (in parallel), and then by stitching these calculations together with the coequalizer (\[eq 4\]). This coequalizer defines $P(K)(u,v)$ as a set of equivalence classes on a set that we’ve computed, namely $$\bigsqcup_{x \xrightarrow{\sigma} y\ \in\ L} P(A)(u,x) \times
P(B)(y,v),$$ for an equivalence relation that is defined by the parallel pair of functions in the coequalizer picture (\[eq 4\]).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In wireless networks, the knowledge of nodal distances is essential for several areas such as system configuration, performance analysis and protocol design. In order to evaluate distance distributions in random networks, the underlying nodal arrangement is almost universally taken to be an infinite Poisson point process. While this assumption is valid in some cases, there are also certain impracticalities to this model. For example, practical networks are non-stationary, and the number of nodes in disjoint areas are not independent. This paper considers a more realistic network model where a finite number of nodes are uniformly randomly distributed in a general $d$-dimensional ball of radius $R$ and characterizes the distribution of Euclidean distances in the system. The key result is that the probability density function of the distance from the center of the network to its $n^{\text{th}}$ nearest neighbor follows a generalized beta distribution. This finding is applied to study network characteristics such as energy consumption, interference, outage and connectivity.'
author:
- |
Sunil Srinivasa and Martin Haenggi\
Department of Electrical Engineering\
University of Notre Dame\
Notre Dame, IN 46556, USA\
Email: [*{ssriniv1, mhaenggi}@nd.edu*]{}
title: 'Distance Distributions in Finite Uniformly Random Networks: Theory and Applications'
---
Introduction
============
Motivation
----------
In wireless channels, the received signal strength falls off with distance according to a power law, at a rate termed the large scale path loss exponent (PLE) [@Rappaport]. Given a transceiver distance $d$, the signal power at the receiver is attenuated by a factor of $d^{-\alpha}$, where $\alpha$ is the PLE. Consequently, in wireless networks, distances (or equivalently, the distribution of distances for random networks) between nodes strongly impact the signal-to-noise-and-interference ratios (SINRs), and therefore the link reliabilities as well. The knowledge of the nodal distances is therefore essential for several important areas such as the throughput and performance analysis and the design of protocols and algorithms.
In wireless networks, nodes are typically scattered randomly over an area or volume, and the distance distributions follow from the spatial stochastic process governing the locations of the nodes. For the sake of analytical convenience, the arrangement of nodes in a random network is commonly taken to be a homogeneous (or stationary) Poisson point process (PPP). For the so-called “Poisson network” of intensity $\lambda$, the number of nodes in any given set of Lebesgue measure $V$ is Poisson with mean $\lambda{}V$, and the number of nodes in disjoint sets are independent of each other. Even though the PPP assumption can lead to some insightful results, practical networks differ from Poisson networks in certain aspects. First, networks are formed by usually scattering a fixed (and finite) number of nodes in a given area (or very close to it). If nodes are uniformly randomly distributed, they form a *binomial point process* (BPP), which we describe shortly. Secondly, since the area or volume of deployment is necessarily finite, the point process formed is non-stationary and often non-isotropic, meaning that the network characteristics as seen from a node’s perspective is not homogeneous for all nodes. For example, the mean farthest neighbor distances are larger for nodes lying near the boundary than for the ones closer to the center. Furthermore, the number of nodes in disjoint sets are not independent but governed by a multinomial distribution. Fig. \[fig:BPP\_vs\_PPP\] shows a realization of the two processes with the same density and depicts that the PPP is clearly not a good model, since there may be more points in the realization than the number of nodes deployed. In particular when the number of nodes is small, the Poisson model is inaccurate. The main shortcoming of the Poisson assumption is, however, the independence of the number of nodes in disjoint areas. For example, if all users or nodes are located in a certain part of the network area, the remaining area is necessarily empty. This simple fact is not captured by the Poisson model. This motivates the need to study and accurately characterize finite uniformly random networks, in an attempt to extend the plethora of results for the PPP to the often more realistic case of the BPP. We call this new model a “binomial network”.
Formally speaking, a $d$-dimensional BPP is formed as a result of distributing $N$ points independently uniformly in a compact set $W\subset\mathbb{R}^d$ [@Stoyan]. The intensity of this process is defined to be $N/\nu_d(W)$. For any Borel subset $V$ of $W$, the number of points in $V$, $\Phi(V)$, is binomial($n,p$) with parameters $n=N$ and $p=\nu_d(V)/\nu_d(W)$, where $\nu_d()$ is the standard $d$-dimensional Lebesgue measure. Accordingly, $$\Pr(\Phi(V)=k)={n\choose{}k}p^k(1-p)^{n-k}.$$ By this property, the number of nodes in disjoint sets are joint via a multinomial distribution. This network model applies to mobile ad hoc networks and wireless networks with infrastructure, such as cellular telephony networks or sensor networks.
In this paper, we study the Euclidean distance properties in a general $d$-dimensional binomial network. We also provide results on network characteristics such as energy consumption, interference, outage and connectivity based on the distance distributions.
Related Work
------------
Even though the knowledge of node locations in wireless networks is important for their analysis and design, relatively few results are available in the literature in this area. Moreover, much of existing work deals only with moments of the distances (means and variances) or characterizes the exact distribution only for very specific system models.
In [@Miller1], the probability density function (pdf) and cumulative distribution function (cdf) of distances between nodes are derived for networks with uniformly random and Gaussian distributed nodes over a rectangular area. [@Rose] studies mean internodal distance properties for several kinds of multihop systems such as ring networks, Manhattan street networks, hypercubes and shufflenets. [@Haenggi1] provides closed-form expressions for the distributions in $d$-dimensional homogeneous PPPs and describes several applications of the results for large networks. [@Miller2] derives the joint distribution of distances of nodes from a common reference point for planar networks with a finite number of nodes randomly distributed on a square. [@Tseng] considers square random networks and determines the pdf and cdf of nearest neighbor and internodal distances. [@Vural] investigates one-dimensional multihop networks with randomly located nodes and analyzes the distributions of single-hop and multiple-hop distances.
Distance Distribution to the Neighbors
======================================
In this section, we determine the distribution of the Euclidean distance to the $n^{\text{th}}$ nearest point from the origin for a general $d$-dimensional isotropic BPP. It will be established that this random variable follows a generalized beta distribution. We also comment on the distances to the nearest and farthest neighbors and the void probabilities.
In a point process consisting of $N$ nodes uniformly randomly distributed in a $d$-dimensional ball of radius $R$ centered at the origin, the Euclidean distance $R_n$ from the origin to its $n^{\text{th}}$ nearest neighbor is distributed as a generalized beta distribution i.e., $$\begin{aligned}
f_{R_n}(r)&=&\frac{d}{R}\frac{B(n-1/d+1,N-n+1)}{B(N-n+1,n)}\beta\left(\left(\frac{r}{R}\right)^d;n-\frac{1}{d}+1,N-n+1\right),\quad{}r\in[0,R],\nonumber\end{aligned}$$ where $\beta(x;a,b)$ is the beta density function[^1] defined as $\beta(x;a,b)=\frac{1}{B(a,b)}x^{a-1}(1-x)^{b-1}$. $B(a,b)$ denotes the beta function and is expressible in terms of gamma functions as $B(a,b)=\Gamma(a)\Gamma(b)/\Gamma(a+b)$.
Consider the BPP with $N$ points uniformly randomly distributed in a $d$-dimensional ball $W$ of radius $R$ centered at the origin $o$, i.e., $W=\mathbb{B}_d(o,R)$. The volume of this ball $\nu_d(W)$ is equal to $c_dR^d$ [@Stoyan], where $$c_d=\frac{\pi{}^{d/2}}{\Gamma(1+d/2)}$$ is the volume of the unit ball in $\mathbb{R}^d$. The important cases include $c_1=2$, $c_2=\pi$ and $c_3=4\pi/3$. The density of this process is equal to $N/(c_dR^d)$.\
The complementary cumulative distribution function (ccdf) of $R_n$ is the probability that there are less than $n$ points in the ball $\mathbb{B}_d(o,r)$ : $$\bar{F}_{R_n}(r)=\sum_{k=0}^{n-1}{N\choose{}k}p^k(1-p)^{N-k},\quad{}0\leq{}r\leq{}R,$$ where $p=c_dr^d/c_dR^d=\left(r/R\right)^d$.
$\bar{F}_{R_n}$ can be written in terms of the regularized incomplete beta function as $$\bar{F}_{R_n}(r)=I_{1-p}(N-n+1,n),\quad0\leq{}r\leq{}R,$$ where $$I_x(a,b)=\frac{\int_0^xt^{a-1}(1-t)^{b-1}\textrm{d}t}{B(a,b)}.$$ The pdf of the distance function is $f_{R_n}=-\text{d}\bar{F}_{R_n}/\text{d}r$ and we have $$\begin{aligned}
f_{R_n}(r)&=&-\frac{\text{d}}{\text{d}r}I_{1-p}(N-n+1,n)\nonumber\\
&=&-\left(\frac{\text{d}(1-p)}{\text{d}r}\right)\frac{(1-p)^{N-n}p^{n-1}}{B(N-n+1,n)}\nonumber\\
&=&\frac{d}{R}\left(\frac{r}{R}\right)^{d-1}\frac{(1-p)^{N-n}p^{n-1}}{B(N-n+1,n)}\nonumber\\
&=&\frac{d}{R}\frac{(1-p)^{N-n}p^{n-1/d}}{B(N-n+1,n)}\nonumber\\
&=&\frac{d}{R}\frac{B(n-1/d+1,N-n+1)}{B(N-n+1,n)}\beta\left(\left(\frac{r}{R}\right)^d;n-\frac{1}{d}+1,N-n+1\right)
\label{eqn:pdf}\end{aligned}$$ for $0\leq{}r\leq{}R$. The final equality casts $R_n$ as a generalized beta-distributed variable.
For the practical cases of $d=1$ and $d=2$, we have $$f_{R_n}(r)=\frac{1}{R}\beta\left(\frac{r}{R};n,N-n+1\right)$$ and $$\begin{aligned}
f_{R_n}(r)=\frac{2}{R}\frac{\Gamma(n+\frac{1}{2})\Gamma(N+1)}{\Gamma(n)\Gamma(N+\frac{3}{2})}\beta\left(\frac{r^2}{R^2};n+\frac{1}{2},N-n+1\right)\end{aligned}$$ respectively.
Fig. \[figure:distance\_pdf\] plots the distance pdfs for the cases of $d=1$ and $d=2$.
Remarks: {#remarks .unnumbered}
--------
1. [The void probability $p^0_B$ of the point process is defined as the probability of there being no point of the process in the test set $B\subseteq{}W$ [@Stoyan]. For a BPP with $N$ points distributed over a set $W$, it is easy to see that $$p^0_B=\left(1-\nu(B)/\nu(W)\right)^N.$$ For the isotropic BPP considered above, when the test set is $B=\mathbb{B}_d(0,r)$, we have $p^0_B=I_{1-p}(N,1)=(1-p)^N$. ]{}
2. [Of interest in particular are the nearest and farthest neighbor distances. The nearest neighbor distance pdf is given by $$f_{R_1}(r)=\frac{dN}{r}\left(1-\left(\frac{r}{R}\right)^d\right)^{N-1}\left(\frac{r}{R}\right)^d,$$ and the distance to the farthest point from the origin is distributed as $$f_{R_N}(r)=\frac{dN}{r}\left(\frac{r}{R}\right)^{Nd},\quad0\leq{}r\leq{}R.$$ Both these are distributed as the generalized Kumaraswamy distributions [@Kumaraswamy]]{}.
3. [For a one-dimensional BPP, $f_{R_n}(r)=f_{R_{N-n+1}}(R-r)$, and therefore knowledge of the distance pdfs for the nearest $\lceil{}N/2\rceil$ neighbors gives complete information on the distance distributions to the other nodes.]{}
We wish to compare the distance distributions for a BPP and a PPP. However, note that in general, the PPP may have fewer points than the number dropped. In order to make a fair comparison, we condition on the fact that there are at least $N$ points present in the PPP model. The following corollary establishes the distance pdfs for such a conditioned PPP. Also note that conditioned on there being exactly $N$ points present, the PPP is equivalent to a BPP [@Stoyan].
Consider a PPP of density $\lambda$ over a finite volume $\mathbb{B}_d(o,R)$. Conditioned on there being at least $N$ points in the ball, the distance distribution from the origin to the $n^{\text{th}}$ nearest neighbor ($n\leq{}N$) is given by $$f'_{R_n}(r)=\frac{\lambda{}dc_dr^{d-1}\left(A_{n-1}(r)\left(1-\sum_{k=0}^{N-n-1}B_k(r)\right)\right)}{1-\sum_{k=0}^{N-1}e^{-\lambda{}c_dR^d}(\lambda{}c_dR^d)^k/k!},\quad{}r\in[0,R],
\label{eqn:cond_pdf}$$ where $A_k(r):=e^{-\lambda{}c_dr^d}\left(\lambda{}c_dr^d\right)^k/k!$ and $B_k(r):=e^{-\lambda{}c_d\left(R^d-r^d\right)}\left(\lambda{}c_d\left(R^d-r^d\right)\right)^k/k!$.
The complimentary conditional cdf of $R_n$ is given by $$\begin{aligned}
\bar{F'}_{R_n}(r)&=&\Pr\left(\Phi\left(\mathbb{B}_d(o,r)\right)<n|\Phi\left(\mathbb{B}_d(o,R)\right)\geq{}N\right)\nonumber\\
&=&\frac{\Pr\left(\Phi\left(\mathbb{B}_d(o,r)\right)<n,\Phi\left(\mathbb{B}_d(o,R)\right)\geq{}N\right)}{\Pr\left(\Phi\left(\mathbb{B}_d(o,R)\right)\geq{}N\right)}\nonumber\\
&{\substack{(a)\\=}}&\frac{\sum_{k=0}^{n-1}\Pr\left(\Phi\left(\mathbb{B}_d(o,r)\right)=k\right)\Pr\left(\Phi\left(\mathbb{B}_d(o,R)\setminus{}\mathbb{B}_d(o,r)\right)\geq{}N-k\right)}{\Pr\left(\Phi\left(\mathbb{B}_d(o,R)\right)\geq{}N\right)}\nonumber\\
&=&\frac{\sum_{k=0}^{n-1}A_k(r)\left(1-\sum_{l=0}^{N-k-1}B_l(r)\right)}{1-\sum_{k=0}^{N-1}e^{-\lambda{}c_dR^d}(\lambda{}c_dR^d)^k/k!},
\label{eqn:cond_cdf}\end{aligned}$$ where $(a)$ is obtained from the property that the number of points in disjoint sets are independent of each other. It is easy to see that $$\frac{\text{d}}{\text{d}r}A_k(r)=\left \{\begin{array}{ll}
\lambda{}dc_dr^{d-1}\left(A_{k-1}(r)-A_k(r)\right)&k>0\\
-\lambda{}dc_dr^{d-1}A_0(r)&k=0
\end{array} \right.
\tag{i}$$ and $$\frac{\text{d}}{\text{d}r}B_l(r)=\left \{\begin{array}{ll}
\lambda{}dc_dr^{d-1}\left(B_l(r)-B_{l-1}(r)\right)&l>0\\
\lambda{}dc_dr^{d-1}B_0(r)&l=0.
\tag{ii}
\end{array} \right.$$ Therefore, we have $$%\frac{\text{d}}{\text{d}r}\sum_{k=0}^{n-1}A_k(r)=-\lambda{}dc_dr^{d-1}A_{n-1}(r) \text{ and }
\frac{\text{d}}{\text{d}r}\sum_{l=0}^{N-k-1}B_l(r)=\lambda{}dc_dr^{d-1}B_{N-k-1}(r).
\tag{iii}$$ The details of the remainder of the proof are straightforward but tedious and are omitted here. Since the pdf of the conditional distance distribution is $f'_{R_n}=-\text{d}\bar{F'}_{R_n}/\text{d}r$, one basically has to differentiate the numerator in (\[eqn:cond\_cdf\]), and after some simplifications using (i)-(iii), it will be seen that the conditional distance pdf is identical to (\[eqn:cond\_pdf\]).
Fig. \[figure:distance\_pdf\] depicts the conditional pdfs of the distances for one- and two-dimensional Poisson processes and compares it with the distance distributions for a BPP.
When a large number of nodes are distributed randomly over a large area, their arrangement can be well approximated by an infinite homogeneous PPP. The PPP model for the nodal distribution is ubiquitously used for wireless networks and may be justified by claiming that nodes are dropped from an aircraft in large numbers; for mobile ad hoc networks, it may be argued that terminals move independently of each other. We now present the following corollary to the earlier theorem, which also reproduces a result from [@Haenggi1].
In an infinite PPP with intensity $\lambda$ on $\mathbb{R}^d$, the distance $R_n$, between a point and its $n^{\text{th}}$ neighbor is distributed according to the generalized Gamma distribution. $$f_{R_n}(r)=e^{-\lambda{}c_dr^d}\frac{d(\lambda{}c_dr^d)^n}{r\Gamma(n)},\quad{}r\in\mathbb{R}.$$
If the total number of points $N$ tends to infinity in such a way that the density $\lambda=N/(c_dR^d)$ remains a constant, then the BPP asymptotically (as $R\to\infty$) behaves as a PPP [@Stoyan]. Taking $R=\sqrt[d]{N/c_d\lambda}$ and applying the limit as $N\to\infty$, we obtain for a PPP, $$\begin{aligned}
f_{R_n}(r)&=&\lim_{N\to\infty}\frac{d}{R}\frac{\left(1-p\right)^{N-n}p^{n-1/d}\Gamma(N+1)}{\Gamma(N-n+1)\Gamma(n)}\\
&=&\frac{d}{r\Gamma(n)}(\lambda{}c_dr^d)^n\lim_{N\to\infty}\left(1-\frac{\lambda{}c_dr^d}{N}\right)^{N}\frac{N(N-1)\ldots{}(N-n+1)}{N^n}\\
&=&e^{-\lambda{}c_dr^d}\frac{d(\lambda{}c_dr^d)^n}{r\Gamma(n)}.\end{aligned}$$ for $r\in\mathbb{R}$. This is an alternate proof to the one provided in [@Haenggi1].
For a PPP, we can also specify how the internodal distances are distributed. By the stationarity of the PPP and Slivnyak’s theorem [@Stoyan], the distance between the origin and its $n^{\text{th}}$ nearest neighbor is the same as the distance between an arbitrary point and its $n^{\text{th}}$ nearest neighbor. [@Haenggi1] studies distance distributions in a PPP in more detail and shows that the nearest neighbor distribution for any node is exponential and Rayleigh-distributed for the one-dimensional and two-dimensional PPP respectively. However, since the BPP is non-stationary, it is not straightforward to specify how the internodal distances are distributed in that case.
Moments of $R_n$
================
We now use the distance pdf to compute its moments. The $\gamma^{\text{th}}$ moment of $R_n$ is calculated as follows[^2]. $$\begin{aligned}
\mathbb{E}[R_n^{\gamma}]&=&\frac{d}{R}\frac{1}{B(N-n+1,n)}\int_{0}^R\left[r^{\gamma}\left(\frac{r}{R}\right)^{nd-1}\left(1-\left(\frac{r}{R}\right)^d\right)^{N-n}\right]\text{d}r.\nonumber\\
&{\substack{(a)\\=}}&\frac{R^{\gamma}}{B(N-n+1,n)}\int_0^1t^{n+\gamma/d-1}(1-t)^{N-n}\text{d}t\nonumber\\
&=&\frac{R^{\gamma}}{B(N-n+1,n)}B_x(n+\gamma/d,N-n+1)|_0^1\nonumber\\
&{\substack{(b)\\=}}&\left \{ \begin{array}{l l}
\frac{R^{\gamma}\Gamma(N+1)\Gamma(\gamma/d+n)}{\Gamma(n)\Gamma(\gamma/d+N+1)}&\quad{}\text{if }n+\gamma/d>0\\
\infty{}&\quad{}\text{otherwise}
\end{array} \right. \nonumber\\
&=&\left \{ \begin{array}{l l}
R^{\gamma}n^{[\gamma/d]}/(N+1)^{[\gamma/d]}&\text{if }n+\gamma/d>0\\
\infty&\text{otherwise},
\end{array} \right.
\label{eqn:moments}\end{aligned}$$ where $B_x[a,b]$ is the incomplete beta function[^3] and $x^{[n]}=\Gamma(x+n)/\Gamma(x)$ denotes the rising Pochhammer symbol notation. Here, $(a)$ is obtained on making the substitution $r=Rt^{1/d}$ and $(b)$ using the following identities: $$\begin{aligned}
B_0(a,b)&=&\left \{ \begin{array}{l l}
0\quad{}&\mathcal{R}e(a)>0\\
-\infty\quad{}&\mathcal{R}e(a)\leq{}0,
\end{array} \right.\end{aligned}$$ and $B_1(a,b)=B(a,b)$ if $\mathcal{R}e(b)>0$.
The expected distance to the $n^{\text{th}}$ nearest neighbor is thus $$\mathbb{E}(R_n)=\frac{Rn^{[1/d]}}{(N+1)^{[1/d]}},$$ and the variance of $R_n$ is easily calculated as $$\begin{aligned}
\text{Var}[R_n]=\frac{R^{2}n^{[2/d]}}{(N+1)^{[2/d]}}-\left(\frac{Rn^{[1/d]}}{(N+1)^{[1/d]}}\right)^2.\end{aligned}$$
Remarks: {#remarks-1 .unnumbered}
--------
1. [For one-dimensional networks, $\mathbb{E}[R_n]=Rn/(N+1)$. Thus, on an average, it is as if the points are arranged regular lattice. In particular, when $N$ is odd, the middle point is located exactly at the center on average.]{}
2. [On the other hand, as $d\to\infty$, $\mathbb{E}[R_n]\to{}R$ and it is as if all the points are equidistant at maximum distance $R$ from the origin.]{}
3. [In the general case, the mean distance of the $n^{\text{th}}$ neighbor varies as $n^{1/d}$ for large $n$. This follows from the series expansion of the Pochhammer sequence [@Graham] $$n^{[q]}=n^q\left(1-\mathcal{O}\left(1/n\right)\right).$$ Also, the variance $\to{}0$ as $n$ increases for $d>2$. This is also observed in the case of a Poisson network [@Haenggi1].]{}
4. [The mean internodal distance between the $i^{\text{th}}$ and $j^{\text{th}}$ nearest neighbors from the origin is simply given by (assuming $i<j$) $$\mathbb{E}[d_{ij}]=\frac{R\left(j^{[1/d]}-i^{[1/d]}\right)}{(N+1)^{[1/d]}}.$$]{}
5. [For the special case of $\gamma/d\in\mathbb{Z}$, we obtain $$\mathbb{E}[R_n^{\gamma}]=R^{\gamma}{n+\gamma/d-1\choose{}\gamma/d}\bigg{/}{N+\gamma/d\choose{}\gamma/d}.$$]{}
Applications to wireless networks
=================================
We now apply the results obtained in the previous section to binomial networks. For the system model, we assume a $d$-dimensional network over a ball $\mathbb{B}_d(o,R)$, where $N$ nodes are uniformly randomly distributed. Nodes are assumed to communicate with a base station (BS) positioned at the origin. The attenuation in the channel is modeled by the large scale path loss function $g$ with PLE $\alpha$, i.e., $g(x)=\|x\|^{-\alpha}$. The channel access scheme is taken to be the slotted ALOHA with contention parameter $p$.
Energy Consumption
------------------
Since battery life in wireless nodes are limited, energy consumption is a key design issue. The energy that is required to successfully deliver a packet over a distance $r$ in a medium with PLE $\alpha$ is proportional to $r^{\alpha}$. Therefore, the average energy required to deliver a packet from the $n^{\text{th}}$ nearest neighbor to the BS is given by (\[eqn:moments\]), with $\gamma=\alpha$. This approximately scales as $n^{\alpha/d}$ when the routing is taken over single hops. When $\alpha<d$, its more energy-efficient to use longer hops than when the PLE is greater than the number of dimensions.
Localization
------------
In this section, we investigate conditional distance distributions and study their usefulness to localization algorithms, where it is beneficial to use as few beacons as possible for the estimation process.
Assume that the nodes in the network can talk to each other and determine how many other nodes are closer to the BS than they are, based on the strength of the received signal from the BS. However, the channel may be time-varying and hence, it might not be possible for all the nodes to determine their exact locations. In such situations, a few beacons can be employed that estimate or even precisely measure their distances from the BS. What can be said about the distance distributions of the other nodes given this information?
Suppose we know that the $k^{\text{th}}$ nearest neighbor is at distance $s$ from the center. Then, clearly, the first $n-1$ nodes are uniformly randomly distributed in $\mathbb{B}_d(o,s)$ while the more distant nodes are uniformly randomly distributed in $\mathbb{B}_d(o,R)\setminus{}\mathbb{B}_d(o,s)$. Following (\[eqn:pdf\]), the distance distributions of the first $k-1$ nearest neighbors from the origin can be written as $$\begin{aligned}
f_{R_n}(r|R_k=s)&=&\frac{d}{s}\frac{B(n-1/d+1,k-n)}{B(k-n,n)}\beta\left(\left(\frac{r}{s}\right)^d;n-\frac{1}{d}+1,k-n\right),\quad{}n<k\end{aligned}$$ for $0\leq{}r\leq{}s$, which again follows a generalized beta distribution.\
For the remaining nodes i.e., for $n>k$, we have in $r\in[s,R]$, $$\begin{aligned}
f_{R_n}(r|R_k=s)&=&-\frac{\text{d}}{\text{d}r}I_{1-q}(N-n+1,n-k)\\
&=&\frac{dr^{d-1}}{R^d-s^d}\frac{\left(1-q\right)^{N-n}q^{n-k-1}}{B(N-n+1,n-k)}\end{aligned}$$ where $q=(r^d-s^d)/(R^d-s^d)$.\
The moments of $R_n$ are also straightforward to obtain. Following (\[eqn:moments\]), we see that for $n<k$ and $n+\alpha/d>0$, $$\mathbb{E}[R_n^{\alpha}|R_k=s]=\frac{s^\alpha{}n^{[\alpha/d]}}{(k+1)^{[\alpha/d]}}.$$ For $n>k$, we have $$\begin{aligned}
\mathbb{E}[R_n^{\alpha}|R_k=s]&=&\int_s^R{}\frac{dr^{\alpha+d-1}}{R^d-s^d}\frac{\left(1-q\right)^{N-n}q^{n-k-1}}{B(N-n+1,n-k)}\text{d}r\\
&=&\frac{1}{B(N-n+1,n-k)}\int_0^1q^{n-k-1}(1-q)^{N-n}\left(q\left(R^d-s^d\right)+s^d\right)^{\alpha/d}\text{d}t\\
&=&\frac{s^\alpha}{(n-k)B(N-n+1,n-k)}F_1\left(n-k;n-N,-\frac{\alpha}{d};n-k+1;1,1-\frac{R^d}{s^d}\right),\end{aligned}$$ where $F_1[a;b_1,b_2;c;x, y]$ is the Appell hypergeometric function of two variables[^4].
Often, it is easiest to measure the nearest neighbor distance. Give this distance as $s$, we have for $n>1$, $$\begin{aligned}
f_{R_n}(r|R_1=s)=\frac{dr^{d-1}}{R^d-s^d}\frac{\left(1-\left(\frac{r^d-s^d}{R^d-s^d}\right)\right)^{N-n}\left(\frac{r^d-s^d}{R^d-s^d}\right)^{n-2}}{B(N-n+1,n-1)}\end{aligned}$$ for $r\in[s,R]$. Also, the mean conditional distances to the remaining neighbors are $$\begin{aligned}
\mathbb{E}[R_n|R_1=s]&=&\frac{s}{(n-1)B(N-n+1,n-1)}F_1\left(n-1;n-N,-\frac{1}{d};n-1;1,1-\frac{R^d}{s^d}\right).\end{aligned}$$
Interference
------------
In order to accurately determine network parameters such as outage, throughput or transmission capacity, the interference in the system $I$ needs to be known. The mean interference as seen at the center of the network is given by $$\begin{aligned}
\mu_I&=&\mathbb{E}\left[\sum_{n=1}^N(T_nR_n^{-\alpha})\right]=\sum_{n=1}^N\mathbb{E}[T_n]\mathbb{E}\left[R_n^{-\alpha}\right],\nonumber\\
&=&p\sum_{n=1}^N\mathbb{E}\left[R_n^{-\alpha}\right],\end{aligned}$$ where $T_n\in\{0,1\}$ are i.i.d Bernoulli variables (with parameter $p$) representing whether the $n^{\text{th}}$ nearest node transmits or not in a particular time slot.
Setting $\gamma=-\alpha$ in (\[eqn:moments\]), we can conclude that the mean interference is infinite for $d\leq{}\alpha$. When the number of dimensions is greater than the PLE, we have $$\mu_I=\frac{pR^{-\alpha}\Gamma(N+1)}{\Gamma(N+1-\alpha/d)}\sum_{n=1}^N\frac{\Gamma(n-\alpha/d)}{\Gamma(n)}.$$ One can inductively verify that $$\sum_{n=1}^k\frac{\Gamma(n-\alpha/d)}{\Gamma(n)}=\frac{\Gamma(k-\alpha/d)}{\Gamma(k)}\frac{k-\alpha/d}{1-\alpha/d}\quad\forall{}k\in\mathbb{Z},$$ and we obtain after some simplifications, $$\mu_I=\frac{NpdR^{-\alpha}}{d-\alpha},\quad{}d>\alpha.$$ The unboundedness of the mean interference at practical values of $d$ and $\alpha$ (i.e., $d<\alpha$) actually occurs due to the fact that the path loss model we employ breaks down for very small distances. In fact, it exhibits a singularity at $x=0$. In order to overcome this, some authors employ a modified path loss law given by $\min\{1,\|x\|^{-\alpha}\}$. Employing the modified path loss law, the mean interference is given by [@Srinivasa1] (assuming $R>1$) $$\mu_I=\left \{ \begin{array}{ll}
\frac{Npd}{R^d}\left[\frac{1}{d}+\frac{R^{d-\alpha}-1}{d-\alpha}\right]&\quad{}d\neq{}\alpha\\
\frac{Npd}{R^d}\left[1/d+\ln(R)\right]&\quad{}d=\alpha \end{array}
\right.$$
Connectivity
------------
We now study the connectivity properties of the binomial network. Define a node to be connected to the origin if the SINR at the BS is greater than a threshold $\Theta$. Let the nodes transmit at unit power and assume noise to be AWGN with variance $N_0$. In the absence of interference, the probability that the BS is connected to its $n^{\text{th}}$ nearest neighbor is $$\begin{aligned}
&=&\Pr(R_n^{-\alpha}>N_0\Theta)\nonumber\\
&=&1-\Pr(R_n>(N_0\Theta)^{-1/\alpha})\nonumber\\
&=&\left \{ \begin{array}{ll}
1-I_{1-p'}(N-n+1,n)&\Theta>R^{-\alpha}/N_0\\
1&\Theta\leq{}R^{-\alpha}/N_0,
\end{array} \right.\end{aligned}$$ where $p'=\left(\left(N_0\Theta\right)^{-1/\alpha}/R\right)^d$. Fig. \[fig:connectivity\] plots the connectivity probability in a two-dimensional binomial network with $25$ nodes.
Outage Probability
------------------
An outage $\mathcal{O}$ is defined to occur if the SINR at the BS is lower than a certain threshold $\Theta$. Let the desired transmitter be located at unit distance from the origin, transmit at unit power and also not be a part of the original point process. Assuming that the system is interference-limited, we have $$\begin{aligned}
\Pr(\mathcal{O})&=&\Pr[1/I<\Theta]\nonumber\\
&=&\Pr\left[I>1/\Theta\right].\end{aligned}$$ Considering only the interference contribution from the nearest neighbor to the origin, a simple lower bound is established on the outage probability as $$\begin{aligned}
\Pr(\mathcal{O})&\geq{}&\Pr\left(T_1R_1^{-\alpha}>1/\Theta\right)\nonumber\\
&=&p\Pr\left(R_1<\Theta^{1/\alpha}\right)\nonumber\\
&=&\left \{ \begin{array}{ll}
p\left(1-\left(1-\frac{\Theta^{d/\alpha}}{R^d}\right)^{N}\right)&\Theta\leq{}R^{\alpha}\\
p&\Theta>R^{\alpha}.
\end{array} \right.
\label{eqn:outage_lb}\end{aligned}$$ The empirical values of success probabilities and their upper bounds (\[eqn:outage\_lb\]) are plotted for different values of $N$ in Fig. \[fig:outage\_bound\]. As the plot depicts, the bounds are tight for lower values of $N$ and $\Theta$, and therefore we conclude that the nearest neighbor contributes to a major portion of the network interference. However, note that as $\alpha$ decreases, the bound gets looser since the contributions from the farther neighbors are also increased.
Other Applications
------------------
We now list a few other areas where knowledge of the distance distributions is useful.
- [*Routing*: The question of whether to route over smaller or longer hops is an important, yet a nontrivial issue [@Haenggi2; @Sikora]. The knowledge of internodal distances is necessary for evaluating the optimum hop distance and maximizing the progress of a packet towards its destination.]{}
- [*Path loss exponent estimation*: The issue of PLE estimation is a very important and relevant problem [@Srinivasa2]. Several PLE estimation algorithms are based on received signal strength techniques and these require the knowledge of distances between nodes.]{}
Concluding Remarks
==================
We point out that the Poisson model for nodal distributions in wireless networks is not accurate in many practical situations, and consider the more realistic binomial network model. We derive exact analytical expressions for the pdfs of the distances to the $n^{\text{th}}$ nearest neighbor from the origin in a general $d$-dimensional isotropic random network of radius $R$. We also analytically express the moments of these generalized beta-distributed variables in closed-form. Our findings have applications in several problems related to wireless networks such as energy consumption, connectivity, localization, interference characterization and outage evaluation.
[300]{} T. S. Rappaport, *Wireless Communications - Principles and Practice*, Prentice Hall, 1991. I. F Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, “A survey on sensor networks,” *IEEE Communications Magazine*, Vol. 40, Iss. 8, pp. 102-114, Aug. 2002. D. Stoyan, W. S. Kendall and J. Mecke, “Stochastic geometry and its applications,” Wiley & Sons, 1978. L. E. Miller, “Distribution of link distance in a wireless network,” *Journal of Research of National Institute of Standards and Technology*, Vol. 106, pp. 401-412, Mar./Apr. 2001. C. Rose, “Mean internodal distance in regular and random multihop networks,” *IEEE Transactions on Communications*, Vol. 40, No. 8, pp. 1310-1318, Aug. 1992. M. Haenggi, “On distances in uniformly random networks,” *IEEE Transactions on Information Theory*, Vol. 51, No. 10, pp. 3584-3586, Oct 2005. L. E. Miller, “Joint distribution of link distances,” *Conference on Information Sciences and Systems (CISS 2003)*, Mar. 2003. C-C. Tseng, H-T. Chen and K-C. Chen, “On the distance distributions of the wireless ad hoc networks,” *IEEE Vehicular Technology Conference (VTC 2006-Spring)*, Vol. 2, pp. 772-776, May 2006. S. Vural and E. Ekici, ÒProbability distribution of multihop distance in one-dimensional sensor networks,Ó *Computer Networks Journal (Elsevier)*, Vol. 51, No. 13, pp. 3727-3749, Sep. 2007. P. Kumaraswamy, “A generalized probability density function for double-bounded random processes,” *Journal of Hydrology*, Vol. 46, No. 1/2, pp. 79-88, 1980. R. L. Graham, D. E. Knuth and O. Patashnik, “Concrete mathematics: A foundation for computer science,” 2nd ed., Addison-Wesley, 1994. S. Srinivasa and M. Haenggi, “Modeling Interference in Finite Uniformly Random Networks,” *International Workshop on Information Theory for Sensor Networks (WITS ’07)*, Santa Fe, June 2007. M. Haenggi and D. Puccinelli, “Routing in Ad Hoc Networks: A Case for Long Hops,” *IEEE Communications Magazine*, Vol. 43, Iss. 10, pp. 93-101, Oct. 2005. M. Sikora, J. N. Laneman, M. Haenggi, D. J. Costello, and T. Fuja, “Bandwidth- and Power-Efficient Routing in Linear Wireless Networks,” *Joint Special Issue of IEEE Transactions on Information Theory and IEEE Transactions on Networking*, Vol. 52, Iss. 6, pp. 2624-2633, June 2006. S. Srinivasa and M. Haenggi, “Path Loss Estimation in Large Wireless Networks,” submitted to *IEEE Transactions on Signal Processing*, Jan. 2008. Available at `http://aps.arxiv.org/abs/0802.0351`.
[^1]: Mathematica: PDF\[BetaDistribution\[a, b\],x\].
[^2]: Note that $\gamma\in\mathbb{R}$ in general, and is not restricted to being an integer.
[^3]: Mathematica: Beta\[x, a, b\].
[^4]: Mathematica: AppellF1\[a, b1, b2, c, x, y\].
| {
"pile_set_name": "ArXiv"
} |
`hep-th/0511109`\
`PUTP-2182`\
**$\mathcal{N}=4$ SYM to Two Loops:\
Compact Expressions for the Non-Compact Symmetry Algebra of the ${{\mathfrak{su}}}(1,1|2)$ Sector**
<span style="font-variant:small-caps;">Benjamin I. Zwiebel</span>
*Joseph Henry Laboratories\
Princeton University\
Princeton, NJ 08544, USA*
`[email protected]`
**Abstract**
We begin a study of higher-loop corrections to the dilatation generator of $\mathcal{N}=4$ SYM in non-compact sectors. In these sectors, the dilatation generator contains infinitely many interactions, and therefore one expects very complicated higher-loop corrections. Remarkably, we find a short and simple expression for the two-loop dilatation generator. Our solution for the non-compact ${{\mathfrak{su}}}(1,1|2)$ sector consists of nested commutators of four $\mathcal{O}(g^1)$ generators and one simple auxiliary generator. Moreover, the solution does not require the planar limit; we conjecture that it is valid for any gauge group. To obtain the two-loop dilatation generator, we find the complete $\mathcal{O}(g^3)$ symmetry algebra for this sector, which is also given by concise expressions. We check our solution using published results of direct field theory calculations. By applying the expression for the two-loop dilatation generator to compute selected anomalous dimensions and the bosonic ${{\mathfrak{sl}}}(2)$ sector internal S-matrix, we confirm recent conjectures of the higher-loop Bethe ansatz of `hep-th/0412188`.
Introduction
============
Our understanding of the conjectured AdS-CFT correspondence [@Maldacena:1997re; @Gubser:1998bc; @Witten:1998qj] has deepened dramatically following the observation that strings with large quantum numbers can be mapped to certain subsets of operators in the gauge theory [@Berenstein:2002jq; @Gubser:2002tv]. Direct progress has been made in testing the correspondence perturbatively by comparing anomalous dimensions of gauge theory operators and energies of classical and quantum corrected string solitons [@Frolov:2002av; @Frolov:2003tu; @Beisert:2003xu; @Frolov:2003xy; @Callan:2003xr; @Beisert:2003ea; @Kruczenski:2003gt; @Kruczenski:2004kw; @Callan:2004uv]. For reviews see [@Tseytlin:2003ii; @Beisert:2004yq; @Tseytlin:2004xa; @Zarembo:2004hp; @Swanson:2005wz; @Plefka:2005bk].
These comparisons have used and motivated much progress in computing gauge theory anomalous dimensions. Beginning with the proof of one-loop integrability in the ${{\mathfrak{so}}}(6)$ (scalar) sector[^1] [@Minahan:2002ve], computations of anomalous dimensions in planar $\mathcal{N}$ = 4 SYM have been greatly simplified by mapping single-trace gauge theory operators to states of integrable closed spin chains. This mapping allows the use of a Bethe ansatz. In [@Beisert:2003yb], integrability of the complete one-loop planar gauge theory was proven, and the corresponding Bethe ansatz was presented. After evidence of integrability was obtained via higher-loop computations of the dilatation generator in compact sectors [@Beisert:2003tq; @Beisert:2003ys], higher-loop Bethe ansätze were proposed at various orders and sectors [@Serban:2004jf; @Beisert:2004hm; @Staudacher:2004tk]. This line of research culminated with a proposal for the all-loop asymptotic ${{\mathfrak{psu}}}(2,2|4)$ Bethe ansatz [@Beisert:2005fw]. A proof of this ansatz assuming integrability was found recently [@Beisert:2005tm]. Parallel developments have occurred on the string side. Classical string theory in $AdS_5 \times S^5$ was shown to be integrable [@Mandal:2002fs; @Bena:2003wd], integrability was used to solve the classical spectrum in terms of algebraic curves [@Kazakov:2004qf; @Beisert:2005bm], and Bethe ansätze for quantum strings were proposed [@Arutyunov:2004vx; @Staudacher:2004tk; @Beisert:2005fw].
Despite all this progress and much evidence, there is no rigorous proof of higher-loop integrability for any non-compact sector of the gauge theory. Finding the dilatation generator would be a first step towards a proof. However, even that step seems intractable for non-compact sectors. Completing a direct diagrammatic calculation is realistic only at low loop order, as was done at two loops for the fermionic ${{\mathfrak{sl}}}(2)$ sector in [@Belitsky:2005bu]. A more promising approach to higher loops is Beisert’s method, which takes full advantage of superconformal symmetry. However, extending Beisert’s method of computing the dilatation generator from one loop for the complete theory [@Beisert:2003jj; @Beisert:2004ry] to three loops for the ${{\mathfrak{su}}}(2|3)$ sector [@Beisert:2003ys], depended on compactness. A large computer algebra computation was essential. Because the dilatation generator in non-compact sectors is built from infinitely many interactions, a brute force computation of this kind becomes impossible.
In this paper, we overcome this obstacle by developing techniques for higher-loop non-compact sectors. Only using constraints from Feynman diagrammatics and superconformal symmetry, we compute the two-loop dilatation operator for the non-compact ${{\mathfrak{su}}}(1,1|2)$ sector. We also find the corrections to this sector’s symmetry algebra up to $\mathcal{O}(g^3)$. We introduce an auxiliary generator that satisfies special commutation relations with the classical and half-loop symmetry generators. This extension of the symmetry algebra enables us to find and verify solutions of the symmetry constraints at one and one-half loops only using the commutation relations of the extended algebra at zero and at one-half loops. Via our method, higher-loop computations reduce to straightforward algebraic manipulations of commutators. This can be done efficiently even without a computer.
For our computation, it is essential that the ${{\mathfrak{su}}}(1,1|2)$ sector has a hidden ${{\mathfrak{psu}}}(1|1)^2$ symmetry [@Beisert:2004ry], which adds tight restrictions. The representation for the ${{\mathfrak{psu}}}(1|1)^2$ symmetry is trivial at leading order and has an expansion in odd powers of $g$, which is proportional to $\sqrt{\lambda}$. The generators, labeled $\overrightarrow{{{\mathfrak{T}}}}$ and $\overleftarrow{{{\mathfrak{T}}}}$,[^2] change the length of the spin chain, reflecting the dynamic aspect of the full ${{\mathfrak{psu}}}(2,2|4)$ spin chain. $\overrightarrow{{{\mathfrak{T}}}}$ and $\overleftarrow{{{\mathfrak{T}}}}$ anti-commute to the dilatation operator. Therefore, we only need one and one-half loops to find the two-loop dilatation operator. More precisely, using a subscript $n$ for the $\mathcal{O}(g^n)$ corrections, $${{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}_4= {\left\{\overrightarrow{{{\mathfrak{T}}}_3},\overleftarrow{{{\mathfrak{T}}}_1}\right\}}+{\left\{\overrightarrow{{{\mathfrak{T}}}_1},\overleftarrow{{{\mathfrak{T}}}_3}\right\}}.$$
The $\mathcal{O}(g^3)$ solution is built only from the leading (non-vanishing) order representations of the generators and the auxiliary generator, ${{\mathfrak{h}}}$, that acts using the harmonic numbers. Schematically, we have[^3] $$\begin{gathered}
{{\mathfrak{T}}}_3 = \pm {\left[{{\mathfrak{T}}}_1,{{\mathfrak{x}}}\right]}, \notag \\ {{\mathfrak{x}}}\sim {\left\{\overrightarrow{{{\mathfrak{T}}}_1},{\left[\overleftarrow{{{\mathfrak{T}}}_1},{{\mathfrak{h}}}\right]}\right\}}. \end{gathered}$$ It follows that ${\delta{\mathfrak{D}}}_4$ is built only from the ${{\mathfrak{T}}}_1$’s and ${{\mathfrak{h}}}$.
Our solution lifts consistently and naturally to non-planar $\mathcal{N}=4$ gauge theory as well[^4], that is for any choice of the gauge group. In particular, it includes wrapping interactions. These are non-planar contributions to the dilatation generator that survive the planar limit by wrapping around short operators [@Beisert:2004hm]. Because supersymmetry relates shorter states to longer states, wrapping interactions do not contribute until four loops [@Beisert:2003ys; @Beisert:2005fw]. However, generalizing this solution for wrapping interactions to higher loops would provide a missing piece for the comparisons between gauge and string theory. The proposed Bethe ansätze are oblivious to corrections from wrapping interactions[^5].
We also find compelling evidence that integrability persists at two loops in non-compact sectors. The two-loop dilatation generator generates the same bosonic ${{\mathfrak{sl}}}(2)$ subsector S-matrix as assumed for the Bethe ansatz in [@Staudacher:2004tk], and the same anomalous dimensions as computed using this Bethe ansatz in [@Staudacher:2004tk]. A complementary approach to confirming this Bethe ansatz will appear in [@Eden:2005xx]. Evidence for two-loop integrability in the fermionic ${{\mathfrak{sl}}}(2)$ sector was given in [@Belitsky:2005bu].
Section \[sec:psu112\] introduces the $ {{\mathfrak{su}}}(1,1|2)$ sector and the residual symmetry algebra. Section \[sec:g2\] discusses the $\mathcal{O}(g^2)$ solution, and Section \[sec:g3\] discusses the $\mathcal{O}(g^3)$ solution and presents the two-loop dilatation generator. For simplicity, we assume planarity until Section \[sec:non-planar\], and in that section we present the lift to the finite-$N$ solution. In Section \[sec:checks\], after verifying that our solution predicts the same anomalous dimensions as those of the field theory calculations of [@Beisert:2003tq; @Kotikov:2003fb; @Eden:2005bt; @Belitsky:2005bu], we summarize the applications of the solution to compute the bosonic ${{\mathfrak{sl}}}(2)$ subsector S-matrix and some anomalous dimensions. We conclude and discuss directions for further research in Section \[sec:outlook\]. The appendix presents details about the symmetry algebra and proofs of our solution. Finally, we use many results and notations of [@Beisert:2004ry].
The $ {{\mathfrak{su}}}(1,1|2)$ sector \[sec:psu112\]
=====================================================
As explained in [@Beisert:2004ry], it is consistent to restrict to various sectors of the states of $\mathcal{N}=4$ SYM. Under such a restriction, the full ${{\mathfrak{psu}}}(2,2|4)$ algebra splits into three components. One component annihilates all the states in the subsector, and the second component maps states in the subsector out of the subsector. The third component, a subalgebra, acts within the subsector non-trivially, and gives the sector its name. For the case we consider in this work, this subalgebra is ${{\mathfrak{u}}}(1)^2 \ltimes ({{\mathfrak{psu}}}(1,1|2) \times {{\mathfrak{psu}}}(1|1)^2) \ltimes {{\mathfrak{u}}}(1)$. The ${{\mathfrak{u}}}(1)^2$ consists of two external automorphisms, the length $L$ and the hypercharge $B$. The ${{\mathfrak{u}}}(1)$, is the quantum correction to the dilatation generator, ${\delta{\mathfrak{D}}}$, which appears as the central charge for both ${{\mathfrak{psu}}}(1,1|2)$ and ${{\mathfrak{psu}}}(1|1)^2$.[^6] Furthermore, the ${{\mathfrak{psu}}}(1|1)^2$ acts trivially classically. Now we describe the restriction to this subsector and the corresponding symmetry algebra, and we present the leading non-vanishing actions of the algebra on the states of the subsector.
The restriction to the $ {{\mathfrak{su}}}(1,1|2)$ sector {#sec:restriction}
---------------------------------------------------------
To restrict to this sector, we must set the classical dimensions of states simultaneously equal to the following linear combinations of the eigenvalues of the Cartan generators of the $ {{\mathfrak{psu}}}(2,2|4)$ algebra [@Beisert:2004ry], $$\begin{aligned}
\label{eq:restriction}
D_0 &=& s_1+\frac{1}{2}q_2+p+\frac{3}{2}q_1=s_2+\frac{1}{2}q_1+p+\frac{3}{2}q_2 \end{aligned}$$ Here $[q_1, \, p, \, q_2]$ are the Dynkin labels of the ${{\mathfrak{su}}}(4)$ subalgebra and $[s_1, \, s_2]$ are the Dynkin labels of the Lorentz algebra. Combined with the bounds given by the field content, this implies that $D_0$ also satisfies $$\begin{aligned}
D_0 &=& L-B+s_1 = L+B+s_2.\end{aligned}$$
The ${{\mathfrak{psu}}}(1,1|2)$ algebra \[sec:psu112alg\]
---------------------------------------------------------
We introduce a notation for the subset of the ${{\mathfrak{psu}}}(2,2|4)$ generators that generates the ${{\mathfrak{psu}}}(1,1|2)$ algebra, $$\begin{gathered}
\label{eq:pSU112Gen}
{{\mathfrak{J}}}^0(g)=- {{\mathfrak{L}}}+2{{\mathfrak{D}}}_0+{\delta{\mathfrak{D}}}(g), \quad
{{\mathfrak{R}}}^0={{\mathfrak{R}}}^2_2-{{\mathfrak{R}}}^3_3, \notag \\ {{\mathfrak{J}}}^{++}(g)={{\mathfrak{P}}}_{22}(g),\quad
{{\mathfrak{J}}}^{--}(g)={{\mathfrak{K}}}^{22}(g),\quad
{{\mathfrak{R}}}^{\uparrow\uparrow}(g)={{\mathfrak{R}}}^3_2, \quad
{{\mathfrak{R}}}^{\downarrow\downarrow}(g)={{\mathfrak{R}}}^2_3, \notag \\
\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)={{\mathfrak{Q}}}^{2}_2(g),\quad
\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)={{\mathfrak{S}}}_{2}^2(g),\quad
\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)={{\mathfrak{Q}}}^{3}_2(g),\quad
\overleftarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)={{\mathfrak{S}}}_{3}^2(g),\notag \\
\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)=\dot{{{\mathfrak{Q}}}}_{22}(g),\quad
\overrightarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)=\dot{{{\mathfrak{S}}}}^{22}(g), \quad
\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)=\dot{{{\mathfrak{Q}}}}_{23}(g),\quad
\overrightarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)=\dot{{{\mathfrak{S}}}}^{32}(g).\end{gathered}$$
${{\mathfrak{L}}}$ is the length operator; it multiplies a state composed of $L$ fundamental fields by $L$. ${{\mathfrak{L}}}$ commutes with all of the ${{\mathfrak{psu}}}(1,1|2)$ generators. In this sector, it satisfies $${{\mathfrak{L}}}={{\mathfrak{L}}}^2_2-\dot{{{\mathfrak{L}}}}^2_2-2{{\mathfrak{R}}}^1_1.$$ ${{\mathfrak{L}}}^\alpha_\beta$ and $\dot{{{\mathfrak{L}}}}^{\dot{\alpha}}_{\dot{\beta}}$ are Lorentz rotations. ${{\mathfrak{D}}}_0$ is the classical dilatation generator, and ${\delta{\mathfrak{D}}}$ is its quantum correction. ${\delta{\mathfrak{D}}}$’s leading term is at $\mathcal{O}(g^2)$, and its expansion includes even powers of $g$ only. The remaining generators appearing on the right side of (\[eq:pSU112Gen\]) are the following ${{\mathfrak{psu}}}(2,2|4)$ generators: ${{\mathfrak{su}}}(4)$ internal R-symmetry rotation generators ${{\mathfrak{R}}}$, Lorentz translations and boosts ${{\mathfrak{P}}}$ and ${{\mathfrak{K}}}$, and fermionic supertranslations and superboosts ${{\mathfrak{Q}}}$, $\dot{{{\mathfrak{Q}}}}$, ${{\mathfrak{S}}}$, and $\dot{{{\mathfrak{S}}}}$. Appendix D of [@Beisert:2004ry] gives a complete description of the full algebra.
Note that the superscript signs correspond to ${{\mathfrak{su}}}(1,1)$ charge (descended from the ${{\mathfrak{su}}}(2,2)$ Lorentz subalgebra). A generator adds dimension equal to ${{{\textstyle\frac{1}{2}}}}$ (-${{{\textstyle\frac{1}{2}}}}$) of the number of its plus (minus) signs. Similarly, vertical arrows correspond to integer ${{\mathfrak{su}}}(2)$ R-charge, and horizontal arrows correspond to half-integer hypercharge, ${{\mathfrak{B}}}$. Throughout this paper we will work in a basis such that hermitian conjugation requires switching signs and reversing arrows simultaneously. Generators without any arrows or a sign are hermitian.
The ${{\mathfrak{psu}}}(1,1|2)$ algebra is given in Appendix \[sec:psu112comm\]. However, many commutators[^7] can be inferred directly from the notation because the three types of charges are conserved. These conservation rules immediately imply that many commutators vanish. Also, ${{\mathfrak{J}}}^0$ measures ${{\mathfrak{su}}}(1,1)$ charge, and ${{\mathfrak{R}}}^0$ measures ${{\mathfrak{su}}}(2)$ charge.
The ${{\mathfrak{psu}}}(1|1)^2$ algebra {#sec:psu11^2alg}
---------------------------------------
The ${{\mathfrak{psu}}}(1|1)^2$ algebra is generated by $$\label{eq:psu11^2Gen}
\overrightarrow{{{\mathfrak{T}}}}^+(g)=\dot{{{\mathfrak{Q}}}}^{14}(g),\quad
\overleftarrow{{{\mathfrak{T}}}}^-(g)=\dot{{{\mathfrak{S}}}}_{14}(g), \quad
\overrightarrow{{{\mathfrak{T}}}}^-(g)={{\mathfrak{S}}}^1_1(g),\quad
\overleftarrow{{{\mathfrak{T}}}}^+(g)={{\mathfrak{Q}}}^1_1(g).$$ As before, the horizontal arrows correspond to hypercharge. However, these generators carry no ${{\mathfrak{su}}}(1,1)$ charge, and the sign corresponds instead to their commutators with the external automorphism ${{\mathfrak{L}}},$ $$\label{eq:length}
{\left[{{\mathfrak{L}}},{{\mathfrak{T}}}^{\pm}(g)\right]}=\pm {{\mathfrak{T}}}^{\pm}(g).$$
The non-zero commutators are $$\begin{aligned}
\label{eq:psU11^2Alg}
{\left\{\overrightarrow{{{\mathfrak{T}}}}^+(g),\overleftarrow{{{\mathfrak{T}}}}^-(g)\right\}}&={{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}(g), & {\left\{\overrightarrow{{{\mathfrak{T}}}}^-(g),\overleftarrow{{{\mathfrak{T}}}}^+(g)\right\}}&={{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}(g).\end{aligned}$$ Any commutator between non-conjugate ${{\mathfrak{T}}}$’s vanishes, including the squares of the ${{\mathfrak{T}}}$. The product structure of the full symmetry algebra will be used many times in the rest of this work: the generators of ${{\mathfrak{psu}}}(1,1|2)$ and ${{\mathfrak{psu}}}(1|1)^2$ commute with each other.
Fields and states {#sec:States}
-----------------
The fields in this sector are derivatives ${\mathcal{D}}={\mathcal{D}}_{22}$ acting on the fermions $\overrightarrow{\psi}=\Psi_{42}$ and $\overleftarrow{\psi}=\dot{\Psi}^1_2$, or the bosons $\phi^\downarrow=\Phi_{34}$ and $\phi^\uparrow=\Phi_{24}$. We denote $k$ derivatives by a subscript $k$, $$\phi^\updownarrow_k\sim{\mathcal{D}}^k \phi^\updownarrow, \quad \overleftrightarrow{\psi_k}\sim{\mathcal{D}}^k \overleftrightarrow{\psi}, \qquad (k \geq 0).$$
The representation of the symmetry algebra acts on a spin chain. The states of the spin chain are tensor products $${|X_1X_2 \ldots X_n\rangle} \quad \mbox{where} \quad X_i \in \left\{\phi^\updownarrow_k, \, \overleftrightarrow{\psi}_k \right\}.$$ A generic state is a linear combination of these tensor products, with the cyclic identification $${|X_1 \ldots X_i X_{i+1} \ldots X_n\rangle}= (-1)^{(X_1\ldots X_i)(X_{i+1} \ldots X_n)} {|X_{i+1} \ldots X_n X_1 \ldots X_i\rangle}$$ $(-1)^{AB}$ is $-1$ if both $A$ and $B$ are fermionic, and $1$ otherwise.
The leading order representation {#sec:Reps}
--------------------------------
For the leading order representation of the symmetry algebra, at $\mathcal{O}(g^0)$, the generators have one-site to one-site vertices. That is, the generators’ action on the spin chain is a tensor product (given by the sum of its action on each individual site). Also, every time a fermionic generator passes a fermionic field, we must add a factor of $-1$. The non-vanishing actions for ${{\mathfrak{J}}}^{++}$, $\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}$, and ${{\mathfrak{R}}}^{\uparrow\uparrow}$ are $$\begin{aligned}
\label{eq:FieldGen}
{{\mathfrak{J}}}^{++}_0 {|\phi^\updownarrow_k\rangle}&=(k+1){|\phi^\updownarrow_{k+1}\rangle},&
{{\mathfrak{J}}}^{++}_0 {|\overleftrightarrow{\psi}_k\rangle}&=\sqrt{(k+1)(k+2)}{|\overleftrightarrow{\psi}_{k+1}\rangle},
\notag \\
\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow} {|\phi^\downarrow_k\rangle}&=\sqrt{k+1}{|\overrightarrow{\psi}_k\rangle},&
\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow} {|\overleftarrow{\psi}_k\rangle}&=-\sqrt{k+1}{|\phi^\uparrow_{k+1}\rangle}, \notag \\
{{\mathfrak{R}}}^{\uparrow\uparrow}{|\phi^\downarrow_k\rangle}&={|\phi^\uparrow_k\rangle}.\end{aligned}$$ The remaining parts of the leading order representation can be computed using hermitian conjugation and the algebra given in Appendix \[sec:psu112comm\]. To find hermitian conjugates, it is simplest to use the (tensor product of the) diagonal metric $$\langle \phi^\downarrow_m {|\phi^\downarrow_n\rangle} = \delta_{mn}, \quad \langle \phi^\uparrow_m {|\phi^\uparrow_n\rangle} = \delta_{mn}, \quad \langle{\overrightarrow{\psi}_m} {|\overrightarrow{\psi}_n\rangle} = \delta_{mn}, \quad \langle \overleftarrow{\psi}_m {|\overleftarrow{\psi}_n\rangle} = \delta_{mn}.$$ Then the condition for two operators, ${{\mathfrak{J}}}$ and ${{\mathfrak{J}}}^\dagger$, to be hermitian conjugates is $$\begin{gathered}
{\langleX'_1 \ldots X'_m|}{{\mathfrak{J}}}{|X_1\ldots X_n\rangle}= {\langleX_1\ldots X_n|}{{\mathfrak{J}}}^\dagger {|X'_1 \ldots X'_m\rangle}. \end{gathered}$$
Constraints from Feynman rules
------------------------------
Before beginning to discuss quantum corrections to the symmetry algebra, we review the constraints from Feynman diagrams [@Beisert:2004ry]. Most simply, only connected interactions appear. For the planar theory, this implies that interactions will involve replacing a set of adjacent spins of the spin chain with a new set, not necessarily of the same length. In fact, power counting implies that the $\mathcal{O}(g^n)$ term of a generator is the sum of interactions involving a total of up to $(n+2)$ initial and final spin sites. Therefore, quantum corrections will deform the representation from the tensor product structure described in the previous subsection. The symmetry algebra for the ${{\mathfrak{su}}}(1,1|2)$ sector implies that for even n, the length of the spin chain is unchanged, and for odd n, interactions change the length of the spin chain by one.
Feynman rules restrict to parity (or charge conjugation) even interactions. In terms of the spin chain, a parity even generator satisfies, for arbitrary fields $X_j$, $${{\mathfrak{J}}}{|X_1 \ldots X_i\rangle} = {|X^\prime_1 \dots X^\prime_f\rangle} \Rightarrow (-1)^{i+f_i(f_i-1)/2} {{\mathfrak{J}}}{|X_i \ldots X_1\rangle} = (-1)^{f+f_f(f_f-1)/2}{|X^\prime_f \dots X^\prime_i\rangle}$$ where $f_i$ and $f_f$ are the number of fermions in the initial and final states.
Finally, we use the normalization for the coupling constant $$g^2=\frac{{g_{\scriptscriptstyle\mathrm{YM}}}^2 N}{8\pi^2},$$ where $N$ is the rank of the gauge group.
The $\mathcal{O}(g^1)$ solution
-------------------------------
Applying the rules from the previous subsection, at $\mathcal{O}(g^1)$, the corrections will entail replacing one spin site with two, or vice-versa. For instance, a one-site to two-site generator ${{\mathfrak{T}}}$ would act on a generic state as: $$\begin{gathered}
{{\mathfrak{T}}}{|X_1 \ldots X_i \ldots X_n\rangle} = \notag \\
\sum_{a, \, b} c^{X_1}_{ab} {|Y_a Y_b X_2 \ldots X_i \ldots X_n\rangle} + \cdots + \notag \\
(-1)^{(X_1\ldots X_{i-1}){{\mathfrak{T}}}} \sum_{a,\,b} c^{X_i}_{ab} {|X_1 \ldots X_{i-1} Y_a Y_b X_{i+1} \ldots X_n\rangle} + \cdots, \end{gathered}$$ where $Y_a$ and $Y_b$ run over all fields. For fixed $i$, only a finite number of the $c^{X_i}_{ab}$ will be non-zero.
Starting at this order, it is possible to construct interactions that vanish on closed spin chain states because of the cyclic identification. These correspond to gauge transformations in gauge theory. An example of a one-site to two-site gauge transformation satisfies, for all fields $X_i$ and a fixed field $Y$, $${{\mathfrak{T}}}{|X_i\rangle} = {|X_i Y\rangle} - (-1)^{Y X_i} {|Y X_i\rangle}.$$ In fact, the symmetry algebra can be satisfied at this order only up to gauge transformations [@Beisert:2004ry].
Only the ${{\mathfrak{psu}}}(1|1)^2$ generators receive $\mathcal{O}(g)$ corrections since only they change the length of the spin chain. The action of $\overrightarrow{{{\mathfrak{T}}}}_1^{+}$ on states composed of the $\overrightarrow{\psi}_k$ was derived in [@Beisert:2004ry], and it is straightforward to generalize to the full solution by requiring commutation with the ${{\mathfrak{psu}}}(1,1|2)$ algebra (up to gauge transformations).
$$\begin{aligned}
\overrightarrow{{{\mathfrak{T}}}_1}^{+}{|\phi^\updownarrow_m\rangle} &=& \frac{1}{\sqrt{2}}\sum_{k=0}^{m-1} \frac{1}{\sqrt{k+1}}\left({|\overrightarrow{\psi_k}\phi^\updownarrow_{m-1-k}\rangle}-{|\phi^\updownarrow_{m-1-k}\overrightarrow{\psi_k}\rangle}\right), \notag \\
\overrightarrow{{{\mathfrak{T}}}_1}^{+} {|\overrightarrow{\psi_m}\rangle} &=& \frac{1}{\sqrt{2}}\sum_{k=0}^{m-1} \sqrt{\frac{m+1}{(k+1)(m-k)}} {|\overrightarrow{\psi_k}\overrightarrow{\psi_m}_{-1-k}\rangle}, \notag \\
\overrightarrow{{{\mathfrak{T}}}_1}^{+}{|\overleftarrow{\psi_m}\rangle} &=& \frac{1}{\sqrt{2}}\sum_{k=0}^{m-1} \sqrt{\frac{m-k}{(k+1)(m+1)}}\left({|\overrightarrow{\psi_k}\overleftarrow{\psi_m}_{-1-k}\rangle}+{|\overleftarrow{\psi_m}_{-1-k}\overrightarrow{\psi_k}\rangle}\right) \notag \\
&-& \frac{1}{\sqrt{2(m+1)}}\sum_{k=0}^{m} \left({|\phi^\downarrow_k\phi^\uparrow_{m-k}\rangle}-{|\phi^\uparrow_k\phi^\downarrow_{m-k}\rangle}\right). \end{aligned}$$
Up to a single minus sign in the last line, $\overleftarrow{{{\mathfrak{T}}}_1}^{+}$ follows from switching $\overrightarrow{\psi_k}$ and $\overleftarrow{\psi_k}$ in the above expression for $\overrightarrow{{{\mathfrak{T}}}_1}^{+}$, $$\begin{aligned}
\overleftarrow{{{\mathfrak{T}}}_1}^{+}{|\phi^\updownarrow_m\rangle} &=& \frac{1}{\sqrt{2}}\sum_{k=0}^{m-1} \frac{1}{\sqrt{k+1}}\left({|\overleftarrow{\psi_k}\phi^\updownarrow_{m-1-k}\rangle}-{|\phi^\updownarrow_{m-1-k}\overleftarrow{\psi_k}\rangle}\right), \notag \\
\overleftarrow{{{\mathfrak{T}}}_1}^{+} {|\overleftarrow{\psi_m}\rangle} &=& \frac{1}{\sqrt{2}}\sum_{k=0}^{m-1} \sqrt{\frac{m+1}{(k+1)(m-k)}} {|\overleftarrow{\psi_k}\overleftarrow{\psi_m}_{-1-k}\rangle}, \notag \\
\overleftarrow{{{\mathfrak{T}}}_1}^{+}{|\overrightarrow{\psi_m}\rangle} &=& \frac{1}{\sqrt{2}}\sum_{k=0}^{m-1} \sqrt{\frac{m-k}{(k+1)(m+1)}}\left({|\overleftarrow{\psi_k}\overrightarrow{\psi_m}_{-1-k}\rangle}+{|\overrightarrow{\psi_m}_{-1-k}\overleftarrow{\psi_k}\rangle}\right) \notag \\
&+& \frac{1}{\sqrt{2(m+1)}}\sum_{k=0}^{m} \left({|\phi^\downarrow_k\phi^\uparrow_{m-k}\rangle}-{|\phi^\uparrow_k\phi^\downarrow_{m-k}\rangle}\right). \label{eq:leftT1+} \end{aligned}$$ The other two ${{\mathfrak{psu}}}(1|1)^2$ generators, the ${{\mathfrak{T}}}^-$, can then be computed at this order via hermitian conjugation. Up to gauge transformations, and a rescaling of $g$, this solution is completely fixed by the symmetry constraints.
Order $g^2$ \[sec:g2\]
======================
At $\mathcal{O}(g^2)$, ${\delta{\mathfrak{D}}}$ and the ${{\mathfrak{psu}}}(1,1|2)$ generators receive quantum corrections. From (\[eq:psU11\^2Alg\]), it is straightforward to compute $${\delta{\mathfrak{D}}}_2= 2 {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}}=2 {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,\overleftarrow{{{\mathfrak{T}}}_1}^+\right\}}.$$ As first shown in [@Beisert:2003jj], ${\delta{\mathfrak{D}}}_2$ acts by projecting two adjacent sites onto modules of definite ${{\mathfrak{psu}}}(2,2|4)$ “spin” $j$ with coefficient $h(j)$. $h$ gives the harmonic numbers $$h(k) = \sum_{k'=1}^{k}\frac{1}{k'}=\psi(k+1)-\psi(1).$$ The harmonic numbers will play an essential role in ${\delta{\mathfrak{D}}}_4$ as well.
${\delta{\mathfrak{D}}}$ is the only generator we need to compute because, as we now explain, once we know ${\delta{\mathfrak{D}}}$ the full ${{\mathfrak{psu}}}(1,1|2) \times {{\mathfrak{psu}}}(1|1)^2$ algebra’s action is fixed by group theory. Knowing ${\delta{\mathfrak{D}}}$ means knowing its eigenstates and eigenvalues. Multiplets are then formed by states of equal eigenvalues, and the generators of ${{\mathfrak{psu}}}(1,1|2) \times {{\mathfrak{psu}}}(1|1)^2$ must connect the states of a multiplet with factors determined by group theory.
However, for our method of computing ${\delta{\mathfrak{D}}}$, it is essential to compute the perturbative corrections to the ${{\mathfrak{psu}}}(1,1|2)$ generators. They are needed for constraining the one and one-half loop ${{\mathfrak{psu}}}(1|1)^2$ generators, which anti-commute to the two-loop dilatation generator. Moreover, the solution we present below for the ${{\mathfrak{psu}}}(1,1|2)$ generators has an interesting and simple structure. Next we present this solution and discuss its possible modifications and its proof.
The solution \[sec:g2solution\]
-------------------------------
We define two auxiliary generators that play central parts in our solution. ${{\mathfrak{h}}}$ is a one-site generator of harmonic numbers. Its action is $$\begin{aligned}
{{\mathfrak{h}}}{|\phi^\updownarrow_k\rangle}&={{{\textstyle\frac{1}{2}}}}h(k){|\phi^\updownarrow_k\rangle},& {{\mathfrak{h}}}{|\overleftrightarrow{\psi_k}\rangle}&= {{{\textstyle\frac{1}{2}}}}h(k+1){|\overleftrightarrow{\psi_k}\rangle}. \end{aligned}$$ ${{\mathfrak{x}}}$ is a two-site to two-site generator that we can write in two equivalent ways, $$\begin{aligned}
\label{eq:definex} {{\mathfrak{x}}}&=& {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}-{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{h}}}\right]}\right\}} \nonumber \\
&=& {\left\{\overrightarrow{{{\mathfrak{T}}}}^-_1,{\left[\overleftarrow{{{\mathfrak{T}}}}^+_1,{{\mathfrak{h}}}\right]}\right\}}-{\left\{\overrightarrow{{{\mathfrak{T}}}}^+_1,{\left[\overleftarrow{{{\mathfrak{T}}}}^-_1,{{\mathfrak{h}}}\right]}\right\}}. \end{aligned}$$ The equality in (\[eq:definex\]) follows from (\[eq:psU11\^2Alg\]), since $${\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}+{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{h}}}\right]}\right\}} = {{{\textstyle\frac{1}{2}}}}{\left[{\delta{\mathfrak{D}}}_2,{{\mathfrak{h}}}\right]},$$ and the analogous equation for $\overrightarrow{{{\mathfrak{T}}}_1}^-$ and its conjugate is satisfied. Because of this equality in (\[eq:definex\]), ${{\mathfrak{x}}}$ is hermitian.
Let ${{\mathfrak{X}}}^\pm$ represent ${{\mathfrak{J}}}^{++}$, ${{\mathfrak{J}}}^{--}$, or the eight ${{\mathfrak{Q}}}$’s, where we retain only the ${{\mathfrak{su}}}(1,1)$ charge. Then the solution of the symmetry and Feynman diagram constraints is $$\label{eq:g2solution} {{\mathfrak{X}}}^\pm_2=\pm {\left[{{\mathfrak{X}}}^\pm_0,{{\mathfrak{x}}}\right]} +{\left[{{\mathfrak{X}}}^\pm_0,{{\mathfrak{y}}}\right]}.$$ We present an outline of the proof that this satisfies the algebra relations in section \[sec:g2discussion\] and more details in Appendix \[sec:g2proof\].
${{\mathfrak{y}}}$ is a two-site to two-site generator that commutes with ${{\mathfrak{B}}}$, ${{\mathfrak{D}}}_0$ and the ${{\mathfrak{R}}}$’s. Commuting all the generators of the ${{\mathfrak{su}}}(1,1|2) \times {{\mathfrak{psu}}}(1|1)^2$ algebras with a generator such as ${{\mathfrak{y}}}$ maps one solution of the commutation relations to another. It corresponds to the first term in the expansion of the similarity transformation $${{\mathfrak{J}}}\mapsto U {{\mathfrak{J}}}U^{-1}, \quad U = 1 + g^2 {{\mathfrak{y}}}+ \cdots, \quad \mbox{i.e.} \quad {{\mathfrak{J}}}_2 \mapsto {{\mathfrak{J}}}_2 + {\left[{{\mathfrak{J}}}_0,{{\mathfrak{y}}}\right]}.$$ We require ${{\mathfrak{y}}}$ to commute with the ${{\mathfrak{R}}}$’s and with ${{\mathfrak{B}}}$ and ${{\mathfrak{D}}}_0$ to preserve ${\delta{\mathfrak{D}}}$’s manifest R-symmetry and ${\delta{\mathfrak{D}}}$’s eigenstates’ hypercharge and classical dimension assignments. To maintain manifest consistency with the Feynman diagram rules, $U$’s expansion must be in even powers of $n$, consisting of $(\frac{n}{2}+1)$-site to $(\frac{n}{2}+1)$-site interactions. For anti-hermitian (or vanishing) ${{\mathfrak{y}}}$, ${{\mathfrak{X}}}^+$and ${{\mathfrak{X}}}^-$ are hermitian conjugates up to $\mathcal{O}(g^2)$.
Freedom for the $\mathcal{O}(g^2)$ solution
-------------------------------------------
There are two possible sources of freedom for the solution at this order: interactions that vanish on cyclic states (gauge transformations) and homogeneous solutions. We now exclude the former and discuss the latter.
The requirement of even parity rules out the possibility of applying gauge transformations to the solution at this order, since generators are sums of two-site to two-site interactions. This also implies that the algebra is satisfied exactly (not just modulo gauge transformations).
However, at this point, we cannot rule out modification by a homogeneous solution. Under this modification, $${{\mathfrak{J}}}^{++}_2 \mapsto {{\mathfrak{J}}}^{++}_2 + \delta{{\mathfrak{J}}}^{++}_2, \quad \overleftrightarrow{{{\mathfrak{Q}}}_2}^{+\updownarrow} \mapsto \overleftrightarrow{{{\mathfrak{Q}}}_2}^{+\updownarrow} + \delta \overleftrightarrow{{{\mathfrak{Q}}}_2}^{+\updownarrow},$$ and similarly for the hermitian conjugates. In order for the symmetry constraints to remain satisfied, the $\delta {{\mathfrak{J}}}$’s and $\delta {{\mathfrak{Q}}}$’s must not contribute to any commutator of the algebra. For example, $${\left[\delta{{\mathfrak{J}}}^{++}_2,{{\mathfrak{J}}}^{--}_0\right]}+ {\left[{{\mathfrak{J}}}^{++}_0,\delta{{\mathfrak{J}}}^{--}_2\right]} = 0.$$ We have not found any non-trivial homogeneous solutions, or ruled them out. However, from the above discussion regarding ${\delta{\mathfrak{D}}}$, we conclude that once ${\delta{\mathfrak{D}}}_4$ is found, this freedom is fixed. We will find the solution for ${\delta{\mathfrak{D}}}_4$ below. Since the $\mathcal{O}(g^2)$ solution presented in this section is consistent with it, this is the field theory $\mathcal{O}(g^2)$ solution.
Discussion of the proof of the $\mathcal{O}(g^2)$ solution \[sec:g2discussion\]
-------------------------------------------------------------------------------
To prove the solution we must check that the commutators given in Appendix \[sec:psu112comm\] are satisfied at $\mathcal{O}(g^2)$. The proof is based upon substitution of the solution, elementary algebra, and keeping track of powers of $g$. The commutators up to $\mathcal{O}(g^1)$ and the identities presented in Appendix \[sec:g2properties\] are also needed. In particular, many of the commutators are simplified since the entire solution is hermitian and given by commutators of leading order generators and ${{\mathfrak{x}}}$. In Appendix \[sec:g2proof\] we verify a representative set of commutators. Here we do one example in detail.
The commutator of the conjugate ${{\mathfrak{J}}}$’s $${\left[{{\mathfrak{J}}}^{--},{{\mathfrak{J}}}^{++}\right]}={{\mathfrak{J}}}^0$$ has the $\mathcal{O}(g^2)$ component $${\left[{{\mathfrak{J}}}^{--},{{\mathfrak{J}}}^{++}\right]}_2={\left[{{\mathfrak{J}}}^{--}_2,{{\mathfrak{J}}}^{++}_0\right]}+ {\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{J}}}^{++}_2\right]}={{\mathfrak{J}}}^0_2={\delta{\mathfrak{D}}}_2.$$ Expanding the solution for ${{\mathfrak{J}}}^{++}_2$ yields[^8] $$\begin{aligned}
{{\mathfrak{J}}}^{++}_2&=&{\left[{{\mathfrak{J}}}^{++}_0,{{\mathfrak{x}}}\right]} \nonumber \\
&=& {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{j}}}^{++}\right]}\right\}}-{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{j}}}^{++}\right]}\right\}},\end{aligned}$$ where we have defined $${{\mathfrak{j}}}^{++}={\left[{{\mathfrak{J}}}_0^{++},{{\mathfrak{h}}}\right]}.$$ Also, after defining $${{\mathfrak{j}}}^{--}=-{\left[{{\mathfrak{J}}}_0^{--},{{\mathfrak{h}}}\right]},$$ direct computation shows that $${\left[{{\mathfrak{J}}}_0^{--},{{\mathfrak{j}}}^{++}\right]}=-{{{\textstyle\frac{1}{2}}}}{{\mathfrak{L}}}.$$ Using these identities we find, $$\begin{aligned}
\label{eq:J-J+} {\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{J}}}^{++}_2\right]}&=&
{\left\{\overleftarrow{{{\mathfrak{T}}}}^-_1,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{j}}}^{++}\right]}\right]}\right\}}-{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{j}}}^{++}\right]}\right]}\right\}} \nonumber \\
&=& - {{{\textstyle\frac{1}{2}}}}{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{L}}}\right]}\right\}}+ {{{\textstyle\frac{1}{2}}}}{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{L}}}\right]}\right\}} \nonumber \\
&=& {{{\textstyle\frac{1}{2}}}}{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,\overrightarrow{{{\mathfrak{T}}}_1}^+\right\}}+{{{\textstyle\frac{1}{2}}}}{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& {{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}_2 \end{aligned}$$ To reach the second to last line, we used (\[eq:length\]), and for the last line (\[eq:psU11\^2Alg\]) was needed. By hermiticity, $${\left[{{\mathfrak{J}}}^{--}_2,{{\mathfrak{J}}}^{++}_0\right]}= {\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{J}}}^{++}_2\right]}^\dagger.$$ So, finally we have, $$\begin{aligned}
{\left[{{\mathfrak{J}}}^{--},{{\mathfrak{J}}}^{++}\right]}_2&=& {\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{J}}}^{++}_2\right]}+{\left[{{\mathfrak{J}}}^{--}_2,{{\mathfrak{J}}}^{++}_0\right]} \nonumber \\
&=& {\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{J}}}^{++}_2\right]}+{\left[{{\mathfrak{J}}}^{--}_0,{{\mathfrak{J}}}^{++}_2\right]}^\dagger \nonumber \\
&=& {{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}_2 + {{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}_2^\dagger \nonumber \\
&=& {\delta{\mathfrak{D}}}, \end{aligned}$$ as required by the algebra. We used the hermiticity of ${\delta{\mathfrak{D}}}_2$ to reach the last line.
Order $g^3$ {#sec:g3}
===========
With the $\mathcal{O}(g^2)$ solution, we can use the constraints to find the $\mathcal{O}(g^3)$ solution, which consists of corrections to the ${{\mathfrak{psu}}}(1|1)^2$ generators. We now present and discuss this $\mathcal{O}(g^3)$ solution and its proof, the two-loop dilatation generator that follows, and the lift to the finite-$N$ dilatation generator.
The solution \[sec:g3solution\]
-------------------------------
Only the ${{\mathfrak{psu}}}(1|1)^2$ generators, the ${{\mathfrak{T}}}$, receive corrections at this order. Once again the form of the solution depends only on the sign of the generator (though recall that now the sign refers to the commutator with ${{\mathfrak{L}}}$). $${{\mathfrak{T}}}^\pm_3=\pm {\left[{{\mathfrak{T}}}^\pm_1,{{\mathfrak{x}}}\right]} + {\left[{{\mathfrak{T}}}^\pm_1,{{\mathfrak{y}}}\right]} + \alpha {{\mathfrak{T}}}^\pm_1. \label{eq:g3solution}$$ Again, the ${{\mathfrak{y}}}$ commutator is a similarity transformation, and it must be the same as that of the $\mathcal{O}(g^2)$ solution. $\alpha$ corresponds to the coupling constant transformation $$g \mapsto g + \alpha g^3.$$ As at $\mathcal{O}(g^2)$, the solution is hermitian, provided ${{\mathfrak{y}}}$ is anti-hermitian. It is difficult to imagine a simpler solution. Beside the coupling constant transformation and the similarity transformation, the solution at this order is just a commutator with ${{\mathfrak{x}}}$, as was the case for $\mathcal{O}(g^2)$.
As at the previous order, we use a direct method to prove that this solution satisfies the symmetry algebra constraints, and the proof is in Appendix \[sec:g3proof\].
Freedom for the $\mathcal{O}(g^3)$ solution
-------------------------------------------
At this order, we could add gauge transformations to the generators. Furthermore, the solution satisfies the commutation relations only up to gauge transformations.
The case for homogeneous solutions at this order exactly parallels that of the previous order. Under a homogeneous modification, $$\overleftrightarrow{{{\mathfrak{T}}}_3}^{\pm} \mapsto \overleftrightarrow{{{\mathfrak{T}}}_3}^{\pm} + \delta\overleftrightarrow{{{\mathfrak{T}}}_3}^{\pm}.$$ In order for the symmetry constraints to remain satisfied, the $\delta {{\mathfrak{T}}}$’s must not contribute to any commutator of the algebra, both for commutators among the ${{\mathfrak{psu}}}(1|1)^2$ generators and for those with ${{\mathfrak{psu}}}(1,1|2)$ generators. Again, we have not found any non-trivial homogeneous solutions, or ruled them out. However, as for the $\mathcal{O}(g^2)$ solution, the successful checks of our solution with field theory computations implies that such a homogeneous contribution is not part of the field theory solution.
The two-loop dilatation operator
--------------------------------
From (\[eq:psU11\^2Alg\]) we can now compute ${\delta{\mathfrak{D}}}_4$ directly, $${\delta{\mathfrak{D}}}_4=2 {\left\{\overrightarrow{{{\mathfrak{T}}}}^+,\overleftarrow{{{\mathfrak{T}}}}^-\right\}}_4=2 {\left\{\overleftarrow{{{\mathfrak{T}}}}^+,\overrightarrow{{{\mathfrak{T}}}}^-\right\}}_4.$$ It follows that ${\delta{\mathfrak{D}}}_4$ is composed only of the ${{\mathfrak{T}}}_1$’s and ${{\mathfrak{h}}}$, the one-site harmonic number generator. After setting the similarity transformation ${{\mathfrak{y}}}$ to zero, without loss of generality, and using the vanishing of the squares of the ${{\mathfrak{T}}}$’s, we find $$\begin{aligned}
{\delta{\mathfrak{D}}}_4&=& 2 {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{h}}}\right]}\right\}}\right]}\right\}}+ 2{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]}\right\}} \nonumber \\ &=& 2 {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{h}}}\right]}\right\}}\right]}\right\}}+ 2{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]}\right\}}. \nonumber \\ & & \label{eq:D4solution} \end{aligned}$$ In this expression we have left out the coupling constant transformation parameterized by $\alpha$ in (\[eq:g3solution\]), which leads to $${\delta{\mathfrak{D}}}_4 \mapsto {\delta{\mathfrak{D}}}_4 + 2 \alpha \, {\delta{\mathfrak{D}}}_2.$$ However, to match field theory results $\alpha$ must be zero.
Non-planarity and wrapping interactions \[sec:non-planar\]
----------------------------------------------------------
By lifting our expressions for the building blocks of ${\delta{\mathfrak{D}}}_4$ to their non-planar generalization, we will construct a candidate for the finite-$N$ ${\delta{\mathfrak{D}}}_4$. To support our conjecture that this is the correct solution, we will observe that it accurately includes wrapping interactions. The two-loop non-planar solution for the ${{\mathfrak{su}}}(2)$ sector (a subsector of the ${{\mathfrak{su}}}(1,1|2)$ sector) was found in [@Beisert:2003tq]. In that case, there is a unique lift from the planar to the non-planar theory.
The non-planar action for the one-site generators, including the $\mathcal{O}(g^0)$ terms and ${{\mathfrak{h}}}$ are straightforward to obtain. Let the gauge group of the theory have generators $\mathfrak{t_m}$ and metric $\mathfrak{g^{mn}}$. Then, for instance, using the notation of [@Beisert:2004ry] $${{\mathfrak{h}}}= \sum_{k=0}^{\infty} h(k)\, \mbox{Tr}\, (\phi^\downarrow_k \, \check{\phi}^\downarrow_k+\phi^\uparrow_k \, \check{\phi}^\uparrow_k)+h(k+1) \, \mbox{Tr}\, (\overleftarrow{\psi_k} \, \check{\overleftarrow{\psi_k}} + \overrightarrow{\psi_k} \, \check{\overrightarrow{\psi_k}}),$$ where for $X_i \in \{\phi_k^\updownarrow,\,\overleftrightarrow{\psi_k}\}$, we have the expansion $X_i=X^\mathfrak{m}_i\mathfrak{t_m}$, and $$\check{X_i} = \mathfrak{t}_\mathfrak{m}\mathfrak{g^{mn}}\frac{\delta}{\delta X^\mathfrak{n}_i}, \quad \frac{\delta}{\delta X^\mathfrak{m}_i}X^\mathfrak{n}_j=\delta_{ij}\delta^\mathfrak{n}_\mathfrak{m}.$$
The ${{\mathfrak{T}}}_1$ also have a natural generalization for the non-planar theory. $$\begin{aligned}
\overrightarrow{{{\mathfrak{T}}}_1}^+=&\sum_{\substack{0 \leq m \\ 0 \leq k < m}}& \frac{1}{\sqrt{2(k+1))}}\,\mbox{Tr}\,{\left[\overrightarrow{\psi_k},\phi_{m-1-k}^\updownarrow\right]} \check{\phi}_m^\updownarrow \nonumber \\ + & \sum_{\substack{0 \leq m \\ 0 \leq k < m}} & \frac{m+1}{2 \sqrt{2(k+1)(m-k)}}\,\mbox{Tr}\,{\left\{\overrightarrow{\psi_k},\overrightarrow{\psi_m}_{-1-k}\right\}} \check{\overrightarrow{\psi_m}} \nonumber \\ + & \sum_{\substack{0 \leq m \\ 0 \leq k < m}} & \frac{m-k}{\sqrt{2(k+1)(m+1)}}\,\mbox{Tr}\,{\left\{\overrightarrow{\psi_k},\overleftarrow{\psi_m}_{-1-k}\right\}} \check{\overleftarrow{\psi_m}} \nonumber \\ + & \sum_{\substack{0 \leq m \\ 0 \leq k \leq m}} & \frac{1}{\sqrt{2(m+1)}}\,\mbox{Tr}\,{\left[\phi_k^\downarrow,\phi_{m-k}^\uparrow\right]} \check{\overleftarrow{\psi_m}}. \end{aligned}$$
There is a similar expression for $\overleftarrow{{{\mathfrak{T}}}_1}^+$, which can be read from (\[eq:leftT1+\]). For the hermitian conjugates, the ${{\mathfrak{T}}}_1^-$, simply perform the switch $$X_i \leftrightarrow \check{X}_i, \quad \forall X_i.$$ Substituting these expressions into the expressions for ${{\mathfrak{x}}}$, gives its non-planar version. Then the expressions given for ${{\mathfrak{T}}}_3$ and ${\delta{\mathfrak{D}}}_4$ become non-planar. Because the proof that the planar solution satisfies the symmetry constraints is independent of planarity, the non-planar generalization still satisfies the symmetry constraints.
While we do not have a proof that this is the correct non-planar solution, our solution accurately includes wrapping interactions, which can be thought of as special cases of non-planar interactions. Wrapping interactions apply to two-site states, for which the planar solution and the non-planar generalization are equivalent. Since the ${{\mathfrak{T}}}_1$’s map one site to two sites or vice-versa, the action of ${\delta{\mathfrak{D}}}_4$ is well defined even on two-site states. Since acting with the ${{\mathfrak{T}}}^+$’s on two-site states yields three-site states, two-site states are in the same multiplets as three-site states. Therefore, adding special wrapping interactions that only change the anomalous dimensions of two-site states would be inconsistent with the symmetry constraints.
Tests and applications of the solution \[sec:checks\]
=====================================================
Using the solution for the two-loop dilatation operator, we first provide strong evidence that it is correct via direct diagonalization and comparison to rigorous field theory computations. We then use our solution to present strong evidence in favor of integrability by computing the internal S-matrix in the bosonic ${{\mathfrak{sl}}}(2)$ sector and by comparing anomalous dimension predictions of the Bethe ansatz of [@Staudacher:2004tk] with the results of direct diagonalization.
Two-loop planar anomalous dimensions
------------------------------------
Expanding the expression for ${\delta{\mathfrak{D}}}_4$ in terms of interactions, we find the planar anomalous dimensions by direct diagonalization. We first identify the spin chain states of the subspaces of certain (small) values of classical dimension, R-charge, length, and hypercharge. Then we apply $g^2 {\delta{\mathfrak{D}}}_2 + g^4{\delta{\mathfrak{D}}}_4$ to these subspaces and compute its eigenvalues (the anomalous dimensions) and eigenstates. Again, we have used `Mathematica`. We check states with rigorously known anomalous dimensions. These include twist-two operators [@Kotikov:2003fb], a pair of states of length three and bare dimension six [@Eden:2005bt], two excitation states (BMN operators) [@Beisert:2002tn; @Beisert:2003tq], and length-three states built from one type of fermion and from derivatives (in the fermionic ${{\mathfrak{sl}}}(2)$ subsector) [@Belitsky:2005bu]. The states we check, given in Table \[tab:comparefield\], are in complete agreement with these previous computations. Therefore, we conclude that we have found the correct solution for ${\delta{\mathfrak{D}}}_4$. Since our comparison includes length-two states, we find confirmation that no additional wrapping terms are needed.
$\begin{array}{|l|l|l|}\hline
{{\mathfrak{D}}}_0&(R, L, B) &{\bigl({\delta{\mathfrak{D}}}_2,{\delta{\mathfrak{D}}}_4\bigr)}^P\\\hline
4 &(2,2,0)& \begin{array}{l} (6,-12)^+ \end{array} \\
\hline
5 &(3,3,0)& \begin{array}{l} (4,-6)^- \end{array} \\
\hline
6 &(2,2,0)& \begin{array}{l} (\frac{25}{3},-\frac{925}{54})^+ \end{array} \\
\hline
6 & (3, 3, 0)& \begin{array}{l} (\frac{15}{2}, -\frac{225}{16})^\pm \end{array} \\ \hline
6 &(4,4,0)&\begin{array}{l} (2.76393,-2.90983)^+ \\ (7.23607,-14.0902)^+ \end{array} \\ \hline
7 & (5, 5, 0)& \begin{array}{l} (2, -\frac{3}{2})^- \\ (6, -\frac{21}{2})^- \end{array} \\ \hline
7.5 & (0, 3, \pm \frac{3}{2})& \begin{array}{l} (10, -\frac{245}{12})^\pm \end{array} \\ \hline
8 &(2,2,0)& \begin{array}{l} (\frac{49}{5},-\frac{45619}{2250})^+ \end{array} \\
\hline
8 & (6, 6, 0)& \begin{array}{l} (1.50604, -0.830063)^+ \\ (4.89008, -7.30622)^+ \\ (7.60388, -14.8637)^+ \end{array} \\ \hline
9 &(6,8,\pm 1)& \begin{array}{l} (1.17157,-0.4895952)^- \\ (4, -5)^- \\ (6.82843, -12.5104)^- \end{array} \\
\hline
9.5 & (0, 3, \pm \frac{3}{2})& \begin{array}{l} (\frac{133}{12},-\frac{131117}{5760})^\pm \end{array} \\ \hline
10 &(7,9,\pm 1)& \begin{array}{l} (0.935822,-0.304865)^+ \\ (3.30540, - 3.44381)^+ \\ (6, -10)^+ \\ (7.75877, -15.2513)^+ \end{array} \\ \hline
10.5 & (0, 3, \pm \frac{3}{2})& \begin{array}{l} (\frac{761}{70}, -\frac{138989861}{6174000})^+ \\ (\frac{ 761}{60},-\frac{419501}{16000})^\pm \end{array} \\ \hline \end{array}$
The two-loop ${{\mathfrak{sl}}}(2)$ S-matrix and diffractionless scattering
---------------------------------------------------------------------------
We now perform a new two-loop check of the bosonic ${{\mathfrak{sl}}}(2)$ sector Bethe ansatz of [@Staudacher:2004tk]. Instead of only checking anomalous dimension predictions, we also verify a key part of its derivation, the S-matrix. It is straightforward to restrict to the two-excitation ${{\mathfrak{sl}}}(2)$ sector, consisting of states composed only of $\phi^\downarrow$’s and two or fewer derivatives[^9]. We have computed the internal S-matrix as in [@Staudacher:2004tk], which used ideas introduced in [@Bethe:1931hc] and [@Sutherland:1978xx]. A basis for two excitation states is $${|\Psi_{x_1x_2}\rangle}= {|\ldots\phi^\downarrow\stackrel{\stackrel{x_1}{\downarrow}}{{\mathcal{D}}(\phi^\downarrow)}\phi^\downarrow\ldots\phi^\downarrow\stackrel{\stackrel{x_2}{\downarrow}}{{\mathcal{D}}(\phi^\downarrow)}\phi^\downarrow\ldots\rangle},$$ where the derivatives appear at sites $x_1$ and $x_2$ of the spin chain. Then, the Schrödinger equation $$H {|\Psi\rangle} = E {|\Psi\rangle}, \qquad H = g^2 \, {\delta{\mathfrak{D}}}_2+g^4 \, {\delta{\mathfrak{D}}}_4,$$ is solved by the ansatz $$\begin{gathered}
{|\Psi\rangle} = \sum_{1 \leq x_1 \leq x_2 \leq L} \left(f_-(\delta x, p_i)e^{ip_1x_1+ip_2x_2}+f_+(\delta x, p_i)e^{ip_2x_1+ip_1x_2} \right) {|\Psi_{x_1x_2}\rangle}, \nonumber \\ \delta x = x_2-x_1. \end{gathered}$$ $L$ is the length of the spin chain, and the $p_i$ are the momenta of the excitations which scatter off each other. Since the Hamiltonian is short-ranged and translationally invariant, for large $\delta x$ the solutions of the Schrödinger equation reduce to superpositions of one excitation eigenstates, proportional to $e^{i p x}$. The S-matrix gives the phase that one excitation’s wave function acquires when passing the other excitation, $$S(p_2,p_1) = \frac{f_+(\delta x, p_i)}{f_-(\delta x, p_i)}, \quad \delta x > 1.$$ The inequality reflects that the Hamiltonian has interactions involving at most three adjacent sites. This short-range Hamiltonian also leads to the following ansatz[^10], $$f_-(\delta x > 1) = 1,\, f_+(\delta x > 0) = S,\, f_-(1)=f_-, \, f_\pm(0)=f_0.$$ Using $\texttt{Mathematica}$, we have solved the Schrödinger equation using this ansatz and our expression for the dilatation generator. The solution for the energy and the S-matrix is $$\begin{gathered}
E= E(p_1)+E(p_2),\qquad E(p) = 4 \sin^2(\frac{p}{2}) - 8 g^2 \sin^4(\frac{p}{2}) \\ S(p_2,p_1) = S_0 + g^2 S_2 \nonumber \\
S_0= -\frac{e^{ip_1+ip_2}-2e^{i p_2}+1}{e^{ip_1+ip_2}-2e^{i p_1}+1} \\
S_2= \frac{8 i e^{i p_1+i p_2} \sin(\frac{p_1}{2})( \sin (\frac{p_1- 3 p_2}{2})- 4 \sin (\frac{p_1-p_2}{2})+\sin (\frac{3 p_1- p_2}{2}))\sin(\frac{p_2}{2})}{(1-2 e ^{i p_1}+ e ^ {ip_1 +i p_2})^2}. \end{gathered}$$ To two-loop order, this agrees with the solution given by equations (3.3) and (6.4), (4.27), and (3.7) of [@Staudacher:2004tk].
At this point, assuming diffractionless scattering and requiring periodicity yields the Bethe equation for this sector, which can be used to compute anomalous dimensions for states with arbitrary numbers of excitations, as in [@Staudacher:2004tk]. As shown in Table \[tab:sl2\], we find perfect agreement between the predictions of the Bethe ansatz and direct diagonalization of the two-loop dilatation generator. This provides compelling evidence for two-loop integrability in the bosonic ${{\mathfrak{sl}}}(2)$ subsector.
$\begin{array}{|l|l|l|}\hline
D_0&(R, L, B) &{\bigl({\delta{\mathfrak{D}}}_2,{\delta{\mathfrak{D}}}_4\bigr)}^P\\\hline
7 & (3, 3, 0)& \begin{array}{l} (6, -\frac{39}{4})^- \end{array} \\ \hline
7 & (4, 4, 0)& \begin{array}{l} (6, -\frac{21}{2})^\pm \end{array} \\ \hline
8 & (3, 3, 0)& \begin{array}{l} (\frac{35}{4}, -\frac{18865}{1152})^\pm \end{array} \\ \hline
8 & (4, 4, 0)& \begin{array}{l} (4.38277, -5.25026)^+ \\ (8.35923, -16.0680)^+ \\ (11.5913, -23.1031)^+ \\ (\frac{23}{3}, -\frac{1331}{108})^\pm \end{array} \\ \hline
8 & (5, 5, 0)& \begin{array}{l} (4.72931,-7.01464)^\pm \\ (7.77069, -14.4229)^\pm \end{array} \\ \hline \end{array}$
The ${{\mathfrak{su}}}(1|1)$ sector
-----------------------------------
Finally, we provide evidence of integrability including fermions as well. We compute anomalous dimensions for the ${{\mathfrak{su}}}(1|1)$ sector(s), again via direct diagonalization. This sector includes states made of only one type of $\phi$ and only one type of $\psi$, and no derivatives. Again, our findings are in complete agreement with those found assuming integrabilty in [@Staudacher:2004tk]. These anomalous dimensions were also found by direct diagonalization of the compact ${{\mathfrak{su}}}(2|3)$ dilatation operator in [@Beisert:2003ys].
$\begin{array}{|l|l|l|}\hline
D_0&(R, L, B)) &{\bigl({\delta{\mathfrak{D}}}_2,{\delta{\mathfrak{D}}}_4\bigr)}^P\\\hline
7 &(1,5,\pm 2)& \begin{array}{l} (10,-20)^- \end{array} \\
\hline
7.5 &(3,6,\pm 1.5)& \begin{array}{l} (8,-14)^\pm \end{array} \\
\hline
8 &(2,6,\pm 2)& \begin{array}{l} (8,-14)^+ \end{array} \\
\hline
8.5 &(4,7,\pm 1.5)& \begin{array}{l} (7,-12)^\pm \end{array} \\
\hline
9 &(3,7,\pm 2)& \begin{array}{l} (6.39612,-9.3993)^- \\ (9.10992, -17.1028)^- \\ (12.494, -24.4979)^- \end{array} \\
\hline
9.5 &(5,8,\pm 1.5)& \begin{array}{l} (6,-\frac{19}{2})^\pm \\ (8, -\frac{29}{2})^\pm \end{array} \\
\hline \end{array}$
Conclusion and outlook \[sec:outlook\]
======================================
We have found a remarkably simple solution for the two-loop dilatation generator and the one and one-half loop symmetry algebra of a non-compact sector of $\mathcal{N}=4$ SYM. The $\mathcal{O}(g^2)$ and $\mathcal{O}(g^3)$ symmetry algebra corrections are given, with appropriate choice of basis and gauge, by $${{\mathfrak{X}}}^\pm_{i+2}=\pm {\left[{{\mathfrak{X}}}^\pm_i,{{\mathfrak{x}}}\right]}, \quad i=0,\,1.$$ ${{\mathfrak{x}}}$, given by (\[eq:definex\]), only involves the leading $\mathcal{O}(g^1)$ terms for the ${{\mathfrak{psu}}}(1|1)^2$ generators and the harmonic numbers that characterize the one-loop dilatation operator. Furthermore, the two-loop dilatation generator (\[eq:D4solution\]) generates the two-loop ${{\mathfrak{sl}}}(2)$ S-matrix of the Bethe ansatz proposal [@Staudacher:2004tk], and its anomalous dimensions match both this Bethe ansatz proposal and the direct field theory calculations of [@Beisert:2003tq; @Kotikov:2003fb; @Eden:2005bt; @Belitsky:2005bu]. This is additional very strong evidence in favor of the two-loop integrability and Bethe ansatz for the ${{\mathfrak{su}}}(1,1|2)$ sector [@Beisert:2005fw].
Despite the convincing evidence that our solution would be produced by a complete direct field theory computation, it is unknown if the solution is uniquely determined by symmetry and Feynman diagram constraints, as is the one-loop dilatation operator [@Beisert:2004ry]. It would be very interesting to identify the order at which the solution is not completely constrained (if any), and the minimal set of additional constraints required to isolate the field theory solution.
The structure of the solution suggests additional directions of research. It is natural to conjecture that the iterative solution we found can be extended to larger sectors, which necessarily have dilatation generator interactions that do not conserve length and hypercharge, or to higher loops. Such a solution for the four-loop dilatation generator may be especially useful. At that order, non-trivial wrapping interactions for two-site states could be consistent with the symmetry algebra. Because the wrapping interactions are less constrained by inspecting Feynman diagrams, it would seem impossible to compute them just using the constraints. However, it is possible that a higher-loop extension of our iterative solution would not need specific wrapping interactions added to match the field theory solution.
We suspect that there are expressions involving the same building blocks for the higher conserved charges due to integrability. Evidence for this was given in [@Agarwal:2005jj]. Up to two loops, Agarwal and Ferretti showed that the first higher charge for the ${{\mathfrak{su}}}(2|3)$ sector could be written diagrammatically in terms of the dilatation generator. They conjectured that the diagrammatic expression generalizes to the entire theory. Using the solution for the two-loop dilatation generator, it is now possible to check whether their solution [@Agarwal:2005jj] generalizes to a non-compact sector.
Finally, this is not the first time iterative structures have appeared in $\mathcal{N}=4$ SYM. Planar scattering amplitudes have iterative structure at two and three loops [@Anastasiou:2003kj; @Bern:2005iz]. Also, following Witten’s work relating gauge theory to a string theory in twistor space [@Witten:2003nn], recurrence relations between amplitudes involving different numbers of particles have been found [@Cachazo:2004kj; @Britto:2004ap]. The two-loop dilatation generator has some qualitative resemblance to this recursive structure. The ${{\mathfrak{T}}}_1$ are analogous to the three gluon on-shell amplitudes, and ${{\mathfrak{h}}}$ is similar to a Feynman propogator. It would be wonderful if iterative structures could be used to relate the dilatation generator and scattering amplitudes.
Acknowledgements {#acknowledgements .unnumbered}
================
I thank my advisor, Niklas Beisert, for pointing me to the ${{\mathfrak{su}}}(1,1|2)$ sector, for sharing essential $\texttt{Mathematica}$ code, for suggesting many improvements to this article and for numerous enlightening and stimulating conversations.
I also thank Joshua Friess for helpful suggestions. This material is based upon work supported under a National Science Foundation Graduate Research Fellowship. Any opinions, findings, conclusions or recommendations expressed in this publication are those of the author and do not necessarily reflect the views of the National Science Foundation.
The ${{\mathfrak{psu}}}(1,1|2)$ commutators \[sec:psu112comm\]
==============================================================
Matrix representation
---------------------
The matrix representation for the full ${{\mathfrak{psu}}}(2,2|4)$ algebra shown in Appendix D of [@Beisert:2004ry] can be restricted to the ${{\mathfrak{psu}}}(1,1|2)$ sector. As for the full algebra, we parameterize an element $j\cdot {{\mathfrak{J}}}$ of the algebra by the adjoint vector $j$. In this case, we split the matrix into $1|2|1$ (even$|$odd$|$even) rows and columns. We write the representation of ${{\mathfrak{u}}}(1,1|2)$ as
$$\label{eq:U112matrix2}
j\cdot {{\mathfrak{J}}}=\left( \begin{array}{c|cc|c}
j^0 + b - \frac{c}{2} &\overrightarrow{q}^{+\downarrow}&\overrightarrow{q}^{+\uparrow}&j^{++}\\\hline
\overleftarrow{q}^{-\uparrow}& r^0+\frac{b}{2} - \frac{c}{2}& r^{\uparrow\uparrow} &\overleftarrow{q}^{+\uparrow}\\ \overleftarrow{q}^{-\downarrow}&r^{\downarrow\downarrow}& -r^0+\frac{b}{2} - \frac{c}{2}&\overleftarrow{q}^{+\downarrow}\\\hline
-j^{--}&-\overrightarrow{q}^{-\downarrow}&-\overrightarrow{q}^{-\uparrow}&-j^0+b - \frac{c}{2} \end{array} \right)$$ The commutation relations of the generators follow from the matrix representation of ${\left[j\cdot {{\mathfrak{J}}},j'\cdot {{\mathfrak{J}}}\right]}$. The ${{\mathfrak{psu}}}(1,1|2)$ algebra follows from dropping ${{\mathfrak{B}}}$, which is not generated by any commutators of the other generators, and setting the central charge ${{\mathfrak{C}}}$ to zero. All physical fields are neutral with respect to ${{\mathfrak{C}}}$.
We now present a minimal set of commutators; the remaining commutators follow from hermitian conjugation and from combining commutators presented here. We group these commutators for later convenience.
Classical commutators
---------------------
These commutators only involve generators that receive no quantum corrections. $${\left[{{\mathfrak{R}}}^{\uparrow\uparrow},{{\mathfrak{R}}}^{\downarrow\downarrow}\right]}={{\mathfrak{R}}}^0, \qquad {\left[{{\mathfrak{R}}}^0,{{\mathfrak{R}}}^{\uparrow\uparrow}\right]}=2 {{\mathfrak{R}}}^{\uparrow\uparrow}.$$
Central charge commutators
--------------------------
These commutators are simpler to deal with at higher orders in $g$, because ${{\mathfrak{R}}}^0$ receives no quantum corrections, and because $${{\mathfrak{J}}}^0_n={\delta{\mathfrak{D}}}_n \qquad n>0. \label{eq:J0nDn}$$ $$\begin{aligned}
{\left[{{\mathfrak{J}}}^0(g),{{\mathfrak{J}}}^{++}(g)\right]}&=2{{\mathfrak{J}}}^{++}(g),&
{\left[{{\mathfrak{R}}}^0,{{\mathfrak{J}}}^{++}(g)\right]}&=0, \notag \\
{\left[{{\mathfrak{J}}}^0(g),\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right]}&= \overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g), &
{\left[{{\mathfrak{R}}}^0,\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right]}&=-\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g), \notag \\
{\left[{{\mathfrak{J}}}^0(g),\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)\right]}&= \overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g), &
{\left[{{\mathfrak{R}}}^0,\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)\right]}&=\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g) \notag \\
{\left[{{\mathfrak{J}}}^0(g),{{\mathfrak{R}}}^{\uparrow\uparrow}\right]}&=0.\end{aligned}$$
Commutators with ${{\mathfrak{R}}}$
-----------------------------------
$$\begin{aligned}
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right]}&=\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g) &
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},\overleftarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)\right]}&=\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g), \notag \\
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},\overrightarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)\right]}&=0, &
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},\overrightarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)\right]}&=-\overrightarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g), \notag \\
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},{{\mathfrak{J}}}^{++}(g)\right]}&=0, &
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},{{\mathfrak{J}}}^{--}(g)\right]}&=0. \label{eq:R+comm} \end{aligned}$$
Plus-plus commutators
---------------------
$$\begin{aligned}
{\left[{{\mathfrak{J}}}^{++}(g),\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right]}&=0,&
{\left[{{\mathfrak{J}}}^{++}(g),\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)\right]}&=0, \notag \\
{\left[{{\mathfrak{J}}}^{++}(g),\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)\right]}&=0, &
{\left[{{\mathfrak{J}}}^{++}(g),\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right]}&=0, \notag \\
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g),\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right\}}&=0, & {\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g),\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)\right\}}&=0, \notag \\
{\left\{\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g),\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g)\right\}}&=0, & {\left\{\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g),\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g)\right\}}&=0. \end{aligned}$$
${\delta{\mathfrak{D}}}$ commutators
------------------------------------
These commutators, at non-zero order in $g$, yield a multiple of ${\delta{\mathfrak{D}}}$. $$\begin{aligned}
{\left[{{\mathfrak{J}}}^{--}(g),{{\mathfrak{J}}}^{++}(g)\right]}&={{\mathfrak{J}}}^0(g),&
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g),\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)\right\}}&=\frac{1}{2}{{\mathfrak{J}}}^0(g)+\frac{1}{2}{{\mathfrak{R}}}^0, \notag \\
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g),\overleftarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)\right\}}&=\frac{1}{2}{{\mathfrak{J}}}^0(g)-\frac{1}{2}{{\mathfrak{R}}}^0, &
{\left\{\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g),\overrightarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)\right\}}&=\frac{1}{2}{{\mathfrak{J}}}^0(g)+\frac{1}{2}{{\mathfrak{R}}}^0, \notag \\
{\left\{\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g),\overrightarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)\right\}}&=\frac{1}{2}{{\mathfrak{J}}}^0(g)-\frac{1}{2}{{\mathfrak{R}}}^0. \end{aligned}$$
Plus-minus commutators
----------------------
$$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g),\overleftarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)\right\}}&={{\mathfrak{R}}}^{\downarrow\downarrow}, &
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g),\overrightarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)\right\}}&=0, \notag \\
{\left[{{\mathfrak{J}}}^{++}(g),\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow}(g)\right]}&=-\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}(g), &
{\left[{{\mathfrak{J}}}^{++}(g),\overrightarrow{{{\mathfrak{Q}}}}^{-\downarrow}(g)\right]}&=-\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}(g). \label{eq:+-comm} \end{aligned}$$
Proof of the $\mathcal{O}(g^2)$ solution \[sec:g2proof\]
========================================================
We will verify that (\[eq:g2solution\]) is a solution by checking representatives of the minimal set of commutators given in the last section. We first present some necessary identities. Throughout this proof, we set ${{\mathfrak{y}}}$, the similarity transformation, to zero, without loss of generality.
Properties of the $\mathcal{O}(g^2)$ solution \[sec:g2properties\]
------------------------------------------------------------------
Since ${{\mathfrak{h}}}$ commutes with ${{\mathfrak{J}}}^0_0$ and the ${{\mathfrak{R}}}$’s, we have $$\begin{aligned}
{\left[{{\mathfrak{x}}},{{\mathfrak{J}}}^0_0\right]}&=0, & {\left[{{\mathfrak{x}}},{{\mathfrak{R}}}^{\uparrow\uparrow}\right]}&=0, \notag \\
{\left[{{\mathfrak{x}}},{{\mathfrak{R}}}^{\downarrow\downarrow}\right]}&=0, & {\left[{{\mathfrak{x}}},{{\mathfrak{R}}}^0\right]}&=0. \label{eq:J00x} \end{aligned}$$ Since ${{\mathfrak{h}}}$ commutes with ${{\mathfrak{L}}}$, by (\[eq:psU11\^2Alg\]) $${\left[{{\mathfrak{x}}},{{\mathfrak{L}}}\right]}=0. \label{eq:xl}$$ Next, we define $$\begin{aligned}
{\left[{{\mathfrak{J}}}^{++},{{\mathfrak{h}}}\right]}&={{\mathfrak{j}}}^{++}, & {\left[\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow},{{\mathfrak{h}}}\right]}&=\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow}, \notag \\
{\left[\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow},{{\mathfrak{h}}}\right]}&=\overrightarrow{{{\mathfrak{q}}}}^{+\uparrow}, & {\left[\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow},{{\mathfrak{h}}}\right]}&=\overleftarrow{{{\mathfrak{q}}}}^{+\uparrow}, \notag \\
{\left[\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow},{{\mathfrak{h}}}\right]}&=\overleftarrow{{{\mathfrak{q}}}}^{+\downarrow} \label{eq:lowercase} \end{aligned}$$ As usual, the conjugates of these new generators are labeled with a minus instead of plus, and arrows pointing in the opposite direction. A useful simplification depends on the following vanishing commutators. $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}}&=0, & {\left\{\overrightarrow{{{\mathfrak{q}}}}^{+\uparrow},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}}&=0, \notag \\
{\left\{\overleftarrow{{{\mathfrak{q}}}}^{+\uparrow},\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}}&=0, & {\left\{\overleftarrow{{{\mathfrak{q}}}}^{+\downarrow},\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}}&=0. \label{eq:vanishing} \end{aligned}$$ Then, from (\[eq:g2solution\]) and (\[eq:definex\]) we find $$\begin{aligned}
\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}&={\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow}\right\}}\right]}, & \overrightarrow{{{\mathfrak{Q}}}_2}^{+\uparrow}&={\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,\overrightarrow{{{\mathfrak{q}}}}^{+\uparrow}\right\}}\right]}, \notag \\
\overleftarrow{{{\mathfrak{Q}}}_2}^{+\uparrow}&={\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,\overleftarrow{{{\mathfrak{q}}}}^{+\uparrow}\right\}}\right]}, & \overleftarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}&={\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,\overleftarrow{{{\mathfrak{q}}}}^{+\downarrow}\right\}}\right]}. \label{eq:Q2} \end{aligned}$$
The following equalities will be essential. $$\begin{aligned}
{\left[{{\mathfrak{J}}}^{++}_0,{{\mathfrak{j}}}^{--}\right]}&={{{\textstyle\frac{1}{2}}}}{{\mathfrak{L}}}, & {\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overleftarrow{{{\mathfrak{q}}}}^{-\uparrow}\right\}}&=\frac{1}{4}(2 {{\mathfrak{B}}}- {{\mathfrak{L}}}-{{\mathfrak{R}}}^0), \notag \\
{\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow},\overleftarrow{{{\mathfrak{q}}}}^{-\downarrow}\right\}}&=\frac{1}{4}(2 {{\mathfrak{B}}}- {{\mathfrak{L}}}+ {{\mathfrak{R}}}^0), &
{\left\{\overleftarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overrightarrow{{{\mathfrak{q}}}}^{-\uparrow}\right\}}&=-\frac{1}{4}(2 {{\mathfrak{B}}}+ {{\mathfrak{L}}}+{{\mathfrak{R}}}^0), \notag \\
{\left\{\overleftarrow{{{\mathfrak{Q}}}_0}^{+\uparrow},\overrightarrow{{{\mathfrak{q}}}}^{-\downarrow}\right\}}&=-\frac{1}{4}(2 {{\mathfrak{B}}}+ {{\mathfrak{L}}}- {{\mathfrak{R}}}^0). \label{eq:essential} \end{aligned}$$
Finally, we have $${\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overleftarrow{{{\mathfrak{q}}}}^{-\downarrow}\right\}}=-{{{\textstyle\frac{1}{2}}}}{{\mathfrak{R}}}^{\downarrow\downarrow}, \quad {\left[{{\mathfrak{J}}}^{++}_0,\overleftarrow{{{\mathfrak{q}}}}^{-\uparrow}\right]}=0 \quad \mbox{and} \quad {\left[{{\mathfrak{J}}}^{++}_0,\overleftarrow{{{\mathfrak{q}}}}^{-\downarrow}\right]}=0. \label{eq:finally}$$
Central charge commutators
--------------------------
We check the first commutator in detail. $$\begin{aligned}
{\left[{{\mathfrak{J}}}^0,{{\mathfrak{J}}}^{++}\right]}_2 &=& {\left[{{\mathfrak{J}}}^0_0,{{\mathfrak{J}}}^{++}_2\right]} + {\left[{{\mathfrak{J}}}^0_2,{{\mathfrak{J}}}^{++}_0\right]}\nonumber \\
&=& {\left[{{\mathfrak{J}}}^0_0,{\left[{{\mathfrak{J}}}^{++}_0,{{\mathfrak{x}}}\right]}\right]}+ {\left[{\delta{\mathfrak{D}}}_2,{{\mathfrak{J}}}^{++}_0\right]} \nonumber \\
&=& {\left[{\left[{{\mathfrak{J}}}^0_0,{{\mathfrak{J}}}^{++}_0\right]},{{\mathfrak{x}}}\right]} + {\left[{{\mathfrak{J}}}^{++}_0,{\left[{{\mathfrak{J}}}^0_0,{{\mathfrak{x}}}\right]}\right]} \nonumber \\
&=& 2{\left[{{\mathfrak{J}}}^{++}_0,{{\mathfrak{x}}}\right]} \nonumber \\
&=& 2 {{\mathfrak{J}}}^{++}_2. \end{aligned}$$
In the second line, we used (\[eq:J0nDn\]), and in the fourth line we used (\[eq:J00x\]).
Commutators with ${{\mathfrak{R}}}$
-----------------------------------
Here is one example of these proofs. $$\begin{aligned}
{\left[{{\mathfrak{R}}}^{\uparrow\uparrow},\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}\right]}_2 &=& {\left[{{\mathfrak{R}}}^{\uparrow\uparrow}_0,\overleftarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}\right]} \nonumber \\
&=& {\left[{{\mathfrak{R}}}^{\uparrow\uparrow}_0,{\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},{{\mathfrak{x}}}\right]}\right]} \nonumber \\
&=& {\left[{\left[{{\mathfrak{R}}}^{\uparrow\uparrow}_0,\overleftarrow{{{\mathfrak{Q}}}_0}^{+\downarrow}\right]},{{\mathfrak{x}}}\right]} \nonumber \\
&=& {\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{+\uparrow},{{\mathfrak{x}}}\right]} \nonumber \\
&=& \overleftarrow{{{\mathfrak{Q}}}_2}^{+\uparrow}. \end{aligned}$$ We used (\[eq:J00x\]) again, as well as the leading order part of (\[eq:R+comm\]).
Plus-plus commutators {#sec:++}
---------------------
The commutators of two plus generators vanish at this order since they vanish at leading order. For example: $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow},\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}\right\}}_2 &=& {\left\{\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow},\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow}\right\}}+{\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overrightarrow{{{\mathfrak{Q}}}_2}^{+\uparrow}\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},{{\mathfrak{x}}}\right]},\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow}\right\}}+{\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},{\left[\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow},{{\mathfrak{x}}}\right]}\right\}} \nonumber \\
&=& {\left[{\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overrightarrow{{{\mathfrak{Q}}}_0}^{+\uparrow}\right\}},{{\mathfrak{x}}}\right]} \nonumber \\
&=& 0. \end{aligned}$$
${\delta{\mathfrak{D}}}$ commutators
------------------------------------
We proved the first commutator in Section \[sec:g2discussion\]. For the other four commutators one must change signs appropriately and use (\[eq:Q2\]) and (\[eq:essential\]).
Plus-minus commutators
----------------------
The commutator of $\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}$ and $\overleftarrow{{{\mathfrak{Q}}}}^{-\downarrow}$ vanishes at $\mathcal{O}(g^2)$ as required by (\[eq:+-comm\]). In fact, both $$\label{eq:Q+Q/} {\left\{\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}, \overleftarrow{{{\mathfrak{Q}}}_0}^{-\downarrow}\right\}}\quad \mbox{and} \quad {\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow}, \overleftarrow{{{\mathfrak{Q}}}_2}^{-\downarrow}\right\}}$$ vanish. We will prove this for the first commutator. The proof for the second follows the same steps. First we need $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow},\overleftarrow{{{\mathfrak{Q}}}_0}^{-\downarrow}\right\}} &=& {\left\{{\left[\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},{{\mathfrak{h}}}\right]},\overleftarrow{{{\mathfrak{Q}}}_0}^{-\downarrow}\right\}} \nonumber \\
&=& {\left[{{\mathfrak{R}}}^{\downarrow\downarrow},{{\mathfrak{h}}}\right]}+{\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overleftarrow{{{\mathfrak{q}}}}^{-\downarrow}\right\}} \nonumber \\
&=& {\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overleftarrow{{{\mathfrak{q}}}}^{-\downarrow}\right\}} \nonumber \\
&=& -{{{\textstyle\frac{1}{2}}}}{{\mathfrak{R}}}^{\downarrow\downarrow}. \end{aligned}$$ We used (\[eq:finally\]) for the last line. Using (\[eq:vanishing\]), we show the first commutator in (\[eq:Q+Q/\]) vanishes. $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow},\overleftarrow{{{\mathfrak{Q}}}_0}^{-\downarrow}\right\}} &=& {\left\{{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow}\right\}}\right]},\overleftarrow{{{\mathfrak{Q}}}_0}^{-\downarrow}\right\}} \nonumber \\
&=&
{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow},\overleftarrow{{{\mathfrak{Q}}}_0}^{-\downarrow}\right\}}\right]}\right\}} \nonumber \\
&=&
{{{\textstyle\frac{1}{2}}}}{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,-{{\mathfrak{R}}}^{\downarrow\downarrow}\right]}\right\}} \nonumber \\
&=& 0. \end{aligned}$$
Using (\[eq:vanishing\]), \[eq:Q2\]), and $${\left\{\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},\overrightarrow{{{\mathfrak{q}}}}^{-\uparrow}\right\}}={\left\{\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow},\overrightarrow{{{\mathfrak{Q}}}_0}^{-\uparrow}\right\}},$$ one can show that anti-commutator of $\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}$ and $\overrightarrow{{{\mathfrak{Q}}}}^{-\uparrow}$ vanishes.
As a preliminary step for verifying the next plus-minus commutator, we compute $$\begin{aligned}
{\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{{\mathfrak{j}}}^{++}\right]} &=& {\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{\left[{{\mathfrak{J}}}^{++}_0,{{\mathfrak{h}}}\right]}\right]} \nonumber \\
&=& {\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{+\uparrow},{{\mathfrak{h}}}\right]}-{\left[{{\mathfrak{J}}}^{++}_0,\overleftarrow{{{\mathfrak{q}}}}^{-\uparrow}\right]} \nonumber \\
&=& \overleftarrow{{{\mathfrak{q}}}}^{+\uparrow}-{\left[{{\mathfrak{J}}}^{++}_0,\overleftarrow{{{\mathfrak{q}}}}^{-\uparrow}\right]} \nonumber \\
&=& \overleftarrow{{{\mathfrak{q}}}}^{+\uparrow}. \end{aligned}$$
The last line works because of (\[eq:finally\]). (\[eq:finally\]) also implies that $\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow}_2$ and ${{\mathfrak{J}}}^{++}_0$ commute. Using this and (\[eq:vanishing\]), we find $$\begin{aligned}
{\left[\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow},{{\mathfrak{J}}}^{++}\right]}_2 &=& {\left[\overleftarrow{{{\mathfrak{Q}}}_2}^{-\uparrow},{{\mathfrak{J}}}^{++}_0\right]} + {\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{{\mathfrak{J}}}^{++}_2\right]} \nonumber \\
&=& {\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{j}}}^{++}\right]}\right\}}\right]}-{\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{j}}}^{++}\right]}\right\}}\right]} \nonumber \\
&=& {\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{{\mathfrak{j}}}^{++}\right]}\right\}}\right]}-{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{Q}}}_0}^{-\uparrow},{{\mathfrak{j}}}^{++}\right]}\right\}}\right]} \nonumber \\
&=& {\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,\overleftarrow{{{\mathfrak{q}}}}^{+\uparrow}\right\}}\right]} \nonumber \\
&=& \overleftarrow{{{\mathfrak{Q}}}_2}^{+\uparrow}. \end{aligned}$$
Equivalent steps show that $${\left[\overrightarrow{{{\mathfrak{Q}}}}^{-\downarrow},{{\mathfrak{J}}}^{++}\right]}_2 = \overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}.$$
Implied commutators
-------------------
Since the $\mathcal{O}(g^2)$ solution we presented (\[eq:g2solution\]) is hermitian (for ${{\mathfrak{y}}}$ set to zero), like the leading order representation of the generators, it immediately follows that all the conjugate equations to those of Appendix \[sec:psu112comm\] are satisfied at this order. The algebra then implies that the remaining commutators are satisfied. This completes our verification of the symmetry constraints.
Proof of the $\mathcal{O}(g^3)$ solution \[sec:g3proof\]
========================================================
We will first check that the solution (\[eq:g3solution\]) commutes with the ${{\mathfrak{psu}}}(1,1|2)$ generators, and then we will check that it satisfies the ${{\mathfrak{psu}}}(1|1)^2$ algebra. Repeatedly, we will use the vanishing of the $\mathcal{O}(g^1)$ commutators without further comment. Again, we set ${{\mathfrak{y}}}$ to zero without loss of generality. We also set $\alpha$ to zero without loss of generality, since a solution of the symmetry algebra remain a solution after a coupling constant transformation.
Commutators with the ${{\mathfrak{psu}}}(1,1|2)$ generators
-----------------------------------------------------------
### Central charge and ${{\mathfrak{R}}}$ commutators
To prove that the commutators of ${{\mathfrak{T}}}$ with the ${{\mathfrak{R}}}$’s or ${{\mathfrak{J}}}^0$ vanish we need only that the ${{\mathfrak{R}}}$’s receive no quantum corrections, (\[eq:J00x\]), (\[eq:J0nDn\]), and $${\left[{{\mathfrak{T}}}_1,{\delta{\mathfrak{D}}}_2\right]} = 0.$$ This equation can be inferred from (\[eq:psU11\^2Alg\]) or from the fact that ${\delta{\mathfrak{D}}}$ commutes with all other generators.
### Plus-plus commutators {#sec:cross++}
The proof for the commutators of $\overleftarrow{{{\mathfrak{T}}}}^+$ or $\overrightarrow{{{\mathfrak{T}}}}^+$ with any of the plus ${{\mathfrak{psu}}}(1,1|2)$ generators works in the same manner as in Appendix \[sec:++\], since commuting with ${{\mathfrak{x}}}$ generates the first quantum correction to all generators involved.
### Minus-plus commutators
We will show how these commutators vanish using $\overleftarrow{{{\mathfrak{T}}}}^-$. With appropriate switching of ${{\mathfrak{psu}}}(1,1|2)$ generators, the same proofs work for $\overrightarrow{{{\mathfrak{T}}}}^-$. Using (\[eq:Q2\]), we find $$\begin{aligned}
{\left\{\overleftarrow{{{\mathfrak{T}}}}^-,\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}\right\}}_3 &=& {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}\right\}} + {\left\{\overleftarrow{{{\mathfrak{T}}}_3}^-,\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow}\right\}} \nonumber \\
&=& {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}\right\}} - {\left\{{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{x}}}\right]},\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow}\right\}} \nonumber \\
&=& {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}\right\}} + {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{Q}}}_0}^{+\downarrow},{{\mathfrak{x}}}\right]}\right\}} \nonumber \\
&=& 2 {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,\overrightarrow{{{\mathfrak{Q}}}_2}^{+\downarrow}\right\}} \nonumber \\
&=& 2 {\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,\overrightarrow{{{\mathfrak{q}}}}^{+\downarrow}\right\}}\right]}\right\}} \nonumber \\
&=& 0. \end{aligned}$$
The last equality follows from the general identity, $${\left\{Q,{\left[Q,R\right]}\right\}}=0 \quad \mbox{if} \quad Q^2=0.$$
The same reasoning shows that the commutator of $\overleftarrow{{{\mathfrak{T}}}}^-$ with $\overrightarrow{{{\mathfrak{Q}}}}^{+\uparrow}$ vanishes.
Since $\overleftarrow{{{\mathfrak{T}}}}^-$ commutes with $\overrightarrow{{{\mathfrak{T}}}}^-$ and $\overleftarrow{{{\mathfrak{T}}}}^+$, (\[eq:vanishing\]) and (\[eq:Q2\]) imply that $${\left\{\overleftarrow{{{\mathfrak{T}}}}^-,\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}\right\}}_3 =0 \quad \mbox{and} {\left\{\overleftarrow{{{\mathfrak{T}}}}^-,\overleftarrow{{{\mathfrak{Q}}}}^{+\downarrow}\right\}}_3 =0.$$
The commutator of $\overleftarrow{{{\mathfrak{T}}}}^-$ with ${{\mathfrak{J}}}^{++}$ actually is already fixed to zero at this order because ${{\mathfrak{J}}}^{++}$ is generated by $\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow}$ and $\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}$. We show this using the commutators given in Appendix \[sec:psu112comm\], $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow},\overleftarrow{{{\mathfrak{Q}}}}^{+\uparrow}\right\}} &=& {\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow},{\left[\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow},{{\mathfrak{J}}}^{++}\right]}\right\}} \nonumber \\
&=& {\left[{\left\{\overrightarrow{{{\mathfrak{Q}}}}^{+\downarrow},\overleftarrow{{{\mathfrak{Q}}}}^{-\uparrow}\right\}},{{\mathfrak{J}}}^{++}\right]} \nonumber \\
&=& {\left[{{{\textstyle\frac{1}{2}}}}{{\mathfrak{J}}}^0+{{{\textstyle\frac{1}{2}}}}{{\mathfrak{R}}}^0,{{\mathfrak{J}}}^{++}\right]} \nonumber \\
&=& {{\mathfrak{J}}}^{++}. \end{aligned}$$
### Implied commutators
Using hermiticity and closure of the algebra, one can conclude that at $\mathcal{O}(g^3)$ all of the ${{\mathfrak{psu}}}(1|1)^2$ generators commute with all of the ${{\mathfrak{psu}}}(1,1|2)$ generators.
Commutators among the ${{\mathfrak{psu}}}(1|1)^2$ generators
------------------------------------------------------------
Using hermiticity, it will be sufficient to check the following equations:
$$\begin{gathered}
{\left[{{\mathfrak{L}}},{{\mathfrak{T}}}^+\right]}_3= {{\mathfrak{T}}}^+_3, \label{comml} \\
{\left\{\overrightarrow{{{\mathfrak{T}}}}^+,\overleftarrow{{{\mathfrak{T}}}}^-\right\}}_4={\left\{\overleftarrow{{{\mathfrak{T}}}}^+,\overrightarrow{{{\mathfrak{T}}}}^-\right\}}_4, \label{eq:halfdD} \\
{\left\{\overrightarrow{{{\mathfrak{T}}}}^+,\overrightarrow{{{\mathfrak{T}}}}^-\right\}}_4=0, \quad \mbox{and} \quad
{\left\{\overrightarrow{{{\mathfrak{T}}}}^+,\overleftarrow{{{\mathfrak{T}}}}^+\right\}}_4=0, \\
(\overrightarrow{{{\mathfrak{T}}}}^+)^2_4 = (\overleftarrow{{{\mathfrak{T}}}}^+)^2_4=0.
\end{gathered}$$
Both sides of (\[eq:halfdD\]) are equal to ${{{\textstyle\frac{1}{2}}}}{\delta{\mathfrak{D}}}_4$ by (\[eq:psU11\^2Alg\]).
### Commutators with ${{\mathfrak{L}}}$
(\[comml\]) follows from the corresponding leading order commutator and (\[eq:xl\]).
### ${\delta{\mathfrak{D}}}$ commutators
In fact, $${\left\{\overrightarrow{{{\mathfrak{T}}}_3}^+,\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}}={\left\{\overleftarrow{{{\mathfrak{T}}}_3}^+,\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}}. \label{eq:U+-T+-}$$ With the conjugate equation, this implies (\[eq:halfdD\]). Here is the proof of (\[eq:U+-T+-\]). $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{T}}}_3}^+,\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}}&=&{\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{x}}}\right]},\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]},\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]},\overleftarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^-,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]}\right\}} \nonumber \\
&=& {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{x}}}\right]}\right\}} \nonumber \\
&=& {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,\overleftarrow{{{\mathfrak{T}}}_3}^+\right\}}. \end{aligned}$$ To obtain the second, and second to last lines we have used (\[eq:definex\]) and applied the algebraic identity $${\left[Q,{\left\{Q,S\right\}}\right]}=0 \quad \mbox{if} \quad Q^2=0.$$ The proof also uses repeatedly the vanishing of the commutators of non-conjugate ${{\mathfrak{T}}}_1$’s.
### Non-conjugate ${{\mathfrak{T}}}$ commutators
The same reasoning as in the previous section works to show that $\overrightarrow{{{\mathfrak{T}}}}^+$ and $\overrightarrow{{{\mathfrak{T}}}}^-$ commute. $$\begin{aligned}
{\left\{\overrightarrow{{{\mathfrak{T}}}}^+,\overrightarrow{{{\mathfrak{T}}}}^-\right\}}_4&=& {\left\{\overrightarrow{{{\mathfrak{T}}}_3}^+,\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} + {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,\overrightarrow{{{\mathfrak{T}}}_3}^-\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{x}}}\right]},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} - {\left\{\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{{\mathfrak{x}}}\right]}\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{x}}}\right]},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} + {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{x}}}\right]},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& 2 {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{x}}}\right]},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{\left\{\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left[\overleftarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& {\left\{{\left[\overrightarrow{{{\mathfrak{T}}}_1}^-,{\left\{\overleftarrow{{{\mathfrak{T}}}_1}^+,{\left[\overrightarrow{{{\mathfrak{T}}}_1}^+,{{\mathfrak{h}}}\right]}\right\}}\right]},\overrightarrow{{{\mathfrak{T}}}_1}^-\right\}} \nonumber \\
&=& 0. \end{aligned}$$
For the commutator of $\overrightarrow{{{\mathfrak{T}}}}^+$ with $\overleftarrow{{{\mathfrak{T}}}}^+$, see appendices \[sec:++\] or \[sec:cross++\].
The squares of $\overrightarrow{{{\mathfrak{T}}}}^+$ and $\overleftarrow{{{\mathfrak{T}}}}^+$ vanish at $\mathcal{O}(g^4)$ for the same reasons that $\overrightarrow{{{\mathfrak{T}}}}^+$ commutes with $\overleftarrow{{{\mathfrak{T}}}}^+$ at this order.
[10]{}
J. M. Maldacena, *“The large [N]{} limit of superconformal field theories and supergravity”*, , [`hep-th/9711200`](http://arXiv.org/abs/hep-th/9711200). S. S. Gubser, I. R. Klebanov and A. M. Polyakov, *“Gauge theory correlators from non-critical string theory”*, , [`hep-th/9802109`](http://arXiv.org/abs/hep-th/9802109). E. Witten, *“Anti-de [S]{}itter space and holography”*, , [`hep-th/9802150`](http://arXiv.org/abs/hep-th/9802150). D. Berenstein, J. M. Maldacena and H. Nastase, *“Strings in flat space and pp waves from [$\mathcal{N} = 4$]{} super [Y]{}ang [M]{}ills”*, , [`hep-th/0202021`](http://arXiv.org/abs/hep-th/0202021). S. S. Gubser, I. R. Klebanov and A. M. Polyakov, *“A semi-classical limit of the gauge/string correspondence”*, , [`hep-th/0204051`](http://arXiv.org/abs/hep-th/0204051). S. Frolov and A. A. Tseytlin, *“Semiclassical quantization of rotating superstring in [$AdS_5 \times
S^5$]{}”*, , [`hep-th/0204226`](http://arXiv.org/abs/hep-th/0204226). S. Frolov and A. A. Tseytlin, *“Quantizing three-spin string solution in [$AdS_5 \times S^5$]{}”*, , [`hep-th/0306130`](http://arXiv.org/abs/hep-th/0306130). N. Beisert, J. A. Minahan, M. Staudacher and K. Zarembo, *“Stringing spins and spinning strings”*, , [`hep-th/0306139`](http://arXiv.org/abs/hep-th/0306139). S. Frolov and A. A. Tseytlin, *“Rotating string solutions: [A]{}d[S]{}/[C]{}[F]{}[T]{} duality in non- supersymmetric sectors”*, , [`hep-th/0306143`](http://arXiv.org/abs/hep-th/0306143). C. G. Callan Jr. et al., *“Quantizing string theory in [$AdS_5 \times S^5$]{}: Beyond the pp- wave”*, , [`hep-th/0307032`](http://arXiv.org/abs/hep-th/0307032). N. Beisert, S. Frolov, M. Staudacher and A. A. Tseytlin, *“Precision Spectroscopy of AdS/CFT”*, , [`hep-th/0308117`](http://arXiv.org/abs/hep-th/0308117). M. Kruczenski, *“Spin chains and string theory”*, , [`hep-th/0311203`](http://arXiv.org/abs/hep-th/0311203). M. Kruczenski, A. V. Ryzhov and A. A. Tseytlin, *“Large spin limit of $AdS_5\times S^5$ string theory and low energy expansion of ferromagnetic spin chains”*, , [`hep-th/0403120`](http://arXiv.org/abs/hep-th/0403120). C. G. Callan Jr., T. McLoughlin and I. J. Swanson, *“Holography beyond the [P]{}enrose limit”*, , [`hep-th/0404007`](http://arXiv.org/abs/hep-th/0404007). A. A. Tseytlin, *“Spinning strings and AdS/CFT duality”*, [`hep-th/0311139`](http://arXiv.org/abs/hep-th/0311139), in: *“From fields to strings: Circumnavigating theoretical physics”*, ed.: M. Shifman, A. Vainshtein and J. Wheater, World Scientific (2005), Singapore. N. Beisert, *“Higher-loop integrability in [$\mathcal{N}=4$]{} gauge theory”*, , [`hep-th/0409147`](http://arXiv.org/abs/hep-th/0409147). A. A. Tseytlin, *“Semiclassical strings and AdS/CFT”*, [`hep-th/0409296`](http://arXiv.org/abs/hep-th/0409296), in: *“String theory: from gauge interactions to cosmology”*, ed.: L. Baulieu, J. de Boer, B. Pioline and E. Rabinovici, Springer (2005), Berlin, Germany, 410p. K. Zarembo, *“Semiclassical Bethe ansatz and AdS/CFT”*, , [`hep-th/0411191`](http://arXiv.org/abs/hep-th/0411191). I. J. Swanson, *“Superstring holography and integrability in $AdS_5\times S^5$”*, [`hep-th/0505028`](http://arXiv.org/abs/hep-th/0505028). J. Plefka, *“Spinning strings and integrable spin chains in the [A]{}d[S]{}/[C]{}[F]{}[T]{} correspondence”*, [`hep-th/0507136`](http://arXiv.org/abs/hep-th/0507136). L. N. Lipatov, *“High-energy asymptotics of multicolor QCD and exactly solvable lattice models”*, , [`hep-th/9311037`](http://arXiv.org/abs/hep-th/9311037). L. N. Lipatov, *“Evolution equations in QCD”*, in: *“Perspectives in hadronic physics”*, ed.: S. Boffi, C. Ciofi Degli Atti and M. Giannini, World Scientific (1998), Singapore. J. A. Minahan and K. Zarembo, *“The [B]{}ethe-ansatz for [$\mathcal{N}=4$]{} super [Y]{}ang-[M]{}ills”*, , [`hep-th/0212208`](http://arXiv.org/abs/hep-th/0212208). N. Beisert and M. Staudacher, *“The $\mathcal{N}=4$ [S]{}[Y]{}[M]{} integrable super spin chain”*, , [`hep-th/0307042`](http://arXiv.org/abs/hep-th/0307042). N. Beisert, C. Kristjansen and M. Staudacher, *“The dilatation operator of [$\mathcal{N} = 4$]{} super Yang-Mills theory”*, , [`hep-th/0303060`](http://arXiv.org/abs/hep-th/0303060). N. Beisert, *“The su(2$/$3) dynamic spin chain”*, , [`hep-th/0310252`](http://arXiv.org/abs/hep-th/0310252). D. Serban and M. Staudacher, *“Planar [$\mathcal{N}=4$]{} gauge theory and the [I]{}nozemtsev long range spin chain”*, , [`hep-th/0401057`](http://arXiv.org/abs/hep-th/0401057). N. Beisert, V. Dippel and M. Staudacher, *“A novel long range spin chain and planar [$\mathcal{N}=4$]{} super [Y]{}ang- [M]{}ills”*, , [`hep-th/0405001`](http://arXiv.org/abs/hep-th/0405001). M. Staudacher, *“The factorized [S]{}-matrix of [C]{}[F]{}[T]{}/[A]{}d[S]{}”*, , [`hep-th/0412188`](http://arXiv.org/abs/hep-th/0412188). N. Beisert and M. Staudacher, *“Long-range psu$(2,2|4)$ [B]{}ethe ansaetze for gauge theory and strings”*, [`hep-th/0504190`](http://arXiv.org/abs/hep-th/0504190). N. Beisert, *“The su(2$/$2) Dynamic S-Matrix”*, [`hep-th/0511082`](http://arXiv.org/abs/hep-th/0511082). G. Mandal, N. V. Suryanarayana and S. R. Wadia, *“Aspects of semiclassical strings in [$AdS_5$]{}”*, , [`hep-th/0206103`](http://arXiv.org/abs/hep-th/0206103). I. Bena, J. Polchinski and R. Roiban, *“Hidden symmetries of the [$AdS_5 \times S^5$]{} superstring”*, , [`hep-th/0305116`](http://arXiv.org/abs/hep-th/0305116). V. A. Kazakov, A. Marshakov, J. A. Minahan and K. Zarembo, *“Classical / quantum integrability in [A]{}d[S]{}/[C]{}[F]{}[T]{}”*, , [`hep-th/0402207`](http://arXiv.org/abs/hep-th/0402207). N. Beisert, V. A. Kazakov, K. Sakai and K. Zarembo, *“The algebraic curve of classical superstrings on [$AdS_5 \times
S^5$]{}”*, [`hep-th/0502226`](http://arXiv.org/abs/hep-th/0502226). G. Arutyunov, S. Frolov and M. Staudacher, *“[B]{}ethe ansatz for quantum strings”*, , [`hep-th/0406256`](http://arXiv.org/abs/hep-th/0406256). A. V. Belitsky, G. P. Korchemsky and D. Mueller, *“Integrability of two-loop dilatation operator in gauge theories”*, [`hep-th/0509121`](http://arXiv.org/abs/hep-th/0509121). N. Beisert, *“The complete one-loop dilatation operator of $\mathcal{N}=4$ super [Y]{}ang-[M]{}ills theory”*, , [`hep-th/0307015`](http://arXiv.org/abs/hep-th/0307015). N. Beisert, *“The dilatation operator of [$\mathcal{N}=\mathord{}$4]{} super Yang-Mills theory and integrability”*, , [`hep-th/0407277`](http://arXiv.org/abs/hep-th/0407277). J. Ambjorn, R. A. Janik and C. Kristjansen, *“Wrapping interactions and a new source of corrections to the spin-chain / string duality”*, [`hep-th/0510171`](http://arXiv.org/abs/hep-th/0510171). C. Sieg and A. Torrielli, *“Wrapping interactions and the genus expansion of the 2- point function of composite operators”*, , [`hep-th/0505071`](http://arXiv.org/abs/hep-th/0505071). B. Eden and M. Staudacher, to appear. A. V. Kotikov, L. N. Lipatov and V. N. Velizhanin, *“Anomalous dimensions of [W]{}ilson operators in [$\mathcal{N} = 4$]{} [S]{}[Y]{}[M]{} theory”*, , [`hep-ph/0301021`](http://arXiv.org/abs/hep-ph/0301021). B. Eden, *“A two-loop test for the factorised [S]{}-matrix of planar [$\mathcal{N}
= 4$]{}”*, [`hep-th/0501234`](http://arXiv.org/abs/hep-th/0501234). N. Beisert, *“[B]{}[M]{}[N]{} operators and superconformal symmetry”*, , [`hep-th/0211032`](http://arXiv.org/abs/hep-th/0211032). H. Bethe, *“On the theory of metals. 1. Eigenvalues and eigenfunctions for the linear atomic chain”*, . B. Sutherland, *“A brief history of the quantum soliton with new results on the quantization of the [T]{}oda lattice”*, . A. Agarwal and G. Ferretti, *“Higher charges in dynamical spin chains for [S]{}[Y]{}[M]{} theory”*, [`hep-th/0508138`](http://arXiv.org/abs/hep-th/0508138). C. Anastasiou, Z. Bern, L. J. Dixon and D. A. Kosower, *“Planar amplitudes in maximally supersymmetric [Y]{}ang-[M]{}ills theory”*, , [`hep-th/0309040`](http://arXiv.org/abs/hep-th/0309040). Z. Bern, L. J. Dixon and V. A. Smirnov, *“Iteration of planar amplitudes in maximally supersymmetric [Y]{}ang-[M]{}ills theory at three loops and beyond”*, , [`hep-th/0505205`](http://arXiv.org/abs/hep-th/0505205). E. Witten, *“Perturbative gauge theory as a string theory in twistor space”*, , [`hep-th/0312171`](http://arXiv.org/abs/hep-th/0312171). F. Cachazo, P. Svrcek and E. Witten, *“MHV vertices and tree amplitudes in gauge theory”*, , [`hep-th/0403047`](http://arXiv.org/abs/hep-th/0403047). R. Britto, F. Cachazo and B. Feng, *“New recursion relations for tree amplitudes of gluons”*, , [`hep-th/0412308`](http://arXiv.org/abs/hep-th/0412308).
[^1]: Integrability in four-dimensional Yang-Mills theory was first observed by Lipatov [@Lipatov:1994yb; @Lipatov:1997vu].
[^2]: Actually, there are four generators, $\overrightarrow{{{\mathfrak{T}}}}^\pm$ and $\overleftarrow{{{\mathfrak{T}}}}^\pm$, but the additional sign index is unimportant for the points we make here.
[^3]: The right side of the second equation actually has two terms like the one shown, using the two pairs of generators with opposite signs and directions of arrows.
[^4]: I thank Niklas Beisert for explaining this to me.
[^5]: For recent promising work on the nature of wrapping interactions in AdS-CFT see [@Ambjorn:2005wa]. Also see [@Sieg:2005kd].
[^6]: We label this sector with ${{\mathfrak{su}}}(1,1|2)$ since this equals ${{\mathfrak{psu}}}(1,1|2) \ltimes {{\mathfrak{u}}}(1)$, which is the minimal algebra containing the full manifest symmetry.
[^7]: Note that, for simplicity, we call both commutators and anti-commutators, commutators; of course, the “commutator” of two fermionic generators is actually an anti-commutator.
[^8]: We set ${{\mathfrak{y}}}$ to zero without loss of generality.
[^9]: Of course, by $R$-symmetry, the sector with $\phi^\uparrow$’s has the same S-matrix and anomalous dimensions.
[^10]: For simplicity we drop the $p_i$, but all functions still depend on them. Note that $f_-$ and $f_0$ are unphysical. They will transform non-trivially under a similarity transformation for ${\delta{\mathfrak{D}}}$, unlike $S$ and $E$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A ray-rotation sheet consists of miniaturized optical components that function – ray optically – as a homogeneous medium that rotates the local direction of transmitted light rays around the sheet normal by an arbitrary angle \[A. C. Hamilton *et al.*, arXiv:0809.2646 (2008)\]. Here we show that two or more parallel ray-rotation sheets perform imaging between two planes. The image is unscaled and un-rotated. No other planes are imaged. When seen through parallel ray-rotation sheets, planes that are not imaged appear rotated[.]{}'
address: 'Department of Physics and Astronomy, Faculty of Physical Sciences, University of Glasgow, Glasgow G12 8QQ, United Kingdom'
author:
- 'Alasdair C. Hamilton and Johannes Courtial'
bibliography:
- '/Users/johannes/Documents/work/library/Johannes.bib'
title: 'Imaging with parallel ray-rotation sheets'
---
=1
Introduction\[introduction-section\]
====================================
Sheets composed of miniaturized optical components can perform [interesting]{} ray-optical transformations. Such [sheets can]{} act very differently from the [optical component on its own]{}. [A sheet comprising a pair of confocal]{} lenslet arrays, for example, can act approximately like the interface between optical media with different refractive indices, including negative refractive indices [@Courtial-2008a]. Another example is a Dove-prism sheet, an array of Dove prisms that flips one transverse component of the local direction of transmitted light rays (for example the $x$ component) [@Hamilton-Courtial-2008a]. If two parallel Dove-prism sheets, one immediately behind the other, are arranged such that their flip directions are orthogonal (“crossed Dove-prism sheets”), they flip both transverse components of the local light-ray direction – they act like the interface between optical media with opposite refractive indices [@Courtial-Nelson-2008]. If the two sheets’ flip axes are not orthogonal, the two Dove-prism sheets rotate the direction of transmitted light rays by twice the angle between the flip axes [@Hamilton-et-al-2009]. Such ray-rotation sheets are without wave-optical analog [@Hamilton-Courtial-2009].
When seen through a ray-rotation sheet, light rays originating from a given point appear to come from a different direction [@Hamilton-et-al-2009]. The direction change is such that a point light source is not geometrically imaged, unless the point light source lies in the plane of the ray-rotation sheet, when the sheet merely changes the direction in which light rays leave the source. [In this sense, a]{} ray-rotation sheet performs trivial imaging of the sheet plane into itself.
We investigate here geometric imaging with two or more parallel [idealized]{} ray-rotation sheets, separated by finite distances. We show that such parallel [idealized]{} ray-rotation sheets perform [unusual]{} imaging: two planes are imaged into each other with magnification $+1$ (and no image rotation); no other plane is imaged. [An optical system consisting of two or more parallel ray-rotation sheets has no optical axis. Therefore, the imaging quality does not vary across the image plane. An increase in aperture size in principle leads to no loss of ray-optical imaging quality. These imaging characteristics are quite different from those that can be achieved using (conventional or Fresnel) lenses.]{}
\[ray-rotation-sheets-section\]Ray-rotation sheets
==================================================
![\[ray-flipping-fig\]Optics of a single Dove-prism array. (a) A Dove-prism array is formed by a stack of Dove prisms that form a sheet. (b) Each individual Dove prism flips one of the transverse direction components of light rays passing through it, here the vertical component. (c) If the Dove prisms are miniaturized, the overall effect of a Dove-prism sheet on the flipped direction of transmitted light rays is equivalent to that of a planar interface between two materials with opposite refractive indices, $+n$ and $-n$.](DovePrismArray.pdf)
In all of this paper apart from this section we treat ray-rotation sheets in terms of their [idealized]{} effect on transmitted light rays, namely rotation around the local sheet normal. Here we [review]{} briefly one particular design of a ray-rotation sheet – two almost coplanar Dove-prism sheets – and [how transmission through such sheets can be described to a very good approximation as pure light-ray rotation]{} [[@Hamilton-et-al-2009]]{}.
A stack of very thin Dove prisms can form a Dove-prism sheet, as shown in Fig. \[ray-flipping-fig\](a). Each individual Dove prism flips one transverse component of the direction of transmitted light rays (Fig. \[ray-flipping-fig\](b)). In general, the transverse positions at which a light ray enters and exits the prism are different – the prism offsets the light rays, whereby the size of the offset is of the order of the size of the prism aperture. By miniaturizing the Dove prisms, this offset can be made almost arbitrarily small while still maintaining the Dove-prism sheet’s direction-flipping property. The effect is then equivalent to that of the interface with two materials with opposite refractive indices, but only as far as the flipped ray-direction component is concerned. (The effect of two Dove-prism sheets in parallel planes but with crossed flip directions is a flipping of both transverse ray-direction components, which in turn is ray-optically equivalent to an opposite-refractive-index interface.)
![\[ray-rotation-fig\] [Example of local light-ray rotation with Dove-prism sheets. (a) Structure of two parallel Dove-prism sheets, $S_1$ and $S_2$, that are rotated with respect to each other by $\alpha/2 = 45^\circ$. The two sheets individually flip the transverse direction of transmitted light rays, but with respect to different axes. (b) Successive flipping of a light-ray direction, $\mathbf{d}$, first by sheet $S_1$ (left frame), resulting in the intermediate direction $\mathbf{d}^\prime$, then by sheet $S_2$, resulting in direction $\mathbf{d}^{\prime \prime}$. Dove-prism sheets are represented by semi-transparent, light-blue, squares (seen from a 3D position from where they appear as parallelograms). Light-ray directions are represented by red arrows; the planes containing the light-ray directions and the local Dove-prism-sheet normal (dashed black lines) are indicated by semi-transparent red rectangles. The flip of the transverse light-ray direction is with respect to an axis in the sheet plane (dotted line). (c) A plot of the transverse light-ray directions (orthographic projection of the light-ray directions into the sheet plane or any other transverse plane) reveals that the two successive flips of the transverse ray direction are equivalent to rotation through an angle $\alpha$.]{}](rayRotation.pdf)
A pair of Dove-prism sheets, one immediately in front of the other, can form a ray-rotation sheet. The Dove-prism sheets are in parallel planes, but the directions of the Dove prisms in the two sheets – and with them the two sheets’ flip axes – are rotated with respect to each other around a sheet normal (Fig. \[ray-rotation-fig\](a)). [By analogy with]{} image rotation[, which]{} can be achieved by flipping the image twice with respect to different axes, ray rotation can be achieved by flipping the ray direction twice [(Fig. \[ray-rotation-fig\](b)).]{} In both cases the rotation angle is twice the angle between the flip axes. In order to achieve a ray-rotation angle $\alpha$, the two Dove-prism sheets therefore have to be rotated with respect to each other by $\alpha/2$.
A ray-rotation sheet based on Dove prisms suffers from a number of imperfections. These include a limited field of view; the small, but nevertheless non-zero, ray offset mentioned above; and diffraction effects, particularly in the case of small Dove prisms. A few of these issues are studied in more detail elsewhere [@Hamilton-et-al-2009].
[In the remainder of this paper we consider idealized ray-rotation sheets.]{}
\[twoRayRotationSheet-sec\]Geometric imaging with ray-rotation sheets
=====================================================================
If an optical system re-directs all the light rays from a point light source at point $L$ such that all the light rays that have passed through the optical system intersect again in another point, $L^\prime$, then the optical system images $L$ into $L^\prime$. $L$ is called the object, $L^\prime$ its image. Both $L$ and $L^\prime$ can be real or virtual: in the former case the actual light rays intersect, in the latter case their continuations.
![\[side-views-figure\] Trajectories of light rays that originate from three point light sources, $L_1$ to $L_3$, and pass through ray-rotation sheets. All light-ray trajectories are shown in red, apart from the trajectory of one light ray from each light source, which is highlighted in blue. The ray-rotation sheets are shown in light blue; the first rotates the direction of transmitted light rays through an angle $\alpha$ around the sheet normal, the second through an angle $\beta$. In the case of a single ray-rotation sheet, light rays from a point light source not located in the sheet plane generally do not intersect again after passage through the sheet (a). The same is true for passage through two ray-rotation sheets (b), unless the light source is located in the one transverse plane that is imaged by the two sheets, in which case all the light rays originating from the light source, $L_3$, intersect again in a point $L_3^\prime$ (c). The points at which the highlighted trajectories intersect the first sheet are marked $A_1$ to $A_3$, those where they intersect the second sheet are $B_2$ and $B_3$. Frames (d), (e) and (f) show the orthographic projections into a transverse plane of the light-ray trajectories respectively shown in (a), (b) and (c). The figure is drawn for $\alpha = \beta = 150^\circ$.](sideViews.pdf)
Fig. \[side-views-figure\](a) shows a cone of light rays leaving a point light source, $L_1$, and passing through a ray-rotation sheet with rotation angle $\alpha$. After passage through the sheet the rays form a twisted bundle; no two rays in the bundle intersect, so the sheet does not image $L$. This is in fact typical of [individual]{} ray-rotation sheets. Except in special cases (ray-rotation angles $\alpha = 0^\circ$ and $\alpha = 180^\circ$), [a ray-rotation sheet does]{} not image point light sources in any plane other than the sheet plane, which is imaged again into the sheet plane[, in the sense discussed in section \[introduction-section\]]{}.
Fig. \[side-views-figure\](b) shows the trajectories of a cone of light rays from another point light source, $L_2$, passing through two parallel ray-rotation sheets with rotation angles $\alpha$ and $\beta$, respectively. Like the typical single-ray-rotation-sheet case, passage through two ray-rotation sheets results in a twisted bundle of non-intersecting rays. The sheets do not produce an image of the light source, and this situation is again typical.
Fig. \[side-views-figure\](c) shows the special case in which imaging takes place. This time the light source, $L_3$, is positioned at a distance from the first ray-rotation sheet such that, after passage through both sheets, all light rays originating from $L_3$ intersect again in the same point, $L_3^\prime$, the image of $L_3$.
Fig. \[side-views-figure\](d), (e) and (f) show these three cases from a different perspective. These plots are orthographic projections of the ray bundles and sheets into a transverse plane. From the projections of the twisted ray bundles in the first two, non-imaging, cases, Fig. \[side-views-figure\](d) and (e), it is immediately clear that there is no point at which all the rays from the same point light source intersect. The third, imaging, case (Fig. \[side-views-figure\](f)) is different. When following the projection of the highlighted ray trajectory it can be seen that, after transmission through both ray-rotation sheets, it passes through the projection of the point light source. As this is the case not only for the highlighted light ray, but for light rays leaving the point light source in any direction, *all* light rays intersect at the point where the projections of the ray and the point light source intersect.
![ \[imaging-trajectory-figure\]Geometry of a light-ray trajectory in the imaging case. The trajectory [(solid dark-blue arrow)]{} is the same as that highlighted in Fig. \[side-views-figure\](c) and (f). [(a) Three-dimensional representation. The ray-rotation sheets (rotation angles $\alpha$ and $\beta$, respectively) are shown in light blue. $\theta$ is the angle between the ray trajectory and the normal to both ray-rotation sheets. (The angle is shown in the form of a red segment.) The dashed line is the sheet normal through the point light source, $L_3$, and its image, $L_3^\prime$. The distances $o$ (between $L_3$ and the first sheet), $s$ (between the two sheets), and $i$ (between the second sheet and $L_3^\prime$) are marked by double-sided arrows. (b) Orthographic projection into a transverse plane. The angles $\alpha$ and $\beta$ by which the direction of the projected ray changes are simply the angles by which the two ray-rotation sheets rotate the ray around the local sheet normal. The relevant angles are marked as red segments.]{}](imagingTrajectory.pdf)
We now study this quantitively. We concentrate on one light-ray trajectory in the imaging case (Fig. \[imaging-trajectory-figure\](a)). First we notice that, as a ray-rotation sheet rotates the light-ray direction around the sheet normal, it does not change the angle between the light ray and the sheet normal. If two or more successive ray-rotation sheets share the same sheet normal, the angle between the ray trajectory and the direction of the normal remains constant. This is the case in our parallel ray-rotation sheets. We call the angle with the sheet normal $\theta$. The transverse and longitudinal distances, respectively $\Delta r$ and $\Delta z$, travelled by a light ray that is inclined by an angle $\theta$ with respect to the sheet normal (which we choose to be the $z$ direction), are then related through the equation $$\frac{\Delta r}{\Delta z} = \tan \theta.
\label{polar-angle-equation}$$
For the light-ray trajectory’s orthographic projection into a transverse plane (Fig. \[imaging-trajectory-figure\](b)), equation (\[polar-angle-equation\]) implies that the side lengths of the triangle $L_3 A_3 B_3$ (where $A_3$ and $B_3$ are the points where the trajectory intersects the first and second sheet, respectively – see Figs \[side-views-figure\] and \[imaging-trajectory-figure\]) are the product of the corresponding $z$ distance and $\tan \theta$. We call the distance between the point light source, $L_3$, and the first sheet the object distance, $o$, we call the separation between the first and second sheet $s$, and we call the distance between the second sheet and the image of $L_3$, $L_3^\prime$, the image distance, $i$. Basic trigonometry then leads to the following object distance, given two sheets with ray-rotation angles $\alpha$ and $\beta$, separated by $s$: $$\label{o-equation}
o = - \frac{\sin\beta}{\sin(\alpha+\beta)} s.$$ Similarly, the image distance is $$\label{i-equation}
i = - \frac{\sin\alpha}{\sin(\alpha+\beta)} s.$$
![ \[imaging-trajectories-cone-and-fan-figure\][Focussing of light rays with different angles $\theta$ with the common sheet normal by a pair of parallel ray-rotation sheets (shown as light blue, semi-transparent, squares). The red light rays leave the point light source, $L_3$, with an angle $\theta = 17^\circ$ with respect to the sheet normal; they are the light rays shown in Fig. \[side-views-figure\](c), which is calculated for the same ray-rotation angles ($\alpha = \beta = 150^\circ$), the same separation between the ray-rotation sheets, and the same position of $L_3$. The blue light rays leave $L_3$ at angles $\theta = 0^\circ, 5^\circ, 10^\circ, ... 40^\circ$ with respect to the sheet normal. All light rays intersect again in the same point, $L_3^\prime$, the image of $L_3$. To facilitate appreciation of the 3D structure of the ray trajectories, the supplementary material contains a movie (MPEG-4, 1.4MB) of the same arrangement seen from different viewing positions.]{}](imagingTrajectoriesConeAndFan.pdf)
There are a number of things to note about these equations for object and image distance. Firstly, they are independent of the angle $\theta$. This means that rays that leave the light source at any angle $\theta$ with the common sheet normal are imaged at the point $L_3^\prime$ (provided, of course, they pass through both ray-rotation sheets) [ (Fig. \[imaging-trajectories-cone-and-fan-figure\])]{}. So $L_3^\prime$ is the geometrical image of $L_3$. [The independence from the angle $\theta$, together with the fact that no approximations were used in finding that $L_3$ is imaged into $L_3^\prime$, implies that idealized ray-rotation sheets perform ray-optically perfect imaging. Specifically, an increase in aperture size leads to no loss of ray-optical imaging quality. Obviously, combinations of particular implementations of ray-rotation sheets are limited by how closely the particular implementations approximate idealized ray-rotation sheets.]{}
[Secondly]{}, equations (\[o-equation\]) and (\[i-equation\]) hold not only for the particular position of light source $L_3$, but for any light-source position in the plane a distance $o$ in front of the first [ray-rotation]{} sheet[, which is then imaged to a corresponding position in the plane a distance $i$ behind the second ray-rotation sheet]{}. This means the entire plane is imaged. [(Similarly, the fact that any point light source that is a distance other than $o$ in front of the first ray-rotation sheet results in a twisted bundle of non-intersecting rays means that no other plane is imaged.)]{} As the orthographic projections into a transverse plane of any light source and its image coincide, the image of the plane is the same size as the object and upright, so the magnification of the imaging process is $M = +1$.
![\[simulation-figure\][Simulated]{} view through a pair of ray-rotation sheets. The ray-rotation angle of each sheet is $137^\circ$; in units of the floor-tile length, the sheets are separated by a distance $s=5$, so the object and image distance are $o=i=3.42$. (a) [A green chess piece is placed behind the ray-rotation sheets, in the object plane. The rendering parameters are chosen such that only the image plane is in focus – the picture is a simulation of a photo taken with a camera that is focussed on the image plane and that has a finite-size aperture. As expected, a sharp image of the chess piece can be seen in the image plane.]{} (b) [Additional]{} (upright) chess pieces [are placed]{} in different planes behind the ray-rotation sheets. [As before, the]{} green chess piece is in the object plane (at distance $o$), the other pieces are not ([they are]{} at distances $o/2$ (blue), $2 o$ (orange), and $4 o$ (red)). This time the rendering parameters were chosen so that all planes are rendered in focus[; the frames are now simulations of photos taken with a camera with a negligibly small aperture]{}. [Therefore all chess pieces can be seen in sharp focus, but note]{} that there is still only one object and one image plane that are being imaged into each other by the pair of ray-rotation sheets. [To demonstrate this,]{} three frames were calculated for different camera positions, namely (from left to right) right of the central sheet normal, on the central sheet normal, and left of the central sheet normal. [This horizontal camera movement makes the chess pieces that are not in the object plane appear to move vertically, which cannot be reconciled with these chess pieces being imaged, as discussed in the main text.]{} The figure was calculated using the ray-tracing software POV-Ray [@POV-Ray], which simulated transmission through the detailed structure of two ray-rotation sheets, each consisting of two Dove-prism sheets [@Hamilton-et-al-2009], each in turn consisting of 1000 Dove prisms. ](simulation.pdf)
Figure \[simulation-figure\](a) shows a simulation of a chess piece in the object plane of two ray-rotation sheets, seen through the sheets. [Note that all frames in Fig. \[simulation-figure\] are ray-tracing simulations of the view through the detailed structure of ray-rotation sheets consisting of Dove-prism sheets (section \[ray-rotation-sheets-section\]).]{} The (simulated) camera is focussed on the image plane, and a sharp image of the chess piece can be seen there, as expected.
Fig. \[simulation-figure\](b) shows a simulation of chess pieces in different planes, [again]{} seen through two ray-rotation sheets. This time the camera is not focussed on any particular plane, so all pieces appear sharp. That only the chess piece in the object plane is actually imaged can be seen when looking at the same scene from [different angles]{}: if the camera is moved sideways, the pieces that are not imaged appear to move sideways with respect to each other – as images at different distances should –, but they also appear to move up and down, as no stationary object or image would do when the camera is moved sideways. [(Note that the camera movement simulated in Fig. \[simulation-figure\](b) can be interpreted as seeing the same scene from different positions within a much larger aperture. The fact that any objects that are not imaged appear to move would then lead to those objects appearing blurred on a photo taken with a lens with such a much larger aperture.)]{}
[Thirdly, the absence of any center of magnification (as the magnification is $M = +1$) or any rotation axis (contrary to what might be expected from transmission through a combination of ray-rotation sheets, the image is not rotated with respect to the object) implies that there is no optical axis. This also follows from the lack of any preferred axis in each of the constituent ray-rotation sheets. ]{}
[These first three characteristics of imaging with parallel ray-rotation sheets are quite different from those of any system of lenses, irrespective of whether or not the lenses are thick or thin and whether or not the lenses are in the form of Fresnel lenses, which bear a resemblance to Dove-prism sheets. Specifically, any system of lenses images not only points in one plane, but points in any position (unless the lens system contains a planar diffuser, in which case only the planes into which the diffuser plane is imaged are imaged into each other, whereby the diffuser plane contains an intermediate image); and any lens has an optical axis, which leads to a preferred axis. The lack of an optical axis makes ray-rotation-sheet imaging more akin to imaging with a “perfect” lens formed by the planar interface between media with refractive indices of equal magnitude but opposite sign [@Pendry-2000]; in fact, transmission through just such an interface is the same as idealized light-ray rotation through $180^\circ$ [@Hamilton-et-al-2009]. In other ways, for example imaging of only a single plane, imaging with parallel ray-rotation sheets differs from imaging with a perfect lens.]{}
[Fourthly]{}, equations (\[o-equation\]) and (\[i-equation\]) can give positive and negative object and/or image distances, respectively corresponding to real and virtual objects/images.
Finally, as the magnification between object and image plane is $+1$, these planes are principal planes [@Smith-1978]. The effect of the two sheets is then that of a single sheet with ray-rotation angle $\alpha + \beta$ (Fig. \[imaging-trajectory-figure\](b)) and principal planes in the object and image planes. In this way, we therefore recover the result that a single ray-rotation sheet trivially images the sheet plane into itself. This argument can also easily be extended to $N$ ray-rotation sheets with rotation angles $\alpha_1, \alpha_2, ..., \alpha_N$ and separations $s_1, s_2, ..., s_{N-1}$: they act like a single ray-rotation sheet with a ray-rotation angle $\alpha = \sum_j \alpha_j$ and principal planes (that is, object and image planes with magnification $M=+1$) at distances $$o= - \frac{\sum_j s_j \sin \left( \sum_{k=j+1}^N \alpha_k \right)}{\sin \alpha}, \quad
i = - \frac{\sum_j s_j \sin \left( \sum_{k=1}^j \alpha_k \right)}{\sin \alpha}$$ in front of the first sheet and behind the last sheet, respectively.
\[conclusions-section\]Conclusions
==================================
[Combinations of ray-rotation sheets can perform geometrical imaging that is, in principle, perfect and which is different from the geometrical imaging performed by simple combinations of lenses.]{} [The unusual imaging properties of ray-rotation sheets]{} might well find applications, for example in range finding.
Acknowledgments {#acknowledgments .unnumbered}
===============
Many thanks to John Nelson for help with the POV-Ray simulations. A.C.H. is supported by the UK’s Engineering and Physical Sciences Research Council (EPSRC). J.C. is a Royal Society University Research Fellow.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give axioms in the language of rings augmented by a 1-ary predicate symbol $Fin(x)$ with intended interpretation in the Boolean algebra of idempotents as the ideal of finite elements, i.e. finite unions of atoms. We prove that any commutative unital ring satisfying these axioms is elementarily equivalent to a restricted product of connected rings. This is an extension of the results in [@elem-prod] for products. While the results in [@elem-prod] give a converse to the Feferman-Vaught theorem for products, our results prove the same for restricted products. We give a complete set of axioms in the language of rings for the ring of adeles of a number field, uniformly in the number field.'
address:
- 'St Hilda’s College, University of Oxford, Cowley Place, Oxford OX4 1DY, UK'
- 'School of Mathematical Sciences, Queen Mary, University of London, Mile End Road, London E1 4NS, and School of Mathematics, University of Edinburgh, James Clerk Maxwell Building, Peter Guthrie Tait Road Edinburgh EH9 3FD, UK'
author:
- Jamshid Derakhshan
- 'Angus Macintyre${}^{\dag}$'
bibliography:
- 'bibadeles.bib'
title: Axioms for Commutative Unital Rings elementarily Equivalent to Restricted Products of Connected Rings
---
**Introduction** {#sec-introduction}
================
This paper is a natural sequel to [@elem-prod] and the main results and proofs are natural extensions of those in [@elem-prod]. In many cases we will simply refer to the material from [@elem-prod]
[@elem-prod] deals with the model theory of products of connected unital rings, and can be construed as providing a partial converse to the Feferman-Vaught Theorem [@FV] in the special case of products $\prod_{i\in I} R_i, i\in I,$ where $R_i$ are connected commutative unital rings and $I$ an index set (Recall that a commutative ring $R$ is connected if $0,1$ are the only idempotents of $R$). The converse concerns the issue of providing axioms for [*rings elementarily equivalent to rings $\prod_{i\in I} R_i$ as above*]{}. The solution of this problem is given in [@elem-prod] and, inter alia has applications to non-standard models of PA (first order Peano arithmetic) in [@PDAJM].
In this paper we start with rings $\prod_{i\in I} R_i, i\in I,$ as above, but work with certain subrings, namely restricted products with respect to a formula $\varphi(x)$ of the language of rings (in a single variable $x$), defined as the set of all $f\in \prod_{i\in I} R_i$ so that $\{i: R_i \models \varphi(f(i))\}$ is cofinite. Provided that $\varphi(x)$ defines a unital subring of each $R_i$, the above subset is in fact a subring of $\prod_{i\in I} R_i$ (not in general definable).
We obtain, for restricted products, results exactly analogous to those of [@elem-prod] for products. Given $\varphi(x)$, we provide axioms in the language of rings augmented by a predicate $Fin(x)$, and prove that any commutative unital ring satisfying these axioms is elementarily equivalent to a restricted product, with respect to $\varphi(x)$, of connected rings. The standard interpretation of $Fin(x)$ in any Boolean algebra, and in particular in the Boolean algebra of idempotents of $R$ is the ideal of finite elements, i.e. finite unions of atoms.
The canonical example of a such a restricted product, very important in number theory (see Cassels and Frohlich [@CF]), is $\A_K$, the ring of adeles over a number field $K$. Here $I$ is the set of normalized absolute values $v$ on $K$ up to equivalence, $R_i$ is the completion $K_v$ of $K$ at $v$, and $\varphi(x)$ is a formula of the language of rings that defines, uniformly for all $v$, the valuation ring $\cO_v$ of $K_v$. (That there is such a $\varphi(x)$ is nontrivial, and it is an important result that there is an $\exists \forall$-formula $\varphi(x)$ that works uniformly for all $\cO_v$, and hence for all adele rings uniformly in $K$, see [@CDLM] and [@DM-ad]).
In the case of adeles $\A_K$, the set of idempotents with finite support is definable by a formula of the language of rings independently of $K$ (cf. [@DM-ad], and a new proof given at the end of this paper in a ring-theoretic situation). Thus we can derive axioms in the language of rings for the adeles, uniformly in $K$.
**The Boolean algebra of idempotents of a ring**
================================================
We shall denote the language of rings by $\cL_{rings}=\{+,.,0,1\}$ and the language of Boolean algebras by $\mathcal{L}_{Boolean}=\{\vee,\wedge,\neg,0,1\}$. We start by recalling the definition of a restricted product of structures with respect to a formula (cf. [@DM-ad], [@DM-supp], [@DM-ad2]).
\[def-rest\] Let $\cL$ be a language and $(\cM_i)_{i\in I}$ a family of $\cL$-structures. Let $\varphi(x)$ be a $\cL_{rings}$-formula in the single variable $x$. The restricted direct product of $\cM_i$ with respect to $\varphi$ (also called product restricted by $\varphi$) is the subset of the product $\prod_{i\in I} \cM_i$ consisting of all $f$ such that $\cM_i \models \varphi(f(i))$ for all but finitely many $i\in I$.
We denote this restricted product by $\prod_{i\in I}^{(\varphi)} \cM_i$. It is a substructure of the generalized product defined by Feferman and Vaught in [@FV] provided that $\varphi(\cM_i)$ is a substructure of $\cM_i$ for all $i\in I$. The results in [@FV] and [@DM-ad], [@DM-ad2], [@DM-supp] yield general quantifier eliminations for such restricted products, where $\cL$ is any many-sorted language. One can deduce, among other results, quantifier elimination for adeles, and results on definable subsets of adeles and their measures.
Atoms and Stalks
----------------
We follow as much as possible the development from Section 1 of [@elem-prod].
Let $R$ be a commutative unital ring. The set $\{x: x=x^2\}$ of idempotents is a Boolean algebra, denoted by $\B$, with operations $$e \wedge f=ef,$$ $$\neg e=1-e,$$ $$0=0,$$ $$1=1,$$ $$e\vee f=1-(1-e)(1-f)=e+f-ef.$$ It carries a partial ordering defined by $e\leq f\Leftrightarrow ef=e$ (which is $\cL_{rings}$-definable). The [*atoms*]{} of $\B$ are the minimal idempotents (with respect to the ordering) that are not equal to $0,1$. (In fact we [*assume*]{} $0\neq 1$).
Note that if $R$ is a product, over an index set $I$, then $\B$ is isomorphic to the Boolean algebra of subsets of $I$ via “characteristic functions”.
\[lem1\] For any $e$ in $\B$ we have $R/(1-e)R \cong eR \cong R_e$, where $R_e$ is the localization of $R$ at $\{e^n: n\geq 0\}$.
See Lemma 1 in [@elem-prod].
We call $R_e$ the stalk of $R$ at $e$. Of special important are the $R_e$ for [*atoms*]{} $e$.
Now we gradually impose axioms on $R$, in order to get a converse to Feferman-Vaught for restricted products.
$\B$ is atomic.
This holds if $R$ is a restricted product of [*connected*]{} rings. One does not even need the restricting formula $\varphi(x)$ to be definable. Moreover $R$ and the unrestricted product have the same idempotents. The basic example is $\A_K$ embedded in $\prod_{v} K_v$.
Now we go through a series of consequences of the current axioms, and additions of new axioms.
\[lem2\] If $f\in \B$, and $f\neq 0$, then $f=\bigvee\{e: \text{e ~ an~ atom}, ~ e\leq f\}$, (where $\bigvee$ is union or supremum).
This is Lemma 2 in [@elem-prod].
We turn to Boolean values and follow 1.3 of [@elem-prod].
Let $\Theta(x_1,\dots,x_n)$ be a formula of the language of rings, and $f_1,\dots,f_n\in R$. Then $[[\Theta(f_1,\dots,f_n)]]$ is defined as $$\bigvee_{e} \{e: e~\text{an~atom},~R_e\models \Theta((f_1)_e,\dots,(f_n)_e)\}$$ provided $\bigvee$ exists in $\B$. Here $(f)_e$ is the natural image of $f$ in $R_e$ (or, seen from perspective of Lemma \[lem1\], $f+(1-e)R$).
$[[\Theta(f_1,\dots,f_n)]]$ exists (as an element of $\B$).
If $R$ is a product of structures then $\B$ is complete, however completeness of a Boolean algebra is not a first-order property.
Axiom 2 is a substitute for completeness (and follows from it).
Axiom 2 is true in a restricted product of connected rings with respect to a given formula $\varphi(\bar x)$.
Boolean Values and Patching
---------------------------
The $[[\Phi(f_1,\dots,f_n)]]$ are in $\B$, and occur in [@FV] in the context of products, with a different notation. The $[[...]]$ notation comes from Boolean valued model theory \cite{}.
The next Lemmas come from 1.4 of [@elem-prod].
\[lem3-5\] Let $\Theta_1,\Theta_2,\Theta_3$ be $\cL_{rings}$-formulas in the variables $x_1,\dots,x_n$. Then for any $f_1,\dots,f_n \in R$,
- $[[(\Theta_1 \wedge \Theta_2)(f_1,\dots,f_n)]]=[[\Theta_1(f_1,\dots,f_n)]]\wedge [[\Theta_2(f_1,\dots,f_n)]]$,
- $[[(\neg \Theta)(f_1,\dots,f_n)]]=\neg [[\Theta(f_1,\dots,f_n)]],$
- $[[(\Theta_1 \vee \Theta_2)(f_1,\dots,f_n)]]=[[\Theta_1(f_1,\dots,f_n)]] \vee [[\Theta_2(f_1,\dots,f_n)]]$.
These statements are Lemmas 3-5 in [@elem-prod]
These are some of the ingredients used in inductive proofs of result in [@FV].
We add another axiom, taken from 1.4. of [@elem-prod]
For any atomic formula $\Theta(x_1,\dots,x_n)$ of the language of rings, $$R\models \Theta(f_1,\dots,f_n) \Leftrightarrow \B\models [[\Theta(f_1,\dots,f_n)]]=1.$$
This is evidently true in restricted products, no matter what $\varphi(x)$ is.
Now we [*fix*]{} a $\varphi(x)$, and aim for axioms true in restricted direct products $R$ with respect to $\varphi(x)$.
We come now to a fundamental point. Classically the notion of restricted product appeals to the absolute notion of which is not, of course, first-order. We are aiming for first-order axioms in some natural formalism. As already suggested, we are going to use an idea from Feferman -Vaught [@FV] of working with Boolean algebras $\B$ with a distinguished subset $\mathcal{F}in$, which in the case of the power set algebra is the ideal of finite sets. In the case of a Boolean algebra of idempotents, $\mathcal{F}in$ will be the set of finite idempotents, as explained earlier.
We will shortly be concerned with other interpretations of a predicate symbol for $\mathcal{F}in$, indispensable for understanding nonstandard models of our axioms (and in particular nonstandard models of the theory of the adeles.
But first we use provisional “axioms” where “finite” really means finite, and “finite idempotents ” really mean finite idempotents, and “cofinite” really means cofinite.
We could avoid this ,and pass directly to the general case. But we prefer to discuss a provisional axiom connected to the kind of patching used in [@FV]
For all $\Theta(x_1,\dots,x_n,w)$, $f_1,\dots,f_n$, there is a $g\in R$ such that if $$[[\exists w (\varphi(w) \wedge \Theta(f_1,\dots,f_n,w))]]$$ is cofinite in $$[[\exists w \Theta(f_1,\dots,f_n,w)]],$$ then $[[\exists w \Theta(f_1,\dots,f_n,w)]]$ is cofinite in $[[\Theta(f_1,\dots,f_n,g)]]$.
This is clearly true in restricted products with respect to $\varphi(x)$ (use Axiom of Choice).
[@elem-prod] has a simpler Axiom 4 for the unrestricted product case. That is not needed here.
From now on, we will get involved with not only $\B$, but with the ideal $\mathcal{F}in$ in $\B$ consisting of finite elements of $\B$, i.e. finite unions of atoms.
We have to enrich the first-order language of Boolean algebras by a 1-ary predicate symbol $Fin(x)$. For our purposes $\B$ will be atomic as above, and $Fin(x)$ interpreted as the ideal of finite support idempotents. The interpretation of $Fin(x)$ in a Boolean algebra of sets, e.g. the powerset $\P(I)$ of a set $I$ is the (Boolean) ideal of finite sets.
However, note that Axiom $4^+$ is [*not*]{} first-order. Any anxieties about this should be removed by considering the result that the theory of the class of all infinite atomic Boolean algebras in the language of Boolean algebras augmented by $Fin(x)$ is axiomatizable and complete (and admits quantifier elimination). This is proved first by Tarski but we give a new proof with explicit axioms in [@DM-bool]. See also Section \[sec-bool\] below. [@DM-bool] contains a unified treatment that includes further expansions by predicates for “congruence conditions on cardinality of finite sets”.
We return to this matter later, reformulating Axiom $4^+$ in terms of $Fin(x)$.
Partitions
----------
In order to sketch a proof of a useful generalization of [@FV] to our more restrictive situation (rings $R$ satisfying the axioms listed above) we need to review several notions of partition used in [@FV].
In a Boolean algebra $B$ a partition is a finite sequence $<Y_1,\dots,Y_m>$ of elements of $B$ such that $$Y_1 \vee \dots \vee Y_m=1$$ and $Y_i \wedge Y_j =0$ if $i\neq j$. (We do not insist that each $Y_i\neq 0$, but do insist that the sequence is finite).
We note that in the definition of partition “finite” will always mean finite.
For a first-order language $L$, a fixed $m$, and $L$-formulas $$\Theta_1(x_1,\dots,x_m),\dots,\Theta_m(x_1,\dots,x_m)$$ the sequence $<\Theta_1,\dots,\Theta_m>$ is a [*partition*]{} if the formulas $\Theta_1 \vee \dots \vee \Theta_m$ and $\neg(\Theta_i \wedge \Theta_j)$ (where $i\neq j$) are logically valid.
(This is of course ultimately a special case of Notion 1).
The basic lemmas about disjunctive normal form in propositional calculus, when applied to formulas $$\psi_1(x_1,\dots,x_m),\dots,\psi_l(x_1,\dots,x_m)$$ give constructively a partition whose elements are propositional combinations of the $\psi_i$’s. This is used crucially in [@FV].
The final result we need before sketching [@FV] for all our rings is.
\[lem6+\] (Analogue of Lemma 6 in [@elem-prod]) Suppose $Y_1,\dots,Y_k$ is a partition of $\B$. Suppose the sequence $$<\Theta_1(x_1,\dots,x_m,x_{m+1}),\dots,\Theta_k(x_1,\dots,x_m,x_{m+1})>$$ is a partition. Suppose $f_1,\dots,f_m\in R$ and $$Y_j\leq [[\exists x_{m+1} \Theta_j(f_1,\dots,f_m,x_{m+1})]]$$ for each $j$. Suppose in addition that for each $j$ $$Y_j \wedge \neg [[\exists x_{m+1} \varphi(x_{m+1}) \wedge \Theta_j(f_1,\dots,f_m,x_{m+1})]]$$ is finite. Then there is a $g$ in $R$ so that $$Y_j\subseteq [[\Theta_j(f_1,\dots,f_m,g)]]$$ for all $j$.
Apply Axiom $4^+$ to each $Y_j$ and $\Theta_j$, $f_1,\dots,f_m$ to get $g_j\in R$ so that $Y_j\subseteq [[\Theta_j(f_1,\dots,f_m,g_j)]]$. Now let $g=\sum_j g_j.e_j$, where $e_j$ is the idempotent corresponding to $Y_j$.
Later we re-do this in terms of $Fin$ (subject to Tarski’s axioms).
The Augmented Boolean Formalism {#sec-bool}
-------------------------------
We adjoin to the first-order language for Boolean algebras $\cL_{Boolean}$ a unary predicate symbol $Fin(x)$. Denote the augmented language by $\cL_{Boolean}^{fin}$. The standard interpretation of $Fin(x)$ in a Boolean algebra is the ideal $\mathcal{F}in$ of finite elements, i.e. finite unions of atoms. However, the class of such $(\B,\mathcal{F}in)$ is not elementary, and it is important for us to give a computable set of axioms complete up to specifying the the number of atoms below an element. We do this in [@DM-bool] (as part of a new expansion of $\cL_{Boolean}$), but the original work was done by Tarski (see [@FV]). Here are the essential points.
Let $T$ be the theory of infinite atomic Boolean algebras in the language $\mathcal{L}_{Boolean}$ expanded by the definable relations $C_k(x)$ ($k=1,2,\dots$) with the interpretation that $x$ has at least $k$ atoms $\alpha \leq x$. Tarski proved that in this language the theory of infinite atomic Boolean algebras is complete, admits quantifier elimination, and is decidable (see [@FV], [@DM-bool]). The axioms for this theory state that the models are infinite Boolean algebras and every nonzero element has an atom below it. Let $\sharp(x)$ denote the number of atoms $\alpha$ such that $\alpha \leq x$.
Now we further expand the given language by adding the predicate $Fin(x)$ with the above interpretation in any Boolean algebra, and obtain $\cL_{Boolean}^{fin}$. We add to the axioms of $T$ the axioms stating that $\mathcal{F}in$ is a proper ideal, the sentence $$\forall x (\neg Fin(x) \Rightarrow (\exists y)(y<x \wedge \neg Fin(y) \wedge \neg Fin(x-y))).$$ and, for each $n<\omega$, the sentence $\forall x (\sharp(x)\leq n \Rightarrow Fin(x))$. This defines an $\cL_{Boolean}^{fin}$-theory $T^{fin}$.
\[bool1\][@FV],[@DM-bool] The theory $T^{fin}$ of infinite atomic Boolean algebras with the set of finite sets distinguished is complete, decidable and has quantifier elimination with respect to all the $C_n$, $(n\geq 1)$, and $Fin$ (i.e. in the language $\mathcal{L}_{Boolean}^{fin}$). The axioms required for completeness are the axioms of $T$ together with sentences expressing that $Fin$ is a proper ideal, the sentence $$\forall x (\neg Fin(x) \Rightarrow (\exists y)(y<x \wedge \neg Fin(y) \wedge \neg Fin(x-y))).$$ and, for each $n<\omega$, the sentence $\forall x (\sharp(x)\leq n \Rightarrow Fin(x))$.
[**Note:**]{} This is important for measurability of definable sets in adele rings as in [@DM-ad].
In [@DM-bool] we prove that $\mathcal{F}in$ is not definable in the language of the theory $T$.
Modifying the Preceding (provisional) Axioms for $R$
----------------------------------------------------
Our axioms given so far are not first-order. To rectify this, we first do the following. Work with rings $R$ together with a distinguished ideal in $\B$. In the adelic cases this ideal will be the ideal of finite idempotents, but we will also be interested in other ideals. We write $\mathcal{F}in$ for the distinguished ideal. We modify axiom $4^+$ to Axiom $4^{Fin}$ (which is still not first-order). We let $\mathcal{F}in$ be the set of realizations in $R$ of the predicate $Fin(x)$.
. There is an ideal $\mathcal{F}in$ in $\B$ so that $(\B,\mathcal{F}in)\models T^{fin}$, and such that for all $\Theta(x_1,\dots,x_n,w), f_1,\dots,f_n$ there is a $g\in R$ such that if $$[[\exists w \Theta(f_1,\dots,f_n,w)]] \cap \neg [[\exists w (\varphi(w) \wedge \Theta(f_1,\dots,f_n,w))]]\in \mathcal{F}in,$$ then $$[[\exists w \Theta(f_1,\dots,f_n,w)]]\cap \neg [[\Theta(f_1,\dots,f_n,g)]]\in \mathcal{F}in.$$
This is clearly true in classical products restricted by $\varphi(x)$ (use Axiom of Choice).
In Lemma \[lem6+\] we need to change “finite” to “in $Fin$”, and the proof goes through, getting
\[lem6fin\] Suppose $Y_1,\dots,Y_m$ is a partition of $\B$. Suppose the sequence $$<\Theta_0(x_0,\dots,x_k,x_{k+1}),\dots,\Theta_m(x_0,\dots,x_k,x_{k+1})>$$ is a partition. Suppose $f_1,\dots,f_k\in R$ and $$Y_j\subseteq [[\exists x_{k+1} \Theta_j(f_0,\dots,f_k,x_{k+1})]]$$ for each $j$. Suppose in addition that for each $j$ $$Y_j\setminus [[\exists x_{k+1} \varphi(x_{k+1}) \wedge \Theta_j(f_1,\dots,f_k,x_{k+1})]] \in \mathcal{F}in$$ Then there is a $g$ in $R$ so that $$Y_j\subseteq [[\Theta_j(f_1,\dots,f_k,g)]]$$ for all $j$.
We now have Axioms 1-3 and Axiom $4^{fin}$. Note that $\varphi(x)$, the restricting formula, is fixed.
There is one last Axiom 5.
$\forall x (Fin([[\neg \varphi(x)]]))$.
Call the resulting axiom set $\cA_{\varphi}$, axioms for $\varphi$-restricted products.
We have given axioms for pairs $(R, \mathcal{F}in)$, and we shall next prove that we have a Feferman-Vaught type theorem.
**The Feferman-Vaught Theorem**
===============================
The Main Theorem
----------------
\[main-th\] Let $\varphi(\bar x)$ be an $\cL$-formula. Let $R$ a commutative unital ring satisfying the axioms $\cA_{\varphi}$. Then for each $\cL_{rings}$-formula $\Theta(x_0,\dots,x_m)$ there is, by an effective procedure, a partition $$<\Theta_0(x_0,\dots,x_m),\dots,\Theta_k(x_0,\dots,x_m)>$$ of $\cL_{rings}$-formulas, and an $\cL_{Boolean}^{fin}$-formula $\psi(y_0,\dots,y_k)$ such that for all $f_1,\dots,f_m$ in $R$ $$R\models \Theta(f_0,\dots,f_m) \Leftrightarrow$$ $$(\B,\mathcal{F}in)\models \psi([[\Theta_0(f_0,\dots,f_m),\dots,\Theta_k(f_0,\dots,f_m)]]).$$
In [@FV], there is a standard inductive proof for this by induction on the complexity of $\Theta$ for the case of generalized products. These are the products that are equipped with extra relations making it into a generalized product. That proof can be modified to go through for the case of restricted products with respect to a given formula $\varphi$ (which is a substructure of a generalized product). This modification can be made to work in the case of our rings $R$.
If $\Theta(x_0,\dots,x_m)$ is a quantifier-free formula, then we can take the Boolean formula $[[\Theta(x_0,\dots,x_m)]]=1$. Then for all $f_1,\dots,f_m\in R$, $$R\models \Theta(f_0,\dots,f_m) \Leftrightarrow \B\models [[\Theta(f_0,\dots,f_m)=1]].$$
Now suppose that $\Theta$ is of the form $$\exists x_{m+1} \Theta^*(x_0,\dots,x_m,x_{m+1}),$$ assuming the result known for $\Theta^*$.
Now for any $f_1,\dots,f_m\in R$, $$R\models \exists x_{m+1} \Theta^*(f_0,\dots,f_m,x_{m+1})$$ if and only if
$(*_1)$ for some $g\in R$ $$R\models \Theta^*(f_0,\dots,f_m,g).$$
By the inductive hypothesis, there is a partition $<\theta'_1,\dots,\theta'_{k'}>$ and a Boolean formula $\Phi'$ (both associated to $\Theta^*$) such that, $(*_1)$ is equivalent to
$(*_2)$ for some $g\in R$ $$R\models \Phi'([[\theta'_0(f_1,\dots,f_m),g]],\dots,[[\theta'_{k'}(f_1,\dots,f_m),g]]).$$
Now we us put $k=k'$, and $$\theta_j=\exists x_{k+1} \theta'_j, \ 0\leq j\leq k',$$ and define the following Boolean formula $$\Phi(z_0,\dots,z_k)=\exists y_1,\dots \exists y_k Part_k(y_0,\dots,y_k) \bigwedge_{j} y_j \leq z_j \Leftrightarrow$$ $$Fin(y_j\setminus [[\exists x_{k+1} \varphi(x_{k+1}) \wedge \theta'_j(f_1,\dots,f_k,x_{k+1})]]) \wedge \Phi'(y_0,\dots,y_k).$$
We show that $(*_2)$ is equivalent to
$$(*_3)\ \ \ \ \ \ \ \B\models \Phi([[\theta_0(f_1,\dots,f_k)]],\dots,[[\theta_k(f_1,\dots,f_k)]]).$$
Assume $(*_2)$. Define $y_j=[[\theta'_j(f_1,\dots,f_k,g)]]$. Then for each $j$ $$y_j\leq [[\theta_j(f_1,\dots,f_k)]].$$ Since $<\theta_1,\dots,\theta_k>$ is a partition, $Part_k(y_0,\dots,y_k)$ holds and $(*_3)$ follows.
Conversely, suppose that $(*_3)$ holds. Then there are elements $b_j$ that form a partition of $\B$, and for each $j$ we have $$b_j \leq [[\exists x_{k+1} \Theta_j(f_0,\dots,f_k,x_{k+1})]],$$ and such that $$\B\models Fin(b_j\setminus [[\exists x_{k+1} \varphi(x_{k+1}) \wedge \Theta_j(f_1,\dots,f_k,x_{k+1})]])$$ and $$\B \models \Phi'(b_0,\dots,b_k).$$
By Lemma \[lem6fin\] there is $g\in R$ such that for all $j$ $$b_j\subseteq [[\theta'_j(f_1,\dots,f_k,g)]].$$ Since $<b_1,\dots,b_k>$ and $<\theta'_1,\dots,\theta'_k>$ are both partitions, for all $j$ we have $$b_j=[[\theta'_j(f_1,\dots,f_k,g)]].$$ This proves $(*)_2$.
For $R$ as above, with restricting formula $\varphi$, $$R \equiv \prod_{e\text{~atom~of~}\B}^{(\varphi)} R_e,$$ the restricted product with respect to $\varphi$.
Both rings have the same idempotents, the same ideal $\mathcal{F}in$, and same $R_e$ ($e$ an atom), the same $\varphi$, and satisfy the axioms $\cA_{\varphi}$.
Note the effectivity and uniformity of Theorem \[main-th\] in $\varphi$ and all rings satisfying the axioms $\cA_{\varphi}$.
Ring-Theoretic Definability of $\mathcal{F}in$
----------------------------------------------
In [@DM-ad] we show that the ideal $\mathcal{F}in$ is $\cL_{rings}$-definable uniformly in all $\A_K$, $K$ a number field. In fact there is ring-theoretical definition of $\mathcal{F}in$ for a large class of rings satisfying our axioms (and the definition is in a clear sense uniform in $\varphi$).
We have to require the following of $R$ and $\varphi$.
$(\sharp$): Suppose $e\in \mathcal{F}in$. Then then there are $g,h\in R$ so that $$[[e=1]]\subseteq [[gh=1 \wedge \varphi(g) \wedge \neg \varphi(h)]].$$
Note that this is true when $R=\A_{\Q}$.
Now we proceed to an $\cL_{rings}$-definition of $Fin$ assuming $\sharp$.
Suppose first $f\in R$ is an idempotent and $[[f=1]]\notin Fin$. Then there is no $g,h\in R$ with $$[[f=1]]\subseteq [[gh=1 \wedge \varphi(g) \wedge \neg \varphi(h)]] \subseteq [[\neg \varphi(h)]] \in Fin.$$ On the other hand, suppose $[[f=1]]\in Fin$. Then by $(\sharp)$ there exist $g,h \in R$ with $$[[f=1]]\subseteq [[gh=1 \wedge \varphi(g) \wedge \neg \varphi(h)]].$$ So we have.
\[def-fin\] Suppose $R$ satisfies $\sharp$. Then $Fin$ is definable.
$e=[[e=1]]$, so $$e\in \mathcal{F}in \Leftrightarrow \exists g \exists h (e\leq [[gh=1 \wedge \varphi(g) \wedge \neg \varphi(h)]]).$$
| {
"pile_set_name": "ArXiv"
} |
=0.5in =0.15in =6ex =2ex
plus 0.2pt minus 0.2pt plus 0.2pt minus 0.2pt
The Husain-Kuchař Model:\
Time Variables and Non-degenerate Metrics\
J. Fernando Barbero G., Alfredo Tiemblo and\
Romualdo Tresguerres
Centro de Física “Miguel Catalán”,\
Instituto de Matemáticas\
y Física Fundamental, C.S.I.C.\
Serrano 113 bis, 28006 Madrid, Spain\
November 26, 1997\
ABSTRACT
We study the Husain-Kuchař model by introducing a new action principle similar to the self-dual action used in the Ashtekar variables approach to Quantum Gravity. This new action has several interesting features; among them, the presence of a scalar time variable that allows the definition of geometric observables without adding new degrees of freedom, the appearance of a natural non-degenerate four-metric and the possibility of coupling ordinary matter.
PACS number(s): 04.20.Cv, 04.20.Fy
Introduction
============
In the long quest to understand General Relativity (G.R.) the use of toy models has a long tradition. This is especially true in Quantum Gravity and Quantum Cosmology where they have allowed to obtain some, otherwise very difficult to get, information. However, this does not come without a price because one is usually forced to introduce very strong simplifying assumptions and, quite often, some of the key features of the theory are lost. Though a final judgement on the success of this approach can only be made once a consistent Quantum Gravity theory is found, it is possible, in principle, to get some clues on how well one is doing by considering widely different toy models.
Bianchi models (see, for example, [@Ryan]) are obtained by imposing homogeneity conditions on the gravitational variables. Their high symmetry has the consequence of killing most of the degrees of freedom of the full theory leaving only a finite number of them. They have been widely used in Quantum Cosmology mainly because the equations obtained upon quantization are more or less tractable.
There are other (less known) toy models that achieve the goal of simplifying the theory by going in the opposite direction: [*adding*]{} degrees of freedom. Chief among them is the Husain-Kuchař model [@HuKu] (H-K in the following). This model is quite interesting because it has some of the features that make G.R. so difficult to deal with in the quantum regime, in particular diffeomorphism invariance, but is significantly simpler because it lacks the Hamiltonian constraint (another important source of difficulties in full G.R.). This has the effect of [*increasing*]{} the number of degrees of freedom per space point from two to three.
To illustrate with a picture the different and complementary roles played by these two approaches one can make the following analogy: Portray G.R. as a complicated, knotted, two-dimensional-surface $\Sigma$ embedded in ${\rm I\!R^3}$. Working with Bianchi models is something akin to trying to get information about $\Sigma$ by looking at a finite number of points on it. The H-K model, on the other hand, is like trying to gather information by studying the whole ${\rm I\!R^3}$. Clearly some crucial features are lost in both approaches but, still, they provide useful and complementary views about $\Sigma$.[^1]
The H-K model, in its usual formulation (see [@Fer1] for some alternative descriptions), can be conveniently derived from an action principle very close to the self-dual action [@SJS] from which the Ashtekar approach to classical and quantum G.R. [@Ash] can be found. The phase space of the Hamiltonian description of both theories is the same: it is coordinatized by a $SO(3)$ connection and a densitized (inverse) triad canonically conjugate to it. Their crucial difference is the absence of a Hamiltonian constraint in the H-K model. The usual interpretation of this lack of “dynamics" is the following: By using the frame field in terms of which the H-K action is written[^2] one can build a degenerate four-metric $g_{ab}$ and a densitized vector field $\tilde{n}^a$ (that can be de-densitized by means of an auxiliary space-time foliation). The lack of dynamics can be seen as the fact that the Lie derivative of $g_{ab}$ in the direction of $n^a$ is zero.
The four-dimensional metric that we can build from the frame field in the H-K action is degenerate. This can lead to the erroneous conclusion that the model describes [*only*]{} degenerate four-metrics; a fact that has induced some authors to claim, for example, that ordinary matter cannot be coupled to the model. We will show that this is not the case in due time but at this point we urge the reader to think about the following paradoxical situation: The fact that the Hamiltonian constraint is missing from the H-K model means that the constraint hypersurface of G.R. in the Ashtekar formulation is contained in the H-K one, hence, every solution to G.R (for example Minkowski space-time) is a solution to the H-K model. How can we then describe these G.R. solutions in terms of the fields present in the H-K action if we only have a $4\times 3$ frame field available?
The solution to this problem that we give in the paper has some unexpected implications that make it quite attractive. On one hand it provides an elegant way to define quantum geometric observables (such as areas and volumes) without having to resort to increasing the number of physical degrees of freedom as in previous approaches [@Hu],[@Rov1]. On the other, it allows the introduction of a kind of time variable in the double sense that dynamics can be referred to it and also that the scalar constraint (that we need now in order to get the correct counting of degrees of freedom) is linear in its canonically conjugate momentum (so that, upon quantization it gives a Schrödinger type of equation).
The main result of the paper is that it is possible to obtain the H-K model from an action principle (also related to the self-dual action) that admits an interpretation in terms of non-degenerate four dimensional metrics. This is achieved by introducing a scalar field that can be interpreted, in a sense that will be made more precise later, as the time variable mentioned before. This will not only solve the paradox presented above but also will provide a means to couple ordinary matter thus enhancing the usefulness of H-K as a toy model. We hope that the possible interpretation of this scalar field as time will help to shed some light on the problem of time in full G.R.
The paper is organized as follows. This introduction is followed by section II where the usual formulation of the Husain-Kuchař model is briefly reviewed. The new action principle, that is the object of this paper, is introduced in section III where we derive it from the well known self-dual action for G.R. The details of the Hamiltonian formulation of our model are spelled out in section IV. There we thoroughly study the derivation of the constraints of the theory and discuss their interpretation. In section V we compare the field equations in both the usual and the new formulation for the H-K model in order to show that they are not in contradiction (a non-trivial fact as the number of equations is different in both cases). Section VI gives a different proof of the equivalence of our “non-degenerate" formulation and the usual one at the Lagrangian level. We also show that the addition of a cosmological constant (made possible in our scheme by the availability of a non-degenerate four-metric) does not lead us beyond the H-K model. We end the paper with section VII, where we give our conclusions and general comments, and an appendix that contains some details of the computations needed to disentangle the constraints in our formulation.
The Husain-Kuchař Model: A Brief Review
=======================================
We review in this section the H-K model in its usual formulation in order to describe its main features and collect the most important formulas for future reference. We start from the action [@HuKu] $$S=\frac{1}{2}\int_{{\cal M}}d^4 x \;\tilde{\eta}^{abcd}\epsilon_{ijk}e_a^i e_b^j F_{cd}^k
\label{1}$$ where our notation is the following: ${\cal M}$ is a four-dimensional manifold ${\cal M}={\rm I\!R}\times\Sigma$ with $\Sigma$ a three-dimensional manifold (that we take compact and without boundary so that we can freely integrate by parts). Curved space-time indices are represented by lower case Latin letters from the beginning of the alphabet. We will make no distinction between 4-dimensional and 3-dimensional indices. The dimensionality of a certain field will be clear from the context. The three and four dimensional Levi-Civita tensor densities will be denoted as $\tilde{\eta}^{abc}$ and $\tilde{\eta}^{abcd}$ respectively (${
\begin{picture}(6,3)(-2,-2)
\put(1,-14){$\tilde{}$}
\put(-2,-2){$\eta$}
\end{picture}}_{abc}$ and ${
\begin{picture}(6,3)(-2,-2)
\put(1,-14){$\tilde{}$}
\put(-2,-2){$\eta$}
\end{picture}}_{abcd}$ are their inverses). We use the convention of representing the density weights of geometrical objects by using tildes above (positive) and below (negative) the stem letter representing them. Internal $SO(3)$ indices, running from 1 to 3 will be denoted by Latin letters form the middle of the alphabet and the internal Levi-Civita tensor as $\epsilon^{ijk}$. We will also use a $SO(3)$ connection $A_a^i(x)$ that defines a covariant derivative acting on internal indices as $\nabla_a\lambda_i=\partial_a \lambda_i+\epsilon_{ijk}A_a^j \lambda^k$ and can be extended to space-time indices by using any torsion-free space-time connection; none of the results that we present in the paper will depend on the extension chosen. The curvature of $A_a^i(x)$ is defined as $F_{ab}^i=2\partial_{[a} A_{b]}^i+\epsilon^{i}_{\;jk}A_a^j A_b^k$. The frame field $e_a^i$ in the previous action is a $4\times3$ matrix; we will reserve the name triad for its projection on the 3-dimensional slices used in the Hamiltonian formalism.
The field equations derived from (\[1\]) are $$\begin{array}{l}
\epsilon_{ijk} e_{[b}^j F_{cd]}^k=0\\
\\
\epsilon_{ijk} e_{[b}^j \nabla_{c}e_{d]}^k=0
\end{array}
\label{2}$$ Some interesting features of (\[2\]) are summarized in the following formulas $$\begin{aligned}
& & {\tilde n}^a F_{ab}^i=0\nonumber\\
& & {\tilde n}^a \nabla_{[a}e_{b]}^i=0\label{3}\\
& & {\cal L}_{n^a} (e_a^i e_{bi})=0\nonumber\end{aligned}$$ where ${\tilde n}^a=\frac{1}{3!}{\tilde\eta}^{abcd}\epsilon_{ijk}e_b^i
e_c^j e_d^k$, $n^a={\tilde n}^a/{\tilde e}$, and ${\tilde e}$ is defined by means of an auxiliary foliation defined by a scalar function $t$ as ${\tilde e}\equiv {\tilde n}^a \partial_a t$. ${\cal L}_{n^a}$ denotes the Lie derivative along the direction defined by $n^a$. The first two equations in (\[3\]) explain why we do not have a dynamics in the model [@HuKu] (the projections of the field equations onto the direction normal to the spatial slices are zero) while the last one, which is a consequence of the others, displays this lack of evolution as the fact that the Lie derivative of the degenerate four-metric $e_a^i e_{bi}$ along $n^a$ is zero.
The meaning of this model is best understood in the Hamiltonian framework. In order to define it we introduce a foliation by means of a scalar function $t$ and a congruence of curves (nowhere tangent to the surfaces of the foliation) parametrized by $t$ whose tangent vectors we denote $t^a$. By doing this we have that $t^a\partial_a t=1$ and, hence, the time derivatives can be interpreted as the Lie derivatives along the direction defined by $t^a$. We can write (\[1\]) as $$S=\int dt\;\int_{\Sigma}d^3 x\;\left\{{\dot A}_a^i\left[{\tilde \eta}^{abc}\epsilon_{ijk}e_b^j e_c^k\right]+A_0^i\nabla_a\left[{\tilde \eta}^{abc}\epsilon_{ijk}e_b^j e_c^k\right]+e_0^i\left[{\tilde \eta}^{abc}\epsilon_{ijk}e_a^j F_{bc}^k\right] \right\}$$ where the dots denote time derivatives of the fields (Lie derivatives along the direction defined by $t^a$), $A_0^i\equiv t^a A_a^i$, and $e_0^i\equiv t^a e_a^i$. After following the usual Dirac procedure [@Dir] one finds out that the phase space of the model is coordinatized by an $SO(3)$ connection $A_a^i$ and a canonically conjugate densitized triad ${\tilde E}^a_i$. The first class constraints are $$\begin{array}{l}
\nabla_{a}{\tilde E}^a_i=0\\
\\
{\tilde E}^a_i F_{ab}^i=0
\end{array}$$ The first constraint (Gauss law) generates internal $SO(3)$ rotations whereas the second (known as vector constraint) generates spatial diffeomorphisms[^3]. As we can see there is no scalar constraint so that we have three degrees of freedom per space point.
From The Self-Dual Action to the Husain-Kuchař Model
====================================================
In this section we introduce a modified action principle for the Husain-Kuchař model that allows us to use four dimensional, non-degenerate metrics in order to describe it. We take as the starting point the self-dual action[^4] of Samuel, Jacobson, and Smolin [@SJS] $$S=-\frac{1}{2} \int_{\cal M}d^4 x\;{\tilde\eta}^{abcd}e_a^I e_b^JF_{cdIJ}^-
\label{6}$$ where now $e_a^I$ is a genuine tetrad field and $I=0,\ldots, 3$ are $SO(4)$ indices, $F_{ab}^{IJ-}$ is the curvature of an anti-self-dual connection $A_a^{-IJ}$ defined by $ F_{ab}^{IJ-}=2\partial_{[a}A_{b]}^{-IJ}+2A_{[a}^{-IK}A_{b]K}^{-\;\;J}$. Following [@Fer2] we write $$A_a^{-IJ}\equiv \left[
\begin{array}{cc}
0 & A_a^j\\
& \\
-A_a^i & -\epsilon^{ijk}A_{ak}
\end{array}\right]
\hspace{3cm}e_a^I\equiv\left[-\frac{1}{2}v_a\;\;\;\;\;\;e_a^i\right]$$ So that (\[6\]) becomes $$S=\frac{1}{2}\int_{\cal M}d^4x\;{\tilde\eta}^{abcd}\left[
v_a e_b^iF_{cdi}+\epsilon^{ijk}e_{ai}e_{bj}F_{cdk}\right]
\label{8}$$ As we can see the (anti)-self-dual action can be obtained by adding a term involving a 1-form field $v_a$ to the usual Husain-Kuchař action (\[1\]). A full discussion of (\[8\]) can be found in [@Fer2].
In the view of the previous formula it is natural to wonder what happens if instead of taking $v_a$ as a general one-form one considers it to be the gradient of a scalar $\nabla_a \phi$. Do we still have G.R. or something else? Let us consider then the following action $${\hat S}=\frac{1}{2}\int_{\cal M}d^4x\;{\tilde\eta}^{abcd}\left[
-e_a^iF_{bci}\nabla_d\phi +\epsilon^{ijk}e_{ai}e_{bj}F_{cdk}\right]
\label{9}$$ Before attempting to unravel its physical meaning, some preliminary remarks are in order. First of all the action is no longer $SO(4)$ invariant[^5] although it is obviously $SO(3)$ invariant. Second, we see now that $S$ is linear in the time derivatives of $\phi$ so we expect to have a scalar constraint linear in its canonically conjugate momentum (that after quantization will lead to a Schrödinger type of equation). It is natural to wonder if (\[9\]) could be an action for gravity (with an explicit time variable given by the scalar field $\phi$). The answer turns out to be in the negative though, at the end of the day, one discovers that (\[9\]) is still interesting in its own right. In order to check whether (\[9\]) describes G.R. or not we consider the field equations coming form (\[6\]) (remembering that we take now $e_a^0=\nabla_a\phi$). The field equation obtained by varying with respect to $A_{IJ}^-$ is $$\left[\nabla_{[a}\left(e_b^I e_{c]}^J\right)\right]^-=0
\label{10}$$ From (\[10\]) we find out immediately that $A_{IJ}^-$ is equal to the anti-self-dual part of the $SO(4)$ connection $\Gamma_a^{IJ}$ compatible with $e_a^I$ defined by $${\cal D}_a e_b^I=\partial_a e_b^I-\Gamma_{ab}^c e_c^I+\Gamma_a^{IK}e_{bK}=0
\label{11}$$ where $\Gamma_{ab}^c$ is the Christoffel symbol of the four-metric $g_{ab}\equiv e_a^I e_{bI}$. Notice that, generically, the determinant of $e_a^I$ $$\det e_a^I=\frac{1}{3!}{\tilde\eta}^{abcd}(\nabla_a\phi)
\epsilon^{ijk}e_{bi}e_{cj}e_{dk}$$ is different from zero so that we can invert (\[11\]) to write $\Gamma_a^{IJ}$ in terms of $e_{a}^{I}$ and its derivatives. By substituting $A_{a}^{-IJ}=\Gamma_{a}^{-IJ}[e,\phi]$ back in (\[6\]) we get $$S=\int_{\cal M}d^4y\;\sqrt{g[e,\phi]}R[e,\phi]$$ where $R$ is the scalar curvature of $g_{ab}\equiv e_a^I e_{bI}=\nabla_a\phi\nabla_b\phi+e_a^i e_{bi}$. If, by choosing $e_a^i$ and $\phi$ we can generate arbitrary and non-correlated $g_{ab}[e,\phi](x)$ and $$\delta g_{ab}[e,\phi](x)=\int_{\cal M}d^4y\;\left[\frac{\delta g_{ab}(x)}{\delta e_c^i(y)}\delta e_c^i(y)+\frac{\delta g_{ab}(x)}{\delta\phi(y)}\delta\phi(y)\right]$$ then $S$ must be an action for full G.R., otherwise, it is something else. At a certain point with coordinates $x$ it is indeed true that both $g_{ab}$ and $\delta g_{ab}$ can be chosen to be anything we want. However, it is not clear that the same conclusion is true [*for all the points*]{} in a neighborhood of $x$ due to the restrictions that we have imposed to the form of some of the components of the tetrads (in fact the main result of the paper shows that $g_{ab}$ and $\delta g_{ab}$ [*are not*]{} completely arbitrary in all the points of $\Sigma$).
Hamiltonian Formulation for the New Action
==========================================
By introducing a foliation as in section II we can write $${\hat S}=\int dt\;\int_{\Sigma}d^3 x\;\left\{{\dot A}_a^i{\tilde \eta}^{abc}\left[\epsilon_{ijk}e_b^j e_c^k-e_{bi}\nabla_c\phi\right]+A_0^i\nabla_a\left[{\tilde \eta}^{abc}\left(\epsilon_{ijk}e_b^je_c^k-e_{bi}\nabla_c\phi\right)\right]+\right.$$ $$\hspace{3cm}\left.+\frac{1}{2}{\dot\phi}\,{\tilde \eta}^{abc}e_a^iF_{bci}+e_0^i{\tilde \eta}^{abc}\left[\epsilon_{ijk}e_a^j F_{bc}^k-\frac{1}{2} F_{abi}\nabla_c\phi\right] \right\}\equiv\int dt\;L(t)$$ We denote ${\tilde\pi}^a_i(x)$, ${\tilde\pi}_i(x)$, ${\tilde\sigma}^a_i(x)$, ${\tilde\sigma}_i(x)$, and ${\tilde p}(x)$ the momenta canonically conjugate to $A_a^i(x)$, $A_0^i(x)$, $e_a^i(x)$, $e_0^i(x)$, and $\phi(x)$ (with Poisson brackets given symbolically by $\{q,p\}=1$). We find the following primary constraints $$\begin{aligned}
& & {\tilde\pi}^a_i+{\tilde\eta}^{abc}\left[e_{bi}\nabla_c\phi-\epsilon_{ijk}e_b^je_c^k\right]=0\label{15a}\\
& & {\tilde\pi}_i=0\label{15b}\\
& & {\tilde\sigma}^a_i=0\label{15c}\\
& & {\tilde\sigma}_i=0\label{15d}\\
& & 2{\tilde p}-{\tilde\eta}^{abc}e_a^iF_{bci}=0\label{15e}\end{aligned}$$ The Hamiltonian and the total Hamiltonian are $$\begin{aligned}
& & \hspace{-1cm}
H=\int_{\Sigma}d^3x\;\left\{e_0^i{\tilde\eta}^{abc}\left[
\frac{1}{2}F_{abi}\nabla_c\phi -\epsilon_{ijk}e_a^jF_{bc}^k\right]+A_0^i\nabla_a\left[
{\tilde\eta}^{abc}\left(e_{bi}\nabla_c\phi -\epsilon_{ijk}e_b^je_c^k\right)\right]\right\}
\label{16}\\
& & \hspace{-1cm}
H_T=H+\int_{\Sigma}d^3x\;\left\{\lambda_a^i\left[
{\tilde\pi}^a_i+{\tilde\eta}^{abc}\left(e_{bi}\nabla_c\phi -\epsilon_{ijk}e_b^j e_c^k\right)\right]+\right.\label{17}\\
& & \hspace{5.4cm}+\lambda^i{\tilde\pi}_i+\mu_a^i{\tilde
\sigma}^a_i+\mu^i{\tilde\sigma}_i+\left.\zeta\left[2{\tilde p}-{\tilde\eta}^{abc}e_a^iF_{bci}\right]\right\}
\nonumber\end{aligned}$$ where $\lambda_a^i(x)$, $\lambda^i(x)$, $\mu_a^i(x)$, $\mu^i(x)$, and $\zeta(x)$ are arbitrary (at this stage) Lagrange multipliers. The conservation under the evolution defined by $H_T$ of the primary constraints (\[15a\]-\[15e\]) gives the following secondary constraints $$\begin{aligned}
& & \nabla_a\left[{\tilde\eta}^{abc}\left(e_{b}^{i}\nabla_c\phi -\epsilon^{ijk}e_{bj}e_{ck}\right)\right]=0\label{18}\\
& & {\tilde\eta}^{abc}\left[\epsilon_{ijk}e_a^jF_{bc}^k-\frac{1}{2} F_{abi}\nabla_c\phi\right]=0\label{19}\end{aligned}$$ and the following conditions on the Lagrange multipliers $$\begin{aligned}
& & \hspace{-2cm}
{\tilde\eta}^{abc}\left[\left(\frac{1}{2}
\delta_{ik}\nabla_b\phi+\epsilon_{ijk}e_b^j\right)\left(\mu_c^k-\nabla_c
e_0^k-\epsilon^{klm}e_{cl}A_{0m}\right)-\zeta\nabla_b e_{ci}-\epsilon_{ijk}e_{0}^{j}\nabla_be_c^k\right]=0\label{20}\\
& & \hspace{-2cm}
{\tilde\eta}^{abc}\left[\left(\frac{1}{2}\delta_{ik}
\nabla_b\phi-\epsilon_{ijk}e_b^j\right)\left(\lambda_c^k-\nabla_c A_0^k\right)-\frac{1}{2}\zeta F_{bci}+\frac{1}{2}\epsilon_{ijk}e_0^jF_{bc}^k\right]=0\label{21}\\
& & \nonumber\\
& & \hspace{-2cm}
{\tilde\eta}^{abc}\left[\left(\mu_a^i-\nabla_a e_0^i-\epsilon^{ijk}e_{aj}A_{0k}\right)F_{bci}+2\left(\lambda_a^i-\nabla_a A_0^i\right)\nabla_b e_{ci}\right]=0\label{22}\end{aligned}$$ The conservation in time of (\[18\]) and (\[19\]) does not generate new secondary constraints but only the following conditions on the Lagrange multipliers $$\begin{aligned}
& & \hspace{-.5cm}
{\tilde\eta}^{abc}\left\{\nabla_a\left[\left(\frac{1}{2}
\delta_{ik}\nabla_b\phi+
\epsilon_{ijk}e_{b}^j\right)\mu_c^k\right]-(\nabla_a\zeta)
(\nabla_b e_{ci})-
\right.\label{23}\\
& & \hspace{6cm}\left.-\left(\frac{1}{2}\delta_{ik}\nabla_a\phi+\epsilon_{ijk}e_a^j
\right)\epsilon^k_{\;\;lm}\lambda_b^l e_c^m\right\}=0\nonumber\\
& & \hspace{-.5cm}{\tilde\eta}^{abc}\left\{\left(\frac{1}{2}\delta_{ik}
\nabla_a\phi-\epsilon_{ijk}e_{a}^j\right)\nabla_b\lambda_c^k-\frac{1}{2}\epsilon_{ijk}\mu_a^jF_{bc}^{k}+\frac{1}{2} F_{ab}^i\nabla_c\zeta\right\}=0\label{24}\end{aligned}$$ In principle, one expects that some combination of the second class constraints will be first class. The way to find out if this is the case is to solve the equations for the Lagrange multipliers. As we show in the appendix it is possible to find $\mu_a^i$ from (\[20\]) and $\lambda_a^i$ from (\[21\]) and write them in terms of $\zeta$, $e_a^i$, $e_0^i$, $A_a^i$, and $A_0^i$ $$\begin{aligned}
& & \mu_a^i=\nabla_a e_0^i+\epsilon^{ijk}e_{aj}A_{0k}+{
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{a\;\;b}^{\;\;i\;\;j}
{\tilde\eta}^{bcd}\left(\zeta\nabla_c e_d^j+\epsilon_{jkl}e_{0}^{k}\nabla_c e_d^l\right)\label{25a}\\
& & \lambda_a^i=\nabla_a A_0^i-\frac{1}{2}{
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{b\;\;a}^{\;\;j\;\;i}{\tilde\eta}^{bcd}
\left(\zeta F_{cdj}-\epsilon_{jkl}e_{0}^{k}F_{cd}^l\right)\label{25b}\end{aligned}$$ where ${
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{a\;\;b}^{\;\;i\;\;j}$ (which is calculated in the appendix) satisfies ${\tilde P}^{a\;\;b}_{\;\;i\;\;j}{
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{b\;\;c}^{\;\;j\;\;k}=\delta^a_c
\delta^i_k$. We have made the ansatz that the triad is non-degenerate (and we will continue to do so throughout the paper). After some tedious algebra it is possible to verify that (\[22\]-\[24\]) are identically satisfied by the previous $\mu_a^i$ and $\lambda_a^i$. We leave (some) of the details for the appendix.
We want to stress here the importance of paying attention to the conditions on the Lagrange multipliers that appear in the Hamiltonian analysis. If one knows beforehand what a theory means, one can usually skip the arduous solution of the consistency equations as one does not need to know the explicit form of the Lagrange multipliers once all the first class constraints have been identified. However, it is true, in general, that the Lagrange multiplier equations themselves may imply additional constraints (they are non-homogeneous linear equations) so, if one does not know the meaning of the theory one is dealing with, great attention must be paid to these equations in order to avoid missing some of the constraints and completely fail in the interpretation of the theory.
Substituting (\[25a\]-\[25b\]) in $H_T$ we get $$H_T=\int_{\cal M} d^3x\;\left\{e_0^i\left[{\tilde\eta}^{abc}\left(\frac{1}{2}
F_{abi}\nabla_c \phi-\epsilon_{ijk}e_a^j F_{bc}^k\right)-\nabla_a{\tilde\sigma}^a_i-\epsilon_{ijk}{\tilde\sigma}^a_l
{
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{a\;\;b}^{\;\;l\;\;j}{\tilde\eta}^{bcd}\nabla_c e_{d}^{k}-\right.\right.$$ $$\begin{aligned}
& & \hspace{1cm}\left.-\frac{1}{2}\left({\tilde\pi}_j^a+
{\tilde\eta}^{abc}( e_{bj}\nabla_c\phi-\epsilon_{jkl}e_b^k e_c^l)\right){
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{d\;\;a}^{\;\;m\;\;j}
{\tilde\eta}^{def}\epsilon_{imn}F_{ef}^{n}\right]-\nonumber\\
& & \hspace{1cm}-A_0^i\left(\nabla_a{\tilde\pi}_i^a+
\epsilon_{ijk}e_a^j{\tilde\sigma}^{ak}\right)+\lambda_i
{\tilde\pi}^i+\mu_i{\tilde\sigma}^i+\label{26}\\
& & \hspace{1cm}+\zeta\left[2{\tilde p}-{\tilde\eta}^{abc}e_a^iF_{bci}+{\tilde\sigma}^a_i
{
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{a\;\;b}^{\;\;i\;\;j}{\tilde\eta}^{bcd}\nabla_ce_{dj}-
\right.\nonumber\\
& & \hspace{1cm}-\left.\left.\frac{1}{2}{
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{b\;\;a}^{\;\;j\;\;i}
{\tilde\eta}^{bcd}
F_{cdj}\left({\tilde\pi}_i^a+{\tilde\eta}^{aef}( e_{ei}\nabla_f\phi-\epsilon_{ikl}e_e^k e_f^l)\right)\right]\right\}\nonumber\end{aligned}$$ The terms proportional to $e_0^i$ and $A_0^i$ together give a first class Hamiltonian and the terms proportional to $\zeta$, $\lambda_i$, and $\mu_i$ are first class constraints (each of them). Of course, we have also all the remaining constraints provided by (\[15a\]-\[15b\], \[18\], \[19\]). The first class constraints ${\tilde\pi}^i=0$ and ${\tilde\sigma}^i=0$ imply that $A_0^i$ and $e_0^i$ are arbitrary functions so we can just remove ${\tilde\pi}^i=0$ and ${\tilde\sigma}^i=0$ from (\[26\]). Furthermore, as now $A_0^i$, $e_0^i$, and $\zeta$ are arbitrary and $H_T$ is first class, the expressions that they multiply (linear combinations of first and second class constraints) must be first class constraints. In this way we get three sets of first class constraints plus the following independent second class constraints
$$\begin{aligned}
& & {\tilde\pi}^a_i+{\tilde\eta}^{abc}\left[e_{bi}\nabla_c\phi-\epsilon_{ijk}e_b^je_c^k\right]=0\label{27}\\
& & {\tilde\sigma}^a_i=0\label{28}\end{aligned}$$
These are very easy to deal with. In practice it is enough to remove ${\tilde\sigma}^a_i$ from the first class constraints and write $e_a^i$ in terms of $\phi$ and ${\tilde\pi}^a_i$ by solving (\[27\]) $$e_a^i=\frac{1}{4{\tilde{\tilde\pi}}}{
\begin{picture}(6,3)(-2,-2)
\put(1,-14){$\tilde{}$}
\put(-2,-2){$\eta$}
\end{picture}}_{abc}\left\{\pm
\left[2{\tilde{\tilde\pi}}-({\tilde\pi}^d_l\nabla_d\phi )^2\right]^{1/2}\epsilon^{ijk}
{\tilde\pi}^b_j {\tilde\pi}^c_k-2
({\tilde\pi}^d_k\nabla_d\phi){\tilde\pi}^{bk}{\tilde\pi}^{ci}
\right\}$$ where ${\tilde{\tilde\pi}}\equiv\det {\tilde\pi}^a_i$. The final Hamiltonian description is very simple. The phase space is coordinatized by the canonically conjugate pairs ($A_a^i$, ${\tilde\pi}^a_i$) and ($\phi$, ${\tilde p}$)[^6] and the first class constraints are $$\begin{aligned}
& & \nabla_a{\tilde\pi}^a_i=0\nonumber\\
& & {\tilde\pi}^b_i F_{ab}^i+{\tilde p}\nabla_a \phi=0\label{30}\\
& & {\tilde p}\mp\frac{1}{2}\left[2{\tilde{\tilde\pi}}-({\tilde\pi}^d_l\nabla_d\phi )^2\right]^{-1/2}\epsilon^{ijk}
{\tilde\pi}^a_i{\tilde\pi}^b_jF_{abk}=0\nonumber\end{aligned}$$ They are the Gauss law, that generates $SO(3)$ gauge transformations, the vector constraint that (essentially) generates diffeomorphisms, and a scalar constraint linear in ${\tilde p}$. They are first class constraints. It is convenient to write them in “weighted" form $$\begin{aligned}
& & G(N^i)=\int_{\Sigma}d^3x\;N^i\nabla_a{\tilde\pi}^a_i
\nonumber\\
& & V(N^a)=\int_{\Sigma}d^3x\;N^a\left({\tilde\pi}^b_i
F_{ab}^i+{\tilde p}\nabla_a\phi\right)\label{31}\\
& & S(N)=\int_{\Sigma}d^3x\;N\left\{{\tilde p}\mp\frac{1}{2}\left[2{\tilde{\tilde\pi}}-({\tilde\pi}^d_l\nabla_d\phi )^2\right]
^{-1/2}\epsilon^{ijk}{\tilde\pi}^a_i{\tilde\pi}^b_jF_{abk}
\right\}\nonumber\end{aligned}$$ The three-dimensional diffeomorphisms are generated by the combination of the Gauss law and the vector constraint $D(N^a)\equiv G(N^a A_a^i)-V(N^a)$. We can write now the constraint algebra $$\begin{aligned}
& & \left\{G(N^i),G(M^i)\right\}=G([N,M]^i)\hspace{1.5cm}
{\rm with}\hspace{1.5cm}[N,M]^i\equiv\epsilon^{ijk}N_jM_k
\nonumber\\
& & \left\{G(N^i),V(M^a)\right\}=0\nonumber\\
& & \left\{G(N^i),S(M)\right\}=0\label{32}\\
& & \left\{D(N^a),D(M^b)\right\}=D(-[N,M]^a)\hspace{.5cm}
{\rm with}\hspace{.5cm}[N,M]^a\equiv N^b\partial_b M^a-M^b\partial_b N^a
\nonumber\\
& & \left\{D(N^a),S(M)\right\}=S(-N^a\nabla_a M)\nonumber\\
& & \left\{S(N),S(M)\right\}=V\left[(N\partial_aM-M\partial_aN)\frac{4{\tilde\pi}^a_i{\tilde\pi}^{bi}}
{2{\tilde{\tilde\pi}}-({\tilde\pi}^d_l\nabla_d\phi )^2}\right]\nonumber\end{aligned}$$ Several remarks are now in order. First, we see that the constraints are first class. As we have 20 canonical variables per space point in $\Sigma$ and seven first class constraints we have three degrees of freedom per space point –one more that in G.R.–. Second, the Poisson bracket of the scalar constraint with itself closes and gives the vector constraint. This is in agreement with what one would expect from the arguments given by Hojman, Kuchař and Teitelboim in [@HKT] where they showed that the algebra of space-time deformations implied a constraint algebra of the type given by (\[32\]). Third, the structure of the scalar constraint is quite suggestive; it has two terms, one linear in ${\tilde p}$ and another proportional to the scalar constraint in the Euclidean Ashtekar formulation for G.R. This may signal a previously unnoticed relation between the Husain-Kuchař model and G.R.
From (\[30\]-\[31\]) we can interpret the model very easily. It is enough to impose the gauge fixing condition $\phi=0$ (admissible because $\left\{\phi(x), {\tilde p}(y)\right\}=\delta^3(x,y)$) to get rid of the scalar constraint and the $\phi$ dependent part of the vector constraint to recover the constraints of the usual H-K model, namely $$\begin{array}{c}
\nabla_a{\tilde\pi}^a_i=0\\
\\
{\tilde\pi}^a_i F_{ab}^i=0
\end{array}
\label{33}$$ This means that in our formulation of the model the gauge orbits have one extra dimension so, in rigor, the models are equivalent only modulo gauge transformations. At this point the reader may have the temptation to think that, after all, it is trivial to add a scalar constraint to (\[33\]) in order to have a time variable (just take ${\tilde p}=0$ and add the term necessary to generate diffeomorphisms on $\phi$ and ${\tilde p}$ to the vector constraint). The formulation thus obtained is, obviously, equivalent to ours (and can be derived from the action (\[9\]) by removing the derivatives of $\phi$ with an integration by parts). However, it is much less obvious (and less trivial) the fact that with a suitable choice of a scalar constraint one gets, not only a time variable, but also a way to interpret the H.K model as a theory for non-degenerate four-metrics at the Lagrangian level.
The four-dimensional Picture: Non-degenerate 4-metrics
======================================================
The four dimensional field equations coming from the action (\[9\]) are $$\begin{aligned}
& & {\tilde\eta}^{abcd}\left\{F_{abi}\nabla_c\phi-2\epsilon_{ijk}e_{a}^{j}F_{bc}^{k}\right\}=0\label{34a}\\
& & {\tilde\eta}^{abcd}\left\{\nabla_{a}e_{bi}\nabla_c\phi+
2\epsilon_{ijk}e_{a}^{j}\nabla_{b}e_{c}^{k}\right\}=0
\label{34b}\\
& & {\tilde\eta}^{abcd}(\nabla_a e_{b}^{i})F_{cdi}=0\label{34c}\end{aligned}$$ If we have a solution to these equations we can build a four-metric from the tetrad given by $(\nabla_a\phi, e_a^i)$ as $g_{ab}=\pm\nabla_a\phi\nabla_b\phi+e_a^i e_{bi}$. Notice that it is possible to write both Euclidean and Lorentzian 4-metrics by choosing the sign in front of the $\nabla_a\phi\nabla_b\phi$ term. In general one expects that $g_{ab}$ is non-degenerate as can be checked by simply showing some solutions to (\[34a\]-\[34c\]) such as $$A_a^i=0,\hspace{2cm}\phi=x^0\hspace{2cm}e_a^i=\left[
\begin{array}{ccc}
0 & 0 & 0\\
1 & 0 & 0\\
0 & 1 & 0\\
0 & 0 & 1
\end{array}\right]
\label{35}$$ As it can be seen (\[35\]) provides both the Euclidean and the Minkowski metric in ${\rm I\!R^4}$. We see that we can solve the (apparent) paradox presented in the introduction by using the scalar field that is present now in the field equations to build non-degenerate four-metrics.
It is interesting at this point to compare the new equations (\[34a\]-\[34c\]) with the old ones (\[2\]). For starters we seem to have one more equation now that we had before; however, as we show below, this equation is not independent of the others and, also, any solution to (\[2\]) is a solution to it. In the following we use a procedure similar to the one that appears in section III of [@HuKu]. Let us write $$\begin{array}{l}
E_{ab}^i\equiv\nabla_{[a}e_{b]}^i\\
\\
{\tilde n}^a
\equiv\frac{1}{3!}{\tilde\eta}^{abcd}\epsilon_{ijk}e_b^i e_c^j e_d^k\\
\\
{\tilde\eta}^a_i
\equiv-\frac{1}{2}{\tilde\eta}^{abcd}\epsilon_{ijk} e_b^j e_c^k\nabla_d\phi
\end{array}$$ Now ${\tilde n}^a{\tilde n}^b E_{ab}^i=0$ implies that there must exist ${\tilde E}_j^{\;\;i}$ such that $${\tilde n}^a E_{ab}^i=e_b^j{\tilde E}_j^{\;\;i}$$ Notice that $e_a^i$ satisfy ${\tilde n}^a {\tilde E}_a^i=0$ so that any linear combination of the $e_a^i$ such as $e_a^j{\tilde E}_j^{\;\;i}$ will also satisfy ${\tilde n}^a{\tilde E}_j^{\;\;i} e_a^j=0$. By the same reasoning there must exist ${\tilde F}_j^{\;\;i}$ such that $${\tilde n}^a F_{ab}^i=e_b^j{\tilde F}_j^{\;\;i}$$ We define also (we suppose ${\tilde n}^d\nabla_d\phi\neq 0$)$$e^{kl}\equiv\frac{1}{({\tilde n}^d\nabla_d\phi)^2}\epsilon^{ijk}{\tilde\eta}^a_i
{\tilde\eta}^b_jE_{ab}^l,\hspace{1cm}
f^{kl}\equiv\frac{1}{({\tilde n}^d\nabla_d\phi)^2}\epsilon^{ijk}{\tilde\eta}^a_i
{\tilde\eta}^b_jF_{ab}^l$$ We can extract all the content from the equations (\[34a\]-\[34c\]) by multiplying the first two by $$\epsilon^{ijk}{\tilde\eta}^a_i{\tilde\eta}^b_j
{\tilde\eta}^c_k,\hspace{2cm} \epsilon^{ijk}{\tilde n}^{[a}{\tilde\eta}^b_j
{\tilde\eta}^{c]}_k$$ and the scalar equation (\[34c\]) by $$\epsilon^{ijk}{\tilde n}^{[a}
{\tilde\eta}^b_i{\tilde\eta}^{c}_j
{\tilde\eta}^{d]}_k$$ (which is proportional to ${\tilde\eta}^{abcd}$).
The result that we obtain from (\[34a\]) is that $F^{ij}=-\frac{1}{2}\left(f^{ij}-\frac{1}{2}\delta^{ij}f\right)$ and $f^{ij}$ is symmetric and from (\[34b\]) that $E^{ij}=\frac{1}{2}\left(e^{ij}-\frac{1}{2}\delta^{ij}e\right)$ and $e^{ij}$ is symmetric, where $e$ and $f$ are the traces of $e_{ij}$ and $f_{ij}$ respectively. In terms of $e^{ij}$, $f^{ij}$, $E^{ij}$, and $F^{ij}$ the scalar equation (\[34c\]) gives $e^{ij}F_{ij}+f^{ij}E_{ij}=0$; we see now that all the solutions to (\[34a\]) and (\[34b\]) are solutions to the scalar equation and, hence, it is redundant.
If we consider now the standard H-K equations we see that there is no scalar equation there. It is possible to extract the content of (\[2\]) by using the procedure introduced above. The only difference now is that we need an auxiliary scalar function (for example the one that gives the foliation used in the passage to the Hamiltonian formulation) to define ${\tilde\eta}^a_i$. We immediately find the result that appears in [@HuKu] $e^{[ij]}=0$, $f^{[ij]}=0$, $E^{ij}=0$, and $F^{ij}=0$, so that now it is also true that the scalar equation (\[34c\]) is satisfied. In order to compare the solutions to (\[2\]) and (\[34a\]-\[34c\]) one must take into account the new symmety present in the model due to the introduction of $\phi$.
From the Old to the New Husain-Kuchař Model: Equivalence at the Lagrangian Level.
=================================================================================
Although we have seen from the Hamiltonian analysis that the new and all formulations of the H-K model are strictly equivalent it is instructive to understand this from an independent point of view because the actions (\[1\]) and (\[9\]) look quite different (in fact one could claim that (\[9\]) is really “closer" to the self-dual action for G.R. than to the H-K action).
The key idea to show this equivalence is the last result of the previous section, i.e. the fact that every solution to the ordinary H-K equations (\[2\]) also satisfies (\[34c\]). This means that nothing changes if we add this condition to the action (\[1\]) with a scalar Lagrange multiplier $\phi$. In this way we get $${\hat S}_1=\int_{\cal M}d^4 x\,{\tilde\eta}^{abcd}\left[
-\phi F_{ab}^i\nabla_c e_{di}+\epsilon_{ijk}e_a^i e_b^j F_{cd}^k\right]$$ which is obviously equivalent to (\[9\]). Actually we can go even further. From the H-K equations it is straightforward to show that (\[2\]) implies $${\tilde\eta}^{abcd}\nabla_a(\epsilon_{ijk}e_b^ie_c^j e_d^k)=0$$ so that even the action[^7] $${\hat S}_2=\int_{\cal M}d^4 x\,{\tilde\eta}^{abcd}\left[
\phi F_{ab}î\nabla_c e_{di}+\epsilon_{ijk}e_a^i e_b^j F_{cd}^k+\psi \epsilon_{ijk}\nabla_a (e_b^i e_c^j e_d^k)\right]$$ describes the H-K model. This last action admits an interesting interpretation. If we choose $\psi(x)=-\frac{\Lambda}{3!}\phi(x)$ with $\Lambda$ a real constant and consider the tetrad $e_a^I\equiv(\nabla_a\phi, e_a^i)$ whose inverse is given by $$e^a_I\equiv\frac{1}{\det e_a^I}\left[
\begin{array}{c}
{\tilde n}^a\\
{\tilde \eta}^a_i
\end{array}
\right]$$ with ${\tilde n}^a$ and ${\tilde\eta}^a$ as defined in the previous section we see that the added term is, in fact $$\int_{\cal M}d^4 x\,\Lambda(\det e_a^I)$$ that is, a cosmological constant term. This is the simplest (and trivial) instance of a matter coupling to the H-K model using the, now available, non-degenerate four metric.
An additional curious fact is that the previous term is equivalent to $$\int_{\cal M}d^4x\,\Lambda (\det e_a^I)e^a_J e^{aJ}\nabla_a\phi\nabla_b\phi$$ i.e. the coupling of the $\phi$ field to the “non-degenerate H-K model" as a free scalar. A Hamiltonian analysis of these last actions with a “cosmological constant" following the lines of section IV shows their equivalence with the usual H-K model.
Conclusions and Perspectives
============================
As we have shown in the paper it is possible to describe the Husain-Kuchař model with an action principle for non-degenerate metrics. We have accomplished this by introducing a scalar field in such a way that adds no new degrees of freedom. This scalar plays, in a sense, the role of a time variable not only because we have now a Hamiltonian constraint that is linear in its canonical momentum but also because it allows dynamics to be referred to it. Our proposal should be compared to those of other authors (especially [@Hu] and [@Rov1]). In these papers a scalar field is included as a means to define quantum gauge invariant observables, quoting Rovelli “matter observables which can be used to dynamically determine surfaces, the areas of which, we can measure". Our contribution in this respect is that we have managed to achieve this goal without introducing new degrees of freedom in the model. We find it quite appealing that in this process we get a nice interpretation of the scalar $\phi$ as time. Not only can we do this but also, as a side result, we have now the possibility of coupling ordinary matter to the model. This provides a type of theories that lie in between those that have a matter evolving in a non-dynamical background and full G.R. We think that a lot can be learned from looking at these theories; we plan to study them in the future. Notice, by the way, that we have the choice of coupling the matter fields to Euclidean or Lorentzian metrics, depending on the choice of the sign in the first term of the four-metric $g_{ab}=\pm \nabla_a\phi \nabla_b\phi+e_a^i e_{bi}$.
We want to remark at this point that not knowing beforehand what the meaning of the action (\[9\]) is, one should be very careful in order to avoid missing constraints crucial for the interpretation of the theory. That is why we have paid so much attention to the solution of the equations for the Lagrange multipliers. Also, we emphasize again the contradiction in claiming that the Husain-Kuchař model only allows for the existence of degenerate four-metrics whereas it is obviously an extension of both Euclidean and Lorentzian G.R. We believe that we have clearly solved this seemingly paradoxical fact in the paper.
Appendix
========
As we have said in the main text of the paper we have paid special attention to the solution of the Lagrange multiplier equations (\[20\]-\[24\]). The strategy that we have followed is simple. First solve (\[20\]) for $\mu_a^i$ and (\[21\]) for $\lambda_a^i$ and plug the result in the remaining ones. The result that we have obtained (for non-degenerate triads) shows that once we write these Lagrange multipliers in terms of $\zeta$, $A_{0i}$, $e_{0i}$, $A_{ai}$, $e_{ai}$, and $\phi$ the remaining equations are identically satisfied.
In order to solve (\[20\]) and (\[21\]) we need to compute the inverses (that we denote ${
\begin{picture}(9,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$P$}
\end{picture}}_{a\;\;b}^{\;\;i\;\;j}$) of the $9\times 9$ matrices $${\tilde P}^{\;a\;\;c}_{\;\;\;i\;\;k}(x)
\equiv{\tilde\eta}^{abc}\left(
\frac{1}{2}\delta_{ik}\nabla_b\phi+\epsilon_{ijk}e_b^j\right)$$ and its transponse $${\tilde P}^{\;c\;\;a}_{\;\;\;k\;\;i}(x)
\equiv{\tilde\eta}^{abc}\left(-
\frac{1}{2}\delta_{ik}\nabla_b\phi+\epsilon_{ijk}e_b^j\right)$$ where $^a_{\;\;i}$ are “double indices" that take the nine different values that make these matrices $9\times 9$. The best way to build their inverses is to explicitly solve the equation $${\tilde M}^{\;a\;\;b}_{\;\;\;i\;\;j}X_b^j\equiv{\tilde\eta}^{abc}
\left(\delta_{ij}v_c+\epsilon_{ijk}e_c^k\right)X_b^j={\tilde J}^a_i
\label{a1}$$ First we introduce the inverse triad $e^a_i$ such that $e^a_i e_a^j=\delta_i^j$ and write ${\tilde\eta}^{abc}={\tilde e}\epsilon^{ijk}e^a_ie^b_j e^c_k$ (here ${\tilde e}$ is the non-zero determinant of the triad). Introducing this in (\[a1\]), expanding, and using the notation $$X_{ij}\equiv e^a_i X_{aj},\hspace{5mm}X\equiv e^a_i X_{a}^{i}\hspace{5mm} J^a_i\equiv{\tilde J}^a_i/{\tilde e},\hspace{5mm}J_{ij}\equiv e_{ai} J^a_{j},\hspace{5mm},
J\equiv e_{ai} J^{ai}$$ we get $$\epsilon^{lmn}e^a_l X_{mi}v_n+\epsilon^{lmk}\epsilon_{ijk}e^a_l X_{m}^{\;\;j}=J^a_i$$ which, after multiplying by $e_{al}$ transforms into an equation that only involves objects with internal indices. $$\epsilon_l^{\;\;mn}X_{mi}v_n+X \delta_{il}-X_{il}=j_{li}
\label{a4}$$ Let us now take the trace of (\[a4\]), multiply it by $\epsilon_{ilp}v^p$ and by $v_i v_l$. We find the following three equations $$\begin{aligned}
& &-\epsilon^{ijk}X_{ij} v_k+2X=J\label{a5a}\\
& & X_{ij}v^iv^j-v^2X-\epsilon^{ijk}X_{ij}v_k=-\epsilon^{ijk}J_{ij}v_k\label{a5b}\\
& & X v^2-X_{ij}v^iv^j=J_{ij}v^iv^j\label{a5c}\end{aligned}$$ where $v^2\equiv v_iv^i$. Adding (\[a5b\]) and (\[a5c\]) and using (\[a5a\]) gives $$X=\frac{1}{2}\left[J+\epsilon^{ijk}J_{ij}v_k-v^iv^jJ_{ij}\right]$$ This means that we know how to express $X$ in (\[a4\]) in terms of $J_{ij}$ and $v_i$. If we look now at how the indices in the remaining $X_{ij}$ appear we see that the $i$ index is at both the second and the first place. If we could find the way to have both $i$ indices at the second place the remaining equation would be very easy to solve by inverting a simple $3\times 3$ matrix. To this end we need to know the expression for $X_{[ij]}$ in terms of $J_{ij}$. This can be computed by multiplying (\[a1\]) both by $v_a$ and $\epsilon_{ilm}e_a^m$ and eliminating tangent space indices as before. One gets $$X_{[ij]}=J_{[ij]}-\frac{1}{2}\epsilon_{ijk}J^{lk}v_l$$ Using this result back in (\[a4\]) we have $$X^{k}_{\;\;i}(\delta_{jk}-\epsilon_{jkl}v^l)=\frac{1}{2}\delta_{ij}(J+\epsilon^{pqr}
J_{pq}v_r-v^pv^qJ_{pq})+\epsilon_{ijk}J^{lk}v_l-J_{ij}
\label{a8}$$ Multiplying (\[a8\]) by $$\frac{1}{1+v^2}\left[\delta_{nj}+v_nv_j+\epsilon_{njs}v^s
\right]$$ and reintroducing the triads we finally get $$\begin{aligned}
& & {
\begin{picture}(11,3)(-2,-2)
\put(1,-12){$\tilde{}$}
\put(-2,-2){$M$}
\end{picture}}_{a\;\;b}^{\;\;i\;\;j}=
\frac{1}{2{\tilde e}(1+v^2)}\left[\delta_{nl}+v_nv_l+
\epsilon_{nlr}v^r\right]\times\nonumber\\
& & \times\left[\delta^{il}\delta_{k}^{j}-2\delta^{i}_{k}\delta^{jl}+\delta^{il}\epsilon^j_{\;\;mk}v^m+2\epsilon^{ilj}v_k-v_kv^j\delta^{il}\right]e_a^n e_b^k
\label{a10}\end{aligned}$$ The inverses of ${\tilde P}^{\;a\;\;b\;\;}_{\;\;i\;\;j}$ and its transponse are immediately obtained from the (\[a10\]). With them it is possible to check by direct substitution that the consistency equations (\[22\]-\[24\]) are identically satisfied. In practice the best strategy to do this is the following. First, eliminate the tangent space indices by multiplying by suitable combinations of inverse triads, then use the constraints (\[18\]) and (\[19\]) in the form $$\left[-\frac{1}{2}\delta_{ik}\nabla_a\phi+\epsilon_{ijk}e_a^j\right]
{\tilde\eta}^{abc}\nabla_b e_{ci}=0$$ $$\left[+\frac{1}{2}\delta_{ik}\nabla_a\phi+\epsilon_{ijk}e_a^j\right]
{\tilde\eta}^{abc}F_{bci}=0$$ In order to check (\[23\]) it is very useful to use the following identity $${\tilde\eta}^{abc}\left(P_{ai}^k\epsilon_{klm}-{\hat P}_{akl}\epsilon^k_{\;\;mi}\right)e_{c}^{m}\Lambda_{b}^{l}=0$$ where $$P_{aik}\equiv v_a\delta_{ik}-\epsilon_{ijk}e_{a}^{j}$$ $${\hat P}_{aik}\equiv v_a\delta_{ik}+\epsilon_{ijk}e_{a}^{j}$$ and $\Lambda_b^l$ is arbitrary.
[**Acknowledgments**]{}
The authors want to thank our colleagues G. Immirzi, J. Julve, J. Leon and G. Mena by their useful comments on this paper. J.F.B.G. wants to thank also J. M. Martín García for a very enlightening discussion. J.F.B.G. and R.T. are supported by C.S.I.C. contracts.
[50]{} M. P. Ryan Jr. and L. C. Shepley, [*Homogeneous Relativistic Cosmologies*]{} (Princeton University Press 1975). V. Husain and K. Kuchař, Phys. Rev. D[**42**]{} (1990) 4070. J. F. Barbero G., Int. J. Mod. Phys. D3 (1994) 379. J. Samuel, Pramana J. Phys. 28 (1987) L429.
T. Jacobson and L. Smolin, Phys. Lett. B196, (1987) 39. A. Ashtekar, Phys. Rev. Lett. 57 (1986) 2244.
A. Ashtekar, Phys. Rev. D36 (1987) 1587.
A. Ashtekar [*Non Perturbative Canonical Gravity*]{} (Notes prepared in collaboration with R. S. Tate) (World Scientific Books, Singapore, 1991). V. Husain, Phys. Rev. D[**47**]{} (1993) 5394. C. Rovelli, Nucl. Phys. B405 (1993) 797. P.A.M. Dirac, [*Lectures on Quantum Mechanics*]{}, Belfer Graduate School of Science Monograph Series number Two, Yeshiva University, New York, 1964. J. F. Barbero G. Phys. Rev. D51 (1995) 5498. S. Hojman, K. Kuchař, and C. Teitelboim, Annals Phys. 96 (1976) 88.
[^1]: This analogy is, actually, a little bit more than that because the Hamiltonian formulation of G.R. can be understood as the study, in phase space, of the hypersurface defined by the constraints.
[^2]: Being a $4\times 3$ matrix it is neither a tetrad nor a triad!
[^3]: Diffeomorphisms are actually generated by a linear combination of the Gauss law and the vector constraint.
[^4]: We actually use anti-self-dual fields for calculational purposes.
[^5]: Because the gradient of a scalar function does not transform as the zero component of a $SO(4)$ vector [@Fer2].
[^6]: This is the symplectic structure given by the Dirac brackets.
[^7]: There are even more possibilities that we do not discuss here.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Kibok Lee$^{*}$Kimin Lee$^{\dagger}$Kyle Min$^{*}$Yuting Zhang$^{*}$Jinwoo Shin$^{\dagger}$Honglak Lee$^{\ddagger*}$\
$^{*}$University of Michigan, Ann Arbor, MI, USA\
$^{\dagger}$Korea Advanced Institute of Science and Technology, Daejeon, Korea\
$^{\ddagger}$Google Brain, Mountain View, CA, USA\
bibliography:
- 'ref.bib'
title: Hierarchical Novelty Detection for Visual Object Recognition
---
=1
Acknowledgements {#acknowledgements .unnumbered}
================
This work was supported in part by Software R&D Center, Samsung Electronics Co., Ltd., Kwanjeong Educational Foundation Scholarship, Sloan Research Fellowship, and DARPA Explainable AI (XAI) program \#313498. We also thank Zeynep Akata, Yongqin Xian, Junhyuk Oh, Lajanugen Logeswaran, Sungryull Sohn, Jongwook Choi, and Yijie Guo for helpful discussions.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Off-equilibrium dynamics of a three-dimensional lattice model with nearest- and next nearest-neighbors exclusions is studied. At equilibrium, the model undergoes a first-order fluid-solid transition. Non-equilibrium filling, through random sequential adsorption with diffusion, creates amorphous structures and terminates at a disordered state with random closest packing density that lies in the equilibrium solid regime. The approach towards random closest packing is characterized by two distinct power-law regimes, reflecting the formation of small densely packed grains in the long time regime of the filling process. We then study the fixed-density relaxation of these amorphous structures towards the solid phase. The route to crystallization is shown to deviate from the simple grain growth proposed by classical nucleation theory. Our measurements suggest that relaxation is driven mainly by coalescence of neighboring crystallized grains which exist in the initial amorphous state.'
author:
- 'H. Levit'
- 'Z. Rotman'
- 'E. Eisenberg'
title: |
Jamming, relaxation and crystallization of a super-cooled fluid\
in a three-dimensional lattice
---
Introduction {#introduction .unnumbered}
============
Random sequential adsorption (RSA) processes are well-studied examples of off-equilibrium dynamics [@evans93]. In these processes, particles are sequentially introduced into a system at random locations, and get adsorbed if this location is compatible with the other particles. For models with inter-particle hard-core exclusion, the process ends when all deposition options are exhausted. Typically, RSA processes terminate at a limiting density far lower than $\rho_{CP}$, the closest packing density of the respective system. For example, random deposition of dimers on a one-dimensional (1D) lattice creates sites whose two nearest neighbors are filled. These sites can never be filled and the process jams with terminal density $\theta_J=0.864664$ [@flory39].
When adsorbed particles are allowed to diffuse, jammed states can be relaxed by particle movement (RSAD models). Jamming density in RSAD models is naturally higher and in some cases closest packing density is reached. In some cases, such as the 1D lattice RSAD and in the square lattice exclusion models N1 and N2 (particles deposited on a square lattice, where each particle excludes deposition on its four nearest-neighbors (N1) or up to second nearest neighbors (N2)), the density converges to $\rho_{CP}$ via a $t^{-1/2}$ power law [@wnp93m; @wnp93p; @eisbar98]. In contrast, in the 2D N3 lattice model (exclusion up to third nearest neighbors) the density converges via a $t^{-1}$ power law to a limiting density lower than $\rho_{CP}$, which is termed the [*random closest packing density*]{} $\rho_{RCP}=0.171626< \rho_{CP}=0.2$ [@eisbar99].
Random closest packing states are characterized by the absence of mobile particles (but a few rattlers). Thus, diffusive relaxation from these off-equilibrium states is impossible. However, off-equilibrium diffusive relaxation can be studied by terminating RSAD filling process at a predetermined density, lower than the random closest packing one, and following relaxation dynamics towards crystallization. Better characterization of the crystallization process is of interest (see e.g. [@binsta76; @heerklei83; @Auerfren01]), as some of its aspects, most notably the crystallization rate, are not well explained by standard nucleation theory [@Auerfren01]. In addition, recent works have highlighted the possibility that while classical nucleation theory describes well the nucleation at concentrations slightly above the coexistence region, highly saturated super-cooled fluids undergo different mechanisms of nucleation. These include spinodal nucleation and a third mechanism, yet to be fully understood, which is characterized by small displacements of the particles [@pusey1; @pusey3].
Here we study a three-dimensional (3D) model on the cubic lattice, where the inter-particle interaction is limited to hard-core exclusion up to the second-nearest neighbors, henceforth termed the 3DN2 model. This can be thought of as a system of hard particles of volume $4$ sites, whose closest packing configuration is a body-centered cubic (BCC) arrangement with a lattice constant equal to 2 lattice sites (Fig. \[model000\]). The closest packing [*particle density*]{} is therefore $\rho_{CP}=0.25$. In the following we measure all densities relative to this maximal particle density ($n=\rho /\rho_{CP}$), i.e. we use [*volume density*]{}. We start by exploring the equilibrium behavior of the 3DN2 model, and show that it undergoes a first-order phase transition, with a density gap between $n_f=0.413$ and $n_s=0.519$. We then look at the RSAD dynamics, and show that unlike the 2D N3 model, where the random closest packing density lies within the density gap, the RSAD dynamics of 3DN2 terminates at density $n_{RCP}=0.81$, well within the solid phase. Thus, RSAD dynamics enables us to compress the system deep into the super-cooled regime, while keeping the system microscopically amorphous.
Finally, we study the diffusive relaxation of the amorphous state as a function of density. The study of the off-equilibrium relaxation of disordered super-cooled configurations towards the equilibrium solid has attracted much interest, mainly in the context of super-cooled glass-formers near the glass transition [@ediger; @cavagna; @berthier; @roteis09]. In contrast, here this relaxation process leads into the equilibrium crystal. Yet, it turns out that several characteristics of the dynamics are reminiscent of glassy relaxation. In fact, it has been argued that in a number of systems an ideal glass transition could be related to the kinetic spinodal [@pusey1; @2dlj; @swift; @cavspin1; @cavagna03; @roteis10], in the sense that spontaneous crystallization of many stable micro-crystallites could lead to a kinetic arrest due to the competition between the different crystalline orders. In comparison, the 3DN2 system presents no such competition, and thus the kinetic spinodal is manifested only in the form of a modified solidification process. Indeed, we find that the route to crystallization in the high densities, well above the equilibrium transition density, is characterized by most particles moving as little as needed to shift to the sublattice on which the system crystallizes, in agreement with recent works on glassy hard spheres system [@pusey3]. Furthermore, we observe that a leading mechanism is the alignment of neighboring micro-crystallites, which exist in the amorphous [*initial state*]{}, against each other.
![A hard core particle which enforces the model’s exclusion rules (left). The central cube is one cubic-lattice site and the whole particle spans 2 lattice sites in each direction. Several configurations are shown in order to demonstrate the allowed nearest neighbors and the closest packing configuration - a BCC lattice (right).[]{data-label="model000"}](fig1a.png "fig:"){width="4cm" height="4cm"} ![A hard core particle which enforces the model’s exclusion rules (left). The central cube is one cubic-lattice site and the whole particle spans 2 lattice sites in each direction. Several configurations are shown in order to demonstrate the allowed nearest neighbors and the closest packing configuration - a BCC lattice (right).[]{data-label="model000"}](fig1b.png "fig:"){width="4cm" height="4cm"}
![(Color online) The phase transition. (a) A histogram of the density at constant $z\simeq z_{c}$ showing the system divides its time between two, high (solid) density and low (liquid) density, states, meaning the transition is of first order. (b) The average density in the vicinity of the transition as measured in lattices of different linear size. The inset shows the finite size scaling of $z_{c}$ according to the locations of the peaks in the compressibility (dashed line) and of the dips in Binder’s cumulant (Eq. \[binc\]) (solid line). (c) and (d) are the measured $\kappa(z)$ and $U_L(z)$, respectively, and (in the insets) the finite size scaling of their critical values.[]{data-label="thermo_pt"}](fig2a.png "fig:"){width="7cm" height="5cm"}\
![(Color online) The phase transition. (a) A histogram of the density at constant $z\simeq z_{c}$ showing the system divides its time between two, high (solid) density and low (liquid) density, states, meaning the transition is of first order. (b) The average density in the vicinity of the transition as measured in lattices of different linear size. The inset shows the finite size scaling of $z_{c}$ according to the locations of the peaks in the compressibility (dashed line) and of the dips in Binder’s cumulant (Eq. \[binc\]) (solid line). (c) and (d) are the measured $\kappa(z)$ and $U_L(z)$, respectively, and (in the insets) the finite size scaling of their critical values.[]{data-label="thermo_pt"}](fig2b.png "fig:"){width="7cm" height="5cm"}\
![(Color online) The phase transition. (a) A histogram of the density at constant $z\simeq z_{c}$ showing the system divides its time between two, high (solid) density and low (liquid) density, states, meaning the transition is of first order. (b) The average density in the vicinity of the transition as measured in lattices of different linear size. The inset shows the finite size scaling of $z_{c}$ according to the locations of the peaks in the compressibility (dashed line) and of the dips in Binder’s cumulant (Eq. \[binc\]) (solid line). (c) and (d) are the measured $\kappa(z)$ and $U_L(z)$, respectively, and (in the insets) the finite size scaling of their critical values.[]{data-label="thermo_pt"}](fig2c.png "fig:"){width="7cm" height="5cm"}\
![(Color online) The phase transition. (a) A histogram of the density at constant $z\simeq z_{c}$ showing the system divides its time between two, high (solid) density and low (liquid) density, states, meaning the transition is of first order. (b) The average density in the vicinity of the transition as measured in lattices of different linear size. The inset shows the finite size scaling of $z_{c}$ according to the locations of the peaks in the compressibility (dashed line) and of the dips in Binder’s cumulant (Eq. \[binc\]) (solid line). (c) and (d) are the measured $\kappa(z)$ and $U_L(z)$, respectively, and (in the insets) the finite size scaling of their critical values.[]{data-label="thermo_pt"}](fig2d.png "fig:"){width="7cm" height="5cm"}\
Thermodynamics {#thermodynamics .unnumbered}
==============
To characterize the model’s thermodynamics, we used Monte-Carlo (MC) simulations of a grand canonical ensemble. In each MC step, particles are allowed to be adsorbed on the lattice (while adhering to the exclusion rules) or desorbed from it. The probability of adsorbing/desorbing a particle is governed by the chemical potential $\mu$. We started the simulations with an empty lattice and raised the activity $z=e^{\beta\mu}$ monotonically (compressing, or cooling), equilibrating the system after each raise. At the critical regime, the distribution of the density for a finite system (at a constant $z$) is bimodal (Figure \[thermo\_pt\](a)), reflecting the transitions between two phases of different density - higher density solid and lower density fluid. Thus, the transition is a first order one.
We used finite size scaling, with lattices of sizes $L=12,14,16...$ $30$, to extrapolate the transition parameters at the thermodynamic limit [@binlan84]. This analysis (Figure \[thermo\_pt\]) verified that the transition is indeed first order. The compressibility, $$\kappa =\left( \langle n^2\rangle -\langle n\rangle ^2 \right) \cdot L^3,$$ exhibits a rounded peak at the transition, whose height is proportional to the volume, as expected for a first-order transition. Binder’s cumulant, the forth order moment of the form $$U_L=1-\frac{\langle n^4\rangle }{3\cdot \langle n^2\rangle ^2},
\label{binc}$$ also conforms to the first-order behavior - it is constant far from the transition, and near it it has a dip whose height is proportional to $L^{-3}$ [@binder81]. The location of the transition, can be extracted from either the positions of $\kappa$’s maxima or of $U_L$’s minima. Both converge (the latter faster than the first) towards a value of $z_c=1.7010(5)$. The limiting densities of the liquid and solid phases are those around which the distribution of the density peaks, $n_f=0.4131\pm0.003$ and $n_s=0.5195\pm0.0005$ accordingly, in agreement with a previous thermodynamic study [@Panagiotopoulos]. Here too, the finite size correction exhibits the non-anomalous $O(L^{-3})$ behavior.
Approach to Random Closest Packing {#approach-to-random-closest-packing .unnumbered}
==================================
We have studied the RSAD dynamics [@eisbar98] of the 3DN2 system in extensive MC simulations, using lattices of linear size between $64$ and $256$ (over $3$ million particles). RSAD dynamics is equivalent to a grand canonical dynamics with an infinite chemical potential, resulting in particle deposition whenever possible, and no desorption. As the only energy scale in the problem is the chemical potential, infinite chemical potential is equivalent to an infinitely fast cooling. First, we performed random sequential deposition on the (initially empty) lattice. This filling process happens at time zero and terminates at density $0.49362\pm0.00004$, in the equilibrium coexistence regime. When all adsorption possibilities are exhausted, the system is let to diffuse, i.e. particles attempt to move to a randomly selected nearest neighbor site. Whenever diffusion creates a space large enough for a particle to be added without violating the exclusion rules, a particle is immediately placed there. A rejection-free algorithm was used to reduce run-time and allow simulation of larger lattices.
![Random sequential adsorption with diffusion (RSAD). The volume density is plotted vs. time on a logarithmic scale. The results are averaged over 1500 realizations, for lattice of size $L=256$ (finite size effects were minute, thus we present only the results for $L=256$). Relative standard error is less than $0.0005$ for all $t$.[]{data-label="rcp256"}](fig3.png){width="8cm" height="5.3cm"}
![(Color online) Density vs. time along the RSAD process. The residual density (distance, in density, to the termination random closest packing density) is plotted, in order to highlight the two power-law regimes: the earlier with a residual density proportional to $t^{-3/2}$ (dashed line), and the later one exhibiting a $t^{-1}$ behavior (dotted line).[]{data-label="rcp256fit"}](fig4.png){width="8cm" height="5.3cm"}
Measurements of volume density (Fig. \[rcp256\]) clearly show that the process jams at a limiting density $n_{rcp}= 0.81095\pm0.00004$, lower than the closest packing value, but much higher than the transition region. The long-time approach towards $n_{rcp}$ follows the asymptotics $n(t)=n_{rcp}-A/t$ (see Fig. \[rcp256fit\]). However, at shorter times, one observes a regime in which the residual vacancies density follows a $t^{-3/2}$ rule. The long-time RSAD dynamics can be described by a model of holes diffusion (facilitated by diffusion of the neighboring particles) [@eisbar97]. The vacancies in the lattice can be relaxed by particle deposition only when multiple holes colocalize, which can be described as an annihilation process from the holes’ perspective. The simplified diffusion-annihilation description predicts that the vacancies density should follow a $t^{-3/2}$ law in 3D ($t^{-1}$ in 2D), according to the probability of two diffusing holes to meet (or the return probability of a random walk). Here we find that the intermediate-time behavior agrees with hole diffusion in 3D while the long-time $1/t$ behavior suggests that hole diffusion is effectively 2D. Analyzing the microscopic processes, we found that this effective dimensional reduction results from the formation of stable, perfectly packed, clusters. The holes left in the systems are trapped between clusters and perform 2D diffusive motion along the boundaries of the clusters, resulting in a slower density gain, $n_{rcp}-n(t)\sim t^{-1}$, associated with the effectively 2D diffusion. A similar phenomenon was observed for the 2D N1 and N2 models, where long-time RSAD relaxation was governed by 1D processes [@eisbar98].
The existence of packed clusters at long times is corroborated by measurements of the static density-density correlation function $$g_{stat}({\bf R})=\frac{\langle \rho({\bf r})\rho({\bf r+R})\rangle - \langle \rho({\bf r}) \rangle^2}{\rho(1-\rho)},$$ where $$\rho=\frac{1}{V}\sum_{{\bf r}_i} \rho({\bf r}_i,t)$$ is the total (time independent) number density. Taken along the principal axes and main diagonals, the correlation function is characterized by the peaks and dips manifesting the closest packing BCC order in short range: peaks for even integers and dips for odd integers (principal axes) as well as peaks for multiples of $\sqrt{3}$ (diagonals). The closely-packed clusters grow in the short time regime until they are stopped by neighboring clusters residing in other sublattices. The decay of the correlation function shows that the typical size of the clusters towards the end of the filling process is about $5$ sites. The clusters reach their maximum size when the density is a little over $0.8$, coinciding with the later time regime when density gain starts converging through the $t^{-1}$ rule (Fig. \[rcpgs\]).
![(Color online) The static correlation function $g_{stat}({\bf R})$ at different times along the filling process, taken along the principal axes and main diagonals and plotted against $R$=$|{\bf R}|$. The positive peaks (at $R$’s fitting the closest packing configuration) decay exponentially with distance. The correlation length reaches $\lambda \simeq 5.3$ sites at $\rho_{RCP}$ (regardless of lattice size), as shown by the fit to a pure exponent (dashed line). []{data-label="rcpgs"}](fig5.png){width="8cm" height="5.3cm"}
Relaxation Dynamics {#relaxation-dynamics .unnumbered}
===================
As the random closest packing density lies deep within the solid phase, RSAD dynamics may be used to create a disordered configurations at densities deep within the equilibrium solid regime. In order to study off-equilibrium relaxation of the super-cooled fluid, we therefore employ RSAD dynamics to fill the system up to some fixed density, and then let the system relax diffusionally, keeping the number of particles fixed. An important characterization of the dynamics is given by particles’ mobility. Off-equilibrium systems are often characterized by a sub-diffusive behavior. For example, some glassy systems exhibit caged dynamics where particles are trapped in small regions, within which diffusion takes place. Such systems display a short-time diffusion within the cages, followed by a period characterized by sub-diffusion as a result of the localization of the particles. The transition rates between cages determines the long-range, long-time, mobility and could be either diffusive or sub-diffusive depending on the nature of the barriers between the cages. Here, the average squared displacement displays a similar behavior, as shown in Fig. \[dynR2128\], only that the later stages are inter-mixed with the solidification process itself. The system starts at a disordered state so that the particles are caged and localized - free brownian movement is only measured at very short times, of order of one time unit, meaning the average cage size is of order a single lattice site. Then, after some (density-dependent) time, structural relaxation begins with movement between cages, and the diffusion measured is an anomalous diffusion resulting from the distribution of cages’ size. The higher the density, the lower is the exponent characterizing the sub-diffusive motion (see Fig. \[dynR2128\]). Finally, as the system crystallizes, particles get localized permanently and mobility diminishes.
![(Color online) The mean squared displacement averaged over all particles in ten realizations of a system of size $L=128$, as a function of density. The log-log plot highlights the diffusive regimes: in times of order of a single time step, the particles perform free diffusion within cages. As the cages’ sizes are of order of $1$ site, this is followed by a sub diffusive stage, where the particles are still mostly bound to their cages, and vibrate around their origins. At later times, the diffusivity is dominated by the movement of particles between cages and greatly depends on density. For $10^3<t<10^5$, we see that at densities within the transition gap the behavior is again diffusive (with a diffusion constant 100 times smaller than that of the first regime), but systems that were super-cooled into the solid regime exhibit sub-diffusive behavior, characterized by a power-law $<R^2>=Dt^{\alpha}$ with $\alpha<1$. Finally, for long times, diffusion is diminished as the systems goes through solidification (The longer times are inaccessible in the gap densities as some of the particles already reach the limit of the finite system.[]{data-label="dynR2128"}](fig6.png){width="7.5cm" height="5.3cm"}
The evolution of the relaxation process can be quantitatively studies through the density-density autocorrelation function $$C(t)=\frac{1}{\rho(1-\rho)}\left( \frac{1}{V}\sum_{{\bf r}_i}
\rho({\bf r}_i,0)\rho({\bf r}_i,t)-\rho^2\right),$$ where ${\bf r}_i$ runs over all $V$ sites, and $\rho({\bf r}_i,t)$ is one (zero) if at time t there is (not) a particle at site ${\bf r}_i$. This correlation function (or rather its long-time limit) is often used as the Edwards-Anderson order parameter in glassy materials. Beyond the glass transition, finite correlations are expected even for infinitely long times. Apparently, the model exhibits the two-step relaxation that is typical of glassy systems (Fig. \[ctc128\]). Due to the aforementioned caging phenomenon, there is very little relaxation in the short time regime as the particles mostly move back and forth within their cages. This leads to the formation of a plateau in the autocorrelation function, known as the $\beta$-regime. As density increases, the plateau extends to longer times. It is followed by the structural relaxation in the $\alpha$-regime, where cages are broken and restructured, and the correlation function decays to zero. While some of the simulations ended having finite $C(t)$ a detailed finite size scaling shows that this is a finite size effect. The $\alpha$-decay takes the form of a stretched exponential, expressing the spatially heterogeneous character of the system.
![(Color online) $C(t)$ in an $L=128$ Lattice, for densities ranging from $0.45$ to $0.8$. The results are an average over several hundred realizations (there are less realizations in the long time regime, hence the larger error bars). Black line indicating the time at which half of the particles are on the same sublattice.[]{data-label="ctc128"}](fig7.png){width="7.5cm" height="5.3cm"}
Despite the apparent similarities to the behavior of glassy systems, it should be stressed that the long-time relaxation observed here is qualitatively different from the alpha-phase relaxation seen in glassy materials. As indicated in Fig. \[ctc128\], the long time relaxation overlaps with the solidification of the system. The nature of this process and the reason it is accompanied by spatially long-ranged relaxation can be understood looking at the microscopic picture.
![(Color online) Particles’ displacements in a high-density ($n=0.75$) system at a time it has already crystallized (almost all particles are lying on the same BCC sublattice). Most particles have moved less than $\sqrt{3}$ sites.The inset shows the displacement correlation length as estimated by the decay of the displacement correlation function to $0.2$. Displacement correlation is short ranged even for times longer than crystallization time. Data presented for a system of size $L=128$ and density $n=0.75$ (over 100 realizations), and is typical for other high densities and reasonable correlation cutoffs.[]{data-label="dyndisplace"}](fig8.png){width="7.5cm" height="5.3cm"}
Microscopic Picture of the Crystallization Dynamics {#microscopic-picture-of-the-crystallization-dynamics .unnumbered}
===================================================
Detailed analysis of microscopic processes underlying the dynamics reveals some interesting features of the relaxation process. We focus on the total displacements of particles along the process (Fig. \[dyndisplace\]), and find that in the high-density regime a considerable number of the particles move only a distance of $1$ to $\sqrt{3}$ sites away from their $t=0$ location, i.e. up to one shift along each of the axes. Moreover, the particles’ movement distance is highly correlated between neighboring particles, i.e. different regions in the model have their own characteristic distance. The sizes of these clusters can be quantified by the spatial correlations of the displacement $$C_D({\bf r},t)=\langle R({0,t})R({\bf r},t) \rangle - \langle R \rangle^2,$$ which exhibit short-range correlations. In order to obtain an estimate for the length scale characterizing these displacement correlation we look at the distance at which $C_D({\bf r},t)$ decays to an arbitraty cutoff value ($0.2$). The inset of Fig.\[dyndisplace\] presents the time-dependence of this length scale. Not only are the estimated correlation lengths short-ranged compared to the static correlation measured in t=0, but they also do not grow over time. It is important to notice that even though the exact value of the length scale is as arbitrary as the threshold used, these two observations are not.
The above observations are inconsistent with simple nucleation theory picture of crystallization, according to which growing nucleation seeds are continuously fed by their fluid environment. Instead, we observe the alignment of micro-crystallites (present at $t=0$) to be the main mechanism leading to crystallization. The particles forming the boundaries of the clusters are effectively caged by the surrounding clusters, so that their movement is somewhat restricted (resulting in the sub-diffusive behavior described above). This microscopic picture fits with the global dynamic measurements described above: the displacement correlation length observed above reflect nothing but the size of the clusters moving cooperatively to re-align themselves against each other. The crystallization mechanism described here, characterized by minute displacements of particles along the process, is reminiscent of a similar phenomenon recently observed for the hard spheres glassy system [@pusey3]. It is believed that the transition to this regime has to do with the density being so high (or the super-cooled fluid being over-saturated) such that the kinetic spinodal is reached. Under these conditions, micro-crystallites of any size are stable. Thus, virtually all of the micro-crystallites naturally occurring in the initial amorphous state survive the relaxation process, and crystallization is achieved by re-alignment of these against each other. This highlights another possible connection to the glassy behavior. It was suggested that in some systems an ideal glass transition occurs at the kinetic spinodal, as micro-crystallites having competing crystal order are all stable and thus no global order may be achieved [@2dlj; @swift; @cavspin1; @cavagna03; @roteis10]. The fate of a deeply super-cooled fluid is then determined by the nature of the micro-crystallites occurring when kinetic spinodal is hit. If they are easily alignable, one would end up having a solid, while when they are not easily alignable, a stable amorphous glass would emerge.
Finally, we comment about the crystallization time. We define the crystallization time as the time (measured from the end of the RSAD process) till half of the particles occupy the same sublattice (The equilibrium occupation fraction of the leading sublattice depends only weakly on the density, changing from $0.9$ at the onset of crystallinity to $0.96$ for $n=0.6$ and approaching unity for high densities). Measuring this time as a function of density, one observe a non-monotonous picture (Fig. \[dynsolidt\]). As density increases, the preference of the solid phase over the disordered phases increases. At the same time, the dynamics becomes increasingly sluggish. The result of these two effects combined is the non-monotonous behavior. The long relaxation time in high-density systems also stems from the fact that they are prone to the formation of two or more macroscopic domains which require a long time to align with one another. The increase in the average solidification time is enhanced by the finite size effects - the larger the lattice, the longer it takes for the system to solidify.
![(Color online) The time of solidification, defined as the first time at which more than half of the particles are on the same sublattice, averaged over $77$ realizations, as a function of system size and density. Realizations in which two or more macroscopic domains are formed are averaged only if one of the domains contains more than half of all particles.[]{data-label="dynsolidt"}](fig9.png){width="7cm" height="5cm"}
Conclusion
==========
The 3DN2 model presented here exhibits a non trivial RCP dynamics. The density gain in the long-time regime, controlled by holes’ diffusion, is going through two distinguished regimes corresponding to an unordered dense liquid state, followed by a state of many micro-crystallites in which the particles mainly perform 2D diffusion on the clusters’ boundaries. Keeping the density fixed, the system relaxes to a solid. In the high-density regime, beyond the kinetic spinodal, even small micro-crystallites are stable. Relaxation is then brought about through alignment of the small solid clusters against each other. It then follows that the dynamics of the system is heterogeneous, similarly to super-cooled glass-formers. The crystallization process, is dramatically different than the one described by classical nucleation theory.
[24]{} J.W. Evans, Rev. Mod. Phys. [**65**]{}, 4 (1993) P.J. Flory, J. Am. Chem. Soc. [**61**]{}, 1518 (1939) J.S. Wang, P. Nielaba and V. Privman, Mod. Phys. Lett. B [**7**]{}, 189 (1993) J.S. Wang, P. Nielaba and V. Privman, Physica A [**199**]{}, 527 (1993) E. Eisenberg and A. Baram, Europhys. Lett. [**44**]{}, 168 (1998) E. Eisenberg and A. Baram, J. Phys. **A 33**, 1729-1734 (2000) K. Binder and D. Stauffer, Adv. Phys. [**25**]{}, 343-396 (1976) D.W. Heermann and W. Klein, Phys. Rev. Lett. [**50**]{}, 1062-1065 (1983) S. Auer and D. Frenkel, Nature [**409**]{}, 1020-1023 (2001) P.N. Pusey, E. Zaccarelli, C. Valeriani, E. Sanz, W.C.K. Poon and M.E. Cates, Phil. Trans. R. Soc. **A 367**, 4993 (2009) E. Sanz, C. Valeriani, E. Zaccarelli, W.C.K. Poon, P.N. Pusey and M.E. Cates, Phys. Rev. Lett. **106**, 215701 (2011) M.D. Ediger, Annu. Rev. Phys. Chem. **51**, 99 (2000) A. Cavagna, Phys. Rep. **476**, 51 (2009) L. Berthier and G. Birolli, Rev. Mod. Phys. **83**, 587-645 (2011) Z. Rotman and E. Eisenberg, Phys. Rev. **E 80**, 060104 (2009) G. Johnson, A.I. Mel’cuk, H. Gould, W. Klein and R.D. Mountain, Phys. Rev. E **57**, 5707 (1998) M.R. Swift, H. Bokil, R.D.M. Travasso and A.J. Bray, Phys. Rev. B **62**, 11494 (2000) A. Cavagna, I. Giardina and T.S. Grigera, Europhys. Lett. **61**, 74 (2003) A. Cavagna, I. Giardina and T.S. Grigera, J. Chem. Phys. **118**, 6974 (2003) Z. Rotman and E. Eisenberg, Phys. Rev. Lett. **105**, 225503 (2010) K. Binder and D.P. Landau, Phys. Rev. **B 30**, 1477 (1984) A.Z. Panagiotopoulos, J. Chem. Phys. [**123**]{}, 104504 (2005) K. Binder, Z. Phys **B 43**, 119-140 (1981) E. Eisenberg and A. Baram, J. Phys. [**A 30**]{}, L271 (1997)
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.